Simultaneous Assimilation of Multiple Data into a Conceptual Rainfall-Runoff Model using Variational Methods for Hydrological Forecasting Applications

Abstract:

Data assimilation methods applied to hydrological applications have primarily focused on assimilating streamflow and, more recently, soil moisture observations. Few cases actually assimilate both observations, and even fewer incorporate additional observations into the assimilation procedure. This is despite extensive developments in remote sensing information. Most research on data assimilation has focused on the implementation of sequential assimilation using Kalman filters. We present an alternative approach using variational methods based on Moving Horizon Estimation (MHE) to simultaneously assimilate streamflow data and remote sensing information obtained from the Satellite Application Facility on Support to Operational Hydrology and Water Management (H-SAF) community, namely snow-covered area, snow water equivalent and soil moisture. This approach enables a highly flexible formulation of distance metrics for the introduction of noise into the model and the agreement between simulated and observed variables.

The application of MHE on data assimilation is tested at two data-dense test sites in Germany and one data-sparse environment in Turkey. The assessment of results is based on the lead time performance of state variables of the conceptual rainfall-runoff model, i.e. not limited to the performance of streamflow forecast but also applicable to snow and soil moisture forecast skills. Results show a potential improvement on the performance of the forecasted streamflow when using a perfect time series of state variables generated through the simulation of the conceptual rainfall-runoff model HBV. The assimilation of H-SAF data, in combination with streamflow, reduces the performance of the forecasted streamflow compared to the assimilation using only streamflow data. However, other forecasted quantities such as the snow water equivalent or soil moisture are improved. Recommendations based on the test cases are given following the length of the assimilation window.