Evaluation of ensemble precipitation forecasts generated through postprocessing in a Canadian catchment

Flooding in Canada is often caused by heavy rainfall during the snowmelt period. Hydrologic forecast centers rely on precipitation forecasts obtained from numerical weather prediction (NWP) models to enforce hydrological models for streamflow forecasting. The uncertainties in raw quantitative precipitation forecasts (QPFs) are enhanced by physiography and orography effect over diverse landscape, particularly in the western catchments of Canada. A Bayesian post-processing approach called rainfall-post processing (RPP), developed in Australia (Robertson et al., 2013; Shrestha et al., 2015), has been applied to assess its forecast performance in a Canadian catchment. Raw QPFs obtained from two sources, Global ensemble forecasting system (GEFS) Reforecast 2 project from National Centers for Environmental Protection (NCEP), and Global deterministic forecast system (GDPS) from Environment and Climate Change Canada (ECCC) are used in this study. The study period from Jan 2013 to Dec 2015 covered a major flood event in Calgary, Alberta, Canada. Post-processed results show that the RPP is able to remove the bias, and reduce the continuous ranked probability score of both GEFS and GDPS forecasts. Ensembles generated from the RPP better depict the forecast uncertainty.