On the NOAA's Hurricane Predictions

Steven Milloy asks, What Hurricane Season?, in an article at FoxNews.com. According to Milloy:

On May 22, the National Oceanic and Atmospheric Administration (NOAA) issued its "2006 Atlantic Hurricane Outlook," forecasting an "80 percent chance of an above-normal hurricane season, a 15 percent chance of a near-normal season, and only a 5 percent chance of a below-normal season." NOAA called for "a very active 2006 season, with 13-16 named storms, 8-10 hurricanes, and 4-6 major hurricanes."

The NOAA adjusted the forecast downward in August but still predicted only a five percent chance of a below normal hurricane season. Milloy correctly observes that the hurricane season, which ended yesterday, was clearly below normal. But he goes further with the following questions and statement:

How can so many smart folks be so wrong? What's the lesson to be learned from NOAA's big whiff? Could it possibly be that predicting weather and climatic events isn't so easy?

...

Despite the vast collective expertise of NOAA scientists, immense quantities of atmospheric and oceanic data, and unprecedented computing power, NOAA failed miserably in predicting weather events a mere six months into the future -- and reiterated those same ill-conceived predictions at mid-season.

While it is clear that the NOAA or the media created the expectation of a severe hurricane season, Milloy is wrong to decide based on this one season that the NOAA "failed miserably." He reminds me of an Air Force fighter pilot that once challenged one of my predictions. The pilot was to fire a Sidewinder missile at a remotely controlled target aircraft on a test range. Before the test I provided him with launch parameters (position and speed relative to the target) within which he would have an estimated 0.9 probablity of destroying the target. After the test, in which he fired the missile within the launch parameters but failed to destroy the target, he declared rather strongly that my estimated kill probability was wrong. I told him that it very likely was, but that he had no way of knowing that. I pointed out to him that I did estimate a one in ten chance that he wouldn't destroy the target. (I don't think he was convinced. Some fighter pilots can't accept the possibility of losing when given a 90 percent chance of winning.) If the pilot repeated the test firing with the same lauch paramters 100 times and got only 50 kills (for example), then he would have a strong case that my estimate was wrong.

The NOAA did predict a one in twenty chance that the 2006 hurricane season would be below normal. So there is no way that Milloy could know that their predictions are wrong unless he could repeat this season enough times to show that the 5 percent chance of a below normal season should have been much higher. But he obviously can't do that.

However, I do agree with Milloy that we shouldn't rely too heavily on mathematical models to predict something as complex as weather patterns or phenomena, especially not for more than a few days into the future. My experience with modeling systems much less complex than weather convinced me that models are useful for studying trends but aren't very good at predicting specific outcomes. It is insanity to spend billions of dollars and depress the economies of countries around the world based on weather models projected decades into the future. If the models are that reliable then the NOAA should be able to more accurately tell us if it's going to rain this weekend.

Comments

Make a comment: