Doktorarbeit
Interaction between surface water and ground water under climate change: A process-based modeling approach
Betreuer: Sven Frei, Stefan Peiffer
This Research merges physically based modelling with natural tracer observations to quantify present day and future groundwater support to streams in the Große Ohe catchment, Bavarian Forest, Germany. Understanding how climate change reshapes exchanges between surface water and groundwater in forested headwater catchments is vital for regional water security and ecosystem resilience. To address this need, we conducted these coordinated studies:
Study 1 – Process-based climate-change experiment A fully integrated HydroGeoSphere model was developed to simulate three-dimensional variably saturated flow under an ensemble of EURO-CORDEX climate scenarios (RCP 2.6, 4.5 & 8.5). The model quantified how declining recharge and earlier snowmelt may convert today’s gaining reaches into losing ones, shortening the perennial network by ~25 % by 2080.
Study 2 – Radon as Natural Tracer
We investigated runoff generation and stream–aquifer exchange in Germany’s Große Ohe catchment by pairing Radon-222 (²²²Rn) surveys with a fully coupled hydrological-transport model that explicitly simulates ²²²Rn. Three summer campaigns (2018–2019) provided longitudinal ^222Rn profiles used for model calibration and validation. Simulations reproduced spatial and temporal patterns well (R² ≤ 0.86) and confirmed a strong relationship between exchange fluxes and activities (R² = 0.60). These persistent hot spots of groundwater underscore their importance for sustaining summer baseflow. Findings inform climate-robust management.
Study 3 – Physics-Informed Neural Networks (PINNs) Model
This study used Physics-Informed Neural Networks (PINNs) for estimating submarine groundwater discharge (SGD) exchange rates from heat data in tidally dominated settings. Using two published synthetic benchmarks and temperature logs from a tidal creek, the PINN inversely solves Darcy fluxes, capturing K1 (24 h) and M2 (12 h) signals without the phase lags seen in VFLUX or 1DTEMPRO. Spectral analysis verifies sub-daily accuracy, demonstrating that PINNs overcome sharp transitions and high-frequency variability that confound conventional heat-tracer models.