Although we may see a few hit-or-miss snow showers through Wednesday, the work week ahead looks mainly dry, which means we're back to looking far into the future for the next storm. As such, now is a good time to discuss the experimental downscaled products from the North American Ensemble Forecast System (NAEFS) that we provide on weather.utah.edu.
Ensemble forecast systems are now de rigueur for medium-range forecasting and are quickly reaching that status for short-range forecasting. They are run by several major forecast centers. The NAEFS is comprised of ensemble forecasts produced by the U.S. National Centers for Environmental Prediction (NCEP) Global Ensemble Forecast System (GEFS) and the Canadian Meteorological Center Ensemble (CMCE).
The resolutions of the GEFS and CMCE fail to adequately resolve the influence of topography over the western U.S., especially over the interior. For example, the GEFS precipitation climatology (left panel below) is far smoother than the climatological precipitation analysis by the PRISM Climate Group (right) over the western U.S.
|Courtesy Trevor Alcott|
|Courtesy Trevor Alcott|
The NAEFS-experimental forecasts on weather.utah.edu use this approach. Although both the GEFS and CMCE are comprised of 20 members, we only use 10 from each ensemble system. Special thanks to one of my former students, Trevor Alcott, who wrote much of the code to do this.
The plume diagram below shows the NAEFS-experimental forecast produced at 0000 UTC (5 PM MST) yesterday afternoon. Note the precipitation with the front last night, then a period of mainly dry weather, and then the potential for precipitation toward the weekend.
One often sees a tendency for clustering amongst ensemble members, especially in the early stages of the forecast period. For example, the GEFS members in the graph above were wetter overnight than the CMCE members, yielding strong clustering. This is a very common problem in ensemble prediction. Ensembles based on a single modeling system like the GEFS usually are what we call underdispersive. They underestimate the full range of possible outcomes. Adding one or more additional modeling systems to the mix typically diversifies the ensemble and yields better probabilities than obtained from a single ensemble system.
What are the advantages of such an approach? Well, the resulting forecasts are better than one can obtain without downscaling, especially over the western interior where the terrain is very fine scale. Model validation work by another one of my students, Wyndam Lewis, suggests that a 5-day GEFS forecast with climatological downscaling is as good as a 3-day GEFS forecast without downscaling.
What are the disadvantages? Well, you use climatology and you get climatology, no matter what the situation. Sometimes we see situations where the mountain effects are more limited than climatology (e.g., during a frontal passage), whereas in others it can be much stronger (e.g., post-frontal snowshowers). The approach also does not correct for model biases. For example, if the GEFS is too wet in your region, you're going to get a downscaled forecast that on average is too wet.
So, overall the downscaling helps, but we can do better. That's what keeps us in business.