You are invited to an upcoming Geoscience seminar to take place in the Geoscience Auditorium on Feb 19 @ 15:00. The seminar will be given by Julian Koch, a senior researcher at GEUS in the Department of Hydrology. Below is the title, abstract and brief bio. There will be cake and coffee

Abstract: 

Assessment of water resources is essential to strengthen societal resilience to hydrological extremes such as floods and droughts and to protect ecosystems. At the Geological Survey of Denmark and Greenland, national scale hydrological modelling is carried out to support these needs. The National Hydrological Model for Denmark (DK-model, https://dennationalehydrologiskemodel.dk/) is a key tool that provides hydrological data for historical reference periods, current conditions, short-term forecasts, and climate change impact assessments. The DK-model is an integrated, physically based model covering the entire land area of Denmark at a spatial resolution of 100 × 100 m, integrating groundwater, surface water, and land use processes.

The DK-model provides a comprehensive overview of Denmark’s water resources and supports climate adaptation planning via the HIP platform (www.hipdata.dk). However, its spatial resolution and accuracy make it only suitable as a screening tool, while local applications often require higher spatial detail and higher accuracy with respect to observations. To address this, we have developed machine learning and hybrid modelling approaches that enhance both accuracy and spatial resolution. For groundwater, we implemented a gradient boosting decision tree model to predict typical groundwater depth at 10 × 10 m resolution for summer and winter conditions. These maps are available on HIP and provide an important decision basis for climate adaptation planning. However, they represent only average conditions and lack temporal dynamics. To overcome this limitation, we developed a real-time hybrid modelling framework that combines the temporal dynamics of the DK-model with machine learning trained on model residuals. This approach reduces not only model bias, but can also be applied to increase spatial resolution. Similarly, for streamflow, we implemented a hybrid scheme using a recurrent neural network to predict residuals of DK-model streamflow simulations. Both hybrid models operate in real time and provide enhanced hydrological information for assessing and mitigating hydrological extremes.

 Bio:

Julian Koch is Senior Researcher at the Geological Survey of Denmark and Greenland (GEUS) – Department of Hydrology, where he develops modelling approaches to support water resources management and climate adaptation. His research focuses on integrating machine learning with physically based models and multi source geospatial data, including satellite remote sensing and hydrogeological data, to improve understanding and prediction of groundwater and surface water systems. He obtained his PhD in Geology from the University of Copenhagen in 2016, following MSc and BSc degrees in geoscience and geography and joined GEUS in 2016 as postdoctoral researcher.