At the Canadian Meteorological Centre (CMC), humidex site forecasts are currently calculated using dry-bulb temperatures and dew-point spreads interpolated from numerical models, either the Regional or Global Deterministic Prediction System (respectively RDPS and GDPS) depending on the time of validity.

Dew-point spreads from those models present important biases and errors, therefore statistical forecast models using CMC Updatable Model Output Statistics (UMOS) system were developed for each site with valid dew-point temperature observations. Same as the currently operational UMOS dry-bulb temperature predictand, Multiple Linear Regression was applied on a 2-year sample stratified seasonally and 3-hourly. A reduction of variance criterion was used for predictors' selection. UMOS dry-bulb and dew-point temperature forecast models were developed independently; therefore a coherence scheme was developed to resolve cases where UMOS dew-point temperature exceeded dry-bulb temperature.

For dissemination purposes, public weather forecasting requires forecasts of dry-bulb and dew-point temperatures at sites where no observations are available. In the past, a weighted average of the sums-of-square-and-cross-products (SSCP) matrices of nearby stations was used to produce a synthetic SSCP matrix representing a fictitious site. For other sites not included in the processes discussed above, further interpolation was done using kriging and the average of all UMOS forecasts as a trial field.

In order to improve forecast performance and simplify maintenance, the synthetic matrix approach was dropped and the old kriging software was replaced by MIST (Moteur d'Interpolation STatistique), an optimum interpolation procedure already used in the Canadian Precipitation Analysis (CaPA) system. Among other advantages, MIST has the capacity to use the GEM-Regional outputs as trial fields and includes a leave-one-out cross-validation (LOOCV) procedure. For comparison purposes, a LOOCV procedure was also developed for the synthetic SSCP matrix methodology.

Verification results of UMOS and RDPS dew-point temperature forecasts using an independent 1-year sample will be presented showing that UMOS brings significant improvements in terms of bias and error. In addition, results from the synthetic SSCP matrix and MIST LOOCV procedure will be presented to demonstrate the MIST procedure's legitimacy.