Soil moisture as a key predictor for regional groundwater levels: a deep learning study from Brandenburg, Germany

Groundwater is the primary source of freshwater worldwide, providing essential supplies for drinking water, agriculture, and industry. Reliable forecasting of groundwater levels (GWLs) at the regional scale is crucial for climate-resilient water management, as availability is strongly shaped by climatic and hydrogeological conditions. This study focuses on Brandenburg, one of Germany’s driest regions, where groundwater is the primary source of drinking water and recurrent droughts increasingly stress resources. We present the first regional deep learning model for GWLs in Brandenburg, leveraging a one-dimensional convolutional neural network combined with a long short-term memory (1D-CNN-LSTM) ensemble trained on 217 monitoring wells using meteorological data, soil moisture (SM) from multiple depths, and landscape predictors. Benchmarking shows that the recurrent memory of the 1D-CNN-LSTM layers effectively captures delayed and cumulative groundwater responses across heterogeneous hydrogeological conditions. A key innovation is the explicit integration of SM as a proxy for vadose zone dynamics, which substantially improves model accuracy. The model achieved strong regional performance (R ² = 0. 72, NSE = 0. 59, RMSE = 0. 11), with SM reducing bias in GWL estimates, particularly during the drought years 2019–2023. SHAP analysis identified SM from shallow and deep layers (0–30 cm and 0–90 cm) as the most influential predictors, surpassing all climatic inputs while confirming hydrogeological plausibility. While individual static features had limited predictive value, their combined inclusion enhanced regional generalisation. Some residual overestimation persisted under extreme low-GWL conditions, highlighting opportunities for further improvements in low-flow representation. This framework balances model complexity with data availability using nationwide open-source datasets and provides a methodological basis for assessing groundwater dynamics under climate stress and for guiding future improvements in regional groundwater modelling.