UMOS statistically post-processed Forecast of the Regional Deterministic Prediction System (RDPS-UMOS-MLR)

Statistical post-processing of weather and environmental forecasts issued by numerical models, including the Regional Deterministic Prediction System (RDPS), reduces systematic bias and error variance of raw numerical forecasts. This is achieved by establishing an optimal relationship between observations recorded at stations and co-located numerical model outputs. The Updatable Model Output Statistics (UMOS) system at Environment Canada carries out this task. The statistical relationships are built using the Model Output Statistics (MOS) method and a multiple linear regression (MLR) technic. The weather and environmental variables being statistically post-processed by UMOS include air temperature and dew point temperature at approximately 1.5 meters above ground as well as wind speed and direction at 10 meters above ground or at the anemometer level in the case of a buoy. The absence of a statistically post-processed forecast can be caused by a missing statistical model due to insufficient observation data quality or quantity. In addition, the absence of a post-processed forecast for wind direction could also be due to weak forecasted wind components preventing the calculation of reliable results. The forecasts of wind speed and direction are produced from independent statistical post-processing models. Geographical coverage includes weather stations across Canada. Statistically post-processed forecasts is available at the same frequency of emission as the numerical model producing the raw forecasts and at 3-hourly lead times for the RDPS.

• Statistical post-processing
• Machine learning
• Multiple linear regression
• UMOS
• Environmental forecast
• Point forecast
• Weather forecasts
• Air temperature
• Wind