Bryan Norcross is a Hurricane Specialist at WPLG-TV, Miami, and the Weather Channel.
Hurricane forecasters and research scientists are justifiably proud of the decreasing errors in storm forecasts. On average, the forecasts produced by the National Hurricane Center predicting where the center of a storm is going to be one to five days in the future are getting better with time.
Massive increases in the capacity of modern computers — the number of calculations they can do every second — are largely responsible for the improved forecasts. More calculations translate into higher-resolution forecasts, meaning small-but-sometimes-important features in the atmosphere are less likely to be missed. This trend is likely to continue, at least for a while.
While the decreasing average errors in hurricane forecasts are a good news story, I contend that we’re looking at the wrong metric. The real question is: How often does the forecast go drastically wrong?
The scariest hurricane scenario is not Irma or Maria. We knew those storms were going to be bad well before they hit. There was plenty of time to prepare. The real scariest scenario is a mega-storm threat that rapidly develops in the day or two before it slams the coast.
Very few people could gin up hyper concern — and therefore go through the slog of hurricane preps — with a tropical storm a few days offshore. And complex coastal areas cannot properly prepare for a devastating storm if the process starts a day or two before landfall.
The decreasing average forecast error implies that sometimes the errors are larger and sometimes they are smaller than average, of course. The good news is that we know the types of storm scenarios that are often forecast more poorly than average.
Storms that are weak, disorganized, and/or slow-moving are prone to much larger forecast errors than well-developed hurricanes that are moving in an established atmospheric flow.
What would happen if a Category 4 or 5 storm hit while coastal residents were in the process of evacuating? It’s an unnerving thing to consider, but the scenario is far from unimaginable.
In 1992, four days before Category 5 Hurricane Andrew crashed into the Miami suburbs, the disturbance had dissipated. Hurricane hunters could not find a circulation. Three days before it hit, Andrew was a 50-mph tropical storm. Two days out it became a 75-mph hurricane. No one imagined it would hit as a Category 5 the next night.
Modern science and technology would not save us if Hurricane Andrew happened again. Forecasts would hopefully be a bit better, and we might have hints that the storm might rapidly intensify, but a strengthening Category 5 at landfall would still be a surprise.
A more frightening Category 5 scenario played out in the middle Florida Keys in 1935. Two days before the strongest hurricane in the U.S. record book hit land, it was a 60-mph tropical storm. One day out it was hurricane with 105 to 115 mph winds. Nobody thinks that an evacuation could be completed in a repetition of this scenario. A monstrous surge of water plowed across the Middle Keys from the Atlantic to the south and Florida Bay to the north. Sustained winds are estimated to have reached 185 mph.
The point is, two of the three most intense hurricanes known to have made landfall in the United States fit the category of “difficult to forecast.” Forecasts for replays of Andrew or the 1935 storm would likely be prone to forecast errors greater, perhaps significantly greater, than the average errors we are so pleased are decreasing.
The problem is exacerbated by the cone — the ubiquitous graphic we use to convey where the worst of the hurricane is most likely to go. The size of the cone is the same for all hurricanes — the width is statistically based on the National Hurricane Center forecasts over the last five years so that about two-thirds of the time the storm will track within the cone.
But, as we’ve seen, disorganized systems are more likely to be the outliers with larger-than-normal forecast errors, so they are more likely to track outside the confines of the cone.
Imagine a 1935 or 1992 scenario where the fast-developing, super-intense hurricane wanders outside the cone and slams a community that thought it was out of the danger zone just a day or two earlier. Outlier scenarios are what we need to worry about.
In 2017, Hurricane Harvey was another kind of outlier. When it reorganized in the Gulf of Mexico, it was forecast to come ashore as a strong tropical storm, just over 48 hours before it hit as a Category 4 hurricane. No computer model initially forecast the spectacular strengthening, and, therefore, neither did the NHC forecast. Thankfully, the storm hit a lightly populated part of the Texas coast, so people had time to prepare when the forecast properly called for a strong storm about a day-and-a-half before landfall.
So while it’s a great scientific accomplishment that the average errors continue to trend down, and it’s also true that very-well-forecast hurricanes like Maria can cause cataclysmic damage, the scariest scenarios involve the outliers. As we saw in 2017, forecasts can be drastically wrong because modern meteorological science still has significant limitations.
Fortunately, we can anticipate the possibility of problem storms. When the system is disorganized or drifting around, beware. The future intensity may well be un-forecastable, and there’s a decent chance the cone based on average errors is not going to tell the story.