Why the GFS Model is so Crappy? Problem Finally Identifed?

[Mike Kovalchick]

I came across this blog discussion and thought it has some implications for chasers.

Most know through personal observations that the European Model outperforms the GFS model. (Of course-very few have access to the full suite of products from the ECMWF which is very frustrating.)

(Link to Model accuracy comparision GFS vs ECMWF http://www.emc.ncep.noaa.gov/gmb/STATS/html/aczhist.html )

Wouldn't it be nice if somebody would just "fix" the GFS so our chasing wishcasting would have more substance than just a dream ? Of course, in order to fix it, the problem with the model needs to be identified.

It looks like the problem has indeed been identified. If you input the ECMWF initialization scheme into the GFS Model..presto-the GFS actually spits out something useful! Unfortunately, the fix may not be that easy to implement.

Refer to this link:

http://www.accuweather.com/mt-news-.../2009/02/why_are_the_models_so_inaccurate.asp

 

[Adam Lucio]

Interesting read thanks for sharing. I tend to be a GFS guy myself. I dont look at other models that much unless its a chase related setup. It would be interesting to see the outcome if the two agencies were to work together though.

Although if a model could spit out a forecast with 100% accuracy, what would people need meteorologists for

 

[rdale]

There doesn't seem to be much backup for that claim however... I'd like to see an official link if it can be found - this is the same blog that compared today's "storm" in the northeast to the Superstorm of '93 saying the impacts would be almost as bad.

Oops.

 

[Mike Kovalchick]

There doesn't seem to be much backup for that claim however... I'd like to see an official link if it can be found - this is the same blog that compared today's "storm" in the northeast to the Superstorm of '93 saying the impacts would be almost as bad.

Oops.

It does refer to a recent AMS presentation on the subject:

http://www.emc.ncep.noaa.gov/gmb/wd23ja/presentations/dropout/AMS%20Oral%20Presentation%20Ballish.ppt

However, it's a bit too cryptic for me to follow. Perhaps, you can make sense of it?

Regarding the Superstorm..I think the European model did show a big storm moving up through Western New York at something like 970 mb in one of it's model runs so I assume that is where they came up with that prediction. Obviously, the European model isn't perfect either.

 

[rdale]

I think page 6 has the best summary... There were times when using ECM background (OVRLY) resulted in a BIG improvement in the GFS, but some of those events had no change at all. If the initialization was the only reason GFS loses to Euro, you'd think it would always make the model run better (or at least a majority) but that's not the case. Good work, but I don't get the impression that "we have found THE reason and THE fix"

 

[Dan Dawson]

Yeah, this is hardly a "fix", as a fix implies that something is broken (i.e., a software bug). The GFS simply uses a somewhat less sophisticated data assimilation system to provide initial conditions to the model, as compared with the ECMWF. There are many practical (as well as political) reasons why this is true. For one thing, 4DVAR, which is what the EC uses, is very computationally expensive. The GFS uses a less expensive but also less accurate 3DVAR scheme. The U.S. runs many other models other than the GFS and have to divide up computer time accordingly. The ECMWF, to my knowledge, focuses almost exclusively on their medium range model.

Thus, there is a tradeoff between accuracy and computing time. The GFS, for all its faults, the argument goes, is not that much more inaccurate than the EC to justify going to 4DVAR for the initialization. Obviously this is a debatable point, and I would love to see the U.S. adopt the 4DVAR (or EnKF, which is another attractive initialization and data assimilation scheme) for many of its operational model runs.