Report comment

(And “why don’t you” questions seem to be fundamentally different than “why do you” questions.)

Report comment

Report comment

Report comment

Tamino did a synthetic-data look at a similar effect. It definitely does reduce the size of peaks, but not enough to get rid of them – if there was anything remotely similar to the current warming we would see it in Marcott. AFAIK MBH98 had enough uncertainty on the shaft to allow that sort of thing to happen undetected.

It’s worth noting that MBH98 didn’t just simply average proxies, the PCA method used would reduce the impact of things like the Tiljander proxy dataset (for those following at home, Mann’s statistical method for calibrating proxy data against observational data to determine the relationship put the Tiljander sediment data series the wrong way up. IIRC there’s evidence of the proxy being disturbed by agricultural runoff or something in the calibration period MBH98 used and that was the reason? It might also have been that Tiljander is just not a good proxy. Don’t recall. Didn’t have a large impact on the resulting reconstruction).

Report comment

And if you follow a more Mannian “kitchen sink” approach and toss in every proxy you can think of with no attempt to verify temperature sensitivity or even check that the signal ends up being used right-side-up, that dampens the signal even more – some component of your signal is either random or is non-random noise introduced by non-temperature-related components. Adding more data only helps if it’s GOOD data you’re adding.

Taking these sort of factors into account, you CAN use a proxy to suggest what times may have been warmer or colder than other times during the length of the proxy record, but you have to keep in mind that the overall variability is being dampened and/or clipped so we can’t directly compare the top (or bottom!) values to the modern-day instrument record. You might try to fudge this by adding an uncertainty envelope, but if so, this envelope should add MORE uncertainty at the extremes rather than a uniform amount throughout.

It’s not worth my trying to debunk the ENTIRE set of skepticalscience talking points here in the forgotten margins of Scott’s blog, so I’m going to leave it at that for now. Thanks for being reasonably civil; I’m sure the issue will come up again.

Report comment

Sure, tree rings do weird stuff outside of their optimal range, but that optimal range is different from the optimal range of other proxies, and if you’re doing something more sophisticated to combine them than simple averaging you can use that fact.

This feels a bit having-and-eating, too – because you then go on to compare glacier proxies to the present day (p.s., your intuition is probably driven by most of the glacier records being northern-hemisphere, because most of the land – particularly land in places that form glaciers – is in the northern hemisphere. That said, Ljunquist strongly suggests that the peak of the MWP in the northern hemisphere wasn’t warmer than the present day on average, and modern-day NH temperatures are even higher than that, and modern-day NH land temperatures – probably the most direct comparison – higher still…). There /are/ several paleoclimate studies that don’t use tree rings. They show essentially the same result as all the rest.

I’m still curious as to how you reconcile low CS values with a large MWP, as well. Was there a gigantic forcing going on at the time? What was it? Do you think climate has gigantic natural variability that somehow isn’t present in the modern day and also doesn’t imply an easily-perturbed climatological system? Do you think that CS is highly nonlinear, dropping off rapidly past present day temperatures? The last one is at least self-consistent, but as far as I’m aware there’s zero evidence for it, it would be Weird in a pretty significant way (all the major identified feedbacks don’t have any obvious nonlinearities at present-day temperatures to several degrees above it), and it also seems like a convenient fallback in the same way “The dragon is flour-permeable” is a convenient fallback.

EDIT: I guess there’s also the question of whether you consider comparing UAH/RSS to HADCRUT/GISS legitimate is there, as well. These are completely different mechanisms for observing temperature, with different limits, different error modes, different systematic failures, even different responses to particular modes in the climactic system (it’s pretty well demonstrated that ENSO events show up much more dramatically in the satellite datasets than in the surface datasets).

EDIT EDIT: I probably shouldn’t let this go by either:

(Loehle’s proxy is probably more accurate in terms of variability than the others in part BECAUSE it uses fewer (and better-qualified) proxies – using more inherently is likely to dampen the signal. But that’s another argument for another time; I’m done with this one.)

This is utterly ridiculous. It’s well-known that regional temperatures are much more variable than global average temperature – compare any annual temperature data series (say, greenwich to any global average dataset (say, GISTEMP. Keep your eyes on the scale!).

That’s because when individual regions get warm or cold, it’s not because the Earth as a whole has gotten warm or cold, it’s because energy within the climate system has shifted around – place A is warmer than average because place B is colder than average, and the total amount of energy is roughly conserved (obviously there is still global variability, just less).

Using small numbers of not-very-well-distributed proxies (and not area-weighting!) and then combining them via simple averages and not weighting the more informative proxies is a recipe for displaying as much variability as possible. Loehle’s ‘reconstruction’ is a NH-extratropical reconstruction dominated by the proxies with the highest variability and the most points, and it’s much less reflective of global average surface temperature than any of the more mainstream ones as a result. They’re not even well-selected proxies – some of the ones he used have a grand total of three datapoints in the time period he’s reconstructing. One of them might not have any data points in the the period.

Report comment

The problem is that proxy data is NOT temperature data, so you can’t treat it as such.

Let’s use the MBH tree rings as a simple example. We’ll ignore the divergence issue and just reason about them from first principles. So: The way we convert tree rings into a temperature history is that we go out and find some old trees near the treeline that we believe have been temperature limited – meaning they seem to grow better when it’s warmer during the growing season. It doesn’t matter WHY they grow better when it’s warm. Maybe there’s more snowmelt or more rain when it’s warm so they get more water. Maybe there’s a different level of cloud cover when it’s warmer so they get more (or less?) sun. The important thing is that we have some reason to think there is a consistent {warmer => better growth} relationship and we already know better growth produces thicker tree rings, so if we measure the rings seen in a core sample and assume a linear temperature/growth relationship, we can translate that info into something kind of like a summer temperature record. Correct so far?

However (notes Forest Ecology expert Loehle in several papers, including one I linked above), the temperature/growth function even for temperature-limited trees can’t be linear. (If it were, trees could grow an infinite amount if it were infinitely hot.) Rather, the relationship has more of an inverted-U shape. Which is to say that if we are looking at any one specific tree in one specific location with its specific local constraints, there must exist some OPTIMAL TEMPERATURE that would maximize that tree’s growth. Being significantly colder OR WARMER than optimum means less growth. Still with me?

So if you’ll indulge me with a thought experiment: let us imagine that the MWP is global and includes a period MUCH WARMER than today, so much warmer that it was ABOVE the optimum growing temperature for the tree we’ve sampled.

What would that do to the shape of our chart? Answer: as that peak MWP temperature approaches, our “temperature” chart (that until then had been pretty accurate) would suddenly go flat or even show a DECLINE in “temperature”. If we compared this result to the charts from other proxies in other parts of the world, we might then conclude the MWP was “only a local phenomenon” and that it did NOT peak at levels above the current day.

The mere fact that our “temperature” is being measured by a PHYSICAL PROCESS that has a physical maximum possible result means very-warm past temperatures must at some point get CLIPPED.

And there is no scaling factor that can correct for this clipping – there’s no way to tell the difference near the top of the peak of the MWP between “the temperature trend actually flattened or declined a bit here” versus “the temperature kept rising but exceeded the ability of this proxy in this location to tell us it was doing that so it LOOKS like it flattened or declined a bit here”

So when you just tack on the instrumental record at the end, you end up comparing a recent temperature record that WASN’T clipped at the high end to an older one that was. Or at least might have been. Which is great propaganda if you want to claim temps are “unprecedented”, but terrible science.

If my hunch that the MWP was as warm or warmer than today is true, the simplest prediction it makes is that when we update the proxies we WON’T see record-breaking tree rings to match our current record-breaking temperatures. And sure enough, we don’t – we instead see The Divergence Problem.

There are dozens of other problems with the big non-Loehle reconstructions, but this one alone is damning enough. (Applying the logic of this example to other proxy records is left as an exercise for the reader.)

As for WHY I suspect the MWP was about as warm as today, you can get that impression directly from glacier evidence without need of detailed multiproxy reconstructions. To wit: nearly everywhere around the world that we find retreating glaciers, as they retreat they reveal evidence that treelines were higher roughly a thousand years ago than they are today or that the now-covered area wasn’t covered back then.

(Loehle’s proxy is probably more accurate in terms of variability than the others in part BECAUSE it uses fewer (and better-qualified) proxies – using more inherently is likely to dampen the signal. But that’s another argument for another time; I’m done with this one.)

Report comment

Report comment

Report comment