Error Characterization of Coupled Land Surface-Radiative Transfer Models for Snow Microwave Radiance Assimilation

Snow microwave radiance assimilation (RA) or brightness temperature data assimilation (DA) has shown promise for improving snow water equivalent (SWE) estimation. A successful RA study requires, however, an analysis of the error characteristics of coupled land surface-radiative transfer models (LSM/RTMs). This paper focuses on the Community Land Model version 4 (CLM4) as the land-surface model and on the microwave emission model for layered snowpacks (MEMLS) and the dense media radiative transfer multilayer (DMRT-ML) model as RTMs. Using the National Aeronautics and Space Administration Cold Land Processes Field Experiment (CLPX) data sets and through synthetic experiments, the errors of the coupled CLM4/DMRT-ML and CLM4/MEMLS are characterized by: 1) evaluating the CLM4 snowpack state simulations; 2) assessing the performance of RTMs in simulating the brightness temperature (TB); and 3) analyzing the correlations between the SWE error (ε_SWE) and the TB error (ε_TB) from the RA perspective. The results using the CLPX data sets show that, given a large error of the snow grain radius (ε_re) under dry snowpack conditions (along with a small error of the snow temperature (ε_Tsnow)), the correlations between ε_SWE and ε_TB are mainly determined by the relationship between ε_re and the snow depth error (ε_dsnow) or the snow density error (ε_ρsnow). The synthetic experiments were carried out for the CLPX region (shallow snowpack conditions) and the Rocky Mountains (deep snowpack conditions) using the atmospheric ensemble reanalysis produced by the coupled DA Research Testbed/Community Atmospheric Model (CAM4). The synthetic experiments support the results from the CLPX data sets and show that the errors of soil (the water content and the temperature), snow wetness, and snow temperature mostly re- ult in positive correlations between ε_SWE and ε_TB. CLM4/DMRT-ML and CLM4/MEMLS tend to produce varying RA performance, with more positive and negative correlations between ε_SWE and ε_TB, respectively. These results suggest the necessity of using multiple snowpack RTMs in RA to improve the SWE estimation at the continental scale. The results in this paper also show that the magnitude of ε_re and its relationship to ε_SWE are important for the RA performance. Most of the SWE estimations in RA are improved when ε_SWE and ε_re show a high positive correlation (greater than 0.5).