Analysis of Systematic and Random Error Components in Soil Moisture Estimates from Land Surface Models and Satellite Retrievals

The importance of soil moisture as a critical variable in the climate system is well established. Soil moisture regulates the water and energy exchange between the land surface and the overlying atmosphere. Therefore, realistic characterization of soil moisture helps to better understand the interaction of land surface and atmosphere. It is a long-established fact that the soil moisture estimates from different sources (i.e., in-situ, model, and satellite) show large discrepancies in both their temporal variability and long-term means. As a result, the global climatology of soil moisture is undetermined. Characterizing random error in soil moisture estimates has already been addressed in numerous studies. In particular, data assimilation approaches are often used to combine the information from models and satellite retrievals. These methods, however, only deal with the correction of the random errors components. Most data assimilation and intercomparison studies ignore the systematic error component in the soil moisture estimates, though such biases are often the source of important signals due to the significant heterogeneity and human modifications of the land surface.

In this study, we use an error decomposition methodology to partition the errors in satellite retrievals and land surface model soil moisture estimates into systematic and random components.

The results indicate that the systematic error in each of these soil moisture estimates is significantly higher than the random error component. At the domain-averaged scale, the systematic error component accounts for 88% and 63% of the total mean squared error, for SMAP and SMOS retrievals. Similarly, across the LSMs, the contribution of the systematic error varies between 78% to 95%, averaged over the NLDAS2 domain. As a result, methods such as data assimilation focused on improving the random error component are likely to provide small improvements. The study also highlights the need for developing alternative methods for improving the inherent “observability” and information content of soil moisture estimates.