Guest Post by Willis Eschenbach
I wrote a post called I Used To Be Snow White (… but then I drifted) a week or so ago about a study titled “Impact of Declining Arctic Sea Ice on Winter Snowfall” (PDF) that claimed to link low arctic ice levels with high snow levels. To recap, their specific claims were:
Abstract
While the Arctic region has been warming strongly in recent decades, anomalously large snowfall in recent winters has affected large parts of North America, Europe, and East Asia. Here we demonstrate that the decrease in autumn Arctic sea ice area is linked to changes in the winter Northern Hemisphere atmospheric circulation that have some resemblance to the negative phase of the winter Arctic Oscillation. However, the atmospheric circulation change linked to the reduction of sea ice shows much broader meridional meanders in mid-latitudes and clearly different interannual variability than the classical Arctic Oscillation. This circulation change results in more frequent episodes of blocking patterns that lead to increased cold surges over large parts of northern continents. Moreover, the increase in atmospheric water vapor content in the Arctic region during late autumn and winter driven locally by the reduction of sea ice provides enhanced moisture sources, supporting increased heavy snowfall in Europe during early winter, and the northeastern and mid-west United States during winter. We conclude that the recent decline of Arctic sea ice has played a critical role in recent cold and snowy winters.
I showed that if there is such an effect, it is not visible using the snow data for the whole US. I thought this would settle it. But folks said, and fairly, that I wasn’t really dealing with their claim. They said I needed to deal with a) the regional nature of their claim, involving northeastern US and Europe, and b) the temporal nature of the claim, comparing only winter snowfall to autumn sea ice. So I decided to take a look at the winter snow data for the northeastern US compared to autumn sea ice.
Unfortunately, I couldn’t find area-wide data for the northeastern US, but I did find something better. This is one of the longest continuous records of snowfall in the northeastern US—the century and a half long record of the snowfall in Central Park in New York City.
Figure 1. Winter snowfall (December/January/February) for Central Park, New York. There is a slight but not statistically significant decrease in winter snowfall over the last century and a half.
So … how well does this correlate with the arctic ice levels? Well, not to put too fine a point on it … no better than my first look at the question.
Here’s the comparison of the snow and ice. I have standardized both of them so that we can compare them directly.
Figure 2. Central Park winter snow (December/January/February) versus Arctic autumn ice (September/October/November). Data have been standardized to allow a direct comparison
As you can see, there is little correlation, and the numbers bear that out. There is a weak statistical relationship (r^2 = 0.13) which is not significant at the p<0.05 level.
I thought that because there is no trend in the snowfall data, perhaps I might get better significance if I detrended the ice data. This would highlight the year by year variations that are theoretically responsible for the year-by-year variations in snowfall. Figure 3 shows that relationship, with the ice data inverted to better illustrate the relationship.
Figure 3. Central Park winter (DJF) snow totals versus inverted, detrended Arctic autumn (SON) ice levels.
Now, this is a very interesting figure, because it illustrates the way that our eyes find patterns when none are there. At first glance, this looks like it is a pretty good relationship. But in fact, that is an illusion. The mathematical analysis says that the r^2 is even worse, only 0.02, and like the previous graph, it is also not statistically significant, in fact the significance is worse (p ≈ 0.4).
Upon closer examination, we can see why that is so. For example, from about 1990 to 1995 when ice decreased, snow generally increased … but not proportionally. When the ice was extremely low there was a little more snow, and when the ice was only a little low, there was a lot more snow. For there to be a relationship, it needs to be proportional. Also, although in general the snow seems to change with the ice, in fact on a year by year basis, there are huge excursions. Look at 2011, for example, very low ice, but in contradiction to their claim, there’s also very low snow.
So I have found the same thing on a regional level using their autumn ice/winter snow claim, that I found when I looked at the data for the entire US for the full year. If there is any association between winter snowfall in the northeastern US and the autumn ice levels, it is very, very weak. There certainly is no sign of it in the Central Park records.
All the best,
w.
Discover more from Watts Up With That?
Subscribe to get the latest posts sent to your email.
I just tried Lake Placid, New York snowfall as well, still no joy. R^2 = 0.15, p = 0.07.
w.
And Rangeley, Maine, came the closest yet, with p = 0.054 … so near, and yet so far …
Weak. There is a very, very weak association between some sites and the ice, but I haven’t found one yet that is significant at the p less than 0.05 level …
w.
Interesting Willis. I have long been of the the belief that a cold winter would be followed by a warm summer, and that a mild winter was followed by a mild summer. That was my personal observation of how the weather worked for me. Or so it seemed. I didn’t check against records.
My reasoning was simply that the average does not change much from one year to the next, so to compensate for a high, there had to be a subsequent low.
In this case it looked reasonable to make the same assumption, but on a regional basis – too little ice here, dump some more snow there to even the balance.
I wonder if that idea has any merit?
Regards
Vincent
Willis in your comment on March 13, 2012 at 11:06 am under your PS you say:
“This means that the effect is not operating on a year by year basis … which makes it very likely that the somewhat greater (but still not statistically significant) correlation using the non-detrended data is only an artifact.”
Well said. – Personally, I hope we are not still around at a time when the North Atlantic and The Arctic Oceans get a better chance to “equalibralize”. – Then – we may see one “heck of an “Effect” – Year after year.
O H Dahlsveen says:
March 13, 2012 at 1:25 pm
No clue what that means, O H. What is “equalibralize” and how do oceans do that? Why would we see an “Effect” from that, and what “Effect” would we see?
What am I missing here?
w.
Willis,
You can not seriously be defending the use of Central Park, NY as a proxy for snowfall in the entire NE US. Had Mann or Jones tried any such silly trick, you’d have blasted them off the map. I don’t care what your finding is, you just can’t claim Central Park snowfall as representative of the whole NE US and Europe.
Monhonk Lake, NY has its own special weather, in a different sort of way. You would have to have visited there to understand why that is. It is kinda like a ‘lost world’ thing. But, I’d say more representative than Central Park, at least.
Kip
PS: Anyone REALLY interested in this idea should read the entire original paper, it is only a few pages long, including lots of illustrations and graphs — shorter, I believe, than Willis’ misdirected refutation.
available at: http://curryja.files.wordpress.com/2012/03/pnas.pdf
“For there to be a relationship, it needs to be proportional.” Ah, not quite correct. For there to be a linear relationship, it needs to be proportional. Of course, witha a noisy signal, and not an enormous amount of data, good luck finding a norlinear relationship.
Sorry Willis I must admit that “equalibralize” is a made up word that does not belong in anybody’s dictionary (I am a sinner).
I could have said: “should the exchange of heat between the Arctic- and the Atlantic Oceans become more intense than it is at the moment” – Then, if the Arctic Ocean becomes completely “Ice free” and the North Atlantic becomes cooler, it may have an impact on precipitation which may very well come down as snow. –
Well, after all the “Ice-berg” that sank the Titanic in 1912 was floating around as far south as what is equal to the northern border between Spain and Portugal – (circa). – So, there must be cold water currents flowing southwards from the Arctic.
I am quite certain it is not CO2 that gets rid of the Arctic Sea Ice and if it is not The Sunshine either, then there is just the “Gulf Stream” or the North Atlantic left to blame. – I know the Gulf Stream runs all along the coast of Norway – so why not all the way to the Arctic?
Hmmm. – I am probably still not making much sense.
Geoff Withnell says:
March 13, 2012 at 5:17 pm
Indeed. What I really meant was that the effect needs to be “dose related” in some linear or non-linear sense.
w.
Kip Hansen says:
March 13, 2012 at 5:07 pm
Kip,
You can not seriously think I am defending the use of Central Park, NY as a proxy for snowfall in the entire NE US.
I always encourage people to quote my words. Not sure where you got that impression, but it’s wrong.
I had used central park because it was the data I had. A couple of people said the reason why it didn’t correlate was UHI. So I showed a several other sites, and they didn’t correlate either.
My response was to say that people were wrong when they said UHI was the problem with central park. Clearly, other sites don’t correlate either, so UHI is not the explanation.
You accuse me of being “misdirected” … you should check your own compass before issuing those kinds of unsubstantiated allegations, you’re way off the path yourself.
w.
Willis,
‘I had used central park because it was the data I had’. Previously, you said: ‘Unfortunately, I couldn’t find area-wide data for the northeastern US, but I did find something better. This is one of the longest continuous records of snowfall in the northeastern US—the century and a half long record of the snowfall in Central Park in New York City.’ This use of the ‘I was searching under the streetlight because that’s where the light is’ excuse for using Central Park is not logically strong. Reading the original study shows clearly they are not talking about single site snowfall, but snow cover–snowy winters–over large areas of the Northern Hemisphere (in the US, northeastern and mid-west), Europe, and East Asia.
It is your refutation that is misdirected (‘…..Willis’ misdirected refutation.’ ), not you personally.
The misdirection is that you are trying to refute a study that finds:
‘The results of this study add to an increasing body of both observational and modeling evidence that indicates diminishing Arctic sea ice plays a critical role in driving recent cold and snowy winters
over large parts of North America, Europe, and east Asia. The relationships documented here illustrate that the rapid loss of sea ice in summer and delayed recovery of sea ice in autumn
modulates not only winter mean statistics (i.e., snow cover and temperature) but also the frequency of occurrence of weather events (i.e., cold air outbreaks).’
by finding non-correlation(s) with single-site snowfall records in a very small part of the area reportedly affected by the effect claimed, instead of area-wide on a sub-continental basis, which is what the study is about.
So now you have done two analyses on two data sets (Whole US and Central Park) neither of which was the subject–or point–the original study–and and claim to find the study wanting?
For other readers, the original study is at available at: http://curryja.files.wordpress.com/2012/03/pnas.pdf
Kip Hansen says:
March 14, 2012 at 3:04 pm
Thanks, Kip. You’re not following the story. So far I have looked at 1. The Northern Hemisphere as a whole, 2, Just North America, 3. Just Europe and Asia, 4. Central Park, 5. Mohonk Lake, 6. Lake Placid, and 7. Rangeley, Maine.
In not one of these is there a statistically significant relationship between autumn (SON) Arctic ice and winter (DJF) snowfall, whether detrended or not. Not one.
I have also made a call for information on some other snowfall datasets. You have not provided one.
You are also not following the paper as to which areas are supposed to be affected. The claim is supposedly valid for the “northeastern united states”. I’m not finding it. If you know where it can be found, please enlighten us. Their figure 1 shows the putative effect as existing as follows:
So I’m calling BS on your statement that I haven’t investigated enough datasets. Look at their claims! When I first looked at that, I started out by saying that with that amount of red, surely, surely it would be easily visible in the datasets on a global level … but it’s not. Not even close.
So I figured the North America dataset should surely show it … but it doesn’t, not even close. And so on down the line. All of New England is bright red, it should show up there … but I can’t find it.
Now, if you have another snow dataset that you’d like me to look at, trot it on out, I’m your man. If not, I stand by my conclusion—arctic ice cover is only a minor player in the question of winter snowiness.
w.
PS—Curiously, if they are right, it reduces one of the oft-made feedback claims. This is the claim that reducing the arctic ice reduces the albedo, and thus warms the world.
But if reducing ice area results in increasing snow area as they claim, that will tend to balance out the albedo effect. Gotta love the climate, it’s weird either way.
tokyoboy says:
March 12, 2012 at 6:51 pm
Good luck with the book. 🙂
Hope you win some converts.
DaveE.
Willis, Willis, Willis — ‘Central Park, 5. Mohonk Lake, 6. Lake Placid, and 7. Rangeley, Maine.’ all = single site snowfalls. Your article says: ‘ …it is not visible using the snow data for the whole US’ so maybe that’s what you mean by one of your first two ‘1. The Northern Hemisphere as a whole, 2, Just North America,’ . I think it’s neat that you’ve looked at ‘just Europe and Asia’ as well. In efforts to replicate a study, isn’t it important to look at the same or equivalent data? Have you actually looked at the data that Jiping Liu, Judith A. Curry, Huijun Wang, Mirong Song, and Radley M. Horton used in their study? (They say it is available at http://climate.rutgers.edu/snowcover ) Do you have reason to doubt their data? Have you asked Liu, Curry, Wang, Song or Horton for clarification?
One more thing. I object to the kind of maps shown above as their Figure 1B. Here’s the problem. When you look at the map, it says “regions within contours denote the regression above 95% confidence”.
The difficulty is, on a purely random but fairly smoothly varying map like Figure 1B, you’d expect about 5% of the total area to be a “region within contours” which is above 95% confidence.
What is usually not noticed is the underlying significance question has changed, it is different for maps.
For a single result, the relevant question is “Is this result significant at p less than 0.05”, that is to say the 95% level.
But for a map, the corresponding question is, “Is the amount of ‘significant’ area significantly different from 5% of the total area”. That’s a more subtle and harder question to answer.
So I’m willing to look at a map like their Figure 1B, but in my head I’m asking “Is the “significant’ area more than 5% and if so by how much?”.
w.
I posited some time ago, (over a year,) that reduced Arctic ice cover was a negative feedback.
The paper under discussion seems to validate that posit.
DaveE.
Willis,
Just to wrap this up….I usually admire your work. In this case, I felt you’d wrong-footed yourself with Central Park–representative ONLY of snowfall in NYC.
To address Lui et al, or any other study carefully done by seriously intelligent scientists, it is first necessary to attempt to replicate what they have done–using the exact same data sets, the exact same methods, and see if you derive the same results, then determine if their conclusions follow from those results.
Instead, you have used different data sets that measure different phenomena, applied different analytical methods, and not-surprisingly, arrived at different results. This is fine and good if you wish to present your study for perusal/review/publication to be considered alongside of Lui et al.
What you can’t do though is to claim to have done something scientific regards to Lui et al. With what you have done so far, such a claim is improper–the attempt at this claim is what I privately refer to as ‘slash and hack’ criticism, a wholly unscientific endeavor, usually seen at other venues (RC and JR come to mind).
It suits you better to either do a professional job or leave it alone.
Kip Hansen says:
March 15, 2012 at 6:05 am
I gave results for three widely separated stations in NYC—Central Park, Mohonk Lake, and Lake Placid. Not only that, but I JUST POINTED THAT FACT OUT TO YOU, saying you weren’t following the story … so now, when you repeat your incorrect statement, it is no longer an innocent error.
I have no problem being busted for what I’ve done. But stop trying to bust me for things I haven’t done.
Thanks, Kip. At your suggestion, I’ve written to the Rutgers snow extent folks requesting gridded data.
However, I disagree with your claim that to address a study one needs to first replicate it. If there are obvious mistakes, then is more than sufficient to point them out. For example, the authors appear to have ignored autocorrelation in their calculations of significance … which in itself invalidates the entire study.
In this case, they claim (see their figure 1 above) that the effect is a) very widespread, and b) at its strongest in New England.
So far, I’ve shown that a) the effect is not widespread, and b) four widespread sites in New England, which is supposed to be strongly affected, show NO AFFECT AT ALL.
You seem to think that this is meaningless. It is not. Yes, as you point out, none of this falsifies the effect.
But as Thoreau noted, “Sometimes circumstantial evidence is very strong, as when you find a trout in the milk.”
So, many thanks for your push for me to actually take the final step. I will be surprised to find anything. You comment about studies done by “serious intelligent scientists”. There are three assumptions in there, and in climate science I find that the usual case is that at least one of those three assumptions is not met.
In the current example, for instance, I find no mention of autocorrelation in their work. I also find no mention of the fact that in a mapped continuous variable, we expect to find a spurious “significant” result over 5% of the map.
Those are huge lacunae, Kip, but they are depressingly common errors in climate science. Here’s another one:
They claim they are “examining observational data”, but in fact they are doing no such thing. They are regressing observations against climate model output (NCEP2). Claiming that the output of climate models is “observational data” is deception, Kip—either they’ve deceived themselves or they are deceiving us … and neither one is a good sign.
So you want to tell me again about how these are “serious, intelligent scientists”? Because serious intelligent scientists don’t claim to use observations and then use model results. Serious intelligent scientists adjust for autocorrelation …
So I make no assumptions about the quality of the work. I’ve been disappointed too many times. And no, sometimes it’s not necessary to replicate an entire study if there are obvious errors … like ignoring autocorrelation, for example.
I’ll report back when I hear from the Rutgers folks, and thanks again for your support,
w.
Willis — Sorry that you feel hassled and or ‘busted’. I was trying to smooth things out a bit but it seems you are sort of beside yourself this late in the thread.
I am only trying to steer you to what I consider a more scientific, and thus less emotional and less ‘slash and hack’ approach, which, in the end, I fully believe would save a lot of time, unnecessary calculation, and a lot less attack-and-defend rhetoric in the comments.
If you suspect–‘For example, the authors appear to have ignored autocorrelation in their calculations of significance’–an obvious egregious error in calculation or method, the more usual collegial action would be to query the corresponding author. The matter could be settled in a quick email exchange.
Such statements as ‘four widespread sites in New England, which is supposed to be strongly affected, show NO AFFECT AT ALL.’ are not useful. In a medical study that found a widespread general improvement in a cohort of ten thousand individuals, calling out four individual’s personal results as examples of the whole would be meaningless. We have already discussed that one of the ‘individuals’ (Central Park) you have picked is already strongly suspected to be unrepresentative of anything except itself.
Your hubris and your disappointment–‘either they’ve deceived themselves or they are deceiving us … and neither one is a good sign.’– seem to have clouded your better judgement.
I recommend the Caribbean–I am on my 42′ cat in Honeymoon Bay, Water Island, St Thomas, USVI. Come for a visit, we’ll put you up in the spare cabin.
Kip Hansen says:
March 15, 2012 at 2:58 pm
Say what? I said nothing about being hassled, not one word. And I said I don’t mind being busted for what I’ve done, but not for what I haven’t done. For example, you claimed that I had only looked at two sites when I’d looked at seven and offered to look at others. So you busted me for what I didn’t do … nor did you ever apologize or even acknowledge your error.
To the contrary, you repeated your untrue claim after I had pointed out that it was untrue … and now you think I’m out of line for protesting about that?
I appreciate that you are trying to assist me, Kip, it is appreciated, and your idea sounds reasonable. However, I have been alternately ignored and flat out lied to by so many scientists that I’ve given that tack up. No matter whether they say yes or no or just don’t answer, I can’t gainsay them. They have not released their code, which would make it easy to tell what they’ve done, so I can’t even check their statements. So what’s the point in asking?
You say it would be more “collegial” to query the corresponding author … but that assumes that I can trust their word, and in the climate world, that’s a sucker’s bet. However, let me give it one more time, the “old college try” … I’ll report back with my results. No response yet from the Rutgers snow folks …
First, no way there are ten thousand snow measuring sites in New England, there are maybe fifty or a hundred.
Second, this is not a medical study. It is a claim that the snowfall is affected over the entire area, a type of claim that is rarely made in medical studies. Yes, I agree, four widely separated sites don’t prove anything … but they definitely send up a red flag. You see, although human beings show little correlation person-to-person, one dies and the next one is fine, precipitation is much more highly correlated. I may get sick and the guy next to me may not. But it would be unusual for the snowfall records to be uncorrelated in that manner. That’s why all of New England is red, and it is why you never see that kind of thing in a medical study.
For example, the correlation 1979-2011 of Central Park (DJF) with Mohonk Lake is 0.70. The correlation of Mohonk with Rangeley, Maine is 0.57.
Not only that, but the correlation of Central Park snow amount with North American snow area is 0.49, and that of Rangeley amount with North American snow area is 0.52. Mohonk is a bit lower at 0.33, but still respectable.
So your medical example is not a parallel at all, because people are not correlated like snowfall is. As you point out, it’s certainly possible I picked four of the stations in the area that are NOT correlated with New England snow area … but given their high correlation with each other and with North American snow area, the odds are not good that somehow I’ve picked four total outliers.
Again, I fear you read things into my words that are not there, and you misquote me in the bargain. What I said was:
So is your point that claiming that the output of climate models is “observational data” is not self-deception, but in fact is good, solid scientific practice?
Or do you agree that they are deceiving themselves and others by calling model output “observational data”?
In either case, I’m not “disappointed” by that in the slightest, I’ve been playing this game far too long for that. Actually, I’m not even surprised in the slightest that they can’t tell climate model output from observational data, that’s far too common these days to even merit comment.
Can’t tell you how much fun that sounds to this landlocked sailor, and I can’t thank you enough for your most generous offer, Kip. It would be great, perhaps some other time, I do love to sail catamarans.
But my next tropical excursion is already scheduled, and it is going to be for surfing purposes, so regret greatly that I have to pass until I can build up more vacation time.
All the best to you,
w.
Kip Hansen says:
March 15, 2012 at 2:58 pm
To which I replied:
I’m back to report that, as I said I would, I sent the following email to the corresponding author on the 15th:
So far … crickets.
w.