Network News

X My Profile
View More Activity
The new Washington
Post Weather website
Jump to CWG's
Latest Full Forecast
Outside now? Radar, temps
and more: Weather Wall
Follow us on Twitter (@capitalweather) and become a fan on Facebook
Posted at 12:45 PM ET, 09/ 8/2008

Some Hurricane Tracks Trickier Than Others

By Steve Tracton

With Hurricane Gustav, the National Hurricane Center (NHC) set a high standard for forecasts this tropical season, thanks to a huge assist from the computer models. It's a standard that may be hard to live up to with Ike -- now weakened to a Category 2 hurricane as it batters Cuba, but likely to strengthen before a possible late-week or weekend U.S. landfall along the Gulf Coast -- and other storms that may arise.

Indeed, the forecasts for Hurricane Gustav were exceptionally good, as Jason noted last week in his post headlined, "Computer Models Nailed Gustav's Track." Why were the forecasts for Gustav so good? It turns out some storms are easier to predict than others.

Keep reading for more on the predictability of hurricane tracks. Also, see our full forecast through the weekend.

First, some brief background. As indicated by a recent policy statement of the American Meteorological Society, "Hurricane Forecasting in the United States," NHC forecast track errors during 2001-2005 averaged about 75 miles for 24-hour forecasts and 136 miles for 48‑hour forecasts --
approximately half what errors were in 1990. At the four- and five-day ranges, average errors are about 225 and 300 miles, respectively. Not surprisingly, the average error belies the fact there is substantial variability in accuracy from one storm to the next, and over the history of any given storm.

Average track errors, as derived from verification of previous years' forecasts, are reflected in the famed (some would say, infamous) "cone of uncertainty" (example below) that NHC uses to illustrate the uncertainty about the official forecast track. The line through the center of the cone represents the most likely track based largely on the "consensus" of computer models and experience of NHC's expert forecasters.


National Hurricane Center forecast track and cone of uncertainty for Gustav from Aug. 29 (left) and Aug. 31 (right). Courtesy NOAA.

In the case of Gustav, which made landfall along the central Louisiana coast the morning of Sept. 1, the NHC forecast was pretty close to having the location of landfall "nailed" as early as Thursday morning, Aug. 28, but occurring about 24 hours later than observed. Good, but good enough? The error in timing may or may not have affected New Orleans' execution of evacuation plans for Gustav. More generally, however, a 24-hour timing error in prediction of hurricane landfall could be critical in decision making and risk analysis.

Also, could New Orleans have counted on being spared a direct hit with a forecast (very accurate as it turned out) that had Gustav tracking about 75 miles to its west? Probably not, because the cone of uncertainty Thursday morning indicated that landfall could be anywhere from southeast Texas to the Florida Panhandle. Moreover, by design there is even a 30-40% chance of landfall occurring outside the cone. So while I would credit it as a good "heads-up" forecast, it was hardly a screaming message of confidence. That's not to say, however, that even a low-confidence forecast of a relatively small chance of a direct hit by a hurricane isn't cause for taking preparatory action.

Even the official forecasts less than one day from landfall indicated it could occur over a range of coastline not much smaller than the Thursday morning predictions. On the positive side, landfall was predicted to be in just about the same location in all forecasts subsequent to Thursday morning. To this extent, I'd agree the forecast was "nailed."

But now, for the rest of the story! To this there are two primary components.

First, there is very little skill in predicting the rapid changes of intensity in hurricanes, or how far hurricane force winds extend outward from the eye of a storm. Thus, even a perfect track forecast leaves as unknowns the extremely important and likely consequential aspects of wind strength and the geographical range of its effects, including the magnitude of the storm surge. But, this will be the subject of another post.

Second is the inherent predictability, or "forecastability" if you like, of the storm track. In this regard it turns out that Gustav was a relatively easy case.

As mentioned above, the predicted track -- center line within the cone of uncertainty -- usually represents the consensus of tracks provided by a set, or ensemble, of computer models. How similar or different the tracks predicted by the various models is another method of estimating the forecast uncertainty. This measure of uncertainty (one that is not taken into account by the cone) is specific to the particular storm and forecast cycle.


Model forecast tracks for Gustav from Aug. 25. Courtesy Colorado State University. Click here to enlarge.

For example, the figure to the right shows the predicted tracks for Gustav by a variety of models from the evening of Aug. 25. Why are these forecasts so different? Yeah, it's the usual suspect in such things, namely, the "butterfly" effect, where differences in approximations and assumptions between models, and/or the initial conditions fed into the models, can multiply as the models moves ahead in time. The result is that, within the set of model predictions, forecasts of storm track increasingly diverge with increasing forecast time. The rate and degree of divergence, or forecast "spread," varies from storm to storm, and from forecast cycle to forecast cycle for a given storm. The more the model predictions diverge, the larger the uncertainty, and vice-versa. (For the geeky and more absorbed by all this, see my presentation on uncertainties in weather forecasting.)

An ensemble of forecasts can also be generated from a single model by running the model many times, each time based on slightly different initial conditions. For example, output from the European Center for Medium Range Forecasts (ECMWF) ensemble model illustrates the range of predictability that is possible. In the example below of two different typhoons over the west Pacific, it's clear that one case (left) is essentially unpredictable -- the storm could go just about anywhere, while in the other case (right) there is much less divergence amongst outcomes, and therefore the storm is more "forecastable."


Model forecast tracks for Typhoon Kirogi in 2005 (left) and Typhoon Tokage in 2004 (right). Courtesy ECMWF.

In the case of Hurricane Katrina (below), a difference in just one day was the difference between high (left) and low (right) levels of uncertainty in the storm track.


Model forecast tracks for Hurricane Katrina from Aug. 25, 2005 (left) and Aug. 26, 2005 (right). Courtesy ECMWF.

Just as some mountains are easier to climb than others, and some golf putts are a sure bet while others are virtually impossible, the track of some storms is easier to predict than others -- that's just the way it is! Why are some cases more predictable than others? I can provide some reasonable hand waving explanations in some instances, but the basic science underlying this question is not yet adequately understood.

Getting back to Gustav, early in the storm's history there was considerable uncertainty, as shown by the second image in this post, as well as the ECMWF model ensemble (not shown). The latter's emphasis was more toward the Yucatan.


Model forecast tracks for Gustav from Aug. 29. Courtesy Colorado State University. Click here to enlarge.

It wasn't long, however, before both sets of ensembles converged on a remarkably narrow and focused target, as illustrated by the adjacent image. Conclusion: Gustav was an unusually easy case for predicting storm track! Hanna, on the other hand, proved not so easy -- even one day before reaching the Southeast U.S. coast, models had not converged on a landfall location (See model guidance for Hanna here, here and here.)

As for Ike, still several days away from a possible Gulf Coast landfall, the jury is still out as to how predictable of a storm it will turn out to be. More likely than not, though, the range of uncertainty associated with the storm's predicted track will encompass a range of possible impacts for residents and decision makers near the path of the storm. Just as state-of-the-art medicine very often can't provide a diagnosis with complete certainty (see here, p. 10-11), the science of weather forecasting (especially hurricanes) is often far from a sure thing.

By Steve Tracton  | September 8, 2008; 12:45 PM ET
Categories:  Tropical Weather  
Save & Share:  Send E-mail   Facebook   Twitter   Digg   Yahoo Buzz   Del.icio.us   StumbleUpon   Technorati   Google Buzz   Previous: Photography: Hanna's Sunset
Next: PM Update: Pleasant Today, Stormy Tomorrow

Comments

Steve, really like your post. I found it very educational! Thanks!!

Posted by: Camden, Capital Weather Gang | September 8, 2008 1:44 PM | Report abuse

One more follow on to Hanna windup: Plot of surface pressure and winds as Hanna passes over a buoy in upper Chesapeake Bay

See:
http://cwg.jbjonesjr.com/storage/Hanna_WindPress_trace.jpg

Posted by: Steve Tracton | September 8, 2008 4:33 PM | Report abuse

Whoa hold on a second... there is variability in forecasting storms? Even big ones? Wow, who knew? Actually Steve that was a great read. Thanks.

Posted by: John - Burke | September 8, 2008 4:34 PM | Report abuse

Where can I find a map of the tracks from the different computer models?

I find them much more informative than the generic cone of destruction that the NHC publishes.

Posted by: Michael | September 8, 2008 5:07 PM | Report abuse

Michael: Computer model tracks are available at this site: http://euler.atmos.colostate.edu/~vigh/guidance/

Posted by: Jason, Capital Weather Gang | September 8, 2008 5:28 PM | Report abuse

Ike is looking a little weird to me here in Tampa...also looking weird - what is that stuff in the far western gulf? Looks a little suspicious...or tropical perhaps?

I'm sure the models have better intuition that this MS program meteorology program dropout! I just wonder if Ike falls into the more "unpredictable" category?

Posted by: tampa edski | September 8, 2008 6:35 PM | Report abuse

Model ensembles indicate a much larger range of uncertainty (forecastability) than the above referenced ensemble of models (or NHC "cone").

Specifically the range of landfall possibilities extends eastward along the Gulf coast with a secondary maximum of strike probability through the Florida Panhandle. If IKE should actually hit this area, there will likely be much consternation in regard to the official forecast - not necessarily on what the NHC cone shows, but because most people probably miss the small print explaining that 30-40% of the time, the actual landfall is expected to fall OUTSIDE the cone.

Posted by: Steve Tracton | September 8, 2008 6:40 PM | Report abuse

Another illustration of the forecast tracks is at
http://tropics.hamweather.com/index.html .

Posted by: PaKo | September 8, 2008 7:05 PM | Report abuse

Wow, cool article, Steve! I always find it interesting to look at the model tracks and the actual tracks of storms after they have passes. The CWG has posted them sometimes, do you know a link?

Posted by: Sterling | September 8, 2008 8:41 PM | Report abuse

Sterling:
The forecast model tracks can be seen at the site mentioned above by Jason. A good site for the actual tracks is: http://www.stormpulse.com/

On the left toward the bottom click on appropriate case in season summary

Posted by: Steve Tracton | September 8, 2008 9:37 PM | Report abuse

The comments to this entry are closed.

 
 
RSS Feed
Subscribe to The Post

© 2012 The Washington Post Company