Parallelization of an Optimization Algorithm for Beneficial Watershed Management Practices

The optimization of beneficial management practices (or beneficial management practices, BMPs) is a typical case of complex geo-computation; a computation-intensive search for optimal solutions of watershed BMPs through many iterative watershed model simulations. This paper presents a parallelization of the epsilon non-dominated sorted genetic algorithm (ε-NSGA-Ⅱ), an increasingly widely-used algorithm for BMPs optimization. The proposed parallel optimization algorithm was designed based on a master-slave parallelization strategy and implemented using the message passing interface (MPI). A case study executed on an IBM cluster for the Meichuan Jiang watershed (about 6366 km²) in the Lake Poyang basin shows that the proposed parallel BMPs optimization algorithm performs well. When the count of cores used in the case study increased (8~512 cores), the proposed parallel optimization algorithm delivered a higher speedup ratio. The speedup ratio reached 310 when 512 cores were used. In this case study, the parallel efficiency of the proposed parallel BMPs optimization algorithm decreased with an increase of the count of cores. The parallel efficiency ranged from 0.61 to 0.91, demonstrating that the proposed algorithm achieves good parallel performance.