Integrating environmental and LULC drivers of groundwater droughts in groundwater-dependent ecosystems: a machine learning (XGBoost)-SEM analysis with ecosystem implications

Abstract Background Investigating the influencing factors of groundwater drought offers critical insights for the sustainable management of groundwater-dependent ecosystems (GDEs). The Upper Zambezi Catchment hosts a large-scale alluvial aquifer system, which is vulnerable to the effects of climate change to sustain GDEs. The study aims to: (a) characterize the spatial-temporal distribution of groundwater drought in the catchment, (b) identify hydrological and terrestrial drivers affecting groundwater drought, (c) rank the drivers according to their impact on the groundwater distribution/system, and (d) explore groundwater management actions under drought conditions i.e. disaster risk management. Methods Influencing factors, which include meterological drought indicators (such as Standardized Precipitation Evapotranspiration Index, SPEI), teleconnection factors (ENSO, PDO and AMO), and anthropogenic factors (land use and land cover (LULC)), were investigated and quantitatively compared based on Spearman correlation analysis and a decision tree machine learning model (extreme gradient boosting, XGBoost). Structural Equation Modelling (SEM) was then used to explain latent (important) factors in the nexus of climate variability—LULC dynamics to groundwater response. Results The study reveals that LULC types, particularly water bodies, cropland and bare land, exert the greatest influence on groundwater drought responses under teleconnection patterns attributed to ENSO, rather than through changes in the net water balance. This highlights the critical role of surface cover dynamics in shaping subsurface hydrological responses, with significant implications for the sustainability of groundwater-dependent ecosystems. Conclusions This study is novel in its application of XGBoost and SEM to unravel the complex nexus between climate variability, LULC, and groundwater dynamics within an ecosystem context, under data-scarcity conditions. This understanding is not only critical for sustaining groundwater availability but also for preserving the integrity and functioning of groundwater-dependent ecosystems.