Multi-model predictions of local climate change with uncertainty assessment using generalized likelihood uncertainty estimation and Bayesian model averaging

A number of general circulation models (GCMs) have been developed to project future global climate change and their outputs are widely used to represent local climate conditions to predict the effect of climate change on hydrology and water quality. Unfortunately, projected results for future climate change are different and it is not known which set of GCM data is better than the others. The objective of this work is to present a Bayesian approach consisting of generalized likelihood uncertainty estimation (GLUE) and Bayesian model averaging (BMA) for the estimation of local climate change with uncertainty assessment. This method is applied to Cannonsville Reservoir watershed. GCM data contributing to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC AR4), under a range of emission scenarios (20C3M, A1B, A2, and B1) are used. The GCM data for the 20C3M scenario are used to calculated the posterior probability using GLUE, while outputs for future scenarios (A1B, A2, and B1) are then processed using BMA which is a statistical procedure that infers a consensus prediction by weighing individual predictions based on the posterior probabilities obtained by GLUE, with the better performing predictions receiving higher weights than the worse performing ones. The method has the advantage of generating more reliable predictions than original GCM data. The results also indicate clearly the high reliability of the GCM data for daily average, maximum and minimum temperatures, but the reliability for daily precipitation and wind speed is low. The application supports the method presented.