Comment by luc4sdreyer | Hacker News Reader

Comment by luc4sdreyer | Hacker News Reader

Parent Full thread

> This seems very hard to believe - mainstream science struggles to predict accurately even 5 years out

In general, this might be true. Predicting the future is harder than it seems. But it depends completely on the type of system (chaotic or not), and how well we understand it. We can predict the position of the Earth thousands of years into the future.

Regarding climate models, you might be underestimating how good they were in the early 80's. Here are old models' projections compared with the actual temperature records up to 2017 [1]. I'm not an expert so I don't know if they excluded early bad models, but they included all the IPCC models.

The graph that AOC showed was pretty small and grainy. It's also compressed along the x-axis (140 years), and has a high aspect ratio. It's possible that she, or her aides, picked the best one for the presentation. But given the performance of the models at the time, it's not unreasonably good. It actually seems worse than the 1990 IPCC model, but that's just my opinion.

[1] https://www.carbonbrief.org/analysis-how-well-have-climate-m...

The models used today have all predicted more warming than actually happened, so I'm very skeptical models got progressively worse over the decades instead of better.

Also note it's impossible for models in the 80's to have been accurate because the temperature data for the past itself has been continuously adjusted by climatologists since then, so any models calculated back then would be using temperatures now considered wrong (and not by a small amount!). Either the old models are wrong or the old temperature records are wrong, but they can't both be true simultaneously.

> The models used today have all predicted more warming than actually happened

Do you have a good source for this assertion? I mean, there are still people claiming that there is no warming at all, so "all" models seems like too strong a claim.

Your information on is out of date. None of the models predicted the big lag in temperature increase in the oughts but since 2015 air temperatures have jumped up to right where the consensus predictions said they would be.

Yes, scientists have been busy trying to refine their models and understanding of temperatures. But my understanding is that, quantitatively, the inaccuracies in the old temperature data was pretty small compared year to year variation. No prediction is ever perfect, even in QED you have one part in a trillion error, but instead you look at whether the result is in the error bars and whether the error bars are narrow enough to be useful.

If the models didn't predict the lag, it stands to reason that the models weren't accurate, even if current temperatures align with them. That is to say that the current alignment is just coincidental.

If I remember correctly, the lag was likely caused by not correctly accounting for deep ocean heating. That's a huge missing piece of the simulation, so how could the models from back then possibly have been accurate?

Yes, our models were inaccurate for the oughts and early 2010s. There are tons of positive and negative feedback loops which our climate models are probably very badly calibrated for. But those models did have accurate predictions of what the temperature would be for 2019 so you have to give them credit for that.

But honestly it's the inaccuracy of current climate models that make me so concerned about global warming.

> If the models didn't predict the lag, it stands to reason that the models weren't accurate, even if current temperatures align with them. That is to say that the current alignment is just coincidental.

Or the lag was a minor factor that only briefly overrode the primary drivers of climate change, which have since reasserted themselves. I think we both know what the smart bet is.

The lag isn't a minor factor though. We are talking a decade plus of incorrect predictions. Newer models that have taken deep ocean warming into account are probably more accurate.

The oceans are now significantly warmer than earlier models predicted. There's no way that doesn't impact the outcome.

Actually newer models don't predict the lag / the pause at all.

Instead what happened is they decided the temperature records that showed a pause were wrong, went back and adjusted them so there's no longer any real pause if you look at the latest temperature datasets. The entire thing was written off as a giant measurement error.

They are climate models, not weather models. Buried somewhere in the IPCC reports is the sorts of time space they apply to. I vaguely remember 10 years or so. I do distinctly remember thinking during the pause if something dramatic didn't happen by 2016 they were in serious trouble, because that was at the end of the range.

> That is to say that the current alignment is just coincidental.

Sort of. The model is (heatIn - heatOut - seaAbsorption) = surface weather. They got the seaAbsorption wrong, something they later figured out with the Argo network as you noted. But that means they got the primary driver of global warming, heatOut dropping as CO2 increases, right. Which sadly means if you wait long enough, their predictions were always going be right because in the end heatIn - heatOut is all that matters.

> The models used today have all predicted more warming than actually happened

Source? Although all of this depends on what you mean by "more", as in what you assume the error bars on the models are.

> temperature data for the past itself has been continuously adjusted

Source?

> Also note it's impossible for models in the 80's to have been accurate

It's not actually. Models don't depend on the temperature. The models could have been (and in fact, were) accurate despite the alleged wrong inputs.

> Either the old models are wrong or the old temperature records are wrong, but they can't both be true simultaneously.

It doesn't really matter in the context of the legal argument about what Exxon thought to be true in 1982.

Reply on news.ycombinator.com