3 ms·

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. Th

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.

guscost 14y ago
Bingo.
oscilloscope 14y ago
We don't even have a particularly good data model. If you're interested in the water cycle, you'll have to construct that data from dozens of agencies in hundreds of governments. If the data was collected, linked together and published in an accessible form, more people could experiment with statistics, machine learning and visualization. Right now the data is in labyrinthine web forms, SOAP services, XLS files, GIS systems, etc. The barrier to entry needs to be lower.
macavity23 14y ago
Agreed. And even then, the system you're dealing with is so 'chaotic' (i.e. massive effects can be caused by very small changes in data) that you get diminishing returns the more data you have. Take the weather forecast as a (very) slimmed-down version of this problem. Yes, weather forecasts are better now than they were 20 years ago, but how much better? In the UK now, the 24h forecast is pretty accurate most of the time, but the three-day is very iffy, and a week out is still a crap shoot. That's despite a metric fuckton of satellite data and some serious hardware (the UK met office numbercruncher consumes more than a megawatt!). We're always going to be making guesses here. It seems, to my non-climatologist brain anyway, that the most up-to-date models produce predictions that are fairly consistent with the real world data we have. I would say the political and economic problems are much more important, and harder to solve.

guscost 14y ago

Bingo.

oscilloscope 14y ago

We don't even have a particularly good data model. If you're interested in the water cycle, you'll have to construct that data from dozens of agencies in hundreds of governments. If the data was collected, linked together and published in an accessible form, more people could experiment with statistics, machine learning and visualization. Right now the data is in labyrinthine web forms, SOAP services, XLS files, GIS systems, etc. The barrier to entry needs to be lower.

macavity23 14y ago

Agreed. And even then, the system you're dealing with is so 'chaotic' (i.e. massive effects can be caused by very small changes in data) that you get diminishing returns the more data you have. Take the weather forecast as a (very) slimmed-down version of this problem. Yes, weather forecasts are better now than they were 20 years ago, but how much better? In the UK now, the 24h forecast is pretty accurate most of the time, but the three-day is very iffy, and a week out is still a crap shoot. That's despite a metric fuckton of satellite data and some serious hardware (the UK met office numbercruncher consumes more than a megawatt!). We're always going to be making guesses here. It seems, to my non-climatologist brain anyway, that the most up-to-date models produce predictions that are fairly consistent with the real world data we have. I would say the political and economic problems are much more important, and harder to solve.