Pareto-optimal MPSA-MGGP: a new gene-annealing model for monthly rainfall forecasting

Abstract

Rainfall is considered the hardest weather variable to forecast, and its cause-effect relationships often cannot be expressed in simple or complex mathematical forms. This study introduces a novel hybrid model to month ahead forecasting monthly rainfall amounts which is motivated to be used in semi-arid basins. The new approach, called MPSA-MGGP, is based on integrating multi-period simulated annealing (MPSA) optimizer with multigene genetic programming (MGGP) symbolic regression so that the hybrid model reflects the periodic patterns in rainfall time series into a Pareto-optimal multigene forecasting equation. The model was trained and verified using observed rainfall at two meteorology stations located in north-west of Iran. The model accuracy was also cross-validated against two benchmarks: conventional genetic programming (GP) and MGGP. The results indicated that the proposed gene-annealing model provides slight to moderate decline in absolute error as well as noteworthy augment in Nash-Sutcliffe coefficient of efficiency. Promising efficiency together with parsimonious structure endorse the proposed model to be used for monthly rainfall forecasting in practice, particularly in semi-arid regions.