Validation of climate model output using Bayesian statistical methods

The growing interest in and emphasis on high spatial resolution estimates of future climate has demonstrated the need to apply regional climate models (RCMs) to that problem. As a consequence, the need for validation of these models, an assessment of how well an RCM reproduces a known climate, has also grown. Validation is often performed by comparing RCM output to gridded climate datasets and/or station data. The primary disadvantage of using gridded climate datasets is that the spatial resolution is almost always different and generally coarser than climate model output. We have used a Bayesian statistical model derived from observational data to validate RCM output. We used surface air temperature (SAT) data from 109 observational stations in California, all with records of approximately 50 years in length, and created a statistical model based on this data. The statistical model takes into account the elevation of the station, distance from coastline, and the NOAA climate region in which the station resides. Analysis indicates that the statistical model provides reliable estimates of the mean monthly SAT at any given station. In our method, the uncertainty in the estimates produced by the statistical model are directly determined by obtaining probability density functions for predicted SATs. This statistical model is then used to estimate average SATs corresponding to each of the climate model grid cells. These estimates are compared to the output of the RCM to assess how well the RCM matches the observed climate as defined by the statistical model. Overall, the match between the RCM output and the statistical model is good, with some deficiencies likely due in part to the representation of topography in the RCM.