Ensemble Averaging Methods for Quantifying Uncertainty Sources in Modeling Climate Change Impact on Runoff Projection

The study discusses future runoff projection and uncertainty arising from different choices of global circulation models (GCMs), different scenarios of future emission, and different statistical downscaling methods for two watersheds in south Arkansas and north Louisiana. Three ensemble averaging methods, simple model averaging (SMA), reliability ensemble averaging (REA), and hierarchical Bayesian model averaging (HBMA), are used to compare the projected ensemble average and variance of future runoff derived from hydroclimate models. Contributions of individual sources of uncertainty are quantified by the analysis of variance (ANOVA) method and the HBMA method. An ensemble of 78 climate change projections, derived from 13 GCMs from the coupled model intercomparison project phase 5 (CMIP5), two representative concentration pathways (RCPs), and three statistical downscaling methods, are used as the forcing input to the hydrologic evaluation of landfill performance model version 3 (HELP3) to project future runoff. The result shows that HBMA performs slightly better than SMA and REA in reproducing the historical mean annual cycle of runoff. Fall runoff would increase and winter and spring runoff would decrease toward the late century. Uncertainty analysis by both ANOVA and HBMA concludes that GCMs are the major source of uncertainty, followed by the downscaling methods and then the emission scenarios. GCM uncertainty is more significant in spring and summer than other seasons. Downscaling method uncertainty shows increases in fall and winter. Emission scenario uncertainty is shown to be significant only in winter and spring for the late century. (C) 2016 American Society of Civil Engineers.