Exploring Bayesian model averaging with multiple ANNs for meteorological drought forecasts

Forecasting drought is essential for water resource management when policymakers encounter a water shortage and high demand. This research utilizes the Bayesian averaging model (BMA) based on multiple hybrid artificial neural network models including ANN- water strider algorithm (WSA), ANN-particle swarm optimization (ANN-PSO), ANN-salp swarm algorithm (ANN-SSA), and ANN-sine cosine algorithm (ANN-SCA) to forecast standardized precipitation index as one of the most important indices of drought. The models were used to forecast Standardized Precipitation Index (SPI) SPI (1), SPI (3), SPI (6), and SPI (12) in the Wadi Ouahrane basin of Algeria. The WSA, SSA, SCA, and PSO were applied to set model parameters of the ANN model. The inputs were lagged El Niño–Southern Oscillation (ENSO), Pacific decadal oscillation (PDO), North Atlantic oscillation index (NAO), and southern oscillation index (SOI). The gamma test was integrated with WSA to identify the best input scenario for forecasting drought. The BMA for forecasting SPI (1) improved the MAE attained by the ANN-WSA, ANN-SSA, ANN-SCA, ANN-PSO, and ANN models 26, 33, 38, 42, and 46%, respectively in the testing level. The MAE of BMA for forecasting SPI (6) was 40, 42, 46, 48, and 62% lower than those of ANN-WSA, ANN-SSA, ANN-SCA, ANN-PSO, and ANN-PSO. Also, the BMA and ANN-WSA had the best accuracy among other models for forecasting SPI (6) and SPI (12). This study indicated that the WSA, SSA, SCA, and PSO improved the accuracy of the ANN models for forecasting drought.