Comment by rjdagost | Hacker News Reader

Comment by rjdagost | Hacker News Reader

Parent Full thread

For a little while I was interested in paleoclimatology and I started studying how researchers developed these historical temperature reconstruction curves. Basically you take “proxies” for temperature, things like tree rings, coral growth, sediment layer thickness, etc. and you posit a relationship between the proxy measurements and temperature. Measurements that better fit the hypothesized model get more weight in the temperature reconstruction.

The main problem is that no one knows if the hypothesized models (temperature vs. tree ring thickness, sediment layer thickness, etc.) have any validity. Why should tree rings grow linearly with temperature- what about a huge number of other variables? The temperature “signal” is at best very weak. Decoupling the temperature signal from other variables (humidity, elevation, tree species, CO2 concentration, etc.) is a hugely difficult multivariate statistics problem and most of the researchers I've corresponded with do not seem to appreciate this fact.

Another major problem is that proxies get weighted in the temperature reconstruction based on how well they fit the hypothesized temperature model. The weighting is where things get really shaky. Right away, you have biased your results in such a way that that nonconforming data is discarded / de-weighted. Note that there is not anything wrong with these nonconforming samples that invalidates them; they just don't fit the proposed models very well. Typically researchers develop their own methods for weighting samples in the temperature reconstructions and many of these methods would make a statistician cringe. Often during the weighting phase you find that a handful of proxy samples (out of a sample size of hundreds) get the vast majority of the weight in the temperature reconstructions. Sometimes the weight of a temperature reconstruction in a location is almost completely computed from a few proxy samples taken halfway around the world at a completely different elevation.

If you try to discuss these fundamental problems with climate scientists you'll typically get told that you're not a scientist (not true) or that you're a "denialist" being paid by big oil (not true). I am convinced that these temperature reconstructions are useless at best.

I worked in a dendrochronology/dendroclimatology lab[1] for several years and can assure you that the researchers involved are well aware of the issue of multiple influences on tree ring growth. There are a couple of ways that they go about dealing with it. NB I was more involved in precipitation reconstructions than temperature reconstructions, but I know enough about the process to discuss it here.

The first is experimental design, much like is necessary in any aspect of science: Annual tree growth is definitely a response to a variety of regional, site-specific, and tree-specific factors for any given species in a location. So in order to increase the 'signal to noise' ratio for a given signal (temperature, precip, fire, disease or whatever reconstructions), you look for combinations of species and sites for which the factor you want to study is the limiting growth factor; in other words, other things the tree needs to grow are present in abundance and one is in shortage. If you want to make a precipitation reconstruction, you find a site that has very little soil moisture storage, such as trees basically growing out of fractures in steep bedrock. If you want to study temperature, you pick species like Bristlecone Pine that grow at high elevations (which are typically relatively wet), so that the temperature/length of growing season is the annual limiting growth factor. That sort of thing.

The second is that researchers use many other forms of measurements than simply ring width to tease these problems out. For example, many people use oxygen isotopic ratios[2], latewood (the dark outer part of the tree ring) density measurements[3], and other measurements in tree rings, both because of the unique and accessible response of the tree, and because (unlike, say, sediment cores) the tree rings are much more quickly, accurately and reliably dated at the single-year precision.

Third, there is a lot of statistical analysis thrown at the problem. I didn't work a lot with this, or see people in my lab do it, but I am aware of researchers doing multivariate statistics techniques such as principal component analysis[4]. I'm sure there is a lot more, but you can do a Scholar search as easily as I can.

Fourth, one of the strengths of the multiproxy technique is that many of them are independent, and they all show a lot of correlation to the temperature records during the instrumental calibration period. There is some divergence before this period, but not as much as would be expected otherwise. There is also the problem of weighting, like you mention, and I don't think anyone is comfortable with it but I don't think it's as ad-hoc as you make it out to be. A review (by a statistician) that discusses this is here[5]; I didn't read it in depth.

[1] http://www.uark.edu/misc/dendro/

[2] http://link.springer.com/article/10.1007/s00382-013-1674-3

[3] http://journals.ametsoc.org/doi/abs/10.1175/JCLI-D-11-00139....

[4] http://onlinelibrary.wiley.com/doi/10.1111/j.1600-0706.2010.... [5] http://www.stat.unc.edu/postscript/rs/Sensitivities-Submitte...

You know, nearly everything is a "proxy". Kitchen thermometers use bending metal as a "proxy" for temperature. All that matters is accuracy and precision. This is how science works, at its base. There's no such thing as "sense data".

True, the signals from tree rings (etc.) are noisier than those from kitchen thermometers. Let's consider, for a moment, how you could verify the precision and come to a conclusion as to the accuracy of those signals: well, when you get the same information coming from multiple seemingly independent sources, it greatly increases your confidence in that information. Hence, if you're doing your modeling right, if you get tree ring data from all over the world, and soil samples from all over the world, and find that, sure enough, there is a common signal visible across all of them, then you say "well, the only thing that can explain this global correlation is the modulation of some global parameters". And you present your data and conclusions to a bunch of people whose job it is to poke holes in it and prove you wrong, and after a bunch of time eventually people whose job it is to prove each other wrong end up roughly agreeing on a small sliver of information that lies at the intersection of all their understandings of the world.

Signals in physics can be noisy too (fortunately, in physics you can take millions of samples). But to say "well, these error bars are too big, so therefore we shouldn't pay any attention to the trend at all -- we can decide to do something when we have much higher confidence in the precise figures. In the meantime, I'm going to oppose tighter fuel economy standards for vehicles, and revenue-neutral carbon taxation, and aggressive subsidization of alternative energy R+D, ... " because god forbid we should all drive around in smaller cars, or that we should stop subsidizing air travel, or that we should too quickly improve solar cell efficiency.

It's definitely not the case that climate scientists are uninterested in discussing methodological problems with paleoclimate proxies. A considerable amount of the literature is taken up with doing precisely that, and it takes up a good portion of the Paleoclimatology section of the IPCC report: http://www.ipcc.ch/pdf/assessment-report/ar4/wg1/ar4-wg1-cha...

I know that some folks in paleoclimatology are interested in these core methodology problems. But these concerns are not conveyed to a wider audience. When a temperature reconstruction graph is released to the public there is no mention at all of these problems and one can easily get the impression that these reconstructed temperatures can be known almost as accurately as going outside and taking a reading from a thermometer.

Can the public understand that?

Excluding math, we use proxies for just about all science. With different assumptions and different criteria. I don't deny that there are politics, assumptions and margins of error but they do put effort in to minimizing them and doing real science. Take a look at pharmaceuticals, they don't even know why a lot of medicine works, in fact there is debate on if it even does... But people still buy it and take it.

The public can't understand that, but it certainly forms opinions based on it. It votes, too...

The public is, honestly, incapable of understanding, or even caring, about underlying issues in proxy temperature modeling. A large amount of the public is incapable of distinguishing between weather and climate.

Proxy temperatures, statistical methods, atmospheric modeling, etc, are all scientific areas whose knowledge can be gleamed at sufficiently well for a person with some form of higher education in science or math education (but only barely; remember that this stuff is being researched by people who have devoted their entire lives to the scientific specialty). But that is an astoundingly low proportion of the population. And even then it is clear that, as methodological errors have diminished and methods for more accurate models have appeared, that they all tend towards the same conclusion.

We should be worrying significantly more about the small but non-trivial (and ever-increasing) probability of an extinction event within the next decades rather than pretend that methodological errors, at this point, are a significant issue in identifying the climate trends we have spotted in the last decades.

Okay. What do you say of purely geological phenomena, like glacier coverage. You can look at hydrostatic rebound, erosion patterns, etc to see where glaciers usually are. And you just have to look at pictures within the last 50 years to see where they're not. Much harder data. Same for ecological ones like flora and fauna existing in and being habituated in environments that are no longer very compatible climatologically. And for anyone in their forties or older, there's always your own damn lying eyes.

There's a billion pieces of supporting evidences for global warming. As far as I have ever heard, there's not a single piece of evidence against it, just piles and piles of FUD thrown at each model that tries to explain it.

Noone is arguing that there WAS a warming trend. The onous to prove its due to humans lies on the chicken littles. It has not yet been accomplished.

Actually, the parent was arguing that we can't trust the one figure that is in TFA. It's a picture of a warming trend.

Surprisingly enough, the fear, uncertainty, and doubt about anthropogenic warming claims have come from people who receive money from carbon-intensive industry groups. It's these people who keep on claiming that it "hasn't been accomplished". And since FUD is easy to do, the FUD-purveyors are winning in the public sphere.

What kind of evidence would convince you, exactly?

No one needs to prove it's due to humans; it doesn't matter what it's due to, it only matters that it's bad for us and we can do something about it. Natural is not equal to good, claiming it matters if humans caused it is as stupid as refusing to stop an asteroid from hitting us because it's not a man made disaster.

Heh. I've been using the same asteroid example for years.

Actually, everyone but you acknowledges the warming trend, even groups that oppose AGW theories. The fashion du jour is to blame it on volcanoes or solar output cycles. Catch up!

How accurately can we track CO2 concentration over long timespans? I get that climate science is hard and produces nebulous answers based on arguably nebulous data, but when I read about about ocean acidification and see the long-term graphs from respected organizations [1] as a layman it's really quite alarming.

I.e. maybe the climate won't spiral out of control, but it seems reasonable to be concerned about the alarming levels of CO2. As a new parent I'd love to have my dire fears of catastrophe awaiting future generations assuaged by rational hard science, but (as a layman) everywhere you look the picture is pretty bleak.

1: http://www.esrl.noaa.gov/gmd/ccgg/trends/history.html

AFAIK one of the main ways to reconstruct CO2 concentration is to study samples of air trapped inside old ice. This looks like a pretty direct and reliable method to me.

Some more discussion of the problem here: http://www.skepticalscience.com/Tree-ring-proxies-divergence...

Reply on news.ycombinator.com