Validation of a High-Resolution Numerical Weather Prediction Land Surface Scheme Using Catchment Water Balances

An adequate representation of the interaction between the land surface and the atmosphere is critical for both numerical weather prediction and climate models. The surface energy and mass balances are tightly coupled to the terrestrial water cycle, mainly through the state of soil moisture. An inadequate representation of the terrestrial water cycle will deteriorate the state of the land surface model and introduce biases to the atmospheric model. The validation of land surface models is challenging, as there are very few observations and the soil is highly heterogeneous. In this paper, a validation framework for land surface schemes based on catchment mass balances is presented. The main focus of our development lies in the application to kilometer-resolution numerical weather prediction and climate models, although the approach is scalable in both space and time. The methodology combines information from multiple observation-based datasets. Observational uncertainties are estimated by using independent sets of observations. It is shown that the combination of observation-based datasets and river discharge measurements close the water balance fairly well for the chosen catchments. As a showcase application, the framework is then applied to compare and validate four different versions of TERRA ML, the land surface scheme of the COSMO numerical weather prediction and climate model over five mesoscale catchments in Switzerland ranging from 105 to 1713 km2. Despite large observational uncertainties, validation results clearly suggest that errors in terrestrial storage changes are closely linked to errors in runoff generation and emphasize the crucial role of infiltration processes.