Improving Spatial Patterns Prior to Land Surface Data Assimilation via Model Calibration Using SMAP Surface Soil Moisture Data

Abstract

AbstractPrior to their application in a land surface data assimilation (DA) system, remotely sensed (RS) soil moisture (SM) products are typically rescaled to remove systematic differences with respect to comparable SM estimates obtained from a land surface model (LSM). This preprocessing of RS SM data—commonly referred to as bias correction—implicitly discards spatial information present in the RS SM retrievals. Here, based on dense SM network measurements obtained within the Huai River Basin of China, we demonstrate that L‐band Soil Moisture Active Passive (SMAP) L3 surface SM retrievals provide a better representation of spatial variability in time‐averaged SM fields than a LSM—suggesting that, as typically applied, bias correction is neglecting important spatial information present in RS SM products. To maximally use RS information, we propose an alternative bias correction approach for SM DA that provides the benefits of traditional rescaling while also improving modeled SM spatial patterns. Specifically, the regional mean of SMAP SM is first corrected using modeled SM of the Variable Infiltration Capacity (VIC) model. Then, two key VIC soil parameters (EXPT and BULKD) are calibrated to match the relative subregional spatial variability captured by the SMAP SM product. Results show that our model calibration method successfully improves VIC SM spatial patterns and pixel‐wise time series. Furthermore, these SM improvements translate into enhanced VIC streamflow estimates. Overall, results suggest that the proposed bias correction framework can improve current land surface DA systems by maximally utilizing spatial information contained in RS SM products.