ECMWF by far best model for tropics days 3-5 in 2006

[LarryWx]

Please refer to p. 25's table 4 at the following link, which shows 2006 model tropical Atl. verifications on average:

http://www.nhc.noaa.gov/verification/pd ... n_2006.pdf

Although not the best within 48 hours and actually next to last at 12 hours, ECMWF did improve from 12 to 48 and was the best by far at 72-120 hours:

1) ECMWF at 72 hours:
- Was < 50% of both the UKMET/CMC 72 hour errors
- Was < all other models' 72 hour error by at least 20%
- Was slightly better than CMC's 36 hour error!

2) ECMWF at 96 hours:
- Was only ~40% of UKMET/CMC 96 hour errors
- Was < ALL other models' 72 hour error!
- Was < CMC's 48 hour error!

3) ECMWF at 120 hours:
- CMC error 2.8 times ECMWF error at 120!!
- Was < all other models' 120 hour error by at least 29%
- Was < all other models' 96 hour error except GFDL
- Was < CMC/UKMET 72 hour error by 20-25%!!

So that last statement means that, on average, a 120 hour ECMWF run had averaged a 20-25% smaller error than the corresponding CMC/UKMET runs made a whopping 48 hours later!! Folks, that is incredible!

With numbers like this for 72-120 hours, the several hour delay in getting ECMWF guidance is trivial imho.

[Extremeweatherguy]

The ECMWF sure hasn't had a very good start this year. It messed up with Erin (too far south) and it has already messed up with Dean (trying to show him hitting Florida instead of the Caribbean when he was still in the central Atlantic). It might improve, but so far it is doing a bad job.

BTW..this is just one year (2006). What were the top models in '05, '04 and '03?

[LarryWx]

The ECMWF sure hasn't had a very good start this year. It messed up with Erin (too far south) and it has already messed up with Dean (trying to show him hitting Florida instead of the Caribbean when he was still in the central Atlantic). It might improve, but so far it is doing a bad job.

BTW..this is just one year (2006). What were the top models in '05, '04 and '03?

1) Since it was only a short-lived cyclone in the Gulf, Erin's projections never reached as far as the 3-5 day period, which was the ECMWF's strongest period in 2006 relative to other models.

2) Regarding Dean being shown to hit FL, that was more than five days out. In at least one case, that was ~ten days out. So, once again this has little to do with the 3-5 day period performance itself.

3) Regardless, I've seen no concrete stats to explicitly tell me that the ECMWF is doing poorly relative to other models in 2007 to this point. On the other hand, there is this NHC study with concrete stats for 2006. Keep in mind that the GFS had the east coast above FL hit by the future Dean several times way out in the future. It is like a crapshoot that far in advance.

4) I haven't seen a similar study for 2003-5. If I had, I'd certainly analyze them as well.

[flwxwatcher]

Larrywx.. Thanks for the info! It will be interesting to see what the 2007 Model verification brings. With both the GFS and EURO going through changes.

[LarryWx]

Larrywx.. Thanks for the info! It will be interesting to see what the 2007 Model verification brings. With both the GFS and EURO going through changes.

You're welcome. Yes, the changes could make a difference. Regardless, I feel justified in already being able to call the Euro the 3-5 day forecasting champ for Dean provided there are no big surprises over the next couple of days. It clearly killed the wild GFDL imho for the 3-5 day runs reaching into the Gulf.

[flwxwatcher]

Yes, the GFDL showed its right bias with Dean and was a disapointment.

[Typhoon Hunter]

ECMWF is incredible.

It called Xangsane's landfall in central Vietnam perfectly last year. AND it called 240hrs out the exact landfall point of typhoon Sepat on Taiwan. Amazing!