Hurricane Prediction Accuracy

[Pangloss]

According to this article from AP, inaccuracy in hurricane path prediction has been cut in *half* since 1998, due to better supercomputers (and presumably their programming).

 

http://www.newsmax.com/archives/articles/2004/9/16/84808.shtml

 

Interesting stuff.

[Pangloss]

I'm sorry, would this have been better placed in the news forum? I'm still trying to get the hang of the place.

[Dave]

It's as good a place as any, has stuff to do with computers.

 

What I'd say is that a lot of the time, the algorithms they use can be somewhat flawed and to predict the weather more accurately you do need more complex algorithms (and hence more powerful supercomputers).

 

They still can't get the predictions right over here though

[drz]

I was arguing with a creationist once upon a time, trying to defend science. Anyhow, she stopped me in my tracks (years ago) by mentioning "How can you trust science's claims on evolution, when, they can't even predict the weather from one day to the next?"

 

And I hate to say it, but she is right. Weather prediction sucks. Seems like they were better at it 10 or 15 years ago. They have like, 30 possible tracks for the hurricane to go, even though when Ivan first formed I was guesinng land fall in texas, New Orleans or Alabama, and was pretty close considering I didn't use the super computers.

 

But I also think the news likes to tell people what they want to hear. When we have a nice, moderate temp day, the weather reporter trys to act like he created this wonderfull day for us.

[Dave]

Quite. Weather prediction is an extraordinary complex beast, and the algorithms behind it (and the supercomputers) are stupidly complex. The trouble is, these are always based on looking at what has happened to similar weather patterns in the past, and a lot of the time very small factors can have extremely large effects - the weather patterns may look almost identical to another earlier pattern, but there are extremely minute differences in some variables which throws the entire thing.

[Pangloss]

Interesting points.

 

One of the little (tiny, eentsy, fleeting, miniscule) bits of science I caught during the Month o' Hurricanes here in Florida was when one of the weather guys mentioned that the way they pick the "official predicted path" is by looking at all the different models and seeing which one has most accurately predicted the storm's path so far, over the last X number of days.

 

I can certainly understand that approach, but I couldn't help but chuckle at the pessimism of doing it that way, essentially saying that they're not picking the most accurate one, but rather the least inaccurate one (so far), and to heck with all that wind data! (hehe)

[Cap'n Refsmmat]

[Dave]

I guess we'll soon see if they get it right.

[Pangloss]

I don't know if it's just because of current events, but I've been wondering lately about the similarity between hurricane path prediction and the problems with voter opinion polls (Gallup, Rassmusen, etc). In a sense the problems are similar, in that they both attempt to model essentially unmodelable problems. Both are dependent on essentially random events. Both make assumptions that those events will not, in fact, be random. The pollsters assume that voters will answer the poll accurately (why wouldn't they?), and forecasters assume that the winds will continue and not change their behavior (why would they?).

 

But of course even the most die-hard ideologue is a free individual who might tell a pollster whatever he or she chooses, for whatever reasons they might have at that moment. In the case of hurricanes there's no free will involved, the problem is more one of being able to model in sufficient detail. You never know what you might be missing, such as a localized temperature change in an unmonitored area.