Advanced Journal of Environmental Science and Technology

ISSN 2756-3251

Advanced Journal of Environmental Science and Technology | Vol. 17 (6), pp. 14-26, June, 2026

Research Article

Sub-Seasonal Drought Forecasting and Soil Moisture Memory Analysis Across Sub-Saharan Agro-Ecological Zones Using Explainable Deep Learning

Kwaku Mensah-Bonsu¹*, isabelle Laurent², Prashant Sharma³

¹Department of Meteorology and Climate Science, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana
²Laboratory of Dynamic Meteorology, Sorbonne University, Paris, France
³Centre for Climate Change Research, Indian Institute of Tropical Meteorology, Pune, India

Corresponding Author: Kwaku Mensah-Bonsu — [email protected]

Abstract

Accurate sub-seasonal (2-to-6-week) agricultural drought prediction is critical for rainfed smallholder farming systems vulnerable to mid-season dry spells, yet dynamical numerical weather models exhibit rapid skill decay beyond two weeks. This study develops a spatiotemporal Transformer-LSTM deep learning model trained on 40 years of reanalysis meteorology, satellite-derived root-zone soil moisture, and sea-surface temperature teleconnection indices across West and East African agro-ecological zones. At a 4-week lead time, the model improved Standardized Precipitation Evapotranspiration Index (SPEI) anomaly correlation skill by 29% over operational ensemble dynamical forecasts. SHapley Additive exPlanations (SHAP) attribution revealed that antecedent root-zone soil moisture memory and Atlantic Niño sea-surface temperature gradients govern forecast predictability, providing actionable early-warning lead times for drought-resilient crop management.

Keywords: sub-seasonal drought forecasting; explainable deep learning; soil moisture memory; Sub-Saharan Africa; climate teleconnections; agricultural resilience

Manuscript Timeline: Received 04 May, 2026; Revised 28 May, 2026; Accepted 04 June, 2026; Published 09 June, 2026.

Citation: Mensah-Bonsu, K., Laurent, I., and Sharma, P. (2026). Sub-Seasonal Drought Forecasting and Soil Moisture Memory Analysis Across Sub-Saharan Agro-Ecological Zones Using Explainable Deep Learning. Advanced Journal of Environmental Science and Technology, 17(6), pp. 14-26.