Process based modeling of regional water and energy fluxes using multi-scale observation patterns
In this DFG TR32 project we estimate the spatio-temporal distribution of soil moisture content, groundwater levels, evapotranspiration and sensible heat fluxes and also river discharges for the Rur catchment. These estimates are the result of optimally combining ParFlow-CLM model predictions and measurement data by means of sequential data assimilation. Such an approach yields an improved characterization of the land surface states and parameters, which results in an improved initialization of the lower boundary condition of atmospheric models.
Cosmic Ray Sensor in Rollesbroich
We improve sequential data assimilation algorithms and implement them together with the hydrologic model ParFlow-CLM for the Rur catchment. Three data assimilation algorithms are of interest to be implemented: 1) A version of the Ensemble Kalman filter (EnKF), which, as compared to the classical EnKF, is more robust against deviations from normality of the variables, 2) the Ensemble Kalman Smoother (EnKS) with an optimized time window for assimilating data from multiple time steps and 3) a particle filter (PF) that is, as compared to the classical PF, more CPU-efficient for large-scale applications like the one in this project section. These algorithms are tested, and detailed calculations for many scenarios are made with the algorithm that performs best. Data assimilation are used to optimally combine model predictions from ParFlow-CLM and measured data. In the data assimilation experiments, in addition to the model states (e.g., soil moisture contents), the model parameters are updated. The focus of the data assimilation experiments is on the assimilation of multi-scale soil moisture data. An algorithm will be formulated that assimilates these multi-scale soil moisture data in one step, without the need to iterate. Besides soil moisture data from TDR probes and from satellite images (e.g., ALOS), also soil moisture data from cosmic ray probes are assimilated. These experiments are carried out on the supercomputer of the Jülich Supercomputing Centre. The analyses result in an optimal characterization of land surface states variables and parameters at different spatial and temporal scales providing improved input for predictions with hydrologic models and an improved initialization of the lower boundary condition of atmospheric models.
Prof. Harrie-Jan Hendricks Franssen
Dr. Heye Bogena
Dr. Carsten Montzka