Hacker News Reader

Comment by patrickgzill | Hacker News Reader

Not two weeks ago I posted this: === We were told back in the late 1980s and early 1990s that huge computer models were written, that ran on supercomputers, to model and predict climate change.

The fastest Cray in 1990 could do (I think) 60 GFlops when loaded with 32 processors, which is roughly the sustained GFlops as a high end i7.

So clearly it is not a case of "just need a faster supercomputer and we will have an accurate prediction".

Reality is, from what I can tell, is that we don't have an accurate mathematical model of all the pieces that go into global climate to run on a supercomputer, no matter how powerful. ===

No one challenged me and I got several upvotes and follow-on posts in agreement. link: http://news.ycombinator.com/item?id=4761364

Did something change in the meantime?

View on HN

I don't really think that the idea that something must be perfect to be useful is correct. Trying to find a perfectly accurate model is impossible, so you must qualify your meaning of the phrase. Mathematics has a whole subject named statistics which spends quite a lot of its time dealing with the proportional incorrectness of models. If you look at the graphs or figures in the report (not just the headline of +4 degrees celcius), you will notice that almost all of them have a degree of uncertainty; figures without uncertainty values are as good as useless.

Also, if you are going to use "No one challenged me and I got several upvotes and follow-on posts in agreement" as a pseudo appeal to authority, I don't think that the readership of a Hacker News comments section really qualifies as authority on this particular issue.

Edit: Oh, apparently models have already predicted temperatures since 1900 correctly via hindcasting, along with changes in climates after significant events such as the eruption of Mount Pinatubo. And so far, they seem to be on-track with empirical data: http://www.skepticalscience.com/images/SLR_models_obs.gif, though I fully understand if you want to discard that graph due to it's source; the risk of selection bias is very high when taking a shallow look at a large and deep subject.

I am not using an "appeal to authority" argument, but I figure the HN readership would have pointed out my errors. (It's certainly happened before :-) )

We do not understand even the basics. For instance, we do not fully understand how the Gulf Stream works.

The problem is that we can't just sum or multiply the uncertainty values and try to arrive at a consensus.

Perhaps an analogy can be made to trying to model the US or Europe's financial system - clearly even the largest banks, spending large amounts of money on IT systems and modeling over the past number of years, have grossly mis-calculated things. And the financial world is arguably less dynamic than Earth's climate system.

http://www.climateprediction.net/

Reply on news.ycombinator.com