A Decomposition-Based Hybrid Prophet–LSTM Framework for SPEI-12 Drought Forecasting in Kano State, Nigeria

Drought persistence in the Sudan–Sahel transition zone of Northern Nigeria poses a substantial risk to agricultural productivity. This study develops a Hybrid Prophet–Long Short-Term Memory (LSTM) architecture to address the existing research gap in near-term predictive capacity for non-stationary hydroclimatic time series. Utilizing Kano State as a case study, the Prophet algorithm was employed to extract deterministic trends from the Standardized Precipitation Evapotranspiration Index (SPEI-12) derived from CRU TS v4.09 data (1980–2024), while an integrated LSTM network modeled the stochastic residuals. Diagnostic results indicate a statistically significant trend toward moisture recovery (p < 0.0001). Comparative analysis demonstrated that the hybrid model significantly outperformed standalone baselines, achieving a Nash–Sutcliffe Efficiency (NSE) exceeding 0.87 and a 67.2% reduction in Root Mean Square Error (RMSE). Furthermore, the framework accurately simulated hydroclimatic transitions with a directional accuracy exceeding 87%, confirming high predictive reliability. Projections for the 2025–2030 period indicate a continued positive moisture shift of approximately 0.9 SPEI units. These findings underscore the technical necessity of decoupling non-linear noise from deterministic signals to resolve complex drought dynamics. Consequently, the proposed framework serves as a robust tool for near-term climate prediction. Scaling this methodology across diverse agroecological zones is recommended to enhance national drought early warning systems and regional climate resilience strategies.