Approximating within-GCM uncertainty for hydrologic climate change impact assessments

Abstract:

Hydrologic climate change impact assessments primarily rely upon the set of General Circulation Model (GCM) projections of past and future climate available within the CMIP3 and CMIP5 datasets. These projections provide insight into likely future climates under the enhanced greenhouse effect from a range of GCMs. However, the number of runs from each GCM for a given scenario is small, usually <10 runs per GCM, which limits any assessment of the impact of within-GCM uncertainty on projected hydrologic climate change. Within GCM uncertainty is the variability in GCM output that occurs when running a scenario multiple times but each run has slightly different, but equally plausible, initial conditions or parameter settings. Precise quantification of within-GCM uncertainty requires numerous, for example 1000, runs of each scenario for each GCM, which are currently unavailable. Here we demonstrate a method to approximate within-GCM uncertainty using non-stationary stochastic data generation. At 17 catchments around the world we generate 100 stochastic replicates of catchment average monthly GCM precipitation and temperature from 5 GCMs from CMIP3 (20C3M and A1B scenarios), apply a bias-correction and input them into a calibrated hydrologic model to obtain 100 projections of runoff at each catchment. We investigate the contribution of within-GCM uncertainty to uncertainty in projected precipitation, temperature, runoff and reservoir yield for two 30-year windows centred around 1980 and 2030. We show the impact of within-GCM uncertainty is amplified from precipitation into runoff and that within-GCM uncertainty is an important component of the uncertainty in hydrologic climate change impact assessments.