An Effective Monitoring of Evolving Groundwater Drought via Multivariate Data Assimilation and Machine Learning

Abstract Groundwater drought represents one of the most pervasive and difficult‐to‐monitor forms of water scarcity, threatening the reliability of freshwater supply for over 2 billion people worldwide, agricultural productivity, and ecosystem health. Despite its critical importance, monitoring groundwater drought with high spatial and temporal resolution remains challenging due to limited in situ observations, coarse‐resolution satellite data, and uncertainties in models. In this study, we introduce an observation‐informed approach for producing daily groundwater drought maps at 1/8° resolution across the contiguous United States (CONUS). Leveraging high‐performance computing, we jointly assimilate Soil Moisture Active Passive soil moisture and GRACE‐FO terrestrial water storage data into the Noah‐MP land surface model to enhance the representation of groundwater–surface water interactions while accounting for uncertainties, enabling a more accurate representation of groundwater drought dynamics. Considering the spatial and temporal complexities of drought patterns, we employ the Growing Neural Gas, a machine learning‐based pattern recognition algorithm, to identify emergent, evolving, and region‐specific behaviors of groundwater drought. The results reveal the onset of distinct and persistent dry clusters in recent years across the contiguous United States (CONUS), identifying the severe groundwater drought conditions that notably impacted large regions of both the Western and Northeastern CONUS. Our findings highlight the need to reassess groundwater resilience strategies, especially as droughts intensify and persist over large domains.