"Our greatest responsibility is to be good ancestors."

-Jonas Salk

Monday, May 24, 2010

Eschenbach has Half a Point

This isn't to justify most of what Willis Eschenbach does, but he has a point in yesterday's WUWT article "On Being the Wrong Size", or at least a half a one.

GRACE measures the gravity of the planet, and it provides information that the size of the Greenland ice sheet is declining. Eschenbach's half a point, quoting an article on Grist by Seth Shulman:
So the next time you read something that breathlessly says …

“If this activity in northwest Greenland continues and really accelerates some of the major glaciers in the area — like the Humboldt Glacier and the Peterman Glacier — Greenland’s total ice loss could easily be increased by an additional 50 to 100 cubic kilometers (12 to 24 cubic miles) within a few years”

… you can say “Well, if it does increase by the larger estimate of 100 cubic km per year, and that’s a big if since the scientists are just guessing, that would increase the loss from 0.007% per year to around 0.010% per year, meaning that the Greenland Ice Cap would only last until May 23rd, 12010.”

OK, so scientists aren't "just guessing", let's leave that aside. And it really isn't clear what the prior expectation was.

In fact the story is not the rate of ice loss, nor the ice sheet as a whole, but the confirmation of estimates of accelerating mass loss at the fringes:
The team found that uplift rates near the Thule Air Base on Greenland's northwest coast rose by roughly 1.5 inches, or about 4 centimeters, from October 2005 to August 2009. Although the low resolution of GRACE -- a swath of about 155 miles, or 250 kilometers across -- is not precise enough to pinpoint the source of the ice loss, the fact that the ice sheet is losing mass nearer to the ice sheet margins suggests the flows of Greenland outlet glaciers there are increasing in velocity, said the study authors.
and
"When we look at the monthly values from GRACE, the ice mass loss has been very dramatic along the northwest coast of Greenland," said CU-Boulder physics Professor and study co-author John Wahr, also a fellow at CU-Boulder's Cooperative Institute for Research in Environmental Sciences.

"This is a phenomenon that was undocumented before this study," said Wahr. "Our speculation is that some of the big glaciers in this region are sliding downhill faster and dumping more ice in the ocean."
As you see, this isn't "just guessing", and GRACE has higher resolution than the subcontinental scale that Eschenbach is going on about. The press release doesn't provide maps. Here's a nice little piece at geoinformatics.com that does.

Here's what Grace sees:



Here's some independent evidence about the melt season, which you can see maps nicely onto the GRACE data and gives you a good idea of the resolution of GRACE.



What we see is Greenland softening at the edges. Ice is sort of a glassy substance, that flows much faster as it warms, so we see the beginnings of a possible failure mechanism for the whole ice cap structure. That is what we should be worrying about, and it means that Eschenbach is, willingly or unwillingly, performing some sleight of hand here.

But ultimately, he is objecting not to the press release, but to Grist's take on it. And here Eschenbach's point stands on its own.
Finally, the original article that got my blood boiling finishes as follows:

The good news for Luthcke is that a separate team using an entirely different method has come up with measurements of Greenland’s melting ice that, he says, are almost identical to his GRACE data. The bad news, of course, is that both sets of measurements make it all the more certain that Greenland’s ice is melting faster than anyone expected.

Oh, please, spare me. As the article points out, we’ve only been measuring Greenland ice using the GRACE satellites for six years now. How could anyone have “expected” anything? What, were they expecting a loss of 0.003% or something? And how is a Greenland ice loss of seven thousandths of one percent per year “bad news”? Grrrr …

I’ll stop here, as I can feel my blood pressure rising again. And as this is a family blog, I don’t want to revert to being the un-reformed cowboy I was in my youth, because if I did I’d start needlessly but imaginatively and loudly speculating on the ancestry, personal habits, and sexual malpractices of the author of said article … instead, I’m going to go drink a Corona beer and reflect on the strange vagaries of human beings, who always seem to want to read “bad news”.
Yuppers. There is something to what he says.

But it's only half true. Pretty much exactly half true.

I'm not fond of false symmetries, it makes for such an easy target for an essay. In this case, I'll make an exception; I see a very close to perfect symmetry here.

Lots of other people cherry pick information to support their point of view. The site Eschenbach writes for is a fine example. If only he weren't guilty of cherry-picking in the opposite direction, if only other people didn't only want to read "good news", we might be able to make some progress.

Greenland is melting detectably and contributing detectably to sea level rise. The quantity is now reasonably well constrained. That's good news, scientifically. It's slightly bad news as far as sustainability is concerned (the change might still have been undetectable, but it isn't.) It's too early in the record to detect any acceleration (*). If and when it accelerates, we'll be in a position to detect that, too. Grist does not have a real basis for "faster than anyone expected", but Eschenbach does not have a basis for being sanguine about it either.

By the way, you will note the increase in mass in the Greenland interior. That is increased snowfall. While this mitigates the net melt and the sea level rise a bit, it's consistent with expectations from global warming: increased winter temperature =>increased winter column moisture => increased snowfall. It is a negligible term in the force balance so far, but in the long run it would also increase the pressure gradient and tend to further accelerate the glacial flow.

Update: (*) It's clear that Greenland once must have been in mass balance just from basic mathematical principles, so arguably any loss at all must be an acceleration. In fact, early measurements in the mid 20th c. did seem to show mass balance, though they were very crude.

My point is that the GRACE record itself shows no acceleration of mass loss over a decadal time scale. I'm basing that on this figure, from Velicogna and Wahr, which I believe is a mass total, not a mass flux. Sorry if this was unclear.



I have little doubt that ice sheet mass is retreating and accelerating. The retreat is not yet rapid, and this is the closest thing to a legitimate point of Eschenbach's. The GRACE record is just too short to provide a convincing demonstration of acceleration in itself, though it certainly is suggestive.

Some of these guys are just down the hall from me. I guess I should just ask them!

Update: It's looking a little stronger on "acceleration"; thanks to a correspondent. Here's the latest Velicogna. I don't call this a slam dunk, myself.



Update
: Gavin has a really nice piece on RC that addresses some of the issues raised here.

27 comments:

Anonymous said...
This comment has been removed by a blog administrator.
Paul said...

The quote that tripped Eschenbach’s trigger is from S.A. Wahr, the lead author of the first study which used GPS data on isostatic rebound, to confirm the Grace data on mass loss. The quote does not appear in the Grist article. It does appear in the following. You refer to a press release. Is this it?

http://news.mongabay.com/2010/0328-hance_greenland_melt.html

I can’t find anything to quibble with in either the press release or the Grist article, which is a profile of NASA GRACE Scientist, Scott Luthcke. “faster than anyone expected” is a quote from Luthcke. I think it is a reasonable statement particularly in view of John Cook’s typically thorough and very timely post on this topic.

http://www.skepticalscience.com/news.php?n=203

He is reporting on a May 16 Nature letter that reports a dramatic acceleration in rebound and mass loss beginning in 1990.

Finally, I am confused by your statements:

"It's too early in the record to detect any acceleration. If and when it accelerates, we'll be in a position to detect that, too."

Is the “it” you are referring to, Greenland’s contribution to sea level rise?

Paul Middents

Michael Tobis said...

Yes, or the total mass balance of the ice.

A fair point, see the update above.

Anonymous said...

The mass loss of the Greenland ice sheet accelerated from the 1980's to 2000 as well, this will not be shown by the GRACE measurements but is evident from the size of melt area . In terms of the <a href="http://glacierchange.wordpress.com/2010/03/27/petermann-glacier/"Petermann Glacier, this is a slow moving thin floating ice tongue than can collapse but not really accelerate much in its lower 80 km which is the floating part.

Aaron said...

Ice is “ice”, not a “glassy substance”. Glass has a softening point as does ice. However, ice also has an abrupt melting point. You can pave a floor with glass, and it will be smooth, but you will not be able to “ice skate on it”. Ice is different. Ice near its melting point has the rare ability to flash from solid to liquid. Warming ice under pressure can do this just as surely as stepping on to an ice rink with freshly sharpened skates produces a film of water under the skate blades. Warming ice under pressure can result in the rapid formation of cracks in the ice filled with water. Then, the two sides of the crack can move relative to each other with great ease, putting pressure on other parts of the ice, resulting in “ice dynamics.

This is a melting point phenomenon. It initiates abruptly. It will not be like anything we have seen when Greenland was colder.

While ice is a poor conductor of heat, Greenland is surrounded by water, including streams of water under it and water vapor over it. Water and water vapor are very good at advecting heat. Greenland will come to 0C pretty much all at once (geologically speaking). Already, moulins are marching up its flanks. Models (with gridding) miss this kind localized hydrology.

Recently, the top of Greenland was wet. My guess is that this melt season we will see liquid water and moulins on Greenland as high as 3,000 meters (a thousand meters higher than last melt season. Remembering that water falling through a moulin releases heat, we have the rapid warming of the core of Greenland.

I am not worried. As long as the Arctic Sea Ice remains intact, I doubt if sea level rise from Greenland will come close to a cm per year. After the Arctic Sea Ice becomes seasonal, then sea level rise from Greenland ice dynamics could easily become several cm per year. However, the Arctic Sea Ice should remain somewhat intact for the next couple of years.

Jim said...

The Antarctic ice mass loss appears to be supralinear: Antarctic Ice Mass Loss, 2002-2009

Steve Bloom said...

Hmm, well, that acceleration may or may not be statistically significant but it sure does put WE's assertions on much shakier ground. That's what you get for taking any of those people at face value, Michael.

On a related subject, I had been assuming that the ESA GOCE mission, in theory more sensitive than GRACE due to its much lower operating altitude (requiring aerodynamic fins and an ion engine to compensate for atmospheric drag), would be able to add detail to the ice sheet picture either on its own or via interpolation with GRACE data, but according to a paper I just this doesn't seem to be the case. NASA does have a GRACE follow-on planned, but I don't know whether much more can be expected from it. OTOH, given GRACE plus altimeters plus GPS elvation ground-truthing, it seems to me there's enough to support the fundamental conclusions that have been drawn regardin Greenland.

Regarding Grist, failing to understand that the real problem is what happens to the volume of ice loss over the next few decades rather than the next few years strikes me as a forgivable lapse unless, like WE, we prefer to avoid admitting there's any problem until we're watching the ice sheet collapse in the rear-view mirror.

Willis Eschenbach said...

Michael, first, thanks for your take on my essay. A few points.

First, when someone says "“If this activity in northwest Greenland continues and really accelerates some of the major glaciers in the area — like the Humboldt Glacier and the Peterman Glacier — Greenland’s total ice loss could easily be increased by an additional 50 to 100 cubic kilometers (12 to 24 cubic miles) within a few years”, that's called guessing. There's a couple "ifs" and a "could easily" in there, which means we're no longer talking about facts. Yes, if it continues, and if it really accelerates then loss could easily be increased. But it is equally true to say that if it doesn't continue and if it decelerates that loss could easily be decreased. That's called guessing, and to me has no place in science.

Second, the first Velicogna figure says "The results shown here have not been corrected for PGR or for the effects of hydrological or oceanic leakage." So it is not clear exactly how accurate the figures are.

Third, per the second Velicogna figure, the apparent (although not significant) acceleration is all in the first two years. For the last four years of the record, there is no apparent acceleration. A quatdratic fit to those years is indistinguishable from a linear fit.

Fourth, the overall record is so short (six years) that there is no statistically significant difference between a quadratic and a linear fit. Thus we cannot yet say if there is acceleration. The lineal nature of the last four years of data argues against it. So as you say, acceleration is not a "slam dunk". It will be some years before we can even determine if there is an acceleration or not, much less get an estimate of its size.

So it is irresponsible for a scientist to say the loss is accelerating. That is nothing but hype and alarmism. We don't know that it is accelerating, it is just another guess.

My point was much simpler, however. Acceleration or not, the size of the ice loss is trivial.

All the best,

w.

Willis Eschenbach said...

There's another oddity, which is the difference between the Velicogna 2006 and 2010 results. These vary quite widely, with a number of them being fairly small, while others are on the order of 200 gigatonnes.

The changes are so large that about half (19 out of 41) of the month to month changes are of different signs in the two datasets.

I don't understand this at all. You would think that ocean leakage and PGR would affect all of the months about the same ... but the effect is not constant in any way.

Since 200 gigatonnes is about the size of the yearly loss, that's a huge difference. I find this very curious. Anyone have any explanations for that?

Mark said...

Willis, you claim that for the "the overall record is so short (six years) that there is no statistically significant difference between a quadratic and a linear fit. Thus we cannot yet say if there is acceleration."

However in the 2009 paper

http://www.agu.org/journals/gl/gl0919/2009GL040222/

Velicogna considered the statistical significance of the acceleration

"[17] To verify that the improvement obtained with the quadratic model is significant we used an F-test [e.g., Berry and Feldman, 1985]. The F-test show that the improvement obtained with the quadratic fit is statistical significant at a very high confidence level (99%)."

If you can see something seriously wrong with that I suggest you submit a comment to GRL.

Michael Tobis said...

I really have my doubts about the statistical analysis in this paper.

Of course we expect acceleration on the basis of our understanding of the physics of the situation. But that's not what this flavor of statistics is about. It's about just looking at the data. Now, I think that approach is overrated in our field, but that's beside the point.

Visually, the "acceleration" is not compelling. Usually the issue with statistics is that the eye sees correlations that aren't there, not the other way around.

So I find it peculiar.

Mark said...

So, MT, you'll be a co-author on Willis's comment? :-) It's a bit beyond me, I'm afraid.

Martin Vermeer said...

Michael,

there could well be statistically significant acceleration on a six year time scale -- but that wouldn't be climatologically relevant. It would be natural variability. And I suspect it would go away if the autcorrelation structure of the data were properly taken into account.

The only significant thing that GRACE can give us from this one short mission is the rate of mass loss itself, assuming it was zero before AGW started in earnest. Say this amounts to 0.5 mm of sea level per year. That is 5 cm per century, or 12000 years to melt all of Greenland -- as Willis says. Now of course we have at present 0.8C above pre-industrial. If temps stay constant at that level, then also the rate of melt will stay constant; but you know and I know that that's not going to happen. And then we're looking at decimetres by 2100, for Greenland alone.

The nonlinearity of this is decadal. We're gonna need that GRACE follow-up mission.

Michael Tobis said...

Let me try to make it clear.
Willis wants us not to worry.

I don't want us not to worry.

Both of us, and Martin who has been a very perceptive commentator here in the past, think the GRACE record is too short to indicate acceleration.

The paper itself claims that the record is sufficient.

I am not particularly expert in this flavor of statistics. (It's not the way Norbert Wiener used statistics at all.)

I think many scientists, in our field as well as many others, tend to be too cook-booky about how to use statistics. I suspect this may be a case in point.

Do I intend to coauthor a comment with Willis? Seriously?

No. He shows no signs of having the statistical skills to take this on. It wouldn't help me personally nor would it add credibility to any point I was trying to make.

Please understand that whether the claim in V's point 17 is valid or not is pretty much orthogonal to whether we should worry about acceleration in Greenland.

I agree 100% with what Martin said above.

Paul said...

The works cited by Skeptical Science in my comment above show clear (to me at least) GPS evidence of acceleration in isostatic rebound starting in 1990 which they relate to GRACE and find consistent with mass loss. Does this have any bearing on the statistical analysis you and willis find questionable?

Paul Middents

Michael Tobis said...

Willis' point and my original half-approval were about the existing scale of the melt, not about acceleration.

I think the GRACE record by itself looks unconvincing as evidence of acceleration. I find V's point 17 surprising and probably ought to roll up my sleeves and look into it.

I think there's little doubt that there is acceleration, but still the actual acceleration, i.e., a quantitative measure of the second derivative of mass vs time, is very uncertain.

The early data is very coarse but indeed supports this point of view as Paul suggests.

Mark said...

No, MT, I'm not really suggesting that you and Willis co-author a paper. Velicogna presents evidence for accelerating mass loss from Greenland and Antarctica in the period 2002-2009. I am not competent to assess the statistical analysis, but they look convincing to a non-expert like me and have passed peer review. Therefore I am inclined to accept her finding, notwithstanding expressions of casual skepticism from you and Willis.

(I would make the comment that a lot of the variability in the mass vs time plots is a seasonal variation, which can be removed, Doubtless there are statistical traps and subtleties in such a removal.)

Velicogna's analysis is based entirely on the short time series she has available. The acceleration may continue in the future or it may not. This analysis on its own cannot tell us which is the more likely.

Willis Eschenbach said...

Mark said...
Willis, you claim that for the "the overall record is so short (six years) that there is no statistically significant difference between a quadratic and a linear fit. Thus we cannot yet say if there is acceleration."

However in the 2009 paper

http://www.agu.org/journals/gl/gl0919/2009GL040222/

Velicogna considered the statistical significance of the acceleration

"[17] To verify that the improvement obtained with the quadratic model is significant we used an F-test [e.g., Berry and Feldman, 1985]. The F-test show that the improvement obtained with the quadratic fit is statistical significant at a very high confidence level (99%)."

If you can see something seriously wrong with that I suggest you submit a comment to GRL.


I haven't had much luck, despite very good statisticians agreeing with me, with comments to GRL. However, let me go over the problem.

Velicogna et al. figure their statistics based on their fit with the "filtered" data. This is the data shown in red in Figure 1 of their paper. Unfortunately, they have ignored the effect of autocorrelation on their statistics. The filtered series that they use is extremely autocorrelated, which means that their statistics are extremely wrong.

Michael Tobis, you say:

[Willis] shows no signs of having the statistical skills to take this on. It wouldn't help me personally nor would it add credibility to any point I was trying to make.

You need to be cautious about your estimation of the skills of people you have never met. In particular, you, and the authors of the paper, should take a hard read of Doug Nychka's paper, ""Confidence intervals for trend estimates with autocorrelated observations" available here.


The short version is that autocorrelation reduces the number of degrees of freedom. Doing a filter like Velicogna et al. did reduced the number of degrees of freedom in their test data to approximately ... 1. As a result, their tests are meaningless. They used the right test (an F-test on the adjusted R^2 values), but they used 74 (and 73) degrees of freedom in the test, which gave them an incorrect answer.

w.

Michael Tobis said...

Willis' suggested explanation is exactly where I would look.

I suspect, specifically, that they are treating the smoothed series as having n degrees of freedom whereas it ought to have (I think) n/12.

Martin Vermeer said...

Willis, Michael, I agree. BTW Willis I don't know where you got DoF = 1 from. Michael's estimate based on the smoothing she is seen to be doing, makes sense although it is an upper bound... I would think n/6 though.

One way to get an idea of the number of DoF in the data is to look at the residuals (relative to the parabolic fit) and see how often they wiggle between positive and negative.

Another thing I have a problem with is the estimate of 0.17 +/- 0.05 mm/yr^2 she gives for the combined effect of Greenland and Antarctica. That would produce 1.4 meters by 2100... without even looking at the other contributions. I don't buy it.

About submitting a comment, why not. And why not together with an experienced scientist, might improve your luck. It is important to get things like this in the record. Alternatively, write
to Isabella, she might consider a corrigendum.

We (scientists) always criticize contrarians for not doing science, but when they try, and happen onto something, I think we have a duty to acknowledge it. Who knows, the attitude might be contagious.

Aaron said...

Actually, if one posts a good idea here, some hungry grad student will see it, write it up, and publish.

Mark said...

Martin, I agree with everything you say except this

"Another thing I have a problem with is the estimate of 0.17 +/- 0.05 mm/yr^2 she gives for the combined effect of Greenland and Antarctica. That would produce 1.4 meters by 2100... without even looking at the other contributions. I don't buy it."

That would only be true if you extrapolated the quadratic to 2100. Why would anyone extrapolate a quadratic fit based on a sample of 7 years of data out to 91 years beyond the sample end-points? In either direction?!

Martin Vermeer said...

Mark, well yes, precisely my point. But if one interprets the acceleration seen as climatological -- i.e., a small part of a much longer global warming trend -- then such extrapolation would make some sense. And reduces to absurdity.

Or do you object to using a quadratic rather than, e.g., a physically more sensible exponential (which would give higher values still)? Yes, then I agree. I am reminded of Loehle's latest CO2 thingy.

Willis Eschenbach said...

Martin said...
Willis, Michael, I agree. BTW Willis I don't know where you got DoF = 1 from. Michael's estimate based on the smoothing she is seen to be doing, makes sense although it is an upper bound... I would think n/6 though.

I got it from the paper that I cited. Please read the Nychka section, which spells out the exact details.

Martin Vermeer said...

Willis, you'll have to spell it out for me...

Willis Eschenbach said...

Martin said...
Willis, you'll have to spell it out for me...

Sure. The effective sample size is equal to:

n_eff = n * (1 - r - .068/sqrt(n)) / (1 + r + .068/sqrt(n))

where "n" is the sample size and "r" is the lag-one autocorrelation. As autocorrelation goes up, the effective sample size (and thus the number of degrees of freedom) goes down.

Anonymous said...

You may find this interesting,

Who is Willis Eschenbach?

As of 2012 Mr. Eschenbach has been employed as a House Carpenter.

He is not a “computer modeler”, he is not an “engineer” and he is certainly not a “scientist” (despite all ridiculous claims to the contrary).

"A final question, one asked on Judith Curry’s blog a year ago by a real scientist, Willis Eschenbach…"