New paper blames about half of global warming on weather station data homogenization

Stay braced for the team of rapid response and little science, how dare they!! …sarc.

Statistical homogenisation as practiced currently, without experimental justification is totally out of whack with reality. Better methods need to be developed.

“Conclusions”
# 2 nails it !!

Very good. Glad to see this peer-reviewed. Results exactly as expected. See my page.
Now can we look at getting Climate Science (the scientists, communicators, and politicians) to embrace the Twelve-Step Program?

But…. wasn’t this the whole point of BEST?

Next subject: a systematic comparison between stations dropped and not dropped during the last decades of the twentieth century.

The full presentation is excellent, beautiful, graphic, comprehensible, and full of statistics too. I hope Steve McIntyre runs a thread to get confirmation of its statistical significance. It doesn’t mention your own groundbreaking work here, but I’d like to think that recognition is implicit in the focus on things like discontinuities, instrumentation, and actual comparisons of some of the worst differences between raw and homogenized temperature graphs.

“Hansenko”
Ouch! 🙂

It’s good to see research like this (which disputes the monopolistsic consensus) is being accepted and published.
The problem statement in their presentation bothered me a bit because it seemed to say that if two nearby instruments’ readings differ, then one of them must be wrongr:

The problem
● Historical and contemporary climatic time series contain inhomogeneities –
errors introduced by changes of instruments, location etc.

What if the weather was different at the two locations? But, reading further, I saw that “microclimate changes” are considered in this process:

Homogenization methods do not take into consideration some characteristics of
hydroclimatic data (long-term persistence, microclimatic changes, time lags).

I can offer a “miicro” example of one of these microclimate changes, from the data of my NWS/CWOP station.
http://www.findu.com/cgi-bin/wxpage.cgi?call=AF4EX
On July 15 you can see a rather large and rapid drop in mid-afternoon temperatures (10F decrease in a 2-3 hours) caused by a small local rain shower. Yesterday (July 16) an even bigger drop, due to a big shower (almost 2 inches of rain)
But other stattions saw it differently. Two nearby CWOP stations, CW2791 (2.4 miles) and DW1247 (12.6 miles) both reported the July 16 anomaly, but DW1247 didn’t report a big anomaly on July 15 because it didn’t report a mid-afternoon shower.
http://www.findu.com/cgi-bin/wxnear.cgi?call=AF4EX
Of course all such readings are subject to measurement error, and these CWOP stations certainly can’t claim perfection in their accuracy. But it should be clear that the large July 15 temperature anomaly at AF4EX was “real weather” and only observable within a radius of a few miles.
I believe that these mesoscale readings are important, for example, for observing and predicting squall lines, derechos and such.
Also,I don’t believe that instruments in large cities, subject to the urban-island heating effects, should be moved. They should report the heat retained and re-radiated from out planet from these warmer areas. But these readings should be weighted, with larger cooler rural areas having more weight, to give a more accurate picture of the planetary radiation balance.

“weather station data homgenization” homgenization?
Add this to the UHI effect and it does not leave much, if any, warming trend at all>

Seems like there are paper(s) ascribing about 40-50% of the original increase value to natural variation. Now is that 50% of the total – what ever that might be – or is that about 0.4 deg C ? If it’s the 0.4 deg C, then aren’t we just about out of warming? Thermostat controls work by allowing small variations it the controlled variable to set the values of the control variables. Because of this and because of the fact that there should be some temperature rise caused by co2 concentration, there should be something left over in the way of an increased temperature. Otherwise, we are headed in the wrong direction and in for the truly serious consequences of cooling.

Makes me feel all warm and fuzzy. Recall that I recently pointed out that one perfectly reasonable interpretation of the recent “run” of 13 months in the top 1/3 of all months in the record, warming trend or not, is that the data is biased! That is, a very small p-value argues against the null hypothesis of unbiased data. That is all the more the case the more unlikely the result is made, so the more “unusual” it is — especially in the absence of anything otherwise unusual about the weather or climate — the more the p-value can be interpreted as evidence that something in your description of the data is horribly wrong.
At that time I thought about the temperature series and almost did a similar meta-analysis (but was too lazy to run down the data) on a different thing — GISS and Hadley have, IIRC, made a number of adjustments over the years to the algorithm (secret sauce) that “cooks” the raw thermometric data into their temperature series. I fully admit to having little but anecdotal recollection of a few of them, mostly from reading about them on WUWT. Anthony has duly noted that somehow, they always seem to warm the present and cool the past, on average.
Let’s say there have been 8 of those adjustments, and all of them warmed the present and cooled the past to the point where they quantitatively increased the rate of warming over the entire dataset but especially the modern era. There is no good reason to think that the thermometers used back in the 30’s were systematically biased towards cold temperature — the default assumption is (as always with statistics) that the error in thermometric readings from the past with the exception of cases that are clearly absurd or defective series (the one kept by the alcoholic postmaster of Back-of-the-Woods, Idaho, for example, back in the 1890s) is unbiased and indeed, probably normal on the true reading. In particular, it is as likely to be a positive error as a negative one, so the best thing to do is not adjust it at all.
In this case the probability that any sort of data correction will produce net warming — even of a little tiny bit — should be fifty-fifty. This is the null hypothesis in hypothesis testing theory: If the corrections are unbiased, they have an even chance of resulting in net warming.
Next one computes the probability of getting 8 warmings in a row, given the null hypothesis: . This is the probability of getting the result from unbiased corrections. Even if there have been only 4 corrections (all net warming) it is 1/16. It’s like flipping a coin and getting 4, or 8, heads in a row.
In actual hypothesis testing, most people provisionally reject the null hypothesis at some cutoff. Frequently the cutoff used is 0.05, although I personally think this is absurdly high — one in twenty chances happen all the time (well, one in twenty times, but that’s not that infrequently). 4 in a thousand chances not so often — I’d definitely provisionally reject the null hypothesis and investigate further upon 8 heads in the first 8 flips of a supposedly unbiased coin — that coin would have to behave very randomly (after careful scrutiny to make sure it didn’t have two heads or a magnet built into it) in future trials or betting that it is unbiased is a mug’s game.
TFA above has extended the idea to the actual microcorrections in the data itself, where of course it is far more powerful. Suppose we have 3000 coin flips and 2/3 of them (2000) turn out to be heads. The null hypothesis is that the coin is unbiased, heads/tails are equally likely p = 0.5. The distribution of outcomes in this case is the venerable binomial distribution, and I don’t have to do the actual computation to know the result (because I do random number generator testing and this one is easy). The p-value — probability of getting the result given the null hypothesis is zero. The variance is or 3000/4 = 750. The square root of 750 is roughly 27. The observation is 500 over the mean with n = 3000. 500/27 = 18, give or take. This is an 18 sigma event — probability a number close enough to zero I’d get bored typing the leading zeros before reaching the first nonzero digit.
Now, I have no idea whether or not there are 3000 weather stations in their sample above, but suppose there are only 300. , , , . The probability of a six sigma event is still — wait for it — zero. I wouldn’t get bored, exactly, writing the zeros, but the result is still in the not-a-snowball’s-chance-in-hell-it’s-unbiased range.
Visibly, there look like there could easily be 300 dots in the map above, and frankly, who would trust a coin to be unbiased if only 100 flips produced 2/3 heads? One is finally down to only a 3 sigma event, sorta like 8 heads in a row in 8 flips, p-value of well under 1 in a 100. I don’t know Steirou, but Koutsoyiannis is, I predict, going to be the worst nightmare of the keepers of the data, quite possibly eclipsing even Steven McIntyre. The guy is brilliant, and the paper sounds like it is going to be absolutely devastating. It will be very interesting to be a fly on the proverbial wall and watch the adjusters of the data formally try to explain why the p-value for their adjustments isn’t really zero, but is rather something in the range 0.05-1.00, something not entirely unreasonable. Did our drunken postmaster weathermen in the 1920s all hang their thermometers upside down and hence introduce a gravity-based bias in their measurements? Silly them. Is the modern UHI adjustment systematically positive? I never realized that cities and airports produces so much cooling that we have to adjust their temperatures up.
Or, more likely, it is simply something that Anthony has pointed out many times. The modern UHI adjustments are systematically too low and this is biasing the trends substantially. One could even interpret the results of TFA is being fairly solid proof that this is the case.
rgb

If I remember correctly, some of the alarmists were saying that solar forcing could only account for about one third of the warming seen this century. If one third is solar, and one half is an artifact, then what’s left for the anthropogenic component? This was assuming that there was zero internal forcing from ocean currents. What’s the latest estimate from the alarmists for the solar component?

The data doesn’t need homogenization. If one location is hotter or colder than a neighbouring location, that’s weather. Raw data is the only data that’s worth anything. Once you bend the data out of shape, it becomes worthless.

The two citations are plainly wrong.The abstract is not a ”new peer reviewed paper recently presented at the European Geosciences Union meeting.” It is an conference abstract by E. Steirou and D. Koutsoyiannis, which was presented at the session convened (organized) by Koutsoyiannis himself.
http://meetingorganizer.copernicus.org/EGU2012/oral_programme/9221
Conference abstracts are not peer reviewed, you cannot review an abstract of 260 words. At EGU all abstracts, which are halfway about geoscience are accepted. Their purpose is to select, which people get a talk and which ones get a poster presentation.
REPLY: I was of the impression that it was “in press” but I’ve change the wording to reflect that. Hopefully we’ll know more soon. – Anthony

Something that troubles me about the USHCN v2 homogenization process is that it apparently assumes a linear relationship between station temperatures. That is OK when dealing with measurement errors but not, in my opinion, when creating the influence of each station’s temperature in the areal product.
Considering that heat flows from hot to cold, it would seem to me that a more representative temperature relationship between a hotter station and a colder one would be a function of the square of the distance between the stations. The quantity of heat represented by the difference between the two station temperatures disburses radially from the hot station and the temperature of the surface layer air would drop in proportion to the diffusion of the differential heat. For instance, the temperature difference at a point half way between the hot station and the cold station would be 1/4 the two station difference. With the current process, the temperature difference half way would be 1/2 the two station difference.
If vertical mixing is also considered, the influence of the hot station on the path to the cold station would be even more quickly diminished. My opinion is that the current homogenization process tends to give too much weight to the hot stations on surrounding areas. The process appears to be manufacturing additional heat where none exists which then is misinterpreted as global warming due to CO2.
But maybe I am missing something in my understanding of the homogenization process.

Misspelling! Homogenization not homgenization

Even if one attributes all (yes, 100%, LOL) of that 0.42 degrees Celsius trend to CO2 increase that would put sensitivity to 2XCO2 between 1 and 2 degrees Celsius.
Assuming:
dF = 5.35ln(CO2f/CO2i)
dT=S(dF)
Using values @ the century endpoints:
dF = 5.35ln(375/300)=1.19
S=0.42/1.19×3.7=1.3
And allowing for some lag in the system(s) by using a CO2 from several years prior to 2000:
dF = 5.35ln(360/300)=0.97
S=0.42/0.97×3.7=1.6
More evidence that the 3-5 degree Celsius sensitivity to 2XCO2 claim is exaggerated.

I suppose the question I have is what algorithm did they use for “homogenisation” and do they use it on
1) “suspect” stations only
2) all stations using the original neighbouring station data.
I guess case 1 otherwise the effect would be a spatial smoothing as in case 2 (again assuming on the routine: sounds like simple IDW mean). Case 2 wouldn’t have a bias unless there were more anomalously high temperature stations than low temperature ones. But then this would give you the suspected spatial average anyway.
Anyways, this seems like bad stats especially if they didn’t do a QC plot (interpolated vs actual) to see how the “high fideility” stations fair when their values are interpolated from other nearby – high fideility stations – using the same/similar algortihm. If they didn’t do this then the adjustments are all BS based on just a load of assumptions.

Re: Victor Venema says:
July 17, 2012 at 6:45 am
“The two citations are plainly wrong.The abstract is not a ”new peer reviewed paper recently presented at the European Geosciences Union meeting.” It is an conference abstract by E. Steirou and D. Koutsoyiannis, which was presented at the session convened (organized) by Koutsoyiannis himself.”
What about it Anthony? I’d already blown this one out to everyone I knew calling it “peer reviewed and presented at….”

Bill Illis says:
July 17, 2012 at 6:51 am
This is why the NCDC and BEST like using the Homogenization process.
“The Steirou and Koutsoyiannis paper is about GHCN Version 2. The NCDC has already moved on to using GHCN-M Version 3.1 which just inflates the record even further.”
==================================================================
All please read this as Bill illis is 100% correct. Version 3.1 is far worse. Also, the TOBS comment is spot on.
———————————————————————————————-

Lucy Skywalker says:
July 17, 2012 at 4:25 am
Very good. Glad to see this peer-reviewed.

It appears to be a seminar presentation. These are not typically peer reviewed. I don’t know about this one, but I doubt it.

Actually, I would expect that unbiased homogenization would result in more temperature trends being decreased, not an equal proportion of increases and decreases (because one of the goals of homogenization should be to erase UHI). So, I would suggest that the temperature record is more corrupt than the authors suggest.

Erm, this wasn’t published or peer reviewed (just presented at a conference). I have no idea how they got the results they did, since simply using all non-adjusted GHCN stations (in either v2 or v3) gets you pretty much the same results as using the adjusted data. Both the Berkeley Group and NCDC have relatively small adjustments on a global level:
http://rankexploits.com/musings/wp-content/uploads/2010/09/Screen-shot-2010-09-17-at-3.20.37-PM.png
http://curryja.files.wordpress.com/2012/02/berkeley-fig-2.png
What is the justification for choosing the specific stations that they did? I could easily enough pick 181 stations that show that homogenization has decreased the trend, but I’d rather just use all the stations rather than an arbitrary subset to avoid bias.
REPLY: I was of the impression that it was “in press” but I’ve change the wording to reflect that. Hopefully we’ll know more soon.
I wouldn’t expect you’d consider any of this, so I’m not going to bother further right now. – Anthony

Should send such papers to Dr. Angela Merkel. She has warned these days that global temperature could increase by 4°C soon, as her chief climate change advisor, Mr. Schellnhuber, has told her. And many Germans are proud to show the world how to run a de-carbonized economy successfully … one day.

Isn’t this all a bit immaterial anyway. The whole thing about measuing the Earth’s average temperature seems crazy to me. Measuring changes on the order of a few tenths of a degree is futile given that the methods employed – if I understand them.
First there is the spatial average. How do you grid temperature data and at what resolution? Do you use IWD mean, natural neighbours, BSpline, Kriging (and which type), declustering (cell vs polygonal) etc.
Then you have to decide whic hprohection system you use or whether you use angular distances but then which geoid do you use.
It’s bias upon bias.
Surely this would be just funny if so much money wasn’t spent on it.

Not being a climatologist or meteorologist, I have little understanding of the logic behind where monitoring stations are placed. I do know from observing my car’s outside temperature indicator that on exceptionally calm, sunny days the indication varies significantly between treeless areas and wooded areas. This variation is minimal, if even detectable at all, on windy or cloudy days.
If the intent of a monitoring station is to measure a temperature representative of the air temperature in the general area in a way that is not influenced by wind speed or direction and cloudiness, then the task is not at all a simple straight forward one.
As a former metrologist, thoroughly familiar with thermometers and their calibration, I would think quality instrumentation was used at monitoring stations, making random instrumentation errors small compared to the variations experienced due to the physical location of the monitoring station. I am guessing that the homogenization process used only corrects for random errors – those having an equal probability of being positive or negative, and will not take out a bias due to physical location.

There’s some sort of Climate Change Uncertainty Principle that the super high temps are reality and happening but not where they’re being measured, and a cat.

This probably sounds like a broken record by now. But I’ve been claiming for about four years that half of the warming trend is due to “adjustments”. And a portion of the remainder is due to coming out of an LIA. This again reinforces the climate sensitivity numbers that were produced by Lindzen and Spencer. The climate sensitivity is somewhere between .5C and 1.2C per CO2 doubling.

How are Steirou and Koutsoyiannis regarded by the warmists? Are they regarded as sceptics, or what?

The Fundamental Premise, that “Homogenization is necessary to remove errors introduced in climatic time series,” is intellectually dishonest.
It presumes a priori knowledge of what “error-free” data would look like, and totally disregards the physical fact of natural variability and its contribution to the uncertainty of whatever ‘trends’ are later inferred.
Either ‘smoothing’ OR ‘trending’ can be performed on a raw data set, but to perform BOTH is wrong and inevitably produces misleading results.

Careful Zeke, realclimatescientists tried the same sort of attacks before against Koutsoyiannis and got their ears pinned back.
http://climateaudit.org/2008/07/29/koutsoyiannis-et-al-2008-on-the-credibility-of-climate-predictions/
He published an updated version in 2009 or 2010 to shut them up. I’m just sayin’

Since UAH satellite temperatures show an increase of ~0.4 deg over the past 30 years then ALL warming over the past century must have been since the 1970 – i.e just about the time the CO2 effect would be expected to become distinguishable from natural variability.
This paper does nothing to debunk the CO2 effect. On the contrary it suggests that TSI, Svensmmark, PDO and other natural effects are negligible over time. Note Satellite readings are not contaminated by UHI.

@West, Mosher
According to this
http://www.ipcc.ch/ipccreports/tar/wg1/345.htm
the IPCC shows 3.5°C ECS from equlibrium to equilibrium (of temperature and CO2) for a doubling of CO2 in 70 years. The lag is hundreds of years. IPCC-TCR (from equlibrium to end of doubling after 70 years) is 2°C.
Since CO2 was already increasing in the year 1900, the first calculation of West goes from CO2 increasing to CO2 increasing, which gives the same temperature difference as from equilibrium to equlibrium (3.5°C according to IPCC for doubling, thus years 70 to 140).
Therefore I see no need to consider any lag and think 1.3°C is the West-ECS.
(P.S. I assume that IPCC-ECS is the same thing as Mosher-ECR)

Steven Mosher says:

John West.
You just calculated the transient climate response. ( TCR) at 1.6.
the ECR ( equillibrium Climate response) is anywhere from 1.5 to 2x higher.
so if you calculate a TCR ( what you did) then you better multiply by 2…
Giving you 3.2 for a climate sensitivity. (ECR)

No, I calculated the TCR @ 1.3 & took a stab @ ECR (minus feedbacks) by taking out about a decade of CO2 increase. Admittedly, not a “climate science” approved method, but for some reason I think around 10 years is plenty enough time to realize the temperature difference from a change in forcing, considering we see temperature changes daily and seasonally from changes in forcing with short lags (see Willis’ post); and whether feedbacks are positive, negative or neutral is still a very open question in my mind. Why should I accept (on faith?) an extremely long lag and net positive feedbacks when I can’t observe that in the real world and have no substantial evidence for it? It’d be like believing in the North American Wood Ape (Bigfoot). Anyway, taking your method of 1.5-2x the ECR would be from 1.95 to 2.6 IF all the warming is due to CO2 increase.

Alexej Buergin says:

temperature difference as from equilibrium to equlibrium
Excellent point!
The temperature in 1900 is in equilibrium [as much?] as the temperature in 2000 and [as much?] as the temperature will be in 2100. Hmmm.

If one is buying a thermometer, and expects any degree of accuracy at all, it would be wise to check its calibration at two points bracketing the temperature range of interest.
After all, a broken clock is right twice a day.

Steven Mosher says:
July 17, 2012 at 8:07 am
John West.
You just calculated the transient climate response. ( TCR) at 1.6.
the ECR ( equillibrium Climate response) is anywhere from 1.5 to 2x higher.
so if you calculate a TCR ( what you did) then you better multiply by 2…
Giving you 3.2 for a climate sensitivity. (ECR)
>>>>>>>>>>>>>>>
You cannot make this statement unless you know to some degree of precision what the time constant is, which you do not. Further, even if you knew the time constant for one particular forcing, you must also know the time constants for all other forcings that are active to the point of still being significant, what their sign is, and how far along each is in terms of a total of 5 time constants. There are WAY too many factors all in place at the same time, we know the time constant of pretty much none of them, let alone which ones are at the beginning of the cycle and which ones are at the end.

John Finn
Since UAH satellite temperatures show an increase of ~0.4 deg over the past 30 years then ALL warming over the past century must have been since the 1970 – i.e just about the time the CO2 effect would be expected to become distinguishable from natural variability.
This paper does nothing to debunk the CO2 effect. On the contrary it suggests that TSI, Svensmmark, PDO and other natural effects are negligible over time. Note Satellite readings are not contaminated by UHI
John…….I’m not sure you meant to let that cat out of the bag did you? You are saying then that all this paper is suggesting is that the general assumptions of the surface records are 0.3 – 0.4 ‘higher’ than they should be? Fair enough…I actually agree with you. Which means then that the surface record should really match UAH at 0.4C since 1979. Fair enough, I agree with you.
Funny thing though, is we have those same surface records showing half a degree swings in the early and mid part of last century….as you say, without any CO2 input, so natural variation can swing 0.4C. Or…..if we junk half the range of surface records as suggested above….natural variability equals around 0.2C……..that leaves you some room for CO2 since the 1970s to give you another 0.2C.
hmmmm……….isn’t that pretty much what Lindzen etc have been saying all along…CO2 effect trivially true but essentially meaningless? Forgive me if I don’t get too worked up about 0.2C over 30+ years….

There is another aspect to temperature trends once explained to me by Steven Mosher when he was demoloshing an error I had made in regard to analysing NASA/GISS gridded data. I’ve been thinking about it since, and it seems this would be an appropriate thread to bring it up in.
IIRC, Mr Mosher explained to me that GISS doesn an “in fill” of missing grid data provided that over some percentage (50% I think) is avaiable for that grid cell for the time period in question. As an exampled, if a grid cell had temps of 10, nodata, and 11, GISS would for those three time segments, “in fill” the nodata with the average of the other two, for a temp of 11.5. At the time, it explained how GISS could show a value in a grid cell truncated to a specific point in time when taking a different time segment would show that grid cell as being empty. But the over all method has bothered me even since and this thread reminded me of it.
If the explanation as I understand it is correct, then is can have no other effect on a global basis across the entire time series but to warm the present and cool the past. That the earth’s temperature has been warming since the LIA is reasonably well accepted. If that is the case, we have to consider that there are is no more data being added to the “cold end” of the GISS record. We’ve got what we’ve got. But there IS data being added at the WARM end of the graph which is in the present. Each year, this extends the to timNe series of gridded data for which we have more than 50% of the gridded data over the entire time period. This in turn allows for older data with missing grid cells to be “in filled” that would otherwise be blank. But because the “in fill” is predicated upon new data that is BY DEFINITION warmer than the old data, the linear trend that calculates the previously empty grid cell in at the beginning of the temperature record must calculate an increasingly colder temperature for that cell. To illustrate:
1, 2, 3, 4, 5, 6 (year)
N,10, N, N 10, N (deg)
In this series, we have 6 points in time, with 4 missing values. The series as a whole is not a candidate for “in filling”. However, if we ran GISS’s reporting program to look at only poiints 2,3,4 and 5, we would meet the 50% threshold, and report that grid cell considered for those time periods ONLY as 10, 10, 10, 10.
Now let’s add a one more years of data” assuming newer years are warmer than older years”
1, 2, 3, 4, 5, 6, 7 (year)
N, 10, N, N, 10, N 12 (deg)
We still cannot “in fill” the whole record because we have less than 50% of the data points. But what if we were to only look at the last four? GISS would “in fill” those by calculating a trend from the two data point that alread exist giving:
4, 5, 6, 7 (year)
9, 10, 11, 12 (deg)
Looks almost sensible, does it not? But wait! what if we looked only at years 2,3,4,5 and applied the same technique? We’d get 10, 10, 10, 10! Year 4 would be a 10, not a 9! By adding one more year of data to the original 6, and looking only at the last 4, we can “cool” year 4 by one degree. Let’s expand the data series with some more years:
1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 (year)
N, 10, N, N, 10, N, 12, 13, 14, 14, 14. (deg)
This is where I think the “system” as I understand it starts to fall down. If one looks at years 5 through 11, “in filling” a value of 11 deg for year 6 makes a certain amount of sense. What gets broken is that we now have enough data to in fill for the enture series. So, while if one looked at years 1 through 5 as raw data, one would most likely conclude that the first 5 years or so were “flat”. But in fill based on a linear trend due to the recent years being added to the record and “in fill” the whole thing, one would get something like:
1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 (year)
9, 10, 9, 9, 10, 11, 12, 13, 14, 14, 14. (deg)
By all means, please correct me if my understanding of this whole snarled up system of “in filling” is wrong. But if I do understand it correctly, I don’t see how temps where are higher today being added each year to the temperarture record can have any other effect than to “cool the past” as the number of grid cells with data gets added in warmer periods of times and so increasingly biases the extrapolation to earlier time periods when we have less data. If we were adding data at the same rate to both ends of the scale, this would make some sense. But since we’re adding data at the WARM end (the present) only, cramming a linear trend through the data to “in fill” data at the “cold” end can only cool the past, and with no real justification for doing so.

Willis Eschenbach says:
“Victor, thanks for your comments. Nobody is saying that the inhomogeneities do not have a bias. What was said was that if you have a number of phenomena biasing the records (you point out several different ones above), you would expect the biases to cancel out, rather than reinforce. They may not do so, but that has to be the default assumption.”
As explained in a bit more detail above radiation errors in early instruments explain much of the effect and has a clear bias. With a small number of causes for inhomogeneities and these changes happening to most stations in a network, it would be very unlikely that they all cancel out.
Willis Eschenbach says:
“So for you to now claim that “homogenization improves the quality of climate data” is a total misrepresentation of the results. According to your study, some types of homogenization improved the temperature data, most didn’t improve the precipitation data, and people often misused the homogenization software.”
Thank you for the correction. Yes, the word “temperature” would have been more accurate as “climate”. Precipitation is a real problem, not only for homogenization also for climate modeling. As the discussion here was about temperature I had not thought of that.
I think the statement that homogenization improves temperature data is a fair one-sentence summary. Scientist know which the good algorithms are and these algorithms are used to homogenize the important data sets. For example, the method you care about most, the one used to homogenize the USHCN dataset performed very well. Some more obscure or new methods produced problems and some people homogenizing data for the first time made the data more inhomogeneous. In a scientific study you mention such detail, in a normal conversation you typically do not. For the details you can read the open-access study.

John Finn says:
July 17, 2012 at 10:00 am
Since UAH satellite temperatures show an increase of ~0.4 deg over the past 30 years then ALL warming over the past century must have been since the 1970
mikef2 says:
July 17, 2012 at 11:17 am
Funny thing though, is we have those same surface records showing half a degree swings in the early and mid part of last century….as you say, without any CO2 input, so natural variation can swing 0.4C.
I agree with Mike. They say a picture is worth a thousand words. Here is the ‘picture’.
http://www.woodfortrees.org/plot/hadcrut3gl/from:1900/plot/hadcrut3gl/from:1912.33/to:1942.33/trend/plot/hadcrut3gl/from:1982.25/to:2013/trend

Steven Mosher;
Sure one can make this statement with a precise knowledge of the time constant. You need to do more reading on how these can be estimated from data. You dont need precision and we dont have precision, by the ECR is like to be 1-2x of the TCR. Fingers crossed there is a really good paper on this, hoping it gets submitted by the end of the month. Until then you get to figure out the math on your own.
>>>>>>>>>>>>
I’ll be very interested in the paper when it comes out. In the meantime, I submit to you that I am capable of calculating a time constant if, and ONLY if, I have sufficient data on ALL the processes involved and EACH of their time constants and EACH of their progress through 5 time constants. I think that’s pretty problematic. We’ve got dozens, perhaps hundreds or thousands of physical process all going on at the same time, and many causing feedbacks both positive and negative to each other. Isolating ONE factor (forcing from CO2 doubling for example) from all the others requires that I know what all the others are, what their time constants are, and when they started, and how they related to each other from a feedback perspective. This makes the hunt for park matter and the Higg’s Boson look like Grade 1 arithmetic. Now knowing what ALL the forcings are, what the time constant for EACH is, and WHEN each began makes the calculation of any single forcing from the data a fools errand. IMHO.

Mosher: “Look if people want to twist and turn the numbers to make this century as cold as the LIA”
NOAA has 30 states where the warmest month is in the 1890s. 11 of those were May.
Why should I believe you that the temperature was coming up from a lower value before 1895?
Is it possible that it was warmer for some period of time before 1895? If the US temperature record started in 1890 or 1880 would there before records from those decades?
http://sunshinehours.wordpress.com/2012/07/11/noaa-warmest-months-for-each-state-june-2012-edition/

John Day,
As I understand it, the kind of event you describe – a low temperature on a single day because of a localized thundershower – is not the sort of thing anyone attempts to “homogenize” out of the record. It is clearly a legitimate weather event.
What I believe people are looking for are instances where, for example, someone plants a large area of forest surrounding the weather station, where previously it stood in open fields. So the microclimate of the station changes, not just transiently but permanently. But the surrounding climate has not changed. So to take the changed temperatures as evidence of changed climate in that region would be wrong.
This is why most climate analyses look for evidence of changes in siting, instrumentation and microclimate by looking for step changes (not transient spikes) in the records from one site that are not matched by similar step changes in records from other nearby sites. They then attempt to correct for those changes, or throw the data out altogether.
Again, as I understand it (which isn’t very far) the BEST approach had the great merit of not attempting to “homogenize” any data series by adjusting the numbers. They still looked for suspicious step changes in data series, but whenever they found them, they simply split the historical series into two – one before and one after – and treated them as though they were completely separate weather stations.
Other people know a lot more about this than I do and I’m sure they will correct me if I’m wrong. I’m also sure there are some good and easily understood introductions to this area out there on the web, and I may now go and try to find one to refresh and expand my knowledge in this area.

Re: Billy Liar
There is no such thing as a Mount Molehill. I’ve repeatedly asked for just one single example of Mount Molehill and all I ever hear is silence.

@Steven Mosher
I’ve always been a fan of Richard Muller, he strikes me as being open and amenable to critical viewpoints. To me, this is more important than necessarily agreeing with him on all detail points.
If you have his ear, the lesson I would take from this paper would be to check his methods against the presence of long-term persistence. You could probably word it better than me but the justification would be as follows:
1. There is dispute among climate scientists about the importance of long term persistence in temperature time series, but most climate scientists who have tested for it have tended to agree that it is present (or at least, failed to reject its presence). E.g. Koutsoyiannis, Cohn, Lins, Montanari, von Storch, Rybski, Halley etc. etc.
2. This paper shows that statistical methods which intuitively work reasonably well on short-term persistent time series can fall apart when faced with data containing long-term persistence.
3. As such, it would be prudent to test any methods applied to temperature series to check to see if the methods are effective in the presence of long-term persistence.
This may mean that there are useful lessons that can be learned from this study, even if the results are not directly applicable.

Hmm, of course, when I say “paper” I mean “presentation”… we should aim to be accurate in these details.

Steven Mosher
I know you’ve had a lot of quick fire posts here to answer but if your could spare a second:
As a member of the BEST team(?), one of the many things that you often hear from Prof. Muller is the use of Kriging to grid their data (correct?). I have asked this of others but never got an answer. Why have the Kriging variance maps never been released, surely these are of huge importants – then again may be not. For example, if for each year 50+% of the gridded points (or blocks?) have kriging variances equal to the error of the set (beyond the range of spatial correlation) one would have to wonder if there is much point in continuing to put together a time series where the spread in values is less than/similar to the dominant kriging variances for each year in that time series.

“Removal of outliers”…with the ratio of urban to rural stations, and the commonly observed temperature differential leads one to assume that readings from rural stations are more likely to fall into the outlier category. Perhaps giving an even greater bias.

SNHT — standard normal homogeneity test
PHA — progressive hedging algorithm or pairwise homogenization algorithm
Assuming one of these acronyms is what Mosher is tossing about and saying that GHCN uses to adjust the temperatures. Which one?

Steven Mosher says:
July 17, 2012 at 11:12 am
Situation: When have station named Mount Molehill. It is located at 3000 meters above sea level. It records nice cool temperatures from 1900 to 1980. Then in 1981 they decide to relocate the station to the base of Mount Molehill 5 km away. Mount Molehill suddenly because much warmer.
But won’t they rename the station? Nope! they may very well keep the station name the same.
But won’t the latitude and longitude change? Nope. it depends entirely on the agency recording the position, until recently many only reported to a 1/10 of a degree ( 10km) So, what you get, IF YOU ARE LUCKY, is a piece of metadata that says in 1981 the altitude of the station changed.
Now, my friends, how do you handle such a record. a station at 3000 meters is moved to 0 meters and suddenly gets warmer? That’s some raw data folks. Thats some un adjusted data.
anybody want to argue that it should be used that way??
—————————————————
The location is the station. By definition, you can not move a location.
You have closed a station and ended a series. You have opened another station and started another series.
Homogenization does not apply.

“Steven Mosher
You just calculated the transient climate response. ( TCR) at 1.6.
the ECR ( equillibrium Climate response) is anywhere from 1.5 to 2x higher.
so if you calculate a TCR ( what you did) then you better multiply by 2…
Giving you 3.2 for a climate sensitivity. (ECR)”
Because the Earth does not rotate, nor does it orbit the sun.

Nick Stokes says:

I can’t see the justification for using a small subset of stations – it’s easy enough to do them all.

And only six of the stations in the diagram are north of 60 degrees latitude, where most of the warming is happening.

Alvin W says:
July 17, 2012 at 2:29 pm
I’ve been disappointed that the website that used to let you see how the number of stations dropped off dramatically at the time the temperature went up is no longer accessible.

The URL that you provided will work if you replace the hyphen with an underscore, like so:
climate.geog.udel.edu/~climate/html_pages/Ghcn2_images/air_loc.mpg
That’s also how it’s recorded in 2005 at archive.org

One point that seems to be missed here is that in their presentation the authors are not claiming that the true temperature rise is only 0.4C. They seem to be saying that’s only what the raw, unadjusted data shows. They are saying that the true temp rise is somewhere between that, at the low end, and the 0.7-0.8C rise being claimed after adjustments. The implication is that these adjustments have to be scrupulously reviewed and justified, and after that review a new record including all appropriate adjustments should be generated. That’s of course what BEST was supposed to do. Perhaps these guys can assemble a team to accomplish that task. Hope they can find funding for the project.

Nick: “I got 0.66 ° per century.”
F or C?
I’ve been looking at the Washington State raw daily data. And drop everything with a Quality flag = blank.
What filtering did you do?
I get 0.0032F / decade from 1895 to 2011. NOAA gets .05F / decade for the same period.
The interesting thing is the monthly mean’s are all over the place.
January 0.294
February 0.284
March 0.065
April -0.236
May -0.135
June -0.156
July -0.115
August -0.013
September 0.206
October -0.09
November -0.036
December -0.029

Re: Ben U 4:28 PM
Thank you so much.

Every time we revisit this subject I just imagine the dedicated person in the 1930s trudging out to the station every day and recording the temperature, squinting to get that last tenth of a degree and wondering what they would think if they knew 80 years later some armchair climatologist was going to “adjust” their reading up three degrees.

John Finn says:
July 17, 2012 at 3:36 pm
I note you’ve chosen to use the Hadley land surface record (1912-1942) to illustrate your point, but surely this data is contaminated….CONCLUSION: There was no warming between 1900 and 1980, i.e. the trend was flat.
Actually Hadcrut3 is both land and water. Of course I could not use the satellite data since it does not go way back. However the trend according to Hadcrut3 is 0.00528699 per year between 1900 and 1980. See:
http://www.woodfortrees.org/plot/hadcrut3gl/from:1900/to:1980/plot/hadcrut3gl/from:1900/to:1980/trend
So we will have to come to some other conclusion.

Trevor says:
July 17, 2012 at 4:46 pm
I am amazed to see the “No global warming” crowd “warming” to the idea of global warming,….
____________________________________
Of course there is “Global Warming” we are in a interglacial, if there wasn’t “Global Warming” NYC would be under a mile of Ice.
Here is a graph of Temperature and CO2 over the Past 400 Thousand years Note the earth warms and cools and the present increase in CO2 (on right) hasn’t caused “CAGW” Actually the CO2 sky rockets while the temperatures are MORE stable than in the other four interglacials.
Also see:
Lesson from the past: present insolation minimum holds potential for glacial inception Ulrich C. Müller, Jörg Pross, Institute of Geosciences, University of Frankfurt
In Defense of Milankovitch by Gerard Roe, Department of Earth and Space Sciences, University of Washington, Seattle, WA, USA
Article on above: http://motls.blogspot.com/2010/07/in-defense-of-milankovitch-by-gerard.html
http://wattsupwiththat.com/2012/03/16/the-end-holocene-or-how-to-make-out-like-a-madoff-climate-change-insurer/
People here at WUWT are just not hung-up on CO2 as the “control knob” for the climate. I at least consider CO2 a life giving gas that was dwindling to a dangerously low amount in the atmosphere. The recent evolution of C4 and CAM photosynthesis and the current GREENING of the biosphere show just how bad the situation was getting.
Or there is this paper in the proceeding of The National Academy of Sciences: Carbon starvation in glacial trees recovered from the La Brea tar pits, southern California by Joy K. Ward * , † , ‡, John M. Harris §, Thure E. Cerling † , ¶, Alex Wiedenhoeft ∥, Michael J. Lott †, Maria-Denise Dearing †, Joan B. Coltrain **, and James R. Ehleringer †…..

Bill Illis says:
July 17, 2012 at 5:04 pm
The lower troposphere is supposed to be warming at a faster rate than the surface, particularly in the Tropics where it is supposed be 27.3% higher according to the climate models, but also extending to mid-latitudes like the US.

Roy Spencer says:
if the satellite warming trends since 1979 are correct, then surface warming during the same time should be significantly less, because moist convection amplifies the warming with height.

Chris Wright says:
July 17, 2012 at 4:31 am
“…However, I don’t think it is the result of any organized conspiracy…”
Never attribute to stupidity that which is best explained by a collection of ###****s with an agenda.
Yeah, it’s not the way it actually goes, but with flip side of that you are giving the agenda pushers and easy win while you sit around waiting for better indications.

It would seem to me that if the warming has been that minute then the earth as a system is incredibly sensitive, given the rapid changes in in the Arctic and the quite dramatic fall in soil moisture levels globally (http://climexp.knmi.nl/ps2pdf.cgi?file=data/ipdsi_0-360E_-90-90N_n.eps.gz). If the earth is that sensitive, then we are in just as much trouble as if the temperature increase had been large …

Steven Mosher says:
July 17, 2012 at 12:12 pm
Victor Venema says:
July 17, 2012 at 11:50 am (Edit)
Steven Mosher says:
“Then of course it would make sense to check the report and see how PHA did? Because they are looking at GHCN v2 here, ”
The pairwise homogenization algorithm used by NOAA to homogenize USHCN version 2, is called “USHCN main” in the article. It performed well. It has a very low False Alarm Rate (FAR). As there is always a trade of between FAR and detection power, the algorithm could probably have been more accurate overall. And the pairwise algorithm has a fixed correction for every month of the year. Inhomogeneities can, however, also have an annual cycle. For example, in case of a radiation error, the jump will be larger in summer as in winter. With monthly corrections USHCN would have performed better, especially as the size of the annual cycle of the inhomogeneities in the artificial data used in this study was found to be a little too large.
################
Thanks Victor I’m pretty well aware of how PHA did, but thanks for explaining to others. Most wont take time to read the article or consider the results. Those who do take the time to skim the article for a word ( like PHA ) will not find it. But of course if they were current on the literature they would know that PHA is USHCN main. (hehe. told me everything I needed to know)
Interesting Steve……at 12:12 pm today you chastize people for not being abreast of the latest literature in not knowing what pha main is. An hour earlier at Steve M site, you yourself don’t seem quite to sure what it is. I take “hmmmm” and ” pretty sure ” to be, well, not so sure
“Steven Mosher
Posted Jul 17, 2012 at 10:53 PM | Permalink | Reply
Hmm. yes he describes the process of correcting GHCN v3 as SNHT.
hmm. I pretty sure that the PHA algorithm is not SNHT.”
Apparently you yourself were not up to date on this one, until of course you looked it up and came back to look like a genius. :0)

Now with this revelation and the earlier post about the IPCC, anyone with thoughts on a climategate 3 release. Keep punching while the consensus is dazed.

“mikelorrey says:
July 17, 2012 at 6:27 pm
Soooooo lemme get this right: 25-50% of climate change is due to solar variation, 25-50% is due to ENSO/AMO/PDO/NAO variations, and now 50% is due to homogenization effects. That means 100-150% of warming is now accounted for”
Lets get it right: If 50% is due to homogenization, climate change is 0.4°C and not 0.8°C. Of these 0.4°C 25% would be due to solar variation and 25% is due to ENSO etc. …
That leaves up to 0.2°C for CO2.

Steven Mosher says:
July 17, 2012 at 11:12 am
“Now, my friends, how do you handle such a record. a station at 3000 meters is moved to 0 meters and suddenly gets warmer? That’s some raw data folks. Thats some un adjusted data.
anybody want to argue that it should be used that way??”
What you do is presume that with thousands of stations for every one that moved to a lower altitude another somewhere moved to a higher altitude and they cancel out.
These instruments were never designed for this task in any case. They cover only a tiny fraction of the earth’s surface and are not accurate to tenths of a degree and until recently only recorded two instantaneous temperatures per day. You can’t make a silk purse out of a sow’s ear.
Follow the satellite data. It’s only 33 years but it’s the only network capable of establishing a global average temperature. As of now it shows 0.14C/decade warming and falling as the warming was faster in the earlier part of the record.
http://woodfortrees.org/plot/rss/every/mean:12/offset:0.13/plot/rss/every/trend/offset:0.13/plot/uah/every/mean:12/offset:0.23/plot/uah/every/trend/offset:0.23
When we overlay the satellite record on the AMDO we see a reasonable explanation for why it was rising faster in the early part of the record – the satellite measurements happened to begin coincident with the warming side of a 60-year cycle.
http://woodfortrees.org/plot/rss/every/mean:12/offset:0.13/plot/rss/every/trend/offset:0.13/plot/uah/every/mean:12/offset:0.23/plot/uah/every/trend/offset:0.23/plot/esrl-amo/every
Another 5-10 years of satellite temperature following the AMDO on the downside of the cycle further reducing the current 0.14C/decade should settle the matter one way or the other. It’s not looking for the warmists at this point. IPCC AR1 in 1990 predicted 0.30C/decade warming if CO2 emission was not curtailed. It wasn’t curtailed and less than half the predicted warming actually occured. That’s game over as far as IPCC consensus “skill”. The only thing left to determine is exactly how badly wrong they were and why. The post mortem will be interesting.

The problem with shelter that are open to the bottom is actually also “thermal pollution”. The problem with these shelters is that on days with strong insolation and little wind, the soil heats up and the thermal radiation from the ground heats the thermometer. This is very similar to the case of the UHI where the surface heats the air and then the thermometer.
See figure below, where two Stevenson screens are compared with one Montsouri screen (right), which is open to the bottom and to the North. Any “skeptic” can build such old screen and validate that indeed these measurements were biased and thus need to be corrected to obtain trends in the true climate.
cd_uk says: “If you’re correcting for the UHI then one would expect that most would be lowered not raised.”
Exactly and that the temperature trend is higher after homogenization means that the UHI is less important than all the other inhomogeneities.
cd_uk says: “As for your link to your page on homogenisation. Thanks for that. It only refers to optimisation, many systems of linear combinations (e.g. some type of weighted mean) are derived via optimsation where the process is to minimise the error between the estimated and true value. This would be a type of averaging. But can’t say one way or another given your page. But thanks anyway.”
There are two types of homogenization algorithm, ones that work with pairs of stations, such as USHCN and ones that use a reference time series computed from multiple surrounding stations. This reference is indeed a weighted average of the surrounding stations. (Some people use kriging weights, which is optimal if the time series do not contain inhomogeneities, it still has to be studied whether it is optimal for homogenization.) However, this reference is not used to replace the station data, but a difference time series is computed by subtracting this reference from the candidate station. In this way the regional climate signal is removed and a jump can be detected much more reliably. The jump size found in this difference series is added to the candidate station for homogenization. The homogenized data is thus the original data plus a homogenization adjustment, it is not the averaged signal of the neighbors, there is no smearing the error as Anthony Watt keeps on repeating.

Take one hundred long term rural stations across America making due allowance for any UHI effect, graph them individually for trend and divide by 100. The AGW crowd are all about trends they would be pleased with the results. Or maybe not. Statistical homoginization using the algorithyms of the AGW crowd is some what like the company advertising for an accountant.
The recruiting officer only asked one question, what is two plus two, all failed the answer until one applicant said “what would you like it to be” he got the job.

I noticed each troll talked down from their lofty position in climate science. But in open scientific debate they were exposed as narrowly opinionated and not truth seekers, still hung up on CO2 as the control knob. A part from funding, their foundation continues to crumble beneath them before many.

Victor
Is the point of the article not, as one would expect for UHI effect, that most station data would are adjusted up rather than adjusted down as in the homogenised data? The other point I’d add is that most urbanisation would be gradual and therefore may not be idntified by a discrete jump. Furthermore, if your homogenisation uses neighbouring stations suffering from the same process the effect would be to push the temperatures up.
As for your qualification on temperature homogenisation thanks. I think the result will still be the same – smoothing.
If I’ve got this right:
1) You use interpolated data (for candidate station) to predict the temperature relative to neighbouring stations. This is carried out for each year of the time series.
2) This difference (for each year): diff = observed – interpolated
3) This diff is then added onto the observed to give the homogenised value and thus:
interpolated = diff + observed
whichas you can see is the same as just assigning the interpolated value to the candidate station and hence the smoothing.

izen
I don’t think that is what is going on here. The homogenisation process does appear to be a purely statistical approach that will effectively smooth real/false climatological data indiscriminately. The reason for these adjustments is to remove experimental error, but this should have a 50:50 split and therefore will not affect the final result. This is not what is happening.
As for global records you are correct. The satellite data does corroborate the instrumental record. However, some of the heaviest adjustment (and downward ones) are pre-satellite.
As for BEST they still have to release their Kriging Variance maps for each year (as far as I know). Without these we don’t know what proportion of their gridded data has uncertainties equal to the variance of the set. If the majority have “variances” on the order of the range seen in the time series they produced the chronology has no – casually speaking – statistical significance.

****
Victor Venema says:
July 17, 2012 at 10:38 am
In practice homogenization makes the temperature trend stronger. This is because temperatures in the past were too high. In the 19th century many measurement were performed at North facing walls, especially in summer the rising of setting sun would still burn on these instruments. Consequently these values were too high and homogenization makes the lower again. Similarly, the screens used in the first half of the 20th century were open to the North and to the bottom. This produced too high temperatures on days with little wind and strong sun as the soil would heat up and radiate at the thermometer.
****
Well, you’d need some pretty detailed photos and/or metadata to quantify that.
But let’s assume it for the moment. I’d think the “radiation” aspect would go both ways. Having an opening to the ground & north sky at nite would also lower the min temp on a calm nite (the ground surface cools first), no? Some rather simple experimental setups could prb’ly quantify it in a reasonable time. One would think with the billions & billions of bucks going to “climate research”, a couple setups like that wouldn’t be hard to do?

John Finn : “solar activity was higher in 1940-1980 than it was in 1900-1940”
There are two main types of solar. TSI and bright sunshine.
Bright sunshine changes may have played a role:
http://sunshinehours.wordpress.com/category/sunshine/

With this tiny amount of warming in the surface temperature over the past century, I find it remarkable that the UAH tropospheric temperatures in the past 33 years have risen 0.46 degrees, and that the Greenland and Antarctic land ice packs are melting at accelerating rates, and that the northern hemisphere snow cover is dropping dramatically, and that the sea level continues to rise, and that arctic sea ice area and volume are dropping dramatically, and that the ocean heat content down to 2000 meters is rising inexorably, and that the tundra is melting, and that ……..

beng says: “Some rather simple experimental setups could prb’ly quantify it in a reasonable time. One would think with the billions & billions of bucks going to “climate research”, a couple setups like that wouldn’t be hard to do?”
Has been done:
Böhm, R., P.D. Jones, J. Hiebl, D. Frank, M. Brunetti,· M. Maugeri. The early instrumental warm-bias: a solution for long central European temperature series 1760–2007. Climatic Change, 101, pp. 41–67, doi: 10.1007/s10584-009-9649-4, 2010.
Abstract. Instrumental temperature recording in the Greater Alpine Region (GAR) began in the year 1760. Prior to the 1850–1870 period, after which screens of different types protected the instruments, thermometers were insufficiently sheltered from direct sunlight so were normally placed on north-facing walls or windows. It is likely that temperatures recorded in the summer half of the year were biased warm and those in the winter half biased cold, with the summer effect dominating. Because the changeover to screens often occurred at similar times, often coincident with the formation of National Meteorological Services (NMSs) in the GAR, it has been difficult to determine the scale of the problem, as all neighbour sites were likely to be similarly affected. This paper uses simultaneous measurements taken for eight recent years at the old and modern site at Kremsmünster, Austria to assess the issue. The temperature differences between the two locations (screened and unscreened) have caused a change in the diurnal cycle, which depends on the time of year. Starting from this specific empirical evidence from the only still existing and active early instrumental measuring site in the region, we developed three correction models for orientations NW through N to NE. Using the orientation angle of the buildings derived from metadata in the station histories of the other early instrumental sites in the region (sites across the GAR in the range from NE to NW) different adjustments to the diurnal cycle are developed for each location. The effect on the 32 sites across the GAR varies due to different formulae being used by NMSs to calculate monthly means from the two or more observations made at each site each day. These formulae also vary with time, so considerable amounts of additional metadata have had to be collected to apply the adjustments across the whole network. Overall, the results indicate that summer (April to September) average temperatures are cooled by about 0.4°C before 1850, with winters (October to March) staying much the same. The effects on monthly temperature averages are largest in June (a cooling from 0.21° to 0.93°C, depending on location) to a slight warming (up to 0.3°C) at some sites in February. In addition to revising the temperature evolution during the past centuries, the results have important implications for the calibration of proxy climatic data in the region (such as tree ring indices and documentary data such as grape harvest dates). A difference series across the 32 sites in the GAR indicates that summers since 1760 have warmed by about 1°C less than winters.