Assessment of bias correction techniques for satellite soil moisture data assimilation in the dry season of Southeastern South America

Estimation of surface soil moisture (SSM) is fundamental for hydrological monitoring and agricultural management, particularly in regions affected by recurrent droughts such as central–northeastern Argentina. Data assimilation techniques provide a robust approach to estimating SSM variability by integrating numerical soil models—i.e., land surface models—with observations from satellite-based remote sensors. Satellite missions based on L-band microwave sensors, such as NASA’s SMAP and ESA’s SMOS, provide global SSM retrievals, yet these observations often exhibit systematic biases arising from instrument noise, indirect measurements (i.e., observational operator), and temporal and spatial heterogeneities. Within data assimilation techniques, the ensemble Kalman filter (EnKF) has been widely employed for SSM estimation. This algorithm assumes unbiased observations; thus, bias correction becomes necessary to ensure optimal assimilation. In this study, we evaluate three off-line observational bias-correction techniques within a land data assimilation framework based on the Noah-MP v4.0.1 land surface model and an EnKF. The assessment focuses on the 2022 dry season over the endorheic Pampas region. We introduce a bias correction approach to mitigate sampling errors in cumulative distribution function (CDF) matching: (i) the climatological statistics are  computed using homogeneous soil texture pixels within the bin, and (ii) a 45-day moving temporal sampling window is used to give a smoother evolution of the CDF. During dry periods, we empirically demonstrate that soil moisture probability density functions are statistically distinguishable across different soil textures within the bin. Furthermore the monthly-fixed statistics exhibited jumps during the dry season. This approach is compared with the standard CDF matching and the normal deviate scaling. These three off-line bias correction techniques are applied to correct SMAP and SMOS satellite retrievals prior to data assimilation. We show that this improved statistical sampling for CDF matching has a non-negligible impact on the SSM estimates resulting from  EnKF, particularly during dry periods. The corrected sampling of CDF matching shows better alignment and stronger correlation with the time series of the independent in-situ soil moisture measurements. Overall, the study emphasizes the need for context-aware bias-correction techniques to enhance SSM data assimilation in regions with strong seasonal precipitation variations. Moreover, SSM estimations influence deeper model layers through vertical propagation of the information. These results motivate future work exploring how surface corrections might lead to enhanced subsurface estimates.