Multi-Scale Long-Term Forecasting of Meteorological Drought Using Wavelet-Decomposed LSTM Models across Six Climatic Regions of Tanzania

Abstract Meteorological drought is a recurrent climate hazard that threatens agriculture, water resources, and socioeconomic development in Tanzania. However, accurate drought forecasting remains challenging due to the nonlinear and non-stationary nature of hydroclimatic processes, despite advances in deep learning. This study proposes a hybrid Wavelet-Long Short-Term Memory (Wavelet-LSTM) model for multi-scale forecasting of meteorological drought across six climatically distinct regions of Tanzania. Meteorological variables from the ERA5 reanalysis dataset, including precipitation, temperature, wind components, wind speed, dew-point temperature, and mean sea-level pressure, were used to predict the Standardized Precipitation Index (SPI) at 3- and 6-month accumulation periods. Wavelet decomposition was applied to reduce noise and extract dominant temporal features before model training. The dataset was partitioned into training (70%), validation (15%), and testing (15%) subsets, and model performance was evaluated using the coefficient of determination (R²), correlation coefficient (R), mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), Nash-Sutcliffe efficiency (NSE), Willmott's index of agreement (d), and analysis of variance (ANOVA). The proposed model achieved consistently high predictive accuracy across all climatic regions, with testing R² ranging from 0.86 to 0.97, correlation coefficients between 0.94 and 0.99, NSE values of 0.86–0.97, and WI values of 0.96–0.99, while maintaining low prediction errors. ANOVA results (p > 0.05) indicated no statistically significant differences between observed and predicted SPI values, confirming the robustness and generalization capability of the model. These findings demonstrate that integrating wavelet decomposition with LSTM substantially improves long-term drought forecasting and provides a reliable framework for drought early warning, climate adaptation, and water resources management in Tanzania and other hydroclimatically diverse regions.