PhD defence: Jørn Rasmussen

Jørn Rasmussen defends his thesis,

Data assimilation in integrated hydrological models
Multivariate assimilation and parameter estimation

Supervisors:
Professor Karsten Høgh Jensen, IGN
Adjunct Professor Henrik Madsen, IGN
Research Professor Jens Christian Refsgaard, GEUS

Assessment Comittee:
Professor Peter K. Engesgaard (chairman), IGN
Professor Arnold Heemink, TU Delft, The Netherlands
Professor Harrie-Jan Hendricks-Franssen, Research Centre Jülich, Germany

Abstract:
Integrated hydrological models are useful tools for water resource management and research, and advances in computational power and the advent of new observation types has resulted in the models generally becoming more complex and distributed. However, the models are often characterized by a high degree of parameterization which results in significant model uncertainty which cannot be reduced much due to observations often being scarce and often taking the form of point measurements. Data assimilation shows great promise for use in integrated hydrological models , as it allows for observations to be efficiently combined with models to improve model predictions, reduce uncertainty and estimate model parameters. In this thesis, a framework for assimilating multiple observation types and updating multiple components and parameters of a catchment scale integrated hydrological model is developed and tested using both synthetic data and real observations.
Groundwater head and stream discharge observations are assimilated in an integrated hydrological model, with the aim of updating the groundwater head, stream discharge and
water level, and model parameters. When synthetically generated observations are assimilated significant improvements are obtained in both stream flow and groundwater modeling. However, the successfulness of both the state updating and the parameter estimation is conditioned on a sufficiently large ensemble size, as spurious correlations often had a negative impact on the performance of the data assimilation algorithm. To reduce the impact of spurious correlation, an adaptive localization method is applied, which significantly improved the performance of the assimilation while reducing the computational requirements. Finally, as observation bias is common in groundwater head observations, two bias-aware data assimilation algorithms are tested and were shown to successfully estimate the bias of most observations. The data assimilation framework was applied to real observations and an improvement in stream discharge was obtained compared to a deterministic model without data assimilation and a model that had been calibrated using inverse modeling.

The thesis is available at the PhD administration office, O4.1.409