Wednesday, October 24, 2012

Computer Model Performance

With eyes starting to glance at each computer model run of the GFS and Euro regarding the potential coastal storm...and in winter with every other model hinted storm...the question occasionally pops up about which model performs better overall?

In terms of simple "score" of the models five days out, the European computer model slightly outperforms the GFS and UK and by a a pretty wide margin beats the other minor guidance out there.  The chart below shows the 500mb (mid atmosphere) verification between forecast and reality five days out for the whole of the northern hemisphere, charted over the past 30-31 days.  The Euro and GFS aren't terribly far apart (.892 and .882 are pretty close as of last night's scoring) and both historically outperform the other models, which is why forecasters give these two models a lot more weight than the others.

Verification data is useful to a point -- it gives you an idea of a model's errors and how frequently another computer model is outdoing the model, perhaps factoring in if a storm or major anomaly event is causing the error or if it were a case of garbage in, garbage out.

Breaking it down to each of the major runs of the GFS (just labeled 0, 6, 12, and 18z) and Euro, you can see the performance of the major computer model runs through the course of the day.  The midday run of the Euro over the past 31 days has performed better five days out but the performance average does vary from day to day...and occasionally the models will have some struggles.

These aren't meant to be "be all, end all" in terms of saying a specific model run is outperforming the others but if both forecasts of the Euro outperform the GFS (which they do), it helps forecasters get an idea of which model tends to get the big picture right more often than not.  Past performance does not necessarily result in similar over the future but it does give you an idea of which model does better (the Euro) historically and why some forecasters tend to rely on it a tad more heavily than the GFS.

This information is found on our computer model page if you are ever interested in seeing this at any other point in the winter, summer, or in between.