A Multi-feature Decomposition Transform-Based Gated Recurrent Unit Model for Drought Prediction in Sub-Saharan Africa

Drought in Sub-Saharan Africa (SSA) represents a critical threat to food and economic security since the majority of the SSA region relies on rain-fed agriculture. The traditional approaches used in drought forecasting have failed to capture the nonlinear, noisy nature of precipitation in tropical regions. The main objective of the proposed study was to design and develop a hybrid model called Wavelet Decomposition Transform-Gated Recurrent Unit (WDT-GRU), which could be used for the improvement of meteorological drought forecasting in the different climatic regions of Sudan, South Sudan, and Nigeria. For the proposed study, the researchers used the CHIRPS data for the calculation of the Standardized Precipitation Index (SPI) on three different scales, namely SPI-3, SPI-6, and SPI-12. For this purpose, the proposed framework uses the combination of the Wavelet Decomposition Transformation (WDT) method for data denoising, as well as Gated Recurrent Units (GRU) and Long Short-Term Memory (LSTM). It was observed that the proposed model achieved excellent accuracy in predicting meteorological drought in the three regions, with the R2 precision rate being greater than 0.97. It was observed in the proposed study that regions like Sinkat in Sudan and Ivo in Nigeria are facing chronic drought, while regions like Ezo in South Sudan, which are resistant to moisture, are highly prone to sudden 'flash' droughts. Adam and Nadam are used in the proposed study for the improvement of convergence rates. With this framework's high-resolution predictive insights at the city scale, a powerful tool for proactive climate adaptation and disaster management in Sub-Saharan Africa is provided.