Hurricane forecasting: musings on accuracy, science and politics

Those of us who watched Atlantic hurricanes closely in 2015 saw a number of spaghetti models, predicting the course of the storm, that looked like this:

Weather Underground November 2016

Clear as mud where that thing is headed, isn't it? In other words, there was little consensus among models about where this storm would head, how bad it would get, and who should prepare. I had selective remembered a number of models like this and guessed that I could write a post about how it was a tough year for models. To prepare for writing this, I first went and studied predictions versus outcome.

As it turns out, despite a few spaghetti models that looked like a four year old's scrawls, models were no less accurate this year than in the past four years. (At least not as far as I can determine by studying the charts. It will take another year to hear a detailed number-crunchy analysis from NOAA.)

Every decade, the modeling of hurricanes improves, but we don't have 100% accuracy at even 24 hours out. 120 hours out? We're not anywhere close to predicting that. It might be wise to check out your storm supplies at 120 hours from a predicted landfall in your region, just in case you need to buy more sandbags, plywood, or gas for your generator, but there's no reason to panic or evacuate that far ahead of a predicted landfall. The storm could come ashore 500 miles from where it was first predicted, or it could turn back into mid-ocean, or it could dissipate altogether in five days (as did Invest 93L, the storm in the graphic above.)

Forecasting is not perfect yet. It may never be. Many variables determine the course and strength of a hurricane. Airborne dust, steering winds, sheer, humidity, ocean temperature, ocean depth, and terrain passed over are among the variables that can alter the course and intensity of a tropical storm. Unusual changes occasionally defy the most rational predictions.

The European model (usually labeled ECMWF on spaghetti plots) continues to outperform any other hurricane forecast model for the coming 12 to 108 hours. The reasons for this are, you may be surprised to know, not primarily scientific ones but political. Knowledge about the physics of atmosphere is widely held and not proprietary. The ECMWF is funded better, and steadily, and more computer power is allocated to running that model. More detail can be entered, more calculations performed, and a more accurate model results.

There is nothing new in the influence of politics on weather safety. Weather forecasting has, in the US, always been subject to political whims of Congress who vote funding one year and take it away the next. In one horrible moment in the 1980's, barely averted, weather forecasting was going to be privatized. (Imagine a pre-Internet world where only people rich enough to afford a subscription service could receive the warning of a hurricane or tornado. That would have been the likely result of the proposal.)

The history of the US official weather services is not an entirely lovely one in any case. Arrogance, ego, greed, narrow-mindedness, and even nationalism have cost many lives of citizens over the years. Erik Larson recounts some of this history in his excellent book on the Galveston hurricane, Isaac's Storm. Caribbean Island forecasters understood hurricanes better than anyone else in the early 20th Century--which makes sense, as they saw more of them--but the US weather service refused to accept what those people knew and ignored their warnings of coming storms, refusing to pass along the information to its own citizens.

Other problems in the bureaucracy have caused deaths from tornadoes. The weather service kept local news outlets--who might have had access to on-the-ground storm spotters or purchased new technology (such as Doppler radar) more quickly than the government--from reporting on a coming tornado. It wasn't until 1950 that the official ban on using the word "tornado" in forecasts was lifted. It is rather difficult to hide in your storm cellar from a specific EF5 tornado bearing down on you when all you're being told is that "Bad weather is forecast this evening."

Like so much in politics, not dedicating sufficient, regular funds to modeling hurricanes in the US doesn't even make cold financial sense. Forecasting a hurricane's landfall wrong by 200 miles can cost millions in lost wages, unnecessary preparations, and a hit to tourism. The billions in property damage where a big storm does hit might be unavoidable no matter how precise the models, but loss of life could be averted by issuing evacuations based on better models. (One would think that even short-sighted politicians might understand that, apart from the moral considerations, dead people are notoriously bad at paying income taxes.)

Luckily for Gulf and Atlantic coast residents in the USA, the less short-sighted Europeans are sharing their superior forecasts and should continue to do so for many seasons to come.

Sources include:
Erik Larson, Isaac's Storm, 2000
Nancy Mathis, Storm Warning: The Story of a Killer Tornado, 2007
https://en.wikipedia.org/wiki/Tornado_warning#Early_history
https://www.washingtonpost.com/news/capital-weather-gang/wp/2015/10/06/what-the-european-model-win-over-the-american-model-means-for-weather-forecasting/
http://stormfacts.net/models.htm
http://www.ecmwf.int/en/about/news-centre/key-facts-and-figures