Why can data models predict the future climate but not the weather?
Why are weather forecasts often wrong but modelling of the future climate highly accurate? I would presume a more complex system over a longer period would be harder to predict.
In: 0
The system is equally complex for both, but while a random, unpredictable event can completely change the weather tomorrow, over a longer time those events won’t matter as much anymore and the system is much easier to predict at this larger scale.
That’s just based on probability: If you roll 1 die, I can’t tell you what face it will show. But if you rolled 100 dice 600 times, I can tell you pretty accurately how many of each face you’ll see: about 10’000 each. This looks like a way more complex system, but it isn’t: it’s the original repeated 60’000 times. Could my prediction be wrong? Yes. Is it likely to be far off? No.
(Going beyond ELI5 here: Based on a Chernoff upper bound, the probability than any face shows up more often than 10500 times is less than 0.025%. At just 11000 of one face, you’re at a chance of less than 10^(-13)%.)
Of course the prediction isn’t accurate down to the single die anymore, but we’re talking about 60’000 dice rolls, nobody cares. Just like in climate models nobody cares whether it’s going to rain in Hinterschlumpfhausen, Germany on April 1, 2033 at 6 pm. Only the large scale predictions are actually important (are there 10’000 or 11’000 6s? / will it be on average 2.5° or 2° warmer?)
Latest Answers