Empirical Distributions of Landscape Pattern Indices as Functions of Classified Image Composition and Spatial Structure

Satellite imagery at multiple temporal, spatial, and radiometric resolutions and spatial extents provides a unique opportunity for examining landscape-level, spatial patterns consisting of a finite set of categories mapped onto regular lattices. Landscape pattern indices (LPIs) have become increasingly popular for quantifying and characterizing various aspects of these spatial patterns. This paper examines the influence of image composition (the proportion of categories) and structure (the spatial arrangement of categories) on LPI values. Unlike the case of Moran-type statistics, the distributions of LPIs have not been studied in detail; they are not known, thus making comparisons of LPIs among various landscapes and/or studies uncertain. We designed simulations using conditional autoregressive Gauss-Markov random fields to establish empirical LPI distributions where we systematically varied the proportion of categories and the spatial autocorrelation parameter. Here we report the results for stationary binary landscapes: global distributions and cross-correlations of four LPIs are presented in detail (number of patches, edge density, area-weighted mean shape index, and contagion). We also show how to extend these results to the multinomial and non-stationary case. Our results indicate that the composition and structure of the underlying landscape significantly affect observed LPI values. While the LPI distributions are primarily controlled by composition, they vary non-linearly according to landscape structure too.