Calibration and validation of Rainfall- Runoff Model
Model
A model is a simplified or idealized representation of reality or a system. A model relates
something unknown (the output) to something known (the input). In the case of rainfall runoff
models, the known input is precipitation, and other meteorologic data. The unknown output is
usually runoff.
Need of model
The main reason of requirement of model in water resources Engineering is due to the limitations
of hydrological measurement techniques. We are not able to measure everything we would like to
know about hydrological systems. We therefore need a means of extrapolating from those
available measurements in both space and time, particularly to ungauged catchments (where
measurements are not available) and into the future (where measurements are not possible) to
assess the likely impact of future hydrological change. Models of different types provide a means
of quantitative extrapolation or prediction that will hopefully be helpful in decision making.
The main reasons to apply modelling tools are to
(i) Simulate processes to learn about system behavior,
(ii) support decision making and to
(iii) project future behavior.
Rainfall runoff model
The hydrological cycle has many interconnected components, and response of each component is
also varied in space and time. The amount of surface runoff is influenced by soil properties, land
cover, hillslope, vegetation, and storm properties such as rainfall duration, amount, and intensity.
Runoff is generated by a combination of two mechanisms, saturation excess and infiltration excess.
Saturation excess occurs when the soil becomes fully saturated with water, exceeding the water
holding capacity of the soil; when the surplus rainfall can no longer be held in the soil, the water
is directed to another location through runoff. Infiltration excess occurs when rainfall intensity
exceeds the maximum rate that water can infiltrate into the soil, and water must flow over land as
a runoff.
Rainfall runoff model can be defined as a set of equations that aid in the estimation of the amount
of rainfall that turns into runoff as a function of various parameters used to describe the watershed.
In rainfall-runoff modelling the transformation of rainfall into runoff can be simulated with
different mathematical tools describing runoff generation processes. Hydrological models
represent the physical/ chemical/biological characteristics of the catchment and simulate the
natural hydrological processes. Hydrological models are essentially mathematical models where
the physical processes of hydrologic cycle are described by a set of mathematical equations, logical
statements, boundary conditions and initial conditions, expressing relationships between inputs,
variables and parameters.
Modelling surface runoff is difficult, as the calculations are complex and involves many
interconnected variables. Surface runoff modelling is used to understand catchment yields and
responses, estimate water availability, changes over time, and forecasting. Following diagram
shows the important steps in preparation of Rainfall runoff model.
Important steps in Rainfall Runoff Modelling Process
Characterize Water shed
Basin Model
Develop Precipitation
Meterological model
Account for Losses
Loss model
Transform to Runoff hydrograph
Transformation and base flow model
Routing of Runoff
Channel Routing model
Outflow Runoff Hydrograph at outlet
Rainfall Runoff Model Classifications
There are different ways to classify the models. The main classification is Based on Structure of
the model, physical models and Mathematical models.
Physical models
Physical models are reduced-dimension representations of real-world systems. A physical model
of a watershed, can be constructed in the lab which is a large surface with overhead sprinkling
devices that simulate the precipitation input. The surface can be altered to simulate various land
uses, soil types, surface slopes, and so on; and the rainfall rate can be controlled. The runoff can
be measured, as the system is closed.
It is difficult and uneconomical to build exact scale models of the hydro-climatological systems
on which one could perform experiments to understand the nature of its operations on rainfall and
its eventual transformation into runoff. Alternative to this is the mathematical model.
Mathematical model
It is a quantitative expression of a process or phenomenon one is observing, analysing, or
predicting. Or it is a simplified systems that are used to represent real-life systems and may be
substitutes of the real systems for certain purposes.
This model is a symbolic, usually mathematical representation of an idealized situation that has
the important structural properties of the real system.
Types of Mathematical models.
Mathematical models, can be classified using a number of different criteria. These focus on the
mechanics of the model: how it deals with time, how it addresses randomness etc. which is a
generalization because many models overlap and contain elements from each of the categories.
Event based or Continuous
An event model simulates a single storm. The duration of the storm may range from a few hours
to a few days. The key identifying feature is that the model is only capable of representing
watershed response during and immediately after a storm. Event infiltration models do not include
redistribution of the wetting front between storms and do not account for drying of the soil through
evaporation and transpiration.
A continuous model simulates a longer period, ranging from several days to many years. In order
to do so, it must be capable of predicting watershed response both during and between precipitation
events. For infiltration models, this requires consideration of the drying processes that occur in the
soil between precipitation events. Surface runoff models must be able to account for dry surface
conditions with no runoff, wet surface conditions that produce runoff during and after a storm, and
the transition between the two states. Baseflow methods become increasing important in
continuous simulation because the vast majority of the hydrograph is defined by interstorm flow
characteristics.
Empirical or Conceptual
This type of classification focuses on the knowledge base upon which the mathematical models
are built.
Empirical or black box models contain no physically based transfer function to relate input to
output. In other words, no consideration of the physical processes is involved in such models.
These models are basically input-output based models. Within the range of calibration, such
models may be highly successful.
These types of models are sometimes called "black box" models because they convert input to
output without any details of the actual physical process involved. A common way to develop
empirical models is to collect field data with observations of input and resulting output. The data
is analyzed statistically and a mathematical relationship is sought between input and output. Once
the relationship is established, output can be predicted for an observed input. For example,
observations of inflow to a river reach and resulting flow at a downstream location could be used
to develop a relationship for travel time and attenuation of a flood peak through the reach. These
empirical models can be very effective so long as they are applied under the same conditions for
which they were originally developed.
A conceptual model is built upon a base of knowledge of the pertinent physical, chemical, and
biological processes that act on the input to produce the output. Many conceptual models are said
to be based on "first principles." This usually means that a control volume is established and
equations for the conservation of mass and either momentum or energy are written for the control
volume. Conservation is a basic principle of physics that cannot be broken. Through the writing
of the equations, a model of the process will emerge. Some conceptual models are developed
through a mechanistic view instead of first principles. A mechanistic view attempts to represent
the dynamics of a process explicitly.
Lumped or Distributed
Lumped (conceptual) models occupy an intermediate position between the fully distributed
physically based approach and empirical black box analysis. Lumped models treat the catchment
as a single unit, with state variables that represent averages over the catchment area, such as
average storage in the saturated zone. Such models are formulated on the basis of a relatively small
number of components, each of which is a simplified representation of the process element in the
system being modelled. Parameters of such type of models are calibrated using trial and error
method or automatic optimization technique or combination of both.
Fully distributed (physically based) models are based on our understanding of the physics of the
hydrological processes which control catchment response and use physically based equation to
describe these processes. From their physical basis such models can simulate the complete runoff
regime, providing multiple outputs (e.g. river discharge, phreatic surface level and evaporation
loss) Unlike lumped conceptual models, physically based distributed models do not consider the
transfer of water in a catchment to take place between a few defined storages. Instead, the transfers
of mass, momentum and energy are calculated directly from the governing partial differential
equations.
A distributed model is one in which the spatial (geographic) variations of characteristics and
processes are considered explicitly, while in a spatially averaged model, these spatial variations
are averaged or ignored. While not always true, it is often the case that distributed models represent
the watershed as a set of grid cells. Calculations are carried out separately for each grid cell.
Depending on the complexity of the model, a grid cell may interact with its neighbor cells by
exchanging water either above or below the ground surface.
Deterministic or Stochastic
In deterministic models’ variables are completely defined and outcomes are certain. A
deterministic hydrological model is one in which the processes are modelled based on definite
physical laws and no uncertainties in prediction are admitted. Deterministic models permit only
one outcome from a simulation with one set of inputs and parameter values. It has no component
with stochastic behavior, i.e. the variables are free from random variation and have no distribution
in probability.
A deterministic model assumes that the input is exactly known. Further, it assumes that the process
described by the model is free from random variation. In reality there is always some variation.
Deterministic models essentially ignore variation in input by assuming fixed input. The input may
be changed for different scenarios or historical periods, but the input still takes on a single value.
Such an assumption may seem too significant for the resulting model to produce meaningful
results. However, deterministic models nevertheless are valuable tools because of the difficulty of
characterizing watersheds and the hydrologic environment in the first place.
A stochastic model is a method for predicting statistical properties of possible outcomes by
accounting for random variance in one or more parameters over time. For a selected duration, the
random variance is typically based on variations found in historical data using standardized
techniques. Stochastic models allow for some randomness or uncertainty in the possible outcomes
due to uncertainty in input variables, boundary conditions or model parameters. Instead of using a
single input like deterministic models, stochastic models include the statistics of variation both of
the input and process.
Measured-Parameter or Fitted-Parameter
This distinction between measured and fitted parameters is critical in selecting models for
application when observations of input and output are unavailable. A measured-parameter model
is one in which model parameters can be determined from system properties, either by direct
measurement or by indirect methods that are based upon the measurements.
A fitted-parameter model, on the other hand, includes parameters that cannot be measured. Instead,
the parameters must be found by fitting the model with observed values of the input and the output.
The Muskingum routing model is an example of a fitted parameter model. The K parameter can
be directly estimated as the travel time of the reach. However, the X parameter is a qualitative
estimate of the amount of attenuation in the reach. Low values of X indicate significant attenuation
while high values indicate pure translation. The only way to estimate the value of X for a particular
reach is to examine the upstream hydrograph and the resulting outflow hydrograph.
Calibration of Rainfall Runoff Model.
Model calibration is the systematically adjustment of a model’s parameter values to get a set of
parameters which provides the best estimate of the observed data. In model calibration process we
find out the various optimum model parameters, by comparing model output for a given set of
assumed conditions with observed data for the same conditions.
During setting up of rainfall runoff models the initial parameters required for simulation of Loss,
transformation and routing were need to be calculated from the reference criteria or from the
modelers experience. Then these initial parameters are entered in the model and model is run and
simulation results are compared with observed results. If the simulated results are not up to
acceptable level, then again, the parameters are changed and model is run for this changed
parameters. This process is repeated up to getting the results up to acceptable level.
The procedure followed during the calibration of Hydrological model is given in the schematic in
figure No-
Fig No- Schematic of procedure to be followed in calibration.
Unfortunately, it is not, in general, possible to estimate the parameters of models by either
measurement or prior estimation. Studies that have attempted to do so have generally found that,
even using intensive series of measurements of parameter values, the results have not been entirely
satisfactory. Prior estimation of feasible ranges of parameters also often results in ranges of
predictions that are wide and may still not encompass the measured responses all of the time.
Methods of calibration
Calibration or parameter values are adjusted between each run of the model, either manually by
the modeler or by some computerized optimization algorithm until some 'best fit' parameter set
has been found.
Sensitivity analysis
It is to be noted that the model structure and the input data collected are not error-free. Thus, there
are lot of uncertainties are involved in the model. Hence sensitivity analysis is to be done, which
will provide the modeler sensitivity of each parameter, which will be useful during the calibration
process.
The parameter values determined by calibration are effectively valid only inside the model
structure used in the calibration. It may not be appropriate to use those values in different models
or in different catchments.
Following general guidelines were followed, while calibrating the Rainfall Runoff model;
1. Calibration is to be done from upstream to downstream side of the catchment.
2. One subbasin is Selected and one parameter is adjusted at a time.
3. Sensitivity analysis is done and uncertain parameters were adjusted. Sensitivity
analysis is the process of determining the rate of change in model output with
respect to changes in model inputs (parameters). It is a necessary process to identify
key parameters and parameter precision required for calibration
4. Looking in to the physical evidence, parameter with higher individual impacts is
adjusted.
Validation Of Rainfall Runoff Model.
The next stage after calibration is the validation or evaluation of those predictions. Model
validation is a process of using the calibrated model parameters, to simulate runoff over an
independent period outside the calibration period to determine the suitability of the calibrated
model for predicting runoff over any period outside the calibrated period. In the validation process
the final calibrated model is run for the validation period without changing the model parameters
and the goodness of fit statistics are computed for the validation period.
The validation is generally done using a “split sample process” whereby a the period of available
hydro-meteorological data to be split up in two parts for model calibration and model validation.
Validation may also be carried out within a quantitative framework, calculating one or more
indices of the performance of the model relative to the observations available (if any) about the
runoff response.
Model Evaluation guidelines for quantification of accuracy of simulated model.
The model performance is typically evaluated from the comparison of simulated and the observed
discharge data in terms of mean, standard deviation, maximum daily discharge, and the total
discharge using commonly used indices. At present no comprehensive guidelines are available to
facilitate model evaluation in terms of the accuracy of simulated data compared to measured flow
and constituent values. However, many researchers have recommended various model evaluation
indices, following four quantitative statistics indices are generally used for model evaluation.
1. Pearson’s coefficient of determination (R2):
Coefficient of determination (R2) describe the proportion of the variance in
measured data explained by the model. R2 ranges from 0 to 1, with higher values
indicating less error variance, and typically values greater than 0.5 are considered
acceptable.
2. Nash-Sutcliffe efficiency (NSE)
The Nash-Sutcliffe efficiency (NSE) is a normalized statistic that determines the
relative magnitude of the residual variance (“noise”) compared to the measured data
variance “information”. NSE indicates how well the plot of observed versus
simulated data fits the 1:1 line. NSE ranges between −∞ and 1.0 (1 inclusive), with
NSE = 1 being the perfect simulation. Values between 0.0 and 1.0 are generally
viewed as acceptable levels of performance, whereas values <0.0 indicates that the
mean observed value is a better predictor than the simulated value, which indicates
unacceptable performance.
3. Percent bias (PBIAS)
Percent bias (PBIAS) measures the average tendency of the simulated data to be
larger or smaller than their observed counterparts. The optimal value of PBIAS is
0.0, with low-magnitude values indicating accurate model simulation. Positive
values indicate model underestimation bias, and negative values indicate model
overestimation bias.
4. RMSE to standard deviation ratio of observations (RSR)
RMSE is commonly used model evaluation error index. Although it is commonly
accepted that the lower the RMSE the better the model performance. A model
evaluation statistic ratio of the RMSE to observations standard deviation ratio is
named as RSR. RSR varies from the optimal value of 0, to a large positive value.
The lower RSR, the lower the RMSE and the better the model simulation
performance.
In general performance evaluation criteria for basin scale models are as below: -
Method Very good Good Satisfactory Not satisfactory
R2 R2 > 0.85 0.75 < R2 ≤ 0.85 0.6 < R2 ≤ 0.75 R2 ≤ 0.6
NSE NSE > 0.8 0.7< NSE ≤ 0.8 0.5< NSE ≤ 0.7 NSE > 0.5
PBIAS PBIAS < ± 5 ± 5 ≤ PBIAS < ± 10 ±10 ≤ PBIAS<± 15 PBIAS≥± 15
RSR RSR< = 0.5 0.5 < RSR ≤ 0.6 0.6 < RSR ≤ 0.7 RSR > 0.7
