Evaluation of Spatio-Temporal Variability in Land Surface Models using Objective Regionalization

Land Surface Models (LSMs) provide the link between observed or simulated meteorology and surface energy and hydrological states. This makes them critical both for simulating the coupled climate system and for applying meteorological datasets and models to problems of food security and water resources planning. As heavily parameterized, process-based models, however, LSMs also contribute to uncertainty in coupled climate simulations and in hydrological estimates. This uncertainty can include spatial biases that result from a combination of errors in the meteorological forcing and LSM errors related to parameter maps and limitations in process representation. The presence of these spatial biases means that traditional grid-to-grid verification methods, which implicitly assume that there is no spatial bias, inefficient for extracting useful model information, identifying model weaknesses, and providing diagnostic metrics to improve model performance. Objective regionalization offers a tool for testing the spatial extent and regional connectivity of climatic and environmental changes deduced from numerical model outputs and/or observed data. In this work, we use objective climate regionalization to evaluate the performance of LSMs and to capture relevant similarities/differences between model outputs and observations. A multi-LSM ensemble of simulations is performed using the NASA Land Information System (LIS). Simulated spatial patterns of temporal variability in soil moisture and evapotranspiration, and in the relationship between these predicted fields and meteorological forcing variables, are compared across models to identify systematic differences between LSMs. The model simulations are also compared to satellite-derived observations to assess LSM ability to capture observed spatio-temporal variability. This approach allows us to map model predictions onto observations at the scale of analogous homogeneous regions as opposed to static grid-to-grid geographic associations. This work aims to provide systematic methods and diagnostic metrics for using objective regionalization to evaluate and improve LSMs, and dynamical models in general, through spatial bias correction.