Improving SLEUTH Calibration with a Genetic Algorithm

A review of calibration methods used for cellular automaton models of land use and land cover change was performed. Calibration advances have been achieved through machine learning algorithms to either extract land change rules, or optimize model performance. Many models have now automated the calibration process, reducing the need for subjective choices. Here, the brute force calibration procedure for the SLEUTH CA-based land use change model was replaced with a genetic algorithm (GA). The GA calibration process populates a "chromosome" with five parameter combinations (genes). These combinations are then used for model calibration runs, and the most successful selected for mutation, while the least successful are replaced with randomly selected values. Default values for the constants and rates of the genetic algorithm were selected from SLEUTH applications. Model calibrations were completed using both brute force calibration and the GA. The GA model performed as well as the brute force method, but used vastly less computation time with speed up of about 3 to 22. The optimal values for GA calibration are set as the defaults for SLEUTH-GA, a new version of the model. This paper is a contraction of Clarke (in press), which reports on the full set of results.