ADVANCED GIS

SV425

Spatial Data
Analysis

Spatial Auto-
correlation

T. JERA
Dept of Geomatics Engineering
University of Zimbabwe
 Spatial Auto-Correlation
• The term spatial autocorrelation refers to the presence of systematic spatial
variation in a mapped variable.
• Spatial autocorrelation measures how close objects are in comparison with
other close objects. It can be classified as positive, negative, and no spatial
auto-correlation. positive spatial
• Where adjacent observations have similar data values the map shows autocorrelation
positive spatial autocorrelation. Where adjacent observations tend to have
very contrasting values then the map shows negative spatial
autocorrelation. Where thers is no spatial auto-correlation, it means there is
presence of randomness
• The presence of spatial autocorrelation is important,
• because it is usually taken as indicating that there is something of
interest in the distribution of map values that calls for further negative spatial
investigation in order to understand the reasons behind the observed autocorrelation
spatial variation,
• - because the presence of spatial autocorrelation implies information
• Spatial autocorrelation
redundancy and hasis the tendencyimplications
important of spatial data
for to
thebemethodology
more similar of
or dissimilar than expected
by chance
spatialat certain
data distances or directions.
analysis.
• It is an important concept in geomatics, as it can reveal patterns, trends, and clusters of spatial
 Spatial Auto-Correlation
Tobler’s First Law of Geography
• “Everything is related to everything else, but near things are more related than
distant things.”
• Closer things are more correlated
• Objects further apart are less correlated, and range of interaction-no spatial correlation
beyond
• Moran’s I (Index) is used to measure spatial autocorrelation.
• Positive spatial autocorrelation is when similar values cluster together on a map.
• Negative spatial autocorrelation is when dissimilar values cluster together on a
map.
• A checkerboard pattern is an example where Moran’s I is -1 because dissimilar
values are next to each other.
 Spatial Auto-Correlation
Why Spatial Auto-correlation
• It uncovers the presence and magnitude of spatial patterns, such as clusters, hotspots,
outliers, or gradients.
For example, during a disease outbreak
• Is the disease an isolated case?
• Is it clustered or spreading with dispersion?
• it can be used to test hypotheses and build models by evaluating the spatial randomness,
homogeneity
• Spatial proximity: This is the tendency of similar things to be located close to each other. For
example, businesses of the same type may tend to be located in the same area.
• Spatial heterogeneity: This is the variation in the values of a variable across space. For
example, population density may vary across a city.
• it can increase the accuracy and efficiency of spatial interpolation, prediction, or classification
by utilizing the spatial information.
• it can improve the communication and visualization of spatial data by emphasizing the spatial
relationships and variations.
 Spatial Auto-Correlation
Spatial Randomness
• This is the Null Hypothesis in a Spatial auto-correlation
statistics
• This is the point of departure. It resembles the
absence of any pattern
• When we look for spatial auto-correlation, we are
trying to reject spatial randomness
• Spatial randomness reflects either of 2 characteristics
• observed spatial pattern is equally likely as any
spatial pattern (simultaneous or pattern view)
• value at one location does not depend on values at
other locations
 Spatial Auto-Correlation
Positive and Negative Spatial
Auto-Correlation
• This happens when one rejects spatial randomness
• Like values in neighbouring locations occur more than for
spatial randomness – positive spatial auto-correlation
• Disimilar values (high vs low) values in neighbouring
locations occur more than for spatial randomness –
negative spatial auto-correlation
 Spatial Auto-Correlation
Positive Spatial Auto- •Identifying spatial clusters: Positive spatial
Correlation autocorrelation can be used to identify spatial
clusters, which are areas where the values of a
• When there is an impression of Clustering variable are significantly higher or lower than the
average. Spatial clusters can be used to identify
• Clumps of like values areas of high poverty, high crime, or high risk of
disease.
• The values can either be hotspots or coldspots •Modeling spatial processes: Positive spatial
• Difficult to rely on human perception to find these autocorrelation can be used to model spatial
processes, such as the spread of a disease or the
clusters
diffusion of a new technology. Spatial models can
• It can be caused by spatial interdependence be used to predict how these processes will
unfold over time and to identify areas that are at
risk.
•Making spatial predictions: Positive spatial
autocorrelation can be used to make spatial
predictions, such as predicting where a new
business will be most successful or where a new
disease is likely to spread. Spatial predictions can
be used to make more informed decisions about
how to allocate resources and develop policies.
 Spatial Auto-Correlation
Negative Spatial Auto-Correlation Important Applications of Negative
Spatial Auto-Correlation
• Shows a checkerboard pattern
•Identifying outliers: Negative spatial
• Hard to distinguish from spatial randomness autocorrelation can be used to identify outliers,
• It is often associated with spatial boundaries which are data points that are significantly different
from their neighbors. Outliers can be caused by
• It can be caused by spatial heterogeneity (variation in
errors in data collection or measurement, or they
soil types) and spatial diffusion (spread of a disease can be genuine anomalies.
from one person to another through contact). •Detecting spatial change: Negative spatial
autocorrelation can be used to detect spatial
change, such as the growth of a city or the spread
of a disease. This is because spatial change often
leads to a decrease in the similarity between
neighboring locations.
•Monitoring spatial patterns: Negative spatial
autocorrelation can be used to monitor spatial
patterns over time. For example, it can be used to
track the spread of a forest fire or the decline of a
species.
•Developing spatial models: Negative spatial
autocorrelation can be used to develop spatial
 Spatial Auto-Correlation
Scale and Neighbourhood in SAC
• Choosing the best spatial scale and neighborhood for spatial autocorrelation analysis can be
challenging, as different choices can lead to different results and interpretations.
Scale Neighbourhood
• Spatial scale refers to the level of detail or resolution • Neighborhood refers to the set of spatial units
of the spatial data. that are considered to be adjacent or related to
• It can be measured by the size of the spatial units, a given unit.
such as pixels, polygons, or points, or by the extent • It can be defined by distance, contiguity
of the study area. (physical connectedness of neighboring
• Spatial scale affects the degree of spatial locations) , or other criteria.
autocorrelation, as smaller units tend to have higher • Neighborhood affects the calculation of spatial
autocorrelation than larger ones, and larger areas autocorrelation, as it determines which units are
tend to have more variation than smaller ones. compared and how they are weighted.
• Therefore, choosing an appropriate spatial scale • Therefore, choosing an appropriate
depends on the research question, the data neighborhood depends on the spatial structure,
availability, and the trade-off between accuracy and the spatial heterogeneity (variation in the values
generalization. of a variable across space), and the spatial
dependence of the data.
 Spatial Auto-Correlation
Moran’s I
• Moran's I is a statistical measure of spatial autocorrelation.
• It is used to measure the degree to which the values of a variable at a given location are similar
to the values of the variable at nearby locations.
• Moran's I can range from -1 to 1. A value of 1 indicates perfect positive spatial autocorrelation, a
value of -1 indicates perfect negative spatial autocorrelation, and a value of 0 indicates no
spatial autocorrelation.
• It is calculated as below

Moran's I = (n / (n - 1) * (S - E) / (S + E))

where:
•n is the number of observations
•S is the observed covariance of the variable
•E is the expected covariance of the variable under the null hypothesis
of no spatial autocorrelation