Using Virtual Reality Catchment Simulations as Testbeds for Data Assimilation

The development of new or improved data assimilation techniques to predict states and fluxes in subsurface-landsurface-atmosphere systems (SLAS) requires robust test environments. A Virtual Reality (VR) catchment simulation is used as a virtual truth to develop and evaluate data assimilation algorithms for coupled SLAS models. This approach also mitigates the data scarcity issue as virtual measurements (states and fluxes across compartments) of the VR can be extracted via arbitrary forward operators on varying temporal and spatial scales. In this presentation we describe the development of a high resolution VR catchment using the Terrestrial System Modeling Platform (TerrSysMP). The model uses the Neckar catchment (SW Germany) as a reference, which features topographic, land-use and weather patterns typical for the mid-latitudes. As a first step two VRs are created, the first by coupled atmosphere-landsurface simulations and the second by landsurface-subsurface simulations. The first VR uses the atmosphere-landsurface mode of TerrSysMP, which couples the COSMO atmosphere model and the Community Land Model (CLM) to generate virtual observations from satellites, precipitation radars and meteorological stations. For the years 2007-2013 this VR provides the atmospheric forcing for the second VR which uses the surface-subsurface mode of TerrSysMP (the hydrology model ParFlow and CLM) to better resolve soil and groundwater processes of the Upper Neckar subcatchment.

Two examples illustrate how data assimilation algorithms use virtual observations, extracted at a range of temporal and spatial scales, to reconstruct the true state of the VR. We aim to identify types of observations and measurement strategies which potentially reduce model uncertainty and improve model predictions. A first example demonstrates the reproduction of precipitation fields from random samples of virtual station measurements (including gauge, link and radar) via data assimilation by comparing the results against the model output of the atmospheric model. The second example presents the estimation of land surface temperature and near-surface humidity from virtual satellite observations. Observations are obtained by calculating the brightness temperature for specific sensor wavelengths translated into virtual observations of the satellite. Results are again compared to the VR.

The next step is a VR resulting from the fully coupled TerrSysMP which will then include the feedbacks of subsurface properties with the atmosphere, such as the effects of soil moisture, near-surface temperatures, atmospheric boundary layer heights or the Bowen ratio. This will improve the physical consistency of the VR as more system processes are considered and including impacts on the data assimilation results by exploiting cross-compartmental correlations of state variables.