Climate Audit

Stuart Harmon

once upon a time you dressed so fine
now you don’t talk so loud
how does it feel to be on your own.

Sorry watching BD at Newport

So your are a statistician I am an engineer so please give me a chance:-

Lets say the spread between the maximum and minimum temperature in the Antarctica is 55 degrees in one year?

From the analysis you have provided above the trend/ is o.o2 degrees per decade.

Lets say the maximum spread one day is, my guess 20 degrees.

Assume there are 10 temperature measuring stations.

My guess is the error would be 1 degree ie 5%

So this is not testing the strength of steel the variables are huge.

So my guess is that Professor Steig could be wrong by a factor of 50 plus or minus.

Sorry about the guesswork but Mr GS thinks guessing is good.

If the above is total garbage please don’t post.

Joel

Why? It showed exactly what they expected. I wouldn’t doubt that someone looked at Harry, said “Eureka!” and proceeded to build a paper around it.

Alan S. Blue

There’s a station named Doug that’s practically on top of Harry since ’94.

Are all of these newer stations failing to meet muster due to being more recent additions?

Hugo M

Within “ftp://amrc.ssec.wisc.edu/pub/aws/10min/rdr”, you will find a file “readme.updates”. Looking there for Byrd reveals:

04/02/03 — Data for Byrd (8903) were replaced for March, April, November and December, 2002. The wind direction was corrected.

03/28/03 — Data for Byrd (8903) were replaced for November and December, 1999 and January – September, 2000. The wind speed and direction were not correct and were eliminated.

Concerning Harry:

02/19/04 — Data for Harry (8900) have been added for July, 2000. The file for July actually contained August data.

After some date, some stations, apparently including Byrd have been equipped with a dT sensor. This could be the third temperature sensor sometimes visible on some of the photographed stations, mounted below the big electronic box. “readme.format” says, that the dT signal, if present, is in the 8th column in raw data files

Besides low temperature variance, this could give another hint to stations buried in snow. At least it will clearly mark the periods, when the dT sensor had been buried. Probably there is distinct pattern.

“Readme.format” also contains hints on data erroneously attributed to two stations. If detected, they provided two files with correct station ids, both containing an “x” somewhere in the filename. Possibly the guys condensing these raw data files into antarctic READER monthly means simply missed some of these?

10/01/02 — The data labelled 89110799.r (Gill) is actually for 89120799.r (D-57). The labels have been corrected and Gill for July will be added.

Mike Smith

Also FWIW, note that in the Ferrell Site report of 1984, the laboratory calibration table has a delta of 0.75 C between AWS and MEASURED.

Ferrell Site Report 1984

bender

The adjustment dramtically shifts the geometric centre of the warm patch, from “west Antarctica” to just the peninsula. That is not a trivial change. It’s the difference between “continental-scale” warming and “local” warming.

Of course, either is consistent with GCM predictions … because nothing is inconsistent with them. They’re irrrrefutible.

• Jon
  Posted Feb 6, 2009 at 10:12 PM | Permalink

  Re: bender (#19),

  snip

  Stratospheric warming, rapid sustained warming on the order of several degrees C in years, monotonic warming, etc. are all inconsistent with the AR4 runs. There are enough legitimate issues with the models that one doesn’t need to invent them.

  snip – enough bickering on both sides

• Jon
  Posted Feb 7, 2009 at 3:07 AM | Permalink

  Re: bender (#18),

  I understand the snip, I guess [Psst- I don’t at all. You’re letting a comment stand that is the essence of what I described], but:
  
  [Steve: foodfights bore me; if I miss something, email me or put a pointer rather then continue it. If I left some foodfight in, it’s not intentional.]

  bender, if you’d like to get in touch with modelers to improve process,es I’m happy to help.

  There is a big target to hit and you’re swinging at bystanders.

Ryan O

I’d be interested in your playing with the 1979-2003 period, too, Steve. I’m curious if it may be a case of cherry-picking endpoints to produce a neat looking graphic.

Edouard

Since the real world gets colder than the best IPCC-guesses we need new headlines. This might be just as scientific as cherrypicking for any warming anywhere on earth. Am I the only one who is a little bit disappointed. I thought there was more real science behind global warming :-(((((

Louis Hissink

Steig”The mean trend for Antarctica changes by less than 0.004 degrees C/decade”

We have instrumentation to this resolution?

This is simply BS!.

Louis Hissink

Just a totally, left field thought, but if the ice caps did melt, (only at the south pole), then isotatic requilibriation of the crust would probably mean a zero change in you know what.

This surreal ideas occur among geographers, occasionally when wandering into weather stations, to add meaning to their post modernist obfuscations. In other words they haven;t a freaking clue.

Lucy Skywalker

Looking back to the 2004 NASA paper of Gavin Schmidt and Drew Shindell, the 2009 paper seems to be very much the result of a self-fulfilling prophecy stated in the earlier paper, a Global Climate Model showing Antarctica warming, alongside a standard map of Antarctica cooling. It seems as if Schmidt et al filed away the thought “the Antarctic data must be faulty somewhere” because they all believed in Global Warming. So when data surfaced that appeared to validate warming for Antarctica too, they were naturally keen to publish.

Now consider a more subtle picture, with several factors at work
(1) the planet has, at least from 1970-2000, been warming overall, and the resultant warmer ocean currents have poured in around the appropriately named Racer Rock and have warmed the largely maritime Peninsula
(2) volcanic activity, known in the area, may have contributed
(3) due to the slight lessening of cloud cover in this time frame, while the rest of the planet has been warming, ice-sheets have higher albedo than clouds and tend to cause cooling
(4) dry air helps the cold spikes of winter temperature when cloud cover abates
(5) these winter spikes could have been removed from the record as being too “anomalous”
(6) the majority of the planet has been warming, and some of this rubs off on Antarctica
(7) even with corrections, temperatures may have overall stayed pretty level – but – NO sign of the huge warming predicted by the models.

I follow the unearthing of data to confirm, or challenge, with interest. What seems to matter is the patterns, not just a single trend with the one-bullet headline “ANTARCTICA WARMING”. Perhaps Steig’s paper will eventually show that there has not been so much cooling as many skeptics have been wont to believe. But also, the expected drastic warming seems significant in its absence.

• Andrew Parker
  Posted Feb 7, 2009 at 12:45 PM | Permalink

  Re: Lucy Skywalker (#37),

  This brings to mind a statement by a climatologist who proposed, many years ago (and for this reason I cannot recall his name), that global warming might trigger a new age of glaciation, not to be confused with a new ice age. Would this relate to the statement that “all weather is local?”

wmanny

[#27]

I tried, unsuccessfully, to post the following at RC. Gavin, who often does engage, has not done so this time, though perhaps he is not on duty at the moment.

“People whose only contribution is to malign or cast aspersions, whatever their background, are simply of no use in that endeavour.”

In context, it would seem you are referring to Steve McIntyre. Has he been of no use to the scientific endeavor in this case?

Cassandra King

” hordes of know nothings”? How many know nothings does it take to make a horde I wonder? What is his definition of a ‘know nothing’?

I wonder why a scientist would refer to another person as a ‘know nothing’? The term puzzles me greatly and I find it worrying that a highly educated person would use such a term.

Knut Witberg, Norway,

snip – policy. sorry bout that

Steve McIntyre

#40. I think that the climatology test was a screen applied to AVHRR only.

Hu McCulloch

Re # 39, 40, 46,
I think Steve is right, that the “climatology test” was supposed to be applied only to the satellite readings, which are known to have problems with cloud cover.

The Methods section at the end of the paper discusses both AWS and AVHRR data in its first paragraph. However, only the first 4 sentences pertain to AWS, and the remaining 7 to AVHRR. The last 2 sentences state, “We make use of the cloud masking in ref. 8 but impose an additional restriction that requires that daily anomalies be within a threshold of +/- 10 °C of climatology, a conservative technique that will tend to damp extreme values and, hence, minimize trends. Values that fall outside the threshold are removed.”

That indicates they are referring to AVHRR, not AWS.

The first section of the SI refers exclusively to AVHRR data. Its first paragraph discusses this threshold further, stating that 10°C from the climatological mean was used bececause it gave a better fit than 6°C. The following SI section on the AWS reconstructions makes no reference to truncating outliers in the AWS data or even in the AWS reconstructions.

This doesn’t mean that they didn’t truncate the surface data as well as they satellite data. In order to get Figure S3a, they must have. They just “fixed it”, as Phil puts it, without mentioning it in either the text or the SI.

Peter Hartley

Re # 48

If one truncates the satellite data within a threshold of +/- 10 °C of climatology, it is not clear to me that would not bias the temperature trend calculations. In particular, why should the effects of clouds on the AVHRR data (which is the supposed motivation for truncation) be random? For example, if the clouds predominantly lead to colder readings, the truncation could remove the colder readings. Then a negative trend in cloud cover (fewer clouds over time) would be read as a positive trend in temperatures. More generally, trimming the data in this way might appear symmetric but the effect need not be neutral if the underlying data is skewed in some way and the skewness changes over time.

Gary A.

LINK

“Ferrell site was chosen due to its long climate record of thirty years. There is also
known accumulation to occur at this site from informal observations over the years by AWS
support personnel who have had to repeatedly raise the tower to keep it from being buried.”

“Nascent and Ferrell are located on the flat expanse
of the Ross Ice Shelf, affected by little topography.”

“Ferrell has a higher number of precipitation events
than any other site.” (in their study)

Doc_Navy

Steve,
Just wondering where my last post went?… Has anyone superimposed a map of active Antarctic volcanic locations over Yours and Roman’s temp trend reconstuctions? Watts has a map on his site that when dropped over the AWS data collection sites shows some interesting correlations.
Particularly:
AWS Site: Ferrel———-> Mt. Erebus
AWS Site: Racer Rock —-> Jun Jaegyu, Deception Island
AWS Site: Butler Island –> Nunatucks Group
Doc
PS. If I missed it somewhere, or the comment was moved to a different thread I apologize for restating something already addressed.

Steve: If you want to discuss volcanos, please do so at Anthony’s. Editorially, I don’t want to get involved in this, as I’ve mentioned from time to time. I sometimes edit when I’m tired and I may have pulled a trigger on your post, rather than saying the issue.

David Smith

It appears that the Ross ice typically drifts away from the Pole towards open water at a rate of about 0.5 to 1.0 km per year, per a Wikipedia article. It seems reasonable to think that Ferrell has drifted perhaps 20km towards a warmer clime in its existence. The effect would be particularly noticeable if Ferrell is getting close to the ice edge and the (relative) warmth of open ocean.

If the above is correct and Ferrell has moved then it seems like a poor candidate for a climatology study.

• captdallas2
  Posted Feb 7, 2009 at 3:52 PM | Permalink

  Re: David Smith (#60),

  It appears that the Ross ice typically drifts away from the Pole towards open water at a rate of about 0.5 to 1.0 km per year, per a Wikipedia article. It seems reasonable to think that Ferrell has drifted perhaps 20km towards a warmer clime in its existence. The effect would be particularly noticeable if Ferrell is getting close to the ice edge and the (relative) warmth of open ocean.
  If the above is correct and Ferrell has moved then it seems like a poor candidate for a climatology study.

  From the few people I have talked to that have been down there that is a normal drift rate. That would complicate the study but I don’t think make it a poor candidate. If you don’t allow for site migration it just gives SM another chance to raise a controversial point.

Knut Witberg, Norway,

Unfortunately, the debate about subjectivity is not familiar to sciences like physics and mathematics, probably because it is seen inconceivable that subjectivity problems could occur in the “pure” sciences. We must, however, admit the problem of subjectivity is equally applicable to all sciences, not least the climate science.

Records show that every time data errors are found by the proponents of AGW, the corrected data supports the AGW.

This might very well be the result of manipulation, not consciously by a human or a group of humans, but manipulation by the system itself. Too many climate scientists confirm that an AGW skeptic has reduced chances to get the best jobs, grants etc. and that the subjects are selected in a partisan way. The big, international conferences have a certain resemblance with religious meetings.

An increasing number of people are seeing that this might be a big problem; below a link to an excellent article that points out that the corrections we have seen through the last few years in climate science may represent a biased selection of possible corrections.

http://online.wsj.com/article/SB120847988943824973.html?mod=opinion_main_commentaries

bfwebster

Steve:

Maybe this explains some of the problems with Antarctic monitoring stations. 🙂 ..bruce..

David Smith

Re #64 Hello, Captain! I bet the Keys are fantastic this time of year.

As best as I can tell, Ferrill is about 60km from the nominal edge of the ice shelf. I say “nominal” because Ferrell seems to be near an iceberg calving region. If Ferrill was 80km away from ther coast and is now 60km away, and is approaching substantially warmer open ocean, then I think some amount of warming could be happening. Is it big, small or middle-sized? No idea.

David Smith

Here’s a map of Ferrell and his neighbors –

Ferrell looks to be 40 to 70 km from the nominal edge of the ice sheet (ignore the brown dots for the moment).

And, here’s an interesting satellite image from 2000:

This shows the birth of iceberg C-19 in 2000. I marked the top map with brown dots which approximate (as best as I can) the place where the iceberg separated from the ice sheet. If the dotted line is correct then Ferrell went from about 60 km inland to 30 km inland in short order.

(Side note – Ferrell’s proximity to places named “crevasse” make me wonder if the region is fractured, with occasional exposed water interspersed with the very thick ice.)

Alan Wilkinson

I’ve had a look at raw Uranus, Limbert and Ferrell and the first two are short series and all three have a lot of missing data.

I did a simple-minded trend calculation by looking at each month compared with the median value for that month (so converting the data to monthly anomalies) and then plotting the average of those monthly anomalies for each year.

Two things stand out.

Uranus and Ferrell have large positive trends (0.22, 0.11 deg C/yr) but Limbert has a large negative trend (-0.17 deg C/yr). I’m puzzled that this can be converted into a positive trend by Seig. I think it could only be done by discarding the last 24 months of data (from Feb 2007).

Ferrell especially has a much greater standard deviation for winter month temperatures than summer. The others show a much less marked but somewhat similar pattern.

anna v

Ross McKitrick:
http://www.climateaudit.org/?p=5180#comment-326073
The thread “sweetness …” is frozen to comments, but I thought that it might interest you to know the answer to your request to Jack:

“I need to find if there is any place in the IPCC report where it was checked if models can reproduce the spatial pattern of temperature trends on a gridcell basis over the post-1980 interval. I don’t mean just 2 fuzzy maps put side-by-side, I mean an actual statistical test. From my reading no such test is in there, whether for model E or any other. But if you have any references to the IPCC text or other journal articles I’d be greatly obliged.”

From: 8.1.2.2 Metrics of Model Reliability
The AR4 report to be found in http://www.IPCC.ch
There is no possibility of a statistical test because the IPCC acknowledges that no likelihood function has been established for these models.

The above studies show promise
that quantitative metrics for the likelihood of model projections
may be developed, but because the development of robust
metrics is still at an early stage, the model evaluations presented
in this chapter are based primarily on experience and physical
reasoning, as has been the norm in the past.

An easy illustration is the albedo parameter. Using the toy model in http://junkscience.com/Greenhouse/Earth_temp.html
a 3% error on the albedo, a=.31% T=15C , a=.30% 16C, a=.32 T=14C. Think of the number of parameters entering. The spaghetti graphs would move all over the phase space if there were an attempt to calculate true errors .

Hu McCulloch

Retraction Re #39:

I tried fitting a trend to the original Steig et al AWS recon for Racer Rock using all of 1957-2006, and to my surprise, the trend actually turns out to be positive, despite the spurious negative outliers in 2005-06:


(Click on graph for bigger view.)

The slope is .016°C/decade, and is actually significant: The OLS t-stat is 4.4, but the first order serial correlation of the residuals is 0.196, making the simple Quenouille-adjusted t-stat (multiplying by sqrt((1-rho)/(1+rho)) 3.6. My eyeball told me that those final negative values must have so much leverage that they would pull the trend well into the negative range, but I was clearly mistaken.

The trend for 1957-2003 is of course higher and more significant: 0.022°C/decade, OLS t = 7.5, rho = 0.347, adjusted t = 5.2.

Even though highly significant, these trends are very small: +1.6&degC/millenium and +2.2°C/millenium, respectively, so that neither could be contributing much to the Steig team’s conclusion of +.17°C/decade in W. Ant. in their Figure 2 (for TIR, and slightly higher for the average over AWS recons shown with their dotted line).

David Smith

Re #89 tty, thanks for catching my dating error in #75. C-19 separated in 2002, not 2000.

Here is a map of the current (8 Feb) temperatures near Ross Island. I’ve marked the island with red splotches and the as-drawn ice shelf edge with green dots. Temperatures are degree F.

The warming as one approaches the ocean is apparent. This contrast is even greater in austral winter, I believe (it’s currently austral summer, of course).

tty

Uranus Glacier is a simple scribal error, it should read:

Jan 00 Station: 8920 Uranus Glacier

As is also indicated by the file name. This isn’t exactly the first time in history that a sloppy “0” is transcribed as a “9”.

As to why the Ferrel data from 2002 to 2006 have not been used, I also wonder (se post 88).

tty

Re 99:

That is very likely a correct explanation. There is usually a polynya off the western part of the Ross shelf for much of the winter.

Willis Eschenbach

Perhaps someone could explain why the READER data is so different from the GISS data? As a brief reality check, I just downloaded the “Amundsen-Scott” station records from both READER and GISS. The monthly difference between them has an RMS size of 0.25°C. The range of the monthly differences goes from +2.7°C to -1.7°C.

As an additional oddity, the differences are concentrated in January. This is important because when the monthly anomalies are removed, the January READER anomaly is 0.4°C larger than the GISS January anomaly. This means that when the anomalies are removed, there is an additional 0.4° difference between the two every January. Thus the removal of the anomalies, rather than reducing the error, actually increases it.

Next, from late 1993 to late 2002, the two datasets are exactly identical … why during that time and at no other time?

Does anyone have any insight into the differences between the GISS and READER records? I can’t find anything on the READER site that says they are doing any post-processing or adjusting …

All comments gladly accepted,

w.

tty

Re 101

That Racer Rock reconstruction seems to be a rather nice demonstration that RegEM decreases variance. You can actually pick out the parts that are real data by eye.

David Stockwell

Spot on Willis, as usual.

Robinedwards

This is all very fascinating, but I have a couple of small technical queries. In #101 Hu McCulloch’s graph shows “Temp Anom”. In what respect are the temperatures anomalous, please? Or, what is the basic value on which they are based?

I ask this because in my analyses of climate time series I like to work with what I call “Monthly differences”, meaning the difference between a given month’s value and the average value for the month in question over the whole of the data set. In other words, I am using a simple “deseasonalised” value. This has the desirable effect of reducing the apparent and very loud noise due to typical seasonal differences and thus improving the precision of the analysis.

Willis (in #102) also uses the term “monthly difference”, but I think he’s referring to the month by month difference between the two sources of the Amundsen-Scott data, READER and GISS. “Monthly Anomalies” is also used in #102.

I just want to be sure about these things so that I can make a reasonable try at reproducing the analyses.

A further point – this time on analyses of time series – is about the Quenouille-adjusted t-statistic., given by Hu. Is this adjustment highly regarded by time series specialists, or is it looked upon as something of a “get-out” to avoid having to go through the more complex arithmetic necessary to identify the appropriate time series model and then use full-power specialised software requuired to run the computations? I don’t have such software as yet, so greatly favour the Quenouille adjustment!

Robin

Hu McCulloch

RE Robinedwards #110,

Good questions —

These “anomalies” are computed somehow by Steig et al relative to an average value for the site, for the individual month in question. If you’re only interested in a trend for a single site, the base period doesn’t matter. But if you’re comparing sites with very different coverage (as is the case), it’s important to do this consistently. We can’t know how Steig et al did this until they release the code they used to call the RegEM routines, so for now the world just has to take their numbers on Faith. (Either that or tentatively reject them as not yet Scientific.)

I’m currently playing with computing a trend for Racer Rock from its (corrected) raw data only. There are only 69 monthly observations, with only 4-6 observations per month, and lots of gaps in the data. It would make sense to regress temperature on 12 monthly intercepts, plus a common slope times time, plus a regression error. Each intercept would have a lot of uncertainty to it, because it is based on so few observations, but still they would be identified. However, their uncertainty would be an important factor in dermining the uncertainty of the slope. Most climate studies just pre-adjust their data for seasonality, and then ignore this easily quantifiable source of error.

My main problem here is just reading missing observations in Matlab. DLMREAD doesn’t recognize the – signs used by BAS to indicate missing data in their txt files. It still won’t read the file if I change these to . characters (the GAUSS missing code) or NA (the Excel missing code). Does anyone know a good trick for this?

Computing a composite trend across sites is tricker, since then the base period matters. However, a common base period can be imposed on the trend lines simply by subtracting the mean date of the base period from the time variable before running the regression.

Serial correlation is a big problem for which there is no standard easy solution. The Quenouille adjustment is quick and easy, and a lot better than nothing, so I used it. Here are some issues:

1. Even if the errors are AR(1), estimating rho from the residuals is different than estimating it from the unobserved errors themselves. Don Andrews (Econometrica 1993) had a way to obtain a “median-unbiased” estimate, that I have tried implementing and extending, but it’s really too cumbersome for everyday use.

2. The errors may be higher-order AR or ARMA, or even Fractionally Integrated (UC’s favorite). These are not insurmountable, but again cumbersome for everyday use.

3. Even if the errors are AR(1) and rho is known (instead of estimated with bias from the residuals), Quenouille’s adjustment (which dates back to 1950) is only a limiting value for a worst-case regressor (like a time trend). There is an exact formula involving matrices that is worth using, but Quenouille is still easier.

4. Another 1950s method is the Cochrane-Orcutt (CORC) adjustment. However, this re-estimates the coefficients under the assumption you know rho, which you don’t until you’ve estimated the coefficients. I now think that just adjusting the se’s of OLS coefficient estimates is the safest way to go. EViews and other programs generalize CORC to AR(p) adjustments, with the same drawback.

5. Econometricians often use the Newey-West HAC standard errors for OLS coefficients, and pretend the problem has gone away. However, this greatly undercompensates for serial correlation in most cases, so I dislike it.

6. Mainstream climatologists Santer, Nychka, et al (2008) just use Quenouille, creating a precedent for its use. Ironically, the same Nychka, with Santer and other others, in an unpublished 2000 working paper, suggested a solution to problems 1 and 2 above, but then went back to Quenouille in their recent paper. See threads 4106, 4127, 4163, 4216, and 4499 on CA last year for discussion (or use CA Google for Santer Nychka).

I hope this helps!

• Willis Eschenbach
  Posted Feb 9, 2009 at 2:33 PM | Permalink

  Re: Hu McCulloch (#113), thanks for your interesting comment. Inter alia you say:

  These “anomalies” are computed somehow by Steig et al relative to an average value for the site, for the individual month in question. If you’re only interested in a trend for a single site, the base period doesn’t matter. But if you’re comparing sites with very different coverage (as is the case), it’s important to do this consistently. We can’t know how Steig et al did this until they release the code they used to call the RegEM routines, so for now the world just has to take their numbers on Faith. (Either that or tentatively reject them as not yet Scientific.)

  I’m currently playing with computing a trend for Racer Rock from its (corrected) raw data only. There are only 69 monthly observations, with only 4-6 observations per month, and lots of gaps in the data. It would make sense to regress temperature on 12 monthly intercepts, plus a common slope times time, plus a regression error. Each intercept would have a lot of uncertainty to it, because it is based on so few observations, but still they would be identified. However, their uncertainty would be an important factor in dermining the uncertainty of the slope. Most climate studies just pre-adjust their data for seasonality, and then ignore this easily quantifiable source of error.

  The error in the removal of the average monthly anomalies can be quite large. Let’s look at station ERIN, which is the red dot to the right of Harry, and slightly larger than Harry, in Figure 4. The size of the red dot shows that it is one of the fastest warming stations in the region.
  .
  Now, here’s the GISS data for ERIN:

  Fastest warming in the region? It’s far too short to say anything firm, but the data clearly indicates that it is cooling.
  .
  I show the trendlines for the max, mean, and entire dataset. All of them are strongly negative. I suspect that part of the change in the data, from an actual trend of somewhere around -4°C/decade to some fictitious warming trend, occurs when the monthly anomalies are removed.
  .
  I say this because when I use the normal method to remove the monthly anomalies, the result has a trend which is only -1.5°C/decade, some 3°C greater. This occurs because, as is the case with many Antarctic records, the missing data is concentrated in one season of the year. For example, in the ERIN data, there are 7 records for February and March, but only two for August and September. This introduces large errors into the calculation of the monthly anomalies.
  .
  w.

• Hugo M
  Posted Feb 9, 2009 at 3:04 PM | Permalink

  Re: Hu McCulloch (#113),

  since matlab is using libc, replacing ‘-‘ in BAS data by “nan” might do the trick? Alternatively, there is also the textread() function.

Hu McCulloch

RE Robin, #114,

The raw Racer Rock data (as corrected Feb. 4) is at
http://www.antarctica.ac.uk/met/READER/aws/Racer_Rock.All.temperature.txt.

Enjoy!

Mark T.

Newer versions of MATLAB have replaced textread() and strread() with textscan(). You can also process the data using standard C functions such as fscanf().

Mark

David Smith

Here’s another comparison of Ferrell’s temperature with those of several nearby stations.

Many of the monthly records, including Ferrell’s, have gaps. In this exercise I used only months when Ferrell and the other station both reported values. I took the difference between Ferrell and the other station in each of those months. I then averaged those calculated monthly differences over each calendar year, discarding any year with fewer than six useable months.

The resulting time series gives an impression of whether Ferrell is warming or cooling versus each nearby station. This method is not perfect but I think it is good-enough for this visualization purpose.

Here is a map of the selected stations. I’ll be glad to add other neighbors if someone has one with a reasonably long history.

Here are the comparisons. (Note: the y-axis is in degrees C (each y-axis line is two degrees C). Please ignore the numerical values shown as the plots were positioned closer together for ease of display.)

In each comparison Ferrell has warmed relative to the other station and at about the same rate. I don’t know a good physical explanation of how AGW could trigger such an apparently local effect.

• bernie
  Posted Feb 10, 2009 at 3:08 PM | Permalink

  Re: David Smith (#120), David:
  How does Ferrell compare to proximate manned surface stations using the same approach?
  Also I am not sure why the difference between Ferrell and Gill are smaller than the nearby Scott and McMurdo. Somehow I would have felt that Ferrel would have been on average warmer than Gill which appears closer to the SP. Am I reading the map and scale corrctly?

• Alan Wilkinson
  Posted Feb 10, 2009 at 5:01 PM | Permalink

  Re: David Smith (#120),

  I thought we had clarified that Ferrell is on ice moving towards the open sea quite rapidly as you and tty mentioned previously – while Scott, McMurdo and Whitlock are fixed on islands and Gill is so far from the sea that it is unlikely to be greatly affected by its temperature?

Bernie

Alan:
Re: Alan Wilkinson (#125), If true, I still don’t understand the ordering of this graph. The relationships are almost opposite what I would have anticipated. What am I not seeing?

• Alan Wilkinson
  Posted Feb 11, 2009 at 3:11 AM | Permalink

  Re: Bernie (#126),

  I was just looking at the relative slopes and ignoring the vertical positions.

  Re: David Smith (#127),

  I don’t understand what you don’t understand. Seems to me quite reasonable that Ferrell is warming relative to all the others given its movement towards the open sea and the considerable temperature differential between those near the sea and Gill – way south down McMurdo Sound.

  And there’s a significant difference even between McMurdo and Scott though they are just at opposite ends of the same island I think.

David Smith

Re #124, #125 Bernie, both Scott and McMurdo are manned stations. Ferrel today runs about 2C cooler than Scott, 3C cooler than McMurdo and 6C warmer than Gill. I should have omitted the numerical values on the y-axis, to avoid confusion.
Alan, I believe the best guess is that Ferrell is moving at about 0.5km/yr. That is a drift but such small changes may be important due to proximity to water. As you note, McMurdo, Whitlock and Scott are land-based and Gill is far from the open sea. Those four stations, which partially surround Ferrell (at some distance), have temperature time series unlike Ferrell’s, which is a situation that is hard for me to explain. That’s the gist of what I was exploring. I wish there were useable sites closer to Ferrell, especially to the east, but the pickings are slim.

Mark T

Wish I could do more, Hu. Let me know if you need anything else offline or otherwise.

Mark

David Smith

Re #132 Robinedwards, try the “AWS Map (All Stations)” section, and links to smaller maps, at

AWS Homepage

StuartR

Quite amazingly Patrick Michaels has been given a page on the Guardian Comment Is Free to talk about this issue, the tenor of the responding comments is a sort of dazed confusion at seeing it there that is quite amusing.

http://www.guardian.co.uk/commentisfree/cifamerica/2009/feb/06/antarctic-warming-climate-change

• bender
  Posted Feb 12, 2009 at 10:39 AM | Permalink

  Re: StuartR (#135),
  Maybe you want to say something there about all the ad hominem attacks on Michaels and point to his published paper, in the primary climatology literature, with co-author Ross McKitrick illustrating that up to half the precious AGW may be attributable to more localized effects of socioeconomics/land-use change that are negelcted by NASA? Maybe it’s less. Who knows. The precious GCMs don’t include such effects. Rather, what you have is the same group that builds the models tinkering with the data used to test the models, trying to estimate UHI “contamination” – but not trying all that hard.

  As for Antarctica. Michaels is right about the Steig paper. Steig is another Juckes. Wanders in from far afield to save the planet, – snip

  • bender
    Posted Feb 12, 2009 at 11:22 AM | Permalink

    Re: bender (#138),

    Michaels is right about the Steig paper

    The warming in Antarctica is neither continental, nor does it span the five decades of the study period. It is far more localized in both space and time. Nature. Indeed!

    • thefordprefect
      Posted Feb 12, 2009 at 1:54 PM | Permalink

      Re: bender (#139), As i suggested it would be worth checking the Russian manned sites. Here are 3 of the longest records I could find with their location (all on the cooling side of antarctica!

      These are nomalised to monthly averages of 1961 to 1990 temperatures and all show warming of between 0.5 and 1 deg C
      Both min and max temps are plotted
      Mike

    • RomanM
      Posted Feb 12, 2009 at 3:19 PM | Permalink

      Re: thefordprefect (#142),

      If you stare at something long enough long enough, I guess you can see anything you WANT to. Try covering the part of the plots before 1980 and tell me how much warming you see.

    • thefordprefect
      Posted Feb 12, 2009 at 4:43 PM | Permalink

      Re: RomanM (#144),

      Try covering the part of the plots before 1980

      OK. If you screen out all data before 2000 and after 2008 what do you get then? 😉
      At about 1980 the ozone depth declined rapidly. Ozone is one of them GHGs!!! => Antarctic cooling????

      Mike

    • RomanM
      Posted Feb 12, 2009 at 5:00 PM | Permalink

      Re: thefordprefect (#146),

      Oooh, I like it! Another “pollution caused the cooling from the 40s to the 70’s argument”. Based on what? Changes in a graph? Is that the best reason you can give?

      Look at the ozone graph. Don’t cherry pick the part after 1980. Look at the time before that. The ozone level looks pretty steady. But the temperatires at these stations were increasing! This looks like a “convergence problem”. That must be what they would call the opposite of the divergence problem in paleos. So, now splice in another explanation for what drove the situation before 1980. Ozone is one of them GHGs!!! If all you have is a hammer … Give me a break.

    • Dave Dardinger
      Posted Feb 12, 2009 at 5:14 PM | Permalink

      Re: RomanM (#149),

      To be fair, Roman, the actual situation over much of Antarctica much of the year is that additional GHGs like ozone will cool rather than heat since there’s little or no solar input to be absorbed, re-emitted by the surface and then absorbed by the atmosphere and sent back to the surface, but some heat in the atmosphere which can be speeded out into space. It all depends on the specific layers in the atmosphere and just what the surface is like. Note: this is just what I’ve read, I’m no expert in the area.

    • Jens Andersen
      Posted Feb 13, 2009 at 12:39 AM | Permalink

      Re: thefordprefect (#146),
      This attempt at an explanation of the recent cooling of the Antarctic by a special event (the ozone hole) ignores that the anomalous behavior of the Antarctic is part of a more general pattern. The following quote from physicsforums.com gives a link to an interesting paper:

      “Attached is a link to Svensmark’s December 2006 paper that provides an explanation for the polar sea/saw.

      Svensmark’s paper concerns the Antarctic anomaly, which is also called the polar see-saw. The Polar see-saw is the term used to describe the phenomena where when the Atlantic region warms the Antarctic cools and vice versa. While other hypothesis (such as ocean ents) have been proposed to explain the polar see-saw, they cannot explain why the change is simultaneous. (i.e. If the effect has due to ocean currents one would expect a delay from hemisphere to hemisphere, as the ocean currents take time to change) Svensmark’s paper provides data (bore hole temperatures, see figure 1, and satellite data figure 2) to support his hypothesis that changes in global cloudiness is causing the polar see-saw.

• Bill W
  Posted Feb 13, 2009 at 1:46 PM | Permalink

  Re: StuartR (#135),
  Guardian Comment Is Free
  Interestingly a Monbiot ad hominem attack against P Michaels was removed from the comments by a moderator.
  BW

  • StuartR
    Posted Feb 14, 2009 at 9:49 AM | Permalink

    Re: Bill W (#166),
    Re: bender (#138),
    I saw the first few comments all gobsmacked to a man when I posted the link here, and then later it turned into a wave of ad hominem maybe not coincidentally after Mr Monbiots prompting, I didnt revisit after that. It is interesting that his comment was struck, maybe the ad hominems were so boring and predictable that even the CiF moderators may have found it a bit embarassing.

    • bender
      Posted Feb 14, 2009 at 10:22 AM | Permalink

      Re: StuartR (#175),
      What percentage of those commenters actually examined the same evidence that Michaels was working from? Maybe less than 1%? The rush to judgement is a bad sign for everyone. It’s catching. Look at the way IPCC handled Michaels & McKitrick’s published papers. But you’re right, it’s not just ugly, it’s BORING. Endless sermonizing.

    • StuartR
      Posted Feb 14, 2009 at 10:57 AM | Permalink

      Re: bender (#176),

      “What percentage of those commenters actually examined the same evidence that Michaels was working from? Maybe less than 1%? ”
      I agree less than 1%, but that is based on my very cheap and lazy survey of their responses. The only person who writes for the Guardian with a reasoned (if not glibly irreverant) scientific nature is Ben Goldacre, and he mainly sticks to medical matters. I think the fact that Patrick Michaels was given a chance to write a page is the most interesting thing. I wonder what went on in the editorial meetings that prompted it.

Phil.

brendy:
February 12th, 2009 at 10:30 am
Meanwhile …. up at the North Pole, Chrysophere Today’s satellite images of Artic sea ice show a rather remarkable jump in sea ice between Tuesday (Feb 10) and Wednesday (Feb 11)of this week in the Hudson Bay and the Newfoundland Sea between Greenland and Newfoundland that appears to be on the order of a few hundred thousand square kilometers. Take a look at
http://igloo.atmos.uiuc.edu/cgi-bin/test/print.sh?fm=02&fd=10&fy=2009&sm=02&sd=11&sy=2009

And a jump down on the 9th as well! The satellite images sometimes show a missing swath which the SSMI obviously did that day, note that it didn’t show on the ASMR-E which is what the daily image is from, also no corresponding drop/rise in the graphical data.

bernie

Interesting. Doesn’t the GISS Mirny data go back to 1956? And what happens if you drop 2007 because it only has 8 of 12 months data? (True 2007 is the warmest year on record but it also has the least monthly data!!) When I make these reasonable adjustments based on the available data, I get a warming trend that I bet is not significantly different from zero and in any case is half the rate you display for Mirny.

• thefordprefect
  Posted Feb 12, 2009 at 4:19 PM | Permalink

  Re: bernie (#143), Apologies, you’re right about the start point on Myrny and Novola.. new plots below
  Myrny data for 2007 is complete so is not deleted
  The averaging of data is for 12 months and this is set to zero if less than 9 months have data. Hence start and end of trace show as zero

  The point is these 3 sites are in areas of decreasing (according to AWS) temperature. All 3 show signs of warming.
  Mike

  • bernie
    Posted Feb 12, 2009 at 4:44 PM | Permalink

    Re: thefordprefect (#145), Mirny GISS data indicates that there is no data for Feb, Mar, May and June for 2007. I know GISS infills for missing data and that may be OK for large scale multi-station analyses but I see no point in doing it when you are looking at station records one by one.

    • thefordprefect
      Posted Feb 12, 2009 at 9:59 PM | Permalink

      Re: bernie (#147), the data is from the reference I gave in a previous post, the station data is here
      http://www.aari.aq/data/data.asp?lang=0&station=3
      There is no missing data noted for 2007 for Surface maximum air temperature and Surface minimum air temperature.
      Re: bender (#151), No one in this thread has posted significance levels for their trends! Why should any climate parameters conform to quadratic/linear/other mathematical trends? I gave a linear trend as others on this thread had used this.

      Re: RomanM (#149), Not sure what you are accusing me of. I suggested that ozone is a GHG (true?) Ozone levels over the antactic are constant (relatively) until 1980(approx). Up to this time temperature plots show a significant increase. After this time the “greenhouse effect” decreases as the concentration reduces by 6% over 10 years. This could reduce the ambient temp if it were not also being forced higher by other effects. The ozone levels having stabilised in 1993(approx) the temperature should again be rising. Mirny and Vostok may show this rise.

      Why the animosity to me posting a graphical form of someone else’s data? I believed it added to knowledge of AWS stations showing them to be inconsistent with manual measurements. Please tell me where my simple plots are invalid!
      Mike

    • bernie
      Posted Feb 12, 2009 at 10:17 PM | Permalink

      Re: thefordprefect (#155), thefordperfect:
      It looks like the GISS data differs from the data at the Russian site. Intriguing. I am going to have to look at this more closely.

      Bender: The r2 on these trends appear to be very small. I doubt that the coefficients are in fact significant.

    • bender
      Posted Feb 13, 2009 at 12:58 AM | Permalink

      Re: bernie (#156),

      Bender: The r2 on these trends appear to be very small. I doubt that the coefficients are in fact significant.

      ford: bernie’s on to you.

    • bender
      Posted Feb 13, 2009 at 12:55 AM | Permalink

      Re: thefordprefect (#155),

      Re: bender (#151), No one in this thread has posted significance levels for their trends! Why should any climate parameters conform to quadratic/linear/other mathematical trends? I gave a linear trend as others on this thread had used this.

      1. Humor me. Step it up a notch and give me the stats.
      2. Sounds like you are trying to dodge the issue of WHEN the warming occurred. Why?
      3. Be a sport. Up the ante.

    • thefordprefect
      Posted Feb 13, 2009 at 7:05 AM | Permalink

      Re: bender (#161), etc.
      All I did was to draw some curves from someone elses data. I added some linear trend lines because everyone on this thread was adding them. I was simply showing that permanent station data was different from AWS data. I was not pushing any GW. I am not involved with antarctic environmental studies so I cannot explain the levelling out of increase in some data from 1980. I gave the ozone data as it seemed plausable (I am not the only one to think this). I would be willing to try to understand why ozone depletion is not the cause. I would be willing to listen to other theories as to why the temperature levelled for a bit.

      I was not expecting the Spanish Inquisition!!!

      Because internal climate variability is complex, nonlinear, stochastic, prone to flip-flopping? Or have you forgotten that there’s more to climate than linear deterministic forcing? No, that would be confirmation bias.

      I was trying to point out that I cannot understand why anyone would try and mathematically fit a simple trend equation (linear, quad,exponential, etc.) to any temperature profile. Climate does not follow such simplifications. A moving average is better.

      Warming was occuring around/before 1956. I have no problem with that

      If you really want I will replot over the last 30 years. But I do not see the point and it seems very much like “cherry picking”.

    • Thomas Gray
      Posted Feb 13, 2009 at 2:11 PM | Permalink

      Re: thefordprefect (#165),

      I added some linear trend lines because everyone on this thread was adding them

      This brings up a classic V&V question. What is the validity of the linear trend for this type of data? Does it have any physical meaning?

    • bender
      Posted Feb 13, 2009 at 7:10 PM | Permalink

      Re: thefordprefect (#165),
      snip – be polite to other posters
      .
      Same thing for San Quentin in the other thread. What’s happening in the Arctic is not UHI. What it is is a bona fide mystery. Or have you not been reading your Hansen papers? What’s happening there totally does not fit the GCMs. The experts are mystified by it. Again, something with the oceans. Rather fitting with Tennekes’ remark at Pielke’s blog, wouldn’t you say? Except that the folks at RC and NASA agree: the Arctic Ocean and the Artic Oscillation atmospheric circulation are still very mysterious to us.
      .
      Gavin, feel free to chime in and correct me.
      .
      San Quentin, feel free to educate us about the arctic. Mind you, there’s a few threads here at CA you’ll want to read where the pertinent literature is discussed.
      .
      Methinks the oceans are a tad more complex than you want them to be. Try reading Wunsch, also discussed at CA.

    • thefordprefect
      Posted Feb 14, 2009 at 5:25 AM | Permalink

      Re: bender (#170), before the snip (I can take impolite, Steve!!)

      [Steve- it doesn’t matter whether you can “take impolite” – I don’t want it]

      I think you were talking about adding R2 etc to graphs.

      An example of R2 “invalidness”

      2 simple curves added (triangle amplitudex2) and 2 trend lines matched to resultant

      1. Which represents the trend better in life, which gives the best R2
      2. The linear trend is exactly the trend added to the triangular wave – why does the statistics not give this fact

      Perhaps you can provide a better statistical correlation that shows this on this very simple example.

      I’ll repeat my previous comment. Simple curve fitting a real life multi input data set is of little value.
      Mike

    • bender
      Posted Feb 14, 2009 at 5:50 AM | Permalink

      Re: thefordprefect (#173),

      An example of R2 “invalidness”

      “invalidness”? You’re certainly not quoting me on that one. I stop reading there.
      .
      I asked for real data, not toy examples. Another non-response. And a 6th ordern polynomial? That’s a bit overkill, don’t you think? Lastly, I didn’t just ask for “R2 etc”. I asked specifically for the s.e.s and p-values on the regression parameter estimates. How else you going to know if they’re significantly different from zero?
      .
      Answer those before I go on to read the rest of your comment where YOU ask ME questions.

    • Mitchel44
      Posted Feb 14, 2009 at 2:00 PM | Permalink

      Re: thefordprefect (#165),

      Your graph at 146 shows the trends starting in 1980, yet reductions in Ozone depleting substance production did not take place until 1989, in fact production actually increased up to 1988, then declined sharply after that.

      http://www.afeas.org/overview.php

      Don’t see that as a fit for the graph.

    • thefordprefect
      Posted Feb 14, 2009 at 3:58 PM | Permalink

      Re: Mitchel44 (#178), Ozone is a ghg, hence ozone “hole” could cause cooling: the ozone “hole” began as shown in my previous graph in 1978 at about the point where antarctic warming “ceases”.

      TES observations of tropospheric ozone as a greenhouse gas
      Worden, H. M.; Bowman, K. W.; Worden, J. R.; Eldering, A.
      American Geophysical Union, Fall Meeting 2007, abstract #A51D-0735
      We present satellite observations of the downward radiative flux from tropospheric ozone, for cloud free ocean conditions…. We examine the sensitivity of the outgoing longwave radiation (OLR) in the 9.6 micron band to upper tropospheric ozone and water vapor by separating the data into hemispherical and sea-surface temperature (SST) ranges. For 2006 data, we estimate an annual average downward flux for upper tropospheric ozone of 0.48 ± 0.13 W/m2 with a standard deviation of 0.24 W/m2 for the latitude range between 45°S to 45°N. This estimate includes natural and anthropogenic ozone sources and is higher than the 2007 IPCC average for climate model estimates of anthropogenic tropospheric ozone radiative forcing of 0.35 W/m2. We also observe that water vapor dominates the clear-sky ocean variability of the outgoing IR radiation in the 9.6 micron ozone band for SSTs higher than 299 K, consistent with the “super greenhouse effect”. This underscores the importance of chemistry-climate coupling in forcing predictions for tropospheric ozone.

      Re: bender (#174), the toy graph does not give sensible R2 values How can real world data. A linear curve fit can give only a general idea of rate of change of temp over the considered period.

      If you are trying to get me to admit total ignorance of statistical methods, then I will freely admit it. In my working life I have never had to use statisics!! I cannot even find FREE information describing s.e.s whatever that is (unless you mean SEs Standard Errors?). As you have the tools then why not use them on the simple toy data and show me their worth (or perhaps show me how to calculate your required numbers)?
      Mike

  • bender
    Posted Feb 12, 2009 at 7:06 PM | Permalink

    Re: thefordprefect (#145),
    1. Don’t bother posting trend lines unless you are going to report significance levels on regression parameters.
    2. Try fitting a quadratic.
    3. Nature newsflash: Antarctica warmed 50 years ago. Nothing since. Who knew?
    4. Yawn.

Steve McIntyre

careful with the GISS data. Please use the dset0 versions and specify the version, as, for what we’re doing here, version differences are worth keeping track of.

David Smith

To find the Jones link, click on Amundsen-Scott penguin on the left side of this page , then click the button below this reference –

Jones, P.D. and P.A. Reid. 2001: A Data Bank of Antarctic Surface Temperature and Pressure Data. ORNL/CDIAC-27 NDP-032. Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, U.S. Department of Energy, Oak Ridge, Tennessee

Ryan O

In case anyone was trying to download the text files for the manned stations on READER, the text link for Mario Zucchelli’s info actually directs you to the text file for Terra Nova Bay. Email sent. 🙂

• conard
  Posted Feb 13, 2009 at 9:28 PM | Permalink

  Re: Ryan O (#168),

  The problem is this:
  89662,Terra_Nova_Bay,74.7S,164.1E,92m
  89662,Mario_Zucchelli,74.7S,164.1E,92m

Paul Penrose

Ford,
Using a 30 year window is not cherry picking, but using the currently accepted standard for the definition of “climate”. This is what we are after, correct, climate change? If you wish to do a little more work, then by all means, go ahead and slide a 30yr window across the entire range for each site looking at how the trend changes over time. That would be very instructive and much more informative than a simple linear trend over a basically arbitrary period.

David Smith

Amundsen (manned) and Clean Air (AWS) are located at the nominal South Pole. (I haven’t found just how many kilometers, or meters, apart they are but they are close. Also, since they are located on the “polar plateau”, I assume they are at about the same elevation.)

It’d be interesting to compare such nearby manned and AWS stations, just to see what if anything pops up. So, here’s a comparison of their reported monthly temperatures for 1986-2002, the period for which Clean Air data is available.

(It is shown as Admundsen (current GISS version) minus Clean Air (current READER version), with the y-axis in degrees C difference. Missing months are omitted.)

One of the features I find interesting is the annual cyclical behavior. I suspect that is due to the difference in averaging methods between manned and AWS stations. If true, then that hints that caution may be needed when working with combinations of manned and AWS data. If the cycling is unrelated to averaging methods and is “real”, then that’s quite an interesting microclimate effect.

Another feature is that the average difference seems to shift about 1994. That surprises me. I guess it is microclimate-related but that is a weak guess. It’s a sizeable change.

If, in Antarctica, nearby stations drift like that, or if subtle AWS/manned differences introduce such drift, then I wonder whether small changes over time in the scattered, gapped Antarctica data are meaningful.

• Bernie
  Posted Feb 15, 2009 at 9:47 AM | Permalink

  Re: David Smith (#180), David:
  Are mannned stations manned continuously or are they unmanned during the depth of the Antarctic winter – JASO?

Hugo M

Very interesting, as it documents a varying bias. According to station history:

Site: Clean Air
Operation: Established with ID 8919 on 29 January 1986. The unit is located at the tower in the clean air sector at the SouthPole.

ID 8919 is wrong. Clean Air’s station ID in 1986 ws 8918, according to 10min data. In 1994, Clean Air’s ID was changed to 8987 and the Station Field Report History of 1994 reads:

Site: Clean Air
Performance: Repaired and moved on 24 January, pressure jumps erratically. expecially during the colder months.

Why your plot ends 2002? In Feb 2003, Clean Air got a new station ID:
21356. According to the station history, the whole station was removed two years later, at Jan 24, 2005. But according to BAS READER data , Clean Air has the ID 89208, and no data for 2003-2005, whereas 10min data for 21356/Clean Air is again available:
2003: 43601 of 57050 entries are valid (76%)
2004: 42801 of 52738 valid entries (81%)

Since 1994, the raw data provides dT in the eight column.

• Hugo M
  Posted Feb 14, 2009 at 8:34 PM | Permalink

  Re: Hugo M (#181),David Smith (#180),

  In case you asked just how far Clean Air was eventually moved in 1994, another Field Report has this on Clean Air:

  Clean Air site was moved from it’s previous location to one approximately 30 meters from the South Pole Meteorological tower on 24 January 1994. AWS 8918 was replaced with AWS 8987. A snow temperature profile was also added. The depth of the profile extends to 4 meters.

  So, what I previously named a “dT” (sensor) is a “snow temperature profile”?

David Smith

Now that Harry’s values have been shaved it’s time to take a peek at Uranus (the Glacier, that is).

I’ve been looking at daily data for that site, which includes resultant (averaged) wind direction. Here’s a plot of this daily wind direction, expressed in degrees of compass, from startup in 1986 through late 2000:

It may take a few seconds for the chart to make sense. Think of 180 as a due-south wind, which for Uranus Glacier is an airflow from mainland Antarctica. A value of 90 is an east wind while 270 is a west wind. The values of 360 and 0 are the same compass direction, of course.
The plot shows a reduction in southerly wind and an increase in northerly wind, most pronounced around 1990.

Perhaps there has truly been a reduction in south wind from the cold mainland due to a shift in weather patterns. If so then the cold south wind has been largely replaced by milder maritime airflow from the north.

If that’s true, then is that shift in weather patterns due to AGW and predicted by the models, or is it just a natural shift which affects local temperature?

Or, maybe there is something local which is affecting Uranus, perhaps an indication of a station relocation or glacier movement circa 1990. Maybe the station is now experiencing funneled, possibly downslope winds due to something in the landscape. Perhaps the mast height and sensors have been somehow altered by the weather.

I dunno.

I’ll take a look at other Uranus Glacier daily data to see if it hints at explanations. One thing the other data does show is evidence of a snow-buried temperature sensor in 1988, when diurnal temperature variation dropped to near-zero. I’ll post that plot later.

David Smith

Re #185 Good question. I don’t know, bernie. My assumption is that most are manned year-round but that’s a guess.

Re #184

The 1990 station history for Uranus Glacier has this note:

Wind speed and direction were intermittent after 25 Apr. Site was removed on 5 Nov.

That makes me wonder if the site was relocated in 1990, which would explain the change in wind direction tendencies. I haven’t found a reference which confirms a 1990 move but there are at least three sets of coordinates given for Uranus Glacier over 1986-2002. If I use the set at the Wisconsin website then (per Google Earth) the station is now on a northward-facing slope of about 15 degrees. That could help explain the lack of southerly wind.

On #180, my averaging-method conjecture to explain the annual cycling doesn’t make sense for a site located at the South Pole, which sees the same angle of sunlight throughout 24 hours. So, scratch that thought. The conjecture does make some sense for sites where sun angle varies during the day. I now wonder if the two South Pole sensors are at different heights, as a sensor closer to the ice surface may run warmer in austral summer and colder in winter.

Back on Uranus Glacier, here is a plot of the diurnal temperature range from inception to the end of 2001. (It’s a 30-day average for ease of visualization. “Diurnal range” is defined as the difference between the highest and lowest three-hourly temperature readings.)

Of note are the drops to near-zero diurnal range, which I attribute to the sensor likely being covered/buried in snow.

There is also a hint (poorly displayed on this 30-day average) that, in the days immediately prior to burial, the range increases then drops sharply. I wonder if temperature right at the surface varies more than it does slightly above the surface. The question mark shows such a spike which then fell towards zero (not shown) immediately before the data ended.

There is also a hint of a downward shift in average about 1990. That adds to my sense that the station was relocated in 1990.

• Hugo M
  Posted Feb 16, 2009 at 3:51 PM | Permalink

  Re: David Smith (#186),

  I now wonder if the two South Pole sensors are at different heights, as a sensor closer to the ice surface may run warmer in austral summer and colder in winter.

  The diagram below was taken from an old book about Antarctica. It shows a vertical temperature profile taken at Molodjoshnaja/Antarctica at Oct 11, 1974 17:30 by Alfred Helbig, a metereologist and member of a geophysical research expedition mounted by the former GDR. In a short report describing his responsibilities he also states that (for another, but near point) the strongest gradient was found to be 7.1 [°K/m], at a height between 0.75 and one meter. Nightly gradients have been found biggest. Influential factors were also fall of ground and wind speed. The first column of the table located to the right provides values for the temperature gradient, which he probably regarded to be somewhat representative.

  Image source: Lange, Gerd [ed.]: Bewährung in Antarktika,Antarktisforschung der DDR, VEB F.A. Brockhaus Verlag Leipzig, DDR, 1. Auflage 1982, p. 123

Robinedwards

The Mirny data are unusual in that they run uninterruptedly for a long period, 1956 to the present, with no missing values. They thus present an opportunity for carrying out analyses that might be able to identify features that are perhaps hidden in the sparser data sets that are more typical of of polar regions.

I am therefore putting forward an alternative to the standard analyses which I submit for your comments.

First, I repeat my assertion, expressed in this blog in previous contributions, that fitting of a simple regression (trend line) to a data set that even to the most casual and least critical perusal looks extremely unlikely to be realistically approximated by a linear fit (y=ax+b) is an exercise of limited value. No-one could conceivably by persuaded that it has any useful basis, function or merit in the real world. It is I suppose possible that there is no intention to propose the simple model as a serious contender for explaining climate behaviour. However, this model and its associated or equivalent inferential statistical quantities – like R-Sq and correlation coefficient and sometimes confidence intervals – which are often cited or quoted seems to imply that it is regarded as satisfactory.

So, what is to be done? In the first place one must hypothesise a plausible underlying model that can allow for the existence of climate regimes that do not sit comfortably with the simple linear fit but which might be regarded by many as being “potentially realistic”. My suggestion is that an assembly of contiguous simple linear models (i.e. covering much smaller time periods than the whole 1956 to 2007 data span) might be suitable.

Choice of suitable segments of the data is clearly a problem, but one that can be addressed by making use of the venerable quality control technique of computing the cumulative sum of the data points over the time period of interest. Applying this to the individual monthly averages as reported by observing stations is readily done, but of course yields a sinusoidal plot due to seasonal effects. De-seasonalising by simply subtracting the overall average for each month from the appropriate observations is a simple and adequate method, and it yields a data set that I call a “monthly difference”. This difference has a mean of zero, by definition – and this is a simple check on the arithmetic involved. By forming the cumulative sum of these monthly differences one obtains a data set that sheds a remarkable amount of light onto the scene.

Without going into the interpretation of cusum plots – which requires a paragraph or so but which is in fact simple and logical, I’ll set out some conclusions on Mirny based on this method. The first and most important is that the data indicate /very clearly/ that there have been at least 5 regimes of varying lengths over the whole period. Most of these are marked by step changes (occurring over a period of 1 or 2 months) of magnitudes between 1 and 2 degrees C. A simple explanation or reason for such changes is difficult for me to advance, but a possibility is a change of instrumentation or siting. They could presumably also be real climate effects, but whatever the underlying explanation they are so obvious that I feel they cannot be ignored.

A further simple inference from the cusum plot which has the general shape of a W is that an overall rise in value over the whole time period seems likely. However, in view of the huge “noise” element in this type of data this slope may not be statistically “significant” – whatever that means – and this is borne out by the statistics of the simple trend line.

Due to my sad lack of computer literacy (and knowledge of Windows XP) unfortunately I do not know how to publish diagrams (GIFs) on this blog, so I have to use words and numbers instead.

As noted above, I have used cumulative sum technology to make tentative identifications of five major regimes that appear to have occurred at Mirny. I characterise them as follows:-

Initial stable period was Jan 1956 to June 1958.
Regimes are named 0,1,2,3,4

Classification of MonDif by Regime with 5 Classes and 636 Items in total

I cannot get the formatting right. Sorry! Will insert \ to indicate a new field.

Regime\ Mean\ Std Dev\ No.\ Percent\ Std error\ Period\ Probable step size\\
MonDif \ of total \\ at end of regime
Deg C

0 0.931175 1.6878 29 4.56 0.313416 to June 1958 -2.0

1 -0.542567 1.92737 134 21.069 0.16649 to Aug 1969 +1.2

2 0.16195 2.06277 272 42.767 0.1207 to Apr 1992 -1.9

3 -0.594584 2.01892 99 15.566 0.202909 to Jul 2000 +2.0

4 0.5605 1.78886 102 16.038 0.177124

Sorry about this formatting problem. Hope you can understand it all.
I could send the GIF illustrations to anyone via email – but everyone is anonymous :-((

Note that the step sizes do not correspond with the changes in mean level. This is because all the regimes have a slight upwards slope, so the beginnings and ends of the regimes are displaced from the regime means.

One way analysis of variance of the simple regime-less model yields a between regimes / within regimes F ratio of 2.51, probability 0.114. The slope of the trend line is 0.0083.

Using the Regime model the F ratio is 8.93, probability 5*E-7, showing that the hypothesised regimes are not randomly related to the data. Yes! I knew that because I chose them, but on very strong evidence.

Summarising, my tentative hypothesis is that Mirny data have presented a regime-like pattern, which is surely worthy of more exploration and explanation before firm conclusions on this coastal site’s climate history is put to bed.

I can provide more information if anyone wishes.

Robin

• thefordprefect
  Posted Feb 15, 2009 at 4:37 PM | Permalink

  Re: Robinedwards (#187), To upload an image
  1. best to save as jpg. Scale to be approx 800 pixel wide or less
  2. join an image hosting site eg, http://www.imageshack.us/
  3. on first entry to the site you are given the oportunity to upload a picture by browsing to the image (easiest place to find it would be the desktop). click on [host it]
  4.when uploaded I find it easiest to click [my images] in the long orange bar
  5.click on the small [i] against the right hand edge of the image.
  6. click once on the direct link and copy all the link (url).
  7. return to the thread move the cursor to the place you want the picture in the edit box. Click on the [img] button. If a warning bar occurs at top of page warning abouit scripted windows – click on it and allow scripted windows. re-click the [img] button and a box should appear. delete its contents
  8. paste the copied info from the image hosting site
  9. ok the next box or 2 until the url appears in the text. press return and the image should appear below in the preview.
  NEVER use the [greater than] [less than] angle brackes in a post it causes silly things to happen!
  ALWAYS mark and copy your post (in the entry box) before submitting -when it goes wrong you can always paste back and repost
  It is interesting that we use similar “Monthly differences” my plots are done over 1961-1990 periods not the whole record)

David Smith

Here’s a closeup of March, 1988 thru February, 1989 for Uranus Glacier. It shows the drop in diurnal range as well as the decline in temperature change from one day to the next. Those are consistent, I believe, with a sensor being buried in snow.

And here is a plot of the recorded temperatures for the same period, plus an overlay of the temperatures for the prior twelve months. There is no way to know what the austral winter of 1988 was actually like at Uranus Glacier but the overlay suggests it probably missed the cold spells.

To me, based on the last several posts, Uranus Glacier should probably be viewed as two time series, one from 1986-1990 (with the winter of 1988 missing) and the other from 1992 forward.

David Smith

Re #191 Hugo, thanks for the reference on the vertical change of T near the surface. I enjoy the older, well-written references. If there is an opposite effect in sunlight then we may have an explanation for the cycling, assuming the sensors are at somewhat different heights.

You also mentioned windspeed. I notice in the Ferrell data that higher winter windspeed tends to be associated with higher temperatures. When I detrend Ferrell’s winter temperature data and plot the anomalies against windspeed I get an r-squared of about 0.3, indicating that wind indeed affects polar winter temperature (probably due to the effect of wind on mixing). Makes sense and it makes me wonder if the apparent warming or cooling of some Antarctic places might be caused by changes in windspeed.

• Hugo M
  Posted Feb 17, 2009 at 1:39 AM | Permalink

  Re: David Smith (#192)
  Dave, somewhere within one of those various sparsely written AWS field report (or – history) files I read that during summer, slow wind speed also affects temperature — because sunlight is directly heating up the sensor shields. Using raw data, if one could throw in the dT sensor information and windspeed readings, one could eventually come with a quality weighting? Something like this:

  Winter && slow wind && dT -> const: sensor is near to be buried
  Summer && slow wind && |dT| >> 0: sun heats up the upper sensor

  Besides, there should be at least one type of AWS in use which is equipped with a snow hight sensor, termed Acoustic Depth Gauge Sensor, which could be used as an calibration example.

  Another analytic approach could be using temperature data only if the wind speed is sufficiently high, say above 3 [m/s] and when a high daily variance indicates that the main temperatur sensor is probably not buried. Thus all considered values would represent mean air temperatures, which, as you said, should not so much depend on sensor height above ground.

David Smith

Here’s a plot of wind speed (30-day average) at Uranus Glacier:

(Note: the mauve trend lines should be viewed with caution, as the timing of the data gaps may affect those apparent trends.)

The appearance is of an apparent shift in mean wind speed circa 1990-1991. This is suggestive of a location change. A change in wind speed is also important because wind speed may affect winter temperature (higher wind speed tends to lead to higher near-surface temperature, due to greater mixing).

The two graphs (wind direction and diurnal range) shown earlier in this thread also suggest a shift of some kind in 1990-91.

Finally, here is a plot of temperature (30-day average):

(Note: as with the first plot, the mauve trend lines should be viewed with caution, as the data gaps as well as the different seasons in which the segments start and stop will affect the slope of the trendlines.

The visual suggestion is of a shift in the temperature mean, circa 1990-1991. My conjecture is that this was due to a sensor relocation. Perhaps the data could be used for trending purposes if separated into a 1986-1990 trend and a 1992-2002 trend, but not as a single series from 1986-2002.

David Smith

Re #195 curious, my growing sense is that windspeed is an underappreciated factor in Western Antarctic temperature.

The winter wind seems to have increased somewhat in the western region in recent decades. This is important as the wind disturbs and mixes the very cold air near the ice surface with somewhat warmer air above the cold surface layer (see Hugo’s chart in #191).

The result is that the recorded temperature at the sensor reads a mix of cold surface air and warmer air from above, giving an indicated rise in temperature. It’s not a real rise in the sense as the atmosphere has not warmed, it’s just that the sensor is sampling a different portion of the atmosphere.

I’m still looking over data to see if this hypothesis holds and hope to have a few plots this weekend.

• Philip Mulholland
  Posted Feb 24, 2009 at 7:09 AM | Permalink

  Re: David Smith (#196),

  David

  If you want to find detailed temperature profile data have, you looked at the Dome A website?
  Excellent examples here of temperature rise associated with mixing.
  Note how on each day (e.g.15th Feb) the 1m sensor lags the temperature rise at the 4m level as the cold pool of air gets disturbed in a top down mixing process. When the wind stops on 16th Feb the cooling process re-establishes the temperature inversion and the 1m sensor plunges to the lowest value in the daily profile. The ice surface is continously acting as a massive radiator of heat to space.

• Geoff Sherrington
  Posted Mar 1, 2009 at 3:12 AM | Permalink

  Re: David Smith (#196),

  David, further to the above, at Ferrell I have just looked at the latest wind speed 3-hourly readings for part of 2009. The least significant digit 7 is still completely absent, as you might expect of you summed 10-minute readings with the 7 absent (but not always if you averaged them). Fortunately, the numeral 3 is back with a vengeance, but now, poor old zero has dropped off the radar. There are 277 readings in this dataset so you’d expect each LSD to report 27 times, approx. Here are the actual occurrences:

  0,0 times
  1,42 times
  2,20
  3,94
  4,17
  5,24
  6,35
  7,0
  8,41
  9,4

  If you combine defective wind speed data with defective temperature data as in post 216, I guess you get a defective graph. I’m no longer a practising statistician even by contamination, so that makes me just a defective detective.

  • Hugo M
    Posted Mar 4, 2009 at 8:35 PM | Permalink

    Re: Geoff Sherrington (#218), Geoff, your jump to a thus far reaching conclusion is not compelling. Try this real world example:

    #sample conversion function for Dragon Devices ECT sensors
    raw_to_temp = function(d) {
    t = log( (4095/d) – 1 );
    t_c = 25.02 + t * (-22.84 + t * (1.532 + (-0.08372 * t)));
    t_c;
    }
    T = round(raw_to_temp(500:800),1);
    LSD = abs((T-trunc(T))*10);
    hist(LSD);

    • Geoff Sherrington
      Posted Mar 4, 2009 at 10:39 PM | Permalink

      Re: Hugo M (#219),

      No, Hugo, I simply look at the numbers. You give some conversion code for temperature in a garden potting mix. I gave also an example for wind speed. Presumably it has a different but related algorithm; it has the same problem with missing numerals.

      What I initially infer from the Ferrell data is that the temperature should be reported only in whole degrees, with no figures after the decimal. What would be the case for wind speed?

      Do you consider that the combination of a large number of (rounded?) observations can overcome a lack of measurement resolution? I am sceptical of how temperature change of 0.1 degrees a decade for the Antarctic can be derived from such coarse data, but I am open to education.

      Do you know how many bits cover the designed instrument temperature range in the Antarctic AWS sensors?

bender

Hmm, neat.

bender

Interesting. Would be nice to see some other station data.

curious

Re: David at 204. Thanks David – as a novice I was thinking of tackling this!

I’d also hoped to include the wind direction. I had a look at the Ferrell data and if I’d got my head round it properly it was pretty much continuously offshore within a 30deg splay. I’d be interested how other stations compare. I did a quick check and there do seem to be references about on antarctic wind patterns but haven’t had time to follow up. I’m wondering about what the overall airflow pattern across the continent is – no doubt it is known and understood already but I’m wondering how it relates to AWS data from up to approx 3m height. Thought I’d mention it FWIW – sorry if it is a dead end.

I’d also like to see any realtime or short time step series data from these AWS stations so if anyone knows a source please shout.

curious

Re: David at 207 – Thanks will follow up. C

David Smith

curious, I’ve looked at the archived 10-minute data available here , with the first column being the sequential day of the year and the second column being the sequential (10-minute) reading of the day. I’m pretty sure that data has undergone cleanup. The three-hour data is on the same page, at the bottom.

You might spot-check some examples to see just how the three-hour data is calculated (spot reading every 180 minutes, or average of 10-minute readings or something else).