Improving flood risk modelling in a national water resources model for Denmark

The national water resource model of Denmark (DK model, www.vandmodel.dk) is a distributed transient groundwater-surface water model, set up in MIKE SHE. During its 20+-year-history it has been, amongst others, used to assess the available groundwater resource, as basis for regional models for mapping wellfield capture zones, to evaluate nitrate retention, and to assess the impact of climate change on the hydrological cycle and groundwater.
The work presented here focusses on improving the representation of shallow groundwater levels and water levels in rivers by use of the DK model. This aims at developing a national flood risk model for use in climate change adaptation, disaster risk reduction and water management. Two case studies, each covering about 1000 km2, were used in the model development – the Storå and the Odense Å catchment. For those two areas, the effect of a more detailed model resolution (100 m compared to the original 500 m model grid of the DK model) was evaluated. We incorporated additional groundwater level observations in the uppermost 10 m below the surface in the calibration and validation of the model. For the calibration, we used a modified objective function based on the Continuous Rank Probability Score (CRPS) in order to minimize the adverse effect of outliers in the observations and model structure errors. Furthermore, we applied a hydrodynamic solution for water level simulations in the surface water model (MIKE HYDRO River), together with surveyed cross sections and detailed data on structures.
For the models with the finer 100 m resolution, the shallow groundwater level in depths of up to 10 m below surface could be modelled with an accuracy of 1 to 2 m (mean absolute error for the best 90% of observations), which is slightly better than what can be achieved with a 500 m model resolution. Also stream discharge was improved for the 100 m compared to the 500 m resolution. Water levels in streams could be modelled with an average accuracy of up to 20 cm. The work compared the effect of different model resolutions, downscaling, simple routing and hydrodynamic solution with respect to their applicability to the national scale, and the ability to provide visualisations relevant for screening of areas at risk of flooding from groundwater or rivers. Moreover, a qualitative evaluation of the spatial accuracy of modelled flood water was performed with the aid of remote sensing observations.