Incorporating spatial autocorrelation with neural networks in empirical land-use change models

Land-use data can accurately reflect spatial pattern dependence (ie, spatial autocorrelation) and a nonlinear relationship with driving variables. In this study landuse dynamics in the Paochiao Watershed, Taiwan are forcast for the next fifteen years by incorporating artificial neural networks with spatial autocorrelation (Auto-ANNs) into the conversion of land use and its effects (CLUE-s) model. In addition to spatial autocorrelations of land use, Auto-ANNs-CLUE-s considers the nonlinear relationships between driving factors and land-use patterns. Results of a three-map comparison indicate that the Auto-ANNs-CLUE-s model has a better overall performance than Autologistic-CLUE-s. The Auto-ANNs-CLUE-s is highly applicable for all resolutions from multiresolution validation. The results of landscape metrics demonstrate the prevalence of urban sprawl in the study area. The proposed model is an alternative means of improving land use and environmental planning.