Guest Post by Willis Eschenbach
[see Update at the end of this post]
I got to thinking about the (non) adjustment of the GISS temperature data for the Urban Heat Island effect, and it reminded me that I had once looked briefly at Anchorage, Alaska in that regard. So I thought I’d take a fresh look. I used the GISS (NASA) temperature data available here.
Given my experience with the Darwin, Australia records, I looked at the “homogenization adjustment”. According to GISS:
The goal of the homogenization effort is to avoid any impact (warming or cooling) of the changing environment that some stations experienced by changing the long term trend of any non-rural station to match the long term trend of their rural neighbors, while retaining the short term monthly and annual variations.
Here’s how the Anchorage data has been homogenized. Figure 1 shows the difference between the Anchorage data before and after homogenization:
Figure 1. Homogenization adjustments made by GISS to the Anchorage, Alaska urban temperature record (red stepped line, left scale) and Anchorage population (orange curve, right scale)
Now, I suppose that this is vaguely reasonable. At least it is in the right direction, reducing the apparent warming. I say “vaguely reasonable” because this adjustment is supposed to take care of “UHI”, the Urban Heat Island effect. As most everyone has experienced driving into any city, the city is usually warmer than the surrounding countryside. UHI is the result of increasing population, with the accompanying changes around the temperature station. More buildings, more roads, more cars, more parking lots, all of these raise the temperature, forming a heat “island” around the city. The larger the population of the city, the greater the UHI.
But here’s the problem. As Fig. 1 shows, until World War II, Anchorage was a very sleepy village of a few thousand. Since then the population has skyrocketed. But the homogeneity adjustment does not match this in any sense. The homogeneity adjustment is a straight line (albeit one with steps …why steps? … but I digress). The adjustment starts way back in 1926 … why would the 1926 Anchorage temperature need any adjustment at all? And how does this adjust for UHI?
Intrigued by this oddity, I looked at the nearest rural station, which is Matanuska. It is only about 35 miles (60 km) from Anchorage, as shown in Figure 2.
Figure 2. Anchorage (urban) and Matanuska (rural) temperature stations.
Matanuska is clearly in the same climatological zone as Anchorage. This is verified by the correlation between the two records, which is about 0.9. So it would be one of the nearby rural stations used to homogenize Anchorage.
Now, according to GISS the homogeneity adjustments are designed to adjust the urban stations like Anchorage so that they more closely match the rural stations like Matanuska. Imagine my surprise when I calculated the homogeneity adjustment to Matanuska, shown in Figure 3.
Figure 3. Homogenization adjustments made by GISS to the Matanuska, Alaska rural temperature record.
Say what? What could possibly justify that kind of adjustment, seven tenths of a degree? The early part of the record is adjusted to show less warming. Then from 1973 to 1989, Matanuska is adjusted to warm at a feverish rate of 4.4 degrees per century … but Matanuska is a RURAL station. Since GISS says that the homogenization effort is designed to change the “long term trend of any non-rural station to match the long term trend of their rural neighbors”, why is Matanuska being adjusted at all?
Not sure what I can say about that, except that I don’t understand it in the slightest. My guess is that what has happened is that a faulty computer program has been applied to fudge the record of every temperature station on the planet. The results have then been used without the slightest attempt at quality control.
Yes, I know it’s a big job to look at thousands of stations to see what the computer program has done to each and every one of them … but if you are not willing to make sure that your hotrod whizbang computer program actually works for each and every station, you should not be in charge of homogenizing milk, much less temperatures.
The justification that is always given for these adjustments is that they must be right because the global average of the GISS adjusted dataset (roughly) matches the GHCN adjusted dataset, which (roughly) matches the CRU adjusted dataset.
Sorry, I don’t find that convincing in the slightest. All three have been shown to have errors. All that shows is that their errors roughly match, which is meaningless. We need to throw all of these “adjusted datasets” in the trash can and start over.
As the Romans used to say “falsus in unum, falsus in omnibus”, which means “false in one thing, false in everything”. Do we know that everything is false? Absolutely not … but given egregious oddities like this one, we have absolutely no reason to believe that they are true either.
Since people are asking us to bet billions on this dataset, we need more than a “well, it’s kinda like the other datasets that contain known errors” to justify their calculations. NASA is not doing the job we are paying them to do. Why should citizen scientists like myself have to dig out these oddities? The adjustments for each station should be published and graphed. Every single change in the data should be explained and justified. The computer code should be published and verified.
Until they get off their dead … … armchairs and do the work they are paid to do, we can place no credence in their claims of temperature changes. They may be right … but given their egregious errors, we have no reason to believe that, and certainly no reason to spend billions of dollars based on their claims.
[Update – Alaska Climate Research Center releases new figures]
I have mentioned the effect of the Pacific Decadal Oscillation (PDO) below. The Alaska Climate Research Center have just released their update to the Alaska data. Here’s that information:
Figure 4. Alaska Temperature Average from First Order Observing Stations
In the Alaska Climate Research Center data, you can clearly see the 1976 shift of the PDO from the cool to the warm phase, and the recent return to the cool phase. Unsurprisingly, the rise in the Alaska temperatures (typically shown with a continuously rising straight trend line through all the data) have been cited over and over as “proof” that the Arctic is warming. However, the reality is a fairly constant temperature from 1949-1975, a huge step change 1975-1976, and a fairly constant temperature from 1976 until the recent drop. Here’s how the IPCC Fourth Assessment Report interprets these numbers …
Figure 5. How the IPCC spins the data.
SOURCE: (IPCC FAR WG1 Chapter 9, p. 695)
As you can see, they have played fast and loose with the facts. They have averaged the information into decade long blocks 1955-1965, 1965-1975, 1975-1985 etc. This totally obsures the 1975-1976 jump. It also gives a false impression of the post-1980 situation, falsely showing purported continuing warming post 1980. Finally, they have used “adjusted data” (an oxymoron if there ever was one). As you can see from Fig. 4 above, this is merely global warming propaganda. People have asked why I say the Alaska data is “fudged” … that’s a good example of why.
Willis Eschenbach (15:29:36) :
“You, like Nick, are missing the point. I know how the data was fudged. ”
You are missing the first point. That your reader would have little idea of how the algorithm works, or what inspired it. Should the reader have to wade through a hundred comments before he finds a discussion of what the algorithm actually is, or even a statement that there even is an objective algorithm?
“What I don’t understand is how this is all justified. I keep asking for a reason that anyone would start adjusting a pristine rural record in 1920.”
I have told you the answer to that question. Maybe it was nice and rural in 1920, but their satellite data tell them it isn’t, now. Unless they implement a method like NOAA’s, they can’t figure out when any impact due to urban warming came about. So they essentially toss the station out. I would say this is a sign of being rather over-eager in trying to avoid UHI, if anything. I prefer the GHCN method, over GISS. Especially once GHCN v3.0 is implemented.
“Provide the reason for the adjustments made to Matanuska.”
It got flagged as urban, and that was the end of Matanuska. That’s all. I’d rather they use the NOAA methods and retain more information from Matanuska, but they instead choose to minimise the impact Matanuska as on the trends at the grid point.
“We know the method, which is to force them to agree with their neighbors. But what is the reason to fudge the data that way? ”
Again in GISS, the adjustment is done to remove the impact of the questionable station. As for GHCN, as has been shown very clearly, neighboring stations correlate very well with each other. So if there is a non-climatic influence/event/distortion at one station, you can get an idea of what the climate signal would have been, by looking at the neighbors.
“”Provide the reason for the adjustments made to Matanuska.””
By the way, even if you try to retain the station, you’ll never have enough historical metadata to track down the reason for every discontinuity. Especially outside the US. But a good statistical method will more-or-less correct the errors you know about, as well as those you don’t. It won’t be perfect, but nothing would be.
“It was fudged by a computer algorithm, one that obviously doesn’t work well.”
You keep saying that. You’ve done nothing to actually demonstrate this.
3×2 (16:11:49) :
To reduce the complication, I don’t think these stations are in USHCN. USHCN is a lower 48 thing.
Robert (10:34:27) :
Until then, I don’t think there’s anything left to say.
Please tell me that is in fact your last offer(ing)!
Nick Stokes (16:01:30) :
“to preserve some short term information. I don’t think that is much of a gain; it will, have very little effect at all.”
I feel the same way. I’ve always wondered if much is gained by keeping the short term variations.
Willis Eschenbach (16:02:02) :
Yes, in this case there is correlation in the variance, but the trends don’t match. So they ignore the trend at any station classified as urban. The end.
You should be happy; there’s not much chance of UHI creeping through if you just axe the urban station’s trend.
Of course, it isn’t perfectly axed, but it comes close enough.
Re: 3×2 (Feb 22 16:11),
USHCN (again, as far as I can see) has already adjusted for station moves
USHCN does not cover Alaska.
carrot eater:
Matanuska. The history of the area is well known. There was no development there before the war. The computer adjusted the temperatures during a time when there is no reason to adjust them. Therefore, the algorithm obviously doesn’t work well.
carrot eater (16:15:58) : edit
I assume that the reader … well … reads. If the reader reads, they would have read this in my original post:
So at that point they know there is an algorithm. As to what “the algorithm actually is”, that was beyond the scope of my article. I’m not interested in overly complicating my original post by trying to explain the intricacies of how the algorithm weights the nearby stations by using the formula
distance / -500 +1
If someone wants to know that, they can read further. But it’s far too much detail for my original article.
Don’t like it? Fine. You are welcome to write an article about how wonderful the GISS algorithm is, and all the details of its operation. Then you can explain how reasonable it it to knock three quarters of a degree off of a pristine rural record simply because (gasp!) it had a different trend than its neighbours.
Let me know when you do, post the URL up here, and I’ll come and ask questions. And I won’t attack you for not writing it the way that I would write it …
carrot eater (16:24:05) : edit
Axed??? In Matanuska, whatever UHI there might have been was increased by the adjustment, at the rate of 4.4°C per century. Not sure how you translate that to “axed”.
Willis Eschenbach (16:50:36) :
You’re just repeating yourself now.
First, “The history of the area is well known” is not necessarily true. There is almost never a station that has every single possible bit of relevant historical metadata logged, except for the new US CRN stations. Do you know what station moves there were, TOB changes, instrument changes, shelter changes?
Second, none of that matters because the GISS method isn’t trying to correct for specific inhomogeneities at Matanuska. It’s simply trying to erase it in a sense, in order to minimise the effect it could have on the long term trends at that grid point. For the whole record. Just to be safe, and ueber-cautious. How many times must that be repeated?
Now this may prove to be so over-cautious that over time, they lose so many rural stations to nightlights that they end up undersampled. This would be a problem, but it isn’t a problem in Alaska now.
You are having a forest/trees problem. No adjustment scheme is perfect, and none could be perfect because the required information simply doesn’t exist.
But the global mean simply isn’t that sensitive to these minor differences in processing. You aren’t impressed by the match between GHCN adjusted, GISS adjusted and CRU? Then how about the match between GHCN adjusted and GHCN raw? Also a good match, if you look globally.
Re: Rural stations as drivers in the case of Anchorage and Matanuska.
Can we just be clear here. On the Matanuska data source (some (me) believe).
First. Put the co-ordinates 61.566031,-149.250255 (cut and paste) into Google maps, get the satellite view. Can we take the urban/rural debate from there (all from the same hymn sheet) because I am getting a little confused reading some of these posts.
Next. GISS does not use macro information for station adjustments – it does not have it. That leaves us with internal GISS adjustments, the obvious one being UHI.
Finally. Unless the state of Alaska has re-defined the OED entry for urban, Matanuska is as rural as most any station gets.
If we now accept that Matanuska is rural (don’t know how much more rural it can be short of sticking it in space) – why is any U(rban) H(eat) I(sland) adjustment required at any time for any reason? Even if adjustments were required, GISS does not have the information required to make them.
So – given just the temperature data and the station co-ordinates – you return a complete mystery. WUWT?
Willis Eschenbach (17:06:57) :
Matanuska is ‘axed’ by forcing its long term trends to match its rural neighbors. Using the two-legged mechanism, this can never be done entirely, but the point is to come close enough that removing Matanuska entirely would have little impact on the long term trends.
How close are they coming to this ideal? Well, until you finally put up a complete analysis of surrounding rural stations, you won’t know, will you?
Eschenbach: “What I don’t understand is how this is all justified. I keep asking for a reason that anyone would start adjusting a pristine rural record in 1920.”
Carrot Eater: “I have told you the answer to that question. Maybe it was nice and rural in 1920, but their satellite data tell them it isn’t, now. Unless they implement a method like NOAA’s, they can’t figure out when any impact due to urban warming came about. So they essentially toss the station out. I would say this is a sign of being rather over-eager in trying to avoid UHI, if anything. I prefer the GHCN method, over GISS. Especially once GHCN v3.0 is implemented.”
———
Why make any adjustments at all? The adjustments introduced are entirely artificial, as there is no indication of when urbanization occurred. Why not experimentally verify current UHI by taking multiple meaurements several times a year to determine what the temperature difference is between the thermometer site and rural areas? Perhaps this could be done by field monitoring every 5 years? Otherwise all you’re doing is introducing fudge factors and guessing at trends, which you then confirm via statistical sleights of hand.
How do you know when the UHI effect began to make a difference to the temperature readings when you backfill adjustments into the records? Why then should we believe the graphical representation of past temperatures, which are supposed to show historical trends? Good science still remains observational, although I know that many university-based scientists would prefer to have sensors piping in electronic data that can be monitored at the lab. So much for studying climate, if your entire focus is now on electronically mediated models!
I’m beginning to understand all these statistical and homogenization practices much harped on by Carrot-Eater et al. and described here, and can’t say that I’m impressed by what I read when they finally condescend enough to explain what they are on about. These methods do not build trust as they are based on assumptions (esp. viz. the satellite monitoring of light intensity to determine the urban status of a surface station location), and ignore the value of local science and observations. The entire construct is artificial, and as far as I can tell, that is what Carrot Eater, Robert and their ilk are trying to obfuscate.
Since the whole argument is about global warming, ultimately, surely the raw data could be tested against current urban heat island effect, and any difference in extra warmth over a set period chosen as the origin could then be attributed to climate change? The ultimate requirement is that the sites chosen as the origin points should all be rural at the beginning. In fact, why not focus only on rural sites? (Of course, that would apparently never do for GISS)
The big problem, as has been pointed out elsewhere, is that the surface stations were not conceived or designed for the purpose to which the AGW scientists have turned them. And of course, the alarmist political atmosphere that surrounds this science does not help. What really needs to be done, in order to gain trust, is for the entire exercise to be recommenced with stations located and designed for the purpose of detecting long-term warming and cooling trends (not a meteorological concern prior to the 1960s). Even then, the discovery of such trends says nothing about the human role in global warming or cooling. Meanwhile, from what I have read at this site and experienced in my daily life, very real ‘heat pollution’ does emanate from cities and human construction, and this is going virtually ignored by global warming theorists.
Willis and Stephen, your collective patience amazes me.
3×2 (17:08:31) :
Generally correct, all around. This may be a case where the nightlight procedure has improperly flagged a station as being non-rural. That’s the danger with any such method; it will misdiagnose a station now and then.
But if you’re really worried about UHI, you’re more worried about false negatives than false positives. So we return to Willis’s point about African cities which are dark by night, but still very much cities. What to do about them? I think that’s a good question, and I’ve not read the papers about nightlights carefully enough to see if this is discussed.
carrot eater (11:14:24) :
1,2,3 – or a really good idea would be to stop relying on bulk processing algorithms that give you “something in the expected range” and start dealing with individual stations manually – case by case.
While we have bulk processing and results such as Darwin, Matanuska or wherever, you are not presenting a convincing argument. The method(s) used by GISS plainly has flaws. Overlook them if you wish but HARRY_READ_ME.txt says a lot about QC in the field.
Re: 3×2 (Feb 22 17:08),
If we now accept that Matanuska is rural
No, GISS uses an objective criterion – night brightness. It’s actually a good criterion. Arguing about population is missing the point, because that’s not a good indicator of UHI either. What you want is a measure of local heat release, and artificial lighting is a good indicator. Maybe the satellites are getting it wrong, but it isn’t GISS.
And so, carrot eater (Feb 22 17:19), I think that relates to the issue of dark African cities. They are populous, but may not be evolving a lot of heat.
I think we should bear in mind, too, that the brightness criterion can be related to where the temp station actually is. If it is associated with a big city but some distance from the bright lights, the brightness criterion can pick that up.
There was a time in my past when I modeled (primarily as a SWE) ocean circulation underneath hurricanes. It was always the case that you started with the physics and mathematics it generates. Models were simply the means to see if you had the math right, i.e. can my model demonstrate some correlation to actual observation? If it correlates well enough (it never did) you’ve got the math right. If the correlation is off you go back to the math because you’ve got the physics wrong.
In this entire debate on AGW I just don’t see that kind of ‘science’ happening. In fact it’s more like the entire debate has been hijacked into a different domain where everybody obsesses over the observations while ignoring the physics. Where’s the math that connects night time radiance to temperature? Where’s the math that connects asphalt parking lots to temperature sensors? I posit that it doesn’t exist and never did.
The warmists are simply making use of lazy proxies because the real deal is hard or impossible.
It reminds me more of painting a picture than building a model. My wife, an itinerant artist, keeps laying on colors until she gets the picture she wants. She gets amazing results with no math at all. These models are simply laying on adjustments until they get the result they wanted all along.
In the end, if your science doesn’t create new knowledge or know how then you’re just a movie director aren’t you?
Not billions, not trillions but tens of trillions of dollars are at risk over this hoax! Last I heard was in the vicinity of $24 trillion (US); whereas it will supposedly cost us $22 trillion if we do nothing about it! Hmmm…
This has been my problem all along. If that farm is the data source (and I believe it is) I might go as far as suggesting that the night light procedure does not work. Following from that, if a farm in the middle of nowhere like Matanuska can trigger the process what else has been mis-diagnosed in the current run?
Nick Stokes (16:35:12) : (and CE earleir)
Re: 3×2 (Feb 22 16:11),
USHCN (again, as far as I can see) has already adjusted for station moves
USHCN does not cover Alaska.
Fine, so v2_mean it is then. Any adjustments have therefore been made “blind” during the GISS process. I was rather hoping that you would suggest that adjustments had been applied using actual metadata held by USHCN.
Blind adjustments to a completely rural station for algorithmic reasons it is then. Couldn’t you have said that earlier?
carrot eater (17:07:50) : edit
Sorry, I thought you were following the thread. These were cited above, at Dominic Marcello (09:26:43).
So if analysis showed a breakpoint at any of those dates, sure, I can see adjusting those. But obviously, that’s not what GISS did at all.
However, that wasn’t my point, which I obviously didn’t make clear, my apologies.
My point was that in 1930, Matanuska was rural, and that is well known from the history. And despite the nightlights, it is still rural. On Google Earth, you can see the Agricultural Experimental Station at 61.5656N, 149.249W. Anyone who says that’s urban hasn’t turned on his own nightlights.
So a) Matanuska’s not urban now, and b) regardless of what it is now, it wasn’t urban in 1920 or 1930 or 1940 or 1950. So what justifies adjusting those years? I repeat myself because I still don’t have an answer to that question.
What I don’t understand is how this is all justified. I keep asking for a reason that anyone would start adjusting a pristine rural record in 1920.
I suggested one which you ignored, the relative proximity of a glacier.
Carrot Eater:
No, no, no. They didn’t “toss the station out.” They introduced an artificial warming of 4.4°C per century starting in 1970. This artificial warming is averaged into the final claim of “recent global warming”. How is introducing that huge artificial warming tossing the station out?
That’s why I said it was “fudged”. They didn’t “toss the station out” in any sense.
Nick Stokes (17:46:37) :
I thought about that, but am unconvinced. UHI effects are more than just waste heat, but also change in materials (different heat capacities and albedos), hindered convection or radiation, reduced evaporative cooling, and so on. It’s a complicated brew.
A dim African city can still have some of these items. I need to read up a bit on that literature sometime.
3×2 (17:45:59) :
As I’ve said before, manual adjustments to every last station seems not feasible, not necessarily desirable, and almost certainly not worth the effort over the ~1200 regular reporting stations in GHCN and then again for USHCN.
One, you can use some human judgment, but you’ll never have all the information you need. Then, you’ll end up making a lot of ad-hoc decisions that nobody else could reproduce. And then the people on this site would really go off – it isn’t reproducible, it isn’t science, the adjustment guy has his finger on the scales, etc. And what would it really gain you? Some oddball stations might be treated a bit better; the regional and global trends would rather likely be about the same.
And as it turns out, statistical methods for adjustment can be better than manual ones. For example, the change from old thermometers in the US to the MMTS stations. This used to be dealt with (in USHCN, now) using a set adjustment described in Quayle (1991). But it turns out, this instrument shift doesn’t always have the same effect because the reasons why the instrument shift affects the temp reading varies from station to station. A human would have some trouble dealing with this, but the statistical routine can sort it out.
Willis Eschenbach (18:43:42) :
“No, no, no. They didn’t “toss the station out.” They introduced an artificial warming of 4.4°C per century starting in 1990. How is introducing that huge artificial warming tossing the station out?”
You mean 1970? Anyway, I’ve said this maybe a dozen times now. If you make it such that the longer trends roughly match the other surrounding stations, then the station in question is not affecting the longer trends at that grid point. In effect, it is being tossed out.
So again, what you need to do to assess this is actually look at those surrounding rural stations.
To assess the calculation, you’d need to use the GISS method to find the temperature at that grid box, with and without Matansuka. If the GISS adjustment did what it’s supposed to do, then the longer trends for the combined record at that grid box would not be affected by adding or removing Matansuka.
Willis Eschenbach (18:38:20) :
NOAA’s only going to have that detailed sort of metadata for a US station, so you’ll need to have a method that doesn’t require historical metadata for non-US data, no matter what. Anyway, things like “Thermometers moved but no information where 5/14/48” is hardly the same as “pristine” data. As you admit, it could require some adjustments.
But yes, GISS isn’t even trying to explicitly or individually adjust for those things.
“So a) Matanuska’s not urban now, and b) regardless of what it is now, it wasn’t urban in 1920 or 1930 or 1940 or 1950. So what justifies adjusting those years? I repeat myself because I still don’t have an answer to that question.”
You’ve been given the answer several times. Maybe it isn’t urban now, but it fails the nightlight screen, so it’s possibly questionable. When did it start being possibly questionable? Who knows. You’d need a detailed statistical analysis of the temp record in comparison to neighboring stations, along with tons of historical metadata, to know for sure. NOAA tries to work this out (though without any historical metadata, now). GISS says, forget it, we won’t even try; the neighboring stations are good enough.
Basically, you just keep arguing for an adjustment procedure that requires a ton of manual work, ad-hoc manual decisions, a ton of historical metadata that won’t be available for each station, and no real indication of a significantly improved global result.
If you can just accept for the moment that such an adjustment process just isn’t going to happen, then you see the choices are the GHCN way and the GISS way. And maybe you don’t care, but the fact that they take such different approaches, and still end up with consistent results really does mean that the results are not that sensitive to the details of the processing. Which is a good thing.
Photos of the Matanuska AES:
1917
1918
1920
1930
Air view, 1938
Unknown date, pre-1959