Optimal Exploitation of AMSR-E Signals for Improving Soil Moisture Estimation Through Land Data Assimilation

Regional soil moisture can be estimated by assimilating satellite microwave brightness temperature into a land surface model (LSM). This paper explores how to improve soil moisture estimation based on sensitivity analysis when assimilating Advanced Microwave Scanning Radiometer for the Earth Observing System brightness temperatures. By assimilating a lower and a higher frequency combination, the land data assimilation system (LDAS) used in this paper estimates first model parameters in a calibration pass and then estimates soil moisture in an assimilation pass. The ground truth of soil moisture was collected at a soil moisture network deployed in a semiarid area of Mongolia. Analyzed are the effects of assimilating different polarizations, frequencies, and satellite overpass times on the accuracy of the estimated soil moisture. The results indicate that assimilating the horizontal polarization signal underestimates soil moisture and assimilating the daytime signal overestimates soil moisture. The former is due to improper parameter estimation perhaps caused by high sensitivity of the horizontal polarization to land surface heterogeneity, and the latter is due to the effective soil temperature for microwave emission in the daytime being close to the one at a soil depth of several centimeters but not to the surface skin temperature simulated in the LSM. Therefore, assimilating the nighttime vertical polarizations in the LDAS is recommended. A further analysis shows that assimilating different frequency combinations produces different soil moisture estimates, and none is always superior to the others, because different frequency signals may be contaminated by varying clouds and/or water vapor with different degrees. Thus, an ensemble estimation based on frequency combinations was proposed to filter off, to some extent, the stochastic frequency-dependent biases. The ensemble estimation performs more robust when driven by different forcing data.