Abstract:
Due to representativity differences between modeled and remotely sensed soil moisture estimates, remotely sensed soil moisture observations are typically 'bias-corrected' to match the modeled soil moisture statistics (mean, variance, and often higher order moments) prior to being assimilated. Depending on the bias correction strategy, the assimilated observations can contain a combination of signals of the observation-model discrepancies at seasonal and sub-seasonal time scales. While soil moisture assimilation is often evaluated in terms of anomalies from the mean seasonal cycle, the impact on modeled soil moisture at distinct time scales has not been explicitly established. In this study, at the four densely instruments USDA ARS Watershed sites, nine year time series of remotely sensed (AMSR-E LPRM X-band) and modeled (Catchment) near-surface soil moisture are separated into signals representing the mean seasonal cycle, inter-annual deviations from the mean seasonal cycle, and sub-seasonal-scale dynamics. Both time series are then evaluated at each time scale against the USDA ARS in situ observations. The impact of assimilating the AMSR-E soil moisture into the Catchment model using an Ensemble Kalman Filter is then established at each time scale, using a variety of bias correction methods: i) bulk statistical correction (i.e., CDF-matching) using the nine-year data record ii) bulk statistical correction using each year separately, and iii) a two-stage state and observation bias filter designed to remove seasonal- and longer-scale discrepancies between the model and observations. In addition to informing the selection of bias correction approaches, this study also enhances our understanding of the errors in the AMSR-E and Catchment modeled soil moisture.