by Robert Henson
Ever wonder how the National Hurricane Center
(NHC) devised the "cone of uncertainty" that's now a fixture on the NHC Web site and TV weathercasts? I was delighted to find an article in the May issue of the Bulletin of the American Meteorological Society
devoted to the cone - or, as the authors dub it, the COU. Here's a 4 MB PDF
of the full article, entitled "Misinterpretations of the 'Cone of Uncertainty' in Florida during the 2004 Hurricane Season." The authors, led by Kenneth Broad
(University of Miami/Columbia University), worked closely with NOAA staff to present the closest thing to an insider's look at the COU. The article not only reveals a couple of things I'd long wondered about the cone, but it also includes some perceptive analysis of how a tool designed to convey the uncertainty in a forecast can actually have the opposite effect.
Jason Samenow's ForecastToday:
Partly sunny and not as hot with a 40% chance of late afternoon or evening thunderstorms. Highs in the mid to upper 80sTonight:
Partly cloudy and warm. Lows 70-75.Monday:
Hot and quite humid with a 20% chance of late afternoon storms. Highs 89-94.
For those who aren't familiar with it, the COU is the most prominent part of the standard forecast graphic produced every few hours by the NHC for each tropical depression, tropical storm, or hurricane in the Atlantic and northeast Pacific. Though it's already a fixture, the current COU design has been used operationally only since 2002. Here's an example for 2005's Hurricane Katrina (to the right -- courtesy National Hurricane Center), issued at 11 AM EDT on August 26, three days before landfall. The "skinny black line
" (which former NHC director Max Mayfield has urged us not to take as gospel) is the most likely track for the cyclone center, as projected by a human forecaster based on a wide range of model ensembles and other input. The uncertainty of the outlook is depicted by the white cone-shaped area surrounding the forecast track. Note that in the example from Katrina, New Orleans is already within the cone 72 hours before landfall, even though the forecast track lies well to the east. (Just six hours later, the forecast track was virtually on top of Katrina's final track - an amazing forecast feat.)
For a long time I'd assumed that the cone itself was tailored for each storm. Actually, as the BAMS article notes, it's based on the *average* error at each time step (12 hours, 24 hours, and so on) for *all* storms over the last few years (ten, according to BAMS; five, according to the NHC's online explanation of the graphic). What this means is that valuable data gets left on the cutting-room floor. In some cases, the various computer models used by NHC agree remarkably well
on a given cyclone's track. In other cases, the models diverge wildly
. If the COU could somehow be reworked to reflect the actual model-based spread for a given storm, instead of a generic uncertainty range, it would be more accurate and more useful.
Here's another little-known fact revealed in the BAMS article: the COU's width is set so that about two-thirds of all forecast errors lie within the cone. This implies that, for any given cone graphic, there's roughly a 33% chance that the cyclone center will pass outside the cone during the next five days. Of course, NHC could have set the margin of acceptable error at 40%, 20%, or any other value. It's essentially an arbitrary choice, and 33% is probably as good as any. You wouldn't want storms straying outside the cone 50% of the time, but a cone that corralled 95% of all storms might be large enough to induce heart attacks from Texas to Maine. In any event, the 33% value was something I hadn't yet learned from TV or newspaper coverage, and I suspect many people assume the percentage is lower.
That's a key point of the BAMS article. The way the cone truly works is often ill-explained in the media, and it's easy to miss the NHC Web site's explanation. As with a Rorschach test, people may see in the cone what they want to see, and it's easy for the intended purpose to get lost. The article points out some frightening ways in which the cone's meaning has been mangled by public survey-takers, writers, and broadcasters. This confusion has consequences, as when thousands of people in Charlotte County, Florida, failed to evacuate for Hurricane Charley in 2004. Although Charley's forecast track did shift from Tampa toward Charlotte County only in the last few hours before landfall, Mayfield has noted that the Charlotte Harbor area was in a warning for 24 hours and inside the cone of uncertainty for a full four days. It was that skinny, shifting line inside the cone that appears to have led residents astray - ironically, conveying the very certainty that the cone itself was meant to dispel.
As the BAMS article notes, the NHC's Web site recently invited its readers to comment on several variations of the cone for possible future use. One example showed the uncertainty for each forecast interval as a colored ring (as shown below), with the skinny black line omitted. The end result is a product that's more informative but less graphically appealing (to my eye, anyhow). The BAMS authors suggest a tiered approach, with a simplified graphic for mass consumption and more finely tailored products for specialized users such as emergency managers.Image courtesy National Hurricane Center
In the meantime, perhaps more people will grasp the true meaning of the cone given enough explanation from media and the NHC. After all, it took years for the public to get used to probabilistic precipitation forecasts after they debuted in the mid-1960s. We've certainly come a long way from the time when predicting a "30% chance of rain" sounded exotic.Meteorologist Robert Henson is a writer and editor at the University Corporation for Atmospheric Research, which operates the National Center for Atmospheric Research. He is the author of "The Rough Guide to Weather" (second edition, 2007) and "The Rough Guide to Climate Change" (2006). Henson is filling in for Andrew Freedman this week and next week.