Abstract

Spatial data exhibits spatial autocorrelation (Tobler’s First Law of Geography), negating the common assumption of independent data observations. Accounting for autocorrelation, heterogeneity, and edge effects in spatial inference. Concurrently, unobserved spatial heterogeneity and edge effects further complicate model specification, transforming simple parameter estimation into a non-trivial challenge of spatial variance structure design. To construct robust spatial statistical models, researchers must rigorously address the joint determinants of spatial dependence, scale effects, and spatial process non-stationarity. A failure to understand the data will result in a weak understanding of model outputs and further propagate into limited fairness understanding and reporting of models. The latter is an important consideration in societal implementation of modules.