Computer Model Guidance Information (may want to sticky)

[wxman57]

After seeing the various model runs posted online for years, I've decided to do a little write-up on the various models to help those of you with little knowledge of what you're looking at understand them just a little.

Most of the "tropical" models generally posted on this and other forums are pretty worthless for hurricane forecasting. I've compiled a little list of what some of the models are and which ones merit a closer look. If any of you have questions about some other models, I, or any of the other Pro Mets here can try to answer them. And if anyone has some other models or hints/tips about using any of the models, chime on in!

XTRAP -- simple extrapolated track. Assumes the storm just keeps going at the same speed and direction. No physics at all involved. Generally of no use except as a "what if" scenario.

BAMD -- deep layer Beta Advection Model. Good for strong storms in the lower tropics (south of 20N latitude where steering currents remain constant). It doesn't emply much, if any, physics to predict changes in steering currents. Because of that, it shoudl not be used where fronts/trofs highs/lows move on by (north of 20N latitude. Never use it in the Gulf or western Atlantic.

NHC -- The official NHC forecast.

BAMM -- mid layer Beta Advection Model. For weaker to moderate-strength storms in the deep tropics. Same shortcomings as the BAMD model above. Don't use it north of 20N.

NHCA98E -- Climatology-based "model". Not really a model, just compares the current storm position to previous storms and "guesses" where it might go. Completely useless everywhere.

CLIPER -- CLImatology and PERsistence. No physics, not a real model. It's used by forecasters to determine if their forecasts have any skill at all. If you can't beat climatology and persistence, then you're not a very good forecaster.

LBAR -- Limited BARotropic. Limited alright, limited in its ability to forecast tropical cyclones. Generally ignore it.

UKMET -- Generally not a bad model. It's put out by the United Kingdom met office. It's worth a look at, but tropical cyclones aren't its forte'. It is run twice daily at 12Z (7am CDT) and 00Z (7pm CDT)

GFDL -- Geophysical Fluid Dynamics Laboratory model. This model was designed for hurricanes. It's based upon the latest AVN (American GFS) model. Because of that, though, it may tend to erode ridges too quickly north of stronger cyclones, leading to a right of track bias toward early recurvature. The GFDL did very well with Katrina, better than most other models, in fact. I take a hard look at the GFDL when making a forecast, but it has been all over the place with its forecast for Rita so far.

NOGAPS -- US Navy NOGAPS model. Designed for marine use, mostly. Sometimes does OK with tropical cyclones. Certainly worth a look. Run only twice daily so it's not available as often as other models.

CMC -- Canadian Model. I've never been too fond of this model. It may be good for winter up in Canada, but it's never shown me any expertise with tropical systems. If it gets a storm right, it's just usually pure luck:
http://weatheroffice.ec.gc.ca/model_for ... bal_e.html

NAM/ETA -- The new North American Model (the ETA was renamed recently). One of the worst models out there for the tropics, and just about everything else. Recent "improvements" have really made this model pretty bad for use down south. Ignore it laugh at it, but don't use it to predict the path of a tropical cyclone.

So, of all the models above, the GFDL is one of the better models. But, like I said, it's been all over the place with the Rita track so far and I don't trust it much yet.

I also like to look at the American GFS model for comparison:
http://www.nco.ncep.noaa.gov/pmb/nwprod ... arib.shtml

The GFS model certainly has its shortcomings, but if one knows its strengths/weaknesses, it can be a very good tool. Don't just trust it as-is unless you know what you're doing.

The European model (ECMWF) is a good tool. It's not shown a very good track record with recent storms, however. Until today, it had Rita moving into the BoC or southern Mexico. Here's a link to a site I like to use for the 7-day ECMWF:
http://weather.cod.edu/forecast/loop.ecmwf850t.html

MM5 -- Florida State's MM5 model is definitely worth a look. It's a "home brew" model tweaked by FSU mets. Doesn't have too much of a track record, though:
http://moe.met.fsu.edu/mm5/

Sometimes, the best models are what we call "ensemble" or "consensus" models. Simple "consensus" models just take 2-3 different models and simply average the positions for each time step out to 3-5 days to come up with a "consensus track". Occasionally, this consensus is better than any one model. Some examples of consensus models:

GUNA - GFDL, UKMET, NOGAPS, AVN (GFS)
GUNS - GFDL, UKMET, NOGAPS

More complex "ensemble" models actually analyze a number of models, account for each model's individual weaknesses, combine/average the tracks, then produce a single forecast track. You've probably heard of the Florida Superensemble model. This model has done VERY well with many tropical cyclones. Unfortunately, it's not available to the general public.

Here's a link to a good model page. It has a variety of models, from BAMs to various iterations of the GFDL model, to the GFS model, and some consensus model. Just weed out the "trash" models when looking at it:

http://euler.atmos.colostate.edu/~vigh/guidance/

Click on "Frame 1" of the early model guidance for Rita forecasts.

Oh, and finally, an online friend of mine, Gary Gray, puts out a very good daily analysis of each tropical cyclone. I HIGHLY recommend reading his detailed thoughts on each storm:

http://www.millenniumweather.com/tropical/discuss.html

[smashmode]

Great post with great info..should be stickied!

[skysummit]

Thank You wxman! That was an excellent and informative post.

[skysummit]

Thank You wxman! That was an excellent and informative post.

[mvtrucking]

Very helpful. Thank you Wxman57!

[jes]

Thanks so much. I think I'll print it in case they don't save it on storm2k

[T'Bonz]

Thanks. Very informative. I've saved it off-boards for future reference.

[karenfromheaven]

I'm interested in your thoughts about the GFDN model.

[dixiebreeze]

Excellent! Mods, please Sticky this post.

[wxman57]

I'm interested in your thoughts about the GFDN model.

The GFDN is the US Navy's version of the GFDL. It's initialized off the NOGAPS vs. the GFDL's initialization off the GFS model. So it tends to follow the NOGAPS. A few days ago, the NOGAPS was taking Rita WSW-SW to the BoC and the GFDN followed. I don't think it's a bad model, but I haven't seen it prove itself in recent times.

[Thunder44]

I'm interested in your thoughts about the GFDN model.

The GFDN is the US Navy's version of the GFDL. It's initialized off the NOGAPS vs. the GFDL's initialization off the GFS model. So it tends to follow the NOGAPS. A few days ago, the NOGAPS was taking Rita WSW-SW to the BoC and the GFDN followed. I don't think it's a bad model, but I haven't seen it prove itself in recent times.

Is the GFDN available somewhere for the public to view?

[karenfromheaven]

I'm interested in your thoughts about the GFDN model.

The GFDN is the US Navy's version of the GFDL. It's initialized off the NOGAPS vs. the GFDL's initialization off the GFS model. So it tends to follow the NOGAPS. A few days ago, the NOGAPS was taking Rita WSW-SW to the BoC and the GFDN followed. I don't think it's a bad model, but I haven't seen it prove itself in recent times.

Is the GFDN available somewhere for the public to view?

You can find it in the plots at the excellent Colorate State link wxman57 posted above. A big thank you to PhD graduate student Jonathan Vigh for his hard work!

[dougjp]

Yes please sticky this, and many thanks for this post.