Comparative Evaluation of Statistical Models and Artificial Intelligence for Drought Prediction in Isfahan Synoptic Station

Meteorological drought is a complex natural disaster that occurs everywhere in the world. Predicting the occurrence and severity of drought can be effective in managing water crises and their impacts. The purpose of the current study is to select the most appropriate model from statistical models and artificial intelligence (artificial neural networks) to predict drought in Isfahan synoptic station during the span period of 1990-1920 using the Z-Score index (ZSI). In this study, the capability and efficiency of the SARIMA stochastic linear model and three advanced learning machine models of the Feedforward Neural Networks (FNNs), Multi-layer Perceptron (MLP), and Extreme Learning Machines (ELM) were evaluated based on Root Mean Squared Error (RMSE), Mean Absolute Square Error (MASE) and Mean Absolute Error (MAE). The results showed that among the many models made, the Feedforward Neural Networks (FNNs) model with RMSE of 0.33, MASE of 0.02, and MAE of 0.22 were selected as the best model. Using the superior model, precipitation for the period of 2025-2021 of Isfahan synoptic station was predicted, then based on the ZSI drought index, drought intensity of forecast precipitation data in 3, 6, 9, 12-month time scales, 18, and 24 months was calculated. The results of drought severity predicting showed that severe drought in 3 and 6 month time scales in 2021 and 2023 and in 9 and 18 month time scales in 2024, moderate drought in all time scales in 2024, and weak droughts occurred at the 3, 6, and 24-month time scales in 2024 and 2025, respectively. Overall, the results showed that the use of feed neural network model was more efficient. Since predicting drought at all time scales can reveals drought more accurately, this predicting in turn to facilitate the development of water resources management strategies for management of drought is effective.