Soil moisture initialization for climate prediction: Characterization of model and observation errors

While it has been shown that soil moisture data assimilation techniques can be used to constrain land surface model predictions with remotely sensed soil moisture observations to provide optimal climate model surface and root zone soil moisture initialization, a good understanding and quantification of both model and observation error are required. In this paper we therefore evaluate the catchment‐based land surface model (CLSM) and scanning multichannel microwave radiometer (SMMR) soil moisture estimation errors using long‐term in situ soil moisture measurements available for Eurasia. Generally, the CLSM surface and root zone soil moisture was found to be biased less than 0.08 vol/vol dry in dry climate and frozen soil areas and biased over 0.08 vol/vol (as high as 0.16 vol/vol) wet in wet climate areas. Moreover, the CLSM suffered from an underestimation in surface zone seasonal soil moisture variation. While the SMMR soil moisture estimates were also biased, less than 0.05 vol/vol dry in dry climate and over 0.10 vol/vol (as high as 0.2 vol/vol) wet in wet climate, they generally had accurate seasonal variations. This error characterization study is crucial for practical Eurasian data assimilation, as unbiased observations and model predictions, and reliable knowledge of relative observed and model predicted soil moisture errors are key data assimilation assumptions. This study therefore provides the error information required for data assimilation and emphasizes the need for careful bias representation when assimilating SMMR data into the CLSM. The potential deficiencies in this error assessment are acknowledged and discussed, including the disparate time‐space representation of the various soil moisture sources.