ECMWF Final Report on SMOS brightness temperature activities over land: Monitoring and Data Assimilation

In this report we present the first steps towards testing the use of SMOS neural network soil moisture in the European Flood Awareness System (EFAS) for the SMOS-Operational Emergency Management (SMOS-E) ESA contract (tasks TH1-005, TH1-010 and TH1-015). Currently, EFAS flood forecasts are made from an initial soil moisture produced by the LISFLOOD hydrological model, which is driven by 24-hour precipitation and near-surface temperature and wind observations. In the first part of this report, we compare the EFAS initial soil moisture product with the ECMWF SMOS neural network soil moisture, in order to understand the similarities and differences and the potential of SMOS for improving the initialisation in EFAS. Both soil moisture products are also compared to the ERA5 reanalysis soil moisture. Secondly, we present results of bias correcting SMOS to EFAS through CDF-matching, a necessary step for SMOS soil moisture to be used directly in EFAS.

Results show that SMOS soil moisture has high correlations to ERA5 in most of Europe (>0.6), with values that are similar to North America and Australia where SMOS is thought to have its best performance. This
justifies testing the use of SMOS soil moisture for flood forecasting in Europe. SMOS has low correlations to both ERA5 and EFAS in parts of Northern Europe (Norway, Sweden, Finland, Iceland), however, indicating that there may be less benefits to using SMOS data for flood forecasting in these regions. Results also indicate lower anomaly correlations for SMOS compared to the other datasets, as well as probable biases in both EFAS and ERA5. The lower anomaly correlations for SMOS are likely at least partially due to a higher noise compared to EFAS and ERA5, as well as to sudden drops in soil moisture that were observed in the time series of different gridpoints. These latter drops in soil moisture would need to be filtered before SMOS could be used directly in EFAS, e.g. using a first guess check.

Results of the CDF-matching show that this bias correction method works well for most gridpoints, with the exception of some areas where there are strong interannual variations in the seasonal soil moisture cycle. This occurs particularly in Iceland, and for some gridpoints in North Africa, and in these areas the data would need to be filtered before SMOS soil moisture could be used directly in EFAS. This could also be done using a first guess check.

Finally, in this report we discuss options for the next steps of the project, and make recommendations for different experiments that could be performed to test the use of SMOS soil moisture in EFAS.