mengistubelete/Deep-Challenge-Fund-Ethiopia-

Enhanced drought predication using deep Learning # Deep-Challenge-Fund-Ethiopia- Enhanced drought predication using deep Learning Abstract Drought remains a multifaceted and devastating environmental hazard in the arid and semi-arid lowlands of Ethiopia, including in Afar and Somali regions, which experience extreme climatic vulnerability. This study investigates drought dynamics across six representative districts (woredas) by integrating remote sensing and climate-based metrics, including the Normalised Difference Vegetation Index (NDVI), Temperature Condition Index (TCI), Vegetation Condition Index (VCI), Vegetation Health Index (VHI), Standardised Precipitation Evapotranspiration Index (SPEI), and Composite Drought Index (CDI). Analysis of these indices reveals distinct spatio-temporal patterns characterised by persistent vegetation stress and a lack of ecological recovery. The observed variability, ranging from moderate resilience to acute drought, underscores the limitations of traditional statistical methods and classical machine learning in modelling complex, non-linear thermal-climatic interactions. To address these gaps, we conducted a comparative evaluation of three gradient-boosting frameworks (XGBoost, CatBoost and Light gradient-boosting machine (LightGBM)) against three deep learning architectures (Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM), Bidirectional Long Short-Term Memory (Bi-LSTM), and Transformer). The CNN-LSTM hybrid model demonstrated superior predictive performance, achieving a mean absolute error (MAE) of 0.28, a root mean square error (RMSE) of 0.38, and a coefficient of determination (R2) of 0.82. These results indicate that the integration of convolutional layers for spatial feature extraction with LSTM units for temporal dependencies significantly enhances the modelling of moisture-stressed lowland environments. This research provides a methodological framework for developing advanced, deep learning-based early warning systems to mitigate the socio-economic impact …