Comparing Land Surface Model Performance between Fine-scale and Coarse-scale Assimilation: Leaf Area Index Retrievals versus GRACE / GRACE-FO Retrievals

Land surface models (LSMs) are useful for estimating land surface states and fluxes such as snow water equivalent, soil moisture content, vegetation, and river discharge. Although estimates are continuous in time and space, LSMs are flawed since they lack comprehensive representation of process physics and are dependent on uncertain boundary conditions. One way to improve LSM performance is by conditioning model states on space-borne retrievals using an ensemble Kalman filter framework.

 

In this study, the Noah-MP version 4.0.1 LSM without conditioning (a.k.a., Open Loop; OL) is compared against the same model with conditioning (a.k.a., Data Assimilation; DA). Two different univariate DA experiments are conducted:  1) assimilation using terrestrial water storage (TWS) anomalies from GRACE / GRACE-FO, and 2) assimilation using leaf area index (LAI) retrievals from MODIS. Not only do the experiments assimilate different types of retrievals (i.e., TWS versus LAI), but also, they assimilate products of different spatial (~3° versus ~0.005°) and temporal (~monthly versus ~weekly) resolutions. Experiments are conducted across different watersheds in North America with a particular focus on basins with irrigated agriculture. Modeled states and fluxes from the OL and DA are then compared against independent, ground-based measurement networks including U.S. SNOTEL, Canadian CanSWE product, U.S. SCAN for soil moisture, and USGS measurement gauges for river discharge. Statistical analyses, including bias, RMSE, and normalized information content (NIC) are computed to quantify the marginal improvements via each assimilation experiment. Results provide a basis to better understand the coupling between different state variables (i.e., snow mass, soil moisture, and groundwater) as well as the utility of using coarse-scale and fine-scale retrievals in land data assimilation.