Can Typical Land Surface Model Parameterizations Support the Expected Soil Moisture Assimilation Efficiency?

Abstract

Remote sensing (RS) soil moisture retrievals are frequently assimilated into land surface models (LSMs) to enhance model estimates. However, soil moisture data assimilation (DA) efficiency is highly model-dependent, making it imperative to investigate whether current LSMs can achieve expected DA efficiencies and identify potential model limitations for DA. Here, we examine soil moisture DA efficiency based on a typical LSM by benchmarking it against a reference soil moisture merging scheme (i.e., assigning weights to combine multiple products into a single one). Both the merged and DA soil moisture estimates are comparable since they are based on identical error estimation theory and the same RS soil moisture data sets. In theory, the DA soil moisture estimates should be superior to the merged results—since DA can characterize the temporal variation of model error and propagate DA benefits into subsequent forecast steps. However, ground-based validation results indicate that DA soil moisture performs worse than simply merged results in regions where the LSM is less precise than RS retrievals. Further combing synthetic experiment, we confirm that the unexpected DA results are primarily attributable to land parameterization uncertainty, which leads to an unrealistic representation of soil moisture events (e.g., dry-downs) and significantly hampers the DA application. Given this, soil moisture DA is likely to remain suboptimal in achieving its desired goals. Therefore, this study emphasizes the urgency and necessity of reducing model parameterization uncertainty in land DA systems.