Potsdam Water Talks

Groundwater recharge is a key control on water availability, ecosystem functioning, and the resilience of water resources under climate variability and change. In karst regions, recharge is particularly difficult to quantify because subsurface heterogeneity and preferential flow pathways decouple infiltration signals from traditional surface observations, while data are often sparse. In this talk, I will present two complementary perspectives on recharge estimation in karst landscapes and beyond. First, I will highlight large-scale modelling studies that aim to represent karst-specific processes across broad regions, and I will discuss what we learn when we move from local case studies to continental-scale assessments. Second, I will introduce data-driven approaches that leverage soil moisture information to infer recharge dynamics, offering an observationally anchored alternative to conventional workflows that rely primarily on meteorological forcing and model structure assumptions. I will conclude with a short perspective on future challenges and opportunities.

Much of hydrologic knowledge stems from laboratory experiments and research basins. This has led to a variety of findings on which processes are dominant under specific conditions, and which hydrologic models best represent such processes. 
In this seminar, I will synthesize some of my work over the past decade aimed at generalizing such local hydrologic understanding into broadly applicable findings. The talk will cover model development, large-domain model application, hydrologic classification and synthesis, within the broad theme of hydrology at scale. I will end the talk with a brief overview of current research challenges and outline potential paths toward improving hydrologic prediction at large geographical scales.

Earth system science increasingly relies on models, observations, and data-driven methods to understand complex environmental processes and to make predictions across a wide range of spatial and temporal scales. Yet progress is often constrained by several recurring challenges, including principled model evaluation, attribution in the presence of correlated variables, statistical inference under model misspecification, and the computational difficulty of training physics-based models at scale.

In this seminar, I outline a theory-guided research program that addresses these challenges by combining ideas from mathematics, statistics, physics, and scientific computation. A common theme throughout is a shared scientific strategy: identify a fundamental inferential or computational bottleneck, develop the theory needed to address it, and translate that theory into practical tools for hydrology and Earth system science.

I begin with recent work on scoring rules and model evaluation, which places hydrologic performance metrics on firmer information-theoretic foundations and helps clarify how model skill should be measured, interpreted, and compared. I then turn to satellite-based analyses of land-atmosphere interactions, where soil moisture, precipitation, and related land-surface controls are studied through functional decomposition into hierarchical component functions of first, second, and higher orders. This framework makes it possible to quantify the direct, cooperative, and correlated effects of soil moisture and other land-surface variables on cloud vertical structure and precipitation.

Next, I discuss sandwich-based inference, a framework for parameter estimation and uncertainty quantification under structural model error. Because environmental models are inevitably imperfect representations of reality, this theory provides a more honest basis for learning from data and diagnosing model inadequacy. Finally, I present SAGE, a new framework for physics-based machine learning with process-based hydrologic models that uses analytic gradients to enable efficient continental-scale training while preserving physical interpretability.

Taken together, these examples illustrate how theory-guided inference and learning can improve both prediction and scientific understanding in Earth system science. By bridging model evaluation, attribution under correlated inputs, inference under misspecification, and physics-based machine learning, the talk highlights a broader framework for building models that are not only more efficient and accurate, but also more transparent, diagnostically informative, and scientifically meaningful.

Floodplains are among the most valuable but threatened ecosystems in the world, contributing to biodiversity and human well being by providing a wide range of ecosystem services. These services are provided by the ecosystem’s characteristic by acting as an aquatic and terrestrial ecosystem depending on the river water levels if well connected hydrologically. However, the status of floodplains is poor -world wide and German and like this restoration measures are carried out and requested by the EU restoration law to bring back a good status as well as functioning ecosystems.

This talk highlights two investigations to quantify ecosystem services in the field, namely biodiversity as a supporting ecosystem service (for all other ecosystem services) as well as water purification, a regulating service. Therefore, in the frame of the PhD thesis of A. Kra soil samples were taken from the Gülpe floodplain and analysed in the laboratory. Modelling was carried out the extrapolate these values to areal rates for the whole island. For biodiversity analysis particularly avian fauna was monitored in 2025 over a time of 6 weeks at six sites along the Havel by master students. High species diversity was found at the sites and the overall diversity was high because the six sites differ in the structures and inundation levels providing a wide range of habitats. The only not restored site did not differ in species numbers or indicator values. Denitrification rates were only high at around 12% of the study sites contributing to 91% of the overall retention rate. These were low lying long lasting inundated areas with low sand content and adapted vegetation to these conditions.

The Lower Havel is a special river because inundation was improved by restoration measures but even before floodplains were inundation over longer periods than at most rivers in Germany. However, depending on the restoration aim different management options and water levels should be favoured. Because higher levels can increase denitrification and the abundance of certain bird species, for other plant communities and fauna higher water levels mean a disturbance. For the study site Polder Bölkershof ongoing bird monitoring in 2026 will show how restoration in 2025 effects bird communities.

Effective adaptation to flood risk is increasingly critical as flood-related losses continue to rise. However, societies face inherent limits to their adaptive capacity. Social vulnerability factors—such as low income levels or advanced age—can constrain households’ ability to implement private precautionary measures, thereby increasing the physical vulnerability of their buildings. Consequently, adaptation strategies must be grounded in risk assessments that explicitly account for the dynamics, complex interactions, and feedback mechanisms within human–water systems. Integrating human behaviour into dynamic flood risk assessments poses a significant challenge, particularly given the already high levels of uncertainty involved. Recent advances in this field include the collection and analysis of comprehensive qualitative and quantitative data on changes in flood risk using the paired-event concept in case studies and the development of probabilistic flood loss models based on Bayesian networks. Probabilistic flood loss models offer a key advantage in that they can capture temporal changes in vulnerability while inherently providing quantitative estimates of predictive uncertainty. Further development of these approaches is therefore recommended, including the application of more dynamic modelling techniques in large-scale flood risk analyses. In addition, quantitative assessments of both the potential and the limits of flood adaptation, considering underlying vulnerability drivers, are needed. Such insights are essential for effectively prioritizing adaptation investments.

Understanding how the hydrological cycle responded to past climate change is essential for anticipating future hydroclimatic risks and landscape responses. This talk explores how molecular fossils preserved in geological archives can be used as “molecular raingauges” to reconstruct past changes in the water cycle, with a particular focus on the hydrogen isotopic composition of biomarkers. These compounds record aspects of the water cycle, such as precipitation, evaporation, moisture source, thereby providing insights into hydroclimate variability across a range of temporal and spatial scales.

The lecture introduces the fundamental principles behind biomarker-based paleohydrological reconstructions and highlights how compound-specific hydrogen isotopes can be applied to sediments from continental archives, such as lakes and paleosols. Selected case studies from the last 20,000 years demonstrate that changes in atmospheric circulation can trigger abrupt regional hydroclimate shifts, often with pronounced environmental and geomorphological consequences. The talk also discusses how human activities under changing climate conditions can amplify landscape-scale responses, altering carbon cycling, erosion, and sediment transport.