GRACE/ML-based analysis of the spatiotemporal variations of groundwater storage in Africa

Abstract

Groundwater is a crucial factor influencing the ecological security and social-economic development in Africa. To support comprehensive and sustainable management of groundwater resources in Africa, GRACE and GLDAS data were utilized to extract information about groundwater storage anomaly (GWSA) in Africa and its different basins. Theil-Sen Median method, Mann-Kendall (MK) trend test and the seasonal and trend decomposition LOESS method (STL) were applied to reveal the long-term and seasonal spatiotemporal trends in GWSA. Additionally, an interpretable machine learning algorithm namely Extreme Gradient Boosting and SHAP model (XGBoost-SHAP) was employed to analyze the driving processes of the factors impacting GWSA in different basins. The study results indicate that GWSA in Africa exhibited an overall upward trend, with significant seasonal characteristics. In sub-Saharan African basins, GWSA showed a significant increase trend, with annual growth rates ranging from 2.75 cm/a to 8.02 cm/a. In contrast, a declining GWSA trend was observed in the Sahara region, with an annual decrease rate of 2.62 cm/a. Quantitative analysis identified population density and normalized difference vegetation index (NDVI) as the key factors influencing GWSA. These findings allowed us to categorize the underlying mechanisms driving GWSA across African basins into three types: (1) anthropogenic activity-dominated regions; (2) natural factor-dominated regions; (3) regions controlled by the interaction of natural factors and human activities. Understanding and monitoring the spatiotemporal heterogeneity of GWSA and the differences in driving factors across different basins is critical for a substantial improvement in the management of groundwater resources in the different basins across Africa.