Drought prediction using ensemble models

Managing water resources requires the prediction of droughts. A robust model should be used for drought prediction as it is a complex and nonlinear problem. We used inclusive multiple models (IMM) and optimized radial basis function (RBF) neural networks to predict the standard precipitation index (SPI). The RBF model was trained using the coot optimization algorithm (COOA), salp swarm algorithm (SSA), shark algorithm (SA), and particle swarm optimization (PSO). Next, the outputs of RBF-COOA, RBF-SSA, RBF-SA, RBF-PSO, and RBF models were inserted into the RBF model. In the Wadi Ouahrane basin (WOB), these models were used to predict 1-month SPI (SPI-1), 3-month SPI (SPI-3), 6-month SPI (SPI-6), and 9-month SPI (SPI-9). The best input combinations were determined using a hybrid gamma test. Lagged SPI values were used to predict outputs. For predicting SPI-9, the Nash–Sutcliffe efficiency values ​​(NSE) of the IMM, RBF-COOA, RBF-SSA, RBF-SA, RBF-PSO, and RBF models were 0.94, 0, 93, 0.91, 0.88, 0.80, and 0.75, respectively. The inclusive multiple models outperformed the other models in predicting SPIs-3 and 6. The ensemble models showed high potential for predicting SPI.