Spatial Statistics is the branch of statistics concerned with data whose values or observations are associated with geographic locations and whose spatial arrangement may affect their statistical behaviour. Conventional statistical methods often assume that observations are independent, but geographic observations frequently exhibit spatial dependence because nearby locations can share environmental conditions, social influences, physical processes or measurement characteristics.
Spatial Statistics provides methods for describing spatial patterns, measuring spatial autocorrelation, analysing spatial point processes, modelling spatial variation and estimating values at unobserved locations. It includes both exploratory and inferential approaches. Important areas include global and local measures of spatial autocorrelation, point pattern analysis, geostatistics, spatial regression and spatial covariance modelling. The field is not simply statistics performed on maps. Its distinctive feature is the explicit treatment of spatial relationships, distance, neighbourhood structure and geographic dependence within statistical reasoning.
Spatial autocorrelation is one of the central concepts in Spatial Statistics. It refers to the relationship between similarity of values and spatial proximity. Positive spatial autocorrelation means that nearby observations tend to have similar values, while negative spatial autocorrelation indicates a tendency for nearby observations to be dissimilar. Moran's I is a widely used measure of global spatial autocorrelation, while Geary's C provides another measure based more directly on differences between neighbouring observations.
Local indicators of spatial association can identify local concentrations or spatial outliers. The interpretation of these statistics depends on the spatial weights matrix defining which observations are neighbours and how strongly they are connected. A weights matrix based on shared boundaries can produce different results from one based on a fixed distance threshold. Statistical significance also requires appropriate assumptions and testing procedures, particularly because spatial dependence can invalidate simple randomisation assumptions if handled incorrectly.
Spatial Statistics includes spatial point pattern analysis, which examines the locations of discrete events or objects. Point patterns can be described as clustered, dispersed or approximately random relative to an appropriate reference process. Examples include locations of disease cases, trees, crimes, accidents, retail establishments or animal sightings. Methods such as nearest neighbour analysis, Ripley's K function and pair correlation approaches examine how points are distributed across different spatial scales.
A major issue is distinguishing genuine clustering from clustering produced by variation in observation intensity or the underlying population at risk. Ten disease cases in an area with one thousand people may have a different interpretation from ten cases in an area with one million people. Likewise, crime locations should generally be interpreted relative to the distribution of population, activity or opportunities rather than simply counted as raw points. Point process models can incorporate environmental or demographic covariates and can estimate how event intensity changes across space.
Geostatistics focuses on spatially continuous or spatially structured variables and is strongly associated with spatial interpolation. A central concept is the variogram or semivariogram, which describes how the expected difference between observations changes with separation distance. Nearby observations may be more similar than distant observations, and the range, sill and nugget parameters can describe aspects of spatial dependence.
Kriging methods use a fitted spatial covariance or variogram model to estimate values at unsampled locations while accounting for the spatial structure of observations. Ordinary kriging, universal kriging and other variants make different assumptions about the mean structure and available covariates. Kriging can also provide measures of estimation uncertainty under the specified model. It is important to distinguish the estimated surface from the uncertainty associated with that surface. A visually smooth interpolated map can appear highly precise even where observations are sparse. Spatial Statistics therefore treats uncertainty as a central component of prediction rather than an optional addition.
Spatial regression addresses situations in which relationships between variables are affected by geographic dependence. Ordinary least squares regression assumes that observations are independent, but spatially clustered residuals can indicate that this assumption is violated. Spatial lag models incorporate dependence in the dependent variable, while spatial error models represent spatial dependence in the error structure. Other approaches include geographically weighted regression, which allows relationships to vary across space rather than assuming a single global coefficient.
Spatial regression requires careful consideration of the underlying causal question because spatial association does not establish causation. Model specification, spatial weights, omitted variables, scale and measurement error can all affect results. The presence of spatial dependence can sometimes be scientifically meaningful rather than simply a nuisance. For example, neighbouring regions may genuinely influence one another, in which case the spatial relationship is part of the process being modelled.
Spatial Statistics is also concerned with uncertainty, sampling design and the problem of incomplete geographic observation. Spatial data are rarely collected uniformly across all locations. Sampling may be concentrated near roads, populated areas, accessible sites or previously known features, producing potential sampling bias. Spatially representative designs can include systematic, stratified, random and adaptive approaches, depending on the research objective.
Spatial interpolation is sensitive to the locations and density of observations, and extrapolation beyond the observed region can be substantially less reliable than interpolation within the sampled domain. Measurement uncertainty can also vary spatially. Environmental sensors may have different error characteristics depending on terrain or atmospheric conditions, while administrative data can reflect differences in reporting practices between regions. A sound spatial statistical analysis therefore considers not only the numerical values but also how, where and why those observations were collected.
Spatial Statistics differs from Spatial Analysis in emphasis, although the two fields overlap extensively. Spatial Analysis is a broad term covering computational, geometric, descriptive and modelling operations involving geographic information, whereas Spatial Statistics specifically applies statistical reasoning to spatially structured observations. A GIS buffer operation is spatial analysis but is not necessarily spatial statistics. Conversely, estimating a spatial covariance function is a spatial statistical procedure even if the resulting model is not immediately visualised on a map.
Spatial Statistics also differs from ordinary statistics because spatial dependence and geographic structure are explicitly incorporated into the analysis. It differs from Cartography because a statistical map is a visual representation of results rather than the statistical method itself. These distinctions are useful because spatial data can be analysed incorrectly if conventional statistical assumptions are applied without considering spatial dependence, scale, aggregation and geographic sampling.
Applications of Spatial Statistics include environmental monitoring, epidemiology, agriculture, ecology, geology, climatology, public health, criminology and regional economics. Environmental scientists can use geostatistical models to estimate soil properties between sampled locations, while epidemiologists can analyse disease clustering and spatial variation in risk. Ecologists can study distributions of species and spatial interactions, and geologists can model mineral concentrations or subsurface properties. Public health researchers may examine geographic inequalities while accounting for spatial dependence and population distributions.
The methods are powerful because they allow spatial structure to be quantified rather than merely observed visually. They also have limitations. Results can depend strongly on spatial scale, boundary definitions, weights matrices, sampling design and model assumptions. Multiple testing can produce apparent local clusters by chance, while poor spatial data quality can undermine sophisticated statistical models. Spatial Statistics therefore requires both statistical competence and understanding of the geographic process that generated the observations.
