Development of a Bayesian Probabilistic Model for the NARCCAP Regional Climate Model Ensemble Using Information from the Driving GCMs to Formulate Priors and Application to Hydrology Impacts

English Abstract:

In this talk we will describe the development of a joint Bayesian Probabilistic Model for the climate change results of the North American Regional Climate Change Assessment Program (NARCCAP) that uses a unique prior in the model formulation. We use the climate change results (joint distribution of seasonal temperature and precipitation changes (future vs. current)) from the global climate models (GCMs) that provided boundary conditions for the six different regional climate models used in the program as informative priors for the bivariate Bayesian Model. The two variables involved are seasonal temperature and precipitation over sub-regions (i.e., Bukovsky Regions) of the full NARCCAP domain. The basic approach to the joint Bayesian hierarchical model follows the approach of Tebaldi and Sansó (2009). We will compare model results using informative (i.e., GCM information) as well as uninformative priors. These results will be used in the context of hydrologic model sensitivities to ranges of temperature and precipitation results to determine the likelihoods of future climate conditions that cannot be accommodated by possible adaptation options.