It’s a question everyone asks in regards to hurricane season, and weather in general: What computer model is the best?
Well, what sounds like a simple question actually has a complex answer. First of all, what are we asking about? Track forecast or the intensity forecast? Before I dive into this answer I would recommend you read Dr. Jeff Masters’ explanation on the different kinds computer models and some model jargon. Also, for a more detailed explanation, the National Hurricane Center (NHC) just released the verification of their 2012 forecasts as well as the forecasts of the computer models they use. It is important to note that for the remainder of this piece I will simply focus on the forecasts for tropical cyclones in the North Atlantic Basin.
I will start off with the track forecast.
Over the past few decades the NHC, as well as the computer models, have gotten pretty good at forecasting the track of the storm. The way this is measured can be done in two ways: Either by 1) total forecast error compared to the actual storm track or 2) the forecast skill relative to what we know from climatology. The NHC uses CLIPER5 as a climatology baseline.
Over the last 3 years, the European Model known as the ECMWF, but above listed as the EMXI, has performed better than any other model. In fact, it has virtually the same amount of skill as the NHC official (OFCL above) forecasts. Also to note is that the GFS (GFSI above) and the two main consensus models (TVCN and the FSSE) performed very well. Consensus models are not “true models” but rather some sort of combination of multiple models.
However, the world of modeling is very complex and each year has its ups and downs. Furthermore, models are continually being tinkered with in hopes of improving their forecasts. In fact, last year the GFS received a large upgrade to the way it operates and yielded very promising results:
The NHC stated that the GFS was the top performing model last year, and that is particularly clear looking at the 72 hour forecasts above. In the medium range of 120 hours, the Euro did slightly outperform the GFS. Also of interest is that last year the FSSE consistently outperformed the NHC forecasts and every other model. Even simple statistical models like the BAMM can perform very well in the medium range of 120 hours.
It will be interesting to see what computer models perform well for the upcoming hurricane season. The Canadian model (CMC) just received a large upgrade and has been performing very well for winter weather. Perhaps it will be able to see a large leap in skill this year. Also, will the FSSE continue its dominance over the other models? Will the GFS continue the skill it showed last season? We will see.
I will be updating again with an analysis of computer model intensity forecasts. These forecasts have shown relatively far less skill than track forecasts.