FROM

OBSERVATIONS
*° SIMULATIONS
A Conceptual Introduction to
Weather and Climate Modelling

Antonello Pasini
 FROM

OBSERVATIONS
™ SIMULATIONS
 FROM

OBSERVATIONS
ro
SIMULATIONS
A Conceptual Introduction to
Weather and Climate Modelling

Antonello Pasini
CNR - Institute of Atmospheric Pollution, Rome, Italy

Translated by
Francesca Sofri

^ World Scientific
NEW JERSEY • LONDON • SINGAPORE • BEIJING • SHANGHAI • HONGKONG • TAIPEI • CHENNAI
Published by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.

Original Italian edition:

I CAMBIAMENTI CLIMATICI — METEOROLOGIA E CLIMA SIMULATO
Copyright © 2003 by Paravia Bruno Mondadori Editori

FROM OBSERVATIONS TO SIMULATIONS

A Conceptual Introduction to Weather and Climate Modelling

Copyright © 2005 by World Scientific Publishing Co. Pte. Ltd.

All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be invented, without written permission from the Publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright
Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to
photocopy is not required from the publisher.

ISBN 981-256-475-6

SEGRETARIATO EUROPEO PER LE PUBBLICAZIONI SCIENTIFICHE

The translation of this book has been funded by SEPS: Segretariato Europeo per le Pubblicazioni
Scientifiche, Via Val d'Aposa 7, 40123 Bologna, Italy, seps@alma.unibo.it, www.seps.it

Printed in Singapore by Mainland Press

 In memory of my father Elio,
who first taught me to play with the physical world
 Acknowledgements

Many friends and colleagues helped me with their advice during the
writing of this book: it is impossible to mention all of them. For the
enormous amount of time they dedicated to me, I would like to
particularly thank Pier Francesco Coppola and Fausto D'Aprile.
Special thanks are due to Eugenia Kalnay and Malaquias Pena, who
allowed me to use a code of theirs for the creation of the figure of the
Lorenz attractor that appears on the cover of the book.
 Preface

When my children were little, I found myself several times watching

them with wonder and admiration while they were playing with
mechanical toys. In fact, if you give a child a toy, and the child is curious
enough and has enough time, he or she will eventually open it up in order
to see how it works, then will try to reconstruct it in order to play with it
again. This childish attitude is usually lost with age; but modern
scientists behave exactly like this with the systems they are studying: it is
not by chance that they are often called "grown-up babies".
In a certain sense, this book means to be precisely a journey into the
ideas that have led the scientists who study the weather and climate to
recover this childish outlook.
In the history of science, after the period of Greek philosophers and
their medieval Epigones (during which people confined themselves to
observing reality, looking for regularities that might explain its
behaviour), with Galileo Galilei scientists began to control and
manipulate reality in the laboratory, in order to induce nature to give
specific answers to specific questions. This led to great cognitive
progresses in the domain of the so-called "hard sciences", such as
physics and chemistry.
Obviously, this childish tendency to open up a toy in order to look
inside it — to disassemble it, then to reassemble its parts — is pursued
nowadays in all the areas of science, including the study of the
atmosphere and climate. As a rule, the activity of decomposing a system
in order to study its individual elements and their basic interactions does
not pose any particular problem: in the laboratory, for instance, we can
easily study the absorption of infrared radiation by carbon dioxide

IX
X From Observations to Simulations

molecules (which contributes to the so-called "greenhouse effect"); or,

regarding air as a fluid or as a mixture of gases and water, we can
analyse the movements of portions of air in simplified cases or study the
main thermodynamic processes that take place in the atmosphere. But
when we try to reconstruct the whole "toy" in the laboratory, though this
is usually possible for mechanical systems, we find that it is extremely
difficult for the atmosphere and for the Earth system: we will see this in
the course of our journey.
So, up to a few decades ago, meteorology and climatology were still
purely observational disciplines, characterised by a lot of difficulties in
achieving theoretical syntheses. Then the fruitfulness of the Galilean
experimental method (though transferred to a different set-up) was
retrieved in these fields too, and now computers and simulation models
may be regarded as "virtual laboratories" where the weather and climate
are studied. In a model, formed of equations that represent our theoretical
knowledge (and can be solved numerically) and of variables that refer to
the real data, it is possible to reconstruct the complexity of reality,
though in a simplified manner. In particular, we can simulate the
evolution of the climate system on the basis of scenarios observed in the
past or surmised for the future; and all this can be done in very little time
(tens of hours for decades of real evolution) and with the possibility of
carrying out "numerical experiments".
In this book we will deal precisely with this "methodological
revolution", which underlies our present understanding of the behaviour
of many complex systems, including the climatic one.
In our journey from observations to simulations, we will follow the
typical route of a scientific investigation, and will encourage the reader
to become qualitatively aware of the characteristics of the atmosphere
and of the Earth system, gradually finding different explanatory schemes.
We will also re-examine some classical, well-known concepts such as
"causality" and "prediction", in the light of the models and of new
concepts pertaining to the theory of dynamical systems.
To sum up: this book presents a research into complex systems that
has a huge range of practical applications, and is also contributing to a
substantial change in our outlook on nature.
A. Pasini
 Contents

Preface ix

1. Introduction 1

2. Meteorological and Climatic Observations 9

2.1 The "State" of the Weather 10
2.2 A Definition of Climate 11
2.3 An Overview of Meteorological and Climatic Observations 12
2.4 Conventional Observations 15
2.5 Satellite Observations 16
2.6 Meteorological or Climatic Observations? 21
2.7 Proxy Data 23
2.8 Is There Any Evidence that the Climate is Changing? 27

3. Naive Meteorology, Coincidences and Correlations 33

3.1 Approaching an Analysis of the Data and of Common Experience . 34
3.2 A Naive Interpretation and Its Problems 37
3.3 Coincidences and Correlations in Available Data 42
3.4 Let Us Take Stock of the Situation 50

4. The Theoretical Framework: Knowledge of Single Phenomena and

Complexity of the Earth System 51
4.1 How Can We Read the "Great Book of Nature"? 51
4.2 The Local Approach to the Study of a System 54
4.3 The Interaction between Radiation and Matter and the
Greenhouse Effect 58
4.4 Greenhouse Gases, Clouds and Aerosols 65
4.5 Approaching a Complete Scheme of Warming from the Bottom 68
4.6 Nature of the Ground and Air Warming 73
4.7 An Outline of Oceanic and Atmospheric Dynamics 77
4.8 Feedbacks and Complexity of System 81

xi
xii From Observations to Simulations

5. The Galilean Experimental Method: A Digression? 89

5.1 Aristotelian Physics of Local Motions and the Advent of
Galileo Galilei 90
5.2 The Galilean"Style" 94
5.3 A Galilean Method for Studying the Weather and the Climate? 97

6. Simulation Models 103

6.1 How Many Meanings Does the Word "Model" Have? 103
6.2 The SimulationApproach 107
6.3 Conceptual Novelties in the Simulation Method and Its Use Ill

7. Meteorological Models 117

7.1 The "Perception" of the Weather Forecasting Activity 117
7.2 The Heart of a Meteorological Model: Primitive Equations and
Their Numerical Solutions 119
7.3 Physical Parameterisations 126
7.4 Determination of Initial State and Analysis Procedure 128
7.5 The Products of a Meteorological Model 131
7.6 The Emergence of Deterministic Chaos and Ensemble Integrations 134
7.7 A Few Conceptual Remarks 142

8. Climatic Models 149

8.1 From Weather Forecasting to Climate Forecasting: What Changes? 149
8.2 The Concept of "Attractor" and Climatic Simulations 151
8.3 Approaching the Description of a Coupled and Highly Interacting
Climate System 156
8.4 Experiments for Validation and Sensitivity Testing of a
Climatic Model 161
8.5 Evolutionary Validation and Climatic Forecasts 165
8.6 Simplified Models and Regional-Scale Models 170
8.7 Simulation Results 174
8.8 Further Remarks about Climate Change and Its Study 179

9. Conclusions and Prospects 185

9.1 The Results of Climatic Models and "What Should We Do?" 186
9.2 The Future of Models for Studying the Weather and Climate 189

Bibliography 195

Index 197
 Chapter 1

Introduction

All of us who are a part of the present-day society of global information

are inundated with a constant flow of news of all sorts, often relevant to
events that have taken place a few hours or minutes before, in some
remote corner of the world. Among these reports, sometimes we receive
some news-flashes about natural events of a meteorological or climatic
type, in many cases extreme events (hurricanes, floods, droughts), that,
either directly or as an immediate consequence, have spread destruction
and death.
Sometimes the news is not so immediately tragic, but brings anxiety
about a more remote future. I refer, for instance, to the news of the
separation from the Antarctic continent of an enormous iceberg (called
Larsen B, larger than the State of Rhode Island), in March 2002.
Subsequently other, smaller icebergs broke away, and many people
thought it was reasonable to link these events to the world-wide warming
of the oceans and lower layers of the atmosphere during the last century
(a subject we will discuss further on). Should this trend go on, other
icebergs might be expected to break away; and should the ice that melts
or is released into the sea come from the cap that covers the Antarctic
land, these "plunges" into the ocean would concur in raising the mean
global level of the sea and in causing the interface between the land and
the sea, i.e. the coasts, to recede.
There are also some pieces of news that rarely reach the western
society, because they are not immediately catastrophic and concern
extremely remote places. I refer, for instance, to the fact that an entire
people, the approximately 11,000 inhabitants of the archipelago of
Tuvalu, a small south-Pacific island-state, is negotiating with the

1
2 From Observations to Simulations

governments of other states (in particular Australia and New Zealand), in

order to be allowed to take sanctuary in their territory. This application
was made necessary by the increasingly frequent flooding of their atolls
and by the rising level of the sea, which suggests the possible need for an
evacuation within the next few years.
After the immediate, facile stir caused by certain pieces of news,
there begins (but not for all of them) a stage of deeper elaboration. In
particular, a traditional deterministic analysis leads us to wonder which
causes have produced (or are producing) a certain flood, the detaching of
a certain group of icebergs or the appearance of "environmental
refugees". The rationale behind these questions springs from the
common-sense observation that, once the cause of an undesirable effect
is removed, that effect disappears as well. At this point scientists come
into play.
In the history of science, causal relationships have always been
studied carefully. Within the sphere of Greek natural philosophy, whose
summa is represented by Aristotle's work, there coexist basically two
types of physical causes: efficient cause and final cause. The former is
the current meaning of the word: it is what comes before a certain
phenomenon and determines it; the latter is the goal towards which the
caused thing tends, in the future. In modern science, finalistic thinking —
according to which a certain phenomenon or process takes place because
it tends towards a final situation, its goal — was abandoned1. The
present-day causal outlook establishes a "time's arrow": all causes come
before the effect they produce (they are situated within the light cone of
the past, to put it in relativistic terms). Moreover, within the sphere of
classical (not quantum) physics, the theoretical approach to evolutionary
phenomena is based on differential equations (ordinary or with partial
derivatives), more or less implicitly assuming that the future state of the
system under consideration is univocally determined starting from a
known past state, by means of evolution equations.

'in actual fact, in the mathematical physics of the eighteenth century there survives a
finalistic explicative concept, with the minimum-action principle or Maupertuis'
principle, within the sphere of variational calculus in dynamics: for a critical analysis and
a modern outlook that connects it to a causalistic approach, see, e.g., Yourgrau and
Mandelstam (1979).
 Introduction 3

We will analyse this paradigm further on, after having applied it to

the atmosphere system, or, more generally, to the Earth system. At
present it will be sufficient to point out that, as a rule, in this pattern the
scientific explanation of a phenomenon or of the evolution of a process
with time involves first of all the identification of its causes, then the
analysis of their way of combining to give rise to the phenomenon or
process under consideration.
The area that acts as a prototype for these causality analyses is
classical mechanics, in particular laboratory-controlled experimental
situations2. If we recall the simple dynamics problems we had to solve in
secondary school and the easy experiments we had to carry out, we
notice that in those cases the few forces that act on a material body are
easy to identify, and the total effect they produce on the body (e.g. its
acceleration) are nothing but the vectorial sum of the effects
(accelerations) that each force would produce if it were applied
individually. This property of a physical system is called linearity. If the
system under examination is correctly described by an equation such as
a = F/m , the solution of this equation, in the case of the composition of
several forces, is the sum of the solutions of the individual cases in
which, each time, only the action of a single force is considered. So once
a certain number of concurrent causes has been identified, their
composition is quite simple and the problem under examination is easy
to solve.
In some systemic ecology studies begun in the nineteen-seventies3,
for the cases that have just been described, linear causality is mentioned,
as opposed to a so-called circular causality that is considered typical of
living systems, where it becomes essential to consider the intricate
relationships and interconnections between the various elements of a
system. Though we shall not go so far as to examine the dynamics of
living systems, we must point out that also in the atmosphere, and, more
generally, when dealing with the overall Earth system, the relationships
between causes and effects can no longer be interpreted in terms of the

2
These considerations will be resumed and extended in Chapter 5, when we will discuss
the Galilean experimental methods.
3
E.g., see Bateson (1980).
4 From Observations to Simulations

simple linear causality of classical mechanics. In actual fact, what

undermines the simple linear pattern is the presence of feedback, i.e.
chains of circular two- or multi-element interactions, in which an effect
acts in turn on the cause that has generated it, increasing its effect
(positive feedback) or decreasing it (negative feedback). In Chapter 4,
when we will analyse the current theoretical knowledge of the Earth
system, we will consider some concrete examples of these complex
cause-effect relationships (which are called non-linear).
From what we have just explained, it is evident that there does not
exist an answer, in terms of a linear composition of concurring causes, to
the questions we had previously posed about the floods, icebergs and
"environmental refugees". If to this we add the fact that our theoretical
knowledge of atmospheric phenomena is linked to a description in terms
of systems of equations that can be solved analytically only in very
particular, simplified cases, we can understand why, up to a few years
ago, giving sensible answers to these questions was quite unthinkable. To
make the situation even more complicated, in many cases these
phenomena should be regarded as "extreme events", that is statistically
improbable events, and this makes it difficult also to approach their
description and prediction in terms that are not dynamic but statistical.

Within the picture we have just given of the situation, whose

elaboration lends itself to be carried out from various angles, we must
now delimit the goals that this book has set out to achieve.
When discussing phenomena relevant to meteorology, the climate and
its changes during the last few decades or centuries, the subject can be
tackled from the viewpoint of our scientific knowledge in this area
(essential for any other discussion of the problem), from the viewpoint of
the impact of these phenomena on nature and mankind (including studies
on the vulnerability of the latter and on possible adaptation strategies),
and — if there seems to be the possibility of acting in a concrete manner
to reduce the causes of the most negative phenomena — from the
viewpoint of mitigation studies relevant to what is usually called
sustainable development (a possible example is the development of
energy production methods that have a lower impact on the
environment). The third viewpoint is the one where the decisions to be
 Introduction 5

taken are the most delicate, because they affect the world-wide social,
economic and political sphere4.
The viewpoint adopted in this book belongs to the first of these three
areas. In brief, we will begin by analysing the current scientific
knowledge of meteorological and climatic phenomenology, both from
the angle of observations (Chapters 2 and 3) and from that of the
theoretical description of the Earth system and its atmosphere subsystem
(Chapter 4). In doing this, we will pay attention particularly to the
conceptually relevant aspects that reveal the complexity of the system
under examination and make it so different from the physical systems
that our school education has made familiar to us. This first part of the
book is somehow preparatory to what follows, because it supplies the
grounding required to understand the change of paradigm in the
researches in this field that is described in the second part of the book.
Chapter 5, which at first sight may seem a digression, discusses the
Galilean experimental method. The motivation of this brief excursus into
the physics of the seventeenth century is that the application of this
method was precisely what allowed physics and other so-called "hard"
scientific disciplines to achieve extremely important results and a very
evident progress in the understanding of nature. In the meantime,
meteorology and climatology, as described in the first three chapters,
remained observational disciplines, and had much trouble in attempting a
theoretical description, because of the complexity of the system that was
being studied. At this point it is natural to think that, if we could recover
a Galilean way of carrying out scientific researches, other disciplines that
up to now were purely observational might achieve important progresses
in understanding the phenomena within their province.
Chapter 6, in which simulation models are introduced, analyses
precisely the entrance of some observational disciplines into the category
4
It is not fortuitous that this schematisation follows, rather faithfully, the one proposed by
the IPCC (Intergovernmental Panel on Climate Change) in the three reports it has
recently published on the state of climatic research in the world. The IPCC was
established in 1988 by the World Meteorological Organisation (WMO) and the United
Nations Environment Programme (UNEP) with the main purpose of periodically drawing
up technical reports (and resumes for policy-makers) about the state of the art of
scientific, technical, social and economic knowledge of climate changes and their
consequences. Further on we will refer to one of these reports.
6 From Observations to Simulations

of Galilean-type sciences. Here a new paradigm in the way of performing

a scientific research, simulation, is discussed, particularly as regards the
possibility of constructing a "virtual laboratory" for the study of complex
systems.
In Chapter 7, we will describe the structure and operation of the
models for weather forecasts, highlighting their strong points and
weaknesses. In particular we will discuss the appearance of what is
called deterministic chaos, which leads to a revision of the concept of
deterministic prediction for complex systems, such as the atmosphere.
In Chapter 8, the system that is being studied in its dynamic evolution
is extended to fully include the oceans and some phenomena that are
neglected or not dealt with dynamically, for instance the so-called carbon
cycle (photosynthesis, respiration, storage). The purpose here is not to
predict the weather during the next few days, but to correctly reconstruct
the climates of the past and make it possible to analyse future scenarios
in relation to important climatic variables such as the mean temperature
and the precipitations in a certain area of the world. The positive results
and current limits of the models are evaluated.
The last chapter contains a general discussion of the importance of
the simulation-based approach to the study of the weather and climate.
This approach is evaluated from a conceptual and epistemological5 point
of view, and the prospects of future development within this paradigm,
and out of it, are presented.

The great importance that meteorology has in everyday life and the
enormous publicity that is given to the debate about climate change,
essentially based on the results of predictive models, are accompanied by
a general lack of information on the scientific practice (of which
simulation-based modelling is an integral part) that characterises these
disciplines. I hope, on the one hand, that my book will be able to bridge
this gap, on the other hand that the conceptual and epistemological

5
In the literature relevant to this area, the adjective "epistemological" is used with
meanings that sometimes differ from each other, depending on the exact meaning
ascribed to the noun "epistemology" (e.g., see Greco (1998)). Here "epistemology" is
understood as the critical and philosophical study of nature and of the procedures of
scientific activity.
 Introduction 1

reflections by which it is accompanied will explain the intellectual appeal

of a change of paradigm that has recently allowed these disciplines to be
included in the category of Galilean-type sciences. Furthermore, there is
a lot of talk about complexity and chaos, and one is led to believe that the
phenomena relevant to them are confined in some obscure sector of
physics. On the contrary, the study of the atmosphere as a subsystem of
the Earth system is an ideal and very concrete case study of complex
system, and makes it possible to evaluate the conceptual and practical
scope of a model-based approach to these themes.
 Chapter 2

Meteorological and Climatic Observations

The beginning of meteorology is conventionally traced back to the year

1643, when Evangelista Torricelli constructed the first mercury
barometer that made it possible to demonstrate the existence of
atmospheric pressure. This was also the period of the invention of the
thermometer and of the improvement of the hygrometer (for measuring
humidity), anemometer (for measuring the intensity of the wind and the
direction from which it comes), and pluviometer (for measuring the
amount of rain that has fallen during a certain period of time). In
substance, meteorology, understood as a discipline that studies the
Earth's atmosphere and the phenomena that take place in it, by means of
instrumental measurements, began in Italy at the court of the Medici in
Florence, and was developed within the sphere of the Accademia del
Cimento, the first example of a scientific society established in Europe,
which included a great number of disciples of Galileo Galilei.
In actual fact, it was not possible to speak of meteorology in the
modern sense of the term or to have some hope of also being able to
predict the future conditions of the weather until individual observations
were integrated in spatially extensive networks that made it possible to
have an overall (we might call it synoptical) three-dimensional view of
the state of the atmosphere at a certain instant in time. The forerunner of
this modern vision appeared, once again, in seventeenth-century
Florence: the grand duke Ferdinand II adopted this method in 1654 and
established the first network of meteorological observations (obviously
surface-based), with data coming even from transalpine territories. Upper
air instrumental measurements began much later with the development of
flight, to which the recent great upsurge in the study of meteorology is

9
10 From Observations to Simulations

chiefly due. It is interesting to notice, in any case, that the first sector that
was benefited by meteorology and began to work in synergy with it was
agriculture, once again in the Grand Duchy of Tuscany. In the eighteenth
century, with the Accademia dei Georgofili and the reformist project of
the grand duke Peter Leopold for the development of Tuscan farming,
modern agrometeorology began.
In spite of this brief outline of the beginning of instrumental
meteorology, we do not mean, here, to retrace the historical development
of this discipline1. It will be sufficient for us to give an accurate
description of what is available now, in terms of observations, for
determining the weather and climate in a certain region.

2.1 The "State" of the Weather

In physics, a system is regarded as defined at a certain instant if its

"state" at that instant is known. As a rule, this state is an entity that is not
observable and this fact does not make it possible to extract all the
information about the system by means of measurements. The
unobservability of the state depends on the indetermination inherent in
the measuring process and on the complexity of the system. The state of
a simple dynamic system of interacting particles (without any internal
structure) is known at a certain instant in time if the position and speed of
all the particles at that instant are known; in a simple thermodynamic
system, it is necessary to be acquainted with the quantities of pressure,
volume and temperature. In the former case it is a matter of a
microscopic description in terms of basic constituents, in the latter of a
macroscopic description in terms of mean quantities2. It is well known
that statistical mechanics is a bridge between these two descriptions and
makes it possible to interpret the macroscopic variables on the basis of a
microscopic description: for instance, the absolute temperature of a
portion of gas, which in the statistical mechanical definition is directly

'Some good texts of history of meteorology are available. For its beginning as an
independent discipline, the recent essay-novel by Hamblyn (2001) can be read.
2
For further details about dynamic and thermodynamic systems, it is advisable to consult
a basic physics text, such as that by Fermi (1937). A particularly interesting case is that of
the concept of state in a quantum system: see, for instance, Ghirardi (2003).
 Meteorological and Climatic Observations 11

proportional to the mean kinetic energy of the molecules, is interpreted

as a macroscopic measurement of this energy.
In the atmosphere, where an enormous number of molecules are
present even in a small portion of air, obviously we cannot apply a
dynamic description in microscopic terms. Observational meteorology
therefore produces a thermodynamic description (inevitably an
approximate one) of the physical state of the atmosphere above certain
observation sites, by means of measurements (in terms of instantaneous
values or of mean values over a period of time) of physical quantities
such as temperature, humidity, pressure and wind. This description is
completed by information relevant to the possible presence of
meteorological phenomena such as rain, fog or mist. If several sampling
points can be used on the territory at the same time, these measurements
give an idea of the state of the physical atmosphere system in the domain
represented by that region.
The concept of state will become crucial when we will start dealing
with the evolution of the physical atmosphere system. Then we will take
up that concept again and extend the discussion of its importance. At
present it is sufficient to state that, through our measurements and
discrete sampling, we are only able to achieve an approximate
determination of the state of the system under examination.

2.2 A Definition of Climate

What we have discussed up to now refers to the determination of the

weather. What about the climate? By climate we mean the set of physical
and meteorological conditions that characterise, on the average, a certain
area of the world over a certain period of time: at least 30 years, as
specified by the World Meteorological Organisation. To put it more
accurately, in order to achieve a description of a climate it is necessary to
know the mean values and variability of relevant quantities such as
temperature and precipitations. In this sense, meteorological observations
may be used for determining the climate; the only provision that is
required is their specific post-processing in order to highlight not only
the mean statistical value of the individual variables over various
12 From Observations to Simulations

decades, but also some elements that determine their rate of variability,
such as the scattering of their values around the mean (estimated, for
instance, by means of the standard deviation) and the frequency of the
occurrence of extreme events.
Climatologists, in actual fact, demand more than that, because they
also require information about periods in which collected and coded
instrumental measurements did not exist yet. As we will explain, this is
possible if we rely heavily on our theoretical knowledge of some
phenomena, in order to retrace long historical series of important
variables, such as temperature.

2.3 An Overview of Meteorological and Climatic Observations

The overall situation of the currently available meteorological and

climatic observations is shown diagrammatically in Figure 1. The
observations we have called direct consist of the subjective ones reported
in historical-period chronicles and documents, and of the instrumental
ones that began (as we have already stated) in the middle of the
seventeenth century, first in a sporadic, isolated way, and then became
increasingly integrated in observational networks. The former should be
regarded as impaired by a considerable degree of uncertainty, if nothing
else because of the subjective manner in which the weather is perceived
(the fact, reported in the chronicles of several periods, that people have
always been complaining that spring and autumn have disappeared,
should alert us). The latter, obviously, are more accurate and objective,
and become also more reliable as the observational networks spread all
over the world with more homogeneous instruments.
The observations that have been indicated as indirect in the figure
have been obtained by means of the reconstruction of climatic data
(called "proxy data") from long local historical series of some quantities
that can be interpreted by means of our theoretical knowledge of physical
or biophysical principles. This way the values of meteorological and
climatic variables such as temperature, even in the remote past, can be
inferred indirectly. Some examples of these historical series are those
relative to the annual rings of centuries-old trees, the characteristics of
 Meteorological and Climatic Observations 13

corals and the geologic core sampling of Antarctic ice. Further on we

will briefly return to this topic, in order to elucidate in a more detailed
manner what can be actually obtained from these data and with which
rate of uncertainty.

Subjective
Instrumental Proxy
(historical)

Fig. 1 Diagram of meteorological and climatic observations.

For the time being, it is interesting to focus on the most typically

meteorological data that nowadays are routinely collected worldwide:
our purpose now is to give the reader an overall view of the constant
monitoring to which the atmosphere is subjected. Various sources of
information are available to the meteorologists, ranging from automatic
or manned meteorological ground stations, to radiosondes that rise into
the atmosphere in order to probe the properties of one of its vertical
columns, and to satellites that orbit at various heights in order to observe
the Earth and its atmosphere by means of many types of different
instruments.
We can easily understand the need to have meteorological data
coming from an observational network that is extended over the territory
or even includes the whole world, if we consider that the weather in a
14 From Observations to Simulations

certain region (for instance one of the United States) is often determined
by the thermodynamic characteristics of the air masses that a few days
before were over another region, thousands of kilometres away (for
instance the Pacific Ocean), and have been carried to the region under
consideration by upper air currents. If there are no data about the
thermodynamic characteristics of the air mass over its place of origin,
and perhaps also of the state of the ground over which it is conveyed, it
becomes very difficult to produce a weather forecast even for a period of
only 24 or 48 hours. From a climatological point of view, the problems
due to a lack of data about some regions of the world are equally evident:
in case of conspicuous gaps in the observational network, climatic
changes can be mentioned only with a lower degree of statistical
confidence.
For about 50 years, the international-level efforts to harmonise a
global observational network have pertained chiefly to the World
Meteorological Organisation (WMO), an agency of the United Nations
whose headquarters are in Geneva and that is now co-ordinating a global
observational system, both for meteorological observations and for more
strictly climatic measurements, including chemical and atmospheric-
composition instrumental surveys.
As is always the case when discrete measurements are performed on
continuous processes, the spatial distribution of the observational
network and the timing of the sampling must be calibrated considering
the scale of the phenomena under examination. Since at a meteorological
level the prime goal is to obtain reliable so-called medium-range
forecasts (1 to 7 days), the WMO recommends a spatial resolution of 50
and 300 km respectively for ground level and upper air observations, and
a sampling interval not exceeding 3 and 12 hours, respectively. For
climatic observations, the required space-time resolution is obviously
much lower (looser network and less frequent sampling), because the
processes to be monitored are slower and more homogeneous in space.
An aspect whose importance should be stressed is that the
observations are carried out all over the world in a synchronous manner,
that is at the same hour: in all the planet, any meteorological observation
always refers to the solar hour of Greenwich (GMT: Greenwich Mean
Time). This way, for instance, every day at 12 GMT the thermodynamic
 Meteorological and Climatic Observations 15

characteristics of the atmospheric fluid are sampled in order to obtain

approximate information about its physical state.

2.4 Conventional Observations

We will now briefly consider the meteorological parameters that are

covered by this monitoring. The meteorological stations on the ground
and the ships employed for this service on the seas and oceans issue
observation bulletins at least every three hours (they are called SYNOPs
for the ground stations and SHIPs for the ships). These bulletins contain
the encoded information about the pressure, temperature and humidity of
the air, the direction and speed of the wind, the clouds (in terms of
extension of coverage, height of base and type), the visibility, the
quantity of rain that has fallen during a certain period of time, the
temperature of the surface (of the soil or water), the thickness of the layer
of snow, if any, etc. The bulletins are transmitted to a world-wide
telecommunication network; on the average, there are approximately
15,000 of them at a main hour such as midnight GMT. Plate 1 shows an
example of the global distribution of these observations: it has some
gaps, even considerable ones, on the oceans and on the African
continent. These gaps in the ground level observational network are only
partly bridged by data that come from automatic buoys installed in the
oceans (not shown in the plate).
The upper air observational network is obviously less close, partly as
a consequence of the greater homogeneity of the atmospheric fluid far
from the ground, which allows a lower-resolution monitoring, and partly
because of the high cost of the installation and use of the radiosonde
stations, which in many cases is unaffordable for developing countries.
In these stations, a sounding balloon full of helium gas is raised into the
atmosphere; it carries a box containing electronic instruments that
measure the pressure, temperature and humidity at various vertical
levels. On the basis of triangulation methods such as LORAN or GPS,
which make it possible to locate the ascending system with a 1-metre
precision, every 10 seconds of ascent of the balloon the distance covered
by it during that period is calculated. Supposing that the horizontal
16 From Observations to Simulations

movement of the ascending system follows that of the high-altitude

wind, an estimate of the average wind relevant to the layer that is crossed
(approximately 40 metres thick) is obtained; in actual fact, the speed of
the wind is slightly underestimated. At present, the average number of
soundings performed in the world at a main hour is approximately 700.
The radiosonde network is supplemented by the so-called PILOTs, in
which a small sounding balloon is used only for estimating the speed of
the wind at several altitudes, and by more modern instruments called
"wind profilers", which perform the same measurement by means of an
apparatus on the ground that emits electromagnetic pulses into the
atmosphere and analyses the return echoes and the relevant Doppler
shift. Moreover, observations and measurements at a single vertical level
are usually performed by intercontinental scheduled flights, particularly
on the North Atlantic routes.

2.5 Satellite Observations

All the measurements mentioned up to now (except the ones coming

from the wind profilers, which cannot be specifically discussed in this
book) are conventional measurements, i.e. they are performed with the
classical instruments that exist since the beginning of meteorology
(barometers, thermometers, hygrometers, anemometers and
pluviometers), though in some cases the sensors that are used currently
are different from the original ones. Now, however, there are also other
types of measurements that are called non-conventional: they are chiefly
those performed by the instruments aboard meteorological or Earth
observation satellites. We will briefly explain the additional help offered
by these measurements towards a more accurate monitoring of the
atmosphere.
In this book we cannot fully discuss the physical principles of the so-
called satellite "remote sensing"3. It will be sufficient to state that this

3
For a discussion that is more complete, but still fairly comprehensible to an uninitiated
reader, consult, for instance, Pease (1994).
 Meteorological and Climatic Observations 17

kind of survey analyses the behaviour of the electromagnetic radiation

that is emitted, absorbed and scattered through the atmospheric medium.
Aboard these satellites there are both active and passive instruments: the
former are real RADARs that emit an electromagnetic radiation and
analyse its return spectrum and Doppler shift, if any; the latter may be
regarded as simple digital cameras that are sensitive to various
wavelengths of the ingoing radiation, basically from ultraviolet to
microwaves, including the visible and infrared range. As a rule, the
radiation that reaches the satellite depends on the thermodynamic state of
the surface of the Earth and of the various layers of atmosphere that are
crossed (and on their composition as well). The problem of how to use
the analysis of the radiation received by the instruments aboard the
satellite in order to obtain values of parameters that define the state of the
ground or atmosphere should be tackled in each case on the basis of what
one wants to obtain. Here we will only briefly mention the fact that, as
we shall explain concisely further on, it is possible to estimate some
quantities such as the temperature of the various layers of the vertical
column under examination. The limit of these estimates consists in the
fact that they allow us to determine only mean temperatures of very thick
atmospheric layers; this limit is only partly technological, because it also
involves some limitations inherent in the physics of the problem of
remote sensing.
During the last few years, there has been a boom of initiatives for the
programming, designing, construction and launching of satellites for
meteorology and the observation of the Earth. Without entering into
details, we will now briefly discuss only those observations that currently
allow an extensive monitoring of the state of the atmosphere.
As regards their orbit around the Earth, satellites can be divided into
two great categories: polar ones and geostationary ones. Geostationary
satellites are put into orbit above the equator, at a height (approximately
36,000 km) that allows them to orbit the Earth along the equator at its
same angular speed; this way they always observe the same portion of
the planet. Polar satellites are brought to a lower height (approximately
800-900 km), and their orbit passes near the two poles. While their
orbital plane remains constant with respect to the fixed stars, the Earth
18 From Observations to Simulations

revolves under them4: this way they observe ever-different portions of

the planet. Their orbital period is usually 100 minutes or slightly more,
and during this period the Earth revolves by 25° or slightly more.
From the point of view of observation, geostationary satellites ensure
the constant monitoring of a certain area of the globe, but with a low
resolution (because of the considerable height at which the instruments
are). On the contrary, polar satellites achieve a monitoring that is more
discontinuous but has a higher resolution. As a result of these
characteristics, geostationary and polar satellites complement each other
in the global observational network. At present, for exclusively
meteorological observations there exist 5 geostationary satellites (such as
the European METEOSAT, known also to the general public) e 3 polar
satellites. A further difference between the observations performed by
the two different types of satellites consists in the fact that, whereas
geostationary satellites can collect data at predetermined hours (e.g. 12
GMT) simultaneously over a vast area (basically most of the hemisphere
over which they are), polar satellites are bound to the limited area they
can observe at a certain instant. In order to be able to use the data of the
polar satellites for an estimate of the state of the atmosphere at a certain
instant, it is necessary to find the way to obtain the values relevant to that
instant also on areas over which the satellite has passed a short time
before or afterwards5.
Besides supplying images that may be quite fascinating but are often
in themselves unusable for a quantitative analysis, geostationary and
polar satellites also issue routine data for meteorology: for this purpose
there even exist encoded observation messages that can be identified in
the global meteorological telecommunication system by the acronyms
SATOB and SATEM.
In the so-called "thermal infrared channels" the passive satellite
sensors are sensitive to the temperature of bodies (land, seas and oceans,
clouds); the SATOB, observation message from geostationary satellites,
supplies information about the temperature of the top of the clouds and

4
The same thing happens, for instance, in the experiment of Foucault's pendulum: in both
cases the reference system of the Earth reveals its character of non-inertial system.
5
This problem will be examined in Chapter 7, when the four-dimensional analysis of the
meteorological data will be discussed.
 Meteorological and Climatic Observations 19

the direction and speed of the wind, calculated through the movements of
the clouds, which are taken as a valid tracer of the flow of the
atmospheric fluid6. If the data of a radiosounding performed near the
region to which the SATOB refers are available, it is easy to obtain the
height of the clouds by means of the double-entry chart represented by
the vertical temperature curve supplied by the sounding: this way the
direction and speed of the wind at a certain height can be estimated7.
The SATEM is a message that comes from the polar satellites of the
NOAA series. These satellites carry the TOVS (Tiros Operational
Vertical Sounder), whose instruments measure the so-called radiance
(practically the intensity of the radiation emitted along a vertical path and
detected at the top of the atmosphere) in several ranges of the
electromagnetic spectrum. From the viewpoint of the physics of radiative
transmission in the atmosphere, when the thermal state of the vertical air
column is known, it is easy to obtain the radiance that falls on the
satellite. The inverse problem is trickier, but can be solved successfully
in many cases. When the radiance at single frequencies is known, it is
possible to determine the mean temperatures of several vertical layers of
air. It is necessary to point out that these layers are very thick. So the
TOVS actually achieves a vertical thermal sounding of an air column,
but this sounding is much more averaged than the one achieved by
classical radiosoundings: it has severe limitations when thermal
structures that have a small vertical scale must be measured. These data,
therefore, are very useful when there are no data coming from
conventional radiosoundings, but their vertical resolution may be
insufficient and they may actually be useless when more precise data are
available8.
The real great advantage of satellite observations consists in their
global coverage (shown in Plates 2 and 3), which obviously is not
affected by hostile logistic conditions (oceans, remote environments) or

6
In this case too, as in the estimate of wind obtained on the basis of the movement of the
radiosondes, the speed is usually somewhat underestimated.
'Besides this information, the SATOB also supplies the surface temperature, the
percentage of cloudiness of a certain region, and data relevant to the humidity and to the
ingoing and outgoing radiation.
8
The SATEM also supplies data relevant to the precipitable water content of the clouds.
20 From Observations to Simulations

financial problems (management of the ground network by developing

countries). From this point of view, meteorological satellites make it
possible to eliminate the previously mentioned gaps in the conventional
observational networks at ground level and, even more, in the upper air
ones. In particular, the surveys performed by geostationary satellites
cover, globally and with synchronous observations, all the area
approximately from latitude 60° south to 60° north (above and below
these latitudes, the data are not usable because of what is called a
parallax error, that is because those areas are observed too much "on the
skew" for the data to be valid). The polar satellites, moreover, cover
these high-latitude areas with an excellent frequency and supply more
details in other areas as well; there remains, however, the problem of the
non-synchronicity of their readings. Lastly, the number of observations
both of geostationary satellites and of polar ones is really enormous, and
incommensurably greater than that of conventional observations.
The satellites we have mentioned up to now were designed
exclusively for meteorological purposes. During the last few years,
however, some satellites with more extensive purposes of observation of
the Earth have been launched: some of the instruments they carry also
supply meteorological information. An example of data that are now
used routinely for meteorology is that of the surface wind fields on the
seas and oceans, obtained by means of instruments called scatterometers,
installed aboard the American/Japanese polar satellite ADEOS-II and on
the European ones ERS-2 and ENVISAT.
Without entering into details, we may briefly state that the
scatterometer is a radar that emits electromagnetic pulses and picks up
the return signal reflected by the surface of the sea. The energy of this
signal depends on the state of the sea: a rough or very rough sea reflects
more energy than a sea that is almost calm or only slightly choppy. A
series of devices, which we cannot describe in detail here, make it
possible for this instrument also to obtain the direction and speed of the
wind on the surface of a certain area. This information is crucially
important, both because it cannot be obtained so extensively with
conventional observations, and because the winds are measured near the
interface between the air and the sea, where the influence of the ocean on
the atmosphere appears.
 Meteorological and Climatic Observations 21

2.6 Meteorological or Climatic Observations?

As we have previously stated, the countless meteorological observations

that have been discussed in these pages also have an immediate climatic
value if they are subjected to a post-processing in order to highlight the
mean values, their variability and any possible trend over a period of at
least a few decades. For this purpose, for instance, meteorological
observation stations on the ground and radiosonde stations emit, with a
respectively daily and monthly frequency, specifically climatic bulletins
that summarise and highlight some quantities that are important for the
reconstruction of the climate of the site under examination. However, as
we shall see when we examine our theoretical knowledge of the Earth
system and of the factors that affect the changes in the parameters and
meteorological phenomena, it is necessary to consider the monitoring of
other elements, such as the radiative exchange between the Earth and the
outer space, the concentration of the constituents of the atmosphere, the
characteristics of the oceans (from a physical, chemical and also
biological point of view), and the characteristics of the Earth's
ecosystems (including lakes and rivers, ice, flora and fauna, and the
presence of human activities).
A more strictly climatic atmospheric monitoring is carried out with
the help of observers usually situated in remote areas and by means of
satellite observations: this monitoring obtains information relevant to the
stratospheric ozone and ultraviolet radiation, ozone at the surface, solar
radiation, concentration of certain gases called greenhouse gases, aerosol
and dust in suspension, and acid rain. In particular the satellite surveys,
combined with ground level observations, considerably help us to keep
the overall condition of the entire planet under control, also from the
viewpoint of the monitoring of the oceans and of the Earth's ecosystems.
The examples of satellite observations that have a climatic value are
countless. We should mention the altimetry activity, which began as
early as 1973 with the SKYLAB, and goes on, with increasingly
sophisticated instruments, up to the latest altimeter installed on the
ENVISAT: this has led to a world-wide monitoring of the mean level of
the sea that is characterised by an excellent degree of accuracy. Another
field in which the "eyes" of satellites are very useful is the evaluation of
22 From Observations to Simulations

the extension of the snow and sea ice cover. As regards ice, we should
mention the intensive monitoring undergone by Antarctica during the last
few years: this has led, among other things, to an immediate alert when
the previously-mentioned detaching of the Larsen B iceberg occurred,
and (obviously even more important) to the evaluation of the
disgregation process in that part of the Antarctic pack (see Plate 4, where
this process is shown as it has evolved from 1992 to 2002).
Inter alia, precisely the polar pack is the main object of the
investigation performed by the recent ICESAT of the NASA. On ground
areas that are not covered by water or ice, it is extremely important to
carry out an analysis of the vegetation cover, with particular reference to
the monitoring of the phenomena of drought and desertification, and of
the anthropogenic changes in the use of soil: all this contributes, in
particular, to the estimate of the so-called "albedo", i.e. the ratio of the
energy reflected in the space by the Earth, clouds and atmosphere to the
total incident energy coming from the sun.
Without going into the details of more strictly meteorological satellite
observations, which have also been mentioned previously, we should
point out that it is possible to obtain an estimate of the precipitations on
areas not covered by ground level pluviometers (particularly on the
oceans). The solar activity is also monitored with a greater accuracy than
that offered by ground level observations, since the latter are troubled by
the interposed atmosphere: it has been possible to accurately measure the
total solar irradiation, which was previously called the solar "constant"
and on the contrary was found to have evident fluctuations within an 11 -
year cycle. Moreover, the various instruments on the most recent
satellites (once again we should mention the European ENVISAT) make
it possible to monitor the ozone (both the stratospheric one and the one
closer to the ground), to estimate the emission spectra and concentration
of climatically relevant gases, and to perform a colourimetric analysis of
the seas and oceans. The latter is important for determining the thermal
state of the upper part of the seas and oceans and for obtaining
information about the oceanic part of the so-called carbon cycle, which
will be briefly discussed further on.
As the reader has probably understood, satellites make it possible to
perform observations of the Earth that are undoubtedly more global, and
 Meteorological and Climatic Observations 23

sometimes also more accurate than those carried out by ground level
stations or instruments in the atmosphere. The limit in the use of these
data from the angle of climatic researches consists in the fact that they
are quite recent: at best, as in the case of altimetric observations, we have
historical series that are about 30 years long, but in most cases the
surveys do not go further back than one or two decades. Sometimes we
have data relevant to a very limited number of years.

2.7 Proxy Data

We have explained that nowadays the global observational network

allows an accurate monitoring of the meteorological and climatic health
of our planet; but which instruments do we have for understanding how
the climate was when this network was not so extensive, or even when
there did not exist any instrumental measurements? Climatologists, we
have said, are particularly exacting on this point. We cannot blame them!
How can we judge the changes that are taking place during the last few
decades if we cannot compare them with those of other periods (when,
moreover, human activity was not able to disrupt the balance of nature)?
An attempt to remedy this lack of information has been made by
analysing long local historical series of quantities from which climatic
data can be reconstructed (called "proxy data"). We will now give a few
short examples of these reconstructions.
The data that are potentially most interesting and make it possible to
go far back in time (so far back that sometimes the term
"paleoclimatology" is used), are those that come from the deep oceanic
sediments and those from vertical soundings (called core samplings) in
the very thick layers of ice mostly present in Antarctica and Greenland.
In the former case, an examination is carried out on the remains of the
shells of small animals, such as foraminifers, that have accumulated on
the bottom (these shells are chiefly composed of calcium carbonate). In
the latter case, a direct examination is performed on the ice and on what
has been trapped in its interstices. In both cases a particular attention is
given to the analysis of the oxygen atoms present respectively in the
calcium carbonate and in the ice.
24 From Observations to Simulations

The reason for this is that in nature oxygen appears essentially in the
form of two stable isotopes9: 18 0 and 1 6 0. On the average, the ratio of the
concentration of 18 0 to that of 16 0 is approximately 1/500, but it changes
slightly with variations in environmental factors such as temperature10.
This leads to the conclusion that a careful analysis of the oxygen in the
calcium carbonate stored in the plankton shells and extracted from the
surrounding water may supply information about the temperature of that
water during the lifetime of the little animal under examination. Likewise
the analysis of ice should reveal the temperature of the snow that had
formed in the air above it during the period under consideration.
Obviously these analyses must be combined with a reliable dating
method.
In actual fact the situation is more complicated! Out of the controlled
conditions of a closed laboratory system, it is possible for the ratio
between the two oxygen isotopes to change simply because a certain
quantity of one of the two is removed from the system. In the seas and
oceans that supply oxygen for the storage of calcium carbonate, many
water molecules are removed because they evaporate, while others return
with precipitation. In particular, during the evaporation stage, on the
average more "light" oxygen atoms ( 16 0) than "heavy" ones (180) are
removed. Considering the matter exclusively at a global level, if all the
evaporated water returns to the sea with the precipitation (which
therefore contain more 16 0), the cycle is closed and the ratio between the
isotopes remains constant. But what happens during particularly cold
periods, when the precipitation (snow) thickens the layers of the polar
and mountain glaciers or forms new layers? In this case, the
concentration of 16 0 in the sea decreases and the ratio between the
isotopes changes.

9
The reader is reminded that a chemical element is determined by the number of protons
contained in its nucleus, and that, if this number remains equal, but the number of
neutrons changes, various isotopes of the same element are obtained. In the case under
consideration, 18 0 has two neutrons more than 16 0, i.e. it is heavier.
10
The theoretical change in this ratio is 0.2 parts per million with a 1-degree variation in
temperature: though it is very small, it can be revealed by the currently available
technology.
 Meteorological and Climatic Observations 25

In brief, the effects on the variations in the ratio between the isotopes
under examination are due to changes in temperature, to evaporation and
precipitation, and to the world-wide quantity of accumulated ice: the
only measure that is available to us, the ratio between the isotopes, is
therefore a function of these variables, which, inter alia, are not mutually
independent (for instance, a decrease in temperature usually corresponds
to an increase in the formation of ice). The usual approach — which
consists of unravelling the skein by reversing the problem in order to
infer water surface temperature values from isotope measurements
performed in this highly interacting system — requires a high degree of
theoretical knowledge of the system and of the interactions between the
various phenomena. Considerations of this sort are valid also for core
sampling.
By examining the samples of air trapped in the interstices of the ice,
core sampling also supplies a quantitative evaluation of the presence of
some gases in the atmosphere, in particular carbon dioxide (C0 2 ) and
methane (CH4), during the period under consideration. Coral skeletons,
which are composed of carbonate, supply information about the
temperature of tropical seas, when the same method that has been
adopted for deep-ocean sediments is used. The same technique is also
used to analyse organic sediments in lakes: it is worth pointing out that
this has recently made it possible to link changes in climate to historical
problems, such as the study of the disappearance of ancient
civilisations11.
A comparative analysis of the climates of the past can be performed
also by examining the so-called "fossil pollen" in the sediments
mentioned above. This makes it possible to find out which types of
plants were present during a certain period: from their northward or
southward shift in various geological sites (or from the presence of
pollen belonging to different types of plants in the same site), we can
approximately infer the type of climate of the various eras.
Lastly, plants contribute to the estimate of the climates of the past by
means of the analysis of the annual rings of the trunks. Obviously this
method can go back only a few centuries. However, since trees produce a

"About the disappearance of the Mayas, consult Hodell et al. (1995).

26 From Observations to Simulations

ring every year, at least in the temperate regions where there is a definite
growing season, these historical series have a high temporal resolution.
The growth of trees depends on the temperature and precipitation, so
when there are warm, rainy years the rings are wide, and when there are
cold, dry years they are narrow. If trees are analysed in areas with
distinct characteristics, it is possible to obtain information only about the
temperature or only about the rainfall: for instance, in typically warm and
dry low-latitude regions the critical parameter for the growth of trees is
the rainfall (so wide rings mean rainy years and narrow rings mean dry
years); in typically cold and humid higher-latitude areas, the critical
parameter is the temperature (so wide rings mean warmer years and
narrow rings mean colder years).
Since, as we have explained, all these reconstructions lead to more or
less accurate estimates of climatically important parameters (often
independent of each other and based on different sites), during the last
few years "multi-proxy" syntheses have been considered for the estimate
of the global or hemispheric surface temperature, achieving various high-
resolution reconstructions. This way, it has been possible not only to
obtain estimated values for the temperature over the last 1,000 years, but
also to define the error bars to be attributed to these values, which are
essential for evaluating the reliability of the estimates. For the northern
hemisphere during the last 1,000 years, for instance, the error can be
estimated at ± 0.5°C up to the year 1600 and at ± 0.2°C from 1600 to the
end of the nineteenth century: during the last century the error further
decreases, gradually. The estimates obtained by means of proxy data that
go further back in time are obviously impaired by a greater uncertainty,
and usually also by a lower temporal resolution.
Reconstructions obtained by means of proxy data, in short, are
indispensable for obtaining information about the remote past. Within the
limit of the error connected with an estimate — which is obviously
greater than that connected with direct instrumental measurements —
proxy-data reconstructions in any case can detect marked world-wide
changes in climate. From a conceptual point of view, it is important to
point out that, in order to obtain these reconstructions, evidence based on
empirical observation is not sufficient: it is necessary to rely on a
theoretical knowledge of the mechanisms of the interactions between the
 Meteorological and Climatic Observations 27

various processes and phenomena that take place in the system under
examination. We have seen an example of this in the analysis of marine
sediments, where it was found to be necessary to know the theoretical
balance between 16 0 and 18 0 with changes in temperature and the
dynamics of the water cycle (evaporation, precipitation, ice formation).

2.8 Is There Any Evidence that the Climate is Changing?

Once the entire database we have just outlined is available, it becomes

interesting to investigate whether these data, in themselves, give an
indication of the position of the climatic changes of the last decades in
relation to the variability of the more or less recent past. An analysis of
this type is performed with statistical methods that have been well-
known for a long time and have been applied extensively to climatic data
during the last few years. Since here we cannot go into a detailed
analysis of the statistical methods that are used, we will translate the
results that are obtained into an everyday language, basing our
explanation on the concept of probability and supposing that the reader is
acquainted with it. Moreover, while referring the reader, for a more
detailed analysis, to Chapter 2 of a recent report of the IPCC12, we will
only consider the results that are believed to be verified and are currently
regarded as a common property of the international scientific
community.
By post-processing the observation data for climatic investigation
purposes, we can obtain information about the mean values of some
parameters, their variability with time and the occurrence of extreme
events. On the basis of these parameters, it is possible also to reconstruct
the estimated mean states and trends of the atmospheric and oceanic
circulation. As for the variability of the climate, it can be studied on
various time scales: there exist an interannual variability, a ten-year or
hundred-year variability, and a variability on a paleoclimatic scale. The
first one is characterised by evident oscillation phenomena in the oceanic
circulation interacting with the atmospheric component (such as ENSO,
El Nino Southern Oscillation), or, more directly, in the atmospheric
12
Houghtonefa/. (2001).
28 From Observations to Simulations

circulation (such as the NAO, North Atlantic Oscillation); the third

variability is relevant to the transition from glacial eras to interglacial
periods and vice versa, during the last few million years. If we wish to
attempt to place the climatic changes of the last few decades inside or
outside a natural climatic variability, and to understand whether during
this period the influence of human activities has been decisive for these
changes, we may in actual fact temporarily disregard these two scales of
variability13 (for which the reader is referred to the existing literature14)
and concentrate on the ten-year or hundred-year variability.
While referring the reader, once again, to the previously-mentioned
IPCC report for a more complete overview, we should mention some of
the safest observational results: the statements that follow are regarded as
very probable (there is a less-than-10% possibility of their being
disproved) or practically certain (the possibility of error is less than 1%).
These highly reliable climatic indicators are:
an increase in the temperature of the surface of the sea by 0.4 to
0.8°C since the end of the nineteenth century;
an increase in the temperature of the air over the surface of the sea
by 0.4 to 0.7°C since the end of the nineteenth century;
an increase in the temperature of the air over the land surface by 0.4
to 0.8°C since the end of the nineteenth century;
a massive shrinking of mountain glaciers during the twentieth
century;
approximately two weeks less of ice formation in high-latitude lakes
and rivers since the end of the nineteenth century;
a 10% decrease in the spring snow cover in the northern hemisphere
since 1987, in comparison with the mean values of the period 1966-
1986;

As a matter of fact, while obviously in the paleoclimatological sphere the influence of

man is completely negligible, a debate is currently under way about the degree to which
this influence can affect the cyclic course of the ENSO and NAO and perhaps also
intensify them.
14
About interannual variability, and in particular about ENSO, see Philander (2004).
About variability on a paleoclimatic scale and the glacial eras, information can be found
in the excellent book by Alley (2002).
 Meteorological and Climatic Observations 29

a 5 to 10% increase in the precipitation at high and medium latitudes

since 1900, in many cases due to very intense events;
no significant global change in the frequency and intensity of
tropical cyclones.
Without presuming to give a comprehensive picture of the present
situation in relation to the last few decades or centuries, it is interesting
to consider the sole parameter of the temperature and to examine its
variations during these periods. In order to achieve an integrated
evaluation of the contributions to the global warming or cooling, we can
reconstruct a combined historical series of the temperature of the air near
the Earth's crust and of that of the surface of the sea, obtained by means
of direct instrumental measurements15. This way we can obtain graphic
representations like the one in Plate 5. It shows the so-called annual
thermal anomalies of the period 1860-2000 with reference to the mean
of the period 1961-1990 (taken as zero on the ordinate axis). The
downward red columns indicate colder years in comparison with the
mean of the period 1961-1990; the ones that move upwards towards
positive values indicate warmer years in comparison with the same
mean. In any case, the values of the departures from the mean of 1961—
1990 can be inferred by the length of the columns and by the relevant
values that can be evaluated by interpolating them on the ordinate scale.
Each column in the histogram is associated with an error bar that is
affected by the reliability of the measurements, particularly in relation to
the degree of global coverage of the observational network: as a rule,
therefore, these intervals gradually decrease from 1860 to now16. Lastly,

The reason for this choice will become fully intelligible in Chapter 4, when the
temperature variations on the surface of the land and sea will be discussed. At present it
will be sufficient to state that this historical series is an average of more evident yearly
fluctuations (on the land surface) and less marked variations (on the sea). In any case, the
sign of these fluctuations is the same in almost all the years that have been considered
(except for a small number of events, in which the absolute value of the variations is
extremely small).
16
We must point out, however, that in the error bars, together with the measurement
sensitivity and global coverage, an increasing uncertainty is considered: it is due to the
phenomenon of urbanisation and to the so-called "urban heat islands", which may affect
the measurements obtained in constantly-developing cities. Here the temperature may
increase year after year for local reasons, connected with the fact that the cities, made out
of materials that "entrap" heat (basically asphalt and cement), are becoming more
30 From Observations to Simulations

the black curve shown in Plate 5 basically represents an average trend of

the temperature that makes it possible to "filter" its yearly fluctuations,
since the latter, per se, might be connected only with the intrinsic
variability of the climatic system and therefore not be very valuable for
the purpose of identifying a trend in the global temperature.
The general trend revealed by Plate 5 is undoubtedly that of an
increase. More in detail, there are two plateaux of almost-constant values
from 1860 to 1910 and from 1945 to 1976. Two periods of increase in
temperature by approximately 0.15°C per decade appear from 1910 to
1945 and from 1976 to 2000: these increases are statistically significant,
because they amply exceed the indetermination limits allowed by the
error bars associated with the measurements. To this we may add (to
satisfy the reader's curiosity) that the data relevant to the years 2001 to
2004, not included in the plate, would appear in this histogram as the
warmest years after 1998. Obviously it is not possible to regard the data
relevant to individual years as statistically significant, but, for instance,
the fact that 2001 was a particularly warm year all over the world is
considered important, because, contrary to 1998, when ENSO was going
through the El Nino stage, with the tropical part of the Pacific Ocean
very warm17, in 2001 this part of the ocean was considerably colder.
Is the evidence supplied by the observations summarised in Plate 5
sufficient to allow us to state that, at least during the last 30 years, we
have been experiencing a period of global warming that is not due to the
natural variability of the climate? Though it is not the purpose of this
book to intervene polemically in the current debate about global warming
and its causes, it is worthwhile to give the reader a further contribution of
observations to be considered.
As we have previously explained, the analysis of proxy data, though
it relies heavily on our theoretical knowledge of the system under
examination, supplies information about the global temperature during
the last few centuries. Considering only the northern hemisphere (whose
sites are the sources of approximately 95% of these data), Plate 6 shows

extensive: this may hide the effects of a global warming or cooling. In Plate 5, all this
leads to the fact that, starting from the years of the Second World War, though the
observation network is constantly growing, the error bars tend to increase slightly.
17
For a more accurate analysis of this episode, see Philander (2004).
 Meteorological and Climatic Observations 31

a reconstruction of the temperature anomalies of the last millennium,

with reference to the usual thirty-year period 1961-1990, in association
with error bars. Notice that the trend of the temperature of the first 9
centuries is almost constant (or slightly decreasing) and that there is a
tendency to a marked increase during the twentieth century. Moreover,
all the data of the last few years exceed any previous error bar: this
means that these years have been the warmest in the millennium (in a
statistically significant manner) and that the values of the present
warming exceed the climatic fluctuations typical of the ten- or hundred-
year variability of the climate.
All this obviously does not lead to a univocal conclusion, but at least
it narrows down the range of possibilities to two hypotheses (which we
are not yet able to explore on the basis of the information contained in
this chapter): the present warming may be the result of a broader-scale
natural variability (that we have called paleoclimatic); or it may be the
evidence of a perturbation in the natural variability due to human
activities. In order to be able to lean to one of these two hypotheses, we
need a knowledge that greatly exceeds the limits of what we have
endeavoured to describe in this chapter; in particular, we will have to
break away from a paradigm that is purely based on observation, to
acquire a stock of sound theoretical knowledge about the system under
examination, and also to adopt a new approach to scientific research.
Before tackling this subject, like good detectives who must solve a
difficult case, in the next chapter we will look for further observation-
based clues that may lead us to a track for discovering the motive, i.e. the
causes that may have led to this global warming. It is obviously
premature to try to find out whether the serial killer may kill again in the
future, i.e. whether the global warming may go on in the course of this
century.
 Chapter 3

Naive Meteorology, Coincidences and

Correlations

The previous chapter helped the reader to understand which instruments

and methods of observation are required for achieving a (necessarily
approximate) definition of the thermodynamic state of the atmosphere
system. In doing this, it identified some significant variables in what we
perceive every day as the weather. These variables, analysed on a
broader time scale, make it possible to define the climate of the place
where we live. Finally, the determination of some time series of these
variables gives an idea of how the state of the system has evolved in the
more or less recent past.
Obviously all this information about the system under examination is
extremely important, indeed essential, because it forms the grounding
needed for any further analysis of the system. However, like all empirical
data pertaining to systems slightly more complex than those we studied
at school (which were based on the simple Galilean mechanics), these
data do not allow us to easily find, among the variables, evidence of the
existence of specific relationships that may be regarded as causal and
may therefore lead to an "explanation" of the "operation" of the system
under examination1.

lr
The reader is referred to the concise analysis of the concept of causality in the
Introduction, and reminded that, from a common-sense point of view, a phenomenon is
explained if its causes and their way of combining to produce it are known. Moreover,
the operation of a system is understood if the relationships between the variables that
define its state (both at that time and in their evolution in time) are known. In any case,
the concept of the understanding or intelligibility of a system will be discussed more
thoroughly in Chapter 4.

33
34 From Observations to Simulations

It is clear that the problem can be tackled from the angle of basic
sciences such as physics and chemistry. In this context, for instance, the
air that forms the atmosphere is nothing but a fluid, a mixture of several
gases and water2, plus other constituents such as aerosols (suspended
particles), to which it is possible to apply the theoretical knowledge
relevant to disciplines such as fluid dynamics and thermodynamics.
Obviously this is done, and this will be the subject matter of the next
chapter (together with the recognition that the system to be considered in
order to understand the dynamics of the weather and climate exceeds the
limits of the atmospheric fluid). For the time being, it will perhaps be
more enlightening to perform a different exercise.

3.1 Approaching an Analysis of the Data and of Common

Experience

Probably all of us have noticed that present-day society is often defined

as based on information and knowledge. As we have already remarked,
the flow of news and data that sweeps over us every day is enormous,
and the possibility of finding information intentionally and independently
is almost unlimited, thanks, above all, to Internet and the global network.
However, the use and interpretation of all this information depend on our
knowledge of the individual subjects.
In this context, it is reasonable to believe that having the know-how
required for a certain subject is more important than learning by rote and
"accumulating" uninterpreted data. Possessing know-how about a certain
system, as a matter of fact, means knowing the rules that determine its
behaviour and therefore mastering a paradigm for the explanation and
interpretation of the phenomena that may take place within it. The so-
called "experts", who are often interviewed on television or in the
newspapers, are people who possess a thorough knowledge of a single
area of interest; their domain of competence, because of the extreme
specialisation currently present in science, is usually quite limited.
In actual fact an interesting phenomenon is taking place at present,
and is extremely evident, for instance, in the labour market: rather than a

2
Water in its three aggregation states: solid, liquid and vapour.
 Naive Meteorology, Coincidences and Correlations 35

person's specific know-how in a certain area, what is considered

important is his or her mind-set, moulded by the experience of school
and work. A methodological approach to data analysis and problem
solution, and flexibility in applying this approach to unknown systems,
are regarded as more important than the acquisition of competence in a
specific area. This is particularly to the benefit of science graduates
(above all physicists), who sometimes are forced to change area of
interest, but manage to find jobs just the same, because of their general
qualities as problem solvers. When this occurs, they have to tackle
systems about which there generally is a good empirical knowledge but a
limited theoretical understanding3.
What is so special about the mentality of a young person who
"studies to become a scientist", though perhaps he/she will never
eventually work as a scientist? Undoubtedly it is this person's ability to
analyse a certain system, even if it was previously unknown to him/her,
in order to find a paradigm that explains it and may make it possible to
act upon it. There is a saying that scientists are big babies. It is true that
the behaviour of a scientist when faced with an unknown problem is very
similar to that of a small child when interacting with the external
environment during the course of his logic (inductive and deductive)
development. He looks for regularities and causal relationships, performs
tests if the system allows it, and works out his subsequent actions on the
basis of his previous experience. In particular he develops his own naive
physics, which is sometimes erroneous and, unless it is investigated more
thoroughly, may lead to typical common-sense errors4.
At this point of our discussion, we are really in the condition of a
child who is facing something unknown. Let us therefore attempt to

3
A concrete example is that of some physicists who have started dealing with economy or
finance, applying to this field some methods and techniques that pertain to theoretical
physics: nowadays this is called econophysics. Conferences are held and books are
written about this topic. At a more practical level, banks and companies that act in the
stockmarket are obviously interested in this type of analysis.
4
In international literature, naive physics is now covered by a great number of studies
(see, e.g., Hood (2004)). As regards the difficult relationship between modern physics
(particularly relativity and quantum mechanics) and common sense, chiefly due to the
different physical domains, it is advisable to consult classical writings like Einstein and
Infeld (1967), and the previously mentioned Ghirardi (2003).
36 From Observations to Simulations

analyse the system we are considering, about which we possess a

considerable quantity of empirical evidence, in order to catch some
relationships between the previously-mentioned variables, if possible in a
shrewd, discerning manner. The "big baby" who will accompany you in
this trip has already covered a similar ground, when, after having taken a
degree in physics with a thesis about the unified theories of gravity and
other interactions, he proceeded, initially only for occupational reasons,
to take an interest in meteorology and climate.
The exercise we are about to perform is not pointless. This is
demonstrated by the fact that some people who only have a naive
knowledge of physics — for instance old farmers and fishermen —
manage to forecast the short-term weather in their territory in an accurate
and sometimes surprising way, on the basis of observations they have
accumulated during their lives and of interpretations they have worked
out over the years. In comparison with these experts of local
meteorology, we are at a disadvantage, because we are no longer used to
examining the sky to find clues and forewarning signs. On the other
hand, we possess a quantity of meteorological and climatic data that are
unknown to them and a better knowledge of basic physics, at least as it is
acquired nowadays during the compulsory-schooling years5.
In what follows, we will consider only an extremely limited number
of observations: they come both from assessments that everybody can
make about phenomena connected to the weather, and from instrumental
data about the weather and climate now available to us. This will lead us
to point out some particular coincidences, which we will try, as far as
possible, also to interpret on the basis of common experience or of the
knowledge of basic physics that we can take for granted. Since, as we
have already explained, by definition the climate is the sum of the
meteorological events (in terms of averages of the individual variables,
scattering of values around the averages and number of extreme events),

5
Obviously, even if we are analysing a system that is "unknown" a priori, we are doing it
with the conceptual instruments that are available to everybody: it would be nonsensical
to pretend to be ignorant of certain basic physical properties that by now are common
knowledge. As we shall show further on, this does not avoid the risk of giving incorrect
interpretations of what takes place in the atmosphere.
 Naive Meteorology, Coincidences and Correlations 37

we will start by analysing some meteorological observations, then we

will examine some data more specifically relevant to the global climate.

3.2 A Naive Interpretation and Its Problems

It is common belief, and also an established scientific result, that the Sun
(together with water and the atmosphere) is what makes the existence of
life possible on the Earth. The Sun, in particular, supplies the radiant
energy required to preserve a temperature that is suitable for life. The
feeling of the sunbeams penetrating our body and warming it up is
undoubtedly one of the most appreciated pleasures of life: this is
demonstrated, for instance, by the presence of crowds of Italian and
foreign tourists on the Italian beaches.
This common experience of "body warming" by "absorption" of the
beams coming from the Sun obviously leads us to believe that the same
thing happens to the air masses. This opinion is confirmed by several
pieces of evidence coming from observation. For instance, if we consider
the same period of the year and temporarily disregard the direction from
which the wind comes, we notice that when the sky is clear and the
weather is sunny the air is usually warmer than on cloudy days. In
particular, during the morning the air becomes gradually warmer as the
Sun rises above the horizon.
The latter fact becomes comprehensible when we consider that, if the
quantity of energy emitted by the Sun during a certain lapse of time (e.g.
an hour) is regarded as constant, before it reaches the ground (or our
body), a certain part of this energy is probably absorbed by the
atmosphere through which it passes. At this point, when the Sun is low
on the horizon, the path of its beams in the atmosphere is much longer
than when it is near its zenith. This way, in the morning and evening the
amount of energy (heat) per time unit that reaches the air near the ground
is presumably much smaller than the amount that reaches the ground in
the central hours of the day, when the Sun is higher on the horizon.
The effect we have just described also has obvious seasonal
consequences: it is well known that, at the same hour of the day, during
the summer the Sun is higher on the horizon than it is in the intermediate
38 From Observations to Simulations

seasons and, even more, in the winter. We must also allow for the fact
that the duration of the day is longer in the summer. The result is that the
average temperatures in the summer are higher than in the spring and
autumn, and, even more, in the winter. So these facts, too, seem to
confirm our vision.
A further consequence of what we have said is that we can predict
that the average warming of the air (near the ground) will be greater in
the tropical areas than in the air at the intermediate latitudes and at the
poles (at the same altitude). And once again this prediction is fulfilled. If,
at this point, we consider a phenomenon we have neglected up to now,
the horizontal conveying of air from a region of the globe to another, in
particular between regions at different latitudes, we notice that, at
medium latitudes and in the northern hemisphere, the winds that come
from the south are warm and those that come from the north are cold.
This is consistent with what has been discussed just now about the
warming undergone by the air at several latitudes. Obviously, therefore,
the temperature of the air at ground level in a certain site is determined
by the combined effect of solar irradiation on the place under
examination and of the changes in temperature due to the arrival of air
whose thermal characteristics are different from those of the air
previously present in the site.
This way, on the basis of an analogy with the common experience of
the warming of our body by the Sun and of observations that can
certainly be shared by everybody, we have acquired a certain qualitative,
specific knowledge about the phenomenon of the warming of the air; we
have practised naive meteorology and now possess an explanatory
scheme of this phenomenon.
Within the sphere of physical sciences, it is commonly repeated that
an explanatory scheme (which, if formalised, is more pompously called
"a theory") is regarded as valid until an empirical piece of evidence that
cannot be explained by it is found6. So let us now widen the scope of our

6
In actual fact there may be other reasons for which a theory is dropped even if it is not in
contradiction with empirical data: an example is the excessive complication of its
theoretical system in comparison with that of a more simple, elegant theory. For many
years there has been a controversy about the influence (or absence of influence) of the
cultural, social and economical climate on the development of science and in particular of
 Naive Meteorology, Coincidences and Correlations 39

observation and find out whether other pieces of empirical evidence are
compatible with our idea of the air-warming mechanism.
Up to now we have always considered daytime situations in which
the Sun shines on the atmosphere. What happens during the night, when
the Sun is below the horizon? It is reasonable to suppose that when the
energy that comes from the Sun and causes an increase in temperature is
absent, the air dissipates the accumulated heat towards the outer space
and the ground. If you ask an inhabitant of a continental region (i.e. one
that is distant from the sea) how the winter nights are there7, you will
find that the temperature of the air near the ground is very low when the
sky is clear and starry, whereas it is higher if the sky is overcast.
Disregarding, to begin with, the influence of the stars on the temperature
values (a study we will gladly leave to the astrologers), it will be
necessary, in any case, to consider a new element in our explanatory
scheme. Clouds, which during the daytime screen out a part of the solar
radiation, thus justifying the decrease in daytime warming when the sky
is cloudy, during the night seem to produce the opposite effect. It is
arguable that they act as a screen against the loss of heat (in the form of
radiant energy that comes out of the air molecules) towards outer space,
but we are not in possession of information that allows us to understand
their interaction with this energy (maybe they reflect it back?).
Thus the study of night-time phenomenology shows how important it
is to know something more about the interactions between radiation and
matter, particularly clouds. We must remember, in any case, that the
interaction between the solar radiation and the air has not been studied in
this scheme. This does not mean a priori that our explanatory scheme is
incorrect; but it does confirm that the scheme needs to be re-examined in
the light of the new knowledge, in order also to explain the phenomenon
of the more or less marked night-time cooling.

explanatory theories. We do not mean to take part in this controversy here, and only wish
to state that in any case a conflict with a piece of empirical evidence that is not included
in the explanatory scheme leads to the rejection of that scheme, at least in its original
form (the scheme may sometimes survive with some changes, provided they are not made
ad hoc).
7
As a matter of fact, this phenomenon is not limited to the continental regions, but occurs
wherever there is a solid surface; a continental region is only a place where this
phenomenon is more evident, for reasons that will be explained in the next chapter.
40 From Observations to Simulations

Up to this point we have considered only observations and

experiences that are shared by everybody, including farmers and
fishermen. Though obviously the latter are more attentive observers of
atmospheric phenomena than we are, we have the advantage of being
able to exploit more "exotic" experiences and the currently available
instrumental data. For instance, people who frequently travel by plane
have certainly noticed that the information usually supplied by the pilot
of the plane includes the temperature of the external air at the flying
altitude: it is normally about -60°C. And the data coming from the
radiosondes, which have been discussed in the previous chapter, reveal
that the temperature of the air decreases with the increase in altitude8.
How can we fit these facts into our explanatory scheme?
From our interpretation of air warming by solar radiation there should
follow immediately, at least during the daytime, that there is a practically
uniform warming on a vertical column, or that the warming is actually
greater in the upper layers of the air, where a part of the radiation that
falls first is presumably absorbed, and where clouds (which are at a
lower level) cannot screen out the beams coming from the Sun. Perhaps,
as we did previously for clouds, we should include some new element in
the scheme under consideration: for instance, what is the influence of the
fact that, in the atmosphere, pressure decreases as altitude increases? Can
this piece of evidence (which appears clearly, once again, in the data
coming from the radiosoundings) explain the lower temperature in
conditions of lower pressure at high altitudes? In a closed room, too, the
air is layered and the pressure decreases with height, yet the warmer air
is near the ceiling and the colder one near the floor.
If these data — some of which are not available to fishermen — make
trouble in our explanatory scheme of atmospheric warming, it is natural
to suppose that the solution may come through our greater knowledge of
basic physics, in comparison with the fisherman's knowledge. Since the
time in which Isaac Newton "unified" the terrestrial and celestial
motions, for instance reducing the explanation of ballistic motions to a
particular case of his laws of universal gravitation, physicists have never

At least up to an area, called tropopause, where this tendency is inverted.

 Naive Meteorology, Coincidences and Correlations 41

stopped believing in the universality of the laws of nature9. This means,

for instance, that we believe that masses attract each other in the same
way on the Earth and in space. In particular, in a different ambit, if I
consider a certain fluid, for instance air, I believe that its behaviour is
described by the same laws both in the free atmosphere and in my house.
What may change, obviously, are the initial conditions of the system and
the so-called "boundary conditions" (for instance, the atmosphere does
not have a ceiling, while my room does).
We can believe, therefore, that my room is a good place for studying
the properties of air — the same air that forms the atmosphere. In
particular, the vertical thermal layering is due to an effect known as
Archimedes' principle, whereby a "bubble" of fluid undergoes an upward
thrust if its density is lower than that of the surrounding fluid (by
applying the law of state of gases, it is possible to demonstrate that for
air this is equivalent to the fact that the temperature of the "bubble" is
higher than that of the surrounding fluid). An example of the validity of
this principle is obtained by observing the ascent of the warm air coming
from a hot radiator. An accurate measurement would also reveal that the
warm air that rises from the radiator actually cools down slightly before
it reaches the ceiling; this effect cannot be explained simply by
molecular diffusion, i.e. by the fact that the air on the surface of the
"bubble" gets mixed with the surrounding air. With an extrapolation, we
might infer from this that if there were no ceiling, the ascending warm air
would go on cooling, thus creating a vertical thermal profile in which the
temperature decreases with height. This, however, can occur only if there
is a source of heat in the lower layers.
The above "domestic" example is a prototype of an explanatory
scheme that might enlighten us on the causes of the actual vertical
thermal structure of the atmosphere and on the dynamics of its formation.
But it requires a source that supplies heat from below, and this is not
consistent with the previously-developed paradigm of solar warming,
which explained "most" of the empirical evidence. We have thus reached
a stalemate. Once again it seems to be important not only to perform a

9
About the concept of law of nature, consult the excellent Barrow (1988).
42 From Observations to Simulations

more careful analysis of the motions of air fluid, but also to reconsider
the interaction between radiation and matter in the atmosphere.
For the time being, it is worthwhile to point out that the decrease in
temperature with the increase in height is not the only phenomenon that
is not explained by our scheme. I would like to mention the fact that
(contrary to the previous statement that at the medium latitudes of the
northern hemisphere the winds coming from the north are cold) in the
regions immediately south of the central and western Alps and Prealps
there occasionally is a northern or north-western wind that is warm (the
so-called fohri). This is particularly surprising if we consider that this
wind comes directly from the Alpine range, that is from altitudes where
the temperature is decidedly lower than in the Po Valley.

3.3 Coincidences and Correlations in Available Data

At this point, since we are not able to supply an explanatory scheme that
is self-consistent and reproduces all the above-mentioned phenomena
and properties of the atmosphere, we are at a cross-roads: either we delve
more deeply into a general study (if possible an instrumental one) of the
interactions between radiation and matter, or we consider other empirical
pieces of evidence that help us define the problem better and give us
some clues on the variables that are important in this study. In actual
fact, in the course of the history of science there have repeatedly been
alternations of periods of great theoretical syntheses and periods in which
there prevailed data that came from observation and had not been
completely understood from a theoretical viewpoint. In the latter case,
particularly in observation-based disciplines such as meteorology, the
search for evidence and correlations between data relevant to different
variables helps us to establish correct relationships between them and to
understand their interdependence, though it does not often lead to the
establishment of causal relationships.
In particular, the concept of "correlation" (or "cross-correlation" in
the case of the analysis of the so-called forewarning signs) is precisely
the one that farmers and fishermen use unconsciously when they point
out relationships between observations of different kinds, for instance
 Naive Meteorology, Coincidences and Correlations 43

between the presence of a cloud having a certain shape above a certain

mountain and the appearance of precipitation in the place under
consideration after a certain lapse of time. Beyond any sort of
interpretation of the phenomenon in terms of a more or less naive
meteorology, the discovery that, on the basis of previous experience, for
instance 8 times out of 10 the appearance of that cloud is followed within
3 hours by rain on that area is extremely important and useful. It means,
evidently, that there is a statistically significant link between these two
phenomena, though it is not possible to determine the causal relationship
between them: first of all it is likely that the clouds, that bring rain there,
are other ones; moreover the fact that rain does not come in 100% of the
cases means that some other phenomenon (that has evidently been
overlooked) also affects the appearance of rain in that place.
The concepts of correlation and cross-correlation are by now well
established and can be expressed by mathematical formulae. Without
going into the details, we will simply state that two variables are
positively correlated if algebraically low10 (high) values in one of them
correspond to low (high) values in the other one, at the same instant
(correlation) or at a subsequent instant (cross-correlation). The opposite
is true of an anticorrelation (or negative correlation). As a rule, if two
variables are positively correlated in a very marked manner, either there
actually exists a causal relationship between them, or they are both
affected by other factors (for instance the trend of a third variable) that
"force" them, tuning their behaviours and making them similar. In any
case, the analysis of correlations can allow us to discover, in the system
under examination, the existence of variables that had previously been
overlooked and may turn out to be important.
Confining ourselves to a very limited number of correlations between
different variables, we will now analyse some historical series of global
temperatures, as we had done at the end of the previous chapter. In
particular, we will analyse Figure 2. It shows the well-known anomalies
of the global temperatures11 of the last few decades in the so-called
troposphere, that is in the lower layers of the atmosphere, from the

Sometimes negative values.

n
This time with reference to the average of the period 1979-1990.
44 From Observations to Simulations

ground to an altitude of approximately 8 to 18 km (depending on the

latitude and season), and the corresponding temperature anomalies in the
low stratosphere, right above the troposphere.

a)

0.5

0.0

-0.5

1960 1970 1980 1990

Year

b)

- -

-uJV,/VvA^^/\/y -
- 1 y. -

- t Agung Z Z T } Satellites
t
El Chichon
T %
Pinatubo
"
Balloons

1960 1970 1980 1990 2000

Year

Fig. 2 Curves of global temperature in the troposphere (a) and of global

temperature in the low stratosphere (b) (source IPCC).

The first fact to be considered, though it cannot be immediately

interpreted here, is that while in the lower layers the temperature tends to
increase, in the upper ones it tends to decrease. Apart from this average
characteristic of the temperatures, it is interesting to concentrate on the
events indicated by the arrows and on the immediately subsequent
periods. These events are the three most important volcano eruptions of
the last few decades, which introduced a considerable amount of dust in
 Naive Meteorology, Coincidences and Correlations 45

the atmosphere. This dust became more or less uniformly scattered all
over the world, and settled in the low part of the stratosphere, where it
remained for many months. In correspondence with these events, we can
notice an increase in temperature where the dust has settled, and, though
less evidently, a subsequent decrease in the temperature of the lower
layers. So there seems to be a positive correlation between the amount of
dust (not shown) and the temperature in the stratosphere, and a negative
correlation between the latter (or the amount of dust) and the temperature
in the lower layers.
Beyond the possible interpretation of this evidence within the
previously created naive explanatory scheme12 (which, on the other hand,
was found to have some problems in the interpretation of other data), we
can state with certainty that the presence of something other than air and
clouds between the Sun and the low layers of the atmosphere disturbs the
values of a variable such as the temperature. Therefore, if we are to
understand the dynamics of the global temperature at a certain altitude,
we need to consider the natural events that produce dust.
This observational evidence has a particular significance, because it
shows, for the first time, that elements coming from outside, in this case
from the lithosphere, can enter into the atmosphere system. So not only
the system is not isolated (since its thermal characteristics are constantly
changed from the outside by the Sun), but it is also interfaced, at its
"boundaries", with other systems that can influence it.
Figure 2 and the discussion that follows its presentation are obviously
based on data obtained by means of instruments during the last few
decades. In the previous chapter, however, we explained that we also
have the so-called proxy data, which we discussed chiefly in relation to
the estimates of the global or hemispheric temperature trends during the
last 1,000 years. Here we can mention the fact that the core samplings of
ice in Antarctica and Greenland make it possible actually to go back to

12
It seems that dust absorbs radiant energy, so it causes an increase in the temperature of
the atmospheric layer in which it is present, and at the same times it acts as a screen for
the lower layers, preventing a part of the radiation from reaching them.
46 From Observations to Simulations

much remoter periods, though obviously with a lower temporal

resolution. We also touched upon the fact that, together with the
complicated estimate of the temperature, it is possible to obtain more
direct information about the composition of the atmosphere during the
periods under examination, by examining the air trapped in tiny
interstices in the ice. The examination of the long-term trends of the
temperature and of the concentration of the gases that form the
atmosphere has revealed some peculiarities, such as those shown in
Plate 7.
This diagram reports the estimates, relevant to the last 420,000 years,
of the values of the temperature and of the concentration of carbon
dioxide (C0 2 ) and methane (CH4), as they have been obtained by means
of the analysis of core samplings performed at Vostok, in the middle of
the Antarctic plateau. The first thing that one notices is the very marked
correlation between the three series of data, whose values rise or drop in
the course of time in a practically synchronous manner. Another
noticeable characteristic is a certain cyclic character of the curves of the
three variables.
In the diagram, obviously, it is possible to detect the glacial eras and
the interglacial periods, such as the one we have been experiencing
during the last few millennia (the last glacial peak, which coincided with
a thermal low, occurred approximately 20,000 years ago). The
alternation of these different periods and their approximate cyclicity
seems to suggest that there is an external forcing factor that determines
this pattern. As a matter of fact, at least for temperature, it has been
found that this external influence consists chiefly of the intrinsic types of
cyclicity of the Earth's orbit and of the precession of the Earth's rotation
axis in relation to the plane of the ecliptic ("Milankovic's theory"). Some
of these effects lead to a different quantity of incident solar radiation,
others to a different distribution of it on the surface of the Earth.
By projecting the diagram in Plate 7 into the future and applying
Milankovic's theory, we should find that now we are at the peak of an
interglacial period and are approaching a gradual cooling of the climate.
Recent calculations, however, demonstrated that an initial orbital forcing
factor needed at least to be amplified in order to account for the changes
 Naive Meteorology, Coincidences and Correlations 47

in temperature found in the past, particularly during the deglaciation

stages13.
Is it possible that carbon dioxide and methane play a "causal" role in
the amplification of the orbital effects that we have just mentioned? And,
by the way, does the increase in these gases during the deglaciation
stages precede, follow or accompany the increase in temperature?
Unfortunately the time resolution of the series obtained with core
sampling does not make it possible to answer the second question. As
regards the first question, in this book we have never discussed the
physical and chemical properties of these gases, so we would not know
how to reply. There remains the evidence of the marked positive
correlation between the three variables examined in Plate 7. This, in any
case, may be enough to induce us to consider these three variables as
constitutive elements, potentially important in the system under
consideration.
Now, taking a step backward to return to the discussion at the end of
the previous chapter, we may assert that Plate 7 also gives another
contribution: it probably helps us to exclude one of the two hypotheses
that had been presented. After having noticed, over the last hundred
years, an increment in temperature that exceeds the typical values of the
variability of the climate over decades or centuries, we wondered
whether this might be due to the effects of a natural variability on a
broader scale. The paleoclimatic analysis that has now been carried out
shows that we are already near the highest peak in temperature in the last
420,000 years and should be approaching a colder period. This leads us
to believe that the influence of a natural broad-time-scale variability
characterised by the regularity of astronomic motions (Milankovic's
theory) is not likely to present unpredictable, opposite-trend aspects and
thus to drive the current global warming. Finally, there still remains to be
assessed the hypothesis of a disruption of the natural variability due to
human activities.
In relation to this matter, and also to the observational evidence
obtained by studies such as the core sampling of Vostok, which
establishes a close correlation between the trend of the temperature and

'See, for instance, Petit et al. (1999) and references therein.

48 From Observations to Simulations

that of C0 2 and CH4, we must consider that the concentration of a gas

such as C0 2 in the atmosphere depends on complicated mechanisms that
involve, for instance, photosynthesis processes and leaf "respiration",
and also some mechanisms of oceanic storage. In these cases, the
emission or absorption of carbon dioxide is performed chiefly through an
"exchange" with other systems that are at the interface of the
atmosphere, e.g. the ocean. A result of the fact that C0 2 is emitted also in
all combustion processes is that, among these systems, we are led to give
full consideration to the biosphere and, in particular, to all the human
activities within it that involve combustion (which in most cases
originates from the use of fossil fuel in the production of energy for
industrial processes, transportation and heating). The deforestation
performed by man in order to allow a different use of the soil can also
have a certain influence, because it eliminates trees, which "absorb"
carbon dioxide.
In this context it is interesting to consider Figure 3, which shows,
together with the concentrations, during the last 1,000 years, of C0 2 and
CH4, also those of nitrous oxide (N 2 0) and sulphate-containing dust. All
the diagrams reveal a considerable increase in the concentrations during
the last two centuries. In particular, if we return to Plate 7, a comparison
between the data relative to carbon dioxide and methane during the last
420,000 years and the present data shows clearly that the latter are
considerably higher than any other value detected in the past. This
obviously seems to suggest that the variations are due to a human
"perturbation" of the composition of the atmosphere.
This "impression" is confirmed by some data relevant to
anthropogenic emissions14, which have enormously increased starting
from the period of the industrial revolution. For the time being, however,
this does not allow us to make any inference about the origin of the
warming that has occurred during recent periods, in terms of global
temperature.

Emissions are something different from concentrations. Whereas an estimate of the

former supplies information about the amount of a certain gas or material introduced in
the atmosphere, e.g. as a result of human activities, the measurement of the latter gives us
an idea of what remains dispersed in the atmosphere after all possible interactions with
other systems at the interface of the atmosphere or within it.
 Naive Meteorology, Coincidences and Correlations

Indicators of the human influence on the atmosphere

during the Industrial Era

(a) Global atmospheric concentrations of three well mixed

greenhouse gases

1.5
360
_ 340 1.0
I 320
0.5
OJ 300
° 280
260

1750 Methane ^0.5

0.4
S- 1500
Q. 0.3
" 1250 0.2

" 1000 0.1

0.0
750

0.15
310 . Nitrous oxide
i
J
0.10

290 _
0.05

270
"• • • ••.,.•• r ^ • 0.0

250 -
1400 1600 2000
Year

(b) Sulphate aerosols deposited in Greenland ice

200 - Sulphur

100
r
'^zP^C^ J/
4"
1800
Year

Fig. 3 Concentrations of carbon dioxide, methane,

nitrous oxide and sulphate-containing dust during the last
1,000 years, from a combination of proxy data of the past
and instrumental data of the last few decades (source
IPCC).
50 From Observations to Simulations

Apart from the high correlation found in Plate 7 (and also in the last
few decades) between the trend of the global temperature and the
concentration of carbon dioxide and methane (a subject that will be taken
up again in the next chapter), we do not have, for instance, any cognitive
or explicative element that allows us to link the increase in these gases to
the increase in temperature.

3.4 Let Us Take Stock of the Situation

In this chapter, on the basis of the data available to us, we covered a path
that is typical of a scientific investigation. We saw that these data do not
immediately shed light on the operation of the atmosphere system, and
were compelled to organise them in an explanatory scheme, albeit a
naive-type one. This scheme, however, was found to be in conflict with
other observational evidences. At this point, we acted like the experts of
local meteorology (farmers and fishermen), who look for what we call
correlations and cross-correlations, in order to identify other relevant
elements in the system under examination. Now that we have a more
complete picture of the system, we need a substantial progress in our
theoretical understanding of it. This is what we will endeavour to achieve
in the next chapter.
 Chapter 4

The Theoretical Framework: Knowledge of

Single Phenomena and Complexity of the
Earth System

In the previous chapter we mentioned the fact that any type of

information becomes usable only if it is fitted into an explanatory and
interpretative paradigm. The empirical evidence supplied by
meteorological or climatic observations, in particular, is not sufficient, by
itself, to reveal the connections among the variables of the system, or to
allow us to understand physical phenomena and processes (and even less
to predict their future evolution).

4.1 How Can We Read the "Great Book of Nature"?

Adopting a more radical approach, we might argue that perhaps there is

no such thing as a pure and simple empirical observation, except in the
first attempts to tackle an unknown system. We must consider, as a
matter of fact, that as soon as we begin to be acquainted with the
physical characteristics of a natural system, within the immense quantity
of empirical evidences that Mother Nature makes available to us, the
only ones to which we pay attention are the ones that we consider
important for understanding the system under examination: only these
become real observations1. In this perspective, the observations we
1
This is based on a principle that is quite familiar to neurophysiologists, and that here we
might call "economy principle". Our brain, because of its finite structure, cannot file all
the sensory information it receives (some of it is actually filtered a priori by the
sensitivity thresholds of our senses). Likewise, because of the finite speed with which we
process information, in the solution of a problem we do not consider all the possible

51
52 From Observations to Simulations

perform voluntarily (e.g. instrumental measurements) are guided by the

expectation to increase our theoretical understanding; therefore they are
selected within a definite explicative scheme (that they may help to
corroborate or disprove). Finally, as we have explained in Chapter 2, the
so-called indirect observations that result in proxy data are essentially
based on our theoretical knowledge of certain physical phenomena2.
All this, therefore, indissolubly links observations to the interpretative
scheme of the "operation" of the system under consideration. Here we
mean to discuss precisely the current theoretical scheme relative to
meteorological and climatic phenomena. First, however, we should pose
the problem of what this capability of ours to "understand" a certain
system consists of. How do we read "the great book of nature"? Maybe,
if what we have stated in Note 1 of this chapter is true, natural evolution
and, subsequently, cultural evolution have selected an "economical" way
of summing up in an interpretative framework all the information we
possess about a certain system.
In actual fact it does seem that this is the way things went. While
carefully refraining from entering the field of the study of human
learning and reasoning (which we leave to cognitive scientists), we will
confine ourselves to a brief analysis of how a scientist manages to
interpret and sum up the observational data relative to the sector under
consideration.
As some authors have pointed out3, nowadays our scientific image of
nature is based on computable functions, and, in actual fact, the
intelligibility of the world is due to the fact that we consider it
algorithmically compressible. What does all this mean? On the one hand

hypotheses. Apparently this is how the human brain works. If, for instance, we analyse
the recent wins (and final draw) of the Russian chess player Vladimir Kramnik in his
matches with the super-computer Deep Fritz, whose processing and storing capacity is
enormously greater than the human one (for instance, it can analyse 3.5 million moves
per second), this suggests that natural evolution has been selecting an option that is
economical and still successful.
2
Proxy data are only an extreme, very evident case. In actual fact, the operation of any
sensor is based on theoretical knowledge: for instance, in a mercury thermometer, the
temperature estimate is achieved by exploiting a phenomenon that is theoretically
understood, the expansion of mercury due to changes in temperature.
3
See, for instance, Barrow (1988).
 The Theoretical Framework 53

it shows that (as had already been indicated very incisively by Galileo
Galilei) science uses the mathematical language in order to decipher the
great book of nature: in particular here we cite computable functions, in
which a variable depends on other variables (and possibly on time), from
which its value can be calculated, analytically or by means of numerical
methods using a computer. On the other hand, the comprehensibihty of
the physical world is strongly linked to the hypothesis that, by means of
mathematics, it is possible to describe the relationships between the
variables present in it, in a condensed and economical way. This way, for
instance, if we are in possession of observational evidence relative to the
values of a pair of variables in a certain system (at the same instant in
time or at different instants), once we have found a "physical law" that
binds them mathematically, we have acquired a condensed, economical
way of describing their relationship. Among other things, this allows us
to "predict" the value of one of them on the basis of the value of the
other, even in the case of values that have never appeared before in the
system.
The example mentioned above is an example of algorithmic
compressibility. The theory of any physical process or phenomenon
(with its diagnostic or evolutionary equations)4 supplies precisely this
algorithmic compression. Vice versa, in a world where algorithmic
compression is not possible, phenomena appear to be random, and the
characteristics of that world can be described only by a long list of
sequences of observed phenomena5.
Obviously, it is worthwhile to remind the reader that the concept of
causality is always essential when the physical laws of a system that is
evolving in the course of time are being sought: it turns out actually to be
the pivot of the scientific explanation of any phenomenon or process. We
must, however, mention the fact that there exist some balance laws (also

diagnostic equations interlink two or more variables at the same instant in time;
evolutionary ones express the evolution, in the course of time, of a variable, which may
depend not only on its value at a certain instant in the past and on time, but also on other
variables.
5
We will not advance any further in this analysis. Obviously it would be interesting to
discuss whether the compressibility of the physical world depends on an intrinsic
characteristic of that world (whatever this may imply) or is an "impression" of ours due
to the use of mathematics, understood as a creation of the human mind.
54 From Observations to Simulations

called coexistence laws), when two or more variables are connected by a

mathematical relation at the same instant in time. These laws cannot be
clearly traced back to cause-effect relationships.
An example of this is Boyle's law for perfect gases, which asserts
that, in a transformation in which the temperature does not change, the
product of the multiplication of the pressure by the volume of the gas (at
the same instant in time) remains constant. Moreover, both in diagnostic
and in evolutionary laws, there are relations that connect the values of
the various variables in a correlative (or statistical) manner: their
contribution to the algorithmic compression of a system is evident,
though the fact that we know only these relationships (which are valid
only from a statistical point of view, i.e. impaired by an uncertainty that
sometimes can be quantified) is often attributed to incompleteness in the
description of the system under examination, to the extent that they are
unlikely to be regarded as capable of supplying an explanation of its
behaviour, at least in classical physics.
Here we cannot proceed any further in this analysis, which would
lead us to delve into the concept of determinism. These themes will be
taken up again in Chapter 7, within a narrower sphere, but having at our
disposal an example of a concrete, realistic study. At present it will be
expedient for us to dwell on another "epistemological leaning" of
scientists in the study of the physical world, a leaning to which weather
and climate scientists are not immune.

4.2 The Local Approach to the Study of a System

As we have already stated in the previous chapter, physicists generally

count on the fact that the laws of nature are universal.
This, besides being probably the only way to make it possible to
explain phenomena that take place in areas that cannot be investigated
directly, naturally leads us to prefer local analyses and the instrumental
investigation of small portions of matter. As we have already indicated,
the air that is present in my room is the same that is present in the
atmosphere: if I count on the fact that the regularities I observe and the
laws I discover by analysing its behaviour here and now are the same
 The Theoretical Framework 55

ones that govern its dynamics in the free atmosphere, I may just as well
study its properties in a place where I can use all the available analysing
techniques. In this perspective, the "local" approach to the study of a
physical system naturally encourages the tendency to carry out a
thorough analysis of the basic constituents of a system, for instance air
molecules, which can be easily "handled" in a room or laboratory6.
It is known that this approach to an in-depth study of the constituents
of a system has increasingly led us to improve the ideal microscope with
which scientists examine nature. In particular, the physics of the
twentieth century was characterised by a tendency to study increasingly
microscopic entities, up to the so-called "elementary particles" and their
constituents, which at present seem to be leptons and quarks. All this
research is accompanied by the more or less implicit idea that to know
the structure and behaviour of the individual basic constituents of a
system (whether they be elementary particles, atoms, molecules or air
masses with uniform internal characteristics) means to obtain the key to
understanding the "large-scale" behaviour of the system, derived from a
"composition" of the elementary processes thus revealed.
In this "rush toward the microscopic", several explanatory theories
were gradually developed, often using different languages, i.e. sectors of
mathematics, each of which was suitable for describing a single "level of
microscopicness". In the current scientific practice, each scientist adopts
the theory (with the relevant formalism) that is suitable for explaining the
phenomena at the level under his/her examination. However, many
people believe that the phenomena pertaining to the most macroscopic
levels may be regarded as a composition of more elementary phenomena
pertaining to a more microscopic level. If this is true, it will be possible
to reconstruct the results and predictions of the theories pertaining to the
macroscopic levels by means of the application of a formalism typical of

6
In actual fact, we should pose the problem of what came first: we cannot absolutely take
for granted either that the increasingly local or "microscopic" analysis of the constituents
of a system was originated by the wave of belief in the universality of the laws of nature,
or that, vice versa, this belief actually developed in order to "export" the increasingly
detailed knowledge that was being acquired within the microscopic sphere into larger-
scale spheres that could not be touched directly by investigation. We will gladly skirt this
matter, whose discussion would lead us too far in another direction, deflecting us
considerably from the subject of this book.
56 From Observations to Simulations

the lower level, that is by means of theories whose constituents (for

instance, the variables considered) are more elementary. This has
actually been done in the past. It does not mean, obviously, that the new
theories that have thus been developed are easier to use (for instance for
prediction activities), or that they replace the old ones in working
practice. However, the possibility to explain the macroscopic operation
of a system in terms of microscopic variables, regarded as more
elementary, is considered a crucial step in the understanding of a system.
The belief that each level may actually be understood through the
lower-level constituents and their physical interaction, together with the
tendency to develop theories that make this operation possible, is called
"reductionism". Extrapolating this tendency and carrying it to extremes,
a thorough reductionist believes in the existence of a theory that explains
all the phenomenology present in the physical world, and even more in
the natural world (including life), starting from its elementary
microscopic constituents. The final goal of the reductionist programme is
to achieve this result. Now cosmology promotes the "historicisation" of
this vision, and there is talk about a "Theory of Everything" that explains
the dynamics of the entire known universe, once the physics and, if
possible, the initial conditions of the first instants of the universe are
known7.
An example of the partial realisation of this project, which we may
consider paradigmatic, is the reduction of thermodynamics to statistical
mechanics, where the macroscopic concepts of pressure, density and
temperature are translated into microscopic terms and reduced to the
average statistical properties of the gas molecules considered in each
case8. Macroscopic thermodynamic phenomena are therefore described

7
See, e.g., Barrow (1991) and Hawking (2002). Obviously in this vision it is necessary
also to evaluate (or upvalue) the roles of determinism and of the so-called anthropic
principle (about this, see Barrow and Tipler (1988)).
8
In order to prevent a misunderstanding, it is expedient to specify that here the adjective
"statistical" does not have the meaning previously adopted to define the "uncertain"
character in all the cases of a correlation between two variables. Now, in statistical
mechanics, the macroscopic variables (e.g. temperature) are determined univocally once
the distribution of the corresponding microscopic variables (e.g. the speed of the
individual molecules) has been fixed. The use of statistics is necessary only because of
the very great number of these variables, and has nothing to do with the uncertainty about
 The Theoretical Framework 57

through the properties and interactions of the individual gas molecules.

Among other things, this microscopic explanatory vision has made it
possible to understand phenomena that were obscure before or were even
ignored by thermodynamic texts, for instance the so-called "second-order
phase transitions" in crystals9.
Despite the undeniable success of the reductionist vision in the
history of science, I believe that a few words of caution about the project
of extreme reductionism are required here. As I have already pointed out,
sometimes these microscopic-level theories cannot be applied practically
within a macroscopic system, for instance because of the lack of data
about the state conditions of the individual microscopic elements, as in
the case of the air molecules in the atmosphere. Though on the one hand
this is only an impossibility of practical application and not a principle
(so it does not impair the cognitive value of the microscopic theory), on
the other hand this does not allow a validation of the theory in spheres
where there may be phenomena characterised by a particular,
macroscopically visible dynamics or self-organisation. We may wonder,
for instance, whether a cyclone is (or is not) an emerging entity that can
be explained by the molecular dynamics of air and water particles.
Maybe it is, but we are not able to verify it.
As we will explain, meteorology took a different direction from the
outset, and soon (with a surmise that will be discussed further on in this
chapter) actually decided to disregard molecular dynamics in the
explanation of some medium- and large-scale meteorological
phenomena.
At present, at least in a field of investigation such as that of the
atmosphere and Earth system, scientists are basically aware of the
importance of studying the microscopic structure and interactions of the
elements that form the system. Many people, however, believe that there

the determination of the corresponding macroscopic variable. This uncertainty, which

does actually exist, is due only to the fact that we do not have a precise knowledge of the
individual microscopic variables.
9
The reader is reminded that classical thermodynamics can be applied also to fluids and
solids, besides gases. However, some "anomalous" behaviours, such as the second-order
phase transitions, can be explained only in terms of statistical mechanics: an example of
these transitions is that of a metal to the superconducting state when there is no magnetic
field.
58 From Observations to Simulations

exists a level of complexity and of exchanges among interacting systems

that cannot be described in microscopic terms, either as a result of a
purely practical impossibility, or as a repercussion of processes and
phenomena that actually originate at a macroscopic level. We will return
to this topic further on.
For the time being, while remaining within the sphere of reductionist
tradition and local instrumental analysis, we will closely examine the
elements that may be regarded as the basic constituents of the
atmosphere, i.e. the air molecules, and their interaction with the radiation
that crosses the atmosphere. Among other things, in the previous chapter
various observational clues had suggested the potential importance of
carrying out an investigation of this type in order to obtain an
explanatory picture that is more realistic and (we hope) does not
contradict the observations.

4.3 The Interaction between Radiation and Matter and the

Greenhouse Effect

In order to explain the origin of the electromagnetic radiation that crosses

the atmosphere, we must return to certain fundamental discoveries of
physics in the years before and after 1900, a crucial year for the
understanding of some basic radiative properties of bodies.
If we consider the emission of visible light, it is evident that its
primary source is the Sun, with a flow on the ground that is more or less
variable during the daytime hours, depending on the height of the Sun
above the horizon, the presence of clouds, fog, etc. The objects that emit
light are not many on the Earth: lightning, lava flows, fire, and obviously
man-made light sources. None of these emissions, however, is as
continuous and intense as the light coming from the Sun.
Sunlight has been studied thoroughly, starting from the second half of
the seventeenth century: that was the period of Newton's experiments
with prisms, which revealed the "hidden" colours of white light for the
first time. From the viewpoint of theoretical understanding, the
disputation between the authors of two antagonistic theories, Newton
with his corpuscular theory and Huygens with his wave theory of light, is
 The Theoretical Framework 59

fairly well known . At the time of their first formulation, both these
theories were able to explain the observational data. Then, as time
elapsed, evidence in favour of the wave theory accumulated, particularly
with the discovery of the phenomena of diffraction and interference, up
to the success (which seemed final) achieved with the formulation of
Maxwell's equations, by means of which visible light was described
within the broader category of electromagnetic waves. This, in particular,
led scientists to predict the existence of "non-visible light" such as radio
waves, which differ from what is commonly understood as light only
because of their longer wavelength11. Soon these radio waves were
reproduced and revealed instrumentally by Hertz, while Rontgen
revealed the waves that are now called X-rays, whose wavelength is
shorter than that of visible light. In the meantime, moreover, scientists
understood that an individual body that emits visible light usually also
emits different-wavelength radiation at the same time.
Without going into technical details, or into historical ones, we can
mention the fact that it had already been known for some time that any
body in thermal equilibrium emits a radiation whose spectrum (which
describes the quantity of energy emitted when the wavelength changes)
is determined univocally by the temperature of that body: it is the so-
called "blackbody heat radiation". 1900 was a crucial year, because that
was when Max Planck determined the correct law of the blackbody
spectrum for all wavelengths12. A direct consequence of Planck's
approach is that the wavelength range in which the maximum quantity of
emitted energy falls depends on the temperature of the body. The

Newton's theory described the nature of light, its transmission and its interaction with
matter in terms of light corpuscles, whereas Huygens's theory contended that the nature
of light was that of a wave, with all the phenomena connected to this.
n
An electromagnetic wave is characterised by its wavelength, k, which can be
represented as the distance between two peaks, or points of maximum value, in the wave
train that propagates in space. When an electromagnetic wave falls on a body, it is easier
to reveal the so-called wave frequency, v, i.e. the number of times in which, within a time
unit, the wave oscillates in the place where it is detected: this value is inversely
proportional to the wavelength (v = c / X, where c is the speed of light in vacuum).
12
This, among other things, began a unique thirty-year period in the history of physics,
from the viewpoint of theoretical syntheses, leading, in particular, to the formulation of
quantum mechanics and the solution of the conundrum relative to the dualism between
waves and corpuscles in the interactions between radiation and matter.
60 From Observations to Simulations

immediate application to a body having a surface temperature of

approximately 6,000°C (the Sun) and to another one having an average
surface temperature of about 15°C (the Earth) shows that the Sun sends
us radiation above all in the visible-light range, while the Earth emits
radiation above all in the infrared range (non-visible). A graphic
representation of the blackbody spectra of the Sun and Earth is presented
in Figure 4.
The discovery of a source of radiation other than the Sun is obviously
important in view of an analysis of the interactions between the air
molecules and the radiation present in the atmosphere, all the more if we
consider that the Earth's radiation has characteristics that are different
from those of the Sun's radiation, in terms of wavelengths.
Without going into details, let us examine which is the typical effect
of the impact of radiation having different wavelengths on the gas
molecules that form the atmosphere. The phenomenon on which we will
focus our attention is that of the absorption of radiation by the various air
molecules, because this "entrapment" of energy in the air contributes to
determine its warming13.
In nature, according to the correct precept of quantum mechanics, the
absorption of radiation having a certain frequency (or, similarly, a certain
wavelength) by a molecule takes place only if the value of this
frequency, multiplied by Planck's constant, h, is equal to the difference
in energy between two orbital levels of that molecule. Since each
molecule possesses its own characteristic energy levels, it can be inferred
that each molecule can absorb only radiations having specific
wavelengths.
In semi-classical terms, we can imagine that the effect of the
absorption of radiation by a molecule is like the increase in kinetics in
the rotational and vibrational motions of that molecule: see Figure 5.

Obviously the absorption of radiation is not the only way in which air can be warmed
up. Recalling what we have learnt at school about heat transmission, besides irradiation
we can cite conduction and convection.
 The Theoretical Framework 61

80000000
E
3. (a)
E 70000000

i> 60000000
un

50000000
E
pert

40000000

30000000

i 20000000

! 10000000

Wavelength (pm)

20 30 40

Wavelength (um)

Figure 4. Blackbody spectra of the Sun (a) and Earth (b). The maximum
irradiance of the Sun is in the visible range (approximately 0.5 \an); that of
the Earth is in the infrared range (approximately 10 (xm). Notice also that the
total irradiated power always increases when the temperature of the body
62 From Observations to Simulations

Figure 5. Mechanical model of a molecule: energy absorption creates a

greater frequency in the oscillation of the distance between the atoms
(along the "springs") and in the oscillations of the angle between the
atoms (both indicated by the arrows).

As a rule, ultraviolet solar radiation (UV), conveying a great amount

of energy, leads to a molecular dissociation or ionisation14. However, this
radiation (particularly its most energetic components, UVB and UVC) is
almost totally absorbed by ozone (which has the right energetic levels to
do this) in the higher atmosphere, warming up the latter. Moreover, the
ozone molecule takes part in a rather complex cycle of reactions with
molecular oxygen and atomic oxygen15.
The visible radiation carries less energy than the ultraviolet one, and,
in the atmosphere, the typical phenomenon it produces is its diffusion
due to impacts with the air molecules16; only a very small part of it is
absorbed. There is, moreover, a region of the electromagnetic spectrum
where the solar radiation and the terrestrial one overlap: it is the so-called
"near-infrared" region. Here most of the radiation is absorbed by water
vapour and carbon dioxide. Finally, in the region where there is the peak

In the former case, the molecules are split into atoms; in the latter, an electron is
expelled by the atomic or molecular structure, leaving a positively charged ion.
15
This balance may be disrupted by chlorinated compounds, which cause the well-known
phenomenon of the destruction of the stratospheric ozone layer and the resulting ozone
hole detected during the last few years. Here we cannot discuss these topics in detail.
16
This, as Rayleigh was the first to demonstrate, is why the sky is blue.
 The Theoretical Framework 63

of the Earth's emission (called thermal infrared), approximately 80% of

the emitted radiation comes out of the atmosphere and reaches the outer
space. This range is characterised by some absorption bands due to C0 2
and other minor gases such as methane and nitrous oxide, which —
together with the action due to the presence of water vapour and of
clouds made up of liquid water and/or ice — lead at present to the
absorption of the remaining 20% of the Earth's infrared radiation. The
thermal infrared wavelength range is particularly sensitive to a change in
the concentration of these gases: in particular, detailed theoretical
calculations (forerun by the Swedish chemist Svante A. Arrhenius as far
back as 189617) show that their increase leads to an increase in the
percentage of absorbed radiation and in the warming of the troposphere
(the lowest part of the atmosphere, where there is the greatest quantity of
these gases).
At this point, we must describe the basic mechanism of what is
currently known as "greenhouse effect". This phrase originated when
climatology began to be popularised, and indicates an analogy between
what takes place in a greenhouse and what takes place in the atmosphere.
The essential "transparency" of the air to solar radiation (chiefly the
visible one) and the absorption of a part of the terrestrial radiation
(almost entirely the infrared one) by some gases, called greenhouse
gases, is compared to the transparency of the glass of a greenhouse,
which allows the solar radiation to come in, but, being opaque to infrared
radiation, prevents the long-wave terrestrial radiation from coming out.
As we all know, the temperature of the air in a greenhouse is
considerably higher than that of the outdoor air: it is fairly common to
conclude that this is due to the differential effect of the glass on radiation
having different wavelengths. In actual fact, this warming effect does
exist, but it accounts only for a small percentage of the warming of the
greenhouse, which is mostly due to the mechanical effect of insulation
from the external environment, achieved by the glass. So the analogy
between the atmosphere and a greenhouse is faulty; however, since the

See Arrhenius (1896).

64 From Observations to Simulations

phrase "greenhouse effect" has by now become a part of the media

jargon, it still survives.
A fundamental aspect that must be explicitly stressed when dealing
with the greenhouse effect is the fact that this effect is essentially due to
the presence of some minor gases in the atmosphere and to water vapour:
oxygen and nitrogen, which together form approximately 99% of the
"dry" atmospheric mixture (i.e. not considering water vapour) do not
absorb the infrared radiation. So a constant monitoring of these low-
concentration gases is important for keeping track of possible variations
in the radiative warming of the troposphere. We must also point out that,
contrary to the generally negative connotation of the phrase "greenhouse
effect" in common parlance, in actual fact the processes connected to this
effect are essential for life on the Earth as it is now: it has been
calculated that, if the greenhouse effect were not present, the average
temperature on the planet would be lower than the present one by
approximately 33 °C. In such an extreme condition, the number of known
forms of life that might survive is very small.
What has been understood up to now, though obviously not
concluding our story, nonetheless is extremely important in view of a
global energy balance of the Earth system (formed by land, oceans, ice,
biosphere and atmosphere, each of which can be regarded as a subsystem
that interacts with the others).
From a physical point of view, as a matter of fact, the Earth is a
system upon which there falls an external flow of energy in terms of
solar radiation; it responds to this external "forcing factor" by dissipating
the energy it has thus received and re-emitting it towards space, again in
terms of radiation. Everything that comes into the Earth system and
comes out of it takes the form of radiation: the energy balance between
the incoming radiation and the outgoing one reveals whether the system
is in temperature balance or whether there is an imbalance that promotes
its warming or cooling. If, for instance, in the future there should be an
increase in the concentration of the greenhouse gases, the latter might
entrap a greater quantity of terrestrial long-wave radiation with respect to
the present quantity, and it would not be possible for this radiation to
 The Theoretical Framework 65

reach the outer space; so this would determine a net increase in energy
(heat) within the Earth system18.

4.4 Greenhouse Gases, Clouds and Aerosols

Considering, for the time being, only the atmosphere system, we can
perform a concise survey of the gases that play the most important role in
the greenhouse effect, i.e. carbon dioxide, methane, nitrous oxide, ozone
in the lower layers, and water vapour. The concentration of these
molecules in the atmosphere sometimes depends on reactions between
the various compounds that are present, as in the case of the complex
photochemical reactions that involve ozone near the ground; but more
often it results from their emission or absorption by other subsystems of
the Earth system that have exchanges with the atmosphere. A typical
case is that of water vapour, which is essentially produced by the
evaporation of water surfaces (oceans, seas, lakes, rivers) and by the
phenomenon of evaporation and transpiration by plants, and is removed
by condensation and precipitation phenomena that, as an ultimate
consequence, bring it back to the Earth's surface.
A case that is undoubtedly more complex is that of carbon dioxide,
whose emission and absorption involve several processes that
interconnect various subsystems. The emission of C0 2 is caused, for
instance, by volcano eruptions, the breathing of animals, fire in the
forests, and the combustion of oil and other types of fossil fuel. C0 2 is
absorbed chiefly because it is stored in the oceans and "consumed" in the
photosynthesis that characterises the vegetable world. The concentration
of carbon dioxide in the atmosphere depends on the balance of this
complex cycle, called carbon cycle. The reader should notice, in
particular, that the situation becomes the more complicated to describe
the more the cycle is affected by human activities: on the one hand, man
contributes to the emission of C0 2 through fossil-fuel combustion and
arson activities, on the other hand he also contributes to eliminate some
"absorbers" of carbon dioxide, for instance through the deforestation

Obviously the Earth system would also respond to the increase in energy with other
effects, the so-called "feedbacks", which will be discussed further on.
66 From Observations to Simulations

carried out in order to achieve a different use of the soil. Obviously all
these anthropic actions tend to promote an increase of C0 2 in the
atmosphere.
In order to close the discussion of the global radiation balance, we
must now introduce two further elements that are present in the
atmosphere and have not been considered yet: clouds and aerosols.
Clouds are made up of water or ice, and result from the condensation of
water vapour and its possible subsequent freezing. For a detailed
description of the interaction between clouds and the radiation present in
the atmosphere, we should allow for the specific characteristics of the
clouds (composition, shape, height). Simplifying the matter as much as
possible, we can assert that the basic effect of clouds on solar radiation in
the visible range is that of reflection: they act as a screen. As regards
their interaction with the infrared radiation, particularly the terrestrial-
origin one, they are able to entrap it between themselves and the ground.
The predominance of one of these two effects depends on the specific
situation. Simple theoretical calculations show that during the daytime
the screening of the solar radiation prevails, and, in comparison with a
clear-sky situation, clouds help make the air under them cooler, whereas
during the night, when the solar radiation does not fall on the clouds, the
only effect that is present is the one on the radiation coming from the
ground, which helps to keep the lower layers warmer than they are
during starry nights. All this is consistent with the common experience
(already mentioned in the previous chapter, taking as an example winter
nights in a continental area) of cold nights with a clear sky and warmer
nights with a cloudy sky.
As regards aerosols, we must explain first of all that this term
indicates a considerable variety of "impurities" present in the
atmosphere, ranging from several types of suspended dust particles
(produced, for instance, by volcano eruptions or anthropogenic pollutant
emissions) to salt crystals coming from the seas and oceans, and to
pollen, spores and bacteria. As we have seen in the case of the volcano
eruptions discussed in Chapter 2, these types of aerosol often reach a
very high altitude in the atmosphere, where the interaction with the
radiation refers particularly to the radiation coming from the Sun.
Therefore the direct consequences, for instance, of the presence of a dust
 The Theoretical Framework 67

"cloud" in suspension basically depend on the optical properties of its

constituents; these properties in turn depend on the material of which the
constituents are formed and on their size (roughly called "particle
diameter"). As a rule these "clouds" reflect the visible radiation and act
as a screen for the underlying atmospheric layers, helping them to cool
down; moreover, particularly if the particles are rich in carbonaceous
materials, they lead to a warming of the atmospheric layer in which they
are present. This analysis accounts for the observational results presented
in Chapter 3, Figure 2.
As we have already suggested, the case of dust in suspension, and
more generally that of aerosols, highlight, once again, the exchanges that
take place between the atmosphere and the other subsystems of the Earth
system, as in the case of the material extracted from the lithosphere and
introduced in the atmosphere. And there is also another aspect. This topic
allows us to start explaining that any element present in the atmosphere
interacts in a complex way with other elements. For instance, the
interaction of aerosols with radiation that we have just described is only
one of the consequences that these particles have on the radiation
balance, the so-called "direct effect". As a matter of fact, there exist at
least other two effects on this balance, called "indirect effects". In order
to explain them, we must mention the fact that, in the atmosphere,
impurities also play a particular role, that of catalysing the condensation
of water vapour into droplets of liquid water, facilitating this change of
state, which otherwise would have more difficulty in taking place, even
in cases of saturation or supersaturation (i.e. when the relative humidity
exceeds 100%). To put it in meteorological parlance, aerosols supply
condensation nuclei.
The two indirect effects on the radiation balance mentioned above are
connected to the consequences of this role. In particular, an increase in
the condensation nuclei leads to the presence of a greater quantity of
water droplets in the clouds; the diameter of these droplets is smaller,
and this changes the optical properties of the clouds, causing them to
reflect (i.e. screen) the sunlight more: the net result is a greater
contribution to the cooling of the layers below. The other indirect effect
is due to the fact that, since the droplets are more numerous and smaller
in diameter, it is less easy for them to reach the critical radius and weight
68 From Observations to Simulations

above which they fall to the Earth as rain, so the clouds persist as such in
the atmosphere and screen out the sunlight for a longer time. In this case
too, the cooling of the lower layers is promoted. If the
aerosols/condensation nuclei decrease, obviously, the effects are
opposite.
This way, the theoretical understanding of the interactions between
radiation and the air, which can all be investigated locally and according
to the rules of a reductionist vision, led to the discovery of an effect, the
greenhouse effect, which plays an essential role in the warming of the
lower-layer air. As the reader undoubtedly remembers, radiosonde
observations have revealed that the atmospheric layers near the ground
are usually the warmest ones; the naive explicative scheme developed in
the previous chapter was not able to explain this. Can the greenhouse
effect account for this observational situation? In this sense, the fact that
understanding the interaction between radiation and other elements
present in the atmosphere (specifically clouds and aerosols) can explain
some observed phenomena leads us to hope that the correct paradigm for
the interpretation of the temperature variations in the atmosphere has
been found in the physics of irradiation. In order to answer this question,
we will have to broaden the scope of our investigation once again, so as
to include other elements and physical processes present in the complex
Earth system.

4.5 Approaching a Complete Scheme of Warming from

the Bottom

As we have already mentioned (see Note 13 in this chapter), up to now

we have discussed only one of the possible modes of heat transmission in
the atmosphere: irradiation. Though from a climatic point of view (i.e.
averaging values over long periods) heat transmission by irradiation is
fundamental, for instance in energy balance calculations, it is reasonable
to wonder whether on a shorter time scale it can account for the warming
process in the low atmosphere. Here heat is redistributed between the
various subsystems of the Earth system, in particular between land,
 The Theoretical Framework 69

ocean and atmosphere, on time scales that range from the day-night cycle
to the seasons.
We have explained that the troposphere is basically transparent to
solar radiation, apart from a few absorption bands in the near-infrared
range. Moreover, solar radiation in the ultraviolet range helps warm the
stratosphere , which, however, is basically separated from the
troposphere below it (where we live and where all meteorological and
climatic phenomena take place). This way we can assert that most of the
solar radiation reaches the ground, where it is partly reflected and partly
absorbed. On the basis of what we have explained up to now, the reader
has understood that the warming of the troposphere takes place from the
bottom, by means of the phenomenon of the absorption of long-wave
terrestrial radiation by greenhouse gases. However, this statement, which
is basically true, reveals only a part of the actual situation: it is true that
the warming of the atmosphere takes place from the bottom, but this is
not due only to the absorption of radiation.
We can spot a first clue of this by looking at Figure 4 again: we can
see here that bodies at different temperatures emit different quantities of
energy, which increase as the temperature of the body rises. Now, it is a
common experience that the ground warms up during the day and cools
down during the night: so we can expect an increase in irradiation during
the day and a decrease during the night. In actual fact, though allowing
for the fact that in certain situations the approximation to a blackbody
may be rough and making the required corrections, a quick calculation
allows us to assert that this effect explains only a part of the
meteorologically relevant temperature variations in the lower layers of
the atmosphere. The absorption of radiation, therefore, is not sufficient to
account for short-term temperature variations, for instance during the
day-night cycle.
In this chapter we expatiated (some readers may say far too much) on
the local analysis of the interaction between radiation and the
constituents of the atmosphere; we were led by a reductionist vision,
which in turn is inspired by a belief in the universality of the laws of

Where, in actual fact, the air is extremely rarefied, and the concept of temperature loses
much of its usual meaning.
70 From Observations to Simulations

nature. The same inspiring concept had led us, in the previous chapter, to
devise a "domestic" example of the warming of the air in a room by a hot
radiator. We had noticed, in particular, that the bubbles of warm air that
formed near the radiator moved up as an effect of buoyancy, because
they were warmer and less dense (i.e. lighter) than the surrounding air.
Soon this created a stratification, in which there was warmer, lighter air
near the ceiling and colder, heavier air near the floor. We did not stop at
this first aspect, and proceeded to point out that the vertically rising air
bubbles cooled down slightly before they reached the ceiling. This led us
to conclude that, if the ceiling were removed, the ascending air would
cool down further, eventually creating a vertical thermal profile in which
the temperature decreases with height, exactly as it does in the
troposphere. This simple "domestic" example could therefore "mimic"
what happens in the atmosphere, provided there is a source of heat near
the ground: in that stage, obviously, the cooling of ascending air was
purely an observational assessment, in a situation where pressure
decreased with height, without an evident theoretical explanation.
At this point, however, we know that there does exist a source of heat
in the low layers of the atmosphere: it is the ground itself, when, as a
result of the absorption of a part of the energy conveyed by solar
radiation, it reaches temperatures higher than those of the air that is in
contact with it. So we can advisedly proceed with the parallelism
between what happens in my room and what happens in the troposphere,
now utilising some theoretical knowledge that comes from the basic
physics of heat transmission and from general thermodynamics.
Though here we have always discussed only irradiation, the best-
known mode of heat transmission is undoubtedly conduction, whereby
two bodies that touch each other tend to reach the same temperature.
Among other things, this is the principle on which the measurement of
the temperature of a body by means of a traditional thermometer is
based, as when, for instance, fever is measured in a sick person. These
measurements are considered reliable if the mass of the thermometer is
small and the difference between the initial temperature of the body and
that of the thermometer is not too great, that is if the presence of the
thermometer does not appreciably affect the temperature of the body. In
the case of air that touches a radiator or the ground, it is precisely the air
 The Theoretical Framework 71

that (over a brief period) takes on the role of an ideal thermometer,

because, after a certain time, it ends up by having the same temperature
as that of the body it touches, without causing any appreciable change in
the temperature of the latter.
This way, for instance, during the daytime the air that directly touches
the surface is warmed by conduction. But how are the layers immediately
above it warmed? Recalling the mechanical-statistical concept of the
temperature of a gas, connected to the average kinetic energy of its
molecules, we can surmise that there is a phenomenon (which does
actually exist), molecular diffusion, that causes the warmer air molecules
(which are faster) to diffuse among the colder-air ones (which are
slower) and to transfer a part of their kinetic energy to the latter by
means of molecular impacts, with a net warming effect. This does take
place, but the molecular "mixing" is rather slow, while in the atmosphere
(and in my room) the air is actually mixed up more quickly. The process
that seems most likely to be the one that allows this greater rapidity is the
previously-described vertical ascent of warm-air bubbles.
This process, which has already been discussed from an observational
point of view in an indoor environment, is called "convection". As we
have already stated, it can be regarded as a manifestation of the
Archimedes' principle, therefore it takes place until the ascending air
meets an obstacle (the ceiling of my room), or until, at a certain point in
its ascent, it ends up by being surrounded by warmer air. In the latter
case, the ascending bubble is now colder and heavier than the
surrounding air, and starts undergoing a decided downward thrust, which
first slows down its ascent, then inverts its movement, forcing it to
descend. We must now explain what determines the previously-
mentioned cooling of the air within the ascending bubble.
Here a gravitational effect (to which, in actual fact, also other less-
important thermal effects are added) comes into play: it determines the
amassment of the air that touches the ground and its increasing
rarefaction with the increase in height above sea level. This results in the
fact that, if we mark off some small volumes of air at certain heights, a
pressure that gradually decreases with height is exerted on their
imaginary walls. So if, for some reason, one of these small volumes of
air rises in the atmosphere, the fact that it reaches a lower-pressure area
72 From Observations to Simulations

causes it to tend to expand (because of the greater internal pressure) .

This phenomenon is described well by the first law of thermodynamics.
If we suppose that the "bubble" that rises in the atmosphere is
isolated from the surrounding air (like the one that is physically
delimited by the balloon that contains the gas), the theoretical
determination of the cooling of the bubble on the basis of its expansion is
correct, with an excellent approximation. In thermodynamic terms,
regarding the bubble as insulated means supposing that it does not have
heat exchanges with the outside (in this case the process is called
"adiabatic"), and in particular not considering molecular diffusion in the
surrounding air. The fact that this approximation (called "adiabatic")
leads to excellent results in the calculation of cooling is due to the
different time scales that characterise convection and molecular
diffusion: the latter is very slow and may be disregarded for up to 12 or
24 hours in the study of certain atmospheric phenomena.
So at this point we have achieved a theoretical justification of the
vertical thermal profile characterised by a decrease in temperature with
height that has been observed in the atmosphere and that had remained
an enigma at the end of the previous chapter. It is due to the combination
of the three modes of heat transmission, which co-determine the
warming of the atmosphere from the bottom (obviously to all this we
must add the horizontal redistribution of heat via oceanic and
atmospheric currents, which has not been discussed yet). We can also
mention the fact that the validity of the adiabatic hypothesis suggests that
sometimes a microscopic description of the system (e.g. in terms of
molecular diffusion) is not necessary. Moreover, once we have allowed
also for the vertical movements of air masses in which water vapour is
present and can condense, the adiabatic hypothesis can be extended to
state changes, and considerably helps us to understand the dynamics of
clouds. Finally, it also leads us to correctly explain other phenomena that
had not been understood previously, such as the warm, dry winds that

" This fact is clearly displayed in the sad moment in which a balloon escapes from the
hand of a child. Rising in the atmosphere, the balloon becomes increasingly large, until
the tension on its surface exceeds a critical threshold, causing the balloon to burst.
 The Theoretical Framework 73

sometimes descend from the Alps to the Po Valley . Breezes, too, are
due to the fact that particular convection cells set in on the territory.

4.6 Nature of the Ground and Air Warming

The theoretical scheme we have just outlined describes a type of

atmosphere warming that takes place essentially from the bottom. In this
context, it is clear that the nature of the ground that absorbs the solar
radiation and helps to warm up the air above it in the ways we have
described may be important for a more detailed analysis of this warming.
After all, is it not a common experience that winters are milder in
regions of the world that are in close contact with the sea than in
continental areas that are quite distant from it? Consider the winters in
Rome, Italy, and in St. Louis, Missouri, USA: both these cities are not
very high above sea level, and St. Louis is even further south than Rome
(so, on the average, it receives a greater amount of incident solar
radiation). Yet the winters in these two cities could not be more different:
they are mild and with comparatively high temperatures in Rome, and
harsh, often snowy and with temperatures that may be extremely low in
St. Louis. What determines this difference? Why should the presence or
absence of a large water surface nearby have such an influence on the
climate?
Once again we must consider the interaction between the incoming
solar radiation and the elements that form the Earth system, in particular
the Earth's surface. Speaking of this interaction, we stated previously
that the surface absorbs a part of the visible radiation coming from the
Sun and re-emits radiation essentially in the "thermal infrared" band
towards the atmosphere and outer space. However, we never made any
distinction between the various types of surfaces: it is now time to do so.
Conduction, convection, and, broadly speaking, the irradiation
emitted by a surface depend almost exclusively on the temperature of the
surface, independently of its physical nature. However, surface

21
Here we cannot discuss the physics of fohn or of other local meteorological phenomena:
for this, the reader may consult a text of aeronautic meteorology, because this type of
information is usually quite important for aeronautics.
74 From Observations to Simulations

temperature and its changes with time are strongly affected by the mode
of absorption of solar radiation by surfaces having different
characteristics. More specifically, different materials are characterised by
different responses to irradiation: for some, a small quantity of energy is
enough to raise their temperature by 1°C, while others require much
more energy.
In physics there exists a quantity that measures this "thermal inertia",
i.e. that determines the resistance of each body to warming. This quantity
is called heat capacity. The energy supplied to a body being equal, the
temperature of this body rises more if its heat capacity is low and rises
less if its heat capacity is high22. As a consequence, in a warming/cooling
cycle, a low heat capacity allows the body to respond more quickly to
these external forcing factors, with a change in temperature that follows
the cycle rather closely, though perhaps with a slight delay. Vice versa,
in bodies that have a high heat capacity a great quantity of incident or
outgoing energy is needed to change their temperature appreciably;
therefore, if the warming/cooling cycle is too quick, their temperature
changes only slightly during the times that are characteristic of this cycle.
We will now apply this discussion to the topic of the temperature
variations on the Earth's surface, temporarily disregarding the role of ice
(which will be considered further on). To begin with, we should remind
the reader that the Earth's surface is formed of approximately two thirds
of sea and ocean masses, and one third of continental crustal plates. If we
disregard the differences that undoubtedly exist between the various
types of ground on land areas (deserts, woods, forests, tundras,
savannahs, etc.), we can assert that the heat capacity of the seas and

In mathematical terms, we can put it this way: the energy (heat) supplied being equal,
there is an inverse proportionality between the increase in temperature and the heat
capacity of the body under consideration. This can be expressed by a very simple
formula: AT = Q I C, where Q is the quantity of heat absorbed by the body, AT is the
difference in temperature between the end and the beginning of its warming, and C is the
heat capacity of the body. To make a comparison between the thermal effect of warming
on different materials, it is expedient to refer to the same mass unit: in this case it is
possible to check the thermal inertia of a specific material in comparison with that of
another material, by measuring the so-called "specific heat", c, of the material, which is
defined by the formula c = C I m, where m is the mass of the body under consideration
and C, obviously, its heat capacity.
 The Theoretical Framework 75

oceans is much higher than that of the crustal surface above sea level.
This does not depend on their greater extension on the Earth: using the
terms introduced in Note 22, above, we can assert that it is precisely the
specific heat of the water of which they are formed that is always higher
than that of the land.
Thus, on the basis of these theoretical considerations, it is easy to
predict that the land warms up quickly during the daytime, following the
solar cycle, and cools down with equal rapidity (through a loss of radiant
energy) during the evening and night. Meanwhile, the temperature of the
sea should not be much affected by the day-night cycle, and should
undergo only a slight change on this time scale. Instrumental
observations, and also common experience, confirm that this does
happen. Who has never plunged into the sea during a hot summer day in
order to cool down? Who has not enjoyed the typically juvenile
experience of a midnight swim? In these cases we have undoubtedly
appreciated the temperature of the water, comparatively constant with
respect to the more variable one of the sand (and, partly, of the external
air).
In the land, the day-night cycle is quite evident in the surface
temperature data. It combines with the seasonal cycle, due to the
different average height of the Sun on the horizon, whose action on the
surface is similar to the one discussed in the previous chapter (with the
beams more or less inclined, the amount of radiant energy that reaches an
area unit changes). In the seas and oceans, on the contrary, the day-night
thermal cycle is only a slight perturbation in the much more evident
seasonal cycle, due to the cumulative effect of the enormous amount of
energy that, on the average, is absorbed during the summer and dispersed
towards the outside during the winter23. In any case, the high heat

To be more accurate, we should point out that the difference between the thermal
behaviour of water and that of land is more marked than that revealed by the calculations
made on the basis of the previous theoretical considerations: there actually exists another
effect that combines with the one due to the different heat capacities and enhances the
difference between the two behaviours. It consists of the fact that, whereas in solid land
solar energy is absorbed entirely in the surface layer, in seas and oceans, since they are
formed of liquid water, there takes place a certain mixing with deeper layers, so the
absorbed energy is distributed in a greater mass, determining a variation in temperature
that is smaller than that predicted by the previous considerations.
76 From Observations to Simulations

capacity of the seas and oceans causes their seasonal range of

temperature24, as well as their daily one, to be rather narrow. This leads
to the theoretical prediction, corroborated by observational assessments,
that, the solar radiation being equal (for instance at the same latitude), on
the average solid land is colder than the surface of the sea during the
winter and warmer during the summer.
Because of the previously mentioned mechanism of the warming of
the lower layers of the atmosphere from the bottom, caused by the
Earth's surface, all this leads us to understand how the presence of
expanses of water near a location limits the temperature range of the air,
tempering the climate, both in its seasonal average level and on a daily
time scale. Thus, for instance, at our mid-latitudes, it is possible to find
continental areas that are distant from the seas and oceans and have
particularly cold winters and sometimes torrid summers, if the
temperature values of the air at ground level are compared with those of
coastal areas, as in our previous comparison between the winters in
Rome and those in St. Louis. In the same way, with reference to the day-
night thermal cycle, the presence of the sea leads to a decrease in the
daytime maximum temperatures and to an increase in the night-time
minimum temperatures, with respect to the values in continental areas at
the same latitude25. Breezes contribute particularly to this phenomenon,
during the day by conveying towards the land air that was previously
above the sea, and during the night by blowing from the land towards the
sea.
From what we discussed previously, it is evident, in particular, that
the seas and oceans act as "climatic dampers", because — like the
suspensions of a car cushion the vertical oscillations of a car on rough
ground — the masses of water lower the variability rate of the
temperature variations in the atmosphere, on several space and time
scales. If, more specifically, we consider the global annual averages, we

The temperature range is the difference between the maximum temperature and the
minimum one within a certain period of time that has a certain cyclic characteristic.
25
This is why in the previous chapter we chose a continental area for discussing the
phenomenon of colder winter nights with a clear sky and warmer nights with a cloudy
sky: in these areas of the world the phenomenon is more evident than in other areas.
 The Theoretical Framework 77

can see (as we had already mentioned) that the interannual temperature
variability, presented in average terms in Plate 5, is also damped down
by the contribution of the seas and oceans. Now we can understand that
this is due precisely to their characteristics in terms of thermal inertia.

4.7 An Outline of Oceanic and Atmospheric Dynamics

Our discussion, here, of the thermal phenomena in the atmosphere has

led us to view the situation from a different angle (that of warming from
the bottom) with respect to the angle adopted in the previous chapter, to
contradict some naive ideas that had been presented there, and to achieve
a theoretical framework in which contradictions with observational
experience have practically disappeared. Nevertheless, some temperature
data on a regional scale are still unexplained, and the horizontal
dynamics of the atmosphere has not been tackled yet. So now, without
presuming to do so in a comprehensive manner, we will endeavour to
concisely outline some other phenomena, in order to achieve a more
detailed treatment of the system under examination and to consolidate
our theoretical framework.
It is known that our planet, because of the inclination of its axis,
undergoes the phenomenon of the seasons. This is due to the fact that the
warming of the Earth's surface by the Sun is greater during the summer
than during the intermediate seasons and the winter, because of the
different average height of the Sun above the horizon. In any case, if,
with a rather low time resolution, we consider the average effects over a
year, we can calculate the average energy absorbed by the Earth's
surface at various latitudes: its values turn out to be very high at the
tropics, lower at the intermediate latitudes, and even lower within the
polar circles. The consequent warming of the atmosphere by conduction,
convection and irradiation, according to our theoretical pattern, must
have a similar trend. Though from a qualitative point of view this is
confirmed by observations, from a quantitative point of view there is a
discrepancy between the predictions of our scheme and the average
temperatures of the air at the various latitudes. In particular, the
difference in temperature observed between the equator and the poles is
78 From Observations to Simulations

smaller than the one that might be expected theoretically. How can this
discrepancy be explained?
A first clue for the solution of this riddle comes from the observation
that the difference between the temperature of the oceans at the equator
and that at the poles is also smaller than the one estimated by calculating
sunlight absorption only. This gives rise to surfaces that are less
diversified in temperature, so the resulting influences on the air above
them turn out to be more similar to each other than had been predicted on
the basis of the previous reasoning. On the one hand we would like to
understand how this can happen, but on the other hand we will see that,
though from a qualitative point of view this effect goes in the right
direction of a mitigation of the differences in the temperature of the air at
different latitudes, from a quantitative point of view it is not sufficient to
reproduce the correct values of these differences.
In this case, an effect of horizontal transmission of the heat contained
in the masses of water is present. Once again, the transmission modes by
irradiation and conduction (or molecular diffusion) are not sufficient to
explain the phenomenon. Only the action of vast oceanic currents allows
such an efficient thermal mixing. Many of us are aware of the existence
of the so-called Gulf Stream, which carries masses of water through the
Atlantic Ocean from the Gulf of Mexico to the coasts of north-western
Europe, mitigating the climate in that area, at least in comparison with
the climate on the American shore of the ocean. In actual fact, now
oceanographers are aware of the existence of a circulation in the oceans
(the "thermohaline circulation") that leads to a global mixing,
particularly by means of an enormous underwater "river", the so-called
"conveyor belt", whose route covers the entire planet. As a rule, the
horizontal route of the ocean currents and their vertical dynamics26
depend on differences in temperature and salinity27.
If now we evaluate the surface temperature of the oceans at the
various latitudes and include these new data in our explanatory scheme
of atmosphere warming, we will find that there is still a discrepancy

There are areas in which deep water comes to the surface as a result of ascending
movements, and areas in which surface water sinks to the depths of the sea.
27
It is not possible here to expand on the subject of the general circulation in the oceans.
A clear explanation will be found in Lionello (2005).
 The Theoretical Framework 79

between the theoretical predictions and the results of observations: to be

more precise, from a quantitative point of view there remain a surplus of
temperature in the air at the high latitudes and a deficit at the low ones,
both of them unexplained. This means that there must be another manner
in which heat is exchanged between the tropical areas and the polar ones.
This manner does exist, and is present within the atmosphere, in the
atmospheric currents of its general circulation. It is not possible, here, to
discuss this topic exhaustively. It will be sufficient to mention the fact
that, once again, what is quantitatively important is the macroscopic
vertical and horizontal motion of air cells, rather than molecular
diffusion.
More specifically, it is expedient to point out that, in the description
of the atmospheric dynamics, an important role is played by the concept
of "air mass", a phrase that denotes a portion of air that is large enough
to allow us to disregard molecular diffusion and small enough to allow us
to consider it thermodynamically homogeneous, i.e. characterised by
univocal values of temperature, density and humidity that are
representative of the characteristics of the air of the entire cell. The
movements of the air masses in the atmosphere are obviously driven by
the forces that act on them, as in the case of what we might call
"buoyancy force", which pushes the cells up or down, depending on their
temperature and on the Archimedes' principle.
The horizontal motions are affected by the combined action of a force
called gradient force (which tends to move the air masses from the high-
pressure areas to the low-pressure ones), of an apparent force28 called
Coriolis force (due to the fact that the Earth revolves over a period of 24
hours), of the centrifugal force in non-rectilinear motions, and of a
friction force for the air layers near the ground. These forces, together
with the fact that it is often assumed that, at a certain height in the
atmosphere, air is not compressible29, lead to a qualitatively and
quantitatively satisfactory explanation of the atmospheric movements.

It originates as a consequence of the fact that the Earth's reference system is not
inertial: within it, a body to which no real force is applied does not remain still or move
with a uniform rectilinear motion, but performs more complicated movements.
29
From a mathematical point of view, this is described by an equation called mass
conservation, according to which, once a volume of air has been delimited, the amount of
80 From Observations to Simulations

Let us mention an example. Horizontal and vertical motions may

combine, and some may be the cause of the others. In an area where the
pressure is low near the ground, the combination of the forces mentioned
above tends to cause the air to converge towards the centre of the area;
since air is essentially not compressible, this results in its moving up
vertically. When the cell rises and cools down, the water vapour present
in it tends to condense, forming a cloud. On the contrary, in an area
where the pressure is high near the ground, the air tends to diverge from
the centre of the area; since air basically does not become more rarefied,
this means that a certain quantity of air comes down from the upper
layers. Descending, it warms up30, and the liquid water, if any, that forms
clouds evaporates (so the clouds dissolve). Thus we have also achieved a
qualitative explanation of the fact that low-pressure areas are often
characterised by cloudiness, while high-pressure areas are usually
associated to expanses of clear sky.
A last consideration to be made on the atmospheric dynamics is that,
because of the slowness of the diffusion process, when two air masses
touch each other, they do not mix diffusing into each other, but clash,
forming discontinuity surfaces called "fronts". The cold air usually
pushes under the warm one and the warm air slides over the cold one.
The mutual diffusion process, and obviously the interaction with the
ground below, become important only if two air masses remain
stationary next to each other for a long time.
The observational evidence relative to the shifting of air masses
having certain characteristics towards latitudes different from the original
ones fits perfectly in this explicative framework. These air masses,
remaining at a certain latitude for a long time, take on characteristics due
to their contact with the ground. More specifically, polar air is cold and

air that comes into it must be equal to that of the air that comes out of it. This does not
allow the air to be compressed or to become more rarefied. Obviously, this assumption
can be made in the atmosphere (the height remaining the same) and not in other physical
systems. We all know that, in other conditions, air can undoubtedly be compressed: this
happens, for instance, in a bicycle pump.
30
This mechanism is the inverse of the adiabatic cooling due to ascent in the atmosphere.
Now the air cell, descending, undergoes a greater pressure in the low layers. This results
in warming.
 The Theoretical Framework 81

dry, whereas tropical air is warm and humid . When, because of

atmospheric currents, these air masses move towards different latitudes,
their thermodynamic characteristics remain almost unchanged for a
certain time. In the northern hemisphere, this results in movements of
warm air towards the north and of cold air towards the south, which
reduce the difference between the temperature of the air in the equator
and that in the poles.
From the viewpoint of the physical and mathematical expression of
what we have presented qualitatively, in the same way in which we
previously showed that vertical dynamics and processes can be
understood and described in thermodynamic terms (for instance with the
application of the first principle to cases of adiabatic expansion), now we
can endeavour to describe the horizontal dynamics of air by means of
hydrodynamic concepts and equations. Since, as we have already stated,
the two types of motion combine, the two descriptions must be applied
simultaneously, in order to achieve an overall vision of the atmospheric
motions and processes. From the viewpoint of horizontal motion, air can
certainly be regarded as an actual fluid, provided its properties are
described by means of averaged concepts like the previously cited one of
air mass. It is thus possible to use concepts such as the density and
velocity of an air mass, and to apply to them the basic equations of
hydrodynamics (known as Navier-Stokes equations) on a rotating
reference system and with the forces that act on the air mass.
It seems, then, that the problem has been formulated correctly.
However, one of the essential difficulties that make this description very
complex is the fact that as a rule these equations cannot be solved
analytically.

4.8 Feedbacks and Complexity of System

In this chapter we endeavoured to give the reader a "key" for the

theoretical understanding of the phenomena that take place in the
atmosphere. We were inevitably compelled to isolate the various
phenomena, in order to achieve a better study of their dynamics and of

'This is due to the greater evaporation of water surfaces at low latitudes.

82 From Observations to Simulations

the causes that produce them. Though we tried to follow the thread of air
warming, the picture we obtained has several aspects, so this chapter
may seem rather dispersive to some readers. Indeed the picture that has
thus been obtained is not simple, so the only way to achieve a more
organic vision of what happens in the atmosphere is to understand the
mutual cause-effect relationships that each process or phenomenon has
with other processes or phenomena occurring in the system under
consideration. In the last part of this chapter, therefore, we will
endeavour to recover this more organic vision, limiting the treatment to a
few explanatory examples. By doing so, we could discover that our
system is really "complex".
First of all, in order to "prune" the treatment by eliminating all the
elements that can reasonably be disregarded, it is expedient to point out
that in the Earth system there exist some processes that are characterised
by different evolutionary time scales. For instance, we have already
mentioned the fact that the temperature of the seas and oceans varies
above all with the seasonal cycle and is only very slightly affected by the
day-night cycle. In this respect, therefore, if we wish to study and
understand the cause-effect relationships that involve the shorter-time-
scale processes, on this scale we can often regard the influence of the
hydrosphere roughly as a constant. Likewise, the amount of aerosols
present in the atmosphere or the quantity of frozen surfaces on the entire
globe can vary only by an extremely small percentage (with respect to
their total amount) in the course of a few days; however, as regards the
meteorological phenomena based on a small space and time scale, their
possible quick local variations (for instance, snowfalls on previously bare
ground or the emission of a great quantity of dust or pollutants) should
be evaluated for their consequences on the processes that take place on
this scale in the atmosphere subsystem of the Earth system.
The approach, obviously, is quite different in the study of climatic
phenomena. As a rule, here the specific meteorological phenomena
should be regarded as "fluctuations" around certain averages, while the
processes with a slower temporal evolution are the ones whose
evaluation is most important: they affect these averages, and ultimately
guide the changes in climate.
 The Theoretical Framework 83

Remaining in the by now familiar field of the discussion of the

warming/cooling of the air due to the influence of the terrestrial ground,
let us consider the warming produced by the ascent of the Sun above the
horizon during the morning. As the hours go by, the ascent of the Sun
determines an increasing irradiation that, if it falls on solid ground, leads
to an increase in the surface temperature. This rapidly results in an
increase in the temperature of the lower layers of the air via the
mechanisms that we know well by now (conduction, convection and
irradiation). Should everything boil down to this phenomenon, we would
be able to accurately predict the temperature at the various hours of the
day in a certain location. In actual fact, as we have explained, there are
other elements that disturb this situation, for instance clouds. Now,
clouds may appear in the sky because they are conveyed from more or
less distant regions; but they may also develop or grow locally. In actual
fact it is convection that, in the suitable humidity conditions, leads to the
cooling of the rising air cells, the condensation of water vapour in the
atmosphere and the consequent development of clouds. Among other
things, the rise in ground temperature also results in an increase in the
evaporation and transpiration of plants, and, to a lesser degree, in the
evaporation from water surfaces (whose temperature does not change
much). The most evident effect of cloud formation is that of screening
the sunlight and causing a smaller amount of it to reach the ground,
which consequently tends to warm up less or even to cool down.
We have thus started from a cause — the rise in ground temperature
as a result of the incident solar radiation — that gives rise to a chain of
effects: rise in air temperature, convection, increase in humidity,
condensation. These effects lead to a process, cloud formation, that
results in a decrease in the warming of the surface, or even in its cooling,
via the screening of sunlight and the reduction of its incidence on the
surface. This way we have described a chain of cause-effect interactions
whose last link loops back to its beginning: the last effect acts on the
cause that has generated the entire chain. There is, in short, a feedback
produced by the change in the temperature value on itself. In this case,
the final effect is that of lessening the increase in temperature: this is
called a negative feedback. In other cases, as we will show, the final
effect is that of intensifying the change: this is called a positive feedback.
84 From Observations to Simulations

Figure 6 is a diagrammatic representation of the meteorological

interactions that take place in the Earth system. The variables that
describe the state of the atmosphere and ground are represented by the
rectangles, while the ellipses represent the main meteorological
processes. The direction of the arrows indicates whether a process
changes the value of a certain variable or whether, vice versa, this value
affects a certain process32. Obviously it is possible to trace the feedback
of the change in ground temperature value on itself (by following the two
circular chains: one comprises ground temperature - sensible heat flow -
condensation - clouds - radiation - ground temperature, and the other
ground temperature - evaporation - humidity - condensation - clouds -
radiation - ground temperature). It may also be amusing to count the
number of circular interaction loops. And this is only a simplified
diagram!
Now it is much easier to understand our allusion, in the introduction,
to the existence of a circular causality, in contrast with the linear
causality typical of Newtonian mechanics. Not only the causes do not
add up in a linear manner and do not produce a sum of effects, but they
actually "act back" on themselves in a non-linear manner. To put it more
specifically, if we are studying an evolutionary phenomenon, such as the
tendency of the ground temperature and the consequent tendency of the
low-layer air temperature, from a mathematical point of view we cannot
describe the situation with a single equation33. We must analyse the
various processes by means of separate equations and their interaction in
a system of equations. If, as is often the case, this is not possible (for
instance because the system cannot be solved or the dynamic treatment
of some processes is not easy), it is possible to identify the basic process
that leads to temperature changes and to adjust its values by means of a
separate examination of the other processes that affect it. In any case,

The diagram shows, among other things, the "sensible heat flow". This phrase indicates
the flow of exchanged heat that produces an increase in temperature: it includes the heat
transmission processes by convection and by horizontal movements of the air. Though
the figure generally shows only interactions between variables and processes, we have
drawn a broken line also between the sensible heat flow process and the condensation
process, in order to highlight the connection mentioned in the text.
33
The variable T (ground) cannot be a dependent variable and an independent one at the
same time!
 The Theoretical Framework 85

there is a quantitative balancing problem in the calculation of the

influence of the individual processes on the variable under consideration.

Winds Humidity

Ground Ground
roughness humidity

Figure 6. A diagrammatic representation of meteorological interactions and

feedbacks.

If now we try to consider the situation from a climatic point of view,

that is on a broader time scale, it is not difficult to understand that here
too there are many causal loops among the processes that are important
in the determination of the climate. Let us reflect, for the time being, on a
global level, and suppose that, for some reason, the temperature of the
Earth's surface and of the air start increasing. How would the Earth
system respond to this increase? Undoubtedly the higher temperature
would change the amount of surface covered by ice, causing it to
decrease, because conditions characterised by a temperature lower than
the freezing point of water would be limited to higher latitudes, or, the
latitudes being equal, to greater elevations. This deglaciation effect,
however, is not harmless, because it has an important consequence on the
radiative balance between the Earth and outer space.
86 From Observations to Simulations

We have not followed up the subject of the properties of ice in its

interaction with radiation, but everybody knows that white surfaces
reflect most of the visible radiation that falls on them: so do ice and
snow. In comparison with them, the solid or liquid surfaces of land and
sea reflect much less and absorb a greater quantity of the incoming solar
radiation. Therefore the decrease in frozen surfaces results in a greater
absorption of solar radiation by the Earth's surface, promoting the
warming of the Earth and consequently of the low layers of the air. So
we have discovered a very evident positive climatic feedback:
deglaciation starts a warming process and encourages a further rise in
temperature. Among other things, positive feedbacks are more
"dangerous" than negative ones: the latter tend to lessen the effects of a
change that is under way and to bring the process back to its starting
point, whereas the former tend to increase the departure from the
previous situation, leading towards scenarios that are less known,
because in many cases they have not been observed previously.
As a matter of fact, a change in a variable in the atmosphere usually
affects several cycles. This applies also to an increase in temperature.
Another effect, which at present is considered less important than the
previously described effect of deglaciation, concerns cloud formation. As
we have seen also on a shorter time scale, the intensification of
convection and of the evaporation of the seas and oceans tend to cause an
increase in the quantity of clouds present in the atmosphere, with a net
effect of greater screening of the solar radiation; therefore this is a
negative feedback.
Finally, a further factor that should be considered, particularly on a
regional scale, is that of the dynamics of the previously mentioned
conveyor belt that guides ocean currents. This dynamics depends on the
temperature and salinity of the water in the various parts of the globe.
Deglaciation, with the consequent flow of fresh water into the oceans,
would disrupt this equilibrium. Recent studies34 demonstrated that an
excessive desalination of the northern Atlantic might certainly result in
changes in the structure of deep ocean circulation. Some researchers
even surmised that the Gulf Stream might disappear, leading obviously

'See, for instance, Wood et al. (1999).

 The Theoretical Framework 87

to climatic consequences on the whole of western Europe, where a period

of intense cold might begin, with a trend opposite that of the global
warming context. Among other things, studies on the dynamics of ocean
currents demonstrated that its changes are hardly ever slow and gradual:
in most cases they take place within a short period, because this
dynamics is affected by the exceeding of certain critical thresholds that
determine a sudden change in the circulation35. This leads us to partly
amend our previous statements about the oceans' "stabilising" role
(essentially due to the great heat capacity of water) with respect to the
climatic temperature fluctuations. This role, which remains basically
active, does not, however, exclude sudden changes in single regions of
the globe, resulting from a change in the directions of the ocean currents.
At this point, what should we say about the role of carbon dioxide
and other greenhouse gases in the present global warming phenomenon?
In Chapter 2 we showed some figures that present a striking positive
correlation between the concentrations of C0 2 and the global temperature
trend. After the discussion of the interactions between radiation and air
molecules, we know that the effect of an increase in the greenhouse
gases produces absorption of heat in the troposphere. But how does this
effect combine with the other climatic mechanisms and with feedbacks
that may be present in the system? Well, if we consider the increase in
the concentration of CO2 in the atmosphere as an observational datum,
we can safely assert that it tends to cause an increase in the temperature
of the air in the low layers. If we consider the interaction of the air with
the ocean surfaces, we will find that in the long run this increase in air
temperature will have a warming effect on those surfaces36. But now,
because of a mechanism that we cannot describe in detail here, the
process whereby carbon dioxide is stored in the oceans, and therefore
removed from the atmosphere, is gradually becoming less efficient as the

35
From a theoretical point of view, ocean dynamics is said to undergo "bifurcations" and
to be partly chaotic. The concept of deterministic chaos will be discussed in depth in
Chapter 7.
36
When we study the cumulative effects of the contact between the air and the ocean
surfaces on a large time scale, we can no longer disregard the perturbation it causes in the
"measurement" of the temperature of these surfaces, as we had done previously for the
sake of simplicity.
88 From Observations to Simulations

temperature of the oceans rises. So this last effect tends to further

increase the C0 2 in the atmosphere. We have thus identified a positive
feedback.
If we include water vapour among greenhouse "gases", we can assert
that the ocean warming obviously leads to a greater evaporation,
therefore to a greater presence of water vapour in the atmosphere, and
that this further promotes a rise in the temperature of the troposphere (at
least until we consider condensation and cloud formation, which, as we
have already explained, are elements that lead to a negative feedback).
As the reader has certainly understood on the basis of the information
provided in this chapter, the situation of the processes under way in the
Earth system, and in its atmosphere subsystem in particular, is extremely
complex, if we consider it from the angle of a theoretical understanding.
As we have already indicated, in this book it is not possible to delve
more deeply into this situation. What we have already explained,
however, is sufficient to allow the reader to understand that each variable
and each process present in the system interacts in a non-linear manner
with other variables and processes, often creating circular chains of
cause-effect relationships. Moreover, what occurs in the low layers of the
atmosphere depends crucially on the exchanges at the interface with
other subsystems, such as land, oceans, ice, and the biosphere (which
includes human activities).
As we have previously explained, it is difficult to deal theoretically
with this situation in a formal and quantitative way. Though the
disciplines involved in the understanding of the various phenomena are
all classical and well-established, a classical scientist of a few years ago
would not have been able even to quantitatively weigh the various
contributions to a typical evolutionary phenomenon such as air warming.
Nowadays, on the contrary, this is done. How it is done is one of the
main themes we will tackle in the next chapters.
 Chapter 5

The Galilean Experimental Method: A

Digression?

Summarising in a "skeletal" way the thread of our discourse, we may

state that in the previous chapters we explained on which observational
data meteorology and climatology are based, illustrated our theoretical
knowledge of the individual phenomena and processes occurring within
the Earth system, and were led to the final recognition that these
phenomena and processes interact in a complex manner. More
specifically, under the stimulus of the local study of the properties of
matter, a considerable part of the progresses of our theoretical
understanding of the system under examination came from reflections on
"domestic" observations, in which, for instance, the behaviour of the
atmospheric fluid could be observed more accurately. This way, we were
able to conduct a direct, "first-hand" investigation into phenomena that
normally take place also in the free atmosphere and are an important
component of its dynamics, e.g. convection. Incidentally, we should
point out that, in small environments that are isolated from the outside,
individual processes are probably less influenced by other processes,
contrary to the case of the free atmosphere, where a process is usually
affected by other processes and sometimes there are strong feedbacks.
Thus, for instance, in a closed room convection does not interact directly
with sunlight or with the horizontal motions of air.
The approach to a theoretical understanding of the phenomena under
examination presented in the previous chapter is based on accurate,
sometimes local observations of what occurs spontaneously in nature.
The idea that underlies this attempt at a theoretical synthesis is that of
gaining knowledge by using both mathematical methods that are

89
90 From Observations to Simulations

increasingly suitable for describing the observational situation (where

possible, with more modern, sophisticated techniques), and observations
that are increasingly accurate and continuous (in order to obtain as
plentiful as possible a sample of different "historical" situations). In this
context, obviously, theoretical activity in meteorology and climatology
may be negatively conditioned by the fact that the observational results
are only those presented by nature in its history. In actual fact, we have
seen that some elements of theoretical understanding come from
thermodynamics and fluid dynamics, understood as sectors of physics,
and here this limitation does no longer exist in scientific practice.
As we shall see, the stepping-stone to a better understanding of nature
rested, in some disciplines, on a transition from a tendency to perform
increasingly accurate, "first-hand" observations of natural phenomena or
processes, to a tendency to force what nature presents to us, practically
"manipulating" the system under consideration. In order to examine this
breakthrough in physics and in other so-called "hard" sciences, it will be
expedient for us first to refer to an analysis of the simple systems that are
familiar to all of us and in which the first "experimental" investigations
historically took place. Once we have appreciated the advantages of this
approach, we will be able to pose the question whether it can be
successfully applied also to other more complex systems and to
disciplines (such as meteorology and climatology) that have always been
historically characterised by purely observational research.

5.1 Aristotelian Physics of Local Motions and the Advent of

Galileo Galilei

When we decide to deal with simple systems that are familiar to

everybody, we cannot but think of Galilean mechanics. We will therefore
analyse (though without going into historical details) some elements of
novelty introduced in scientific practice by Galileo Galilei during the
seventeenth century: these elements now underlie all experimental
scientific disciplines.
During the last few years, Galilei's name appeared rather frequently
in the mass media, in relation to the revision process that the Catholic
 The Galilean Experimental Method 91

Church (and Pope John Paul II in particular) has been applying to certain
past instances of the ecclesiastic authorities' behaviour. So it is generally
known that the Italian scientist, on the basis of his observations through
the telescope constructed by him, discovered that celestial bodies are not
perfectly spherical1, studied their motions and eventually got to the point
of supporting the Copernican heliocentric theory against the Ptolemaic-
origin geocentric one. This was the basic reason for which the Inquisition
demanded and obtained Galilei's abjuration.
In actual fact Galilei was not only an attentive observer of the solar
system: besides his well-known discovery of Venus's phases, of the
existence of Jupiter's satellites and of their motions2, he also carried out a
more "local" activity, studying simple mechanical systems. It is precisely
this second activity of his that is regarded by scientists as the most
innovative one, because it introduced an investigation method that is
largely in use today and is often referred to as "Galilean experimental
method".
In the previous chapters we mentioned the fact that pure and simple
empirical data, in themselves, do not shed light on the behaviour of
matter: they must be fitted into a well-defined theoretical scheme. Only
with a "reading" of this type they become usable, both from the
viewpoint of the understanding of the system under consideration and
from that of possible interventions on it. Galilei's activity in the study of
the dynamics of simple mechanical systems revealed, for the first time,
the limits of explanations that were regarded as evident and "natural",
and were directly induced by the observational data and interpreted in the
perspective of the common sense stance of that period. It will be
instructive, therefore, to briefly consider how Galilei analysed the
problem of the fall of bodies, and, more generally, of all local motions,
and how he ended up by introducing a new vision of dynamics in place
of the Aristotelian one.

'Galilei realised, notably, that there are mountains on the Moon. This observation was
very important, because it contradicted the Aristotelian-origin assumption that our
sublunar world is formed of irregular, imperfect matter, whereas the celestial world must
be formed of incorruptible matter, with regular, perfect (spherical) shapes.
2
A11 these observations backed up a Copernican vision, or in any case a non-geocentric
one.
92 From Observations to Simulations

Most of the Aristotelian description of local motions reminds us of

the naive physical image we all developed as children, when we
examined these motions and found regularities in them on the basis of
the repeated observation of everyday experiences. For instance, noticing
that an iron ball falls to the ground more quickly than a wad of cotton,
and examining the fall of various objects of different weights and
densities, we were led by induction to believe that heavier bodies always
fall more quickly than lighter ones. Likewise, when we placed a body on
a horizontal surface, we all saw that, if no force is applied, the natural
state of motion of a body is that of rest; moreover, in this situation the
body certainly moves if one pushes it, but as soon as one leaves off
pushing it, it stops, sometimes after having covered a further stretch
because it is still propelled by the impetus it has received. An important
aspect of this description is that it is constantly corroborated by everyday
experience, and the actions we perform in harmony with this vision are
very useful for practical purposes: for instance, if a heavy object is about
to fall on my foot, I know that I must move my foot away very quickly,
in order to avoid getting hurt.
In Aristotle's system, obviously, this description that we have called
naive was consistently fitted into his vision of the world3. From this point
of view, for instance, the interpretation of the vertical fall of a stone
towards the ground is that, since it is chiefly formed of the earth element,
it tends to reach its natural place; likewise the flames of a bonfire, where
the fire element is predominant, tend to rise in order to reach their natural
place, the sphere of fire. Moreover, Aristotle was certainly not a naive
observer: this is testified by the fact that he also allowed for the
resistance of the medium in which the motion took place, and went so far
as to assert that the velocity, V, of the body whose motion was
considered was proportional to the ratio of the imparted force, F, to the
resistance of the medium, R; simplifying, and adopting a mathematical

3
In Aristotle's vision of the world, while celestial bodies are formed of aether, all
sublunar bodies are formed of a mixture of four elements: earth, water, air and fire. Under
the sphere of the Moon, the four elements are arranged in concentric spheres, according
to their weight: from the bottom up, first there is the sphere of earth, then that of water,
then that of air, and finally that of fire.
 The Galilean Experimental Method 93

form that obviously Aristotle did not use, we may express this with the
formula V=FIR.
This last assertion is precisely what allows us to understand that
Aristotelian physics is not problem-free: everybody knows that if a heavy
body is on a rough table and one pushes it delicately it does not budge: a
greater force is required to make it move. The proportionality relation
that we expressed as V = FIR implies, on the contrary, that the body
should move (though slowly) even if a very small force were applied to
it. What should we say, finally, of a body to which a force has been
applied and that goes on moving even when the force stops being
applied? This is what happens when a pen is pushed violently onto a
table or an arrow is shot from a bow.
In the Middle Ages some scholars of natural philosophy had already
realised that Aristotelian physics had some weak spots. On the whole,
however, the solutions proposed remained within the sphere of the
Aristotelian principles, at least up to Galilei's time4. The Italian
scientist's power of analysis and abstraction led him to thoroughly
criticise, for the first time, the Aristotelian-type theoretical scheme of
local motions and to develop, at the same time, a valid alternative to it.
An attitude undoubtedly peculiar to Galilei, that distinguished him
from natural philosophers of the Aristotelian sort, was his approach to
natural observations and to their current theoretical interpretation. He
often tried to imagine the qualitative results that might be obtained from
the analysis of events that had never been observed before, endeavouring
to draw inferences from them, even only from the angle of internal

4
To mention an example, in the sixth century after Christ, John Philoponos proposed a
change in Aristotle's proportionality relation: it could be expressed as V = F - R and
produced a positive velocity only after threshold R had been exceeded. Moreover, the
problem of the motion of an arrow was usually solved by imagining that the arrow,
moving forward, created a depression on its own butt. Since in Aristotelian physics the
case of vacuum was excluded, where the resistance of the medium was null by definition
and therefore (because of the relation V = FIR) the velocities should have been
unrealistically high or even infinite, the continuation of the motion of the arrow was
interpreted as a result of the force imparted to it by the air that rushed in to fill the
vacuum around its butt. An attempt to solve this problem that broke away from the
Aristotelian outlook was that of the theory of impetus: in this case the arrow, during its
flight, allegedly preserved a sort of memory of the thrust originally imparted to it.
94 From Observations to Simulations

consistency, for the theoretical framework in which the results were

being read.
A typical expression of this attitude of Galilei's was the argument
with which he "demolished" the Aristotelian theory of the fall of bodies.
According to Aristotle, the velocity of the fall of bodies is proportional to
their weight, so heavier objects fall to the ground at a greater speed than
lighter objects; this seems to be confirmed by natural observations.
Galilei tried to imagine a new situation, in which two objects of different
weights are joined to form a heavier body, for instance by tying them to
each other with a string or even gluing them together. What happens to
the new body? According to Aristotle's theory, since its weight is greater
than that of each of its two "parts", it should fall to the ground at a
velocity higher than that of the fall of the heavier part. On the other hand,
in the Aristotelian perspective of a velocity of fall proportional to weight,
the lighter part (whose velocity of fall is lower) would slow down the fall
of the heavier part, which in turn would speed up the fall of the lighter
part: the resulting velocity of the composite body would be intermediate
between that of the heavier body and that of the lighter one.
Thus, with two arguments that are both within the Aristotelian logic,
we obtain two different qualitative results for the velocity of fall of the
composite body. Starting from the assumption of a velocity of fall
proportional to weight, we reach a contradiction: this suggests that the
assumption is not valid! We do not know whether Galilei ever actually
carried out an experiment of this type. In any case, a conceptual
contradiction like this one was already sufficient to address him to a
different path from the Aristotelian one in the investigation and
theoretical interpretation of local motions.

5.2 The Galilean "Style"

So Galilei's first step in the direction of a modern scientific conception

was that of imagining the possibility of observing events expressly
created by human beings, for instance, events that had never occurred
before in nature. In this sense, Galilei spoke of "sensible experiences"5,
5
Galilei (1967).
 The Galilean Experimental Method 95

both for natural observations and for the "experiences" devised and
carried out by him. In the latter case, the "experience", for instance with
the construction of the composite object described above, was expressly
chosen by the scientist, and its purpose was to test a theoretical
explanatory scheme of the phenomenon under examination. Another
element that distinguished the Galilean vision from the Aristotelian one
was the importance that it gave to the quantitative analysis of the
problems by means of an extensive use of mathematics: we have already
mentioned the fact that Galilei believed that the book of nature was
written in mathematical terms.
The second step (the decisive one) towards a modern scientific
conception was taken by Galilei with the method used for choosing and
preparing the "sensible experiences" devised and carried out by him,
which, from now on, we will call "experiments". On the one hand, at the
basis of the choice of an experiment there was a theoretical hypothesis
about what happens in nature, in the field under examination, for
instance the dynamics of moving bodies. On the other hand, at the basis
of the preparation of an experiment there was the awareness that an
individual phenomenon is affected by manifold physical factors, and that
it is possible to separate these factors and eliminate or minimise some of
them, so as to simplify the system, up to the point of being enabled to
study the effect of a single cause on the phenomenon under
consideration.
As regards the theoretical hypothesis that underlay the choice of the
experiment to be carried out, this hypothesis was usually suggested to
Galilei by natural observations and previously conducted "experiences";
however it did not stem directly from them or from the consequent
accumulation of data by simple induction. During this stage, the
scientist's creative act appeared to be essential. In the scientific
biography of Galilei, we can find several ways in which he was led to
develop hypotheses for the law of inertia and the fall of bodies. An
example derives from the consideration that bodies of different weights
fall at different velocities through different media (oil, water and air); by
adding to this the fact that the difference between these velocities
decreases as the density and viscosity of the medium decreases, Galilei
reached, by extrapolation, the hypothesis that in vacuum (where, by
96 From Observations to Simulations

definition, density and viscosity are null) all bodies, heavy and light, fall
at the same speed, independently of their weight6.
However, the creative act would have remained barren had it not been
united to the ability to prepare an experiment that could reveal something
about the validity of the hypothesis thus formulated. Only with this
preparation, Galilei's power of abstraction was put to the test. The
essential characteristic of the Galilean method of investigation consisted
precisely in the combination of the power of abstraction with the ability
to "manipulate" physical reality in order to prepare and carry out an
actual experiment.
A fundamental idea that Galilei expressed in some of his writings was
that the world hides its real mathematical nature behind "spurious"
elements that disrupt regularities and thus make it difficult for us to
understand it. The Salviati-Galilei of the Dialogue1 explicitly maintained
that it was necessary to "deduct the impediments of matter". This
programmatic assertion materialised in the preparation of the
experiments, when the Italian scientist "released" nature from its
impediments, for instance by smoothing off his tables and inclined
planes. This way he kept only the essential structure of physical nature,
reducing to a minimum what, in his opinion, prevented a simple
mathematical reading of the dynamics of local motions, i.e. friction. Thus
he discovered the law of inertia (whereby a body that is not subjected to
forces persists in its state of rest or uniform rectilinear motion), and
found the correct relation between the time and the space covered by
non-horizontal motions: this relation eventually led to a direct
generalisation in Newton's second law of dynamics.
This Galilean approach was obviously influenced by a vision of
reality that was reminiscent of the Platonic one, with the true essence of

6
It may be interesting to point out that Galilei adopted extrapolation as a useful
conceptual practice in other occasions as well: the best-known example is that in which
he had to defend the veracity of his astronomical observations through the telescope.
When someone asserted that the lens had a deforming effect and therefore led to
untruthful observations, Galilei replied that observations through a telescope could be
personally verified on the Earth (by observing an object at a distance, then going there to
check the object). Since this was true for increasingly great distances, it was reasonable to
infer that it was true also for distances such as that between the Earth and the Moon.
7
Galilei (1967).
 The Galilean Experimental Method 97

things situated in an external world (which for Plato was unattainable

and for Galilei could somehow be approached through experiments); and
this, in fact, was how it was interpreted in the past. Without attempting a
philosophical discussion, we wish here to emphasise the methodological
novelty of this approach. More specifically, the analysis of the elements
of the system (the ones that are considered fundamental and the ones that
are considered spurious) and the elimination of some of them (the
spurious ones) can be interpreted as a typical causal-influence analysis
on a single phenomenon. The elimination of the spurious elements
confirms an actual separability of the individual causes that act on the
observed phenomenon: this separability is first surmised theoretically,
then realised practically with the smoothing off of the tables.
Thus Galilei paved the way for a new "style" in scientific
investigations. Scientists were no longer content with carefully observing
nature and inferring its regularities by induction, but advanced original
hypotheses and put them to the test, with experiments in which all the
working conditions were controlled by the experimenter and only rarely
corresponded to those found in observational reality. The latter, in many
cases, was too complex to be interpreted correctly without a suitable
simplification of the conditions in which the observations took place. It
was necessary, therefore, to prepare the system so it could be examined
in simpler conditions, for instance by removing the spurious elements:
this made it possible, more specifically, to determine the effect that a
single cause could produce on a certain phenomenon.

5.3 A Galilean Method for Studying the Weather and the Climate?

With Galilei, in short, domestic observations became laboratory

experiments, in which the initial state of the system under examination
and all the conditions in which the experience took place were decided
by the experimenter. In a laboratory we can obtain conditions that would
never appear in nature, to the point of making it possible to analyse the
behaviour of the system under the influence of a single cause, if
necessary repeating the experiment over and over again. Thus the
determination of a physical law in simplified conditions can easily be
98 From Observations to Simulations

achieved; and a law can be expressed mathematically by means of an

equation.
In this process, we have implicitly assumed that the individual
concurrent causes that affect a certain phenomenon can be separated
from each other; in our case we may even consider eliminating all of
them except one (gravity)8. This way we get to the point of studying an
ideal situation, which would be difficult to find in natural reality. If,
however, now our goal is to achieve a correct description precisely of
this natural reality in its complexity, the next step may certainly be that
of adding another cause in the controlled system and of analysing the
combined effect of the previously studied cause and the added one. For
instance, we can reintroduce the friction component in the system, and
verify how it perturbs the law of inertia or the uniformly accelerated
motion in the phenomenon of the fall of bodies. In a simple dynamical
system, we may find, for instance, that the effect of friction is that of
slowing down motion in comparison with a situation in which there is no
friction, and that friction leads to the addition of a term in the equation
that describes the motion. Proceeding in this direction, we can attempt to
reconstruct natural complexity in the laboratory, where work is carried
out in controlled conditions.
These operations of "decomposition and recomposition" of a system
in the laboratory, together with the controlled manipulation of all the
experimental conditions, have produced enormous results in terms of the
description and understanding of systems, starting from classical physics
and chemistry, and proceeding to nuclear and subnuclear physics. Our
basic knowledge of non-living matter is for the most part due to the
application of the Galilean scientific method.
Considering these premises, it is reasonable to suppose that if we
could apply this method also to the study of meteorology and climatic

In comparison with Galilei's time, nowadays our experimental capability is obviously

much more sophisticated, so we can rely less on extrapolations (for instance, we can
obtain vacuum minus a very small number of molecules per cm3). In this context we
should point out that, perhaps because of the intrinsic limits of the seventeenth-century
experimental apparatus, or more probably because of a cultural inclination of his, Galilei
always showed that he preferred "sure demonstrations" to "sensible experiences": it is
even possible that he never performed some experiments that have been attributed to him
in the past.
 The Galilean Experimental Method 99

changes, our knowledge and our possibility of describing and predicting

what takes place in the Earth system would be enormously increased. Is
it possible to achieve this theoretical and practical result?
To answer this question, we must first of all emphasise the fact that
the purpose of the activity of decomposition of a system in the laboratory
is to acquire as detailed as possible a knowledge of the nature of the
phenomena and of their responses to individual causes. The subsequent
activity of adding other elements to the system is undertaken, on the
contrary, in order to achieve, in controlled, repeatable conditions, a more
realistic description of what actually takes place in nature, where the
various elements of the system interact in a way that is difficult to
interpret by simple observation. So the former activity has a purely
theoretical value, because it aims at increasing the knowledge of basic
phenomena, whereas the latter activity is both theoretically and
practically important, because its purpose is both to study the interaction
between the various elements of the system, and to attempt to "mimic"
the behaviour of the natural system, endeavouring to reconstruct it,
somehow, in controlled conditions.
This double operation of decomposition and recomposition of a
system in the laboratory sometimes gives rise to problems. More
specifically, this activity is usually easy when the phenomena under
examination depend on a small number of causes, perhaps linear and
easily separated, but is extremely difficult when the phenomena are part
of a highly interacting system, where the phenomena "emerge" from a
complex, non-linear mixing of concurring causes. As we explained in
detail in the previous chapter, this is the case of the Earth system9.
Another example of the difficulties that are met when attempting to
examine the behaviour of the atmosphere in the laboratory is the study of
air as a thermodynamic fluid. Obviously in a laboratory it is possible to
study small air masses, if necessary using experimental apparatuses (for

9
Here I would like to avoid any misunderstanding: the Galilean experimental method is
not applied only to mechanics or classical physics, but also to more "difficult" fields,
where investigation requires sophisticated instruments and theories, for instance nuclear
physics: here, however, this is enormously facilitated by the alleged linearity of the
phenomena, which are described by a theory that is "difficult" but linear, such as
quantum mechanics.
100 From Observations to Simulations

instance revolving spheres and cylinders) that force them to perform

motions similar to the real ones. In reality, the air masses involved in the
atmosphere are much larger. How does this fact affect the "translation"
of the experimental results obtained in the laboratory into the
macroscopic domain of the real atmosphere? There even exists a theory,
called theory of similarity, that deals with the mathematical similarities
found between phenomena observed empirically on various scales when
they are correctly scaled up or down and that can be applied to the
"translation" of these laboratory findings.
What should we say, moreover, about the fact, which has also been
already discussed, that some phenomena seem to emerge from complex
macroscopic interactions, without having an equivalent in microscopic
processes? In the laboratory, we usually study with great attention the
elements that form a system (e.g. air molecules) and the processes on this
scale (e.g. molecular diffusion), whereas in meteorology the validity of
the adiabatic hypothesis, discussed in the previous chapter, suggests that
some of these processes may be disregarded, at least in good
approximation, while others take on a dominant role.
These last reflections lead us to believe that whereas on the one hand
the Galilean experimental method can be applied successfully to the
study of the basic elements of the atmosphere and of the Earth system
(for instance as has been done actually for the study of the interactions
between radiation and matter), on the other hand the complexity of the
system seems to elude a local experimental investigation. The ideal thing
would be to have a laboratory as large as the Earth itself, in which it were
possible to manipulate the system, i.e. to prepare its initial conditions and
its "boundary" ones, in order to carry out real experiments and to extract
theoretical syntheses from them, in a simplified, repeatable way.
Obviously this is not possible: nobody is able to perform such a broad
"meteorological experiment" or, even less, such a "climatic experiment",
simply because we cannot achieve the control of all the elements of the
system. It seems, therefore, that we are not able to fully exploit the
Galilean breakthrough, which has been so fruitful in other disciplines.
So up to the nineteen-sixties and nineteen-seventies the study of
meteorology and the climate had a purely observational character and
had a great difficulty in formulating theoretical syntheses, not so much
 The Galilean Experimental Method 101

for the study of the basic elements of the system and of their elementary
interactions, to which the Galilean approach could be applied, as for the
study of the system in its totality and complexity, for which scientists
were compelled to keep to theoretical-empirical rules that had very little
descriptive and predictive power.
During those years, however, something changed; and, from that time
on, the problem we posed here of the retrieval of the Galilean approach
also in meteorology and climatology can be considered in a different
perspective. What has changed? Does there exist a technical ploy that
makes it possible to get around the problems that had not allowed us to
apply the Galilean experimental method completely? Or do we have to
achieve a new breakthrough in scientific practice, a change in our
cultural attitude, in the broadest sense of the word? We will endeavour to
answer all these questions in the next chapter.
 Chapter 6

Simulation Models

In the previous chapter we explained how Galileo Galilei usually worked

out his own idealised representation of the reality he was studying, first
of all formulating hypotheses, then putting them to the test in the
experimental "trial". The final result was often expressed in the
formulation of mathematical relations that described the phenomena of
motion in the controlled conditions of the experiments: Galilei did this,
for instance, in the equation that linked the elapsed time to the covered
distance in the fall of bodies. For the problem that is being studied here,
this equation is a valid example of algorithmic reduction.
This way the mental representation of the course of a certain
phenomenon, after it has been studied experimentally in simplified
conditions, leads to a mathematical formulation that describes the
operation of the simplified system, which is believed to possess the
essential characteristics of the systems seen in reality.
To put it in the language of today, Galilei first constructed a "mental
model" of the course of motions in basic conditions (i.e. after the
spurious elements had been eliminated), then verified this model by
means of a "material model" (the inclined plane), and finally worked out
a "mathematical model" (the equation of motion) that formally and
synthetically described the motions in these simplified conditions.

6.1 How Many Meanings Does the Word "Model" Have?

The concept of "model" is of great importance today in all scientific

disciplines, not only in the physical ones. At the same time, as the reader
has certainly understood from the various adjectives we have added to it,

103
104 From Observations to Simulations

this term can be used with different meanings in scientific terminology

— as it can, actually, also in common language. Any dictionary can
show us the great variety of meanings that may be attributed to this noun.
A model may be defined as a thing or person that is considered
exemplary, and therefore worthy of being imitated; as an original to be
reproduced; as a person who sits for painters or sculptors; as an industrial
prototype; as a garment that is sewn as a unique specimen according to
an original design; and so on.
Proceeding to a more strictly scientific sphere, we will now concisely
analyse the various meanings that coexist at present and have developed
during the course of the history of science.
The original meaning of the word "model" was probably that of a
material reproduction of a physical system on a certain reduced or
enlarged scale: for instance, maps of the Earth, or, in more modern times,
reconstructions of the structures of crystals (enlarged) and of the
atmospheric circulation on revolving spheres (reduced). These
reproductions are useful displays of the systems under examination, but,
as we stated in the previous chapter, it is not always possible, by
changing the scale, to obtain reliable information about the behaviour of
the original natural system. Even the inclined planes and the balls used
by Galilei cannot be regarded as simplified material models that
reproduce, through a simple change of scale, all the characteristics of the
motion of stones rolling down the sides of a mountain (velocity,
acceleration, space covered, time elapsed). These systems became useful
only when their use made it possible to determine some fundamental
laws (such as those of the fall of bodies) that could subsequently be
applied to systems on different scales. In this sense, modern experimental
laboratory reconstructions do not aim at imitating the behaviour of
greater systems: their purpose is to reproduce, in controlled conditions,
the basic physics of systems on the small scale of the laboratory.
The inability of a simple material reconstruction to achieve an
imitation, on a different scale, of the behaviour of a natural system
reveals the need for the construction of a mental model in which previous
hypotheses or theoretical knowledge coexist with an ideal, simplified
representation of the physical system to be studied. This new concept of
model actually appears in various examples in the history of physics: the
 Simulation Models 105

best-known one is perhaps the planetary atom model, in which simple

little balls with a negative electrical charge (electrons) revolve around a
positively-charged nucleus, under the "thrust" of the Coulomb attraction.
The latter is described in the same mathematical form as gravitational
attraction, so the analogy between planets and electrons turns out to be
cogent.
This concept of model as an ideal display guided by our theoretical
knowledge, however, was gradually dropped in the history of physics,
undoubtedly in the case of microphysics: the brief history of the
planetary atom model illustrates the difficulties inherent in the displaying
of microscopic bodies1. On the other hand, in classical physics the
physical elements that are considered are macroscopic, therefore can be
displayed, located and separated. This suggests that it may be possible to
construct an ideal model, at least for some phenomena: for instance, we
can visually imagine the interaction between two air masses, in a limited
environment, under the action of their different thermodynamic
characteristics, with the warm air masses sliding above the cold ones and
the latter pushing under the former, thus "modelling", at least
qualitatively, the frontal interaction we briefly described in Chapter 4.
In any case, both in classical physics and in quantum physics, the real
breakthrough in the understanding of a phenomenon takes place when a
mathematical model of that phenomenon is worked out, that is when a
description of it (and, if possible, also of its future evolution) is achieved
by means of mathematically formalised physical laws that represent an
algorithmic reduction of the apparent complexity of the real system. The
mathematical model is usually formed of one or several equations in
which the individual variables represent the properties of the individual
elements of the real system under examination (for instance, the
temperature and humidity of the air masses that come into contact with
each other). So there exist some actual rules of correspondence that

'Simple calculations of classical electrodynamics demonstrate, for instance, that an

electron that performs a circular motion (and therefore undergoes a uniform acceleration)
quickly loses energy and soon ends up by falling on the atomic nucleus. The quantum
description, moreover, which is the only one to adapt to the properties of microscopic
bodies, shows that they cannot be displayed, located or separated in a classical sense.
106 From Observations to Simulations

interrelate the descriptive variables and the properties of the real objects,
whose interactions can be studied concretely in the laboratory.
Obviously, when one is in possession of the mathematical model of a
phenomenon (obtained, for instance, from the formulation of hypotheses
and from the previous execution of experiments), this model can be
validated again in new experimental situations in the laboratory. The
validation on the real system, though in the simplified conditions of the
laboratory, is always necessary if the model is to be regarded as a
scientific one.
In practical terms, if, for instance, we are in possession of an
evolutionary mathematical model (i.e. a model that is believed to be able
to describe the evolution of a certain phenomenon or process in the
course of time) and wish to find out whether it is a more or less faithful
representation of an experimental reality in the laboratory, we must
usually solve the equations of the mathematical model, by fixing an
initial instant and a final one, and by determining the values of the
variables at the initial instant and all the boundary conditions of the
system under examination. If we consider the example of the two air
masses that have different thermodynamic characteristics and interact,
we will have to introduce into the equations of the mathematical model
the various initial values of velocity, temperature, humidity, etc., of the
two air masses, and also the temperature of the walls of the room that
contains them, both at the initial instant and throughout the experiment.
Once the equations with these initial and boundary data have been
solved, and once the laboratory experiment has been carried out, we can
compare the results predicted by the mathematical model with those
obtained through the measurements. If the values predicted by the model
for the different variables in the different points of the experimental
room are "similar" to the ones that have been detected experimentally
(i.e. if they remain within the error bars that characterise these
measurements), we may assert that our model is a good description of
what takes place in the reality of the laboratory.
What we have stated up to now makes it evident that the
mathematical model represents an element of theoretical knowledge of
the system under examination. This knowledge may be limited to a very
simple real system, which is decomposed in the laboratory up to the
 Simulation Models 107

point of considering a single cause acting on the phenomenon under

investigation (as in the cases of the motions considered in the previous
chapter); or this knowledge may be extended to a mixing of concauses
that act on the phenomenon. In any case, the activity of decomposition
and recomposition of the real system present in the laboratory has an
analogue in the mathematical model, which, for instance in the case of a
recomposition, may be formed of a system of coupled equations. And at
any rate the purpose of the elaboration of a mathematical model
nowadays is very pragmatic: it usually aims at mathematically describing
and reconstructing the experimental reality within a very limited domain.
If a more complete theoretical construct is sought, we have to use the
term "theory", and the range of validity of a theory must be broader, at
least enough to include all the experiences known in the field under
consideration; in a theory all arguments may spring from axioms (i.e.
from basic postulates regarded as certain), and the aim of a theory may
even be that of grasping the essence of the investigated reality (which in
Galilei's view meant approaching the Platonic perfect world). Once a
theory has been adopted, it is possible to work out models that use its
equations to describe particular phenomena or processes: as a rule, when
a laboratory "experience" is available, only a comparison between the
results of the model and the experimental reality can lead to the
corroboration or refutation of the individual equations.

6.2 The Simulation Approach

As we have explained, when the system under examination is a complex

one like the atmosphere or the whole Earth system, the Galilean
experimental method, and the consequent fruitful interchange between
mathematical models and laboratory experiments, can be applied only
locally for studying the basic properties of the individual constituents of
the system and a few simple interactions between them. This is true for
any system characterised by strong feedbacks and manifold concauses
that interact in a non-linear manner on a single phenomenon, in other
words for any complex system. Must we therefore lose all the
108 From Observations to Simulations

fruitfulness that springs from the interaction between theory and

experiment through the concept of mathematical model?
To answer this question, it is expedient to shift our attention
temporarily from the experimental difficulties we have just mentioned to
the theoretical ones that underlie non-linear mathematical models. With
the usual recomposition technique, when we wish to describe a complex
system in which two or more causes produce the change of a certain
quantity in a circular manner (i.e. with a feedback), we consider a system
of two or more non-linear equations, whose solution, with certain initial
and boundary conditions, supplies the model's prediction for the changes
in the variable under examination. Now, the crucial fact here is that these
systems of equations make it possible to achieve analytic solutions only
in very particular, simplified cases that are hardly ever realistic. Up to a
few decades ago, this limitation was added to the difficulties of
experimental reconstruction, and made it utterly impossible to tackle the
problem of working out a mathematical model for a complex system,
drawing from it theoretical prescriptions about the system's behaviour
and evaluating its effectiveness through an experimental comparison.
Today, at least from the angle of theoretical description, the situation
has changed. The introduction of computers has made it possible to solve
these systems of non-linear equations in all physically substantial and
relevant cases. Obviously these solutions are not analytic, but numerical:
they are found by software programmes by solving these systems step by
step, supplying a spatially and temporally discrete evolution of the
variables contained in them.
All in all, we can consider a real system that evolves in the course of
time, and we can describe its behaviour by means of a mathematical
model whose equations are solved numerically by a computer2. If the
variables present in these equations are linked to properties of the objects
of the real system, what we obtain is a "simulation" of the real system's
behaviour on the computer. So we will use the expression "simulation
models" to indicate those mathematical models whose equations depend
on time and on variables that correspond to some actual properties of the

2
If the equations are differential, this is called "numerical integration".
 Simulation Models 109

system under examination, and are solved numerically by means of a

computer.
In the "virtual" world of the simulation model, it is possible to follow,
step by step, the evolution of the variables in space and time, if necessary
displaying their values in a graphic form: in this sense, at least in
classical physics, the simulation models make it possible to meet the
need for visualisation that is particularly felt by scientists and that
characterised the ideal models we discussed previously.
As we have explained, before the introduction of simulation models
the comparison between a mathematical model and a real experiment
depended on the ability of the individual scientists to carry out analytic
calculations in strictly predetermined and simplified conditions, and on
the reconstruction of these situations in the laboratory. Now the
possibilities of experimental verification are becoming more extensive.
And that is not the whole story. Even in cases where it is not possible to
reconstruct in the laboratory the complexity of a macroscopic real
system, a simulation model allows us to consider the variables and
interactions that are regarded as fundamental and to simulate the
evolution of a simplified system, subsequently comparing its behaviour
(simulated in the model) with the one observed in reality. In this case, the
goal may be to verify the correctness of a choice of variables and
interactions that aims at reproducing (more or less faithfully) the
phenomena and processes found in observational reality.
As our explanation so far has made clear, the strong point of the
simulation approach consists chiefly in the fact that it makes it possible
to tackle complex problems from a theoretical point of view. Simple
systems, with only one or a few causes that interact in a linear manner,
can be described by means of equations characterised by analytic
solutions. On the contrary, the "recomposition" of a complex system
starting from the causes and interactions we regard as fundamental — an
activity that, as we have explained in the previous chapter, is extremely
arduous in the real world of the laboratory — can be undertaken more
easily in the virtual world of the simulation model. This world can be
manipulated and controlled at will, and within it we can study how
certain macroscopic phenomena emerge from the interaction of basic
processes.
110 From Observations to Simulations

More specifically, this reconstruction activity can be made

increasingly realistic by introducing in the simulation model some
theoretical elements relative to interactions or processes that are regarded
as less and less important. In this case, obviously, the experimental
validation is not achieved by means of a comparison with the results of
laboratory experiments, but from the observation of a real system.
In the light of what we have just explained, it is clear that the Galilean
experimental method discussed in the previous chapter can be applied
directly only within the local sphere and to the basic properties and
interactions of the constituents of a complex macroscopic system. Should
we wish to reconstruct the complexity of a real system of this type, this
would be possible not in the real world of a laboratory, but in the virtual
one of a simulation model. In the latter, we would start from basic
theoretical elements, the individual equations of the model (the only
elements to be validated by means of a classical experimentation activity
in a real laboratory), and compound them, in order to simulate the
complex interactions that take place in the real system. Though on the
one hand the real world eludes the researcher's control, on the other hand
the simulation model can be extensively manipulated, and there the
scientist can carry out "virtual experiments". In the world of the model
we can determine, a priori, the conditions in which the interactions are to
be studied, and at any time we can change both the equations and their
coupling (this allows us, for instance, to describe the intensity of the
feedbacks theoretically), and also the initial and boundary conditions that
affect the phenomena under examination.
So at this point we have explained the advantages of the application
of the Galilean experimental method to simple systems, we have
illustrated the difficulties of a purely observational approach to the study
of meteorology and science of climate, and we have concisely presented
the simulation approach to the study of complex systems. The
conclusions we have drawn are that, though it is impossible to
reconstruct in the laboratory the complexity of macroscopic systems such
as the atmosphere, it is possible to "transfer" this reconstruction into a
different sphere, that of the simulation model. Within the latter we can
retrieve our theoretical knowledge (through the model's equations), the
description of the system's physical state (through the initial conditions)
 Simulation Models 111

and the interaction with other systems or with the external environment
(through the evolution of the boundary conditions). This way,
experimental activity becomes possible in the world of the model, and its
validation is achieved by means of a comparison with the behaviour of
the real system that is being simulated, albeit in a simplified manner.
Once this change of perspective has been accepted, it seems natural to
regard the simulation approach as the "heir" of the Galilean approach to
the study of reality.

6.3 Conceptual Novelties in the Simulation Method and in Its Use

Given the conceptual importance of the simulation method in modern

scientific practice, before we proceed, in the next chapters, to discuss its
specific use in meteorology and science of climate and to stress the
peculiar, original aspects of these applications, it is worthwhile to dwell a
little more on the aspects of novelty (shared by all the areas of
application) of this approach3.
Summarising concisely what we have already partly discussed in this
chapter, and adding a few elements of further clarification, we may list
the main characteristics of a simulation model, as follows.
There is a one-to-one correspondence between the quantities of the
real system and the variables of the simulation model: what evolves
in reality has an equivalent in what evolves in the model. It is
possible to display graphically and follow over a period of time the
behaviour of these basic variables: thus we obtain a concrete (and
dynamical) equivalent of the ideal models worked out in the past by
physicists.
The real system is assumed to be decomposable into basic processes
and interactions that can be described by single equations, "pieces"
of theory that achieve the algorithmic reduction of these processes
and interactions. The purpose of the simulation model is not to
corroborate or falsify these equations, but to reconstruct the
complexity of the system by means of these equations, to simulate

3
The remarks that follow in this chapter are considerably influenced by some discussions
I had with D. Parisi, whose clarity of mind and depth of thought were a great help to me.
112 From Observations to Simulations

the operation of the real system, and to make it possible to validate

the model by means of a comparison with the large-scale behaviour
of the system. Underlying this way of proceeding, there is the
fundamental idea that, whereas for studying and understanding
simple systems it is essential to decompose the system and analyse
its individual causes in the laboratory, for understanding a complex
system it is necessary for the individual components of the model to
interact so as to correctly reproduce the behaviour of the real system.
In this sense, to reproduce correctly means to grasp (both
qualitatively and quantitatively) the complex non-linear dynamics of
the system under examination.
The simulation of the evolution of the characteristics of the real
system is achieved by means of a numerical solution of the equations
of the simulation model whereby the values of the corresponding
variables in space and time are obtained in a discrete manner4.
With simulation models, the behaviour of a system is studied as a
whole and in its complexity, without isolating it from the
environment with which it interacts. The action of the latter can be
described dynamically, for instance by means of an equation that
outlines its evolutionary behaviour and interacts, in a non-linear
manner, with the other equations, or only as an external forcing
factor that is not affected by feedbacks from the system.
In the world of the simulation model, scientists are completely in
command of the virtual system that simulates the real one. They can
carry out numerical experiments and repeat them at will, with an
extreme ease in changing the theoretical elements of the model and
the situations of the experiments (for instance, by changing the
values of the variables). More specifically, as after Galilei scientists
began to manipulate reality in order to obtain experimental situations
that had never appeared in nature, now in the simulation model
scientists can create possible worlds that have never existed in

4
The solution of the equations of the simulation model takes place in a discrete manner:
there are no continuous solutions like those of the analytic solution. The reader who
wishes to understand more concretely how this can occur is referred to the examples in
the next chapter.
 Simulation Models 113

natural history. As we will explain in Chapters 7 and 8, this

possibility is extraordinarily important for understanding future
environmental scenarios.
Referring again to the fact that in a simulation model it is so easy to
manipulate the studying conditions, in the same way in which it is
possible to hypothesise future scenarios and observe the relative
behaviour of the system, in "historical" sciences it is possible to
reconstruct the past (for instance, different environmental conditions)
and to study the behaviour of the virtual system in the model under
these conditions. This activity is extremely important, because only
for the past we can achieve a validation of the model by means of a
comparison with the behaviour of the real system that is being
simulated.
In the same way, by changing the equations of the model or their
coupling parameters (which define the feedback values), scientists
can easily evaluate the empirical effectiveness of several theoretical
schemes in the evolution of a complex system.
Having said this, I would like to conclude this short excursus in the
world of simulation models with a few more basic remarks that are
particularly relevant to what we will discuss in the next chapters.
It is known that in present-day science specialisation is carried to
extremes. As early as the first half of the twentieth century, the great
Danish physicist Niels Bohr humorously expressed this conviction by
remarking that "an expert is a person who has made all possible mistakes
within a very narrow field". Well, when we deal with complex systems,
we frequently happen to meet with different theoretical descriptions of
individual components of a system, in each of which the language typical
of the specific discipline that studies it has been adopted. If we wish to
reconstruct the behaviour of the system as a whole, we must then "link
up" these languages. In this sense, the working out of a simulation model
for a complex system compels the individual experts to speak the same
language, in particular to devise a unified, interdisciplinary formalism to
be introduced in the model.
The emergence and increasing development of simulation models,
therefore, essentially promoted the tendency towards a unified,
interdisciplinary vision of reality that has been appearing in the scientific
114 From Observations to Simulations

world during the last decades. But the introduction of simulation models
had other consequences as well!
In actual fact, the use of these models in natural sciences and in
sciences of mankind has led to changes in scientific practice. The ease
with which scientists could manipulate the elements of the model led
them to regard simulation models as virtual experimental laboratories
where they could carry out "experiments" that were not possible in
reality. We have already explained, for instance, how difficult it is to
reduce the complexity of a system such as the atmosphere to the narrow
space of a laboratory; we should also consider that it is impossible to
carry out real experiments for very long periods, like those that are
characteristic of climatology. Well, in the virtual laboratory supplied by a
simulation model, space and time can be expanded at will: the speed of
calculation of a computer makes it possible to simulate experiments that
in reality would have to be prolonged for tens or hundreds of years.
If we add to this the capability of a simulation model to perform a
"synthesis" through the recomposition of the phenomena — i.e. the fact
that it can realistically account for the complexity and manifold
interactions and feedbacks of a complex system — we can understand
why these models are becoming essential tools in modern scientific
practice.
Finally, our remarks up to now have made it clear that simulation
models are useful for the study of complex systems. The structure of
simulation models is usually influenced by a tendency that is extensively
present in the history of science and that we have already discussed:
reliance on a reconstruction of the behaviour of a macroscopic system
that is based on the composition of elements and interactions that are
regarded as fundamental and have been validated separately in the
experimental activity of the real laboratory. If these basic elements
belong to a different (lower) level with respect to the phenomena to be
studied, this amounts to a relapse into a form of reductionism; otherwise,
it is nothing but reliance on the possibility of reconstructing the
complexity of reality on the basis of simpler phenomena.
In the next two chapters we will examine some concrete applications
of these simulation models within the scope of this book. We will realise,
however, that from these applications there emerge some important
 Simulation Models 115

conceptual aspects that need to be discussed. In the last chapter we will

concisely return to an examination of the methods of present-day model
making and of possible alternative approaches that may be conceptually
interesting and practically useful within the sphere of the disciplines that
study complex systems.
 Chapter 7

Meteorological Models

In the previous chapter, we introduced the concept of simulation model.

We particularly underlined the peculiarities that suggest that the use of
these models in disciplines characterised by the study of complex
systems leads to a change of methodological paradigm in the cognitive
approach to these systems. Now we will move on to a more concrete
level of discussion, analysing the structure of the models devised for
weather forecasting, and examining the impact of their application.
At the same time, however, within this very cogent and particular
applicative framework, we will see that there emerge some problems
(and relative attempts at a solution) that have a much more general
significance and will enable us to discuss further changes of paradigm in
the scientific approach to the study of complex systems.

7.1 The "Perception" of the Weather Forecasting Activity

In order to concisely evaluate how important the weather is in everyday

life, it may be useful to notice how much this subject is discussed by
ordinary people, that is by people who do not have any specific
professional interest in weather data and forecasts. From this point of
view — apart from cliches, such as the one that in the British Isles this is
everybody's main topic of conversation — it is evident that in the
countries of Northern Europe and the United States "meteorological
culture" is much more developed than in the temperate-climate countries
of the Mediterranean area. This is due to the fact that in the former
countries the weather has a greater impact on everyday life, because
there is a more frequent occurrence of intense cold, snow storms, frost,

117
118 From Observations to Simulations

prolonged rain, and, in the United States, of violent phenomena such as

hurricanes and tornadoes.
In the present-day society of information, a more quantitative index
can be supplied by the presence of weather information on television. In
all the northern countries this information is usually given quite often
and in a very invasive way (rather like commercials). There even exists a
channel, the Weather Channel, that is completely dedicated to world-
wide weather information. On the contrary, in the countries of the
Mediterranean area, including Italy, apart from occasions of particular
cold or heat waves, in which the newscasts take care to underline the
minimum or maximum temperatures in certain locations and to supply
forecasts for the next few days, information about the weather is
restricted to the regular forecast features.
Despite this limited presence of meteorologists on television, during
the last few years, in the countries of the Mediterranean area, there has
been an increase in the popularity of these features: their audience and
viewing figures have risen almost to the levels of those of Saturday-night
shows featuring popular showgirls. It is not clear why this has happened:
some people believe that the cause is modern society's greater need to
plan leisure time, others the fact that during the last few years the climate
seems to have changed and there has been an increase in the frequency of
extreme phenomena — such as cold or heat waves, violent storms and
floods — from which people have to defend themselves. Other people
believe that the meteorological features today are watched more because
the weather forecasts have at last acquired a high degree of reliability.
On the one hand, the recognition of a high statistical reliability of the
weather forecasts shows that there has been an increase in the credit of
meteorology as a scientific discipline that has been making advances
during the last few years1. On the other hand, this increase in credit is not
always combined with an increase in meteorological culture. If one asks
a person why the quality of the weather forecasts has improved, in most

In this context it is significant to notice that up to a few years ago, for instance, in the
makeup of Italian daily newspapers, the weather forecasts were always placed near the
horoscopes, in an obnoxious divinatory couple, often combined also with the lottery draw
news. Now at last some newspapers have made the decision of giving the weather
column its own space, raising it to a rank of greater credibility.
 Meteorological Models 119

cases one will obtain a reply like "Because now with satellites everything
is easier...". Many people actually believe that the forecasts are produced
by the satellites. The reader who has patiently reached this point in the
book knows, on the contrary, that satellites are an instrument of
observation, whose data can be used for a better definition of the state of
the atmosphere system at a certain instant, but that they do not have any
value for prediction purposes, at least on the time scales involved in the
forecasts for the general public. It is clear that the misunderstanding in
which laymen have fallen stems from what they see on television, where
sequences of images from a satellite are often shown: they reproduce the
movements of clouds during the last few hours over a certain territory.
Though over a very short period (a few hours) these movements can be
projected into the future, a forecast of the evolution of these clouds
(formation of new ones or dissolution of old ones) and of their further
movements requires knowledge of the evolutionary laws of the system
and their interaction, therefore a model.

7.2 The Heart of a Meteorological Model: Primitive Equations and

Their Numerical Solution

Here too, as in the previous chapters, we will refrain from adopting a

historical perspective on the development of models in the past. We will
only mention that the first theoretical forecast (which was worked out by
discretising the meteorological variables at an initial instant and some
basic equations, which were then solved "by hand" step by step) was
published by Lewis F. Richardson in 1922: it was a 24-hour forecast,
which turned out to be decidedly wrong. Almost 30 years elapsed before
this early attempt was resumed with some hope of obtaining a more
encouraging result, and with more reasonable calculation times, since in
the meantime the American ENIAC, the first computer on which the
calculations of a meteorological model were performed, had become
available. Richardson's unsuccessful attempt allowed scientists to
understand that it was not sufficient to know the basic equations that
govern the physics of the atmosphere, because it was necessary to
simplify them, in particular by eliminating the phenomena, such as the
120 From Observations to Simulations

propagation of sound waves, that could impair the correctness of the

forecast of the future behaviour of the atmospheric flow. So, for at least
twenty years, very simplified models that used extremely approximate
equations were worked out. Finally, starting from the nineteen-seventies,
there appeared the first prototypes of the present-day models, the so-
called "primitive-equation models". In this book we will describe only
these models.
As we have already explained in the previous chapter, the central core
of a simulation model consists of the equations that represent our
theoretical knowledge about the system we are going to study, in some
cases also in its temporal evolution. In order to describe highly
interacting systems such as the atmosphere, where various causes concur
in producing a single effect and there are feedbacks on the causes, it is
necessary to couple several equations in a single system to be solved
numerically. The variables that are contained in the equations, and whose
future values are to be forecast, are physical quantities that can be
measured in the real system. If, as in meteorological models, the
equations are based on partial derivatives and include temporal
evolution, the second element that is needed is the determination of the
variables at the initial instant in the whole spatial domain considered by
the model. Finally, the boundary conditions (e.g. the state of the ground
in relation to time) and the external forcing factors (e.g. the solar
radiation cycle) temporarily complete the ingredients of this recipe for a
simulation. In actual fact, we will see later on that, in order to hope to
achieve a realistic simulation of the physical evolution of the
atmosphere, it is necessary to add at least another fundamental element.
A preliminary, essential remark must be made at once: standard
meteorological models, at least the ones that aim at producing a correct
medium-range forecast (up to 7-10 days), are characterised by a
dynamical treatment only of the atmosphere subsystem within the
broader Earth system discussed in Chapter 4. In practical terms, this
means that the equations refer only to the dynamics of the atmosphere,
and that the interaction with the other subsystems is expressed as an
interaction with the external environment through the consideration of
forcing factors and boundary conditions, even if they are evolving and
 Meteorological Models 121

partly depend on what takes place in the atmosphere system2. This is due
to the fact that, as a rule, the evolutionary dynamics of the other
subsystems that interact with the atmosphere produces changes in the
latter that are slow in comparison with the time scales involved here. In a
model whose purpose is to deal dynamically with the changes that each
subsystem of the Earth system causes in the other subsystems, and with
the resulting feedbacks, it is necessary to express the individual
dynamics in terms of equations, and then to couple these equations in
systems that can be solved numerically. We will briefly return to this
topic in the next chapter, because in climatic models a more decidedly
dynamical approach is required.
For the time being, it is interesting to concisely explain the basic
equations (now usually called primitive equations) that are introduced
into a meteorological model for a weather forecast. Adopting the
principle of system decomposability, which leads to the determination of
the individual equations, the complexity of the atmosphere system is
usually reconstructed by means of six primitive equations: the first two,
called "diagnostic equations", are laws like those we have called balance
or coexistence laws in Chapter 4, and supply the relationship between
different variables at the same instant; the other four equations, called
"prognostic equations", depend on time and supply the evolution of the
values of the involved variables. Let us briefly examine these equations.
Equation of state of gases: supplies the link between pressure,
density and temperature in an air mass at the same instant.
Hydrostatic equation: this equation is obtained, by means of a scale
analysis3, from the equation of vertical motion, and supplies the
approximate relationship between the density of the air and the
change of pressure in relation to height. As we shall see, it is

2
For instance, if in the world of the model there occurs a snowfall on a previously bare
ground, from that moment on, and for a period that depends on the amount of snow that
has fallen and on the temperature of the air, the boundary condition that determines the
state of the ground is changed, so as to allow the model to correctly simulate the decrease
in absorption of solar radiation on that part of the territory.
3
The scale analysis method makes it possible to determine the relative importance (in
terms of orders of magnitude) of the individual terms of which the equation is formed. In
actual fact, as a result of a scale analysis, some of these terms are disregarded and the
situation is simplified.
122 From Observations to Simulations

considered valid for models where the spatial resolution is not very
high.
Continuity equation: it ensures the conservation of mass. Once a
volume that delimits a portion of air has been determined, for
instance in the case of the hypothesis of the perfect non-
compressibility of the fluid, this equation ensures that if a certain
amount of air gets into the volume under consideration, the same
amount gets out.
Navier-Stokes equation: this is the equation of motion in fluid
dynamics for the horizontal components of wind: from the vertical
part of the complete three-dimensional equation, after a scale
analysis, the hydrostatic equation is obtained.
First law of thermodynamics: it is an evolutionary equation for
temperature that allows for the thermodynamic processes in the
atmosphere, such as the adiabatic warming or cooling of the air due
to vertical movements, the release of latent heat in changes of state,
etc.
Continuity equation for the water (liquid, solid or vapour) contained
in the atmosphere: it is an evolutionary equation that obviously
allows for all the processes connected to the changes of state, i.e.
evaporation, condensation, fusion, solidification, sublimation and
rime formation.
As we have already mentioned, the computer-aided solution of
equations such as those we have just described involves numerical
techniques of step-by-step solution of the equations. If there existed
analytical techniques for arriving at general solutions or even only for
solving significant specific cases, it might be possible, perhaps, to use the
computer with a method more similar to that of a classical
mathematician: nowadays symbolic-calculus software packages that
solve certain equations analytically are available. Unfortunately the
mathematical knowledge about the properties of these equations is still
limited, and the dream of an analytical solution is still far off: to this day
there does not exist a general theorem of existence and uniqueness for
the solutions of the Navier-Stokes equation.
In a situation like this, the successful strategy seems to be the one that
had been outlined by Richardson as early as 1922, that is the
 Meteorological Models 123

replacement, in the equations, of the derivatives with the finite

differences between points at a certain spatial distance (in the case of
derivatives with respect to space) or at a certain time lapse (in the case of
derivatives with respect to time). The concept of the derivative of a
certain quantity is a sort of generalisation precisely of the concept of the
increment or decrement of that quantity in a time or space unit, because it
determines the rate of difference of the quantity for infinitely small
spatial distances or time lapses. Thus, by replacing the derivatives with
finite differences, we achieve a "discretisation" of the equations under
consideration. This means, in particular, that these values will have to be
calculated on a finite number of points in space and on a sequence of
discrete temporal steps. Thus, in the world of the model, the space
continuum and the time continuum that characterise the description of
spatial and temporal evolution in physics are replaced by a three-
dimensional spatial "lattice" and a sequence of discrete temporal steps.
So, in order to concretely bring about this "discretisation" of the
equations, it is necessary to define the so-called "grid", i.e. the spatial
lattice on whose points the values of the variables present in the equation
are to be calculated in the immediate future (at the next discrete time
step). Proceeding step by step, it is then possible to obtain the values of
these variables on all the points of the grid for the forecasting period to
be considered4.
Obviously there exist several types of grids: a natural grid, for
instance, is one that is defined by the points of intersection between
meridians and parallels on the Earth's surface; though appearing to be
completely natural, this grid has the negative characteristic that its points
are very close to each other towards the poles, further apart at medium
latitudes and even more sparse near the equator. The grids that are
chosen usually have a more uniform, equidistant distribution of points.

4
For the sake of completeness, and only for the interested reader, it is worthwhile to
mention the fact that this scheme based on finite differences is not the only one that
allows a discretisation of the equations: there exist also some methods, called "spectral
methods", in which the spatial changes in the variables are expressed as components in
the Fourier space or in spherical harmonics. The results of the two methods are quite
comparable.
124 From Observations to Simulations

Moreover, the reader should notice that the grids are not two-
dimensional, but three-dimensional, since they extend upwards in the
atmosphere, with several vertical levels and a series of concentric two-
dimensional grids around the Earth.
Likewise there exist several techniques for the numerical solution of
the equations on the grid that has been chosen. These techniques tend to
cut down to a minimum the approximations inherent in the discretisation,
and to elude certain problems stemming from the numerical solution of
the equations. Nowadays there exists an entire branch of mathematics,
numerical analysis, that studies these problems. Here, obviously, we
cannot dwell on this topic5.
Returning to our grid, we must remark, that, as a rule, the thicker it is,
the closer the finite differences come to the values of the derivatives of
the space-time continuum. The problem that arises at once, however, is
that of the calculation times: if a three-dimensional grid exceeds a certain
degree of thickness, the number of calculations to be carried out by the
model increases excessively. On the other hand, our most pressing need
when using a meteorological model is to obtain results in a time that is
practical for their use: obviously if a model takes two days to produce a
forecast for the day after, the result, even if it is correct, becomes utterly
useless. Therefore the lower limit for the distance between the meshes of
the grid is shifted, basically following the performance of the new
computers, which are increasingly swift. Nowadays, for a global model
(whose domain of numerical integration is the whole Earth), we stop at a
typical distance between two points of the grid (the so-called "grid
spacing") that is about 50 Km horizontally, while vertically there are
about 60 levels in the first 60 Km of the atmosphere (but these levels are
not equidistant, because they are closer to each other in the lowest layers,
less and less close as the altitude increases, and very distant from each

5
There are no texts in which the techniques of numerical integration, and the dynamic and
physical structure of meteorological models, are explained in a thorough but accessible
way. For the reader who is interested just the same, we recommend Krishnamurti and
Bounoua(1996).
 Meteorological Models 125

other beyond the tropopause)6. Because of this discretisation, each point

of the grid essentially identifies a cell centred on that point, for instance a
parallelepiped whose horizontal sides are 50 Km and whose vertical
sides are a few hundreds of metres longer.
Since the physical quantities of the future are calculated only at the
intersections of the lattice, the grid spacing essentially determines the
resolution of the model: the variations that take place in any variable
between two adjacent points of the grid cannot be represented by the
model if their values exceed the range of the values present on those two
points. It is as if we tried to reconstruct a sinusoid y = sin x with a 2n
period by means of a number of discrete points whose distance between
each other exceeds n on axis x: in this case it is not possible to correctly
represent the shape of the sinusoid. In other words, the discrete
numerical solution of a model necessarily "averages" the real physical
behaviour of the system under examination.
The present limitations in the spatial resolution of a global model are
partly overcome if we focus on a smaller portion of territory: there exist
some models, called "limited-area models", in which the numerical
solution of the equations is limited to a less extensive area, for instance
the Euro-Atlantic region of the northern hemisphere. In these cases, it is
possible to obtain a thicker grid while preserving the same order of
magnitude in the number of grid points. Obviously these models need the
conditions at the boundaries of their area of interest, and these conditions
can be supplied only by the evolution forecast through a global model:
for this reason, limited-area models cannot be regarded as self-sufficient.
In these models it is possible to achieve a horizontal spatial resolution
even only of a few Km. Beneath 10 Km, particularly in cases where the
ground (which is the lower boundary of the atmosphere) is characterised
by complex orographical features, the validity of the hydrostatic
approximation disappears, so it is necessary to use a Navier-Stokes

6
The reader is reminded that the tropopause is the area where the temperature, which
decreases with height in the areas below, starts increasing with height. The tropopause is
situated at an altitude that depends on the latitude, season and meteorological
configuration: at medium latitudes it is, on the average, between 11 and 14 Km above sea
level.
126 From Observations to Simulations

equation in the model also for the vertical component. These models are
called "non-hydrostatic models", and will not be examined in this book.

7.3 Physical Parameterisations

The previous remarks led us to understand that the discrete numerical

solution of the equations of a model based on a grid and characterised by
a certain grid spacing necessarily results in our obtaining an averaged
description of the real evolution of the atmosphere system. This is a
serious problem when there is the need to draw a distinction between the
weather in a certain point and that in another one separated from it by a
horizontal distance smaller than the grid spacing; this difficulty,
however, is obviously present in any model that simulates reality in a
discretised manner on a computer. The real physical problem that must
be taken into account is a result, instead, of the fact that in meteorology
there exist phenomena on very small scales — smaller than the grid
spacing — that heavily influence the value of the quantities present in the
equations solved by the model on the points of the grid. Basically these
phenomena are not "seen" by the system of discretised equations, so, if
everything were limited to a solution of this system on the given grid, the
description of the physical atmosphere system would turn out to be
inadequate, and the validity of the forecast would be greatly impaired.
In order to allow the reader to understand that these small-scale
phenomena do not pertain to marginal aspects of the weather as it is
usually perceived, it is sufficient to point out that one of them is the
presence of thunderstorm clouds: the diameter of a thunderstorm cloud
leading to a typical summer thunder shower is always less than 50 Km,
and often even less than the 10 Km that have been taken as the lower
limit for the grid spacing of a non-hydrostatic model. A model that does
not "see" these clouds obviously would result in a completely wrong
forecast of the amount of precipitation: it is known that the most intense
precipitation comes precisely from convective phenomena.
But this is not the whole story! There is a further complication: the
presence of thunderstorm clouds, which are very tall, obviously also
perturbs the value of other variables "handled" by the model with its
 Meteorological Models 127

dynamic equations. For instance, the quantity of water (in its three states
of aggregation) present in the cells that contain the thunderstorm cloud
and are identified by the neighbourhood of the relative grid points in the
vertical direction is considerably altered on the entire column. Moreover,
the strong upward currents present within a cloud lead to a vertical
conveyance of matter and humidity and to a vertical redistribution of the
temperature. More specifically, there appears a vertical thermal profile
typical of an air mass that rises adiabatically in the atmosphere, with
phase transitions within it from water vapour to liquid water and ice: the
latent heat that is released in these transitions causes the temperature
within the cloud to be usually higher than that of the surrounding air.
Finally, the presence of a thunderstorm cloud results in a decided change
in the balance between the incoming radiation and the outgoing one in
the cells under consideration: sunlight is intercepted, and so is the long-
wave radiation coming out of the ground (this too affects the temperature
values).
The example mentioned here of the possible presence of
thunderstorm clouds highlights the existence of an intense evolutionary
phenomenon on a scale that is not solved by the equations of the model
with the selected grid spacing. We have already explained that this
deficiency is serious from the viewpoint of forecasting, both because of
its immediate consequences (unforeseen strong precipitation) and
because of the perturbations in the other variables that the model causes
to evolve independently on the grid through the solution of the equations.
We will not dwell on other small-scale phenomena that have a similar
effect (though perhaps in a less quantitatively evident way, as in the case
of the mechanical turbulence in the low layers); it is clear, in any case,
that there is the need to express the action due to these small-scale
processes in relation to the quantities on the scale of the grid that are
dynamically treated by the model.
The strategy adopted for this purpose is that of creating routines,
program modules called "physical parameterisation modules", that
describe the evolution of these small-scale phenomena. Usually we have
separate but interacting modules for convection, radiation, turbulence,
etc.; they are used, every now and then, to "adjust" the values of the
variables, which, in the meantime, evolve separately through the solution
128 From Observations to Simulations

of the equations. The use of this interacting set of dynamical equations

and physical parameterisations is the only method with which we can
hope to achieve a correct forecast of the evolution of the characteristics
of the atmosphere system.

7.4 Determination of Initial State and Analysis Procedure

By considering the primitive equations and the physical parameterisation

schemes, both solved in a discrete manner on a three-dimensional grid,
and with a discrete time step, we have introduced our theoretical
knowledge of the atmosphere system into a model that simulates it. We
have already pointed out, moreover, that external forcing factors and
evolutionary boundary conditions make it possible to describe the
relationships between this system and the other systems that define its
external environment (oceans, lithosphere, biosphere, etc.). After the
theoretical framework has been thus defined, the model can be "linked"
to the real system by means of the variables present in the equations,
which have a one-to-one correspondence with certain physical quantities
in the atmosphere.
At this point, if we wish to achieve a simulation of the behaviour of
the atmosphere in a situation that is actually present in nature, all we
need is the determination of the initial state of the atmosphere through
the measurement of the physical quantities introduced in the model as
variables. Obviously, it will be necessary to define the value of these
variables at the initial instant in all the points of the grid that
characterises the model, so as to supply a complete initial condition for
the numerical solution of the theoretical scheme. If we consider that the
meteorological measurements performed on the planet are far from being
arranged on a three-dimensional grid7, we can understand that the
problem of supplying the model with a complete initial condition is far
from easy to solve. In actual fact, any reputable meteorological office
employs a considerable part of its researchers in the optimisation of the

7
The reader will recall, for instance, Plate 1, discussed in Chapter 2, which showed the
highly unhomogeneous horizontal distribution of conventional surface-based observa-
tions and the shortage of observations on the oceans and African continent.
 Meteorological Models 129

methods (and operating process) for obtaining an estimate of the initial

state of the atmosphere in the form of initial data on the grid of the
model.
The first solution that enters our mind when we are looking for the
value of a variable in a point where its measurement is not available is to
obtain this value by means of the interpolation of the known surrounding
values. Actually, given a spatial distribution of values of a certain
quantity (acquired, for instance, by means of measurements), there now
exist some software packages that can interpolate the value of this
quantity onto a point where it is unknown, provided the number of
available measurements is sufficient to ensure the reliability of this
reconstruction. Can we apply this method to the determination of the
values of all the variables of the model on all the points of the grid at the
initial instant of the simulation? Certainly! But if we did it, we would
immediately realise that the results are quite meagre from the physical
point of view.
The key for understanding the reason for this lies precisely in the
physical nature of our data, which are representative of a system
characterised by a certain balance or coexistence between several
quantities at the same instant and by a certain consistency between
spatially close points. In fact, though the data that come from a
meteorological office with gross coding or transmission errors can be
corrected by a quality check that might be introduced also in a simple
interpolation software package, discovering systematic errors and more
subtle ones (instrumental or due to other causes), or obtaining a
physically consistent interpolation requires a different treatment of the
data — a treatment that allows for the physical relationships that
interlink different quantities.
Moreover, we must consider that the amount of available data is often
not sufficient for a satisfactory interpolation: while for conventional
observations we have just mentioned the presence of gaps in the
observational network, for satellite observations we have stressed (in
Chapter 2) the shortage of points in the vertical direction in the SATOB
messages and a similar problem also in the SATEM ones, because of the
highly averaged measurements that are characteristic of the vertical
soundings performed by the TOVS. Furthermore, the data are actually
130 From Observations to Simulations

completed by those coming from the polar satellites, but the latter are
usually not synchronised with the conventional data or with those
coming from geostationary satellites. In order to be able to somehow
compare these data with those of the synoptic hours, we will inevitably
need a temporal "consistency", i.e. evolution equations.
The situation we have thus described leads us, therefore, to drop the
idea of a simple mathematical interpolation of the data, quantity by
quantity, and to prefer a more dynamical way of dealing with this
problem. An approach that might seem natural is that of correcting the
results of the interpolation by using the physical constraints we know.
Let us now briefly discuss what might be a typical strategy for tackling
the problem of determining the values of the variables at the initial
instant on all the points of the grid.
On the whole, if we consider the importance of the physical
constraints in the reconstruction of the discrete spatial distribution on the
points of the grid, we are led to completely overturn the apparently
natural approach to the problem.
In actual fact, instead of starting from a mathematical interpolation
and correcting its results by applying known physical laws, we prefer to
start from the values of the variables, fixed by means of the forecast
results coming from the previous run of the model (6 or 12 hours before).
This way, we obtain a first guess for the fields8 of all the variables at the
initial instant of the new run of the model. Only at this point the
observations come into play, obviously in the role of correctors for the
fields that have been determined with the first guess.
This rather original approach makes it possible to allow fully for the
physical laws of the atmosphere system, both the balance ones and the
evolutionary ones, which are all included in the model. Moreover it
makes it possible to obtain more reliable initial conditions on the areas
where there are few observations and where their simple mathematical
interpolation would give rise to serious problems. Finally, since the

8
Without referring specifically to the strict mathematical definition, within our context of
discrete model making, a field is a set of values of variables on all the points of the grid:
it is a scalar field if the variable under consideration is a quantity that is defined by a
single numerical value (a scalar), or a vectorial field if the field variable is defined by a
vector.
 Meteorological Models 131

obtainment of the first guess is based on the use of an evolutionary

model, which theoretically can supply the fields of the variables at each
temporal step, it is possible also to retrieve the correction contribution
relative to the observational data that have not been read precisely at
instant t0 = 0 of the new run of the model, but have been included, for
instance, in the interval t0 - 3 hours < t < t0 + 3 hours.
The procedure we have thus briefly defined is called "analysis". More
specifically, a very recent achievement of research in this area is the
possibility of including in the analysis observational data coming from
measurements that are not synchronous with the initial instant. In this
case, rather sophisticated mathematical techniques are used9, and the
analysis is called "four-dimensional", because, besides the three spatial
dimensions, the temporal one is also included, in order to account for the
inflow of data relative to instants different from to.

7.5 The Products of a Meteorological Model

Now that we have discussed all the ingredients of the "recipe" for a
weather forecast simulation, we will proceed to the executive stage of
this recipe. The analysis procedure supplies the initial data to the model,
which starts the simulated evolution with time. After the very first period
(a few hours of simulated time), during which, despite the physical
approach of the analysis, there may still be the influence of a few
settling-down problems in the balancing between the fields of the various
quantities10, we will begin to obtain the first reliable, consistent results.
The values of the variables provided at all the points of the grid are
stored in suitable memory areas of the computer in which the model is
"running", at fixed intervals and throughout the simulated-time
predefined for the run of the model (usually 10 days for a global model
and 3 days for a limited-area model).

9
For the reader who is interested, we can summarise by stating that it is essentially a
matter of solving minimisation problems by means of variational calculus and of the
construction of an adjoint representation of the linearised model.
10
This problem is called "spin-up effect".
132 From Observations to Simulations

The files containing the values of all these variables on all the vertical
levels of the model remain usually in the archive of the data processing
centre of the meteorological service or institution that develops and
operates the model, and are available, if necessary, for subsequent
studies. What reaches the users is something different. The technical
users (i.e. the people who need these data in order to process them
further, both for scientific and for purely applicative purposes) receive
digital, coded data1 y on certain vertical levels, for the required areas and
forecast times (usually every 6 hours of simulated time). The other users
normally receive graphic elaborations that supply a selection of the same
pieces of information, but in a form that can be directly understood by
anybody, because they are usually displayed in weather maps that can be
interpreted more or less readily.
Plate 8 shows an example of these weather maps, also called
"prognostic charts", drawn from the global-circulation model of the
European Centre for Medium-range Weather Forecasts (ECMWF). This
figure represents the forecasts for 4 February 2003 at 12 GMT, obtained
by the operational run of the model after 36 hours of simulated time from
the date of the initial analysis. This forecast is illustrated here by
displaying the situation predicted for that date and time, by means of an
upper-level chart (a), a surface chart (b), a cloudiness chart (c) and a
precipitation chart (d). We will now concisely analyse some traits of
these products, which obviously result from graphic-display elaborations
of the data provided by the model and distributed in a discrete manner on
the grid.
The first chart, shown in Plate 8a, describes the situation forecast at
an altitude approximately at the middle of the troposphere. In actual fact,
the processing of upper-level charts is influenced by an inheritance due
to the fact that these charts were originally produced for flight aiding
purposes. Since the altimeters of aircrafts are basically barometers, i.e.
instruments that measure the atmospheric pressure, stabilising the height
of an aircraft during the cruising stage means flying the aircraft at a fixed
pressure and not at a certain altitude above sea level. So in upper-level
charts, instead of placing ourselves at a certain altitude and ascertaining

"in order to minimise file sizes and transmission times.

 Meteorological Models 133

what pressure there is as we move horizontally, we place ourselves at a

certain pressure and display the altitude at which this pressure is present
over the area under consideration12. So each blue line in this figure
represents a line that joins the points where the pressure of 500 hPa is at
the same altitude (represented by a number that indicates tens of metres
of height). Likewise, the red broken lines identify the points that are
characterised by the same temperature at the pressure of 500 hPa. A vast
altitude minimum is visible: it is forecast to be centred over Denmark.
The next chart, shown in Plate 8b, represents the atmospheric
pressure read at sea level (by means of blue lines that connect the equal
pressure points, called isobars). The forecast field of the temperature at
850 hPa (about 1,500 metres) is also displayed by means of red broken
lines, called isotherms, that connect points of equal temperature. There is
a forecast area of low pressure that involves the whole of central Europe
and extends towards Italy, where the pressure is particularly low in the
central and northern parts of the country.
These first two charts are a graphic representation of the fields
forecast for basic variables that are supplied to the model as initial
conditions, then treated dynamically by the model. The actual use of
these charts requires a certain forecasting experience. There are,
moreover, charts that refer to variables that have been reconstructed or
somehow drawn from the basic ones. For instance, from the temporal
trend of the liquid water or ice content it is possible to infer the presence
or absence of clouds on the territory (if necessary, also determining their
altitude): see the example in Plate 8c, which zooms onto Italy and the
neighbouring countries, and represents the cloudiness forecast in terms of
low cloud cover13. Moreover, the amount of liquid or frozen water that
falls to the ground from clouds supplies a precipitation forecast: see

12
This is facilitated by the fact that in meteorological models the vertical coordinate is
often expressed in terms of pressure, because this results in a simplification of the form of
the equations.
13
In Plate 8c the forecast low-cloud cover is indicated in the oktas scale, at whose ends 0
oktas means clear sky and 8 oktas means completely overcast sky. Low clouds are
characterised by the fact that their base is at an altitude of less than 2 Km and are very
significant from a meteorological point of view: for instance, they often lead to
precipitation.
134 From Observations to Simulations

Plate 8d. Contrary to the first three charts, which represent "snapshots"
of the fields expected at the time of the forecast, these charts represent
the distribution of the precipitation expected to fall to the ground in the
12 hours of simulated time that precede the time limit of 12 GMT of 4
February 2003. Obviously these last two charts are much more self-
explanatory than the previous ones, and can therefore be utilised more
directly (though, unlike the layman, a careful forecaster who is
acquainted with the characteristics of the model also knows how to
correct possible defects in the forecast of these quantities).

7.6 The Emergence of Deterministic Chaos and Ensemble

Integrations

We have just shown an example of the graphic display of the fields

provided by a meteorological model. Obviously a model is regarded as
valid if it correctly forecasts what will eventually take place in the reality
of the atmosphere; this validation can be carried out only by means of a
comparison with a posteriori analyses of data obtained through the
observation of the real system. In practice, checking the performance of a
model is an important aspect of the activity of a meteorological office,
because it enables the researchers to understand the effects of possible
changes in the model, in terms of the accuracy of the simulation and
therefore of the validity of the forecast. This validation takes place both
as a daily operation carried out on the new forecasts that are produced
and in an analysis of the behaviour of the model in case studies
characterised by particularly significant meteorological situations.
As we already stated in the previous chapter, when we leave the
controlled conditions of the laboratory and attempt to reconstruct the
complexity of reality, we are inevitably compelled to choose the
theoretical elements we regard as fundamental and to leave out the ones
we regard as secondary and less important for the dynamics of the
system under consideration. Once we have worked out a simplified
model, obviously it can be improved by the addition of further theoretical
elements, so we may be led to believe that, through successive
improvements, it is possible to eventually achieve a reproduction of all
 Meteorological Models 135

the details of the behaviour of the real system. Obviously a project that is
so ambitious can be based only on confidence in the validity of the
theoretical scheme, which in our case is formed of diagnostic and
prognostic differential equations and parameterisation schemes and on
the possibility of rendering our theoretical scheme complete and
univocal.
We may be doubtful about the possibility of achieving the univocality
of the scheme, because in our case it would be a matter of having to
determine in a univocal manner (obviously on the basis of theoretical
considerations) the value of the parameters that currently determine the
coupling of the various parameterisation schemes in the models and
therefore the intensity of the feedbacks. At present these parameters are
fixed by means of an activity that might be called artisanal, balancing
them, i.e. mutually "fine tuning" them, as we do when we fine tune a
television set in order to centre the frequency of a channel and obtain
optimum vision. In any case, at least for the time being, we will pretend
that this problem does not exist, and assume that it is possible to work
out a complete and univocal model, a "perfect" model that can accurately
reproduce the future behaviour of the atmosphere system.
If we could have a model whose theoretical scheme perfectly
reproduces the physics of the atmosphere system, the only element of
uncertainty in the future forecasts would stem from approximations and
errors we might make in the determination of the initial state. An error
here would obviously propagate, step by step, during the discrete
numerical integration. It remains to be explained how this takes place.
Present-day models, like the earlier ones, are characterised by
forecasting errors in the various estimated quantities, and these errors are
usually the greater the longer the simulation time. All this is rather
reasonable, because, since the model is not perfect, it leads to
inaccuracies in the forecast, which, step by step in the discrete numerical
integration, will propagate and get amplified in the course of time. These
errors cannot be distinguished a priori from the ones that stem from
inaccuracies in the determination of the initial conditions. In this
framework, however, it is interesting to point out that the amplification
of the errors is not linear (i.e. practically constant), but is affected by
sudden upsurges that, after a certain time lapse, cause the behaviour
136 From Observations to Simulations

forecast by the model to be completely different from the real behaviour

of the atmosphere system. On what does this fact depend?
In order to study this simulated behaviour, and to hope to be able to
identify the elements that are important for understanding it, we must
separate the contributions due to the model from those due to the initial
situation. More specifically, if we temporarily assume that our model
perfectly simulates the behaviour of the atmosphere, a way of imitating
the propagation of errors due to an incorrect estimate of the initial
situation is that of starting two runs of the model with two slightly
different initial conditions: one of the runs simulates the behaviour of the
atmosphere on the basis of a correct estimate of initial conditions, the
other does the same thing on the basis of a somehow incorrect estimate.
What this simulation experiment reveals is that the error thus
introduced in the initial condition is amplified in a manner that is not
linear or slow and gradual, but "exponential", with a sudden upsurge
after a certain time lapse, exactly as in the comparison between our
working simulations and the behaviour of the real atmosphere. On the
one hand this observation may suggest that our way of representing the
evolution of the atmosphere is basically correct, but on the other hand it
shows that our representation of the atmosphere system by means of a
system of non-linear differential equations and physical
parameterisations is extremely sensitive to initial conditions. In these
theoretical schemes, if we start from two initial conditions that are even
highly similar, after a certain simulated time lapse the solutions will
diverge: this is a phenomenon called "deterministic chaos"14.
Since, because of the discrete, unhomogeneous and partly
discontinuous nature of atmosphere monitoring, the initial condition for
the runs of the model is always identified (through the analysis process)
in an approximate way, we can endeavour to minimise this error but
cannot ever eliminate it. So even if we had a perfect model, the
approximate estimate of the initial state would lead, after a certain time
lapse, to the production of forecasts that considerably differ from the

Though the emergence of deterministic chaos is typical, as a rule, of systems of non-

linear equations, it was discovered precisely in the meteorological field, by Edward N.
Lorenz. He reported his findings in a famous article of 1963 and more recently produced
an interesting account in Lorenz (1994).
 Meteorological Models 137

behaviour of the real system. This limit, which obviously is intrinsic to

the theoretical scheme that has been adopted, leads to the recognition of
a maximum theoretical predictability period beyond which it is not
possible to produce forecasts with this type of model: the period is
approximately 10 to 15 days, though in actual fact the model starts
performing rather badly even before.
Once this limitation, which is completely inherent in our theoretical
scheme, has been recognised, the next goal is to find the way to reduce
its damage to the forecasting activity. This will be the topic of the next
pages. First, however, it is worthwhile to dwell somewhat on the
absolutely fundamental significance of this discovery.
Classical physics is entirely based on a deterministic description of
the evolution of dynamical systems, inasmuch as its time-dependent
equations can be solved in the future and can yield a univocal solution if
the initial conditions from which they start are univocal. Some
philosophers, whose most paradigmatic representative was Laplace,
adopted the philosophical stance of defending the most absolute
determinism, meaning by this that "if an intellect, at a certain instant,
knew all the forces that animate nature and the mutual positions of the
beings that compose it, and if it were so vast as to submit these data to an
analysis, it would condense into a single formula the movements of the
largest bodies of the Universe and that of the lightest atoms: nothing
would be uncertain for it, and both the future and the past would be
present before its eyes"15. Laplace clearly went beyond determinism
when he asserted that all natural phenomena could be reduced to the
(deterministic) laws of mechanics.
In a stance like this one, the concept of determinism leads directly to
that of the univocal prediction of the behaviour of a system. Apart from
previous evidences that had already undermined this concept — such as
the discovery, made by Poincare, that the three-body problem cannot be
solved exactly, and the discovery of quantum phenomena — nowadays,
with the emergence of chaos in systems of non-linear deterministic
equations of classical physics, it is being shown that in systems that are
sensitive to initial conditions a deterministic approach cannot ensure a

'Laplace (1820).
138 From Observations to Simulations

reliable evolutionary uniqueness in the future. So will we have to adopt a

different forecasting strategy? After all, Laplace himself, being aware of
the finiteness of human knowledge and of the difficulty of completing
his own deterministic-mechanistic programme, was the creator of the
probability theory. Let us now endeavour to picture our situation with the
help of classical methods.
If we consider a system of interacting particles (without any internal
structure), the state of this system at a certain instant is defined when we
know the position and velocity of all its particles. If, in a very simplified
situation, our system were formed of a single particle constrained to
move in a definite direction, we might display its state at a certain instant
as a point in a Cartesian coordinate system with axes x and vx. Though
obviously for a particle that moves on 2 or 3 spatial dimensions (and
even more for a system with several particles) it is not possible to
graphically display the situation of its state, this conceptual
representation remains valid: the state of the system at a certain instant is
represented by a point in a multidimensional space (called "phase
space"). Though now there is an uncertainty about the determination of
the state (e.g. at initial instant t0), we can represent the situation by
considering not a point but a hypervolume, whose hypersurface delimits
the zone where there is the initial state we have not been able to
determine univocally. In the case of the single particle constrained to
move along axis x, this new situation can be easily displayed as shown in
Figure 7.
Vx.

VO

!
•
xo x
Figure 7. Display of uncertainty about the initial state of a
particle constrained to move along axis x.
 Meteorological Models 139

Returning now to the world of our meteorological model, here its

state at a certain instant is defined if the values of all the variables at that
instant on all the points of the grid are defined. The total number of these
values is very great: for a global model it is somewhere around several
tens of millions of data. In perfect analogy with what we have just
discussed about particle systems, we can define a "state space" in which
each axis is relative to a single variable on a single point of the grid. This
way, the initial state, i.e. the initial condition determined by the analysis,
turns out to be a point in this multidimensional space. Likewise, the
uncertainty in the determination of this initial condition can be
represented as a hypervolume, more or less extended around this point.
In this perspective, the meteorological model, by predicting the value
of all the variables on all the points of the grid through the numerical
integration of the equations and parameterisation schemes, does nothing
other than determine, in successive time steps, the evolution of this point
in the state space. But obviously the analysis-based initial state is
approximate: it does not represent the real initial state of the atmosphere
exactly. This, however, is likely to remain within the hypervolume that
represents the uncertainty region in the determination of the initial state
of the model.
At this point, obviously, we have no means to univocally determine
the initial state within the uncertainty region and thus to supply the
model with an initial condition that is quite similar to the real one: its
best estimate for us is the one given by the results of the analysis.
However, the evidence that two initial states that are very similar may
lead to very different developments after a certain time lapse seems to
suggest that we should examine the evolution of the entire volume that
determines the uncertainty. This is done in systems of more limited
dimensions, where a function that measures the probability of finding the
state in the various points within the volume is also associated to this
volume16. In certain systems described by non-linear equations we find
that, beyond certain temporal thresholds, this volume becomes stretched
out and greatly distorted, causing the individual points within it to
become even very distant from each other. This means that initial states

"This function is called "probability density function".

140 From Observations to Simulations

that are similar to each other evolve into final states that are quite
different.
It is obvious that the purpose of the study of the temporal evolution of
this hypervolume is not to determine the initial state that results in a
correct evolution (this would be possible only a posteriori), but to
identify the moment in which this volume starts becoming so distorted as
to lead to very different developments of the various states included in it.
After this moment, because of this uncertainty about the initial condition,
the deterministic forecast (which starts from a single point) might no
longer be reliable.
In the world of a meteorological model it is not possible to determine
a probability density function and study its evolution in the course of
time. Nevertheless, the researchers of various meteorological offices
posed the problem of studying the reliability of deterministic forecasts
and identifying the time limits after which the evolution of various initial
states begin to diverge. The idea underlying these studies is that of
perturbing the results of the initial analysis, to the point of determining a
certain number of initial states (approximately 50) that are representative
of a volume of uncertainty; then these initial states are caused to evolve
by means of the model, with a number of runs equal to that of the initial
states under consideration. These operations are called "ensemble
integrations"17.
If now we examine the forecast results for the variables on the points
of the grid of a certain zone, we can find out when the results of the
model are no longer reliable for that zone. We will discover that the
reliability interval depends on the situation of the weather: there are
some situations in which the developments forecast by the runs of the
models based on the various initial conditions remain close to each other
for a long time, and others in which they begin to diverge after a few
days. Figure 8 (from Pasini and Pelino (2000)) shows the results of a
series of ensemble integrations for the forecast of the wind speed at
surface level on the Italian weather station of Brindisi. The reader will

Because of computer-time problems, all these additional runs are usually carried out on
grids whose resolution is lower than that of the operational model. Here we obviously
cannot dwell on these methods. The reader is referred to a review article on this subject
(Buizza, 2001).
 Meteorological Models 141

notice that the trajectories (therefore the wind speed forecasts) remain
rather close to each other during the first three days of simulated time,
but diverge broadly after 72 hours.

TIME (12 - hours interval)

Figure 8. Results of ensemble integrations for a surface-level wind forecast at Brindisi,

Italy, obtained on the basis of perturbations of the analysis of 13 June 1998 at 12 GMT.
The broken line represents the forecast of the operational model (figure reprinted with
permission from Elsevier).

We will overlook the complex statistical processing carried out on the

data yielded by ensemble integrations, and mention only the fact that the
most immediate possible use of these data is to attribute to the forecast of
the deterministic operational model (the one that starts from the non-
perturbed analysis data) an index of forecasting reliability that depends
on the time limit of the forecast18. In situations of high predictability,

18
Further theoretical aspects of the evolution of the state of a complex system in the state
space will be discussed in the next chapter.
142 From Observations to Simulations

such as those that evolve very little (due, for instance, to "blocking
situations", when a high-pressure condition at all levels persists for many
days over a vast area), this index turns out to be high, up to a distant
forecasting time limit, so the forecasters may commit themselves up to
that date. In opposite situations, where reliability decreases quite soon,
the forecasters will have to limit themselves to a forecast of a few days,
because the reliability attributed to the operational model decreases
quickly after that time limit.

7.7 A Few Conceptual Remarks

In this chapter we analysed the structure and performance of

meteorological models, highlighting a very concrete applicative activity,
and also some aspects that are important for the development of
scientific thought, such as those relative to the emergence of
deterministic chaos and the use of ensemble integrations. Now it will be
expedient to dwell a little on the peculiarities that these models show in
the sphere of the simulation paradigm discussed in the previous chapter:
these peculiarities reveal some conceptual difficulties of modern
scientific practice in the study of complex systems.
To begin with, weather forecasting models closely follow the
structure of a typical simulation model, since they are formed of a core of
differential equations based on partial derivatives that represent our
theoretical knowledge of the basic processes and phenomena of the
atmosphere. These equations have been essentially corroborated by fluid
dynamics and thermodynamics researches with laboratory experiments,
so the purpose of the meteorological model is not to verify their validity.
The fact that we start from individual equations, combine them in a
system and use them together with parameterisation schemes means that
we rely on the possibility of reconstructing the complexity of a system
such as the atmosphere through the recomposition of some basic
elements and interactions, in the form in which they have been
understood theoretically after having been studied separately in the
laboratory.
 Meteorological Models 143

In this recomposition activity, though the primitive equations that are

considered are practically the same for all models, the parameterisation
schemes are different and undergo a constant development, whose main
purpose is to improve the description of the processes that take place at a
scale smaller than the grid spacing, including previously overlooked
details and phenomena. In this regard we can assert, from a historical
point of view, that the path towards the correct simulation of the future
behaviour of the atmosphere has been passing through the examination
of constantly new theoretical elements, which had previously been
regarded as secondary and were subsequently added to the scheme of the
models.
Though here we will not deal with the history of meteorological
modelling, it is worthwhile to point out that the first models based on
primitive equations were dry adiabatic ones, i.e. they regarded the
atmosphere as a system where there is no heat exchange with the external
environment, and air as a compound devoid of humidity or water in its
three states of aggregation. Afterwards, the continuity equation for water
was added: it also supplied the precipitation for non-convective
phenomena. Then a parameterisation scheme was added for convection,
but it only led to an adjustment of the vertical thermal profile of the
involved cells and to the production of convective precipitation (in the
model, all the condensed water fell to the ground and there was no cloud
formation). Later on, convection schemes with cloud formation were
introduced, and it became possible to consider parameterisation routines
also for the radiative exchanges between the Earth and the free
atmosphere, and for the influence of the ground19. We may therefore
assert that, in meteorology, the modellers' ability to simulate the
behaviour of the atmosphere has really improved, owing to a
recomposition activity that led them to consider an increasing number of
new elements of theoretical description, with particular reference to
parameterisations and the relative feedback cycles acting on the variables
of the system.

"Several of these transitions led to a very considerable improvement in the forecasts. A

presentation of some of these different theoretical schemes will be found, for instance, in
Krishnamurti and Bounoua (1996).
144 From Observations to Simulations

The models thus obtained are now able to forecast the future
characteristics of the atmosphere, at least within the predictability period
that can be estimated on the basis of the courses of the ensemble
integrations. Above all, they correctly reconstruct the development of
medium- and large-scale systems, such as the low and high atmospheric
pressure configurations and the fronts with the cloudiness associated to
them. The limits of these models consist in the fact that forecasts on very
limited areas (which are sometimes also characterised by complex
orographical features) cannot be solved correctly because of the finite
grid spacing. So sometimes in the world of a model it is not possible to
tell the windward side of a small range of mountains from the leeward
one, where the weather may be quite different.
The recognition that the macroscopic dynamics is captured well by
the meteorological models supports the idea that the complex non-linear
mixture of theoretical elements introduced in the models is sufficient to
account for the complex emergence of large-scale phenomena in the
atmosphere. If we consider, for instance, the fact that the equations of the
model are applied to air cells whose horizontal sides are about 50 Km
long, this is not surprising, because here we are not describing a
macroscopic system in terms of elements belonging to a different
(smaller) scale: in the latter case the model would be a reductionistic one,
but this is not the situation of the meteorological models. Here the
macroscopic concept of air mass, discussed in Chapter 4, can safely be
applied to simulated evolution.
From this point of view, the success of these models (which are
basically the only instrument we have for reliably forecasting the weather
for a period longer than 24 hours) may reassure us by confirming that the
method adopted (i.e. that of unravelling the skein of the complexity of
the atmosphere by means of experiments in a fluid-dynamics and
thermodynamics laboratory and of subsequently recombining these
elements in a model) has been quite fruitful. This is undoubtedly due to
the fact that, in order to reconstruct the behaviour of the system, we have
placed ourselves at its same macroscopic level. However, there do exist
some macroscopic phenomena, such as the outbreak and evolution of
tropical cyclones and hurricanes, that are not captured well by the
models. This appears to be due to two factors: on the one hand the causes
 Meteorological Models 145

that prime these phenomena seem to be on a small scale; on the other

hand, at least in their mature stage, they can be regarded as phenomena
of macroscopic self-organisation that perhaps represent a complexity that
is emerging at a macroscopic level and cannot be reduced completely to
the basic laws of the system.
We have seen that the results of these weather forecasting models are
supplied in digital form — and therefore can be objectively evaluated a
posteriori by means of statistical indices that quantify the validity of the
forecast — or transmitted as prognostic charts or sometimes graphs like
the one we showed for the results of the ensemble integrations on a
station. Weather maps, which allow us to follow the forecasts of the
various fields in successive temporal steps, represent just the evolution of
the ideal model we discussed in the previous chapter, because they make
it possible to display (dynamically and not statically) the behaviour of
the system under examination, in this case the atmosphere.
This immediate display enables us to comprehend the evolution of the
system "at a glance", even without consulting the data. Sometimes
simply comparing these charts with the analysis obtained subsequently at
the time limit provided for the model is sufficient to allow us to
understand whether in that case the model has yielded satisfactory results
or not, even without having to carry out laborious a posteriori statistical
analyses. This is very useful when we analyse the characteristics of the
model and of possible versions of it that have been modified on the basis
of past case studies that were particularly significant from the
meteorological point of view. In this regard, the possibility of having at
our disposal a graphic display of the behaviour of a complex system such
as the atmosphere facilitates the evaluation of the results, and
substantially changes scientific practice in the analysis of the behaviour
of this complex system. Evaluations are often performed on the ideal
model whose graphic evolution is available: a check is carried out to
ascertain in what position the various models place a certain atmospheric
low, what values they forecast for its minimum value, etc.
These last examples help us to understand that we can change the
physics of a model and control its forecasting performance on case
studies of the past. However, the manageability and the flexibility of
models as an instrument allow us to do other things as well: for instance,
146 From Observations to Simulations

to construct possible worlds that do not exist currently in nature. This

may be interesting when we wish or need to act on the natural
environment in order to change it. For instance, let us suppose that we
have decided to build a dam with an artificial lake, for the production of
electric energy: if we have a reliable meteorological model (i.e. a model
that has been validated on the basis of past situations or of everyday
work), we can evaluate, a priori, the environmental impact of the lake in
terms of the changes it may cause in the weather of the surrounding area.
Obviously we can perform tests on the past years for which we have the
meteorological data relative to the area under consideration. First we
must run the model with the state of the ground as it is in reality (these
runs of the model are called "control runs"), then we must change the
state of the ground, creating a lake in the world of the model; at this
point, we can run the model with these new boundary conditions. The
fundamentals of meteorology explained in Chapter 4 lead us to reckon
that the presence of this expanse of water will essentially affect the
radiative exchange between the Earth and the free atmosphere, the
transmission of heat, and the atmospheric humidity content. Basically,
the most evident changes we expect are in the temperature of the air at
surface level, in the precipitation regime, and in the phenomena
connected to them. An analysis of these results may advise us to build
the dam or not to build it, or to change the project in a certain way. In
actual fact, if we perform these tests over a period of several years, the
simulations will become climatic ones, and for these the reader is
referred to the next chapter.
To round off this discussion, we must return briefly to the emergence
of deterministic chaos that has led to the need of developing and using
the method of ensemble integrations. We have already explained that
temporal evolution may move apart two points that were initially quite
near in the state space, causing them to become even considerably distant
from each other after a certain time lapse20. Moreover, their way of
moving away from each other is rather characteristic: after a certain
instant the points diverge quickly (in an exponential manner): this is
called a bifurcation. Where possible, the method for amending this flaw

Obviously this is a distance in a multidimensional space.

 Meteorological Models 147

in the deterministic treatment of a non-linear dynamical system has

consisted in studying the evolution of the probability density function
associated to the volume that quantifies the uncertainty about the initial
state. In the case of meteorological models, an alternative method has
been adopted: that of studying an extensive range of evolutionary paths,
one for each point considered at the initial instant.
Whatever alternative we are compelled to choose, these examples
show that the concrete forecast of the future state of a complex system
cannot be achieved univocally with a deterministic prediction. Both the
methods highlight the need to adopt a probabilistic point of view in a
domain — systems of non-linear differential equations — that was
considered the undisputed realm of determinism. The concrete example
of the resort to ensemble integrations (which also has a share in changing
the scientific and applicative practice in the world of meteorology)
particularly reinforces this vision. In any case, it seems that Laplace's
dream has vanished for good.
 Chapter 8

Climatic Models

In the previous chapter, we explained how it is possible to work out a

simulation model of the atmosphere that can supply weather forecasts for
a certain time lapse. Here, reminding the reader of the definition of
climate given in Chapter 2, we will consider whether it is possible to
work out simulation models that account for the average weather and its
variability over a few decades, both validating them for the past and
applying them to a forecasting of the future climate. As we have always
done in this book, here we will not only evaluate the applicative results
of these models, but also analyse their aspects of conceptual novelty as a
means of investigation that by now has become a part of present-day
scientific practice in the study of a complex system like the Earth.

8.1 From Weather Forecasting to Climate Forecasting: What

Changes?

In the course of Chapter 4, we endeavoured to construct a theoretical

vision, at least a qualitative one, of some phenomena and processes that
take place in the Earth system, particularly highlighting the
characteristics of some feedback cycles, and thus revealing the
complexity of the system. When in Chapter 7 we proceeded to work out
some simulation models for weather forecasting, we focused on the
atmosphere subsystem, supplying an algorithmic reduction of its fluid
dynamic and thermodynamic characteristic features by means of
primitive equations and parameterisation schemes. In doing this, we dealt
dynamically only with the atmosphere subsystem: the representation of
its interactions with the other subsystems of the Earth system was

149
150 From Observations to Simulations

confined to the status of a series of influences of an environment outside

the atmosphere, and these influences were expressed in the model as
boundary conditions and external forcing factors. More specifically, in
this non-dynamical treatment of the relations between the various
subsystems, the influence of the external environment is usually a one-
way one, i.e. it is not subject to feedbacks, except in an artificial and, in
any case, non-dynamical way1.
What made it possible to choose this approach was, notably, the fact
that the changes that take place in the subsystems situated at the interface
with the atmosphere are usually slow in comparison with the
meteorological evolution, on the time scales that are considered (up to 10
days). If we decide to follow a path that will lead us to the elaboration of
models for studying the evolution of the climate, therefore at much larger
time scales, the evolution in time of these subsystems (with their
influences on the atmosphere and relative feedbacks) becomes no longer
negligible. From this point of view, we should fully consider the
dynamics of these subsystems in their interaction with the atmosphere,
through the elaboration of a coupled system of diagnostic and
evolutionary equations. Among other things, the fact that systems such as
the oceans are characterised by inertia factors greater than those relative
to the atmosphere may suggest that they play a stabilising role, thus
promoting the predictability of the characteristics of the overall Earth
system.
A further element that emerged in our discussion of meteorological
models and is important both from a conceptual point of view and from
an applicative one, is the recognition of the existence of a limited
predictability time lapse for forecasts. In this perspective, if for the
weather we cannot produce forecasts over more than 10 days, how can
we hope to achieve a forecast of the climate over a future period that may
extend to several decades?
The key for answering this question lies in the statistical formulation
of the definition of climate and in the consequent concept of climatic
forecast. The weather is defined when the values of certain atmospheric
variables that approximately determine the state of the atmosphere are

'See the particular case described in Note 2 of Chapter 7.

 Climatic Models 151

known; weather forecasts aim at predicting the evolution of this state

with time in an accurate, deterministic way (with the obvious limits
revealed by the emergence of deterministic chaos and the consequent use
of ensemble integrations). When we speak about the climate, on the
contrary, we mean to define a certain period of time by means of average
values and of the variability of quantities that are important in the
atmosphere; climatic forecasts, therefore, do not aim at determining the
precise appearance of various atmospheric states, but at revealing future
"scenarios" in which these quantities take on certain average values and
a certain variability, and in which it is possible, if necessary, to
determine whether one or several "classes" of atmospheric states are
more frequent than others2.

8.2 The Concept of "Attractor" and Climatic Simulations

As a rule, the interest in the average behaviour of a system and in its

statistical variations over a fairly long period of time, in comparison with
the wish to determine the precise behaviour of a system, leads to lower
requirements during the forecasting stage. This type of forecasting,
therefore, is easier to carry out, particularly if there is the possibility of
determining the statistical laws followed by the system. An extreme
example, familiar to all of us, is the situation that arises when we
repeatedly roll a dice. Though we cannot foretell the sequence of the
sides that will come up, over a great number of throws we can predict the
average frequency (1/6 for each side), and also the variability in the
number of times each side comes up.
Obviously, for the system we are discussing in this book we do not
know any well-defined statistical laws such as the law of large numbers
for the distribution of random events: for instance, it is still under debate
whether the course of the climate is driven by a complex, chaotic
dynamics or by a sequence of random events3. In any case, we can adopt
the depiction of the evolution of the system, which is represented by a

2
The possibility of determining these classes will be briefly discussed further on.
3
The interested reader is referred to the article by Pasini et al. (1997) and references
therein.
152 From Observations to Simulations

point in the multidimensional state space, as a trajectory of this point in

the course of time. Temporarily disregarding the problem of the univocal
determination of the initial state, knowledge of the deterministic laws of
the evolution of the system offers the possibility to determine the curve
that the point that represents its state traces within this multidimensional
space in the course of time. Vice versa, a statistical knowledge of the
system would make it possible to know the average distribution of these
points in the state space, their variability and the shape of the "geometric
figure" (if any) formed by them.
Though we do not possess a detailed knowledge of all the laws that
determine the evolution of the system, or an in-depth statistical
knowledge of the system on the basis of purely theoretical explanations,
the fact that there exist some coexistence or evolutionary laws
interlinking the variables indicates that not all combinations of their
values are possible4. In our case, this leads us to presume that, in the
evolutionary history of the system under consideration, its state cannot
cover all the points in the state space (as would be possible if there were
no functional links between the variables), but must be confined in a
subset belonging to it. The geometric figure that identifies this subset is
called "attractor" and determines the points that are physically possible
for the state of the system, given certain external conditions (essentially
forcing factors and boundary conditions).
Statistical knowledge of the system may therefore consist merely in
knowledge of the characteristics of the attractor: the barycentre of this
geometric figure may represent the average state of the atmosphere, and
the extension along the various axes may represent its variability. If,
moreover, we were to identify some parts of the attractor that are
particularly thick with points, i.e. are highly "frequented" by the state of
the system, this would testify to the existence of particularly probable
state classes.
The concepts of state space and attractor are obviously rather abstract
and, because of the high dimensionality of our system, not easy to
visualise. So we will resort to a two-dimensional mechanical analogy.

For instance, in the simple case of two variables, x and y, that are interlinked by a linear
law, only the points belonging to this line are actually possible on plane xy.
 Climatic Models 153

Let us consider a straight, horizontal little track that can be used as a

track for marbles. If we bend this track downwards and upwards, we
obtain a piece of a little roller-coaster like the one shown in Figure 9.

(a) (b)

Etot

Figure 9. Mechanical analogy for the concept of attractor.

Now, if we imagine a condition of ideal frictionless motion, when we

place a marble at a certain point of this track and let it start from a rest
condition, it will run down the track, then run up again to the same height
on the other side; subsequently its motion will be inverted, and it will
return, after a certain period, to its starting point. The path covered by the
marble (indicated by the thicker line in the figure) is determined by the
height from which we have started the motion: as we know from the
principles of mechanics we have studied at school, this determines its
potential energy (which, since the marble starts from the rest condition,
coincides with the total mechanical energy of the system, EtoX). At this
point, if we change the height from which we release the marble, i.e. if
we change the value of Etot, the states that are possible for the marble
change as well, as shown in the two parts of Figure 9.
The curves drawn with a thick line are thus the analogue of the
attractor in the state space: as you can see, when the energy of the system
changes, the extension and shape of the trajectory change as well.
Among other things, in this simple mechanical example we are also able
154 From Observations to Simulations

to determine which are the "most-frequented" stretches of this trajectory,

i.e. the point where we are most likely to find the marble if we look at the
system at any instant: if we apply the law of conservation of total energy
(understood as the sum of kinetic energy and potential energy), we will
find that the velocity of the marble is lower in the highest parts of the
trajectory and higher in the lowest parts. So the stretches where the
marble stays for a longer time are the highest ones: this would probably
be particularly noticeable in the stretch of less steep slope to the right in
Figure 9b.
Up to now, nobody has been able to determine the characteristics of
the attractor of the atmosphere or, even less, of that of the Earth system.
The main cause of this deficiency is the fact that there are few points
available in the state space (for the brief period of extensive monitoring
of the system: to be optimistic, the last century) with respect to the high
dimensionality of the system: in nature we must be content with
historically observed states, which, however, are only a small subset of
the actually possible ones. We must add to this that, as the years went by,
the forcing factors outside the system also changed, making the problem
even more difficult to tackle from a theoretical point of view.
At this point, it may be natural to think that in this situation the
models, since they allow us to reconstruct possible worlds and to repeat
the simulations over an extremely long period of simulated time and with
different initial states, may be able to contribute to the determination of
the attractor of a system. If this is possible, although a precise
determination of the evolution of the system is not obtained, the
statistical properties of this evolutionary system can be recognised. Since
these properties are precisely the ones that characterise the definition of a
climate, the application of certain models may turn out to be essential for
our purposes.
Once the model of a certain physical system has been worked out, for
instance by means of a system of differential equations, in this dynamical
model the attractor is the figure that determines the "asymptotic"
behaviour of the system, i.e. the behaviour that appears after a transient
stage of imbalance of the variables (such as the spin-up stage discussed
 Climatic Models 155

in relation to meteorological models)5. Figure 10 shows the attractor for

Lorenz's three-dimensional system, the first system in which the
phenomenon of deterministic chaos was revealed. Notice the point at the
centre of the figure, from which the motion starts: after a short stretch of
trajectory, the asymptotic behaviour is reached on the attractor. This
attractor is characterised by two wings, which express different classes,
or regimes of motion.

Fig. 10. The Lorenz attractor (figure reprinted from Pasini and Pelino (2000) with
permission from Elsevier).

Incidentally, we would like to point out that the existence of attractors

such as Lorenz's one contributed to the emergence of a "fractal"

To return briefly to the mechanical analogy presented above, here the asymptotic
behaviour of the system coincides with the real behaviour, because we imagined an ideal
system without friction. If we introduce this slowing-down component as well, we will
see that the velocity of the marble is gradually lessened (transient stage), and the marble
ends up by stopping at the lowest point: this fixed point may therefore be the real
attractor for motion in a real system with friction.
156 From Observations to Simulations

geometry, typical of objects characterised by a non-integer

dimensionality6.

8.3 Approaching the Description of a Coupled and Highly

Interacting Climate System

In Chapter 4 we examined some phenomena and processes that take

place in the atmosphere, and highlighted the influence (and relative
feedbacks) to which they are subjected by what occurs in the subsystems
at the interface with the atmosphere. Whereas for the short time limits
characteristic of weather forecasts some of these interactions may be
disregarded, or parameterised in a simple manner, on climatic time scales
this is no longer possible: the simulation model would completely lose its
effectiveness in relation to observational reality. More specifically, on
these time scales the evolution of the subsystems at the interface with the
atmosphere can no longer be defined a priori (through a given
evolutionary law or by means of a representation of its internal dynamics
alone), because it is affected by feedbacks due to changes that have taken
place in the atmosphere in the meantime. Therefore it is necessary for the
dynamics of the various subsystems to be coupled in a single, highly
interacting dynamical system, where what occurs in a subsystem affects
the dynamics of the other subsystems and is affected by feedbacks from
them.
In the history of science, and above all in the course of the recent
development of science, characterised by an extreme specialisation, the
various subsystems of the Earth system were studied within the domains
of separate disciplines. In each of these disciplines, theoretical
knowledge has been advancing, and in these sectors, as in the study of
the atmosphere, during the last few decades there has been an intensive
use of simulation models. In some cases, like those of oceanic circulation
models (though with the obvious differences due to the fluid medium and
to the particular role of salinity), the formalism that is used is very
similar to the one adopted in atmospheric models. In other cases,

6
Obviously here we cannot dwell on fractal geometry. The reader is referred to a book
written by the creator of this discipline (Mandelbrot (1987)).
 Climatic Models 157

particularly when study is addressed to the contributions to the dynamics

of a particular subsystem that are also caused by biological organisms (as
in the development of vegetation, or in the part of the carbon cycle due to
emission or absorption by plants and algae), specialistic studies lead to
very detailed representations of the processes on a local level, but require
further development in order to fully consider and adequately size up the
contribution of these factors in the climatic system on a regional or
global level.
If we wish to make a list of the main components of the climatic
system, we may consider the following subsystems, cycles and processes
of the Earth system, for each of which we possess an individual
theoretical treatment, as it has evolved in the course of the history of
scientific knowledge.
Atmosphere. This is a system we know well, and whose modelling
was the subject of the previous chapter: obviously this knowledge is
crucial for the problem we are considering, because it is within this
system that climatic phenomena are detected, through the change in
the values of its variables within a given time lapse.
Oceans. The extension of the sea interface, the characteristics of its
thermal capacity and contribution of humidity to the atmosphere,
together with the dynamics of ocean currents and cycles such as that
of El Nino, cause the ocean to be a fundamental factor in the
interactions between the atmosphere and the external environment.
Oceanography is a rather ancient discipline; at present it supplies a
theoretical description of this system at the same macroscopic level
as that of the description available for the atmosphere.
Continental surfaces. Like that of the oceans, the behaviour of the
solid interface is essential for understanding the flows of heat,
humidity, etc., exchanged by the ground and the atmosphere. Here,
too, great attention must be given to the changes in the state of the
ground that affect the radiative balance. The reader should notice,
however, that some of the changes in the ground are to be attributed
not to natural dynamics, but to human activities (e.g. the
deforestation or the establishment of new crops).
Cryosphere. This name indicates the subsystem formed of the sea ice
and land ice, whose evolution is now described dynamically. The
158 From Observations to Simulations

dynamics of the formation or melting of ice obviously affects the

radiative exchange between the ground and the atmosphere, makes it
possible to better define the capability of absorption and reflection of
solar radiation in the course of time, and also affects the salinity rate
of the various parts of the oceans.
Aerosols. We already explained that in certain conditions the
lithosphere introduces in the atmosphere dust of various origins (for
instance, volcanic ash from eruptions), and we described the possible
primary, direct effects of this on the radiative balance, the secondary
effects on cloud formation, and the consequences of these effects.
The duration of the persistence of these aerosols in the atmosphere
depends on several factors, including their chemical composition.
Now we are able to describe the life cycle of aerosols formed of
sulphates and also to schematically describe the behaviour of
aerosols of different compositions. It is important to point out,
however, that the emission of these aerosols depends chiefly on
natural events that may be hard to predict, or on anthropogenic
processes.
Carbon cycle. This is the complex cycle that accounts for the
presence and amount of carbon dioxide in the atmosphere. The
presence of CO2 depends on mechanisms of emission, absorption and
storage that involve both physical and biological phenomena in the
oceans and on land: it is known, for instance, that in photosynthesis
C0 2 is absorbed, and its carbon is stored in the wood that forms the
trees. As in the previous case of aerosols, now most of the CO2
emissions have an anthropic origin.
Vegetation. All the vegetable world affects the climate and its
changes, and is affected by them: on the one hand its presence (with
different species) influences the albedo7, the humidity introduced in
the atmosphere with the evaporation/transpiration process, and the
absorption of carbon dioxide; on the other hand its growth is driven
by weather and climate factors that it contributes to change. During

7
The reader is reminded that the albedo is defined as the ratio of the energy reflected in
space by the Earth, clouds and atmosphere to the total incident energy (coming from the
Sun).
 Climatic Models 159

the last few years, there have been some attempts at a dynamic
modelling of the vegetation. Here too, however, it is necessary to
point out that these attempts must include, as data, the changes in
vegetation caused by mankind — first of all the deforestation
activity.
Atmospheric chemistry. The chemicals present in the atmosphere,
most of which are emitted as a result of human activities, interact in a
complex way through physical phenomena (such as diffusion or
deposition on the ground) and chemical reactions. All this is
interesting for the study of possible climatic changes, because the
presence or absence of certain chemical species affects the radiative
balance and other phenomena in the atmosphere. On the other hand,
the changes may give rise to feedbacks on the atmospheric
chemistry: one of several possible examples is the fact that the
occurrence or non-occurrence of certain reactions and their speed
depend on the temperature. A theoretical treatment, necessarily
simplified, is now available, and attempts are being made to
introduce it as an interacting model component within a climatic
model.
Mankind. As we have repeatedly stated, in almost all the points of
this list of the subsystems that form the Earth system, the presence of
mankind affects various important cycles of the system, to the point
that it is reasonable to start thinking of the possibility of beginning to
produce models of human activities in their interactions with the
other subsystems. In this sense, we mean to interpret the presence of
mankind as an integral, fundamental part of the Earth system: though
on the one hand a "human dynamics" is obviously hard to define, as
the so-called "human sciences" have shown, on the other hand
considering the human element within the Earth system may be the
only way to reveal the feedbacks of possible climatic changes on
human activities, and to understand, in an integrated way, the
sustainability and limits of the system.
Under the present circumstances of climatic modelling, a model that
is (so to speak) complete, that integrates in a single dynamical system all
the components we have just discussed, does not actually exist. In
particular, the human element is still outside the model-system that is
160 From Observations to Simulations

being studied dynamically, and acts on the model as a factor that changes
some of the forcing factors of the system. Moreover, as regards the
aspects relative to the atmospheric chemistry and, partly, to the dynamics
of vegetation, the integrated modelling treatment of these subsystems is
still rather problematic, both because of "connection" problems of the
formalisms in the model and because of numerical problems often due to
the different scales of development of the relative phenomena with
respect to the atmospheric and oceanic processes in particular.
In this picture, the "hard core" of present-day climatic models
consists of a strong coupling of a meteorological model with an
oceanographic one, in which the aspects relative to the continental
surfaces and cryosphere are included as an integral part. To this base
there are added some interacting modules relative to the dynamical
behaviour of aerosols (both those containing sulphates and others having
a different composition) and of the carbon cycle. The models thus
obtained are called AOGCMs (Atmosphere-Ocean General Circulation
Models).
The need to carry out simulations at least over several decades
obviously leads to a reduction in the resolution of these models, in
comparison with that of the purely meteorological models discussed in
the previous chapter. At present, the state-of-the-art values for the
various coupled models are, in a meteorological model, an average
distance between the grid points of 250 Km horizontally and 1 Km
vertically (where, however, in many cases an increased resolution is kept
in the lower layers). In an oceanic module, the range is usually from 125
to 250 Km for horizontal resolution and from 200 to 400 m for vertical
resolution. Obviously, as we have already seen for weather forecasting
models, here too parameterisation schemes are used for the processes
that take place on a spatial scale smaller than that of the grid, such as the
convection of clouds in the atmosphere and oceanic convection.
In spite of the difficulty of including the elements that are more
distant from a purely physical treatment, we can safely state that modern
coupled models are an important theoretical training-ground for the
multidisciplinary representation of complex phenomena like those that
take place in the Earth system. More specifically, the need to work out a
model (the only method with which we can hope to attain an
 Climatic Models 161

understanding of the phenomena of this highly interacting system),

compels experts of various disciplines to speak the same language, that is
to work out a unitary formalism for obtaining software to be run on a
computer (everybody knows that computers are stupid and can carry out
instructions only if they are self-consistent and written in a univocal
language).

8.4 Experiments for Validation and Sensitivity Testing of a Climatic

Model

In the course of the brief description of coupled models that we have just
completed, we duly highlighted the importance of regarding the various
subsystems of the Earth system as interacting dynamically in a single
model, particularly in order to avoid one-way influences of the external
environment on the model-system under consideration. This makes it
possible to account for the mutual interactions between these subsystems,
including the feedbacks, therefore to achieve a more consistent and
correct simulation of the evolution of the whole system.
In actual fact, there also exists another motivation for dynamically
joining the various subsystems in a single model. In fact (returning to the
concepts introduced in Section 8.2), if we believe that the model
correctly simulates the meteorological and climatic behaviour of the
Earth system, we can think that the climate is represented by information
that can be extracted from the attractor of the model, in terms of average
state, variability and most frequent states. Strictly speaking, however, the
attractor of a system is a "static figure", that does not change in the
course of time8, only if the external forcing factors and boundary
conditions are fixed: in this case the system is called "autonomous"9. The

And that therefore is outlined increasingly well by observing the state of the system over
increasingly long periods.
9
It is fairly simple to observe how the statistical properties of an autonomous system are
driven by the values of external constraints or forcing factors. If, for instance, we
consider an isolated system formed of a gas in a container to which a certain amount of
energy has been previously supplied (possibly in the form of heat), the velocity of the
particles within a statistical distribution (called Maxwell-Boltzmann distribution) is
determined by this amount of energy. If we supply more heat to the system, then isolate it
162 From Observations to Simulations

most immediate way of not losing anything of the wealth of knowledge

gained from the concept of attractor is therefore that of rendering the
model-system as autonomous as possible, by reducing the evolution of
the external environment to a minimum, i.e. by dynamically including all
the evolutionary subsystems in the model10.
Now that we have acquired this theoretical background and this
vision of climatic models as dynamical systems, we will endeavour to
apply them concretely. The first thing we require of a climatic model,
obviously, is a correct reconstruction of the climate of the past. Let us
therefore consider a period in which the external forcings and the
boundary conditions can be regarded as practically constant, and let us
proceed to simulate the climate over this period.
The first problem we must tackle is obviously that of the sensitivity to
initial conditions: if we start from an initial condition determined by a
(necessarily approximate) analysis at a given instant, after a certain time
lapse the simulated evolution will be quite different from the real
evolution of the system under examination. This problem was explained
for meteorological models in the previous chapter, and now reappears in
the model-system that is extended also to the other subsystems. Now the
problem actually seems even more serious, both because for the distant
past no accurate analyses are available, and because the lower resolution
of a climatic model in comparison with a meteorological one renders the
analysis even more approximate.
In actual fact, as we mentioned previously, the statistical character of
the definition of climate makes it possible to get around this problem: in
particular we can consider the properties of the attractor under the

again, once the internal equilibrium has been reached, the velocity of the particles has
changed (and its average value has increased).
l0
It is evident that in any case there will remain some constraints or forcings outside the
Earth system: the reader should consider the amount of incident solar radiation, which is
precisely what determines the energy supplied to the system. Obviously it is not possible
to establish a perfect parallelism with the case described in the previous note, because
here we are in a non-linear dissipative system with many interacting components.
However, should we suppose that there is a radical change in the so-called solar constant,
we might expect that the system would respond to this different value of the forcing with
a change in its average state, that is with a change in the barycentre of its attractor
(which, we must point out, is not defined only by the temperature values, but also by the
values of the other variables), and possibly also with changes in its shape.
 Climatic Models 163

influence of the specific boundary conditions and external forcings, thus

evaluating the average climatic properties and the variability on the basis
of the statistics of the states of the system in the period under
consideration. Obviously, because of the high dimensionality of the state
space of the model, we will extract only the variables we regard as
fundamental for the reconstruction of the climate.
At this point, however, it is evident that the single run of the model
supplies only an evolutionary trajectory in the state space, a sort of
"sampling" of the attractor, just like the evolution of the real system is a
particular realisation of the possible physical states. So if we have no
hope of being able to reconstruct this real trajectory, because of the
emergence of deterministic chaos, we may just as well increase the
statistics and run the model repeatedly with different initial conditions.
This way we can explore the state space more thoroughly and reconstruct
the attractor with a greater abundance of details11. If we carry out a
considerable number of runs (which here too we can call "ensemble
integrations"), we can determine the average values and possible climatic
variability over the period under consideration, and we can also
understand whether the individual, particular realisation of the climate
extracted from the data of the real system falls into the class of climatic
values simulated by the model. If it does, the model is corroborated by
the real data: this is the validation stage12.
Once the state space has been explored and the climatic variability
has been evaluated by means of the analysis of the sensitivity to initial
conditions we have just described, if a validated and physically self-
consistent model is available, it is possible to evaluate the climatic

"it is worthwhile to remind the reader that the attractor is the figure that determines the
asymptotic behaviour of the model-system, which appears after the transient stage due to
the initial unbalancing problems of the fields. In this regard, we would like to point out
that, in modelling practice, work often starts from all sorts of initial conditions (e.g. a
condition of atmosphere at rest), beginning to consider the data produced by the model
only after the end of this transient phase, which in a climatic model may even be several
years, because of the unbalancing of the hydrological part of the module relative to the
continental surfaces.
12
During the validation of one or several models, it is also possible to carry out studies on
the sensitivity to changes in the modelling representation of the various processes
described by means of equations or parameterisations, if necessary with expressly-made
projects of intercomparison between different models.
164 From Observations to Simulations

response of the model-system to changes in the boundary conditions and

external forcing factors. Assuming that our theoretical knowledge
expressed in the equations and parameterisation schemes has an
extensive domain of validity, this is done in order to analyse the
behaviour of the Earth system thus simulated, once a new balance has
been reached in extreme conditions (doubling or quadruplication of the
concentration of CO2, total deforestation, conditions of "nuclear winter"
due to a thermonuclear war, etc.). Since, as we have seen, the Earth
system is very complex, there actually exist countless factors that affect
the climate, both directly and indirectly, through feedback mechanisms.
In this framework, studies on the sensitivity to boundary conditions and
external forcings may make it possible to distinguish the factors that
have a greater impact on climate change.
This way we can also begin to explore the scenarios that may appear
in the future. As a rule, some runs of the model (called control runs) are
carried out with the values of the forcing factors and boundary conditions
typical of the present period or of a reference sample period. Then the
model is run with some of these parameters changed (usually by bringing
them to extreme values).
Since the model is practically autonomous, once some ensemble
integrations have been carried out it is possible to evaluate the statistical
changes in its average state and in its variability, through the variations
in the two attractors, as they are outlined, respectively, by the control
runs and by the runs with altered forcing factors or boundary conditions.
The parallelism with the situation described in Note 10, i.e. the gas
system to which different amounts of energy are supplied, is rather
cogent. A situation of this type — returning for an instant to the case of
the repeated rolling of a dice — can be obtained by applying a tiny
weight to a side of the dice, which thus becomes a loaded dice. Now,
though the temporal sequence of the draws remains unpredictable, the
frequency of the draws, obtained empirically after the dice has been
rolled for a considerable number of times, will turn out to be different,
and, from now on, predictable for that particular dice.
The analysis performed by means of climatic models, therefore, is
basically statistical; as in other systems, this treatment supplies results
that otherwise would be unattainable. More specifically, in the analysis
 Climatic Models 165

of the sensitivity of a climatic model with perturbation of the forcing

factors and boundary conditions, the first thing to be evaluated is the
significancy of the climatic changes that are found. Since the first runs
determine the averages and the natural variability of the quantities that
are important for the definition of the climate of the control period, it is
necessary to ascertain whether the climate outlined by the runs with
perturbed parameters falls into the range of climatic values of the control
period, or whether it does not. In the latter case, the changes that are
found are statistically significant, and we can recognise an impact of the
change in the values of the forcing factors and boundary conditions on
the climate; in the former case, these values fall into the natural
variability of the climate, so there does not exist a clear signal of an
influence of these external changes on the dynamical system.

8.5 Evolutionary Validation and Climatic Forecasts

The analysis of climatic models carried out up to now was particularly

focused on the evolution of the state of the model-system in the state
space and on the concept of attractor, which, as we explained,
summarises the statistical characteristics of the system. Our discussion
reveals the close connection between the method for studying the climate
on the Earth and the methods adopted for analysing complex systems
that are characterised by smaller dimensions and a greater manageability
(possibly systems of which we have an experimental counterpart in the
laboratory). This attributes to modelling the difficult task of verifying
advanced theoretical physics methods in an extremely concrete and
applicative field, characterised by impact consequences that can be
verified in a very cogent way.
Despite the recognition of this noble origin of the structure and
investigation methods of climatic models, which issue from the
theoretical physics of complex systems, the rigour of theory (as usually
happens in applicative disciplines) must sometimes be "softened"
because of the working conditions, which often require the achievement
of results (albeit partial and approximate) within a reasonable time. Thus,
for instance, in the validation and sensitivity testing experiments that are
166 From Observations to Simulations

actually carried out with present-day models, because of the prohibitive

computer times required, the use of ensemble integrations was rather
reduced, and was limited in most cases to a small number of runs. A
discrete, rather restricted sampling of the attractor was thus obtained, so
it is reasonable to point out that this sampling may turn out to be
insufficient for determining the desired climatic statistics with due
accuracy.
In the theoretical context of the representation of the Earth system by
means of equations and parameterisation schemes, there has been
substantial progress during the last few years. Up to a short time ago, for
instance, in the physical coupling between the oceans and the atmosphere
it was necessary to introduce an exchange of flows of heat and water
whose nature was not physical (i.e. they were not observed in nature), in
order to reconstruct the present climate in a satisfactory manner: this was
called "flow adjustment". Nowadays the improved representation of the
exchanges at the interface between the oceans and the atmosphere makes
it possible to dispense with this artificial adjustment. In any case, we
must remember that, like the meteorological models discussed in the
previous chapter, climatic models contain some parameters that in any
case must be fixed (without losing track of their physical meaning) with
a somewhat artisanal activity of numerical experimentation.
These remarks reveal how difficult it is to preserve an extremely high
degree of theoretical rigour in the application of climatic models to the
validation and sensitivity studies presented in the previous section. In
these studies, the approach to the problem is theoretically correct in any
case, because the boundary conditions and external forcings are regarded
as practically constant (the model is autonomous). This means that we
are examining a situation of equilibrium, where, after the transient stage
due to the effects of the unbalancing of the initial fields, the trajectory of
the state of the model-system lies on its own attractor.
The problem that comes up at once, if we consider the data of the
external forcing factors and boundary conditions in the last century, is
that these factors and conditions cannot at all be regarded as almost
constant throughout the period. Among other things, there is some
evidence of the fact that they have a certain influence on the value of
some quantities that have a share in defining the state of the system, such
 Climatic Models 167

as the quantity shown in Figure 2 of Chapter 2, relative to the changes in

temperature detected after volcano eruptions. If we add to this the fact
that the influences of human activities, in particular anthropogenic
emissions of greenhouse gases, have been increasing on the whole13, we
can understand that, in order to simulate the behaviour of the climate
during the last century, validating a model on these real data, we cannot
postulate conditions of equilibrium within an autonomous model-system.
At this point, however, the whole theoretical construction based on
the concept of the attractor of an autonomous system begins to "totter":
the system can no longer be studied in a state of equilibrium, but must be
studied in evolutionary conditions, i.e. transient ones. The calm,
reassuring situation where, at least in theory, we could run the model for
a long simulated time with the same forcing factors and boundary
conditions, thus determining with great richness of detail the shape of the
attractor that determined its climatic statistics, is no longer permissible. If
we really wish to validate a model with respect to real observations, we
must cause it to reconstruct the climatic statistics of its evolutionary
history: we are in the conditions of an evolutionary validation.
In these conditions it is no longer possible to speak about an attractor
as a static figure in the state space. In actual fact, the problem due to the
fact that forcings and boundary conditions depend on time has been
studied in dynamic systems characterised by small dimensions, and it has
been noticed that the trajectory of the model-system tends to shift
progressively towards attractors having a different barycentre, extension
and shape, though obviously it never lies on them: the system is in
transition and not in equilibrium. This shift takes place sometimes
gradually and sometimes more suddenly, and sometimes depends on the
position in which the point that determines the state of the system is
located in the state space. In essence, it has been demonstrated that an
evolutionary problem of this type can be studied, though the extensive
statistics that in the stationary, equilibrium-based problem sprang from
the fact that the trajectory lay repeatedly on the static attractor must be

13
We remind the reader that in climatic models the mankind subsystem has not yet been
dynamically coupled with the other subsystems.
168 From Observations to Simulations

obtained here by further increasing the use of repeated runs of the model
with different initial conditions.
The outcomes of the evolutionary validation of these coupled models
during the twentieth century, if we include the real data of the
concentrations of greenhouse gases, changes in the soil, and so on, lead
to satisfactory results where averages and global or hemispheric
variability are concerned, while they show discrepancies (even
considerable ones in some cases) on individual regions of the globe. In
particular, the variability with time, on the scale of a few years or a
decade, is caught better if real variations in Sun irradiation and volcanic
aerosol emission are included as forcing factors.
When we have worked out a climatic model and have validated it in
an evolutionary manner over a historical period for which we possess
real data, we can rely on the validity of its theoretical scheme. We are
ready, at this point, to apply the model to the simulation of the future
evolution of the Earth system, in order to forecast possible climatic
changes in a more or less near future, possibly in an accurate, reliable
way.
But here comes the difficult part! Since the system is non-
autonomous, above all because the human influences are an external
factor that is not dynamically coupled with the other subsystems of the
Earth system, the behaviour of these influences in the course of time
(supplied as forcings or boundary conditions in the model-system) are
affected by forecasts, made outside the model, that can be produced only
by the so-called "human sciences". This fact introduces so many
important factors of uncertainty that it is absolutely necessary to
hypothesise various future scenarios.
Since variability in natural forcings is rather unpredictable and is
evaluated statistically, the essential forcing factors that come into the
model-system from the outside are basically the concentrations of
greenhouse gases and anthropogenic aerosols, and the changes in the use
of the soil. Both are driven by the scenario that can be surmised for the
global and regional socio-economic development (which notably
determines the emission of these gases), and are determined after having
considered the feedbacks of the system, which may lead, for instance, to
the absorption of a part of the emissions. Socio-economic development,
 Climatic Models 169

in turn, follows market laws, but is also affected by interactions with the
political forces. This situation is summarised in Figure 11, which shows
the cascade of factors that influence the data to be included in the
climatic model — these data, in turn, obviously influence the results of
the forecasts. The figure also surmises that there is a feedback of the
results of the models on the political forces: considering the situation of
international negotiations, at present this seems to be an excessively
optimistic surmise14.

POLITICAL FORCES SOCIO-ECONOMIC FORECASTS

EMISSION
SCENARIOS

FORECASTS OF FORECASTS OF
CONCENTRATIONS USE OF SOIL

CLIMATIC
MODEL

i CLIMATIC FORECASTS

Figure 11. The socio-economic forcing factors for a climatic model.

14
We will return to this topic in Chapter 9.
170 From Observations to Simulations

Of late, the scientific community that deals with climatic models has
acquired a range of scenarios of socio-economic and emission
developments. The climatic simulations refer more and more often to
runs of an individual model on the basis of the data of these scenarios, so
as to give out a similar range of climatic developments, in which each
forecast is relative to a scenario for economy and emissions that has been
surmised before starting the run. Some examples of these forecasting
results will be discussed in Section 8.7.

8.6 Simplified Models and Regional-Scale Models

As the reader has undoubtedly understood from what we have explained

up to now, the computer power required for carrying out simulations
with coupled atmospheric-oceanic models that dynamically involve the
other subsystems as well (the so-called AOGCMs) is very considerable,
even with a limited resolution. It is so considerable that it turns out to be
extremely difficult to run these models for more than a few decades and
with a number of ensemble integrations sufficient for determining correct
statistics.
In this situation, on the one hand scientists felt the need to have
simplified models that could lead to longer integrations and also facilitate
the testing of the sensitivity to changes in fundamental mechanisms
(these tests are particularly onerous from a numerical point of view). On
the other hand, the low horizontal resolution and unsatisfactory results in
the reconstruction of the real climate on a regional scale (though
extended to subcontinental areas, such as the Mediterranean area,
northern Europe or Saharan Africa) led the scientists to consider models
that were more expressly dedicated to catching details on this scale.
The class of simplified climatic models includes several types of
models. We can only describe them concisely, in the list below.
Energy Balance Models (EBMs). These are the models with the
highest degree of simplification, because they regard the Earth's
atmosphere as a single point, and, in the course of time, evaluate the
global radiative balance (i.e. the balance between the incoming solar
radiation and the outgoing one), under the influence of changes in
 Climatic Models 171

greenhouse gases and sometimes also aerosols. A slightly more

advanced form of these models also evaluates the conveyance of
energy between different latitudes: in this case the evolution of the
temperature for each latitude band is obtained.
Radiative-Convective Models (R-CMs). These models consider the
radiative transformations that take place when energy is absorbed,
emitted or diffused, and the role of convection, which is basically
understood as a factor that modifies the vertical thermal profile of the
atmosphere, with a mechanism similar to the one mentioned in the
previous chapter (Section 7.7), which we may call "convective
adjustment". In R-CMs, it is possible either to regard the atmosphere
as a single vertical column or to introduce a horizontal spatial
distinction (though a rather rough one).
Statistical-Dynamical Models (SDMs). These models usually
combine the horizontal energy transfer of EBMs with the radiative
and convective approach of R-CMs. However, the transfer of energy
from the equator to the poles takes place in a slightly more
sophisticated way, on the basis of theoretical and empirical relations
in the flow between different latitudes. Parameters such as the speed
and direction of the wind are estimated by statistical relations, while
for obtaining an estimate of the horizontal diffusion of energy the
laws of motion are used.
Earth-system Models of Intermediate Complexity (EMIC). This is a
rather diversified class of models, in which the individual models all
have the characteristic of being an attempt to bridge the gap between
AOGCMs and the previously-described simplified models. They are
usually dynamical models characterised by certain simplifications
with respect to the AOGCMs, particularly in their atmospheric and
oceanic modules; but, just the same, they offer the possibility of
introducing a series of bio-geo-chemical cycles, such as the carbon
one, and of evaluating the model's sensitivity to changes in the
forcing factors and theoretical schemes much more easily and for a
longer simulated time, in comparison with what is possible with
AOGCMs.
As a rule, all these models are used for testing, in a simplified way,
new theoretical schemes and previously disregarded feedbacks (possibly
172 From Observations to Simulations

examining the various components of the Earth system one at a time),

and for obtaining scenarios over time lapses that are so prolonged as to
make it impossible to carry out simulations with fully coupled models. In
particular, once one of these simplified models has been validated, its
results can be rendered quantitatively consistent with the averaged results
of an AOGCM, by adjusting certain parameters with a fine tuning
operation15. This makes it possible, for instance, sometimes to use a
simplified model in place of a more complete AOGCM in the long-range
forecasting of climatic scenarios, at least in the cases in which interest is
focused on averaged global- or hemispheric-scale climatic values.
In the opposite case, i.e. when averaged data are not sufficient and
more precise information is needed, on a scale even smaller than that
solved by an AOGCM, it is necessary to adopt "regionalisation"
techniques that make it possible to carry out a more accurate, reliable
investigation on these scales. On a regional scale (by "region" here we
mean a part of a continent that is characterised by a comparative
homogeneity of climate), the climate is affected not only by the average
global values of certain variables and by the oceanic and atmospheric
circulation on a global or hemispheric scale, but also by some more local
forcing factors, for instance the presence of complex orographical
features, great lakes or practically closed sea basins such as the
Mediterranean Sea, the characteristics of the use of the soil, etc.
Moreover, the climate of a particular region may be heavily influenced
by cycles that take place elsewhere in the ocean or atmosphere: this is the
case of the far-reaching influence of the phenomena resulting from El
Nino and the North Atlantic Oscillation (NAO).
By studying these regionalisation techniques, we can hope to obtain a
more detailed simulation of the spatial structure of the temperature and
precipitation, a smaller-scale description of the atmospheric circulation
(e.g. convective systems, tropical cyclones, breeze circulation), and a
representation of the processes whose frequency in time is higher than

5
This term already appeared in the previous chapter (Section 7.6): it was used for
highlighting that part of artisanal activity present in meteorological models. Now it can
also be used for climatic ones.
 Climatic Models 173

that of processes that can be simulated by an AOGCM, e.g. frequency

distributions and intensity of precipitation.
In short, there are three types of techniques that make it possible to
obtain results on a regional scale: application of atmospheric models with
a high resolution that can be modified, if required; regional climate
models (RCMs); and downscaling statistical techniques. Whereas the use
of high-resolution global atmospheric circulation models for a few
decades of simulated time is still in the experimental stage and quite
expensive in terms of computer time, the use of RCMs and downscaling
statistical methods seems to be more established, and leads to rather
generalised improvements in the results.
Regional climate models are basically limited-area models "nested"
within an AOGCM. This means that the integration is carried out on a
limited region of the globe, with a high-resolution grid, and is driven by
the boundary conditions supplied by a global model. This way it is
possible to perform a detailed study of the climatic evolution of a certain
zone. To this day, the large-scale information given by the boundary
conditions is usually not affected by feedbacks due to the smaller-scale
evolution described by the RCM; however, experiments have been
started on a dynamical nesting in which the feedbacks of the regional
scale on the global one lead to the interactive running of the global
model together with the regional one, allowing for the feedbacks due to
the evolution described by the latter.
The statistical downscaling methods are based on the fundamental
assumption that the regional-scale climate is conditioned by two main
factors: the large-scale climatic state, and the regional or local
characteristics of the territory (orographic features, distribution of
land/sea boundaries, soil cover, etc.). At this point, if we manage to
develop a statistical model that links some large-scale characteristic
features to some regional- or local-scale variables over a reference
period, we will be able, over a certain period in the future, to use the
climatic forecasts of the large-scale characteristic features obtained by
means of an AOGCM and to find the values of the regional or local-scale
variables by means of the application of this statistical model. This way
we will be able to obtain smaller-scale climatic forecasts. To do this, we
can use several techniques, ranging from a basic multiple linear
174 From Observations to Simulations

regression to more sophisticated artificial-intelligence methods such as

neural network models.

8.7 Simulation Results

Now that we have established that climatic models are dynamical

systems with attractors and certain properties relative to trajectories in
the state space, and once we have outlined their structure, we can finally
proceed to analyse the main results yielded by their concrete application.
In doing this, we will present only some results that are common to all
currently existing models and are regarded as well-established by the
international scientific community. Moreover, the compactness of this
book and the fact that it is focused on the study of climatic changes in the
present and in the immediate future compel us to overlook some difficult
and prolonged simulations relative to paleoclimatic studies.
First of all, we would like the application of climatic models to help
us to shed light on a point that has appeared several times in this book:
the possibility of determining the causes of the climatic change that has
taken place during the last century, as revealed by the data reported in
Chapter 2. In the subsequent chapters our knowledge of the Earth system
has increased; we have become aware of the various factors that affect
the climate; and we have understood that the system is extremely
complex and that therefore it is not possible to carry out a linear causality
analysis, as is done, for instance, on a simple Galilean mechanical
system. This complexity was what led to the development of simulation
models, first for meteorology, then for the investigation of climate. These
models, starting from laboratory-validated basic laws, essentially
constitute an instrument for unravelling such an intricate skein.
Well, we have said that the first stage of the testing of a model
consists of the validation of the model in the reconstruction of the past
climate, either as an independent system (if it is possible to consider a
period in which the boundary conditions and external forcings can be
regarded as almost constant), or in a transient phase (if the boundary
conditions and external forcings are in evolution). As a matter of fact,
during the last century the changes in the natural and anthropogenic
 Climatic Models 175

influences that affect the Earth system have been so important as to put it
in evolutionary conditions: therefore, using the observational data for
these boundary conditions and forcing factors, it is possible to attempt to
reconstruct the climate by means of simulation models. As we have seen,
in this situation the statistics of the model that is applied is somehow
reduced (the trajectories in the state space no longer lie on a static
attractor), and it will be basically necessary to take ensemble integrations
into account.
With reference to Plate 9, which presents some reconstructions of the
global-scale temperature, we will now briefly examine the results of this
causality analysis. In the three parts of the figure, the red line represents
the anomalies observed in the global temperature with reference to the
average of the forty years between 1880 and 1920; the gray bands
represent the results of the reconstruction of this temperature that have
been obtained by means of four ensemble integrations of an AOGCM.
Part (a) refers to runs of the model in which only the purely natural
forcing factors, i.e. solar radiation and volcanic activity, have been
caused to change (according to actually observed values), while the other
forcings have been left at a constant value. In Part (b) the situation has
been inverted: values that are variable and have been actually observed
are considered for the partly or chiefly anthropogenic forcing factors,
such as greenhouse gases and sulphate atmospheric aerosols, while the
solar radiation and volcanic activity are kept at a constant value. In Part
(c), all the previously discussed forcing factors have been caused to
change, according to the values that have been observed.
The results presented in Plate 9 are particularly interesting. More
specifically, the anthropogenic forcing factors appear to be necessary for
a correct reconstruction of the global temperature data relative to the last
three decades. We must point out, however, that the reconstruction of the
whole series requires the introduction of the variability relative to both
types of forcings. The evolution of the global temperature is thus
reconstructed in a satisfactory way, and the use of climatic models also
leads us to infer that the anthropogenic forcings are a primary cause of
the change in the global temperature.
Now that we have seen how an AOGCM can reconstruct the global
temperature values of the last 140 years, we may pose the problem of
176 From Observations to Simulations

what happens in the reconstruction of the temperature in individual zones

of the Earth. If we limit ourselves to considering the temperature of the
air in large continental or oceanic masses, the situation is fairly
satisfactory; but as soon as we pass to the regional (i.e. sub-continental)
scale, the errors of this type of model become considerable: there are
typical errors of ±3-4°C in the temperature and between -40% and +80%
in the amount of precipitation, in comparison with the observed values.
In this field, we are helped by the regionalisation techniques, which lead
to a considerable reduction of the errors, particularly those in the
reconstruction of the temperature, which now remain within a range of
±2°C.
Once the models have been validated in the reconstruction of the past
climate, it is possible to tackle the problem of forecasting future climatic
scenarios on the basis of the available socio-economic scenarios, as
shown in Figure 11. Because of the variety of socio-economic and
emission scenarios, in practice it has not yet been possible to run the
climatic models that are most expensive in terms of required computer
resources (i.e. AOGCMs) for all these scenarios. In order to evaluate the
prevailing tendency, over the next 100 years, of average quantities such
as the global temperature for all the available scenarios, simplified
models have been applied to this forecast, obviously only after having
optimised them with reference to an AOGCM, by means of fine tuning
operations, as described in the previous section.
An example of the results thus obtained is presented in Plate 10,
which indicates the global temperature forecast up to the year 2100. The
graph explicitly shows the curves relative to the evolutions of the
temperature, forecast on the basis of nine of these scenarios, with the
uncertainty bars relative to the first six. The dark-gray band includes the
set of simulations relative to a single model, with data from no less than
35 different scenarios, whereas the light-gray band includes the set of
results obtained by 7 different models with the data of all the scenarios. It
is evident that all the scenarios suggest a more or less marked rise in the
global temperature in the models. This analysis leads to the conclusion
that, with a reasonable degree of reliability, in the year 2100 we may
expect a rise in temperature ranging from 1.4°C to 5.8°C with respect to
the value of 1990.
 Climatic Models 111

Though these results are extremely significant, it would be desirable

to obtain more, particularly as regards the values of other quantities and
phenomena whose degree of intensity and more or less frequent
appearance contribute to a change in the climate of a certain zone. For
this purpose, obviously, we must resort to the results of simulations
performed by means of AOGCMs. So, though the currently available
computer power has not allowed a wide-spectrum analysis (on all the
scenarios that have been surmised) like the one we have just presented
for the global temperature, it will be worthwhile in any case to briefly
summarise the evaluations of the future climate obtained by means of
simulations with fully coupled models. We will condense these forecasts
into the following basic points:
the temperature of the air near the surface (both of the land and of
the sea) will rise;
the temperature of the air should rise more during the night than
during the daytime, so there should be a drop in the temperature
range;
the temperature of the sea will rise;
the whole troposphere will undergo a warming process;
the low stratosphere will tend to cool down;
the amount of sea ice will decrease;
the snow cover on the continents will also decrease;
the water vapour in the troposphere will increase;
the so-called "heat index"16 on the continental zones will tend to rise,
and this will lead to worse conditions of meteorological and climatic
comfort for mankind, at least at the low and medium latitudes;
the precipitation at the low and medium-low latitudes will decrease
(this, combined with the rise in temperature, will lead to the risk of
an increase in the desertification of certain areas);
the precipitation at the high and medium-high latitudes will increase;
some models forecast an increase in the frequency and intensity of
tropical cyclones;
some models forecast an increase (albeit a slight one) in the
frequency and intensity of heavy precipitation at mid-latitudes.

'This is an index that is obtained by combining temperature with humidity.

178 From Observations to Simulations

The forecasts of the AOGCMs have been refined, of late, by the use
of regionalisation techniques. These recent studies make it possible to
evaluate the changes in temperature and precipitation regimes in
individual subcontinental regions. For instance, in these simulations,
Alaska, northern Canada and Greenland undergo an above-the-average
warming during the winter, while precipitation decreases in a rather
marked manner in the Mediterranean area during the summer. Recent
studies have also revealed that the melting of ice may contribute a great
quantity of fresh water to the northern Atlantic Ocean, and may deviate
or lessen the thermohaline circulation there; so, in a context of general
wanning, northern Europe may undergo a regional-scale cooling17.
So, besides the practically univocal forecasts of a rise in temperature,
changes in other important parameters are also forecast. For instance,
though these results should be regarded only as a general indication, in
the situation of the Mediterranean area (and in Italy in particular), we can
surmise a change in the precipitation regime, with precipitation whose
total amount will perhaps be less abundant, but whose appearance will be
concentrated in single, more violent episodes, with non-beneficial
effects.
To conclude this overview of results, we would like to point out that
the forecasts we have concisely summarised have consequences also on
phenomena or processes that are usually not perceived as strictly
climatic. For instance, the rise in temperature and the decrease of the
expanses of ice both in the sea and on the continents directly affect the
sea level to be expected during the next 100 years. Still referring to the
variety of scenarios surmised for the economic situation and the
emissions, Plate 11 shows some forecasts of the sea level, which will be
clearly rising up to 2100. The six lines are relative to six different
scenarios, while the dark-gray band refers to the averages of the runs of
some AOGCMs, and the light-gray band indicates the scatter of the
individual runs of the models.
Obviously what has been presented in this section is a summary of
the forecasts that are believed to be rather well-established. It is clear that
in a discipline like the study of climatic changes, characterised by an

The reader is reminded of the previously cited Wood et al. (1999).

 Climatic Models 179

"explosion" of scientific studies, almost every day an article appears in

some specialised journal, bringing further contributions to the
understanding of phenomena and to their forecasting in the future. In
particular, some phenomena that had been previously neglected or dealt
with in a perfunctory manner in the models are now analysed more
carefully, particularly with reference to their importance in the
determination of future climatic scenarios. Nowadays, for example, the
need is felt for a more thorough study of the consequences of the water
cycle and of the use of the soil, particularly in relation to the indirect
effects of atmospheric aerosols on clouds.

8.8 Further Remarks about Climate Change and Its Study

What we have presented in this chapter has made it possible at last to

tackle the problem of the study of climate change by means of an
instrument suitable for exploring the complexity of the Earth system.
Some signs of the fact that certain influencing factors might turn out to
be fundamental in determining climatic changes had already been
pointed out starting from Chapter 3. Then the theoretical remarks in
Chapter 4 contributed, at least partly, to the determination of the action
of some of these single factors, considered individually. Only now,
however, within a climatic model, we have been able to evaluate the
complex, non-linear mixture of these diverse concauses of climatic
changes. When this was being done, the anthropogenic influences on the
evolution of the climate were revealed in all their importance: we saw, in
particular, that these influences account for the net rise in the global
temperature during the last thirty years.
Have we thus identified the increase in anthropogenic emissions of
greenhouse gases as the main cause of the changes observed during the
last few decades? Undoubtedly the physical interaction between these
gases is well known, and their role as an important influence on the
climate does not seem to be very controversial. Obviously, as we have
seen, the elaboration of all-embracing climatic models is still a rather
distant goal, since there still exist some processes that need to be
understood better and evaluated more carefully in the models. This
180 From Observations to Simulations

means that, in the future, the inclusion of new feedback mechanisms may
slightly reduce the importance of this role of human activities. We should
point out, however, that the main anthropogenic feedbacks are all
positive: an example is the deforestation activity, which tends to
eliminate absorbers of C0 2 and leads to a further increase in the
concentration of C0 2 in the atmosphere, with the relative consequences
on trapped heat and temperature.
During the last few years, some of these less-known mechanisms
have been investigated better: we should mention the role played by the
oscillations of ENSO on the interannual variability of the climate, or the
more accurate analysis of the carbon cycle on land and in the oceans. At
the present stage of climatic modelling researches, two points that have
not been clarified completely yet are the role of clouds in the feedback
cycles and how cloud formation is influenced by the indirect effects of
changes in aerosol concentration.
In short, the system we are dealing with is complex, and this cannot
be concealed. The future is quite likely to bring further progress in
knowledge and simulation, and this may partly change our vision of
climatic processes. However, what we know at present seems sufficient
to allow us to assert that the influence of human activities on climate
change is considerable, and that impact studies (which are consequent
upon the forecast climatic scenarios, but will not be discussed in this
book) show that it is necessary to act at once in order to avoid the most
objectionable consequences of climatic changes, such as an excessive
rise in sea level that might force the population to leave some coastal
territories, with mass departures from the most threatened zones.
In this context, one of the efforts modellers should make is to
improve the theoretical understanding of the instrument they are using,
endeavouring to come as near as possible to the rigour of the science of
complex systems. This would lead, in particular, to a greater awareness
of the intrinsic qualities and limits of climatic models. In this
perspective, for instance, modellers should delve further into the
"chaoticity" of the system (i.e. the sensitivity of its future evolution to
errors in the determination of its initial state), by analysing the zones of
the state space where two neighbouring states suddenly diverge in an
exponential manner (the so-called "bifurcation points"). While in
 Climatic Models 181

meteorology — where the goal is a forecast, as deterministic as possible,

of the evolution of a state in the future — these bifurcations lead to an
actual unpredictability after a certain time lapse, in climate studies they
lead the system to classes of possible states, whose knowledge
determines the statistics of the system. It is reasonable, therefore, to
assert that while in meteorology modellers "suffer" bifurcations, in
climate studies they can "exploit" them, in order to determine more
extensive climatic statistics over a certain time lapse.
In particular, the change in the frequency (and therefore probability)
of the occurrence of certain classes of states in comparison with others
may be due to a change in the external forcings: in the past, rather
sudden climatic changes occurred several times (the reader is reminded
of Plate 7, which shows the comparative swiftness of the transitions from
glacial eras to interglacial periods). In a non-autonomous system like
this, an analysis of the behaviour of the evolving trajectories (on a no
longer static attractor) may allow us to understand these sudden
transitions.
In the panorama we have thus outlined, the resort to ensemble
integrations on the runs of an individual model is essential. At present, in
actual fact, ensemble integrations are also beginning to be carried out
starting not only from different initial states, but also from different
models: this makes it possible to test the reliability of the results obtained
and, in some cases, to increase the confidence that can be placed in these
results. The ideal objective at which efforts would be aimed if this were
possible is that of achieving a forecast of the distribution of probabilities
for the states of the system in the future; but this is obviously impossible,
because of the multidimensionality of the system. In this context, then, a
result that appears to be attainable is an accurate determination of the
classes (or scenarios) of possible climates under the influence of forcing
factors and boundary conditions of the system. The key to all this is in
the statistical analysis of the dynamical treatment carried out with
climatic models.
After having emphasised that climatic models possess to the utmost
degree the characteristics of simulation models that integrate the
knowledge coming from different disciplines and make it possible to
carry out an experimental activity of construction of possible future
182 From Observations to Simulations

worlds (scenarios), at this point we would like to conclude this chapter

with a few more remarks of a conceptual rather than practical nature.
As we have already stated, models that simulate the behaviour of
complex systems do not have the purpose of corroborating or falsifying
the equations of the model: this activity is left to laboratory experiments
in controlled and simplified conditions. The purpose of these models is to
reconstruct the complexity of reality and to simulate phenomena and
processes on a macroscopic scale. As concerns us, then, what does
validating a climatic model mean? Is it possible to falsify a model of this
type?
Whereas in the realm of experimental laboratory activity there is the
possibility of carrying out crucial experiments that — through a
comparison with the behaviour of the real system — make it possible, for
instance, to draw a distinction between alternative theories or models, in
numerical experimentation, tests can be carried out in order to evaluate
the physical consistency of several models in extreme situations, though
these situations do not usually appear in nature.
In the validation of climatic models, in particular, besides checking
the physical self-consistency of the scheme, an attempt is usually made
to understand the nature of the interactions in the model. The comparison
with the data observed in reality, however, is usually carried out in a
"filtered" manner, that is, for instance, through a comparison with the
meteorological and climatic analyses, which, as we have explained in the
previous chapter, stem from a combination of data and results of
forecasting models. So there remains a difficulty in falsifying a model in
the "Popperian" sense of the term18.
The situation, therefore, is not as clear as in real laboratory
experimentation, which, however, is possible in very simplified
conditions and for studying individual phenomena or processes. So the
example of the study of the climate in a complex system like that of the
Earth is also the paradigm of a new approach to science — a science that

The reader is reminded that a theory or model is falsified if even only one particular
situation is found in which the behaviour prescriptions stemming from the theory or
model clash with the data obtained from natural observations or laboratory experiments.
On this subject, consult Popper (2002), where the possibility of falsification is actually
used as a criterion for a boundary between science and non-science.
 Climatic Models 183

probably would not be regarded as science by Popper: see note 18. In

particular, the choice among the various models cannot be made simply
on the basis of the possibility of their falsification.
This last remark encourages us even more to feel that, with the
analysis of climatic models, we have reached the threshold of a modern
science, where, in order to understand the complexity of the natural
environment, we will inevitably be forced to relinquish a classical
approach to science.
 Chapter 9

Conclusions and Prospects

Now that we have reached the end of this journey in the complex world
of the atmosphere and Earth system, it is expedient for us to take stock of
what we have discovered and to evaluate some prospects of the future
development of scientific knowledge in this field.
The route followed in this book showed the complexity of the natural
environment in which we live, at least as regards the part relative to
meteorological and climatic phenomena and processes. At the same time,
we became aware of the diversity of the systems under consideration, in
comparison with the ones we were able to deal with in a school
laboratory and whose operation we could easily understand, often thanks
to the application of the Galilean experimental method. Understanding
the behaviour of the atmosphere and of the Earth system required a
breakthrough similar to the one that took place in the seventeenth century
with the adoption of the method devised by Galilei: today, simulation
methods are the key for compounding phenomena and elementary
processes, and for thus reconstructing the complexity of reality in the
virtual, controllable world of computers.
In our domain, we have seen that the application of meteorological
and climatic models has led to immensely important practical results, in
particular to forecasts of considerable climatic changes over the next
hundred years.
Moreover, in this excursus on the science of weather and climate we
made several forays into the domain of theoretical physics. In doing so,
we had the chance to recognise, for instance, that the concept of univocal
deterministic forecast can no longer exist for a complex system, and that
a probabilistic description has rightfully come into the (once

185
186 From Observations to Simulations

unchallenged) realm of determinism. An important concept like that of

attractor was similarly added to the statistical definition of a climate, and
turned out to be extremely useful for the determination of the latter. On
the whole, besides changing scientific practice in a substantial way,
simulation models (particularly meteorological and climatic ones) also
contributed to change our outlook on nature.
In this book, we have endeavoured to outline a conceptually
meaningful vision of the scientific approach to the study of the weather
and climate. In doing so, on the one hand we explained some scientific
findings and pieces of knowledge, while on the other hand we discussed
conceptual and epistemological issues1 relative to the nature of the
systems under examination, to the methods used for studying them, and,
ultimately, to problems inherent to the intelligibility of nature.
At this point, in order to complete the picture, we should perhaps
present some conclusions and discuss the prospects of future
development with reference to two important points: the results of
climatic models, with the consequent actions that can be undertaken; and
the present and future developments of the modelling paradigm applied
to the study of the weather and the climate.

9.1 The Results of Climatic Models and "What Should We Do?"

In the previous chapter, we discussed climatic models, and, in particular,

presented the basic results attained through their application to the
forecasting of future climatic scenarios. Without going again into the
details of these results, we may mention the following: the consistency
with which these models forecast a more or less marked rise in the global
temperature over the next hundred years; the consequences, on a
continental or regional scale, of the possibility of an increase in extreme
events (ranging from prolonged drought to conditions of heavy
precipitation); and the rise in the sea level.
We already stated that the AOGCMs are still in the development
stage and do not yet consider all the processes that act in the Earth

'See Note 5 in the introductory Chapter 1.

 Conclusions and Prospects 187

system: this will be briefly discussed in the second part of the present
chapter. However, AOGCMs show the influence of human activities on
the climate2. In this perspective, does anyone intend to do anything? In
particular, since the problem is a global one, has an international
negotiation been activated?
The answer to these questions is obviously "yes". Under the aegis of
the United Nations, in December 1997 a document was drawn up in
Japan, the (by now well-known) Kyoto Protocol, which establishes an
international-level commitment for the reduction of some greenhouse
gases, including C0 2 . The rationale behind this document obviously
stems from the acknowledgement of the contribution of these gases to
the global warming and of the consequent need to limit their emission,
compatibly with the development requirements of the poorest countries
in the world. In practice, once the increase in greenhouse gases has been
identified as a cause of climatic changes, an effort is made to stop or
slow down the warming process by acting on this cause: it is an attempt
at a "mitigation" of the effects.
As a matter of fact, the actual situation of the international
negotiation has reached a deadlock: the Kyoto Protocol has come into
force on 16 February 2005, after its ratification by Russia. With the entry
of this country, the last of the thresholds laid down for the Protocol to
become legally valid has been passed: now the sum of the emissions of
all the countries that have ratified the Protocol is more than 55% of the
total emissions detected in the reference year 1990. However, it seems
that the stance of the Bush government against the ratification of the
Protocol by the United States is thwarting the international efforts: the
United States are currently emitting more than one fourth of the annual
quantity of greenhouse gases emitted in all the world. We must also add
that many scientists consider the Kyoto Protocol important more as an
expression of international will than as an actual measure for mitigating
the phenomenon of global warming: it has been pointed out by several
experts that more drastic measures would be necessary.

2
This is underscored not only by the last report of the IPCC (Houghton et al. (2001)), but
also by a more recent account of the National Academy of Sciences of the U.S.A. for
President Bush (see National Academy of Sciences (2001)).
188 From Observations to Simulations

In this situation, some researchers propose engineering solutions for

confining carbon dioxide underground or in the oceans: for the time
being, these solutions are not very feasible or safe, and are financially too
exacting. The American government, on their part, have initiated a vast
programme of studies on the impact both of anthropic activities and
natural influences on climatic changes, maybe in the secret hope of
demonstrating that natural elements and not mankind are the main causes
of global warming. If this were true, the present pattern of economical
development might be able to grow further without the hindrance of
negative effects on the ecosystem, at least as regards the climate.
What we have explained in the previous paragraph essentially allows
us to understand that human activities lead to an increase in greenhouse
gases and therefore promote a global-scale rise in temperature, creating
feedbacks that are all intrinsically positive. Undoubtedly the magnitude
of this effect in comparison with that of other changes in natural forcings
is open to question, but a shred of common sense and a "precautionary
principle" (advocated by many people) should lead us in any case to
reduce the extent of these anthropogenic causes, also because the Earth
system, being highly non-linear, does not always respond gradually to
changes in forcing factors.
From this angle, the "wait and study" tactics of the American
government causes perplexity. Many people feel that, instead of
spending energies and resources in studies for mitigation projects that
might turn out to be practically unfeasible, it would be better to
concentrate on "adaptation" studies, i.e. on the evaluation of how to
reduce the impacts of climatic changes (by now considered inevitable) on
the territory and population. In actual fact, now international efforts are
focused particularly on studies of this type.
This subject, of course, is partly beyond the scope of this book. It is
interesting, however, to discuss it, and to show, for instance, that there is
a connection between political stances, with their models of socio-
economic development, and scientific climatic activity, which, like all
sectors of modern science, increasingly depends on the financing of
research activities.
 Conclusions and Prospects 189

9.2 The Future of Models for Studying the Weather and Climate

Returning to the main theme of this book and to the modelling approach
we have presented for analysing and forecasting the meteorological and
climatic behaviour of the Earth system, we can now concisely summarise
the importance of this paradigm, highlighting its strong points and
weaknesses, and then proceed to discuss the prospects of future
development of research in this field.
From a theoretical point of view, we have repeatedly emphasised the
conceptual importance of the simulation paradigm applied to systems
such as the atmosphere or Earth system, its capability to reconstruct
reality in a virtual, controllable laboratory, and all the advantages that
can be associated to this in terms of possibilities of experimentation for
understanding complex phenomena and processes.
From a practical point of view, as regards the study of the weather,
we have seen that meteorological models are the only instrument we
have for obtaining detailed forecasts beyond 24 hours. Their
deterministic structure, combined with the use of ensemble integrations,
allows us to evaluate their accuracy and reliability at the various
forecasting time limits. The drawbacks of these models are revealed in
the following circumstances:
in long-range forecasts, when the theoretical predictability limit
(which is intrinsic to a non-linear model and marks the onset of
deterministic chaos) is reached;
in very short-range forecasts, when the problems due to the initial
unbalancing of the fields (the previously mentioned spin-up effect)
and the long computer time required for the analysis and the model
do not make it possible to obtain reliable forecasts that can be used
within an operational time;
on a local scale, because of the limited resolution of the models3;
when there is the need to obtain forecasts of some variables that the
model does not supply directly, such as the meteorological visibility
(important, for instance, for determining fog situations).

3
For instance, this problem is particularly felt in aeronautics, where forecasts on a
particular site, i.e. the runway of an airport, are required.
190 From Observations to Simulations

As regards climatic models, we would like to emphasise that in this

sphere we appreciate, even more, the capability of the simulation method
to interconnect phenomena and processes that arise within different,
mutually interacting systems. From this point of view, it is obvious that
the greatest limit that can be found at present in climatic models is
precisely that of being still rather far from having completed the
reconstruction of the real system: climatic models currently include in
their schemes only the interactions that are considered most important.
Other limits are obviously due to the low horizontal resolution of these
models: this problem has been only partly solved by the regionalisation
techniques.
Considering the excellent results obtained with the application of the
simulation method to weather forecasts and the study of the climate, the
meteorological and climatic research centres scattered all over the world
persist in applying the simulation paradigm, in quest of constant
improvements, which also involve the elimination of the drawbacks we
have listed above. An example that refers to meteorology is the
elaboration of coupled atmosphere-ocean models, which, combined with
the use of ensemble integrations, is now leading to an extension of the
predictability time lapse, to the point of obtaining seasonal forecasts in
terms of scenarios for some quantities such as temperature and
precipitation. Climatically important examples are the attention that is
being given of late to previously-neglected interactions, such as those
relative to the influence of cosmic rays and solar variability on cloud
formation (an attempt is being made to include them in climatic models),
and the development of projects (such as the Japanese Earth Simulator)
in which new-generation supercomputers and high-resolution models, as
all-embracing as possible, will be used for a more complete study of the
Earth system.
In this attempt at working out increasingly complex and sophisticated
models, the scientific enterprise that involves meteorology and climate
science has acquired the standing of a "Big Science". This term indicates
a science whose development requires very substantial funding
(sometimes unaffordable for a single country), which in our case is used
basically for managing major computer centres and for supporting large
teams of researchers and technicians.
 Conclusions and Prospects 191

In my opinion, when a level of applicative development like this is

reached, there begins a process of science policy and sociology whereby
the same type of research tends to be self-nourished in the future. To
explain this more clearly: since supporting these centres for the
researching and operational development of models is quite expensive,
the people who finance them and those who manage them must
guarantee a constantly improving final product; therefore, once a
paradigm has been found that works (and ensures a constant, though
maybe slight, progress), it is unlikely that this paradigm will be
abandoned in order to investigate less-explored paths. Moreover, for the
individual researchers (whose career is linked to the number, and, partly,
to the quality of their publications on international scientific journals) it
is easier to publish researches performed with standard and extensively
accepted methods. This fuels what Thomas S. Kuhn calls "normal
science"4, to the detriment of investigations that are more innovative but
are also more at risk of being unsuccessful.
I would like, obviously, to prevent any misunderstanding: the
simulation paradigm we have discussed has so many important positive
aspects, both from the conceptual point of view and the applicative one,
that the constant improvement of meteorological and climatic models can
only be looked on with favour. The effect we have just described does
nothing but further amplify the success of these models.
In spite of this, some attempts have been made recently to tackle
certain meteorological or climatic problems outside this paradigm, and,
in actual fact, some techniques for bridging certain gaps in these models
are emerging. In many cases, the authors of these works are researchers
who do not belong to big research centres but enjoy a greater scientific
independence and can explore more original, though more risky, paths5.

4
Kuhn(1996).
5
For instance, most Italian scientists, including the author, are in this situation. The fact
that the persistent shortage of funds in the Italian world of research does not make it
frequently possible to compete with big foreign research groups leads the individual
scientists to deal with original subjects or methods, which allow them to carry out high-
level (and perhaps more creative) researches despite the fact that their resources are
scanty.
192 From Observations to Simulations

In actual fact, there exist some domains of meteorology and

climatology where the models we have described are in difficulty at
present, and presumably will be so in the future as well. I refer, for
example, to very short-range local meteorological forecasts. Moreover,
some phenomena of remote correlation, the so-called "teleconnections"
(for instance, between the NAO index in the Atlantic Ocean and the
winter temperatures and precipitation in Europe), suggest that there exist
some macroscopic variables that determine the climate on a certain zone
of the globe.
On the other hand, the method for the reconstruction of reality in a
simulation model is extremely sensitive if it is applied to a non-linear
system characterised by a great number of feedbacks: we have already
mentioned the somewhat artisanal activity with which the various
parameterisation schemes are balanced. This set of problems cannot be
solved in a simple way, and leads to uncertainties that are often hard to
quantify.
All these considerations have led some researchers to choose a
different path from that of the classical models we have described: a path
that does not presume to replace these models, but endeavours to bridge
their gaps and to catch the evolution of the system under examination in
a more direct manner, without getting "bogged down" in the meanders of
the balancing of the various parameterisation schemes and of the many
feedback cycles. For instance, these researchers use artificial-intelligence
methods such as neural network models, non-linear artificial systems that
mimic some simple functions of the human brain6.
With these models, the simulation paradigm is abandoned: here there
undoubtedly exist some elements that have a one-to-one correspondence
with the real system (for instance, the meteorological variables), but they
are not what evolves with time. Learning on the basis of previous
experiences supplied by the researcher, and readjusting the connections
between their own artificial neurons, these models manage to find non-

6
Some examples of applications of this type can be found in regionalisation techniques, in
very short-range forecasts (see, e.g., Pasini et al. (2001)), in the ENSO forecast (see, e.g.,
Tangang et al. (1998)), and in the analysis of climatological data (see Pasini et al.
(2005)).
 Conclusions and Prospects 193

linear correlation laws among the various states of the real system and to
account for its characteristics (and possibly also for its evolution).
This opens the way to a different chapter in the history of modelling,
no less fascinating than the one we have dealt with here. For instance,
imagine a little artificial brain that finds diagnostic or evolutionary laws
in cases that are so complex as to remain unintelligible to us even after
the application of our theoretical knowledge of the problem7. At the same
time, the neural network model may also indicate the fundamental
variables for a dynamical description of the real system under
examination.

At this point, however, the subject becomes excessively vast, and I

should start writing another book. Perhaps I may manage to do so some
day.

7
A case of this type is described in Pasini and Ameli (2003).
 Bibliography

Alley, R.B. (2002). The Two-Mile Time Machine: Ices Cores, Abrupt Climate Change,
and Our Future, Princeton University Press.
Arrhenius, S.A. (1896). On the Influence of Carbonic Acid in the Air upon the
Temperature of the Ground, Phil. Mag., 41, pp. 237-276.
Barrow, J.D. (1988). The World Within the World, Oxford University Press.
Barrow, J.D. (1991). Theories of Everything: The Quest for Ultimate Explanation,
Clarendon Press.
Barrow, J.D. and Tipler, F.J. (1986). The Anthropic Cosmological Principle, Oxford
University Press.
Bateson, G. (1980). Mind and Nature. A Necessary Unity, Bantam Books.
Buizza, R. (2001). Chaos and Weather Prediction — A Review of Recent Advances in
Numerical Weather Prediction: Ensemble Forecasting and Adaptive Observation
Targeting, IlNuovo Cimento, 24C, pp. 273-301.
Einstein, A. and Infeld, L. (1967). Evolution of Physics, Free Press.
Fermi, E. (1937). Termodynamics, Dover Publications.
Galilei, G. (1967). Dialogue Concerning the Two Chief World Systems: Ptolemaic and
Copernican, University of California Press.
Ghirardi, G.C. (2003). Sneaking a Look at God's Cards: Unraveling the Mysteries of
Quantum Mechanics, Princeton University Press.
Greco, J. (1998). Introduction: What Is Epistemology?, The Blackwell Guide to
Epistemology (J. Greco and E. Sosa eds.), Blackwell.
Hamblyn, R. (2001). The Invention of Clouds: How an Amateur Meteorologist Forged
the Language of the Skies, Farrar Straus Giraux.
Hawking, S. (2002). The Theory of Everything: The Origin and Fate of the Universe,
New Millennium Press.
Hodell, D.A., Curtis, J.H. and Brenner, M. (1995). Possible Role of Climate in the
Collapse of the Classic Maya Civilization, Nature, 375, pp. 391-394.
Hood, B.M. (2004). Children's Understanding of the Physical World, The Oxford
Companion to the Mind (R.L. Gregory ed.), 2nd edition, Oxford University Press.

195
196 From Observations to Simulations

Houghton, J.T., Ding, Y., Griggs, D.J., Noguer, M., van der Linden, P.J., Dai, X.,
Maskell, K. And Johnson, C.A. (eds.) (2001). Climate Change 2001: The Scientific
Basis, Cambridge University Press.
Krishnamurti, T.N. and Bounoua, L. (1996). An Introduction to Numerical Weather
Prediction Techniques, CRC Press.
Kuhn, T.S. (1996). The Structure of Scientific Revolutions, 3rd edition, University of
Chicago Press.
Laplace, P.S. (1820). Theorie Analitique des Probabilities, Courcier.
Lionello, P. (2005). Oceans: Their Motions and Role in Climate, Foxwell and Davies.
Lorenz, E.N. (1994). The Essence of Chaos, University of Washington Press.
Mandelbrot, B. (1982). The Fractal Geometry of Nature, Freeman.
National Academy of Sciences (2001). Climate Change Science: An Analysis of Some
Key Questions, National Academy Press.
Pasini, A. and Ameli, F. (2003). Radon Short Range Forecasting through Time Series
Preprocessing and Neural Network Modeling, Geophys. Res. Lett., 30 (7), 1386.
Pasini, A., Lore, M. and Ameli, F. (2005). Neural Network Modelling for the Analysis of
Forcings/Temperatures Relationships at Different Scales in the Climate System,
Ecol. Mod. (in press).
Pasini, A. and Pelino, V. (2000). A Unified View of Kolmogorov and Lorenz Systems,
Phys. Lett, A275, pp. 435-446.
Pasini, A., Pelino, V. and Potesta, S. (1997). Evidence of Structured Brownian Dynamics
from Temperature Time Series Analysis, Nonlin. Proc. Geophys., 4, pp. 251-254.
Pasini, A., Pelino, V. and Potesta, S. (2001). A Neural Network Model for Visibility
Nowcasting from Surface Observations: Results and Sensitivity to Physical Input
Variables,/. Geophys. Res., 106(D14),pp. 14,951-14,959.
Pease, C.B. (1994). Satellite Imaging Instruments, Wiley & Sons.
Petit, J.R. et al. (1999). Climate and Atmospheric History of the Past 420,000 Years from
the Vostok Ice Core, Antarctica, Nature, 399, pp. 429-436.
Philander, S.G. (2004). Our Affair with El Nino: How We Transformed an Enchanting
Peruvian Current into a Global Climate Hazard, Princeton University Press.
Popper, K.R. (2002). The Logic of Scientific Discovery, 15th edition, Routledge.
Tangang, F.T., Tang, B., Monahan, A.H. and Hsieh, W.W. (1998). Forecasting ENSO
Events: A Neural Network-Extended EOF Approach, J. Clim., 11, pp. 29-41.
Wood, R.A., Keen, A.B., Mitchell, J.F.B. and Gregory, J.M. (1999). Changing Spatial
Structure of the Thermohaline Circulation in Response to Atmospheric C0 2
Forcing in a Climate Model, Nature, 399, pp. 572-575.
Yourgrau, W. and Mandelstam, S. (1979). Variational Principles in Dynamics and
Quantum Theory, Dover Publications.
 Index

absorption, 60, 62, 63, 65, 69, 70, 74, balance laws, 53
86,87, 157, 158, 168 bifurcation, 87, 146, 180, 181
adaptation, 4, 188 Big Science, 190
ADEOS-II, 20 blackbody, 59-61,69
adiabatic, 72, 80, 81, 100, 122, 143 Bohr, 113
aerosol, 21, 65-68, 82, 158, 160, 168, Boltzmann, 161
171, 175, 179, 180 boundary conditions, 41, 100, 106, 108,
air mass, 55, 79, 81, 105, 106, 144 110, 111, 120, 121, 128, 150, 152,
air warming, 38-40, 73, 82, 83, 88 161-168, 173-175, 181
albedo, 22, 158 Boyle's law, 54
algorithmic compression, 52-54 Bush, 187
algorithmic reduction, 103, 105, 111,
149 carbon cycle, 6, 22, 65, 157, 158, 160
analysis procedure, 128, 131 carbon dioxide (COz), 25, 46-50, 62,
anomalies, 29, 43, 175 63,65,66,87,88, 158, 164, 180,
anthropogenic emissions, 48, 66, 167, 187, 188
179 causal relationships, 2
anticorrelation, 43 cause-effect interactions, 83
Archimedes' principle, 41, 71, 79 cause-effect relationships, 54, 82, 88
Aristotelian physics, 90, 93 centrifugal force, 79
Aristotle, 2, 92-94 circular causality, 3, 84
Arrhenius, 63 climate, 11, 36
asymptotic behaviour, 154, 155, 163 climatic forecast, 150, 151, 165
atmosphere warming, 78 climatic model, 121, 149, 160-170,
Atmosphere-Ocean General Circulation 174-176, 179, 181, 183, 185, 186,
Model (AOGCM), 160, 170-173, 190, 191
175-178, 186, 187 climatic variability, 163
attractor, 151-155, 161-167, 174, 175, coexistence laws, 54, 152
181, 186 conduction, 60, 70, 71, 73, 77, 78, 83
autonomous (system), 161, 162, 164, continuity equation, 122, 143
166, 167 control runs, 146, 164

197
198 From Observations to Simulations

convection, 60, 71-73, 77, 83, 89, 143, equation of state of gases, 121
160, 171 ERS-2, 20
conveyor belt, 78, 86 evolution equations, 2, 130
core samplings, 45, 46, 47 evolutionary equations, 53, 150
Coriolis force, 79 evolutionary laws, 54, 119, 152, 156
correlations, 42, 43, 50 experiment, 95-98, 100, 103, 106,
corroboration, 107 108-110, 112,114
Coulomb, 105 explanatory scheme, 38-42, 45, 50, 78,
cross-correlation, 42, 43, 50 95
cryosphere, 157, 160 explicative scheme, 52, 68
extreme events, 1,4, 12, 27
decomposition, 98, 99, 107
deforestation, 65, 157, 159, 164, 180 falsification, 182, 183
desertification, 177 feedback, 4, 65, 81, 83-87, 89, 107,
determinism, 54, 137, 147, 186 108, 110, 112-114, 120, 121,135,
deterministic chaos, 6, 87, 134, 136, 143, 149, 150, 156, 159, 161, 164,
142, 146, 151, 155, 163, 189 168, 169, 171, 173, 180, 188, 192
diagnostic law, 54 final cause, 2
diagnostic equation, 53, 121, 135, 150 finalistic thinking, 2
differential equations, 2 fine tuning, 135, 172
direct effect, 67 first guess, 130, 131
discretisation, 123-125 First Law of Thermodynamics, 72, 122
domestic example, 41, 70 flow adjustment, 166
domestic observations, 89, 97 forcing, 120, 128, 150, 152, 154,
downscaling, 173 161-169, 171, 172, 174, 175, 181,
dynamical system, 162, 165, 167, 174 188
fossil pollen, 25
Earth Simulator, 190 four-dimensional analysis, 18, 131
Earth-system Models of Intermediate fractal, 155, 156
Complexity (EMIC), 171 future scenarios, 113, 151, 168, 176,
ECMWF, 132 179, 186
efficient cause, 2
energy balance, 64, 68 Galilean experimental method, 5, 89,
Energy Balance Model (EBM), 170, 91,99-101, 107, 110, 185
171 Galilean mechanics, 33, 90
ensemble integrations, 134,140-142, Galilei, 9, 53, 90, 91, 93-98, 103, 104,
144-147, 151, 163, 164,166,170, 107, 112, 185
175, 181, 189,190 geostationary satellites, 17, 18, 20, 130
El Nino Southern Oscillation (ENSO), glacial eras, 28, 46, 181
27,28,30, 180,192 global temperature, 48, 50, 175-177
EN VIS AT, 20, 21,22 global warming, 31, 47, 48, 87, 187,
epistemology, 6 188
 Index 199

gradient force, 79 linear causality, 3, 4, 84, 174

grand duke Ferdinand II, 9 linearity, 3
grand duke Peter Leopold, 10 Lorenz, 136
greenhouse effect, 58, 63, 64, 68 Lorenz attractor, 155
greenhouse gases, 21, 63-65, 69, 87,
88, 167, 168, 171, 175, 179, 187, material model, 103, 104
188 mathematical model, 103, 105-109
grid, 123-132, 139, 140, 160 Maxwell, 59,161
grid spacing, 124-127, 143, 144 mental model, 103, 104
Gulf Stream, 78, 86 meteorological and climatic
observations, 12, 51
heat capacity, 74, 76 meteorological model, 117, 119, 120,
heat index, 177 124, 131, 133, 134, 139, 142, 144,
Hertz, 59 147, 150, 155, 160, 162, 172, 185,
human activities, 47, 48, 65, 88, 157, 189, 191
159, 167, 180, 187, 188 METEOSAT, 18
Huygens, 58, 59 methane (CH4), 25, 46-50, 63, 65
hydrostatic equation, 121, 122 Milankovic, 46, 47
mitigation, 4, 78, 187, 188
ice core, 13 model-system, 159, 162-168
ICESAT, 22
ideal models, 105, 109, 111, 145 naive meteorology, 33, 38, 43
impetus, 92 naive physical image, 92
indirect effects, 67, 180 naive physics, 35, 36
initial conditions, 41, 100, 108, 110, NAO, 28, 172, 192
128, 130, 135-137, 139, 140, 162, NASA, 22
163, 168 Navier-Stokes equations, 81, 122, 126
initial state, 128, 129, 135, 136, negative feedback, 4, 86, 88
138-140, 147, 152, 154, 180, 181 nesting, 173
intelligibility, 33, 52, 186 neural network models, 174, 192, 193
interglacial periods, 28, 46, 181 Newton, 40, 58, 59, 96
IPCC, 5, 27, 44, 49, 187 nitrous oxide (N 2 0), 48, 49, 63, 65
irradiation, 60, 68-70, 73, 74, 77, 78, non-autonomous (system), 168, 181
83 non-hydrostatic model, 126
normal science, 191
Kuhn, 191
Kyoto Protocol, 187 observation bulletins, 15
oxygen isotopes, 24, 25, 27
Laplace, 137, 138, 147
Larsen B, 1, 22 paleoclimatology, 23
law of state of gases, 41 paradigm, 51, 68, 117, 182, 186, 189,
limited-area model, 125 191
200 From Observations to Simulations

parameterisation, 126-128, 135, 136, sea level, 1,2, 178, 180, 186
139, 142, 143, 149, 160, 163, 164, sensitivity, 161-166, 170, 171, 180
166, 192 simplified model, 170-172
Parisi, 111 simulation, 6, 107, 108, 110-112, 120,
Pasini, 140, 155 128, 129, 134, 136, 143, 154, 160,
Pelino, 140, 155 168, 170, 172, 174, 176, 177, 180,
phase space, 138 189, 190
Planck, 59, 60 simulation model, 5, 103, 108-114,
Plato, 97 117, 142, 149, 156, 174, 175, 181,
Poincare, 137 186, 192
polar satellites, 17, 18, 20, 130 simulation paradigm, 142, 190-192
Pope John Paul II, 91 SKYLAB, 21
Popper, 183 solar constant, 22, 162
positive feedback, 4, 86, 88 sounding, 16, 19
precautionary principle, 188 sounding balloon, 15, 16
predictability, 137, 141, 144, 150, 189, specific heat, 74, 75
190 spin-up, 131, 154, 189
primitive equations, 119, 120, 121, 143, state, 10
149 state space, 139, 152, 153, 154, 163,
probability density function, 139, 140, 165, 167, 175, 180
147 statistical mechanics, 10
prognostic equation, 121, 135 Statistical-Dynamical Model (SDM),
proxy, 12,23,26,45,49,52 171
stratosphere, 44, 45, 69
radiance, 19 sulphate, 48, 49, 158, 160, 175
radiative balance, 85, 157-159, 170
Radiative-Convective Model (R-CM), teleconnections, 192
171 theory, 38, 107, 108, 111
radiosonde, 15, 16, 21, 40, 68 thermohaline circulation, 78, 178
radiosounding, 19 Torricelli, 9
recomposition, 98, 99, 107, 109, 114, TOVS, 19, 129
142, 143 tree rings, 12, 25
reductionism, 56, 57, 114 tropical cyclones, 144, 177
refutation, 107 tropopause, 40, 125
Regional Climate Model (RCM), 173 troposphere, 43, 44, 69, 70
remote sensing, 16 Tuvalu, 1
Richardson, 119, 122
Rontgen, 59 universality, 41, 55
urban heat islands, 29
satellites, 13, 16-18, 119, 129
validation, 57, 106, 113, 134, 161, 163,
scatterometer, 20
165-167, 174, 182
scientific explanation, 3
 Index 201

virtual laboratory, 6, 114 weather forecast, 6, 14, 117, 118, 149,

volcano eruptions, 44, 65, 66, 167 151, 156, 190
Vostok, 46, 47 weather forecast simulation, 131
wind profilers, 16
weather, 10, 11 WMO, 5, 14
world-wide warming, 1