Benefits of a Multimodel Ensemble for Hub-Height Wind Prediction in Mountainous Terrain

While numerical weather prediction (NWP) models are often used to estimate future wind power, no single NWP model is perfect. A better approach is to run many models (an ensemble) and use the average to estimate future wind speeds. The ensemble can also be used derive valuable knowledge of wind-forecast uncertainties. This presentation demonstrates the benefits of using a multi-model ensemble to predict wind speeds at wind-turbine hub heights over using single deterministic ensemble members, and using climatology. We do this for a one-year period at four wind farms in mountainous terrain in British Columbia. The ensemble mean has higher forecast-accuracy than climatology until a forecast horizon of 6.5 days. The ensemble mean is more highly correlated to the observations than climatology through the 7-day forecast horizon tested. Use of the ensemble-mean forecast results in a gain in skill advantage (increase in time that a forecast remains more skilled than climatology) of at least a 1-2 days over use a single, deterministic ensemble member for both forecast accuracy and correlation. For probabilistic forecasts, use of the multi-model ensemble mean is most beneficial to improvements in probabilistic sharpness (narrowing of uncertainty). A comparison of Weather Research and Forecasting (WRF) model ensemble-member forecasts initialized by the National Centers for Environmental Prediction Global Forecast System (GFS) and North American Mesoscale (NAM) models, the Canadian Meteorological Centre Global Deterministic Prediction System (GDPS/GEM), and Fleet Numerical Meteorology and Oceanography Center Navy Global Environmental Model (NAVGEM) showed that the GDPS/GEM provided the best initial conditions for the locations tested for the year-long case study.