Reducing bias-corrected precipitation projection uncertainties: a Bayesian-based indicator-weighting approach

In this paper, a Bayesian-based indicator-weighting approach is developed to reduce the uncertainty resulting from bias-correcting projection outputs from multiple general circulations models (GCMs). The approach decides whether or not a projection from a given GCM output should be used depending on how close output from the GCM's retrospective run was to past observation (bias criterion) and agrees with the consensus (convergence criterion) estimate of all future GCM projections in a "truth-centered" statistical framework. Indicator weights are derived by equating present day versus future changes in mean precipitation of individual GCM output to the one obtained from the posterior distribution of all GCMs using a Markov Chain Monte Carlo algorithm. Use of GCMs outputs in hydrological impact studies requires downscaling and/or bias correction steps in order to account for discrepancies between small and large scale land-atmospheric processes. One of the most popular techniques for bias-correcting retrospective GCM outputs is the cumulative distribution functions matching approach based on observed precipitation. Future GCM projections are then adjusted depending on the bias correction results of retrospective outputs. In this sense, the bias correction process introduces variability/uncertainty into GCM outputs resulting in a wide range of projected values. If more than one GCM is used, the range of variability/uncertainty further increases. The approach that is used to reduce this uncertainty is demonstrated using 23 GCM outputs of CMIP5 model runs for west central Florida.