What the Hurricane?

Animated GIF shows GFS day 9 forecast flop
The moment we had been waiting for all summer finally arrived last week. The single model run to throw the social media world into a frenzy: the day 9 ‘dream’ hurricane from the GFS.

12z GFS August 19, 2014. Source: WeatherBell Analytics

It was amazingly viral for a landfall that was never to be. It was quite possibly the best non-storm setup we had seen in years, and, it was set to make landfall in New Orleans on the 9 year anniversary of Katrina. Of course, this was ridiculous and in the hours and days that followed many came out and said as much. I’ll resist the urge to continue a rant here about media and hype around storms though -- we’ve done that already.

The problem with the frenzy that erupted last week is that models have virtually no skill at nine days out in forecasting development, strength, and track of a system like this. At the time the system was only a tropical easterly wave east of the Caribbean islands. The amount of hype surrounding this system prompted the National Hurricane Center to remind us all to be cool:

The bottom line really is: be alert, be prepared, but also be wary of long-range projections that go beyond what the science can offer. And make the NOAA National Hurricane Center www.hurricanes.gov your calm, clear, and trusted source for official forecast and warning information on tropical cyclones.

Rather than call out the worst offenders here, I’ll just say this: It’s irresponsible, cut it out. Folks living along the Gulf and Atlantic coasts are understandably on edge this time of year, so offering bogus warnings to push your agenda is a really gross way to get followers, views, and clicks. And for those sharing these reports -- please pay closer attention to what sources you are using, and make some effort to find a reliable, trustworthy one.

As forecasters, we should use this type of run to help better understand the pattern, and see what models are trying to sniff out. Why we’re compelled to share images of hurricanes making landfall over a week out is baffling -- especially on Twitter where there aren’t enough characters to explain that it’s a) just a single model run, b) there is a GREAT amount of uncertainty, c) we’re “gearing this to our meteorologist friends that understand these uncertainties” (we’ve all already seen it), d) it’s not in line with its ensembles or other models like ECMWF, and d) it’s the GFS.

Since it’s naive to think we can all agree to stop sharing 10 day forecast maps from a single model… maybe we can agree to try a few things when doing so:

So, how on-target was this pot-stirring model run on 12z Wed, August 19?
The forecast was for 12z this coming Thursday, August 28 (pictured above). By the time Thursday rolls around, now hurricane Cristobal is expected to be far out to sea in the Atlantic (though we may have another disturbance to contend with in the Gulf). So how well did it do? Wrong body of water, no U.S. landfall, and about 1300 miles from where it was forecast to end up.

Below is an animated GIF which shows the forecast progression for 12z August 28, 2014. The initial frame shows what kicked things off last week. Projected landfall: southeast New Orleans.

The consecutive images show all 00z and 12z model runs since then, all forecasts for the same time on August 28, 2014. The first three runs kept the system in the Gulf on the 28th, but even here we see fluctuations of several hundred miles run to run. The following several days the GFS was ushering the system far out into the Atlantic by the 28th, before again trending back west toward Florida. If nothing else, this should help demonstrate the type of confidence any single model run should be given so many days out.

Finally we see some consistency from one run to the next when a center is found by hurricane hunters on August 23. The track at this point shifts back east a little, away from Florida and the eastern United States. Fish storm.

WeatherBell Analytics

Brendan Heberton

Brendan is founder of Weather5280. He is co-founder of FreshyMap, and develops software for geospatial data analysis and visualization.

Denver, Colorado
You can unsubscribe at anytime.