Spatial simulation of landscape changes in Georgia: A comparison of 3 transition models

Abstract

Spatial simulation models were developed to predict temporal changes in land use patterns in a piedmont county in Georgia (USA). Five land use categories were included: urban, cropland, abandoned cropland, pasture, and forest. Land use data were obtained from historical aerial photography and digitized into a matrix based on a 1 ha grid cell format. Three different types of spatial simulation were compared: (1) random simulations based solely on transition probabilities; (2) spatial simulations in which the four nearest neighbors (adjacent cells only) influence transitions; and (3) spatial simulations in which the eight nearest neighbors (adjacent and diagonal cells) influence transitions. Models and data were compared using the mean number and size of patches, fractal dimension of patches, and amount of edge between land uses. The random model simulated a highly fragmented landscape having numerous, small patches with relatively complex shapes. The two versions of the spatial model simulated cropland well, but simulated patches of forest and abandoned cropland were fewer, larger, and more simple than those in the real landscape. Several possible modifications of model structure are proposed. The modeling approach presented here is a potentially general one for simulating human-influenced landscapes.