Spatial Heterogeneity

Spatial heterogeneity refers to the uneven distribution of a variable’s values or systematic variation of outcomes across space.

Spatial heterogeneity means that parts of the machine learning model may vary systematically with geography, that is, the parameters or error terms of the model may change according to the location. The error term change of a model, or the presence of variance in the residuals, is caused by spatial heteroskedasticity, a special type of heterogeneity. For example, the climate and weather might change dramatically across different climate zones, affecting forestation/deforestation and crop yields. If spatial heterogeneity exists as a spatial effect, then it needs to be considered by spatial machine learning models.

In the real world, spatial dependence may sometimes play a bigger role in affecting other independent variables or dependent variables (outcomes). But at other times, spatial heterogeneity may play a bigger role than spatial dependence. However, both play a role in many scenarios. This implies that for a specific use case you may choose different machine learning algorithms for different applications, depending upon the spatial effect.

Oracle Spatial AI takes into consideration the spatial relationships and spatial effects in all its spatial machine learning algorithms. For that purpose, spatial weight is a required parameter for all algorithms. Spatial weight is how spatial relationships are quantified. Spatial effects can be tested using statistics, such as Autocorrelation/Moran’s I, Lagrange Multipliers, and Koenker-Basset Test.

See Spatial Weights, Spatial Autocorrelation, and Metrics for Spatial Regression for more information.