Parallel genetic algorithms for optimising cellular automata models of natural complex phenomena: An application to debris flows

Cellular automata models of natural complex phenomena may depend on a set of parameters which can significantly influence the global dynamics of the simulated events. In order to reliably apply such models for predictive purposes, their parameters have to be estimated with the greatest possible accuracy. However, no standardised optimisation techniques exist in this specific research field. Genetic Algorithms (GAs) offer a possible solution: they are parallel algorithms, and can be easily implemented to exploit the simultaneous use of multiple CPUs, thereby greatly reducing the execution time.An application of a parallel GA to the optimisation of a cellular automata model for the simulation of debris flows characterised by strong inertial effects is presented. The May 1998, Curti-Sarno (Italy) debris flow has been selected as a case study for the optimisation of the model. Theoretical considerations on the dynamics of the adopted GA are discussed, with reference to two different fitness functions applied to an idealised case study.Results demonstrated the usefulness of the approach, in terms of both computing time and quality of performed simulations. Moreover, experiments on the idealised case study pointed out that the simplest fitness function (only based on the comparison of affected areas) could conveniently be adopted for calibration purposes.