Pattern-based calibration of cellular automata by genetic algorithm and Shannon relative entropy

While cellular automata (CA) are considered an effective algorithm to model urban growth, their precise calibration can be challenging. The Shannon relative index (SRI) is an indicator of urban sprawl accounting for dispersion or concentration of built-up/non-built-up areas. This study uses SRIs directly in the calibration of CA as patterns, applying a genetic algorithm (GA). Moreover, the kappa coefficient is used in the calibration process. CA was calibrated using data for 2001 and 2006 and validated using 2011 data to model urban growth in Shelby County, TN. Results indicate that the kappa coefficient achieves the highest value using the proposed method (89.48%) compared with a GA without patterns (86.15%, which underestimates 32.22 km(2)) or logistic regression (85.83%, which underestimates 36.76 km(2)). A more precise calibration of urban growth using the proposed method helps city planners to provide more realistic models for the future of the region.