Joint Assimilation of Soil Moisture and Vegetation Satellite Retrievals into the Noah-MP Land Surface Model

In this study, we investigate the potential of a data assimilation (DA) system which integrates remotely sensed data of surface soil moisture (sfsm) and vegetation to enhance estimates of the Noah-MP land surface model (LSM), which dynamically simulates energy, water, and carbon fluxes. We conducted one experiment with sfsm DA, and two with vegetation DA, i.e., either assimilating optical retrievals of leaf area index (LAI) or microwave-based retrievals of vegetation optical depth (VOD). Additionally, we undertook two ‘joint DA’ experiments, which each incorporate sfsm and one of the vegetation products.

The experiments were compared against a model-only run using independent data to understand the value of assimilating these satellite products separately and together for estimating geophysical variables such as root-zone soil moisture (rzsm), evapotranspiration (ET), and gross primary production (GPP). Additionally, we evaluate key diagnostics of the assimilation system: the ensemble spread, innovations, and increments are compared between single-sensor and joint DA experiments.

We utilized the optical LAI from the Copernicus Global Land Service (CGLS), the microwave X-band VOD from the Advanced Microwave Scanning Radiometer 2 (AMSR2) LPRM version 6, and sfsm from the 36 km Soil Moisture Active/Passive (SMAP) L2 product for assimilation. In order to obtain the same climatology as the model, a seasonal rescaling was performed on the LAI and sfsm retrievals, while a machine learning approach was used to map the modeled LAI and rzsm to VOD.

Our findings highlight that solely assimilating sfsm enhances soil moisture estimates, while using only LAI or VOD predominantly improves GPP and ET estimates. The joint DA experiments, where both sfsm and vegetation observations are assimilated, capture both of these improvements in a single combined DA product. Furthermore, the joint assimilation experiments also have the smallest ensemble spread, and thus assumed uncertainty, in their estimates. Overall, our results underline the potential of multi-sensor DA, in which information from different sources is combined to improve the estimates of several land surface variables simultaneously.