[ackydoodles]

That is because the weather is a chaotic system. It is governed by the mathematics of complex dynamics.

Predictions diminish in accuracy with increasing time because the number of inputs rapidly approaches infinity, and each one of them can turn out to be determinative of the overall behaviour of the system.

It should be no surprise to any scientifically literate person that a weather forecast issued in October has no validity by January. Any resemblance to reality is purely accounted for by chance.

This is not amenable to the Mr. Fix-It mentality that inevitably shows up in these discussions. There is good reason to believe that weather prediction will never improve, no matter what the scientific and technological developments, because the future of a chaotic system is mathematically intractable and genuinely unknowable.

The same argument applies to climate science. I'm sorry to bring you the very bad news that we live in a chaotic universe, not a deterministic one. Get over it. Wear a helmet.

Edit: The mysterious part is that the scientists at the so-called Climate Prediction Center would issue a 3-month forecast. What part of complex dynamics did they not understand? Sometimes an expert is just a guy with slides.

[sp332]

Prediction centers might not be able to tell the exact date and place of a storm that far out, but they can at least tell if a season is going to be bad well ahead of time.

Edit: It says this right in the article: climate forecasters focus on things that change more slowly, such as temperatures of the land and oceans... they try to say whether a given three-month period will be wetter, drier, hotter, or colder than average.

[ackydoodles]

I'm sorry, what part of complex dynamics did you not understand?

It doesn't matter how slowly things change. It doesn't matter how big the inputs are. Complex dynamics teaches us that scale does not matter; very small inputs can and do end up being more important than very large inputs.

I'll say it again: scale does not matter.

No, they cannot tell if a season is going bad well ahead of time. It is a fallacy to think that they can. They can tell that a season is going bad exactly at the moment the season starts to go bad.

Why is it that every time I raise the issue of the unknowable future of chaotic systems that people down-vote me? Is it so hard for you to accept? Check the math and the science: it is sound. I am sorry you do not like it, but the mathematics of chaos is not some fringe idea. It underlies everything. I'm astonished that so few people have noticed this.

[sp332]

Chaos theory does not mean that literally anything can happen. Weather changes are still bounded by energy inputs and outputs. It can't rain if the humidity in the air is too low. The jetstream isn't just going to reverse course anytime soon, although it is shifting slowly. It's physically impossible to have snow if the weather is too warm - complex dynamics can't change that. I'm not saying a forecast will never be wrong, but not every case is an edge case. With good data and a useful model, you can be right most of the time.

[upquark]

You are taking an extreme stance on the topic. There's room for modeling and forecasting even in highly complex dynamical systems, the models are getting better with advances in applied statistics research. They are not perfect, but they are not completely useless either.

For example, here's an article I just found on modeling stochastic nonlinear dynamics in ecological/oceanographic applications: http://arxiv.org/pdf/1211.1717.pdf

As for caterpillars predicting climate patterns, I agree with you that it's unlikely, unless there are some simple environmental indicators the caterpillars learnt/evolved to take into account but we humans haven't paid attention to yet. This doesn't mean the changes are unknowable.

[kbutler]

> they can at least tell if a season is going to be bad well ahead of time.

Proof needed.

e.g., the UK Met office predicted this would be a dry winter...

[sp332]

Every season the Heidke skill score is high, they did well. Here's an example http://www.cpc.ncep.noaa.gov/products/precip/CWlink/ghazards...

[ackydoodles]

A ridiculous measure, because it depends upon the definition of "correct", which is an arbitrary reduction of continuous data to a single axis, and because the scoring is conducted by those who have a vested interest in the outcome, and that's just for starters.

How about applying some statistics, like the probability that a correct forecast (whatever that is) could be accounted for by chance, given the population of forecasts and conditions in which it takes place? P-values have their limitations, but that's no reason to discard them entirely.