Sources of Uncertainty in Replicating Hydro-climatic Extremes

English Abstract:

Effective communication of projected hydrologic extremes starts with a discussion of uncertainty. We perform an analysis using 56 downscaled (2 x observations; 4 x reanalyses; 7 x downscaling methods) results to drive the Variable Infiltration Capacity hydrologic model. Performance of each method is tested for timing and distribution of 3-day peak flow, 7-day low flow and 26 ClimDEX indices. Downscaling methods include bias corrected constructed analogues (BCCA), double BCCA (DBCCA), climate imprint (CI), bias corrected climate imprint (BCCI), bias correction spatial disaggregation based on mean temperature (BCSD), BCSD based on minimum and maximum temperature (BCSDX), and BCCA bias corrected with BCCI (BCCAQ). NCEP1, ERA40, ERAInt and 20CR are the reanalyses downscaled and two separate observational datasets are used as the downscaling target. The ability to replicate the timing and distribution of ClimDEX indices depended on downscaling technique, reanalyses and gridded-observation in order of importance. The strongest method was DBCCA, while the weakest was BCSDX. 3-day peak flow performance was more dependent on reanalysis than downscaling technique. ERA40 results were strongest and NCEP1 weakest. In the case of 7-day low flow, results depended most strongly on downscaling method, CI failed to replicate the distribution, while BCSD and BCSDX failed to replicate the timing regardless of reanalyses; these poorly performing methods over estimated low flow volumes. BCCI and BCCAQ tied for best combined performance for 7-day low flow and 3-day peak flow. Uncertainty can be reduced by careful selection of downscaling methods for those metrics most sensitive to downscaling (7-day low flow), but can’t be reduced where global climate model uncertainty plays a more important role (3-day peak flow).