the Air Vent

Because the world needs another opinion

Spatially Weighted Antarctic Temperatures – No Peninsula

Posted by Jeff Id on April 18, 2009

This post is also running with some improvement at WUWT.

What happens to Steig et al’s warming when you divide Antarctica into two distinct climate zones?


From this post and others I’ve determined that the temperature trend in the RegEM versions of the Antarctic are not entirely created by smearing of the peninsula station’s data.

In order to interpret the RegEM results from the previous peninsula free reconstruction, we need to see a baseline reasonable reconstruction without the peninsula. These trends are based entirely on the surface station data. I saw several questions on WUWT about the improved accuracy of satellite temperatures. The satellite temperatures in this paper are of a different type than UAH or RSS use and these are affected substantially by clouds. The result is a much noisier and less trustworthy dataset than surface measurements.

In my opinion this sort of thing is about the best we can do in determining a total trend for the Antarctic over this timeframe. There are a few tweaks which might help but beyond that we have to accept that we don’t know any better than this method shows.

Now removing the peninsula does have basis in science because the ultra thin strip of land is primarily dependent on ocean temperatures and currents. It will be seen as cherry picking because I’ve clipped the part of the Antarctic warming the most. Before TCO or someone points out that I wouldn’t clip it if it didn’t have warming, keep in mind that I show it both with and without the peninsula and I make no claim that clipping the peninsula is the preferred method. It does make some sense though.

First the full trend.


Figure 1

Spatial trends with clipping region shown in black Figure 2.

Figure 2

Figure 2

As I’ve shown before, the trends from 1967 onward.

Figure 3

Figure 3

Spatial distribution 1967 onward.

Figure 4

Figure 4

If I’ve learned anything from all these plots, it’s that the Antarctic isn’t warming at 0.12 +/- 0.7 C/decade. It just isn’t. The actual trend is much lower than that and since 1967 it has even dropped a little across the continent.

Now from the other reconstructions we have the following.


Compared to the 0.12 that Steig et al. cliams the real trends are pretty low. When the peninsula is removed in the properly weighted reconstruciton presented here the trend drops by(0.52-.39)/.52 x100 =25%. This represents a very large contribution to the average simply sbecause this tiny area has shown so much warming. It doesn’t seem reasonable to adjust the continent upward 25% based on this little strip of land.

Surprisingly, in RegEM the trend changes by an similar amount (0.108-0.074).0.08 x100 =31%. This is impressively similar to me as many have speculated that the positive trend in RegEM is created by smearing of peninsula trends. In my spatial reconstruction above, there is no smearing of the peninsula trend at all and there is a 25% trend drop when the peninsula is removed. This means to me that the RegEM high positive trend is not wholly created by the peninsula. Also, this does not mean the trends aren’t smeared by RegEM, they are.

26 Responses to “Spatially Weighted Antarctic Temperatures – No Peninsula”

  1. Ryan O said

    I think we’re beginning to figure out what is causing what.

    The spatial smearing does not seem to be as much a function of RegEM as it is a function of reducing the continent to three PCs. The three PCs do not provide enough spatial resolution, resulting in the peninsula warming being spread across the continent. RegEM then uses the resulting PCs to infill the pre-1982 portion of the PCs and ground data based on this already smeared result.

    As I alluded to in my post on your Warming the Raw Sat Data topic, I think the fundamental problem is reducing the continent to three PCs. This appears to provide support for that.

    Also, as we should expect, by any method, the average trend for the continent is positive from 1957-2006. The 1950’s were an unusual low for the continent, it should surprise no one that the trend starting from there is positive. The big deal is whether the continent continued to warm after recovering from the 1950’s low. According to Steig, it did. According to your analysis, it did not. (Actually, according to Steig’s own AWS reconstruction, it did not.)

  2. Nicolas Nierenberg said

    I’ve enjoyed following this.

    Just a note on your first graph. “Using” doesn’t have an “e” in it.

  3. TCO said

    Ryan’s hypothesis sounds like the one that we talked about on RC ages ago. I think its important to know what is doing what. Since we are really investigating an algorithm. That is why I hate smearing andf conflating and responding with issue B when A is in discussion. Or when discussing Aa dn B ambiguously so that it is not clear if they are each flaws on own or only a flaw in AB combination.

  4. Ryan O said

    Also, Jeff, your comment on the AVHRR data is apt. As used by Steig & Co., the AVHRR data is not taking equivalent measurements as ground instruments. You’ve brought this up in the past. If the data sets are not measuring the same quantity, then splicing them together is likely to give erroneous results.

    However, if (big if) the nonlinearities can be properly accounted for in a general calibration, then it would be appropriate to use the data. Shuman was able to do this with the 37GHz data. I’m continuing to try to see if I can do it with the AVHRR data. I’m fairly close, BTW.

    If we can perform an algorithm calculation of the AVHRR data (without feeding in any station locations) and then show that the difference between the calibrated data and ground station data is trendless, then we can conclude that they are measuring the same quantity and we would be justified in splicing them.

    If we get to that point, then using the satellite data for a reconstruction would theoretically be valid (assuming the tools we would use – RegEM and PCA) are up to the task.

  5. Page48 said

    I’ve really enjoyed following your work on this. Thanks

  6. Carrick said

    Jeff, as you are probably aware, global climate models suggest that prior to 1980, the dominant source of warming was natural (human generated warming from CO2 was nearly offset by cooling from anthropogenic sulfate emissions).

    If you were to break Figure 1 into 2 episodic periods: 1960-1980 and 1980-current, it looks to me like the period where AGW should be dominant, the central part of Antarctica actually experiences cooling…


    Ryan’s hypothesis sounds like the one that we talked about on RC ages ago. I think its important to know what is doing what. Since we are really investigating an algorithm. That is why I hate smearing andf conflating and responding with issue B when A is in discussion. Or when discussing Aa dn B ambiguously so that it is not clear if they are each flaws on own or only a flaw in AB combination.

    No, we’re investigating a data set. If a particular algorithm is giving an atypical result from that data set, either there’s a reason why it’s better than other approaches, or there’s something wrong with applying it to the data set.

    Either way, it’s important to understand why this atypical result is being produced.

  7. Carrick said

    Had to add my 2 cents
    , sorry. Even if the Antarctic has really been cooling since 1980, that isn’t counter evidence that the Earth’s isn’t warming, nor that humans aren’t playing a role in that warming.

    Fundamentally the problem (to me) is the models are very limited in what they can reliably predict on a regional scale. (If you look at model versus data, very often the data are 180° out of phase with the models.) That doesn’t mean the underlying physics isn’t well understood, just for example that the problem is complex enough that the current generation of computers can’t numerically treat the problem with enough detail to give meaningful predictions…

  8. Jeff Id said

    #7 “Had to add my 2 cents”

    No problem thats what tAV is for.

    The 1982-current in RegEM is actually all satellite data. Visually in the spatially weighted reconstruction you are right though. It looks like a downward trend.

    It will be interesting to redo this when Ryan gets a working recalibrated dataset.

  9. Ryan O said

    #7 Agree with Jeff. The important thing about the stuff that Jeff has been doing (in my opinion) isn’t so much deciding whether Antarctica is warming/cooling/neutral, but how valid it was to use this particular set of mathematical tools in this particular context. If they are not valid, it is important to show that they are not to hopefully prevent the misuse of the tools in the future.

  10. TCO said

    Well no duh. But he sure likes to jump up and down with his rattle before actually learning which features do what.

  11. Fluffy Clouds (Tim L) said

    snow fall here in Michigan next few days….lol

    Jeff, I hope there maybe a way to use your knowledge with these tools to Reconstruct
    WUWT work for the CRN 1,2 vs 3 vs 4,5. I just think there might be something to this.

    BTW It looks to me like the accuracy is better W/O the peninsula added

    Great work!!!!

  12. Geoff Sherrington said

    I do not know if I am light years behind the action or not. From the beginning I was worried that satellites could not give a surface temperature below cloud. I have confirmed that they cannot. Cloud is always present in the Antarctic. Therefore, three problems have to be solved. 1. Discriminate cloud cover from none and 2. Work out the surface temperature under the cloud and 3. develop an algorithm and calibrate it, tp cope with the movement of cloud wrt satellite passes.


    “CASPR is research code. It does a lot of things well but doesn’t do anything perfectly. Many of the algorithms have been validated, some have not. They are all detailed in the Reference Guide. There are three broad problems worth mentioning at the outset. First, everything depends on cloud detection, which sometimes borders on being as much an art as a science when working in the polar regions with the AVHRR. We are, indeed, trying to squeeze water from a stone. We do not claim to have solved the cloud detection problem, but rather provide methods that work reasonably well most of the time. Second, there are fundamental difficulties in estimating some parameters at extreme viewing and illumination geometries common to the high latitudes. We’re venturing into territory that is difficult to model so beware. Third, the retrieval of cloudy sky parameters requires temperature and reflectance values underneath the clouds. CASPR interpolates clear sky values to cloudy areas. This generally works but can result in large uncertainties in very cloudy areas. Cautionary notes are given throughout the Reference Guide. Please do not ignore them!”

    I do not see how any amount of statistical treatment can be precise when so much of the input is a guess. I cannot even get a straight answer from the Antarctic on whether cloud cover warms or cools the surface; or whether the process is time-variable in a way that cannot be reconstructed.

    Can you please briefly educate me on why so much time is being spent on tiny changes to d(noise + guess)/dt?

  13. Ryan O said


    As far as I know, the CASPR algorithm is not used by either Monaghan or Comiso. Each of them have developed their own methods for dealing with cloudmasking which may lessen the impact of clouds. Comiso’s provided data set matches ground stations fairly well. He did not correct for surface reflectivity changes enough, however, and he did no corrections for satellite offsets ( ). Still, even with that, the satellite offsets are less than 0.3 Deg/C, and a first pass at non-linearity corrections changes the overall trend in the areas where surface stations exist by only about 5 percent.

    While cloudmasking (and the resulting infilling) is difficult, I would not call the results completely unreliable. In Comiso’s case (which is what we are discussing here), the results are actually pretty good. They’re closer than UAH/GISS or even RSS/GISS, for example. After having played with the cloudmasked data for some time now, I do think that it is, in principle, possible to splice ground and AVHRR data for a reconstruction and obtain quantitative results.

    To be quite honest, Comiso’s data matches well enough to the ground that I would put more trust in splicing AVHRR with ground than I would place in the GHCN network splicing different instruments at surface stations for input into GISS and CRU.

    With that being said, I think some ground rules need to be established for this type of analysis. One could imagine a “checklist” of items that should be analyzed and accounted for prior to splicing data sets for this purpose. There are several big-ticket items that appear not to have been done (especially the satellite offsets).

    Along with that, the conclusions need to be appropriately moderated by the uncertainties in the data. In the case of Steig et al., many of the uncertainties were either not considered at all or simply not documented in either the main paper or the SI. That practice should stop.

  14. TCO said

    It seems like one ought to be able to see how cloud temp varies versus uncloud temp by a surface measurement (would assume warmer at night and colder in day and stuff like that, but I guess level of clouds matters, also just modes of weather correlated with clouds.) One could then look at if there are trends of more/less clouds over time, or with certain weather patterns…and what that means.

    I would think the benefits get pretty lost though, if you make everything into 3 PCs…

  15. Jeff Id said


    I’m not confident in the trend of the satellite data. Comiso did a good enough job that correlations are amazing compared to the NSIDC form but the trend signal is too small to reliably extract. Since correlations are reasonable I thought it might be possible to reasonably distribute surface temperature data on the sat data using it as a matrix for correlation, while the trend was reasonable the first reconstruction attempt resulted in unnatural correlation/distance.

    I loose interest in broken toys pretty quickly, but in this case there are a few more things to do related to RegEM. Now that I get what’s going on, I’d like to explore RegEM a bit more.

    I’ve said it so many times now but I’m still amazed at how little confirmation was done to see if the math works.

  16. Fluffy Clouds (Tim L) said

    second, reading/thoughts are, you may indeed have a valid number here 0.041/.4C per 100years.
    you believed that (earth) warming was about 0.7C per 100 years.
    assuming MMGW IS NOT present in the Antarctic, and doing the math we have .7 – .4 or 0.3C/100years……. that I would totally believe!!!!! CO2 forcing COULD be that amount.
    Lief has said solar is about .1C so that gets a range of .2 to .4C now man made CO2 is about 10% would get us to 0.02 – 0.04C/100years.
    I believe this as a possibility that man could induce a 0.04C world temperature.

  17. Geoff Sherrington said

    Jeff Id,

    Thanks for your comments on reliability of satellite obs. I take your point about broken toys and wish you well in the EM dissection. When you say that Comiso did a good job, do you have an objective measure to judge against? A good correlation coefficient is a necessary but not sufficient finding.

    My quote from CASPAR was to suggest that similar intractible problems exist for any current method of cloud correction. Was taking mechanisms rather than badges.

    Hi again TCO. Remember that in the Antarctic the sun does not appear for a while each year. Climatically, it’s not a place for intuitive thoughts based on habitated climates, as I think you conclude.

    I’m still wondering what it is that Dr Steig claims to be measuring. Is it the temperature of the Antarctic ice mass, its temperature to a certain depth which might be 1m or 1 mm, the air immediately above the soil/gas interface (which need not be a simple surface, as in driven snow cases), the air in the lower atmosphere to a nominated height above sea level or above ground level – or have I missed a possibility? At a tenth of a degree C!

  18. TCO said

    He’s taking the sat measured temps and extrapolating back in time. I think the satts measure something pretty close to truse surface temp. Given that temps are almost always below freezing, the air temp and very top of the ice are at same temp. The satts are different from the RSS-UAH type satts that measure the lower atmosphere. These satts in Antarctica measure the actual ground temp.

    Warning, my knowledge is all from reading denialist blogs.

  19. Jeff Id said


    I believe you are correct about the intractable problems in the sat data. After I saw the NSIDC version of the data, thanks to Jeff C and realized CASPAR had ‘huge’ noise levels. I’d read the same paragraph you quoted and was in the mood to chuck the whole thing but Comiso’s version had decent correlations to surface stations and a reasonable decorrelation vs distance. This made me think I could do a reconstruction by distribution of surface station data according to correlation and high order RegEM. No dice, the trend is ok and IMO my method is better than RegEM but the quality of the sat data wasn’t sufficient to constrain the surface station information. I want to go back to it one more time to try some different things.

  20. Carrick said


    The 1982-current in RegEM is actually all satellite data. Visually in the spatially weighted reconstruction you are right though. It looks like a downward trend.

    Thanks for the gentle reminder about when the RegEM analysis started and ended.

    How do these researchers not understand basic facts about climate science?

    How could they not know something as basic as when (based on climate models such as GISS) net anthropogenic warming started?

    Showing there is an upward trend prior to 1980, that was then followed by a decrease in temperature makes none of the points they want to make.

  21. Carrick said

    Jeff, I had one followup question to this comment:

    The 1982-current in RegEM is actually all satellite data. Visually in the spatially weighted reconstruction you are right though. It looks like a downward trend.

    Are there satellite data that extend to the surface? I checked RSS and it only extends to TMT for the Antarctic main continent. (TLT is gray…e.g. no data.)

  22. Jeff Id said

    The data used in this reconstruction is IR emission (TIR) which is different from the microwave data from the lower troposphere used in TLT by UAH and RSS. It’s a heck of a lot noisier too, because it can’t sort out what is cloud and what is ground. The whole continent is covered though.

  23. Matt Y. said

    “Showing there is an upward trend prior to 1980, that was then followed by a decrease in temperature makes none of the points they want to make.”

    They got the Nature cover with “Antarctic Warming” in big letters and a graphic showing the entire Antarctic in bathed in red. Mann got to make his sound bites stating that the Antarctic wasn’t bucking the global warming trend and another skeptic’s cannard had been put to bed.

    Given that 99.9% of the people don’t want to know the details, I would say it was mission accomplished for the alarmists. Welcome to politically biased science.

  24. Carrick said

    Thanks for the answer, Jeff.

  25. Jeff Id said

    #24 No problem. I don’t want to give the impression of disinterest, slow responses are never due to not giving a crap. I am a really insanely busy person, besides being dad, one of my “hobbies” is pres. of a company so it takes a lot of time.

  26. Geoff Sherrington said

    Jeff Id,

    You now this already, but the winds in the Antarctic can be ferocious and the surface obscured by high velocity wind-driven snow and ice. So what the satellite IR system sees at a particular time might have acquired its temperature some grid cells away from where it is observed and gained or lost heat in transit. The “apparent” relocated grid cell temperature might not equal the true surface temperature. I have no quantitative measures to help you, I’m sorry.

    In any case, the explanation for cooling by the British Antarctic Survey related to the ozone hole affecting wind circulation and drawing cold winds over the surface. My fundamental question is how the very coldest winds of the Antarctic get even colder in the presence of Global Warming. The BAS inferred quite strongly that wind temperatures – and these are what is measured by ground stations – were the cooling agent. There was little reference in what I have read to the solid surface either cooling or warming under wind influence. More noise, less easy to predict melting.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: