Steps 1–3 allow users to accomplish the bias-correction procedure for both NCEP/GEFS and CMC/GEFS. Note that this procedure contains two options. The first is that the NCEP/GEFS and CMC/GEFS can be grouped together before postprocessing, and then the bias correction is applied to the joint ensemble. The second option is to apply the bias correction to the NCEP and CMC ensembles separately and then the NCEP/GEFS and CMC/GEFS are grouped together after postprocessing. Although the first option is an easy approach, it may not provide the best results since each participating ensemble may have unique biases due to its own ensemble generation configuration. Specific treatments are needed for each participating ensemble. NCEP uses one model and perturbed initial conditions to create its ensemble (Toth et al. 2012; Wei et al. 2008). The model-related systematic errors grow with lead time and it is assumed that the forecast errors obtained from the ensemble mean can stand in for the systematic errors. The NCEP/GEFS biases are estimated from the ensemble mean with respect to the NCEP analysis, and the same bias estimation is applied to each ensemble member during the calibration. On the other hand, the Canadian ensemble includes 20 perturbed forecasts and 1 control forecast. All are performed with the GEM model but use different physics parameterizations, data assimilation cycles, and sets of perturbed observations (http://www.weatheroffice.gc.ca/ ensemble/index_e.html). Therefore, the individual bias is estimated and used to correct each individual member independently. For ease and efficiency, each participating ensemble calibrates its raw forecast against its own analysis.