More accurate forecasting of drought indices using a decomposition-based hybrid machine learning model

Drought is a recurrent natural hazard that poses severe challenges to agriculture, water resources, and socio-economic stability, particularly in arid and semi-arid regions. Accurate forecasting of drought conditions is essential for effective early warning systems and timely implementation of mitigation strategies. This study presents a comprehensive modelling framework for forecasting precipitation-based drought indices namely, the Effective Drought Index (EDI) and the Standardized Precipitation Index (SPI) at 3- and 6-month scales (SPI-3 and SPI-6) for two major drought-prone districts in Maharashtra, India: Ahmednagar and Jalgaon. The study evaluates and compares the performance of multiple forecasting models, including the Autoregressive Integrated Moving Average (ARIMA), Time Delay Neural Network (TDNN), and Extreme Learning Machine (ELM), along with their hybrid variants incorporating Seasonal-Trend Decomposition using Loess (STL) and Ensemble Empirical Mode Decomposition (EEMD). The models were rigorously assessed using standard accuracy metrics, including Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and Mean Absolute Error (MAE). Among all the models, the EEMD-TDNN hybrid emerged as the most effective, exhibiting a quantifiable improvement of 15–30% reduction in RMSE and MAPE across all three drought indices when compared to conventional and other hybrid models. The Diebold-Mariano test confirmed the statistical significance of EEMD-TDNN’s improvement over other models.This superior performance is attributed to the synergy between EEMD’s ability to decompose complex, non-stationary time series into intrinsic mode functions, and TDNN’s strength in capturing temporal dependencies through time-lagged input representations. The proposed EEMD-TDNN model offers accurate drought forecasting, aiding proactive planning for policymakers and resource managers.