Evaluating a High-resolution Operational LDAS/LSM with In-situ Soil Measurements Throughout North Carolina

Evaluating -- and particularly benchmarking -- the performance of land-surface models that assimilate data is challenging for a wide-variety of reasons. With respect to in-situ data, several different sensor classes are often needed to "close" the water budget -- although full closure may still be pragmatically out of reach. Thus, it may be impossible to fully understand which model processes or geophysical parameters are responsible for bias error of varying sign in one versus another particular location. In our case, we have nearly one-year of in-situ soil moisture and temperature data from nearly 40 locations in the state of North Carolina, and for most of those locations also have soil survey data including modest information about aquifer characteristics. For that same period, we have archived 1km scale operational radar-based QPE-assimilating LDAS/LSM data interpolated to the depth of the sensor observations. We are in the process of quantifying bias error, RMSE, correlation coefficients, Index of Agreement, and other discreet statistics while also looking at categorical/threshold based statistics useful for agriculture, such as the reference and wilting soil moisture volumetric contents. Particular attention is being paid to modeled versus measured soil parameters such as porosity and hydraulic conductivity, as well as geographic location, to gain insight into the possible source of model error, including both relative and absolute measures of soil saturation. In this paper, we will present an overall summary of our findings. We hope this presentation will add value to the ongoing validation/verification and benchmarking challenge presented by operational LSMs.