Stories tagged predictions

The National Oceanic and Atmospheric Administration (NOAA) has issued its Atlantic hurricane season forecast. The prediction -- "near-normal" -- fits pretty well with the forecasts from AccuWeather, Colorado State University, WSI Corporation, and the Weather Research Center. What does "near-normal" mean? NOAA is predicting 9 to 14 named storms, with 4 to 7 hurricanes, and 1 to 3 major storms (category 3 or higher on the Saffir-Simpson scale of storm intensity).

But storm prediction is a tricky business. More rainfall over West Africa, warmer sea surface temperatures, and reduced wind shear could encourage more storms. An El Nino pattern in the Pacific or cooler ocean temperatures could discourage them. NOAA will issue another prediction in August, just before the usual peak in the hurricane season.


So said Yogi Berra, and science is proving him right. It turns out that making accurate predictions entails more than taking current conditions and extrapolating them into the future. It’s a specialized sub-field of mathematics, with lots of rules to ensure that predictions have a reasonable chance of being accurate.

Scott Armstrong of the University of Pennsylvania and Kesten Green of Monash University, Australia, examined the forecasts recently made by the UN’s Intergovernmental Panel on Climate Change.

Armstrong and Green rated the methodology used by the panel against 89 principles of good forecasting derived from years of research. They found that the panel report breached 72 of those principles. They concluded that the forecasts the weather was likely to change in many negative ways were worthless.

Now, this doesn’t mean that the Earth isn’t getting warmer—that’s a well-established fact. But, most of the predictions we’ve heard of climate catastrophe due to this warming are based on bad math.

(Freeman Dyson, physics professor at Princeton, makes a similar point in this video. He argues there’s been too much focus on building computer models of climate, and not enough emphasis on collecting actual data to see if the models hold up.)