Improving Neural Network Models for Forecasting Seasonal Precipitation in Southwestern Iran: the Evaluation of OCEANIC-ATMOSPHERIC Indices

Most parts of southern Iran have frequently experienced extreme climate conditions including drought and floods. Seasonal prediction of dry and wet episodes is essential for competent management of limitted water resources during these extreme events. The capability of artificial neural network (ANN) models for forecasting seasonal precipitation was examined for two key stations (Shiraz and Bushehr) in southwestern Iran. Besides precipitation time series, historical records of three climate indicators including the Persian Gulf Sea Surface Temperature (PGSST), North Atlantic Oscillation (NAO), and Southern Oscillation Index (SOI) were used as the predictors. The Auto-Regression with eXtra inputs (ARX) model was firstly used as a linear approach to predict seasonal precipitation one season ahead. The neural network-based nonlinear ARX (NNARX) model was trained and optimized as the next step. Results confirmed the ability of the employed ARX family models in general and the optimized NNARX in particular for successful prediction of seasonal precipitation.