Parallel Data Assimilation Framework coupled to the Terrestrial Systems Model Platform

Illustration of the impact of data assimilation on root zone soil moisture characterization. The vertical axis displays the ratio of RMSE for data assimilation and RMSE for open loop so that values smaller than one imply improvement by data assimilation. The upper graph shows results for different climate types, the figure in the middle results for different plant functional types and the lower figures results for different data types.

TerrSysMP-PDAF is the Parallel Data Assimilation Framework (PDAF) coupled to the Terrestrial Systems Model Platform (TerrSysMP). It shows a very favourable scalability on supercomputers, and has been successfully tested for the assimilation of different types of land surface and subsurface data for improving model predictions.

The Parallel Data Assimilation Framework (PDAF) developed at the Alfred-Wegener Institute (AWI) has been coupled to the Terrestrial Systems Model Platform (TerrSysMP). The Terrestrial Systems Model Platform comprises the atmospheric model COSMO, the land surface model CLM and the subsurface model ParFlow. These three models are two way coupled. TerrSysMP-PDAF has a very good scalability on supercomputers and is therefore suited for data assimilation in combination with terrestrial system models at high spatial and temporal resolution. Currently, TerrSysMP-PDAF is able to assimilate groundwater level, river stage and multi-scale soil moisture (in situ measured by cosmic ray neutron sensors (CRNS) and remotely obtained by satellite) data and can also estimate soil and aquifer hydraulic parameters. TerrSysMP-PDAF has been tested in a number of synthetic and real-world studies.

The assimilation of groundwater level data was tested to update not only groundwater levels but also root zone soil moisture content. Different assimilation methodologies have been compared and it was found that fully coupled data assimilation, where model states in the aquifer domain and in the deep vadose zone are updated in terms of pressure, and model states in the rest of the vadose zone are updated in terms of soil moisture performs better than other data assimilation approaches. The methodology was tested for different climate, soil and plant functional type combinations. The conclusion was that groundwater level data can contribute to improve root zone soil moisture characterization if the groundwater level is between 1 and 8m below the land surface. Slightly better results were obtained for broadleaf trees than for other plant functional types, but the influence of soil and vegetation type on the simulation performance was less important than climate type.

Another study evaluated the assimilation of soil moisture data derived from CRNS for the Rur catchment located in Germany near the borders of Belgium and Netherlands. This study concluded that the assimilation of data from just nine CRNS-probes can already improve the catchment wide soil moisture characterization and reduce the root mean square error (RMSE) from about 0.09 cm3/cm3 (no assimilation) to 0.05 cm3/cm3. The evaluation was made using independent data which were not used in the assimilation. The relatively strong improvement of soil moisture characterization in spite of the limited number of sensors installed can be explained by the spatial correlation in soil properties and meteorological forcings. In another study at the continental (EURO-CORDEX) scale, soil moisture data from satellite (ESA CCI product) were also assimilated to improve the soil moisture characterization and river discharge modelling over Europe. This study showed that also at this large spatial scale an improved assessment of these variables can be obtained.

The assimilation of river stages including the update of spatially distributed fields of Manning´s coefficients was tested in a synthetic study. The study shows that estimation of Manning´s coefficients is feasible and that the information from the rainfall radar is essential to improve estimates. If in addition also soil hydraulic properties are uncertain, the assimilation of river stage data becomes more challenging and it is not always possible to improve both the characterization of spatially distributed fields of Manning´s coefficients and hydraulic conductivities.

TerrSysMP-PDAF is currently extended for the assimilation of more data types including atmospheric data, for applications at the continental scale and enhanced capabilities for coupled data assimilation.