Spatio-Temporal Modeling for Regional Climate Model Evaluation: Eigenvector Filtering Versus Bayesian CAR

A suite of 10-year ensemble-based simulations was conducted to investigate the hydroclimatic impacts due to large-scale deployment of perennial bioenergy crops across the continental United States. Given the deterministic nature of the simulations, uncertainties of hydroclimatic impacts caused by physics parameterizations exist within the ensemble. Hierarchical Bayesian space-time models could be used to evaluate physics parameterizations and biofuel impacts; unfortunately such approaches are computationally expensive. In this work, a computationally efficient scheme based on spatio-temporal eigenvector filtering (hereafter: STEF) is investigated. Three approaches of introducing proxy variables to the STEF model are specified in order to alleviate the spatial confounding problem and a fast approximation of penalized estimators is implemented to select proxy variables efficiently. Monte Carlo experiments are conducted using the proposed methods; results are evaluated against those obtained from three alternative Bayesian spatio-temporal specifications. Finally, STEF is applied to quantify the robustness of simulated hydroclimatic impacts associated with bioenergy crops to alternative physics parameterizations.