Sunday, March 7, 2010

Climate Prediction Accuracy

It is the way of the world. The law of large numbers meets Heisenberg's Uncertainty Principle. Small samples are hard to predict statistically, and large samples are easier. The more exactly you try to predict something, the harder it gets. The farther ahead you try to predict, the less accurate you become. Measuring, or producing something will ALWAYS have a tolerance built in. More accuracy or power means more expensive, more complicated, more time, more money.

If it's raining outside, I can accurately guess 99 out of 100 times that it will be raining in 5 minutes. OK, maybe not when it's on and off like right now in San Diego, but that's beside the point. 3 days ago when I checked it said there was a 30% chance of rain yesterday and today. It's been raining the whole time! The farther ahead we go, the more the law of unintended consequences takes root. Here in San Diego we're in the middle of a BUNCH of climates, near the ocean, have things like Santa Ana winds come in some times, and a lot of water dumping off the mountains to the North and East.

The point is that no number really means anything without context and accuracy being specified. So while I accept that there has been some warming of the climate, I also know it's happened before (geological evidence points to this) . When I read this physics blog posting about an increase in the predicted number of hurricanes, I became predictably annoyed. What does this really mean? Nothing. We have ZERO history prediction history. The blog even highlights that:

That could help to solve an important climate change puzzle but before greater reliance can be placed on Ehrlich's, it needs to show its colours by accurately forecasting the numbers of hurricanes in the next few years. Its predictions do not make for pleasant reading.

I'm calling for people to begin to post their historical accuracy with their predictions. Now that science and politics are so interrelated, we need quality control in science just like any other endeavor. I know hurricanes only happen yearly, but if your "model" is only historical, and you can't post how accurate you've been in the past, your just as good as the tarot card reader.

No comments: