Geostatistics
Introduction
What is Geostatistics?
Key Points:
Geostatistics is a branch of statistics focusing on
spatial or spatiotemporal datasets.
It provides tools to model spatial correlation and
interpolate values at unsampled locations.
Used in environmental science, hydrology, agriculture,
epidemiology, geology, and climate modeling.
Statistics, Geostatistics, Big-data
analysis
Why Geostatistics Matters
Applications:
❑ Estimating mineral reserves in mining
❑ Mapping rainfall or temperature across regions
❑ Predicting soil contamination levels
❑ Modeling disease spread in public health
❑ Precision agriculture (e.g., fertilizer application)
Geostatistical analysis of decadal rainfall (mm)
pattern in Uttar Pradesh
Hg mine located in the Somiedo Natural
Reserve (Spain).
Types of Spatial Data
1. Point Data: Measurements at specific locations
(e.g., soil samples, weather stations)
2. Areal (Lattice) Data: Aggregated over regions
(e.g., population per census tract)
3. Raster/Gridded Data: Continuous fields (e.g.,
satellite imagery, elevation models)
Point Data
Example of geostatistical data. Malaria
prevalence at specific locations in Zimbabwe.
Example of areal data. Household income in
$1000 USD in neighborhoods in Columbus,
Ohio, in 1980.
Raster/Gridded Data
Digital elevation
model (DEM) of Mt.
Everest region
(Source:
Wikimedia)
Sambalpur topographic map, elevation, terrain
Exploratory Spatial Data Analysis (ESDA)
Goals:
Understand data distribution
Identify trends, outliers, and clusters
Check assumptions before modeling
Tools:
Histograms and box plots
Scatter plots
Spatial maps (thematic maps, bubble plots)
Trend analysis (e.g., polynomial surface fitting)
Spatial Autocorrelation – The Core Concept
Definition: The degree to which a set of spatial
features and their values are clustered, dispersed,
or randomly distributed.
Tobler’s First Law of Geography:
"Everything is related to everything else, but near
things are more related than distant things."
This principle is the foundation of all spatial
interpolation methods.
Measuring Spatial Autocorrelation
1. Moran’s I
Measures global spatial autocorrelation
Values from -1 (dispersed) to +1 (clustered); 0 =
random
2. Geary’s C
Similar to Moran’s I, but more sensitive to local
differences
3. Getis-Ord (G)*
Identifies hot spots and cold spots
Notes: We’ll use Moran’s I most often. A significant
positive value means similar values cluster together.
Semi-variogram
What is a Semi-variogram?A semi-variogram (or variogram) is a fundamental
statistical tool used in geostatistics to quantify the spatial variability or spatial
autocorrelation of a measured property (like mineral concentration, soil
moisture, magnetic susceptibility, etc.) over a geographic area.
It tells you how similar or dissimilar values of a
variable are depending on the distance between
them.
Semi-variogram
Semi-Partial
Variogram- Plot / graph of variance
As you increase the distance, you
decrease similarity between
values
Definition: A function that quantifies spatial dependence
by measuring how variance increases with distance.
The semivariogram is the heart of geostatistics. It tells us
how fast spatial correlation decays with distance.
Components of a Semivariogram
1. Nugget: Variance at zero distance
(measurement error + micro-scale variation)
2. Sill: Total variability; level at which
semivariance plateaus
3. Range: Distance at which spatial correlation
becomes negligible
Common Semivariogram Models
Spherical: Most commonly used; levels off at sill
Exponential: Approaches sill asymptotically
Gaussian: Smooth behavior near origin (implies high
continuity)
Linear: No sill; unbounded variance
Pure Nugget: No spatial correlation
Visual Suggestion: Plot all models on one graph for
comparison.
We fit these models to the experimental
semivariogram to use in kriging.
What is Kriging?
Kriging is a smart way of estimating values at
places where you don't have data — using values
from nearby known points plus their spatial
relationships.
It’s like "intelligent averaging" — but it doesn't
just average the closest points like IDW or
nearest-neighbor. It looks at how things usually
vary across space using something called a
semivariogram.
Real-life analogy:
Imagine you’re hiking in a hilly area, and you have
elevation readings at a few known GPS points.
Now, you want to estimate the elevation at a new
spot where you didn't measure.
• With Kriging, you don’t just average the closest
points.
• You ask: “How does elevation usually change
across distance in this area?”
• If elevation changes slowly, distant points still
help. If it changes quickly, you trust closer
points more.
Kriging uses this behavior to weigh the nearby
points properly.
