Fast and accurate optimization of a GPU-accelerated CA urban model through cooperative coevolutionary particle swarms

The calibration of Cellular Automata (CA) models for simulating land-use dynamics requires the use of formal, well-structured and automated optimization procedures. A typical approach used in the literature to tackle the calibration problem, consists of using general optimization metaheuristics. However, the latter often require thousands of runs of the model to provide reliable results, thus involving remarkable computational costs. Moreover, all optimization metaheuristics are plagued by the so called curse of dimensionality, that is a rapid deterioration of efficiency as the dimensionality of the search space increases. Therefore, in case of models depending on a large number of parameters, the calibration problem requires the use of advanced computational techniques. In this paper, we investigate the effectiveness of com- bining two computational strategies. On the one hand, we greatly speed up CA simulations by using general-purpose computing on graphics processing units. On the other hand, we use a specifically designed cooperative coevolutionary Particle Swarm Optimization algorithm, which is known for its ability to operate effectively in search spaces with a high number of dimensions.