Why looking at any one model for a storm is a bad idea
Over the past few days, I have already started to see the chorus of accolades for the European model on its handling of Irma. It did an amazing job with the storm track for many days including the all important 5 day period that the hurricane center forecast is made for. In fact many times in the technical discussions from the NHC they noted that the model was performing well and they were shading their track towards that direction. It should be pointed out the NHC 5 day forecast cone did amazingly well, and the average error appeared to be below the 5-year average error for the cone. Below is an animation of every 5-day forecast cone from the beginning until 8-am Monday morning 9/11/2017. That’s a pretty dang good forecast. 
How the models did:
There are loads of models that we look at they are called numerical weather prediction guidance for a reason. They are guidance and not meant to be a forecast but a tool that a forecaster uses to make a prediction. Too many people think they do the work for you. Like any skill, a trained meteorologist uses these tools to make a forecast. Just the model alone can’t build a forecast(unless it’s an app) just like some wood, a saw and nails can’t just build a house without a qualified contractor to build it.
Track Forecast:
The ECMWF or European model did an outstanding job with Irma. Below is the average position error over 7 days. Remember the NHC only does a 5-day forecast so it doesn’t go out to 7 days. While yes the ECMWF did a great job the GFS honestly wasn’t all that bad until you got to the long range. Note the ensemble mean did way better. These errors from all 4 did really well even though the ECMWF did better overall no doubt about it.
Words of Caution:
So you are thinking let’s just look at the ECMWF all the time!!! Well not so fast even it did poorly in the really long range initially. Remember all those fake/bogus social media posts about the storm slamming into the east coast? Those were in many cases looking at one single model run 10 days out. Here’s a pro tip the models suck at the long range no matter the model. Below is a look at ensemble and operational model runs from 10 days ago for the ECMWF and AVN/GFS they somewhat agreed, and both were very wrong.
What about intensity?
So the ECMWF or European model did a great job with Irma right? Well only on the track, it absolutely sucked big time for the intensity of Irma. It was actually the worst model to look at for intensity forecasts. You were better served looking at other models like the GFS and hurricane models which were much, much better. Plus the human forecaster which even in the challenging intensity forecast did better than the models. See computers won’t take my job, well just yet! I mean the Euro did fantastic with the track but that intensity error especially at just 12 hrs was just horrible.
Moral of the model story:
All models are wrong but some are useful.
George Box
I Love that quote from the statistician George Box because he’s right! Some weather models are better than others and at certain things and at only at certain times. Like anything, in weather, there are no absolutes. You have to have the knowledge and experience to understand that. While the ECMWF did great with Irma’s track, it was horrible for its strength. The blended model tolls combined with a good experienced human forecaster is always going to be the best way to get a quality forecast.
