Machine learning-augmented climate models for predicting groundwater stress zones in Sub-Saharan Africa

Abstract

Groundwater supplies more than 70% of water demand in SSA regions, but is increasingly under threat from climate change and population growth. Conventional groundwater models are data-intensive and not appropriate for Sub-Saharan Africa (SSA) contexts. This systematic review assesses Machine Learning-Augmented Groundwater Models (ML-AGMs) such as Random Forest, XGBoost, and LSTM algorithms to predict groundwater stress zones with minimal data inputs and satellite-based climate data. Several peer-reviewed articles (2015–2025) were analyzed to assess the efficacy, applicability, and scalability to SSA regions. Findings reveal that ML-AGMs provided excellent prediction potential (RMSE error range: 0.356–0.372 m; R-squared: 0.503–0.679) with minimal data inputs. This represents an important innovation where data-scarce groundwater modeling is possible and dynamic. Google Earth Engine and QGIS enhance spatiotemporal analysis and improve communication with stakeholders. However, this review also highlights limitations of ML-AGM concerning data quality, applicability issues, and ethical considerations. Further research recommendations include hybridized models and inclusion of socioeconomic factors. ML-AGM and applications provide scalable and enhanced Sustainable Development Goal (SDG) 6 implementation. However, issues and limitations persist with aquifer diversity and the equitable application of groundwater stress assessments.