Validation of lateral boundary conditions for regional climate models.

摘要

California boasts a population of more than 34 million and is the tenth largest energy consumer in the world. As such, the California Energy Commission (CEC) is greatly concerned about the environmental impacts of global climate change on energy needs, production and distribution. In order to better understand future energy needs in California, the CEC depends upon international climate scientists who use results from simulations of western U.S. regional climate models (RCMs). High-resolution RCMs are driven by coupled Atmosphere/Ocean General Circulation Model (AOGCM) simulations along lateral surface boundaries outlining the region of interest. For projections of future climate, however, when the RCM is driven by future climate change output from an AOGCM, the performance of an RCM will depend to some degree on the merit of the AOGCM. The objective of this study is to provide tools to assist with model validation of coupled Atmosphere/Ocean General Circulation Model (AOGCM) simulations against present-day observations.;A comparison technique frequently utilized by climate scientists is multiple hypothesis testing, which identifies statistically significant regions of difference between spatial fields. In order to use these methods, the AOGCM fields must be interpolated onto the reanalysis grid. In this work, I present an efficient interpolation technique using thin-plate splines. I then compare significant regions of difference using multiple testing procedures of Bonferoni against the false detection rate methodology. A major drawback of multiple hypothesis methods is that they do not account for correlation in the spatial field. I introduce and employ measures of comparison, including the Mahalanobis distance measure, that account for anisotropy within the spatial field. Bayesian techniques are applied to calculate comparison measures between the driver-GCM lateral surface boundaries and the NCEP/NCAR and ERA40 reanalysis data sets. I find that the Mahalanobis measure provides a systematic ranking of model performance against present-day observations.