The North American Land Data Assimilation System (NLDAS) Science Testbed: An Environment for the Systematic Evaluation and Benchmarking of NLDAS Outputs

The North American Land Data Assimilation System (NLDAS) produces hourly land-surface meteorology and surface states, including precipitation, soil moisture and temperature, snow cover/amount, evapotranspiration, runoff, and streamflow. The NLDAS domain extends over North America from 25-53 North on a 1/8th-degree grid. NLDAS is a collaborative project between NCEP/EMC, NASA/GSFC, Princeton University, the University of Washington, NWS/OHD, and NCEP/CPC. The current Phase 2 of NLDAS employs four unique land-surface models (LSMs), each driven separately by the surface meteorological forcing. Phase 2 datasets extend from Jan 1979, and are run operationally at NOAA/NCEP, updated daily in near real-time with a 3.5-day lag. The next phase of NLDAS will include later versions of these LSMs, as well as several more recently developed LSMs that are being brought into the NLDAS project. These models are run using the NASA/GSFC-developed Land Information System (LIS) software framework. The LIS software allows a common driver for all LSMs as well as the assimilation of remotely-sensed soil moisture and snow states and of terrestrial water storage anomalies to improve model states and fluxes.

This work will present results from the NLDAS science testbed, which is an extensive evaluation and benchmarking environment developed to systematically evaluate the results from the NLDAS models. In this presentation, the new/upgraded LSMs will be compared against the Phase 2 versions of the LSMs under the NLDAS configuration. The Land Verification Toolkit (LVT), also developed at NASA/GSFC, will be used to evaluate the LSMs against available observations including streamflow, in situ soil moisture, snow products, surface fluxes, and groundwater. This suite of tests will examine the effects of upgraded model physics through a comprehensive suite of evaluation metrics (i.e., anomaly correlation, RMSE, bias, Nash-Sutcliffe efficiency, Taylor Skill Scores, etc.). Additionally, the benchmarking capabilities within LVT will be used to evaluate how the advanced physics within the LSMs compare with the a priori benchmarks developed from simple regression models. The goal of this work is to quantify the changes in model performance in the NLDAS environment from the new/upgraded LSMs.