Assume A Spherical Cow – therefore all steaks are round

Assume A Spherical Cow

Original image and story (with the risk, of course, that the AGW Wiki Langoliers will go change history… again.)

Fair warning, this posting is a bit long. It covers a lot of history of what I’ve been doing and why, plus for the long suffering, it has graphs in it ;-)

Summary

I am not trying to calculate a “Global Average Temperature”, I am looking for information about the structure of the temperature data that we have for the planet. Basically, my major goal is to see how big the DATA problem is as a first step. In a later process will come analysis of how well GIStemp handles this data bias. And along the way I am in the process of making my own analysis program that shows what the actual warming of the planet (if any) is likely to be. The DATA bias, and my analysis of the data showing what I think is the correct result, will both be used to assess the quality of the processing in GIStemp when the time comes. The first two steps are substantially done, but the last step is still a work in progress (though with some early results).

The basic “problem” with other folks assertion that a “thermometer compared to itself” anomaly process shows there is no bias in the data, so I must be wrong; is that it looks at a solution to the problem and asserts that shows there is no problem. And it looks at my measurements of the problem and asserts I’m saying that is the temperature of the planet. I’m not. So we have a solution to a problem offered as proof there is no problem, and an assertion about what I’m doing that is wrong.

So what AM I doing? “Measuring the data” or “characterizing the data”. That is the very first thing one ought to do in a problem like this. Get a feel for the data. Does it rise in steps? As a smooth 200 year ramp? Are there large dropouts or spikes and drops? Big gaps and holes?

Why do you do this? Because it tells you how to fix the issues that the data has. And all data ‘has issues’. It also gives you a first rough benchmark for the data. It shows you where you are starting and lets you see the size of the problem. And there are lots of problems in the data. About 1/2 of the data are missing. Huge gaps and holes. Short lived instruments catching a biased glimpse in time. Changed technology introducing step function jumps up.

Yes, I think it matters to find that out and put a size on it. That way you have something to measure your performance against as you test your solution. While some folks are willing to just assume their solution is perfect, I prefer to measure them against the starting point.

In the case of GIStemp I’ve run a benchmark that shows that the particular kind of anomaly process it does (an imperfect comparison of one group of thermometers in a baseline period to a different group later) does let instrument change effect leak through. So we know it ‘has issues’, now we need to measure ‘how much’. In that context, knowing what bias is in the base data does matter. Furthermore, even the folks at NASA admit this problem. From the NASA foia emails as discussed here:

https://chiefio.wordpress.com/2010/02/18/nasa-giss-speaks-foia-emails/

Subject: Re: Your Reply to: GISS Temperature Correction Problem?
From: Gavin Schmidt gschmidt@giss.nasa.gov
Date: 19 Feb 2008 14:38:47 -0500
To: rruedy@giss.nasa.gov

I had a look at the data, and this whole business seems to be related to the infilling of seasonal and annual means. There is no evidence for any step change in any of the individual months.

The only anomalous point (which matches nearby deltas) is for Set 2005. Given the large amount of missing data in lampasas this gets propagated to the annual (D-N) mean – I think – with a little more weight then in the nearby stations. The other factor might be that lampasas is overall cooling, if we use climatology to infill in recent years, that might give a warm bias. But I’m not sure on how the filling-in happens.

Gavin

So it isn’t just me saying that data issues can get through the GIStemp “climatology” process, it’s Gavin in an email to the guy doing the maintenance of GIStemp.

For these kinds of reasons, it is valuable to look at the base data before doing any anomaly processes on them (but not as a replacement for same) and see just how much bias is in these base data that could be leaking through such a ‘climatology’ code.

The bulk of the GHCN analysis I have done is exactly this kind of “characterize the data” process. And it finds bias “by altitude”, “by latitude”, “by airport percentage”, and a few others. Those biases do not go away from the data just because someone has a way of fixing it that is not used in GIStemp. Those biases could very easily end up in the work product of GIStemp that is used for policy decisions.

Further, I’ve made my own variation on an ‘anomaly code’ that I’ve used to look at the data ( to find the most likely actual warming or cooling trend). What I’ve found is that the “trends” are quite different from region to region. Whatever is happening has strong local components that change when instruments change, and little “global” component. To average all the regions together “over averages” and hides the most interesting information. Most critics of actually looking at the data make this error. Averages are used to hide information that is in the way. But you don’t want to average too much or you will miss the truth hiding in the data.

Here is a sample of that product, the USA. We’ll come back to it near the bottom. For now just note that it’s not a warming hockey stick (hurrah!) and it shows things we know to be true like a warm 1930’s. (And notice that the present is on the left and the past on the right.) You ought to be able to click on the image to get a larger view.

USA All Data Delta T

And the truth that I see in the data is that there is plenty of room in the data for a warming signal based on instrument change (10’s of degrees C) and some room for a cyclical signal based on natural processes like ocean changes ( single digits of C ) but by the time you have them out of the way, the remainder is just barely able (and sometimes unable) to cover the Urban Heat Island, Airport Heat Island, and other known siting issues. After allowance for them, I see little room for CO2 as an agent. But that is an early assessment based on only a few regions. It will take a bit longer to demonstrate it fully (though a couple of examples are offered below).

At the end of it all, a Hypothetical Anomaly Process showing it can’t find bias left in the data after squeezing it all out is of little interest, and less use. I’m interested in measuring the size of the bias first, then taking it out. And in any case, it is what the real world code used for policy decisions does that matters, not the Hypothetical Code.

The Trouble with Tamino – Spherical Cows

OK, against my better judgment, I’m going to give my evaluation of where someone else has “got it wrong”. I generally have a policy of avoiding such “prove a negative” food fights, so do not expect me to indulge in this behaviour again.
But Just This Once…

The Trouble with Tamino (what I’ll be calling that article – apologies to A. Hitchcock) begins with an error of assumption. He/She/It (that’s the trouble with pseudonymous posters, you don’t know what appellation to give them…. maybe I’ll just go with the grammatically acceptable “singular they”…) They make a false assumption about what was asserted, then attempt to prove their false assumption means something.

In this case, the first thing assumed wrong is that I have asserted that the anomaly trends for the cold stations are different from the warm stations. I have never asserted such a thing. They then run off to prove that something ( that I’ve never asserted) is wrong, so I must be wrong. Well, when I was a kid we called that “A swing and a miss”. Now it is just another “error of assumption”. But right out the gate, the basic assumption of the whole “proof” is broken. I’m not talking about anomalies, I’m talking about the DATA.

Further, I’m not overlooking the anomaly process. I’m deliberately avoiding using it. I want to see the patterns of the data. I do not want to make them disappear at this stage of the examination. To assert that I’ve got it wrong because I’m not doing what someone else wants (because I’m looking for different properties; looking for different answers) is to simply not understand what I’m doing or why.

What I have found is a persistent bias in the spacial and temporal distribution of temperature recording stations used in the GHCN. That bias is real, exists, and is demonstrable (as many of my postings have illustrated “by latitude”, “by altitude” and even “by airport” flag.) That does not mean it is impossible to correct for that bias, but first you must admit the bias is there.

To demonstrate a method of dealing with the bias is not proof that the bias is missing.

The Trouble With Tamino all hinges on the notion that the only way that a bias in data can exist or can have an impact is if the anomalies over time in the average of all of them is divergent between the two sets (the kept and the tossed) and that assumption is simply wrong. Not all processes use “self to self” anomalies, so not all processes will respond to data biases in the same way.

Further, “The Trouble With Tamino” over averages. (That is, it averages too many things together and loses too much information). There is the implicit assumption that if you average a global bunch of anomalies together it means something. Well, if I have Canada getting warmer and Africa getting colder and average them together, all I may find is an accidental zero. So the world is divided into two groups (kept and tossed) and they are found to be about the same anomalies (after a basket of ill defined area adjustments). It can simply mean that the average of everything hides the interesting bits in both cases. If you would really see what is happening, look at the detail. We’ll do that in one case at the bottom of this article.

Then the leap to the conclusion that this means something about what the actual programs used, like GIStemp, do with the data. The anomaly process in GIStemp is different from the theoretical “Self to Self”, so The Trouble With Tamino can say nothing about it. Yet it claims that the bias in the data can have no impact because it does not show up in some other process. The process in GIStemp is not Their process, so they can say nothing about how the bias will be handled by the actual code used in the real world. (The thing that I care about, and WHY I want to know what the actual bias is in the data, not what is left over after someone does some other process).

So we have an ill conceived process that addresses a question orthogonal to what I’m investigating and finds that highly processed data has had the interesting information squeezed out of it. Then leaps to the conclusion that this has meaning about how other processes and programs will handle bias in the data. To quote someone or other “It’s not even wrong”.

Trot Out The Hypothetical Cow

Sidebar: There is a fairly consistent Hypothetical Cow that gets brought on stage each time the anomaly process is discussed. Folks leap to the conclusion that the data products, like GIStemp, do anomalies “Self to Self”. Comparing a thermometer today with itself in the past. GIStemp does not. It does “Random box of thermometers in time A” vs “Different random box of thermometers with adjustments and some fill in and some made up numbers in time B”. While it’s nice to have a Hypothetical Cow that makes perfect soycowburgers, I’m investigating what the actual code that is run does on the actual data it is fed.

At this point the Trouble with Tamino heads off to hypothetical soycowburger land by computing anomalies in some (ill defined but different from GIStemp) way and asserting that this means what I’ve said about GIStemp processes on actual data must be false. Nope. Just comparing Hypothetical Cow soyburgers to real world beef.

To quote from the Trouble with Tamino article:

“Unlike some who claim to have analyzed these data, I combined station records in a proper way, I computed temperature anomalies properly, and I’ve computed area-weighted averages. All these steps are essential for a correct result.”

So here we have a statement that what is being looked at are not the DATA, but “combined” and “computed” things in ways using the “anomalies” in “area-weighted averages”.

And it might even give correct results (though without the method and code it is not possible to say). It is just as possible to be giving bad or changeable results as GIStemp does.

But in all cases it is just another Hypothetical Cow talking about something They have done which is not relevant at all to what GIStemp does with the actual DATA bias.

So if They would like to go work for NASA and fix GIStemp, I’m sure we would all be happy, especially if Their method really does work and correctly deals with all the issues of station drops and holey data and instrument bias.

But for now it is just another Hypothetical Cow making round steaks.

And it speaks not at all to the issue of DATA bias and instrument errors. Maybe it can fix the bias, but it says noting about the presence or absence of the bias. Basically, suppressing a fever with aspirin is not proof that you are well.

Sidebar On GIStemp Issues

Sidebar on GIStemp: I’ve run a benchmark on GIStemp that shows how it responds to station changes. First the AGW Advocates asserted that station changes would do nothing since “The Anomaly Will Fix It”. But this benchmark shows otherwise. Realize that this was NOT run on some fictional data, nor on “my data”. It uses NASA GISS code run on NOAA / NCDC data. The result was that the anomaly map changed. In 5/2007 the USHCN data set became static. Until 11/2009 GISS did not change to the USHCN.v2 set, so the USA was limited to the 136 or so stations in GHCN. I was able to “put back in” the USA stations for that interval and see that the bias leaked through to the “basket to basket” anomaly map. So in the real world of real beef, the thermometer changes DO bias the DATA and the changes in those DATA do make it into changes in the Anomaly Map from GIStemp.

https://chiefio.wordpress.com/2009/11/12/gistemp-witness-this-fully-armed-and-operational-anomaly-station/

Oh, when USHCN.v2 was “put back in” it was made warmer in the process:

https://chiefio.wordpress.com/2010/01/15/ushcn-vs-ushcn-version-2-more-induced-warmth/

So in doing such a benchmark with the intent to use the actual values in the anomaly map, you have to allow for that and do USHCN.v2 up to 5/2007 vs USHCN.v2 up to 11/2009 and not compare USHCN to USHCN.v2 (even if that is the way the GIStemp history will now look comparing older results with results after 11/2009). So the present test mostly just shows that “things change” where exactly ‘how much’ needs a bit more work to unscramble the eggs from NCDC.

How much change of thermometers is there between the two sets of time that GIStemp compares? Quite a bit. This site:

http://www.ncdc.noaa.gov/oa/climate/ghcn-monthly/source-table1.html

Makes it sound like there are thousands and thousands of thermometers used in GHCN. And there are. Just not for very long for most of them. (The total is about 7280 – though you would not know that from looking at their chart. It’s “in the fine print” elsewhere in the data set description. The chart lists things like:

NCAR's World Monthly Surface Station Climatology  3,563
NCDC’s Maximum/Minimum Temperature Data Set  3,179
Deutscher Wetterdienst's Global Monthly Surface Summaries Data Set  2,559
Monthly Climatic Data for the World  2,176
World Weather Records (1971-80)  1,912
World Weather Records (1961-70)  1,858
U.S. Summary of the Day Data Set  1,463
U.S. Historical Climatology Network  1,221

We’re somewhere north of 17,000 already and there is some chart left to go.

So what is the truth like? Well, starting with that 7280 total that actually make it into the GHCN data set, we can look at how many are in a sample baseline year and how many are in the present of 2010 (as of the January data). I used the built in commands of Linux to make two files with the unique count of thermometer records in each year. The following command counts how many are in each of those two files

[chiefio@Hummer data]$ wc -l EMS.u.1960 EMS.u.2010
5253 EMS.u.1960
1113 EMS.u.2010

5253 thermometer records are in 1960
1113 thermometer records are in 2010

Hmmm…. Only 5k in the start of the baseline period. And dropping to 1113 today.

Gee, we’re under 1/10 th of the “small part of the chart” we looked at above; no where near 17,000. And we have a 5 to 1 ratio between the two years. So at least 75% of the stations in the baseline will not be in the present. Those are two very different boxes of thermometers we’re going to be comparing… But just how many of them are in both time periods?

The Linux “diff” command tells you what goes in, and what is taken out, to turn one file into another. Here I look in the output of the command for the number of “pull it out” records vs the number of “put it in” records to tell me how many are changed.

[chiefio@Hummer data]$ grep “< ” DIFF.1960.2010.uniq | wc -l
4319
[chiefio@Hummer data]$ grep “> ” DIFF.1960.2010.uniq | wc -l
179

So 4319 are no longer around in 2010 and 179 new ones are put in. That makes the stable set 1113-179 = 934 (Or done the other way: 5253 – 4319 = 934 )

Our common base between the middle of the baseline and now is: 934 thermometers. Not “thousands and thousands”. ( it is 715 in 1950 just before the baseline starts. Things change during the baseline period too…) I make that about 17.8 % of the starting set survive. The rest of our “box of thermometers” from that baseline year don’t make it. Not exactly a lot of stability in those “Baseline Box” to “Present Box” calculations.

And when you consider that there are 8000 grid/boxes of surface on the globe to be filled, most of those boxes are comparing two very different things to each other. (Those 5253 thermometers get smeared into 8000 grid/boxes in 1960, then the 1113 thermometers in the present get smeared into 8000 grid/boxes to cover the globe. There is about 1 / 10 stable thermometers per grid/box. Then those two sets of smears get compared to make the “Global Anomaly Map”. Not a lot in common between those sets of boxes by that point.)

(It is actually worse than that as GIStemp tosses out a bunch of these records in processing. Anything shorter than a 20 year life span, for example. So the actual number used to make the grid boxes will be lower than these gross numbers. Others may be dropped for other reasons.)

To give you an idea how sparse the coverage can be, notice that most of the dots on this picture are the ‘short lifetime’ dots:

GHCN Stations by Geography and Age

Full size image at: Wikipedia

What we have here is a whole lot more empty space than thermometers, and especially so in the early years (that are not that long ago.)

Back to The Hypothetical Anomaly Wars

Now I can’t say if The Trouble with Tamino has actually “gotten it right” in how They work around the DATA bias with calculated anomalies. And frankly, I don’t really care. Yet Another Hypothetical Cow is completely unimportant to the world.

What matters is that the GIStemp code that folks are using to set policy DOES let the bias through, and the bias comes from imperfect anomaly mapping via “Basket A” to “Basket B” and from using the temperature data for a variety of adjustments and in-fill PRIOR to the anomaly process. ( Homogenizing, averaging USHCN and GHCN with an odd offset, UHI “correction” that often goes the wrong way, etc.)

The AGW Advocates have constantly asserted that “The Anomaly Fixes It” and I’ve constantly asserted that I care about what the data say, not the hypothetical anomaly they would like to imagine is used (but is not). So the first error is to assume that the hypothetical anomaly means anything to what I’ve asserted. It doesn’t. They then run off to show how the anomaly trends are the same for two sets of thermometers and then leap to more conclusions. OK, they have assumed a Hypothetical Spherical Cow and found that all steaks are round. There is supposed to be a surprise here?

What I have asserted is that the DATA (not any anomalies from it nor any other hypothetical cows) have a warming bias introduced INTO THE DATA SERIES (and nothing said about their anomalies) via station changes. I’ve demonstrated this bias in the gross averages of the data over time as compared to individual station data.

(Typically at this point, the AGW Advocates will assume yet another Hypothetical Cow and assert that the averaging of the data is a bad technique for measuring global warming trends and you really ought to do anomalies … so the steaks are round and I’m wrong. That completely misses the point that what I’m measuring is the DATA bias, not some theoretical global temperature construct. To measure what bias is in the DATA, you must look AT the DATA. That is how you find out what problems it has so you can know how to fix them.)

And I’ve even made a “first cut” at my own particular way to “fix the problem” using an anomaly technique similar to “First Differences” that I’ve taken to calling “dT/dt”. The difference is that with FD, each data gap results in a ‘reset’ of the difference to zero. I carry that difference forward until some data do show up. (Mine reacts less well to things like station changes, but preserves the trend across gaps better – at least up to about a 5 year gap.)

https://chiefio.wordpress.com/2010/02/09/dtdt-agw-ddt/

So contrary to the assertion that I don’t know how to ‘do it right’, I’m in fact following a very similar path of doing “Self to Self” anomaly processing to see how well that cleans up the data.

But the notion that showing “you can clean up the bias in a data set via a given process” somehow means that there is no bias IN THE DATA and that some OTHER processes will not have issues with that bias is just leaping off a cliff of conclusion to crash at the bottom of Hypothetical Cow Cliff.

So What Are The Biases in the DATA?

What I found when I looked was a large number of what look like systematic biases introduced into the temperature series in the DATA (with nothing at all said about their anomalies). So we get thermometers leaving the mountains and headed for the beach, and we get thermometers leaving the Canadian Rockies and headed for the Mexican Megathermal Zone. (Please note: not one hypothetical cow here. Actual counts of thermometers by altitude, by latitude, etc. I’m not fond of hypothetical cowburgers, so I avoid them whenever possible). And those changes do bias the base DATA set. They load up the present with hotter places and with places that have flatter seasonal profiles.

You can find the bulk of those investigations and the measurements of how much bias is introduced into the base DATA here:

https://chiefio.wordpress.com/2009/11/03/ghcn-the-global-analysis/

though some of the articles end up in this category listing before I get them added to that “consolidator posting” link above:

https://chiefio.wordpress.com/category/ncdc-ghcn-issues/

Now realize what this is NOT saying. It is not saying that the earth is getting hotter or colder. It is not saying WHY these thermometer changes happen (malice, stupidity, or?…) It is not saying that the TREND of these averages means anything about the TREND of the actual temperature of the planet or of the anomalies for a location. It is only saying that the DATA contains a bias. It is stating the problem, not the conclusion.

(At this point many AGW Advocates commence to tossing rocks about how I’m asserting that the TEMPERATURE of the planet is headed in some direction or other and that I’m doing it wrong because I’m not doing anomalies and grid boxes. That is the error of assumption They made above. They assumed the wrong premise, then attacked it wrongly. I’m “characterizing the DATA”, not talking about the Global Average Temperature. As I’ve said many times, the average of an intensive variable is meaningless anyway. It’s like counting the number of coins in your pocket and not looking at the denominations.)

So when we look at those DATA series, we find them fairly strongly biased by thermometer changes. That still leaves open the question of what impact that bias has on the various temperature series programs ( such as GIStemp and CRUt).

For what it’s worth, while many of the “Averages of the DATA” charts are fairly obviously biased by thermometer dropouts, you can still see some valid things about average temperatures in the graphs, charts, and tables. For long periods of time the thermometer counts can be stable. During those times, we typically see little that could be called “global warming”. In some places you do see warming in some times (often with the advent of airports and cities) but the pattern of the changes is not agreeable to the CO2 Thesis. But mostly you see massive swings from instrument change. There are 10’s of degree swings from instrument change in this global calorimetry experiment and we’re trying to find 1/10’s of degrees of meaning. That is the real lesson of inspecting these charts. The signal is very very well swamped by instrument changes. (There are also visible, in the data tables, single digit degree changes from periodic ‘ripple’ as what looks to be cyclical ocean currents change along with semi-random whole digit jumps from year to year variation. We’re trying to hear a whisper at a KISS Concert.)

One of my favorite examples where you can see things is the Pacific Region.
(Region 5 in the numbering scheme).

When we look at the Pacific, there is not a lot of room to change things. Islands are often small and with only one Airport. So you often get one change from a grass field to a jet tarmac tourist stop and that’s it. Islands, though they have Airport Heat Island bias, are in many ways “Islands of Stability” too. Looking into the pattern of the DATA BIAS, I noticed that it substantially showed up in Australia and New Zealand for the Pacific Region. So I did a set of reports on what the DATA say about the Pacific Basin, and found that with just the Islands being looked at, things are mighty darned flat:

https://chiefio.wordpress.com/2009/10/29/ghcn-pacific-basin-lies-statistics-and-australia/

Now the virtue of a graph like this one:

Pacific Basin - Ex- Australia; Thermometer Count and Temps

The virtue does not come from a belief that the ‘average temperature’ of the region actually started at 18 C then rose to 25 C and held darned near flat until today but with a mysterious plunge to 21 C smack in the middle of WWII. The value comes from realizing that this is the profile of the DATA (and NOT the temperature). It is telling you that the ‘start of time’ is very cold biased. It is telling you that the middle of WWII had a major dropout of warm stations as Japan overran the Pacific Theatre, but not New Zealand and Southern Australia. And if you look at it carefully, you can also see that as long as you have instrument stability, you get about a 25 C average. It is showing you that: stability of instruments matters and does so far more than actual temperature changes.

And what is the one thing we don’t have? Stability of the instruments.

Kiwis To The Rescue

As an interesting example, I used New Zealand. It is particularly interesting because there is a dropout of exactly ONE cold thermometer. It is in the baseline period, but it is missing in the recent past. It is Campbell Island, and it is a cold place. So when you look at the DATA profile for New Zealand with all the stations in, you find a nice warming bias introduced by dropping this one station and keeping all the others. I explored that in this article:

https://chiefio.wordpress.com/2009/11/01/new-zealand-polynesian-polarphobia/

where I showed that if you “stabilize the instruments” by leaving Campbell Island out of the whole time series, the base DATA in New Zealand have no significant warming signal present. If it isn’t in the basic DATA, it can only be introduced via the processes done to that data.

This chart shows the comparison of the average of New Zealand thermometer data with Campbell Island left in the baseline vs removing it from all of the history of New Zealand. I’ve “zoomed in” on the period from the start of the GIStemp baseline (1950) to date. Just as a reminder, the purpose of this graph is NOT so say this is the temperature of New Zealand. It isn’t. It is to say that this is the way the data look in aggregate for New Zealand. While we would expect a warming trend to cause values to rise over time: the absolute value is meaningless. ( The average of an intensive variable is meaningless, which is why a “Global Average Temperature” is meaningless, but that’s a topic for another day. Just think of it this way: the average number of coins in your pocket is meaningless if you don’t look at the denominations. For net heat gain on the planet, we need more than just temperature. Mass, specific heat, heat of vaporization and fusion of water are all missing.)

With that said, for a place of limited range, like New Zealand, we can depend on the “odds” being that most of the “coins in our pocket” will be dimes over much of the period in question… (Put less poetically, there are a limited number of stations in New Zealand and they are substantially airports at sea level in a limited latitudinal range. The major “ringer” is Campbell Island, a very cold southern island that is in the baseline interval but not in the recent history.) So we are looking at the biasing factors in the base data and we are looking at the trend over time with and without that bias.

New Zealand data with and without Campbell Island

As you can see, the major source of a rising trend in the New Zealand data comes from that one lone Island and it being in for a while, then left out.

This does not mean it is impossible to remove that bias in the data, but it does mean that the various ‘climate codes’ run on the data must be able to deal with that fundamental bias in the data. That has not been demonstrated. Hypothetical examples using pure “self to self” anomaly calculations do not answer this issue either. GIStemp does anomaly calculations by comparing “Basket with Campbell Island” in the baseline against “Basket without Campbell Island” now. It is simply assumed that the machinations of GIStemp will be sufficient to correct these two baskets to remove this basic bias. For the other climate codes, we do not have the source code to know how they calculate anomalies (but we do know that there are not a lot of choices as the base data are so sparse in many regions as to make “self to self” anomalies very limited in scope).

Taking an excerpt of the table in the New Zealand link, we can look closely at what happens in 2002 when Campbell Island is dropped:

Thermometer Records, Average of Monthly Data and Yearly Average
by Year Across Month, with a count of thermometer records in that year
--------------------------------------------------------------------------
YEAR  JAN  FEB  MAR  APR  MAY  JUN JULY  AUG SEPT  OCT  NOV  DEC  YR COUNT
--------------------------------------------------------------------------
1880 16.6 17.3 15.7 14.0 11.4  8.4  8.2  8.5 11.0 11.6 14.4 14.8 12.7   6
1881 15.4 17.0 15.6 14.1 11.8  9.8  9.1  9.1 11.3 11.7 13.1 14.9 12.7   4
1882 16.1 15.2 16.0 13.9 11.1  9.9  8.6  8.4 11.0 11.0 12.9 15.9 12.5   4
1883 17.2 17.6 15.8 12.5 11.3  9.4  8.2  8.9  9.5 11.0 12.0 14.2 12.3   4
1884 14.0 14.7 14.2 12.2  9.8  9.0  8.2  9.1 10.2 11.0 12.2 14.1 11.6   4
1885 15.1 16.2 15.4 12.7 10.5  9.9  8.3  8.8  9.8 11.2 13.0 14.1 12.1   4
1886 16.0 17.1 15.4 13.9 11.5  8.2  7.6  8.0  9.6 11.4 13.5 14.5 12.2   4
....
2000 16.2 16.5 15.3 13.8 12.1 10.2 10.0  8.5 10.9 12.3 12.5 16.2 12.9   9
2001 14.8 16.2 15.3 13.6 12.0  9.6  8.2  9.9 11.9 13.1 14.9 17.5 13.1   9
2002 17.8 16.9 16.9 14.3 12.3 11.2  9.6  9.5 11.1 11.0 12.8 15.2 13.2   9
2003 16.9 17.2 17.2 14.2 12.6 11.3  8.6  9.8 11.6 12.3 13.5 16.4 13.5   8
2004 18.2 17.2 15.4 12.6 12.5 10.8  8.9  8.8 10.6 12.4 14.5 14.0 13.0   8
2005 17.5 18.9 16.6 13.3 12.9  9.1  9.8 10.2 11.6 12.7 14.1 17.7 13.7   8
2006 17.9 17.8 15.1 15.4 12.2  8.5  9.4  9.5 11.8 12.6 14.2 14.4 13.2   8
2007 17.2 17.4 17.0 12.2 13.4  9.3  8.9  9.8 11.0 11.7 12.6 15.8 13.0   9

Most of the “low excursions” into 14 C and 15 C in January get ‘clipped out’ in 2002. The annual 11.x and 12.x in the right column becomes 13.x

We’re now consistentently in the 17-18 C range for January and 13.x for the year after the change. Until 2002, you have a more consistently low series.

Looking at a graph of all the data we see a generally flat trend running about 12 C then with a bit of a hockey stick at the end in 2002 when Campbell Island leaves the series. (Though the recent cooling has clipped the end of the stick).

New Zealand, All Stations, 1950 to Date

So I just took Campbell Island out of the whole thing and there is no more warming bias in the recent DATA. This graph shows the much flatter result from 1950 to date. (Prior to 1950 we still have the 12 C base due to a limited number of stations in the set, but once we have decent coverage, we have about 13.5 C and sit there.

New Zealand minus Campbell Island

So what does all this mean? It means that the bulk of all the rise of the numbers comes from instrument change, not from an increase at the individual stations. Further, it means that any temperature series codes (such as GIStemp) run on this data has a built in bias to deal with. We are depending on it to do a very good job…

But even more interesting, it says that with a reasonably stable coverage area, the DATA do not show increases over time conformant with steady increases in CO2. We have one steady number (with low thermometer count) followed by another stable number (with higher thermometer counts). The ‘recent rise” profile only shows up with the dropping of Campbell Island in 2002, though there is still enough instrument change to be masking some trend, one way or the other.

And you find this pattern repeated around the world, though with minor and interesting variations. “AGW” is not about CO2, it’s about instrument changes.

Here is a chart of the data from 1939 to date for folks who like to have the numbers. Particularly look at that right hand “YR” column:

Thermometer Records, Average of Monthly Data and Yearly Average
by Year Across Month, with a count of thermometer records in that year
--------------------------------------------------------------------------
YEAR  JAN  FEB  MAR  APR  MAY  JUN JULY  AUG SEPT  OCT  NOV  DEC  YR COUNT
--------------------------------------------------------------------------
1939 13.8 14.6 14.5 12.7 10.2  9.1  6.0  7.6  9.4 10.8 12.6 15.2 11.4   6
1940 16.5 14.6 15.8 13.2 11.8  9.9  9.3 10.6 11.5 12.5 13.6 16.2 13.0   7
1941 18.0 17.6 17.3 13.7 12.6  9.6  9.7  9.6 11.3 11.9 14.0 15.5 13.4   7
1942 16.1 17.1 16.1 15.2 13.2 10.9 10.4 10.5 11.9 13.3 14.5 15.7 13.7   7
1943 17.1 17.6 16.1 15.1 11.9  8.6  9.7  9.5 11.6 12.7 15.0 16.9 13.5   7
1944 17.5 18.2 17.0 15.5 12.2 10.3 10.0 10.0 10.9 12.9 14.1 15.3 13.7   7
1945 18.0 18.3 16.9 14.7 12.0  9.5  9.1 11.1 11.4 11.6 14.3 14.7 13.5   7
1946 16.4 17.5 16.4 14.9 13.3 11.1 10.5 10.5 11.7 12.4 12.4 15.0 13.5   7
1947 16.0 16.9 16.3 14.7 12.6 10.6 10.2 10.6 11.9 13.2 14.3 16.4 13.6   7
1948 17.9 16.0 15.8 13.9 11.8  9.6  9.5  9.4 10.6 11.5 13.0 14.6 12.8   8
1949 15.1 16.9 15.0 12.8 11.2  9.8  9.5  9.4 10.3 12.4 13.4 14.9 12.6   8
1950 16.3 16.1 14.6 12.7 12.2  9.1  8.4  8.6 10.1 12.0 13.5 14.7 12.4   9
1951 16.2 16.5 16.1 14.1 10.7  8.3  8.8  8.7 10.3 11.9 13.6 14.5 12.5  14
1952 15.8 17.4 15.4 13.9 11.6  9.9  8.6 10.2 11.1 12.6 13.9 15.8 13.0  14
1953 16.0 16.1 15.6 13.8 12.2 10.2  9.0  9.7 10.8 11.8 14.8 16.1 13.0  15
1954 16.6 17.5 16.9 13.7 12.3 10.4  8.9  9.3 10.3 11.9 14.7 15.6 13.2  15
1955 17.5 18.8 17.3 15.5 13.3  9.8  8.9 10.7 11.8 13.6 14.7 16.5 14.0  14
1956 19.0 17.8 15.9 16.8 12.7 11.0  9.6  9.9 11.5 13.5 15.1 16.6 14.1  14
1957 17.8 18.6 17.4 15.1 12.6  9.9  8.7 10.2 11.3 12.2 14.2 15.3 13.6  14
1958 16.5 18.5 17.3 13.3 12.0  9.9  8.6 10.0 10.7 14.0 15.0 17.1 13.6  14
1959 18.1 17.6 16.6 14.5 10.5  9.2  9.2  9.7 11.6 12.1 14.4 16.4 13.3  14
1960 17.3 17.5 15.8 14.0 12.3 10.5  9.5  9.6 11.3 13.4 14.4 15.3 13.4  14
1961 16.9 17.5 15.8 14.2 11.7 10.0  9.3  9.4 10.7 13.6 14.3 16.7 13.3  16
1962 18.1 17.4 16.7 14.6 13.7 11.1 10.0 10.3 11.2 14.1 14.5 15.9 14.0  18
1963 17.2 18.0 15.7 13.4 11.7  9.6  9.2  8.8 11.0 12.4 13.2 15.0 12.9  18
1964 16.2 16.9 16.1 13.7 11.4  9.9 10.0  9.7 10.9 12.4 13.5 16.4 13.1  18
1965 17.9 16.3 15.9 13.6 11.1  9.6  8.3  9.2 10.8 11.4 13.6 15.4 12.8  18
1966 16.8 18.3 16.6 14.3 11.2  9.5  8.8  9.0 10.7 12.1 13.5 15.4 13.0  18
1967 16.7 17.1 16.8 14.4 11.8  9.3  8.8 11.1 10.5 13.0 13.9 16.1 13.3  18
1968 16.8 17.0 17.8 14.5 12.6 10.5  8.8 10.3 10.5 12.1 13.5 14.9 13.3  18
1969 16.8 16.5 16.0 13.4 11.5  9.2  8.3  9.7 12.2 11.6 14.3 17.2 13.1  18
1970 18.2 17.3 17.2 14.9 11.5 10.5 10.0 10.6 11.8 13.0 14.4 16.1 13.8  18
1971 17.6 18.1 16.4 14.5 12.7 11.4  8.9 10.7 11.4 12.7 14.2 16.1 13.7  16
1972 16.3 16.2 17.0 14.3 11.6  8.5  9.3  8.7 11.6 12.7 15.2 14.8 13.0  16
1973 16.8 17.3 16.3 14.0 12.0 10.4  8.7 10.0 11.9 12.9 14.8 16.3 13.4  16
1974 16.6 18.9 15.4 14.8 12.1  9.9 10.0  9.6 11.8 12.7 14.9 17.4 13.7  16
1975 18.4 18.0 17.4 15.0 12.6  9.4  9.1 10.2 11.2 12.8 13.5 15.1 13.6  16
1976 16.9 15.5 16.3 14.3 11.4  9.1  9.2 10.2 10.5 11.8 13.0 15.8 12.8  16
1977 16.3 17.2 16.4 14.4 10.8  9.9  9.4  9.8  9.7 12.0 13.5 15.4 12.9  16
1978 17.5 17.9 16.7 15.6 12.3  9.7  9.7 10.3 10.9 11.7 14.1 16.1 13.5  16
1979 17.4 17.2 17.2 14.3 11.6 10.6  9.6  9.6 11.3 12.5 14.7 16.2 13.5  16
1980 17.4 17.5 15.6 13.9 11.9  9.9  9.0  9.7 11.7 13.1 13.2 15.7 13.2  16
1981 17.8 18.2 17.4 15.3 11.9 11.1  9.7  9.2 10.9 12.6 14.4 17.3 13.8  12
1982 17.6 18.3 16.4 13.2 12.3  9.6  8.8  9.4 10.8 11.4 14.9 15.3 13.2  11
1983 16.2 15.6 16.2 14.0 11.7 10.2  9.0 10.1 11.2 12.7 14.3 15.5 13.1  11
1984 16.1 17.7 17.3 14.2 11.4 10.6 10.0 10.4 10.9 12.2 15.0 17.4 13.6  11
1985 18.7 18.0 15.1 14.4 11.9 10.2 10.3  9.2 10.7 11.6 14.1 16.1 13.4  11
1986 18.3 18.5 16.5 14.9 12.5 10.3  8.4  8.7 10.0 12.5 13.6 15.9 13.3  10
1987 18.2 17.1 15.5 14.1 12.2 10.2  9.3 10.5 10.9 12.9 14.7 16.3 13.5  11
1988 17.2 18.2 16.1 13.4 11.8 10.5 10.3 10.1 12.0 13.3 15.1 16.9 13.7  11
1989 18.3 17.6 16.8 14.2 12.1 10.2  8.9 10.3 11.8 12.9 14.3 14.9 13.5  11
1990 16.5 17.7 16.2 13.6 11.5  8.8  8.4  9.2  9.7 12.2 13.9 15.7 12.8   8
1991 16.5 16.6 15.5 12.9 10.8  8.2  7.9 11.3 10.8 11.7 12.1 14.5 12.4   9
1992 16.8 16.4 14.1 11.4  8.6  8.0  8.7  9.2  8.5 11.6 14.3 14.9 11.9  10
1993 16.3 16.4 15.3 12.5 11.3  9.9  8.8  8.5  9.3 12.7 13.3 14.8 12.4  10
1994 17.7 17.6 15.2 13.0 11.3  7.6  8.1  8.2  8.2 11.1-99.0 15.7 12.2   9
1995 15.1 16.5 14.8 13.9 10.8  7.9  3.0  8.1  9.7 10.9 12.2 15.6 11.5   7
1996 16.5 16.3 15.0 13.3 10.1  7.6  7.4  7.2 10.5 11.2 12.3 14.7 11.8   9
1997 15.2 16.4 14.7 12.4 11.3  9.3  8.4  9.1  9.9 11.8 13.2 15.0 12.2   9
1998 16.9 19.5 16.6 14.1 11.8  9.0 10.1  9.3 11.2 12.9 13.8 15.7 13.4   9
1999 17.8 17.4 16.9 13.8 12.5 10.5  9.8  9.7 11.6 13.6 15.0 15.4 13.7   9
2000 17.1 17.4 16.1 14.7 12.7 10.9 10.5  9.0 11.5 13.0 13.4 17.3 13.6   8
2001 15.6 17.3 16.1 14.3 12.9 10.1  8.6 10.4 11.9 14.0 14.9 17.5 13.6   8
2002 17.8 16.9 16.9 14.3 12.3 11.2  9.6 10.1 11.8 11.5 13.5 15.9 13.5   8
2003 16.9 17.2 17.2 14.2 12.6 11.3  8.6  9.8 11.6 12.3 13.5 16.4 13.5   8
2004 18.2 17.2 15.4 12.6 12.5 10.8  8.9  8.8 10.6 12.4 14.5 14.0 13.0   8
2005 17.5 18.9 16.6 13.3 12.9  9.1  9.8 10.2 11.6 12.7 14.1 17.7 13.7   8
2006 17.9 17.8 15.1 15.4 12.2  8.5  9.4  9.5 11.8 12.6 14.2 14.4 13.2   8
2007 17.2 17.4 17.0 12.2 13.4  9.3  8.9  9.8 11.0 11.7 12.6 15.8 13.0   9
2008 15.9  9.6 12.8 12.8  9.7  9.2  9.2  9.2 11.8 12.5 14.4 16.7 12.0   9
16.9 17.1 16.0 14.0 11.6  9.6  8.9  9.5 10.8 12.3 13.9 15.6 13.0
16.6 16.7 15.6 13.6 11.3  9.2  8.5  9.1 10.6 12.1 13.6 15.4 12.7
[chiefio@tubularbells Temps]$
Such is the impact of One Little Island…

Pacific without Australia and New Zealand

Why this matters can be seen in this table that looks at the DATA of the Pacific Basin with the changing instrumentation of Australia and New Zealand left out. This is a fairly stable set of places in a fairly stable climate zone with all the UHI, pimples and warts left in. If there is no warming signal seen in this data, then it must come from the processing done to the data. And that is the true lesson that you learn from asking the DATA what they have to say, then listening politely to it. There is a strong instrument change signal, a very modest UHI / Airport Warming signal, but very little room left for a CO2 warming signal. I was going to make a graph of this, but a flat line just doesn’t look very impressive.

It doesn’t get much more dead flat than this. Just look at that left hand “YR” column. Wanders around 26.x from top to bottom.

[chiefio@tubularbells analysis]$ more Temps/Temps.LIST.yrs.GAT 

Thermometer Records, Average of Monthly Data and Yearly Average
by Year Across Month, with a count of thermometer records in that year
--------------------------------------------------------------------------
YEAR  JAN  FEB  MAR  APR  MAY  JUN JULY  AUG SEPT  OCT  NOV  DEC  YR COUNT
--------------------------------------------------------------------------
1880 25.7 26.6 26.7 26.8 27.2 26.3 26.3 26.8 26.9 27.0 26.5 26.0 26.6   3
1881 26.0 26.2 26.4 26.9 27.4 26.9 26.9 26.9 26.9 27.4 27.1 26.7 26.8   3
1882 26.9 26.7 26.7 27.2 26.9 26.7 26.6 26.6 26.9 26.7 26.5 26.5 26.7   3
1883 26.2 26.0 27.1 27.2 27.4 27.1 26.9 27.0 26.8 26.6 26.4 26.0 26.7   4
1884 25.5 25.8 26.4 27.0 27.1 26.6 26.1 26.7 26.6 26.8 26.6 25.6 26.4   4
1885 25.8 25.7 26.4 27.4 27.5 27.0 26.7 26.9 27.1 27.3 26.8 26.2 26.7   4
1886 25.5 25.2 26.2 27.2 27.6 27.2 26.9 27.2 27.3 27.1 26.3 26.2 26.7   5
1887 25.3 25.2 25.9 26.7 26.9 26.8 26.6 26.7 26.4 26.4 26.1 25.3 26.2   4
1888 24.7 25.2 26.6 27.7 28.0 27.5 26.7 27.1 27.2 27.2 27.0 26.2 26.8   4
1889 26.2 26.2 26.9 28.4 28.6 27.8 27.1 27.3 27.3 27.2 26.4 25.7 27.1   4
1890 25.7 25.7 26.3 26.8 26.9 26.6 26.2 26.5 26.4 26.1 25.5 25.7 26.2   5
1891 25.9 26.0 26.7 27.4 28.1 27.3 27.0 26.6 27.0 27.3 26.8 26.2 26.9   7
1892 25.8 26.4 27.0 27.0 27.7 27.0 26.8 26.6 26.9 26.8 26.3 25.8 26.7   7
1893 25.5 25.7 26.3 27.4 26.8 26.6 26.5 26.7 26.9 26.7 26.3 25.8 26.4   7
1894 25.4 25.5 26.3 27.0 26.9 26.9 27.0 26.6 26.6 26.8 25.9 25.6 26.4   7
1895 25.8 25.8 26.4 27.2 27.3 27.1 26.6 26.8 27.3 27.5 26.8 26.1 26.7   7
1896 26.4 26.6 27.0 27.2 27.3 26.8 27.0 26.7 27.1 27.6 27.5 27.0 27.0   5
1897 26.2 26.6 27.0 27.4 27.6 27.8 26.8 27.0 27.2 27.4 27.3 26.8 27.1   5
1898 26.5 26.6 26.8 27.4 27.6 26.6 26.8 26.9 27.1 27.0 26.8 26.6 26.9   5
1899 26.3 26.2 26.7 27.1 27.3 26.9 26.2 26.1 26.6 26.8 26.5 26.0 26.6   4
1900 26.1 26.4 26.8 27.2 27.2 26.7 26.2 26.5 26.9 27.1 27.4 27.0 26.8   4
1901 26.9 26.5 27.0 27.9 27.9 27.2 27.0 26.7 27.4 27.4 27.4 26.7 27.2   4
1902 26.7 26.2 27.0 27.5 27.7 27.2 27.1 27.0 27.2 27.6 27.4 27.5 27.2   4
1903 26.3 26.0 27.1 27.6 27.7 27.7 27.0 26.9 27.0 26.9 26.4 25.6 26.8   8
1904 25.3 25.1 26.4 26.8 27.1 26.7 26.4 26.4 26.6 26.8 26.5 26.0 26.3   8
1905 26.1 26.1 26.8 27.9 27.6 27.3 26.8 26.6 26.9 27.4 26.7 26.9 26.9   9
1906 26.8 27.1 27.2 28.0 27.7 27.4 27.2 27.1 26.5 26.2 26.3 26.1 27.0  10
1907 25.8 25.9 26.2 26.6 26.5 26.0 25.7 25.3 25.7 26.2 26.0 25.7 26.0  10
1908 25.9 26.1 26.3 27.0 26.3 25.8 25.5 26.2 25.8 26.1 25.8 26.3 26.1  10
1909 26.2 26.4 26.6 26.8 26.5 26.2 25.5 26.1 25.8 26.2 26.1 25.6 26.2  10
1910 25.9 26.1 26.3 26.6 26.2 25.8 26.0 25.7 26.0 25.9 25.6 26.2 26.0  10
1911 26.1 26.0 26.5 26.6 26.5 26.2 25.6 25.7 25.7 26.0 26.2 26.6 26.1  11
1912 26.3 26.5 26.9 27.0 27.0 26.5 25.7 25.8 26.2 26.1 26.1 26.0 26.3  11
1913 26.1 26.2 26.8 26.7 26.4 26.1 25.7 25.3 25.5 25.5 26.0 26.1 26.0  11
1914 25.7 26.3 26.8 27.2 27.0 26.2 26.4 25.4 25.7 26.0 26.5 26.7 26.3  11
1915 26.2 26.7 26.7 26.4 26.3 26.1 25.2 25.3 25.5 25.8 25.8 25.9 26.0  12
1916 25.5 26.2 26.1 26.3 25.8 25.6 25.3 25.6 25.2 25.4 25.5 25.5 25.7  12
1917 25.4 25.4 25.7 26.2 25.8 25.5 25.4 25.1 25.0 25.1 25.6 25.3 25.5  12
1918 24.7 24.7 25.5 25.8 25.7 25.6 24.8 25.0 25.1 25.4 25.6 25.6 25.3  12
1919 25.8 26.3 26.6 26.9 26.2 25.3 24.7 25.0 26.2 26.4 25.2 25.2 25.8  12
1920 25.5 25.8 25.9 26.3 26.1 25.3 25.1 24.9 25.1 25.1 25.4 25.7 25.5  11
1921 25.9 26.2 26.5 26.0 25.6 25.2 25.4 25.2 25.5 25.8 25.8 25.8 25.7  17
1922 26.1 26.2 26.5 26.7 26.2 25.9 25.2 25.4 25.6 25.9 25.9 26.1 26.0  18
1923 26.2 26.6 26.4 26.7 26.4 25.9 25.5 25.2 25.4 25.9 25.9 26.0 26.0  17
1924 26.2 26.5 26.7 26.7 26.9 26.2 25.9 26.1 26.2 26.1 26.2 26.0 26.3  18
1925 26.4 26.1 26.6 26.9 26.6 25.9 25.5 25.9 26.1 25.9 26.1 25.9 26.2  18
1926 25.8 26.4 26.5 26.8 27.0 26.2 25.8 25.8 26.1 26.3 26.4 26.2 26.3  17
1927 26.1 26.3 26.8 26.9 26.6 26.0 25.4 25.4 25.6 25.7 26.1 26.2 26.1  18
1928 26.5 26.4 26.9 27.0 26.8 25.8 25.4 26.0 26.1 26.1 26.3 26.5 26.3  17
1929 25.6 26.2 26.2 26.6 26.4 25.8 25.1 25.1 25.4 25.7 25.9 25.9 25.8  19
1930 26.3 26.2 26.8 26.8 26.6 25.9 25.7 25.6 25.7 26.1 26.3 26.2 26.2  24
1931 26.4 26.6 26.9 27.1 27.0 26.4 25.8 25.9 26.4 25.9 26.2 25.9 26.4  26
1932 25.9 26.3 26.5 26.8 26.7 26.1 26.0 25.9 25.9 26.2 26.2 26.3 26.2  29
1933 26.4 26.5 26.7 26.9 27.0 26.3 25.7 25.9 26.0 26.1 26.1 26.0 26.3  30
1934 26.1 26.1 26.4 26.9 26.8 26.7 26.0 25.8 25.9 26.2 26.3 25.9 26.3  31
1935 26.2 26.6 26.9 26.8 26.7 26.4 26.0 25.8 26.1 26.5 26.5 26.2 26.4  32
1936 26.3 26.8 26.9 27.0 26.7 26.1 25.8 25.6 25.9 26.0 26.2 26.5 26.3  32
1937 26.6 26.8 27.0 27.0 26.6 26.4 26.0 26.0 26.1 26.3 26.6 26.5 26.5  34
1938 26.6 26.7 26.9 27.1 26.7 26.4 26.3 26.1 26.0 26.2 26.1 26.2 26.4  36
1939 26.2 26.4 26.9 26.7 26.5 25.9 25.7 25.7 25.4 25.7 26.4 26.0 26.1  37
1940 26.0 26.3 26.7 26.9 26.7 25.9 25.7 25.2 25.5 25.9 26.2 26.1 26.1  40
1941 26.4 26.8 26.9 26.7 26.4 25.6 25.2 25.2 25.4 25.7 26.3 26.4 26.1  36
1942 26.8 26.9 27.0 26.4 25.7 25.1 24.2 24.4 24.7 25.5 25.7 26.1 25.7  22
1943 26.4 26.6 26.4 26.0 25.3 24.4 24.0 24.5 24.6 25.4 26.0 26.1 25.5  22
1944 26.3 26.3 26.5 26.0 25.4 24.5 23.7 23.7 24.4 24.7 25.3 26.0 25.2  20
1945 26.5 26.6 26.6 26.1 25.0 24.4 24.3 24.4 24.6 25.0 25.6 26.2 25.4  22
1946 26.8 26.7 26.8 26.6 25.6 25.2 24.9 25.1 24.8 25.5 26.0 26.5 25.9  31
1947 27.0 26.8 27.0 27.1 26.6 26.2 25.6 25.6 25.7 25.9 25.9 26.3 26.3  37
1948 26.3 26.4 26.8 26.8 26.4 25.9 25.5 25.3 25.5 25.9 25.9 26.1 26.1  44
1949 26.0 26.4 26.6 26.7 26.3 26.1 25.7 25.4 25.8 26.1 25.9 26.1 26.1  62
1950 26.2 26.3 26.5 26.6 26.3 26.2 25.5 25.6 25.6 26.0 26.2 26.1 26.1  70
1951 26.0 26.2 26.4 26.9 26.4 26.2 25.8 26.0 26.1 26.4 26.6 26.4 26.3 127
1952 26.4 26.5 26.7 26.9 26.9 26.5 25.9 25.7 25.9 26.3 26.4 26.1 26.3 131
1953 26.0 26.3 26.7 26.9 26.6 26.1 25.7 25.7 25.9 26.4 26.5 26.2 26.2 134
1954 26.3 26.3 26.5 26.8 26.6 26.2 25.8 25.7 25.8 25.9 25.8 25.8 26.1 136
1955 25.8 26.1 26.3 26.5 26.6 25.9 25.5 25.6 25.9 26.0 25.8 25.7 26.0 141
1956 25.7 26.1 26.4 26.5 26.4 26.0 25.6 25.6 25.7 26.1 26.1 25.9 26.0 149
1957 26.0 26.1 26.4 26.8 26.7 26.1 25.7 25.8 25.9 26.1 26.2 26.3 26.2 155
1958 26.3 26.4 26.7 26.9 26.8 26.4 25.8 25.8 26.0 26.2 26.0 26.0 26.3 157
1959 26.0 26.3 26.5 26.6 26.5 26.4 25.8 25.5 25.8 26.1 26.2 26.3 26.2 159
1960 26.1 26.1 26.5 26.8 26.7 26.2 25.8 26.0 26.1 26.2 26.3 26.2 26.2 191
1961 25.9 26.5 26.7 26.9 26.7 26.0 25.7 25.6 25.9 26.1 26.4 26.3 26.2 195
1962 26.1 26.0 26.5 26.7 26.8 26.3 25.9 25.7 26.0 26.5 26.4 26.1 26.2 202
1963 25.6 25.9 26.3 26.7 26.8 26.3 25.8 25.8 26.1 26.2 26.6 26.4 26.2 204
1964 26.7 26.5 26.6 26.9 26.7 26.3 25.9 25.9 26.1 26.2 26.1 25.9 26.3 206
1965 25.7 26.1 26.2 26.5 26.5 26.0 25.5 25.6 26.0 26.3 26.5 26.4 26.1 205
1966 26.2 26.4 26.7 27.0 26.6 26.1 25.9 26.0 26.2 26.3 26.4 26.2 26.3 210
1967 26.0 26.1 26.4 26.6 26.7 26.2 25.8 25.8 26.1 26.3 26.3 26.0 26.2 209
1968 26.0 26.0 26.6 26.6 26.7 26.4 26.0 25.9 26.2 26.3 26.3 26.3 26.3 202
1969 26.4 26.4 27.0 27.1 27.1 26.6 25.9 25.9 26.1 26.4 26.5 26.5 26.5 207
1970 26.5 26.6 27.0 27.1 27.0 26.5 25.9 25.9 26.2 26.4 26.4 26.3 26.5 203
1971 25.8 26.0 26.1 26.5 26.4 25.9 25.6 25.7 26.0 26.1 26.0 26.0 26.0 197
1972 25.8 26.3 26.3 26.6 26.6 26.1 25.8 25.7 25.9 26.3 26.5 26.5 26.2 201
1973 26.5 26.7 26.9 27.1 26.9 26.6 26.1 26.0 26.1 26.2 26.4 26.1 26.5 197
1974 25.8 26.0 26.3 26.6 26.5 26.1 25.8 25.8 26.1 26.2 26.3 26.0 26.1 197
1975 26.1 26.2 26.5 26.8 26.6 26.1 25.7 25.8 26.0 26.1 26.1 25.9 26.2 196
1976 25.9 26.0 26.4 26.3 26.2 25.6 25.2 25.2 25.3 26.1 26.3 26.2 25.9 141
1977 26.3 26.4 26.6 26.7 26.4 25.9 25.5 25.3 25.5 26.0 26.3 26.3 26.1 144
1978 26.3 26.4 26.8 26.7 26.7 25.9 25.6 25.4 25.6 25.9 26.1 26.2 26.1 147
1979 26.3 26.6 26.8 26.7 26.5 26.2 25.6 25.5 26.0 26.2 26.3 26.2 26.2 147
1980 26.5 26.7 26.8 26.8 26.7 26.1 25.7 25.6 25.8 26.2 26.3 26.3 26.3 146
1981 26.3 26.5 26.8 26.8 26.4 26.0 25.6 25.4 25.9 26.2 26.3 26.4 26.2 141
1982 26.5 26.5 26.7 26.6 26.3 26.0 25.3 25.3 25.4 25.9 26.3 26.4 26.1 110
1983 26.5 26.8 27.1 27.1 26.7 26.1 25.7 25.7 26.0 26.3 26.2 26.3 26.4 112
1984 26.2 26.5 26.7 26.9 26.6 26.0 25.6 25.6 25.8 26.1 26.4 26.3 26.2 113
1985 26.3 26.9 27.0 26.9 26.6 25.9 25.3 25.6 25.8 26.2 26.4 26.6 26.3 102
1986 26.5 26.5 26.7 26.8 26.6 26.0 25.7 25.6 25.8 26.2 26.5 26.3 26.3  78
1987 26.1 26.6 26.8 27.0 26.6 26.1 25.3 25.3 25.8 26.3 26.6 26.3 26.2  94
1988 26.8 26.9 27.2 27.1 26.9 26.5 25.7 25.8 26.0 26.2 26.1 25.9 26.4  92
1989 26.3 26.2 26.4 26.8 26.5 25.9 25.7 25.8 25.9 26.2 26.1 26.0 26.1  96
1990 26.4 26.8 26.9 27.0 26.7 26.1 25.8 25.6 25.5 25.9 26.3 26.4 26.3  95
1991 26.5 26.6 26.7 26.9 26.5 25.7 25.8 25.3 25.5 25.9 26.0 26.1 26.1 102
1992 26.2 26.4 26.8 26.8 26.7 26.2 25.8 25.9 26.0 25.9 26.0 26.2 26.2  74
1993 26.3 26.1 26.6 26.8 26.7 25.9 26.0 25.6 25.7 25.9 26.4 26.5 26.2  71
1994 26.5 26.7 27.1 26.9 26.6 25.9 25.5 25.6 25.5 25.7 25.9 26.1 26.2  71
1995 26.3 26.3 26.7 26.9 26.8 26.6 26.2 26.2 25.8 26.1 26.2 26.0 26.3  76
1996 26.1 26.2 27.0 26.8 26.3 25.8 26.0 25.7 25.8 26.0 26.3 26.2 26.2  81
1997 25.8 26.4 26.7 27.1 26.9 26.3 25.8 25.8 26.0 26.2 26.5 26.8 26.4  81
1998 26.9 27.2 27.5 27.6 27.6 26.8 26.2 26.3 26.1 26.9 26.6 26.6 26.9  71
1999 26.5 26.5 26.9 26.9 26.8 26.2 26.0 25.6 25.7 26.0 26.1 26.1 26.3  71
2000 26.3 26.8 26.8 27.0 26.6 26.2 25.3 25.5 26.3 26.6 26.4 26.3 26.3  75
2001 26.7 26.6 27.1 27.2 26.7 26.3 26.0 26.1 26.4 26.3 26.2 26.2 26.5  72
2002 26.5 26.6 27.2 27.1 26.9 26.6 26.5 26.2 26.1 26.4 26.5 26.7 26.6  71
2003 26.4 26.9 27.0 27.4 27.2 26.1 26.1 25.9 26.0 26.2 26.4 26.6 26.5  72
2004 27.0 26.9 27.0 27.4 27.0 26.5 25.9 26.1 25.9 26.2 26.5 26.6 26.6  78
2005 27.0 27.0 27.3 27.1 26.9 26.8 26.1 26.0 26.2 26.5 26.8 26.4 26.7  73
2006 26.5 26.7 27.2 27.2 26.8 26.5 26.2 25.9 25.9 26.1 26.7 26.8 26.5  75
2007 26.6 26.8 27.0 27.1 27.2 26.7 26.3 26.1 26.2 26.7 26.4 26.4 26.6  78
2008 26.3 26.2 26.5 26.8 26.5 26.4 25.7 25.8 26.3 26.5 26.8 26.6 26.4  86
     26.2 26.4 26.6 26.8 26.7 26.2 25.7 25.7 25.9 26.2 26.3 26.2 26.2
     26.2 26.3 26.7 26.9 26.7 26.2 25.9 25.9 26.0 26.2 26.3 26.2

For Country Codes 502 503 504 505 506 508 509 51 52 53 54
[chiefio@tubularbells analysis]$

If the Pacific Basin is not significantly warming from AGW, there is nothing “global” about any change that is happening. And if it isn’t in this base data, it can only be an artifact of the processing done to the data.

Is There A Way To See The Trend Buried In The Instrument Change?

Marshall Islands Airport - Where is the Thermometer?

So here we have the Marshal Islands Airport. Think there might be a bit of warming from all that “not a palm tree” construction? For what it’s worth, here is a view of the buildings at Majuro including the Stevenson Screen (on the far left, click on the link below the image to enlarge it, then zoom in to that building on the left). It’s the white box close to the dark exterior wall. So we have the thermometer next to the building, but with a large expanse of runway nearby. Depending on which way the wind blows you get exactly what?

Majuro site with Stevenson Screen on the left near a building / wall

Original full sized image.

There are several ways to take some of the instrument bias out of the data, but not all of it. Each with different “issues”. One of which I am particularly fond is called First Difference. It looks at a thermometer and uses the first value as a starting point, then measures each change year to year as an offset, a “delta” to the prior value. When data are missing, it resets to a new “starting point” when the data returns. In this way it will automatically compensate for things like instrument changes. But it does not do well on data with lots of “holes” in it (taking a lot of gratuitous ‘resets’ and losing some trend information). It is also rather sensitive to the starting value.

I’ve made a variation on FD that I’ve taken to calling dT/dt and then a variation on dT/dt that starts in the present (so that ‘start of time’ bias is removed and so we’re using our most recent records to set the starting point) and runs backwards in time. By definition, the present will be ‘near zero’ and the past will be seen as a rise or fall compared to us, now. Further, I don’t take a ‘reset’ on a data gap. Since the actual records are just chock full of dropouts, this makes dT/dt less sensitive to such dropouts of data ( it preserves tend information better ) but at the cost of also preserving such things as errors from instrument change bias and UHI effect. But at this stage, I’d rather see those things in the product and ask if they account for what is seen.

So using this tool, we can look at the New Zealand data and see if we can tease out some more valid information about what is actually happening. This is, substantially, the same as the Hypothetical Cow approach that others are taking. It is using an anomaly process (though most likely different from theirs) to see if we can find a hidden trend in this apparently flat basic data. And we do. (This is a good thing. There ought to be some UHI in the cities of New Zealand and since substantially all the thermometers are at airports, we ought to be seeing some impact from all the tarmac and lack of vegetation – even if there are not a lot of takeoffs and landings, the land change of tarmac taxi ways, runways, and parking area along with the removal of the vegetation ought to show some warming). So in this graph, we see about a 1.5 C max rise over the aviation age with a large bolus in 1990 or so when the ASOS type equipment is rolled out. ( I’ve run it on both the biased and the unbiased sets and the results are very similar. Not quite identical, so some of the station change bias can still leak through an anomaly process, but close enough). I don’t know exactly what equipment New Zealand uses, but the mod flags show a change. Notice also that the ‘take off’ point is at about the -1 C line near 1950. Flat before that, then warming 1 C as thermometers move from grass fields to airport tarmac. In sync with airport growth, not in sync with the industrial revolution nor all the fuel burned in WWII.

New Zealand, all data, dT

Also notice that 1934 is a cold time in New Zealand, while it was a warm one in the USA. This speaks to a probable oscillation of some sort between the hemispheres. Since we have really bad spacial coverage of the souther hemisphere, in the early years nearly none, such an oscillator could easily show as a bias in the data we do have, especially if the temperature is treated as a single global thing instead of looking for patterns by continent, or by country, or even by smaller divisions.

OK, so this tool can take basic data, with or without bias, even apparently flat data, and find an underlaying trend ( that is of about the magnitude we would have expected for this location, biased as it is by Airport Heat Islands and the conversion of instruments to the electronic type where the ASOS has been shown to have bias.)

At this point, I’m fairly comfortable with using this tool on a larger set. What does it say about the whole Pacific basin, even WITH Australia and New Zealand included? What can it do to handle that WWII dropout, for example? Note that time runs from recent on the left to long ago on the right.

Pacific Basin Delta T, All Data

Not Bad, if I do say so myself. We are able to find the AHI effect we would expect from all those Airports In The Sun as the Jet Age started planting loads of runways all over the Pacific and we see relative stability during the ‘grass field’ era prior (while the Industrial Revolution was pumping out loads of CO2). Though interesting is that “rise” at the start of time. Looks like it was warm just before GIStemp starts it’s baseline… Also of interest is that 1915 spike to about where we are now. Nothing new under the sun, it would seem, other than some tarmac.

What About A Bigger Place? The USA

Using the same tool on the USA gives a graph that matches what we know of history:

USA All Data Delta T

Here we see the warm 1930’s followed by the “swoon” into the GIStemp baseline between 1951-1980. More interesting in many ways is further back in time. We have a fairly cold mid 1800’s, but a much warmer period early on in the 1700s (but always cycling about the zero line). And we can see how ‘cutting off time’ in 1880 as GIStemp does imparts it’s own “sample bias”. Toward the far past we are getting down to very few thermometers and the volatility starts to rise (less dampened by averaging).

Now we could argue about the “sparsity” of thermometers in the 1700s, but then we would have to address the sparsity today… And we could talk about the instrument error in the 1700s, but then we would need to address all the siting issues found by SurfaceStations.org along with the ASOS bias found in those instruments used at those modern airport locations. And I’d be just fine with that; because, IMHO, it’s all about “Instrument Change” in our little global calorimetry experiment, and not about CO2.

And we could even talk about Hypothetical Cows and my own little experiment in an anomaly process based program. But then we would also need to ask why GIStemp and the other temperature programs fail to find that ‘flat rolling’ pattern but instead find a rising hot pattern for the USA. And I’d be OK with that too… since it was, and is, my original goal.

In Conclusion

So what’s the point of all this?

I said the data had biases introduced into it from thermometer drops, and it does.

I’ve said that bias is toward warming, and it is (with lots of links provided to reports demonstrating that).

I’ve said that when you look at samples with that bias removed, we find the ‘CO2 warming signal’ leaves with it, and demonstrated that with the entire Pacific Region where warming matches the transition from sea planes and grass fields to tarmac and jet ports, but is flat during the industrial revolution.

I’ve said that Hypothetical Cows tell you nothing about real world processes in other codes, and shown that that is true.

Further, I have my own Hypothetical Cow that shows that They have “over averaged” and hidden the interesting bits in the regions; and I’ve shown that CO2 is not needed to explain where we are today.

And at the end of it all, They have a nice Hypothetical Cow, but it has nothing to do with me, my work, nor even very much to do with the real world processing done by the actual temperature series used for policy decisions.

Hypothetical Cows are just not very important. And that is why I generally avoid wasting time admiring them. Besides, I like a Porterhouse steak, not round processed patties of mystery meat …