Bidirectional Coupling between Groundwater and Surface Water — Deep Learning-Based Modeling and Simultaneous Forecasts

• Applicant: Dr. Tanja Liesch
• Subject Area: Hydrogeology, Hydrology, Limnology, Urban Water Management, Water Chemistry, Integrated Water Resources Management
• Term: since 2026
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 588280036

The interaction between rivers and adjacent groundwater systems is a fundamental component of the hydrological cycle. Depending on the hydraulic gradient, water can infiltrate from the river into the aquifer or flow in the opposite direction as baseflow, sustaining river discharge during dry periods. These bidirectional exchange processes influence flood and drought dynamics, the availability of drinking water resources, the stability of floodplain ecosystems, and the operation of water management infrastructure. Despite their importance, river stages and groundwater levels are typically modeled and forecast separately, meaning that their mutual interactions are only partially represented. The aim of this project is to develop a data-driven, deep-learning-based modeling framework that simultaneously predicts river and groundwater levels while explicitly and quantitatively representing the exchange between the two subsystems. The model will learn from long-term time series of river discharge and stage data, groundwater levels, and meteorological drivers (e.g., precipitation and temperature) to capture their dynamic interactions. A dedicated coupling term will be developed to represent the direction and magnitude of water exchange between aquifer and river in a differentiable and physically plausible manner, and to integrate this exchange consistently into both predictions. A key scientific objective of the project is to investigate under which hydrogeological and hydrological boundary conditions the explicit modeling of these exchange fluxes improves the predictability of both systems. To this end, representative pilot sites across Germany will be selected, covering different river types, geological settings, flow regimes, and degrees of anthropogenic influence. The performance of the developed deep learning model will be systematically compared with established physically based coupled models in order to assess predictive skill, modeling effort, computational efficiency, and transferability. By combining methodological innovation in machine learning with a process-oriented analysis of groundwater–surface water interactions, the project aims to advance both predictive capability and scientific understanding. The results will contribute to improved forecasting of flood and low-flow events, more robust water resources management, and a better assessment of climate change impacts on coupled river–groundwater systems.

• DFG Programme: Research Grants