Comparison among three automated calibration methods for cellular automata land use change model: GA, PSO and MCMC

Spatial cellular automata (CA) model is one of the most common approaches to simulate land use change. Generally, CA estimates the transition likelihood from one land use state to another according to local neighbourhood dynamics and global drivers. Logistic regression (logit) method is widely used to calibrate CA models. Recently, several optimization algorithms have been introduced to calibrate CA models. This study compares the performance of three optimization algorithms: (i) genetic algorithm (GA), (ii) particle swarm optimization (PSO), (iii) and Markov Chain Monte Carlo (MCMC). The three algorithms are incorporated into a CA model to simulate urban expansion in Wallonia (Belgium). In addition, we compare the three calibration algorithms with the logit method. The results show that all three algorithms outperformed the logit method. The results also reveal that the performance of GA is slightly better than PSO and MCMC.