Copula based merging of modelled and satellite derived soil moisture fields and hydrometeorological fluxes

Due to it's important role in the global water cycle, soil moisture is of great interest for the scientific community, policy makers, and the general public. Today, state-of-the-art satellite missions like SMOS or SMAP are able to provide global high-resolution (in both time and space) maps of top-layer soil moisture while air-bourne sensors can be applied for small-scale regional studies. However, in order to fully understand how soil moisture contributes to the water cycle, we also need to know about the deeper soils and highly vegetated areas. As remote sensing approaches are able to penetrate the first few centimeters only, we still need in-situ observations and modeling approaches in order to analyze e.g. root-zone or deeper soil moisture. When comparing the different data sources, we still detect large discrepancies between the remotely sensed products, in-situ measurements, and hydrological (or land surface) models in the spatial patterns, dynamics, and magnitude of soil moisture. In this study, we thus aim at a statistical data combination of all these data sources in order to provide a consistent, high-resolution estimate of soil moisture. This is achieved by applying a Copula-based assimilation technique, which makes use of the spatial and/or temporal dependency structure (i.e. the correlation) between different data sources. Due to the heterogeneity of soil moisture, we have to take other climate variables into account, if we want to conclude from single station observations to other locations. This can achieved by using tri (or higher) variate dependency structures, which is a logical and straightforward extension of the widely used bi-variate Copula approaches. In this study, we derive the dependence structure between observed soil moisture and other climatic variables like, e.g., precipitation or incoming radiation, which can then be used for correcting model estimates based on local climatic conditions. It is assumed that such an assimilation approach provides more realistic values compared to classical bias-correction methods, where only the dependency structure between modeled and in-situ observations is considered. We therefore present the methods and preliminary results and show comparisons to classic bias correction methods.