From the told ya so department, comes this recently presented paper at the European Geosciences Union meeting.
Authors Steirou and Koutsoyiannis, after taking homogenization errors into account find global warming over the past century was only about one-half [0.42°C] of that claimed by the IPCC [0.7-0.8°C].
Here’s the part I really like: of 67% of the weather stations examined, questionable adjustments were made to raw data that resulted in:
“increased positive trends, decreased negative trends, or changed negative trends to positive,” whereas “the expected proportions would be 1/2 (50%).”
And…
“homogenation practices used until today are mainly statistical, not well justified by experiments, and are rarely supported by metadata. It can be argued that they often lead to false results: natural features of hydroclimatic times series are regarded as errors and are adjusted.”
The paper abstract and my helpful visualization on homogenization of data follows:
Investigation of methods for hydroclimatic data homogenization
Steirou, E., and D. Koutsoyiannis, Investigation of methods for hydroclimatic data homogenization, European Geosciences Union General Assembly 2012, Geophysical Research Abstracts, Vol. 14, Vienna, 956-1, European Geosciences Union, 2012.
We investigate the methods used for the adjustment of inhomogeneities of temperature time series covering the last 100 years. Based on a systematic study of scientific literature, we classify and evaluate the observed inhomogeneities in historical and modern time series, as well as their adjustment methods. It turns out that these methods are mainly statistical, not well justified by experiments and are rarely supported by metadata. In many of the cases studied the proposed corrections are not even statistically significant.
From the global database GHCN-Monthly Version 2, we examine all stations containing both raw and adjusted data that satisfy certain criteria of continuity and distribution over the globe. In the United States of America, because of the large number of available stations, stations were chosen after a suitable sampling. In total we analyzed 181 stations globally. For these stations we calculated the differences between the adjusted and non-adjusted linear 100-year trends. It was found that in the two thirds of the cases, the homogenization procedure increased the positive or decreased the negative temperature trends.
One of the most common homogenization methods, ‘SNHT for single shifts’, was applied to synthetic time series with selected statistical characteristics, occasionally with offsets. The method was satisfactory when applied to independent data normally distributed, but not in data with long-term persistence.
The above results cast some doubts in the use of homogenization procedures and tend to indicate that the global temperature increase during the last century is between 0.4°C and 0.7°C, where these two values are the estimates derived from raw and adjusted data, respectively.
Conclusions
1. Homogenization is necessary to remove errors introduced in climatic time
series.
2. Homogenization practices used until today are mainly statistical, not well
justified by experiments and are rarely supported by metadata. It can be
argued that they often lead to false results: natural features of hydroclimatic
time series are regarded errors and are adjusted.
3. While homogenization is expected to increase or decrease the existing
multiyear trends in equal proportions, the fact is that in 2/3 of the cases the
trends increased after homogenization.
4. The above results cast some doubts in the use of homogenization procedures
and tend to indicate that the global temperature increase during the
last century is smaller than 0.7-0.8°C.
5. A new approach of the homogenization procedure is needed, based on
experiments, metadata and better comprehension of the stochastic
characteristics of hydroclimatic time series.
Full text:- Presentation at EGU meeting PPT as PDF (1071 KB)
- Abstract (35 KB)
h/t to “The Hockey Schtick” and Indur Goklany
UPDATE: The uncredited source of this on the Hockey Schtick was actually Marcel Crok’s blog here: Koutsoyiannis: temperature rise probably smaller than 0.8°C
Here’s a way to visualize the homogenization process. Think of it like measuring water pollution. Here’s a simple visual table of CRN station quality ratings and what they might look like as water pollution turbidity levels, rated as 1 to 5 from best to worst turbidity:
In homogenization the data is weighted against the nearby neighbors within a radius. And so a station might start out as a “1” data wise, might end up getting polluted with the data of nearby stations and end up as a new value, say weighted at “2.5”. Even single stations can affect many other stations in the GISS and NOAA data homogenization methods carried out on US surface temperature data here and here.
In the map above, applying a homogenization smoothing, weighting stations by distance nearby the stations with question marks, what would you imagine the values (of turbidity) of them would be? And, how close would these two values be for the east coast station in question and the west coast station in question? Each would be closer to a smoothed center average value based on the neighboring stations.
UPDATE: Steve McIntyre concurs in a new post, writing:
Finally, when reference information from nearby stations was used, artifacts at neighbor stations tend to cause adjustment errors: the “bad neighbor” problem. In this case, after adjustment, climate signals became more similar at nearby stations even when the average bias over the whole network was not reduced.
Discover more from Watts Up With That?
Subscribe to get the latest posts sent to your email.
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)
REPLY: Thanks for clarifying what you said, always good to cite – Anthony
Yes, Steve’s position is pretty clear. he thinks the best approach is to start with the best stations and work outward, rather than, using all the data. and trying to correct or throw out the worst. He is, as he says, not to interested in looking at these issues, no matter how many times I ask.
here.
When steve first posted on Menne Zeke suggested that he look at this
https://ams.confex.com/ams/19Applied/flvgateway.cgi/id/18288?recordingid=18288
Not sure if he did, but its easy to watch
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/
Steven Mosher says:
July 17, 2012 at 11:12 am
Why do it the hard way? If the Mount Molehill weather station moves to the bottom of the hill and doesn’t change its name, instead of trying to homogenize the hell out of it, the sensible person would simply end the Mount Molehill record and start the Mount Molehill – New Place record.
What’s not to like?
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.
Victor Venema says:
July 17, 2012 at 11:50 am
Victor, thank you for the clarifications. I’m not sure what you mean when you say PHA performed “well”, since 7 out of the 12 other methods outperformed it per Figures 2 and 3, which would mean to me that it performed about average …
Also, it’s not clear to me that your study looked at the same thing as the Koutsoiannis study. That is to say, if there is a mix of good data and bad data, did your study consider whether the bad data “pollutes” the good data? It seems to me that your study assumed that all data had inhomogeneities, and then looked to see who corrected them better, but I could be wrong.
Please be clear that I have no theoretical problem with correcting inhomogeneities. Nor am I of the opinion that there are huge biases in the recent ground station records, because they agree within error with the satellite records, although that says nothing about the earlier part of the ground station records … and the existing difference between e.g. GISS and UAH MSU is still about 0.2°C per century, about a third of the century-long trend, so while it is small it is not zero.
As always, however, the devil is in the details.
w.
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.
The paper seems to suggest that the homogenization contains homgenization.
@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.
Victor Venema said; “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[m] lower again.”
Suppose a psychologists had taken an IQ test of a large group of children and found that those without a certain education scored on average 5 points less. By ‘homogenisation’ he would give those children 5 bonus points. In practice these differences are handled in statistical models and not by altering the data. For an outsider like me data handling in climate science looks like an incredible mess. Moreover outsiders should also believe that this mess produces a perfect unbiased result. Perhaps one day I will start to believe in miracles.
Hmm, of course, when I say “paper” I mean “presentation”… we should aim to be accurate in these details.
Steven Mosher says:
July 17, 2012 at 12:12 pm
Hehe? Yeah, that’s hilarious. If you knew it was referred to as USHCN main in the paper, then referring to it as PHA can only be described as sneaky, malicious, and underhanded. Some of the readership here is not totally current on every obscure branch of climate science, no surprise, it is a very broad field and no one can stay totally current on everything, myself included … so your response to laugh at us?
So your own knowledge is so encyclopedic and it covers every part of climate science so well that you can afford to heap scorn on others who don’t know the intricacies of some particular topic? Really?
I and others are here to learn, Steven, so abusing and laughing at people who may not know some minor fact that you know is not helpful. It just makes you look arrogant and uncaring. I doubt greatly that you are either, but you are sure putting up a good imitation …
w.
Can anyone confirm if the BEST study has been peer approved and published in a science journal yet
Tonyb
Victor
The point about the UHI effect is that this is a “process” that introduces bias – not experimental error that homogenisation is meant to correct for(?). The adjustments being criticised are for experimental error they are not for a progressive “thermal-pollutant” that moves in one direction. If you’re correcting for the UHI then one would expect that most would be lowered not raised.
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.
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.
Sorry Steven by:
(beyond the range of spatial correlation)
By this I mean:
That most of the gridded values lie at distances from control points that are greater than the range of the variogram models.
Sorry not very clear.
“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.
This is great to see.
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.
climate.geog.udel.edu/~climate/html-pages/Ghcn2_images/air_loc.mpg
I wonder why they do not want us to see it anymore.
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?
Isn’t this similar to what E.M. Smith was saying in his post a couple of weeks back?
http://wattsupwiththat.com/2012/06/22/comparing-ghcn-v1-and-v3/
Willis Eschenbach
@ur momisugly
Steven Mosher
I and others are here to learn, Steven, so abusing and laughing at people who may not know some minor fact that you know is not helpful.
Here is a major fact that every climate scientist should get to know.
300 year temperature record of zero trend.
300 year of regular oscillations
No CO2 effect
No UHI
Just simply a natural oscillation.
http://www.vukcevic.talktalk.net/GSO-June.htm
I can’t see the justification for using a small subset of stations – it’s easy enough to do them all. I did a comparison here of the period from 1901-2005. I chose those years because the trends were cited in the AR4 report. Using the unadjusted GHCN v2 data (no homogenization) I got 0.66 ° per century. The corresponding trends cited in AR4 for CRU, NCDC and GISS, all using homogenization, were, respectively, 0.71, 0.64 and 0.60 °C per century. No big difference there.
On GHCN V2, the effect of homogenization on trends, for all stations, was discussed a few years ago. Here is the histogram of adjusted 70-year trends, and here is the unadjusted.
Here is the histogram of trend differences created by adjustment over the 70 years. It’s not far from symmetrical. The mean is 0.175 °C/century.
The fact that homogenization does not have zero effect on the mean is not a “homogenization error”. It’s the result. No one guaranteed that non-climate effects have to exactly balance out.
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.
sunshinehours1 says:
July 17, 2012 at 10:34 am
John Finn “Note Satellite readings are not contaminated by UHI.”
Are we sure? NASA has no trouble finding huge amounts of UHI from their satellites. </i.
Read your link, The satellites in this case are actually measuring the land surface temps. UAH satellite readings are measurements of temperatures in the troposphere. The latter are unaffected by UHI.