What is the influence of landscape metric selection on the calibration of land-use/cover simulation models?

Various methods have been developed for the calibration of cellular automata (CA), which can produce a plausible cell-by-cell fit between modeling results and observed land-use data. However, traditional cell-based CA models still fail to characterize the aggregate landscape patterns of multiple land-use changes. To address this problem, we introduced a landscape-driven multiple CA model that can consider landscape patterns during the calibration procedure. Genetic algorithm was used to search for the optimal calibration parameters. We further investigated the performance of five important landscape metrics in calibration. Comparisons with two well-accepted cell-based CAs indicated that the modeling results of the proposed method are closer to the observed land-use data. Furthermore, we found that patch cohesion and edge density are appropriate landscape objectives for CA calibration in this study. More importantly, our method can effectively evaluate the performance of different landscape metrics, which could provide useful information for land-use planning.