A Hybrid Artificial Neural Network and Particle Swarm Optimization Algorithm for Statistical Downscaling of Precipitation in Arid Region

Prediction of future climate change is based on output of global climate models (GCMs). However, because of coarse spatial resolution of GCMs (tens to hundreds of kilometers), there is a need to convert GCM outputs into local meteorological and hydrological variables using a downscaling approach. Downscaling technique is a method of converting the coarse spatial resolution of GCM outputs at the regional or local scale. This study proposed a novel hybrid downscaling method based on artificial neural network (ANN) and particle swarm optimization (PSO) algorithm.
Downscaling technique is implemented to assess the effect of climate change on a basin. The current study aims to explore a hybrid model to downscale monthly precipitation in the Minab basin, Iran. The model was proposed to downscale large scale climatic variables, based on a feed-forward ANN optimized by PSO. This optimization algorithm was employed to decide the initial weights of the neural network. The National Center for Environmental Prediction and National Centre for Atmospheric Research reanalysis datasets were utilized to select the potential predictors. The performance of the artificial neural network-particle swarm optimization model was compared with artificial neural network model which is trained by Levenberg–Marquardt (LM) algorithm. The reliability of the models were evaluated by using root mean square error and coefficient of determination (R2).
The results showed the robustness and reliability of the ANN-PSO model for predicting the precipitation which it performed better than the ANN-LM. It was concluded that ANN-PSO is a better technique for statistically downscaling GCM outputs to monthly precipitation than ANN-LM.
Discussion and This method can be employed effectively to downscale large-scale climatic variables to monthly precipitation at station scale.