A comparative study on stochastic and AI-based approaches for predicting meteorological droughts

Abstract Meteorological drought, a recurrent manifestation of climate variability, poses significant challenges to sustainable water resource management and agricultural planning in India. This study investigates the comparative performance of stochastic and artificial intelligence (AI)-based models for forecasting meteorological drought using monthly precipitation data (1981–2021) from Sagar and Chhatarpur districts of Madhya Pradesh, India. The primary objective is to identify the most reliable model capable of capturing complex temporal dependencies in the Standardized Precipitation Index (SPI) series. A suite of models, including Auto-Regressive Integrated Moving Average (ARIMA), Artificial Neural Network (ANN), Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU), and Long Short-Term Memory (LSTM) networks, were employed. The LSTM demonstrated superior predictive accuracy, achieving reductions in root mean square error (RMSE) of 71.4%, 33.0%, 21.3%, and 11.2% over ARIMA, ANN, RNN, and GRU, respectively, in Chhatarpur, with comparable improvements in Sagar district. The Diebold–Mariano test further validated the statistical significance of the LSTM’s performance advantage. The findings underscore the capability of deep learning approaches in capturing nonlinear and long-term dependencies in meteorological drought forecasting. Enhanced forecast accuracy achieved by the LSTM model holds practical value for policymakers, water managers, and agricultural planners by supporting the development of early warning systems, optimizing irrigation scheduling, and improving adaptive drought mitigation strategies. The study strengthens the evidence base for integrating advanced artificial intelligence frameworks into climate risk management and drought resilience planning.