Problems With The Scalpel Method

Link to Dedekind’s post does not work.
Do you assume that the drift is always positive: a tree creating a shadow, new asphalt, new buildings, fading paint, …
Is it possible to find signal that is smaller than the measurement errors. This is the big question of the temperature records.

no conspiracy is required. programmers never look for errors when the data gives the expected result. that is how testing is done most of the time. you only look for problems when you don’t get the answer you expect. so, if you expect warming, you only look for bugs when the results don’t show warming. as a result, bugs that cause warming are not likely to be found – at least not by the folks writing the code.

“…a temperature record subject to periodic or episodic maintenance or change…”. Have any tests been done to determine the magnitude of changes such as repainting compared to daily dirt and dust build-up? I have a white car which progessively becomes quite grey until a good rain storm. I would imagine such dirt build up could have a significant effect on a Stevenson screen, between rain storms?

It is clear beyond doubt (see for example the recent articles on Steve Goddard’s claim regarding missing data and infilling) and the poor siting issues that the surface station survey highlighted, that the land based thermometer record is not fit for purpose. Indeed, it never could be, since it has always been strained well beyond its original and design purpose. The margins of error far exceed the very small signal that we are seeking to wean out of it.
If Climate Scientists were ‘honest’ they would, long ago, have given up on the land based thermometer record and accepted that the margins of error are so large that it is useless for the purposes to which they are trying to put it. An honest assessment of that record leads one to conclude that we do not know whether it is today warmer than it was in the 1880s or in the 1930s, but as far as the US is concerned, it was probably warmer in the 1930s than it is today..
The only reliable instrument temperature record is the satellite record, and that also has a few issues, and most notably the data length is presently way too short to be able to have confidence in what it reveals.
That said, there is no first order correlation between the atmosheric level of CO2 and temperature. The proper interpretation of the satellite record is that there is no linear temperature trend, and merely a one off step change in temperature in and around the Super El Nino of 1998.
Since no one suggests that the Super El Nino was caused by the then present level of CO2 in the atmosphere, and since there is no known or understood mechanism whereby CO2 could cause such an El Nino, the take home conclusion from the satellite data record is that climate sensitivity to CO2 is so small (at current levels, ie., circa 360ppm and above) that it cannot be measured using our best and most advanced and sophisticated measuring devices. The signal, if any, to CO2 cannot be seperated from the noise of natural variability.
I have always observed that talking about climate sensitivity is futile, at any rate until such time as absolutely everything is known and understood about natural variation, what are its constituent forcings and what are the lower and upper bounds of each and every constituent forcing that goes to make up natural variation.
Since the only reliable observational evidence suggests that sensitivity to CO2 is so small, it is time to completely re-evaluate some of the corner stones upon which the AGW hypothesis is built. It is at odds with the only reliable observational evidence (albeit that data set is too short to give complete confidence), and that sugggests that something fundamental is wrong with the conjecture.

Willis Eschenbach says:
June 29, 2014 at 1:03 am
//////////////////
Commonsense would suggest that the majority of recent adjustments should be to cool recent measurements.
Given the effects of UHI and urban crawl, switch to airports etc (with more aircraft traffic and more powerful jet engines compared to props etc) these past 40 or so years, would be that adjustments to the ‘record’ for 2014 through to say 1970 should be such that these measurements are lowered (since the raw data would be artificially too high due to warming pollution by UHI etc).
Yet despite this being the commonsense position, it appears that the reverse is happening. Why???

Richard Verney says:
“If we cannot detect the effects of UHI in the land based thermometer record, given that UHI is a huge signal and we know that over the years urbanization has crept and that station drops outs have emphasized urban stations over truly rural stations, there is no prospect of seeing the far weaker signal of CO2.”
Perhaps the answer is to follow the advice of John Daly and use only those ´rural´ stations with a long record in areas where the response to CO2 ( if discernable) will be at its highest and competition with water vapour at its lowest – in the Arctic and Antarctic regions where the low temperatures give the highest IR output in the region where CO2 has its highest absorption bands. Given that it has been stated that we only need 50 or so records to give an accurate GTA he offers 60 plus sites that meet the criteria of being minimally affected by UHI effects, most of which do not show any long term warming trend with many showing cooling.
http://www.john-daly.com/ges/surftmp/surftemp.htm

Temporal UHI (Urban Heat Island) effect is another continuous drift, this is why BEST was unable to identify it. It also goes into the positive direction, because, although population explosion was already over 2 decades ago (global population below age 15 is not increasing any more), population density, along with economic activity keeps increasing in most places, due to increasing life expectancy and ongoing economic development. Which is a good thing, after all.
It is far from being negligible, UHI bias alone can be as much as half the trend or more.

I cannot remember the number of times I have written on blogs that ” IT IS TOTALLY SCIENTIFICALLY UNACCEPTABLE TO ALTER PAST DATA” unless you have good, scientific and mathematical analysis to allow you to do so without any doubt or favour.
STOP IT !!

Addendum to my 2:45 am post:
The total number of sites was 35,000
but only 10,000 of them had less than 10% missing data
and only 3000 of them had <10% missing data and greater than 38 years long.
To be clear, these numbers came from a superficial look at the raw data files Zeke and Mosher provide links to in a index page. These values are prior to the use of the BEST scalpel. BEST has ready and fast access to post-scalpel segment length. Constructing our suggested census plots by sampled year should be easy and no burden compared to other processing they do.
Another interesting chart that Richard H and I briefly explored is a census map of 1×1 degree grid based upon post-segment length. For instance color code the 1×1 deg cells by the number of segments that exist that are longer that 20 years and cover the year 2005. I hypothesize it would be a pretty blank map. 2×2 deg cells? (that is about 100×125 mile cells).

“the most accurate measurements are those immediately following the change. Following that, there is a gradual drift in the temperature until the following maintenance.”
The drift is always positive, that’s why it doesn’t work. If the drift were random it would work.

The breakpoints should be equally distributed between positive and negative adjustments of roughly the same magnitude.
And the positive and negative breakpoints should be equally distributed through time. In 1910, there should be 200 negative breakpoints and 200 positive adjustments of roughly the same magnitude. And this should be roughly consistent from the start of the record to the end of the record.
This is how bias can be detected. Simple histograms. And then we can determine whether the excess of negative adjustments in 1944 is valid for example. What is the main reason for this.
There should not be a trend through time in the breakpoints unless it can be proven that they should vary through time.
BEST hasn’t shown this that I am aware of. The description of the simple breakpoint/scapel method suggests that the breakpoints should be random through time. Are they? Why haven’t they shown this and explained it?
We know the NCDC has shown their adjustment have a systematic trend through time with a maximum negative adjustment of -0.35C to -0.55C in the 1930s to 1940s. Why? Does the TOBs adjustment vary throughout time? Why? Shouldn’t the other changes vary randomly through time? Why not?
Should people be allowed to change the data without fully explaining it.

Oh I see, so they check each station to see if it was a physical change…or weather
…and if it’s weather, they don’t splice it

SandyInLimousin says:
June 29, 2014 at 3:47 am
==============
i routinely get systems designers telling me that the odds of a particular event are billions to one against, so we can safely ignore it in the design. then I show them with a simple data mining exercise that we routinely have many of these billion to one events. most people are very poor at estimating events when it is in their favor to estimate poorly.

Correcting slice bias is not difficult. Add up all the offsets from slicing. some will be positive and some negative. whatever the residual, add the inverse into the final result to remove any trend created by slicing. but without knowing the offsets due to slicing, there is no way to know if it introduced a trend or not.

Willis, maybe Zeke and Mosher were no longer interested in the discussion, but I did answer your question.

That is why you should not only correct jumps known in metadata, but also perform statistical homogenization to remove the unknown jumps and gradual inhomogeneities. Other fields of science often use absolute homogenization methods (finance and biology), with which you can only remove jumps. In climatology relative homogenization methods are used that also remove trends if the local trend in one station does not fit to the trends in the region. Evan Jones may be able to tell you more and is seen here as a more reliable source and not moderated.
P.S. To all the people that are shocked that the raw data is changed before computing a trend: that is called data processing. Not much science and engineering is done without.

Relative homogenization methods, compute the difference between the station you are interested in and its neighbours. If the mean of this difference is not constant, there is something happening at one station, that does not happen at the other. Thus such changes are removed as well as possible in homogenization to be able to compute trends with more reliability (and the raw data is not changed and can also be downloaded from NOAA). In relative homogenization you compare the mean value of the difference before and after a potential date of change, if this difference is statistically significant a break is detected or in case of BEST the scalpel is set and the series are split. It does not matter whether the means are different due to a gradual change or due to a jump. What matters is that the difference in the means before and after is large enough.
This part of the BEST method is no problem. With Zeke, BEST, NOAA and some European groups we are working on a validation study aimed especially at gradual inhomogeneities. This complements the existing studies where a combination of break inhomogeneities and gradual ones were used for validation.

BEST by year, what is the net total of the difference between the endpoints for all the slices?

pps: one would also need to consider the gridding when apportioning the residuals from slicing. even if they added up to zero in absolute terms, this could still introduce bias when gridded.

Willis I like the scalpel because it creates a break where there is a shift in data bias. That is a good way to start.
Suppose the change was trimming a tree, painting the enclosure or cutting down a nearby hedge. The influence of this class of change is gradual – we can assume it grew to be a problem gradually so the effect is approximately linear.
The bias (up or down) is the difference between the last and first data of the two sets but there would have to be some brave corrections because the week following the change is not the same as the one before it.
Suppose the new data were consistently 1 degree cooler than the old. Doesn’t work. We have to take a whole year. But a year can be a degree colder than the last. Big problem.
If there really was a 1 degree bias, we have to assume the early part of a data set is ‘right’. The step change has to be applied to the old data to ‘fix’ it assuming the change is linear.
Step change = s
Number of observations = n
N is the position in the series
D1 = data value 1
D1c = Corrected value 1
Linear Correction:
D1c = D1+(N1-1)*s/n
D2c = D2+(N2-1)*s/n
D3c = D3+(N3-1)*s/n
Etc
That works for steps up or down.
The data set length can be anything. All it needs is the value and sign of the step.
Other types of episodic change like paving nearby sidewalk need offsets, not corrections for drift because that is a different class of change.

Quoting Willis “quotes” from the article:
“Since the Berkeley Earth “scalpel” method would slice these into separate records at the time of the discontinuities caused by the maintenance, it throws away the trend correction information obtained at the time when the episodic maintenance removes the instrumental drift from the record.”
“As a result, the scalpel method “bakes in” the gradual drift that occurs in between the corrections.”
“So I’d like to know if and how the “scalpel” method avoids this problem … because I sure can’t think of a way to avoid it.
So I’m here to ask it again …”
———–
Willis, I will offer my solution for your stated problem but you will have to determine if it is applicable and if it can be implemented or not.
First of all, a discontinuity “flag” character would have to be chosen/selected that would be appended to the “first” temperature reading that was recorded after said maintenance was performed at/on the Surface Station. Said “flag” would thus denote a “maintenance break-point” in the daily temperature data. …. And for “talking” purposes I will choose the alpha character “m” with both the capital “M” and small “m” having significance.
Next, a new “maintenance” program would have to be written that would “scan” the temperature data file for each Surface Station looking for any capital “M” discontinuity “flags” and if found, it would calculate the “trend” variance between said “M” flagged temperature value and the one previous to it. And via that actual “trend” value said maintenance program would algebraically “add” it to said previous temperature …. and then via an algorithm sequential “decrease” in/of said “trend” value would add said newly calculated “trend” values to all sequentially previous temperature data until it detects a small “m” discontinuity “flag” or a “default” date. The program would then change the originally noted discontinuity “flag” from a capital “M” to a small ”m” thus signifying that “trend” corrections had been applied to all temperature data inclusive between the small “m” discontinuity “flags”.
Of course, the above requires use of the “raw” and/or “trend” corrected data each time said new “maintenance” program is executed.
If that was not what you were “asking for” ….. then my bad, …. I assumed wrong.

@Bill Illis at 5:23 am
This is how bias can be detected. Simple histograms. And then we can determine whether the excess of negative adjustments in 1944 is valid for example
In addition to simple histograms, I think we need to see scatter plots of
Y: BreakPoint Offsets vs. X: Length of segment prior to the breakpoint.
For different 5-year semi-decades.
I can envision it possible for the simple histogram to show no bias in breakpoints, but only because the offsets after short segments counteract offsets of opposite sign after longer segments. A scatter plot of Offset vs prior segment length could show an interesting trend and how it changes for different periods of the total record.
Better yet, why don’t we just have access to a table of:
StationID, BreakPoint date, Segment Length Prior to break, Trend value before Break, Trend value after break.

Willis
You say
So we should use the Fahrenheit figures, rather than daring to ALTER PAST DATA by converting them to the correct Celsius figures
I don’t think anyone would suggest that, especially as most of the world uses Celsius and the USA is an exception, even the UK uses Celsius with Fahrenheit as a bracketed value for older readers (known colloquially as Old Money). For global values one or other has to be converted. Adding a fudge factor to the changed value would be unacceptable as would not noting why the change was made. Changing historical datasets in the way Paul Homewood and at one time only Steven Goddard describe is not acceptable, that is adding a fudge factor to data someone else is going to use and will confirm your results and not your theory.
I’m quite happy for the original unedited data to be all that a government agency publishes, corrections and fudge factors can be added by any researcher. The Data keepers should be just that, not the arbiters of what the data recorder meant when he wrote the figures on the piece of paper, or the automated station meant when it sent the electrons down the wire.

the really interesting thing about the slice method is that when it was proposed it seemed like a good idea. it is only now, much later in the day, that folks are realizing that it is sensitive to a certain class of errors.
it seems likely that pair-wise correction was the same. the researchers were trying to correct a specific problem, and the method worked for the cases studied. it was only much later, when the effects started to diverge from reality, that there was any indication there were problems.
like a computer system that randomly changes 0`s to 1`s and 1`s to 0`s. If it happens quickly enough you can find the error. but if the problem works slowly enough, over time your system will die and there is nothing you can do to prevent it. it is almost impossible to detect slow moving errors.

Willis: “Your method, using “statistical homogenization to remove the unknown jumps and gradual inhomogeneities”, will not fix the bogus trend created out of thin air by the scalpel method.”
I do not have a method, but only validated the methods of others up to now. The normal way of homogenization is not the scalpel method. In the normal way, the neighbours are also used to compute the corrections. This makes the long-term trend of the station with the gradual inhomogeneity similar to the one of the neighbours. I do not expect standard methods to have more problems with gradual inhomogeneities as with jump inhomogeneities.
Willis: “My question was, it is even possible to fix that spurious trend created by the scalpel method, and if so, how are the Berkeley Earth folks doing it?”
I understand their article right, BEST reduces the weight of data with gradual inhomogeneities. I would personally prefer to remove it, but they prefer to be able to say that they used all the data and did not remove anything. If the weight is small enough, that would be similar to removing the data. That is the part of the algorithm, I would study, not the scalpel mentioned in the title of this blog post.
Hello pruf Evan Jones, the quality of station placement is another problem as the one mentioned in this post. I do not want to redo our previous discussion, which would be off topic here. Do you have any new arguments since our last long, civil and interesting discussion?

How many knobs are there?
Show us the effect on the result of turning each knob from zero to ten.
The BEST slice/dice knob is obviously already turned to eleven by Mophead Mosher:
http://youtu.be/4xgx4k83zzc
Does the clear overzealousness of the chopping affect the trend? YES OR NO? We don’t know. So we don’t trust your black box. Where is the online version that lets us play with the settings? This algorithm matches the other series out there only too well early on but then becomes a climate model matching outlier in the last decade. Why? Where is the discussion of this in the peer reviewed literature? I’ve compared it here to HadCRUT3, the Climategate University version put out before Phil Jones joined a Saudi Arabian university that he used as his affiliation in his HadCRUT4 up-adjusted version:
http://woodfortrees.org/plot/best/mean:30/plot/hadcrut3vgl/mean:30
In a field with a bladeless hockey stick making it into top journal Nature you really do have to show your work instead of just releasing software few know how to run, since no, we no longer trust you. You would think you would jump at the chance to convince us further. Maybe start with finding that blade in the Marcott input data, or if you can’t find it, work on getting that paper retracted and all of the “scientists” involved fired if not arrested. That is how it is done in normal non-activist science such as medical research:
“A former Iowa State University scientist who admitted faking lab results used to obtain millions of dollars in grant money for AIDS research has been charged with four felony counts of making false statements, an indictment filed in federal court shows.”
I’m even amazed Berkeley didn’t sue you guys for using their name like Harvard sued a company called Veritas. Legally, it’s very easy and in fact guaranteed that the public will associate the BEST plot with Berkeley University though at least you didn’t also swipe their logo. Your results helps the Obama administration feel justified in stereotyping us skeptics as Moon landing denying members of the Flat Earth Society. So we are asking you for clarification, loudly. Are the other climate model falsifying temperature products wrong or are you wrong? Are climate models and thus climate alarm now falsified or not?

Zeke:

“I haven’t found an example of poor adjustment that creates a biased trend relative to surrounding stations, but I’d encourage folks to look for them.

OK, here’s one: Auckland in New Zealand.
BEST shows 0.99±0.25°C/century from 1910 for Auckland. The correct value (manually adjusting for known UHI and shelter) is closer to 0.5±0.3°C/century.

“Berkeley really doesn’t optimize for getting the corrected underlying data as accurate possible; rather, it focuses more on generating an accurate regional-level field.”

I can’t see how one can get a regionally correct value without first obtaining accurate underlying data, when the regional values are based on the underlying data. For example, I have no doubt that the incorrect Auckland series was used by BEST to adjust other NZ sites, thereby introducing an error.

This example that our host pulled up yesterday would be worth looking at:
Luling TX is also the one Paul Holmwood picked out for other reasons.
Apparently this is a good site with stable MMTS since 1995 yet is seems several discontinuities are picked up in relations to its regional average.
Does that indicate that there is a notable bias in the regional mean?

Willis,
Neighbor (or regional climatology) difference series don’t use correlations for anything. Rather, it uses the difference in temperature over time between the station in question and its neighbors. I realize that correlation often provides very little information about the trend, which is why its not a great indicator of potential bias.
The Savannah example shows some sawtooth patterns in the neighbor difference series, but they are homogenized in such a way that both the gradual trend and the sharp correction are removed.
.
Bob Dedekind,
Thats not a station record, thats a regional record. What specific stations in the Auckland area show sawtooth-type patterns being incorrectly adjusted to inflate the warming trend? Here is a list of Auckland-area stations: http://berkeleyearth.lbl.gov/station-list/location/36.17S-175.03E

Great to see this issue talked about more – I was first aware when GISS published their telling diagrams in 2001 –
How many times does a truth have to be told ? – UHI warming has been cemented into global temperature series by adjusting for steps outward from cities –
http://www.warwickhughes.com/blog/?p=2678

evanmjones says:And he has made me think more deeply beyond the stats to the mechanism in play.”
That is good. That seems to be the main thing that would make the paper a lot more convincing. I have been thinking about this since our conversation, but I am unable to think of a mechanism that could explain the statistical results you found. Especially something that would cause artificial trends due to micrositing in the 1990s, but not since the US climate reference network was installed in 2004. Puzzling to me.
Could you simply call me VIctor Venema? I can’t help it that I had to get a PhD to be allowed to do research. That is the way the system works.
REPLY: You don’t need a PhD to be able to do research and publish papers, I’ve done three now, working on #4, and as many people like to point out, including yourself, I don’t have a PhD and according to many, I am too stupid to be in the same ranks with you. Yet, I do research and publish anyway. If the school of science didn’t have a foreign language requirement, and I didn’t have horrible unsolvable hearing problem, and the Dean of the School of Science wasn’t a prick at the time, and the ADA had been in place, that might have been different. TV/radio where I only had to speak was my salvation, I had a one-time chance and I took it. But, I know in the eyes of many in your position that career path makes me some sort of lowbrow victim of phrenology.
The explanation to the problem you pose is based in the physics of heat sinks, but you’ll just have to wait for the paper. Though, ahead of time to help readers understand, I may post an article and/or experiment to show how the issue manifests itself.
Bear in mind I don’t wish to start a dialog with you at the moment, mainly because you called your view of my religion into question without actually knowing what my view is, and I find that shameful and just as bad as the things you accuse me of. I’m only pointing out that PhD holders are not exclusive to research. No need to reply.
Also, while I can’t prevent it, even though I hold copyright on my own words, I ask that you not turn my comment into another taunt at your blog. It would be a good gesture if in fact you believe what you write about what I should be doing. – Anthony

Bob,
You can see alternative names on the right side of the station page: http://berkeleyearth.lbl.gov/stations/157062
Stations tend to get merged if they have overlapping identical temperature measurements under different names.

Zeke:

“You can see alternative names on the right side of the station page: http://berkeleyearth.lbl.gov/stations/157062“

I’m sorry but I don’t get that. The list is:
ALBERT PARK
AUCKLAND
AUCKLAND AERODROME
AUCKLAND AIRP
AUCKLAND AIRPORT
AUCKLAND CITY
AUCKLAND, ALBERT PAR
As far as I can tell from this, there are two sites, Albert Park and Auckland Airport, but it certainly isn’t clear, because the chart has three red diamonds (Station moves) shown. How do you know when the station move happened? Do you look at metadata? Is the “station move” the same as a splice point?
The elevation is given as 27m. Albert Park is 49m, Auckland Airport is 5m or less. Perhaps it’s the average?

“Stations tend to get merged if they have overlapping identical temperature measurements under different names.”

It is extremely unlikely that Albert Park and Auckland Airport had identical temperatures, simply because there is a well-documented 0.66°C difference between the two. Unless you’re talking about correlations, or anomalies.

Bob Dedekind,
I believe that the difference series in question are calculated prior to adjustments by comparing each station to the raw station records of surrounding stations.
Also, NCDC’s method should be able to pick out similar sawtooth patterns; see the M4 model in Menne and Williams 2009: ftp://ftp.ncdc.noaa.gov/pub/data/ushcn/papers/menne-williams2009.pdf
As I mentioned earlier, this could all be tested better using synthetic data, something that is planned as part of the new International Surface Temperature Initiative: http://www.geosci-instrum-method-data-syst-discuss.net/4/235/2014/gid-4-235-2014.html

@Zeke Hausfather at 12:01 pm
Here are a few examples of sawtooth and gradual trend inhomogeneities seem to be correctly adjusted:
Like Willis, what is the evidence that ANY of the breaks is a correct adjustment? . Much less ALL of them?
Notes:
#Moves, #Other Breaks, (3 Longest Segment since 1960 incllusive,) Difference from Regional
http://berkeleyearth.lbl.gov/stations/169993
SAVANNAH/MUNICIPAL, GA.
2 moves, 8 other breaks, (18, 17, 10) year, -0.5 deg C
http://berkeleyearth.lbl.gov/stations/30748
JONESBORO 2 NE (?Arkansas?)
6 moves (all since 1974), 14 Others, (16, 11, 8) years, -2.0 deg C
http://berkeleyearth.lbl.gov/stations/156164
TOYKO
2 moves (1 in 2006), 5 Others, (40, 15, 8), +1.9 deg C
http://berkeleyearth.lbl.gov/stations/161705
LAS VEGAS MCCARRAN INTL AP (1936-current)
2 moves (1996, 2008), 7 Others, ( 34, 13, 6 ) years, +2.5 deg C over regional
http://berkeleyearth.lbl.gov/stations/33493
FOLSOM DAM, (near San Fran, CA) (1893 to 1993)
?1 move 1957, 11 Others, (18, 15, NA) years, -1.0 deg C
http://berkeleyearth.lbl.gov/stations/34034
COLFAX (near Sacramento, CA) 1891-current
7 moves (6 since 1972), 6 Other breaks, (18, 16, 5) years, -0.1 deg C
This one bears a revisit.
It is a flat regional trend difference with a few years of -1.0.
The Raw Anomaly looks dead flat. BEST says it is 0.43 Deg / Century
After break points applied it is 0.71 deg / Century, Regional is 0.79 deg / Century.

More worrying is the lack in BEST of the six stations specifically identified by Hessell (1980) as good rural New Zealand sites “not known to be significantly affected in any of these ways [sheltering/urbanisation/screen changes]”.
These sites are:
-Te Aroha
-Appleby
-Waihopai
-Lake Coleridge
-Fairlie
-Ophir
Why were these good sites excluded, when poor sites like Albert Park were included?
Should we be worried that Te Aroha’s trend is 0.23°C/century, Appleby’s is 0.52°C/century and Fairlie’s is 0.45°C/century (I haven’t calculated the others yet)?
All these are somewhat less than the average.

In my experience there is no statistical solution to this conundrum. There are only fudge factors.
Problems with combining different datasets which have actual sampling methodology differences cannot be resolved unless you go back and re-sample and re-submit for analysis. If the difference exists in the processing of the sample only, then you need to re-submit the sample, unless there is degradation of the sample over time. (Every case is different). And since weather/climate data is a case which is time -dependant, unless you have a time machine, in my humble opinion you can’t fix this problem, because you can’t re-sample nor re-submit for re-analysis.
What is important is that the data is archived and it is stated clearly what the sampling and methodology was. What you can’t do is throw out the original raw data, or combine different datasets without noting the inherent limitations. If you could, the laws of physics would have to be changed. Sorry can’t be fixed.

Just start first with a histogram of the breakpoint impacts over time. The next step is to figure out why.
We are just arguing about nebulous suppositions but noone is starting at the first point about what the data actually shows.

Willis Eschenbach says:
Do you see how crazy you sound with your absolute dicta?
Stephen did not offer an absolute dicta. He gave a well qualified conditional. You ignored the conditional, and removed it from what he said to make your non-sequitur. He said:
“I cannot remember the number of times I have written on blogs that ” IT IS TOTALLY SCIENTIFICALLY UNACCEPTABLE TO ALTER PAST DATA” unless you have good, scientific and mathematical analysis to allow you to do so without any doubt or favour.”
The part of that beginning with the word “unless” invalidates your counterexamples:

Say that we have a change in the time of observation. Suppose that for years we’ve been taking afternoon temperatures at 3 PM at all of our temperature stations, and then we start taking them at 2 PM.

A documented change in TOBS that is explicitly correctable is a “…good, scientific and mathematical analysis to allow you to do so without any doubt or favour.” Ditto the silliness you put up about F vs C unit conversion.
The circumstance to which Stephen refers is one where the “error” is not known but assumed to exist, or is known to exist but is not explicitly correctable, i.e. where there is no “…good, scientific and mathematical analysis to allow you to [make a correction] without any doubt or favour.”
But you removed that part from your quote of what Stephen had actually said:

But in that situation, if you say that we must accept the observational data exactly as it was recorded because ”IT IS TOTALLY SCIENTIFICALLY UNACCEPTABLE TO ALTER PAST DATA”, then you’ve just put your full weight behind a highly misleading (although totally accurate and unaltered) temperature dataset.

turning his reasonable conditional into your “absolute dicta” strawman.

Reading all this analysis of how to back out climate trends from past temperature data, I’m reminded of a well-known saying about candidates trying to win elections:
If you’re explaining, you’re losing.
It seems unfair that this saying could be relevant, because science is supposed to be all about dispassionate contemplation of what is and is not known, and what can and cannot be measured. But as the whole climate-science fiasco makes clear, as a group activity there is also a political element — getting your work accepted by others as valid and trustworthy — and that goes double when lots of money is involved. So if you find yourself having to “explain” over and over what you’ve done and “explain” over and over why it makes sense, it may be time to try another approach.

evanmjones says: “Bad microsite does not create an artificial trend. It merely exaggerates a real, already-existing trend (warming or cooling). But from 2004 there is little if any trend. So there will be no trend to exaggerate from bad siting after 2004.”
I would call that a description of what your data shows, but not yet an explanation of what happened locally to the measurement.

In my experience there is no statistical solution to this conundrum. There are only fudge factors
=============
the problem is similar to name and address correction in mailing lists. you have a number of similar entries that may or may not be for the same person. how do you tell which entry is the “most correct”, so you can use that while discarding the others.
Do you average all the entries for the same person? no, because that will create problems. the low quality entries will swamp out the good ones, leaving you with poor quality.
Rather, what you need to do is rate the quality of the entries, not based on how similar they are to their neighbors, but rather on how good the source was for the entries. Then you throw out the entries that are from a poor source.
This would appear to be the crux of the solution for temperature. you cannot determine how good a reading is by comparing it with its neighbors, because you don’t know the quality of the neighbors. the comparison is nonsensical. you need to rate the quality of the station based on information about the station itself, and then either use the data if it is high quality, or throw it out if it is low quality.

richardverney: Per Willis
“…As a result, the raw data may not reflect the actual temperatures….”
//////////////////////////
Wrong; the raw data is the actual temperature at the location where the raw data is measured.
What you mean is whether there are some factors at work which have meant that the actual temperature measured (ie., the raw data) should not be regarded as representaive of temperatures because it has been distorted (upwards or downwards) due to some extrinsic factor (in which i include changes in the condition of the screen, instrumentation, TOBs as well as more external factors such as changes in vegetaion, nearby building etc). .

I think richardverney puts the matter badly here, and Willis Eschenbach is closer to the truth. In measurement science it is common to distinguish between “accuracay” of measurement, and “precision of measurement”, where accuracy refers exactly to the question of how close the the true value you may consider the measured value to be. When these two aspects of measurement are addressed by statisticians, they are called “bias” (the complement of “accuracy”, the expected disparity between the expected value of the observations and the true value), and “variance” ( the complement of “precision”, namely the variation in repeated measures of the same quantity.) Mean Squared Error is then the sum of the squared bias and the population variance of the estimators.
“Temperature” of a region is proportional to the mean kinetic energy of the molecules in that region; the measured temperature is always slightly different from the true temperature by some small amount.
Notice that richardverney uses “should not be regarded as representative of” whereas Willis Eschenbach uses “may not reflect actual”. I do not perceive a meaningful difference imparted by richardverney’s “correction”, but he confueses the issue by presenting his rewording as a correction.
Remember always: “accuracy” and “precision”; and their complementary concepts in statistical analysis, “bias” and “variance”. A measuring system with high bias and very low precision will repeatedly and reliably get the wrong answer, and the mean of a large number of independent observations will even more reliably get the wrong answer.
Willis’ question can be restated (I hope, please forgive me if I misunderstand) as: “Does the scalpel process bias the estimates in such a way that the estimated trend is reliably too large?” The possibility that it might do so has to be addressed. That does not say it hasn’t been addressed. I have not (yet) read all the relevant literature on the temperature record.

So we know each beak-point is “most” correct vs the data surrounding. The question then is whether the reading increase from maintenance to maintenance is linear or asymptotic toward a flat line. I would expect the latter, but determining the specific behavior of temperature bias vs time would probably require a multi-year experiment using multiple temperature stations with varying degrees of routine maintenance.
What if then we applied a linear trend from discontinuity point to discontinuity point, and calculated the slope from start to end of the interval. We cannot do the slope using best fit as that will not be correct, it must be the slope using only the start and end points. Now we find the slope from the start point of the period and the start point of the next period. We’re pretty sure the start of the next period is correct and (most likely) the same as the current period so we adjust all points based on the difference of the slopes. This way no start points get shifted down and long term trends are kept correct. The discontinuities are also gone without going all willy nilly on the data also.

Anthony replies @:
June 29, 2014 at 4:00 pm
“REPLY: You don’t need a PhD to be able to do research and publish papers, I’ve done three now, working on #4, and as many people like to point out, including yourself, I don’t have a PhD and according to many, I am too stupid to be in the same ranks with you. Yet, I do research and publish anyway.
But, I know in the eyes of many in your position that career path makes me some sort of lowbrow victim of phrenology.”
——————–
Right you are, Anthony. A big majority of those having been awarded an MA/MS or PhD Degree, …. including their “brainwashed” underlings and admirers, …. all possess a “Rank before Frank” mentality.
They have been nurtured to “bow down” to any “Rank” that is greater than their own …. and to ignore, discredit or defame any and all “Franks” regardless of what they might want to contribute to a conversation.
The per se “purchasing” of a PhD Degree from a reputable college or university is akin to …. some one “purchasing” a BIG toolbox chuck full of all kinds of “specialized” tools from a local Sears, Lowe’s or Home Depot.
Thus, both parties have “proof of ownership” (Diploma-Degree vrs. Sales Receipt) of their big box of “tools” ……… but said “proof of ownership” is neither proof nor factual evidence that said parties are actually capable of using the “tools” contained in their “toolbox”.
And when one of the aforesaid pulls “Rank before Frank” on you, …. you should immediately know what their debilitating “deficiency” problem is.

There is a theoretical possibility that some discrete microsite events (like a parking lot paving or nearby building constructed) can be eliminated by the scalpel,….
In the case of airports, there are commonly construction projects as terminals and tarmacs expand, runways lengthened and added. If you don’t move the temperature sensor, you would be changing the microsite conditions and it arguably deserves a breakpoint. But if you move the sensor away from the construction to restore the micrositing classification back to Class 1, should you count it as a station move and institute a breakpoint?
I would argue, NO.
Breakpoints are not going to change the gradual build up in UHI and activity at the airport. Moving the station has restored and recalibrated the temperature sensor to make it less dependent on nearby sources of contamination. While you can argue that an (unnecessary) breakpoint can be inserted here without introducing bias, I argue that it is a bias against long term records, especially from well maintained sites, which appears to be a rare commodity.

@Evan Jones at 5:57 pm
(Yes, I am both a “station dropper”, and “data adjuster”, good lord have pity on me. Just don’t ask me to homogenize. Some sins transcend the venal.)
LOL.
Well, I’m a petroleum geophysicist, among other things. When it comes to dropping stations, nulling data for multiples, and adjusting timeseries (in the time domain!) for normal moveout, near surface static corrections, and converting to depth, seismic data processors have no peer. Fortunately, we have fold to rely upon —- which is a form of homogenization, to make each source-receiver pair look like others and the same Depth point and very similar to neighbors.
Guilty! Geophysicists commit data sin — discretely, and in bulk.
But we leave the field tapes alone and we document the processing steps.

if you homogenize milk and manure, the end product will still taste like shzt.
you cannot eliminate the manure by comparing one pail of milk with another. what if the neighboring pail of milk is also contaminated? the only way to eliminate the manure is to compare each pail against a known standard.
thus Anthony’s approach of eliminating poorly sited stations is correct, while the various nearest neighbor comparison methods are flat out wrong.

if your neighbors all tell you the same tale, does that make it true? no, because you don’t know the quality of their source. only after you find out if the had a high quality source can your judge if the story is accurate or not.

Nick Stokes: I’ve written a post here which tries to illustrate the fallacy of that. When you are calculating the average for a period of time, or a region of space, that data point was part of the balance of representation of the sample. If you throw it out, you are effectively replacing it with a different estimate. You can’t avoid that. And your implied estimate could be a very bad one.
It’s not good advice.
REPLY: Nick Stokes, defender of the indefensible, is arguing to preserve bad data. On one hand he (and others) argue that station dropout doesn’t matter, on the other he argues that we can’t throw out bad stations or bad data because it won’t give a good result.
If you knew for sure what data were “bad” and what data were “good”, your reply might make sense, but note Nick Stokes’ point that throwing out a “bad” data point is equivalent to imputing (the word most statisticians prefer to “estimating” missing data) a particular value to it, and that might not be the best imputation possible. In almost all cases, including the temperature data sets, all the data are “imperfect to some degree”, and the classification into “bad” vs “good” is an arbitrary simplification. With lots of imperfect but few “bad” data points, using the extant data to impute a value to the “bad” data probably is better than the particular imputation method of dropping the “bad” data.
Nick Stokes’s defense is reasonable, and the whole topic of methods of imputation is much addressed in the statistical literature. How good a particular method is in a particular case often can’t be determined with great confidence from the extant data, but dropping “identified BAD” data is almost always among the least defensible alternatives.
Another venue where this issue arises is in measuring small concentrations (of drugs, metabolites, toxic pollutants, etc) where a large number of values are positive but “below the limit of detection”. Throwing them out can be worse than using them. Two statisticians who have addressed this problem for particular cases are Diane Lambert, PhD (then of AT&T Bell Labs, more recently at Google) and Emery Brown, MD, PhD (then at Mass Gen or Macleans; now at Harvard Medical School.) If anyone is interested, I can get the full references, but the general topic is “missing values and data imputation”.
This is a terrific thread. I would like to thank Nick Stokes for hanging around and presenting a spirited defense of a defensible approach to missing weather values, and Willis for initiating the thread.

Willis Eschenbach: This is that throwing out the bad data increases the uncertainty of the result by reducing N, while replacing the bad data with the average of the remaining data decreases the uncertainty of the result. As a result, Nick’s claim that the two are equivalent is simply not true.
That is a pertinent point. When data are imputed, the number of imputed values has to be subtracted from N in order to avoid the misleading appearance of greater precision. Nick Stokes will speak for himself, but I bet that he knows that.
Now back to “when is imputing better than simply dropping?” Consider for now the estimate of the mean high temperature in the US on July 1. Doing the work, you find that the temperature for Lubbock TX is missing. Simply dropping it is equivalent to replacing it with US mean high temp from all of the other data. A different method of imputation is to replace it with the mean high temp of a region around Lubbock; then use that in calculating the US mean. Which of these imputations yields an estimate closer to the real US mean, given that neither imputed value is exact? Almost for sure, the imputation based on local stations is better than the overall mean. If there is enough other reliable information, the estimate calculated as the Bayesian posterior mean of a well-chosen locale (or based on values highly correlated in general, as with Kriging) can’t be beaten. But you probably can’t know for sure: the proofs depend on assumptions, and the assumptions are almost never exact representations of what you are working with.

For example … IF we can get a lower MSE by infilling data, then we could run the MSE down to zero by using the “virtual stations” approach I outlined above
Let us not forget, when ever you change, alter, or replace a data point, you must add in an uncertainty band to each estimate. Compare each station’s measurement with it’s difference to any krigged trend that doesn’t use that station. At minimum, an infill must add in at least that mean error.
At least that much error. For you should consider the mean error of the cases where you might want to infill, for who would want to infill a station that reads close to the krigged trend. In addition, I would want an estimate of how much the krigged trend would change for a random omission of 20% of the control points.
So, if people are honest about errors and uncertainty added to the dataset as infills are performed, then it is not possible to drive the mse to arbitrarily low values, but mse will soon increase the more you tamper with the data.

I would like inquire in a little more detail what control is used for the krigging of the regional field?
What do you use for the very first krigging?
Since kigging a regional field is necessary to determine outliers and breakpoint, it follows that the first krigged field has no adjustments for it’s control. There may be breakpoints for gaps in the records.
Someway and somehow you identify a station that needs an empirical breakpoint because it diverges from the regional trend by a key threshold. You insert the breakpoint. One semi-long record becomes two semi-short records.
THEN WHAT? Does that altered station go back into the pool of krigging control points?
What are your options?
A) Remove the station from the pool available for krigging? There soon would be no stations left to give a regional trend.
B) Preplace the original station with the one with an extra break point. ( After all, breakpoints only “improve the data.” right? /sarc. ) Before long the krigged field is totally dominated by stations that have been subjected to prior krigging. We now have a perpetual motion machine endlessly krigging regional trends to test for new break points at stations already ground into pulp.

What defines the radius of control for the regional krigging around any given station under study for breakpoints and segment trend testing?
How far might the most distant control point be?
What are the minimum number of stations that a krigging must include?
Is there a Maximum number?
Is there any weighting or inclusion in the control pool based upon the length of the segment(s)?
I think there is a distinct tendency for breakpoints to appear more frequently when there are more stations in the neighborhood in which to generate a regional krigged field of trends. This is at once logical and a red flag.
It is logical in that that it is difficult to determine an empirical breakpoint in a station record if it is the only one for one hundred miles. Early airport stations, such as DENVER STAPLETON AIRPORT fit this bill. But later in a station’s history, there are many neighbors that can quibble with its trend and suggest breakpoints, deserved or not.
It is a red flag in that the density of breakpoints, if deserved by station microsite issues, TOBS, and instrument changes, are unlikely to be a function of the number of other stations within a day’s drive. Yet that appears to be what is happening.
The empirical breakpoints of a station are at least a partial function of the station’s neighbors and appear not to be intrinsic to the station’s own behavior.

EDIT of 10:24am:
It is a red flag in that the density of breakpoints, if deserved by station microsite issues, TOBS, and instrument changes, should be independent of the number of other stations within a day’s drive. Yet it appears that the frequency of breakpoints increases with the number of neighboring stations.

Stephen Rasey: EDIT of 10:24am:
It is a red flag in that the density of breakpoints, if deserved by station microsite issues, TOBS, and instrument changes, should be independent of the number of other stations within a day’s drive. Yet it appears that the frequency of breakpoints increases with the number of neighboring stations.
That is a really interesting comment. Is it possible that there are more thermometers in the regions that have the greatest change?

@Matthew R Marler at 10:34 am
Is it possible that there are more thermometers in the regions that have the greatest change?
“greatest change?”
Change in what?
I think the existence of a breakpoint at a station has less to do with what happens at the station and more to do with the number of and nearness of neighboring thermometers in the krigged field of trends.
I further suspect that as more stations are broken into smaller and smaller segments and returned to the pool of krigging control points (a guess on my account, see above), the krigged field of trends takes on a “fabricated” shape with unrealistically small uncertainty. Which is what perpetuates the processes desire to keep breaking segments into ones shorter than ten years.

Speaking of mse (mean standard error),
Here is a simple question.
Given a 30 day month’s daily Tave measurements
c(9,11,9,11,9,11,9,11,9,11,9,11,9,11,9,11,9,11,9,11,9,11,9,11,9,11,9,11,9,11)
The month’s Tmean is obviously 10.00,
but what is the month’s Tmse?
StDev = 1.000, Count = 30,
mse = StDev/sqrt(count) = 0.182.
Wrong.
It is a trick question.
We never measure any 9 or 11.
We never measure any daily Tave.
We measure 30 daily Tmin and 30 daily Tmax.
So let’s assume that our
30 Tmins are c(4,6,4,6,4,6,4,6,4,6,4,6,4,6,4,6,4,6,4,6,4,6,4,6,4,6,4,6,4,6)
and
30 Tmax are c(14,16,14,16,4,16,4,16,4,16,14,16,14,16,4,16,4,16,4,16,14,16,14,16,4,16,4,16,4,16)
The monthly Tavg from the 30 Tmins and 30 Tmaxs = 10.000
but the StDev = 5.099, count = 60
Tmse = 5.099/sqrt(60) = 0.658 deg C.
That’s a big error bar when we are dealing with trends of 0.20 deg / decade.
It is a huge error bar if we break long temperature records into sub-decade long segments. For this reason, if no other, BEST’s scalpel is making a bad situation worse.
Even if the temperature sensor is pristine, class 1, and error free, the fact that the original data is daily mins and maxes, means there is about a half degree uncertainty in the monthly mean, which will pass through to the monthly anomaly.
Where in all this work about mse are we keeping track of the original Tmse in the monthly anomalies? Keep in mind, every infilled data point must contain not only the Tmse that should be in the station data, but the Tmse present in the control points of the krigging field. As you infill the data, you are adding noise you must not ignore.

Stephen Rasey, lots of good comments.
Yes, estimate the imputed value and the error of its estimate. With multiple imputation, you draw samples from the data (where the exist) and from the distributions of the imputed values, when estimating the precision of the overall estimate.
The key feature of Samaniego’s book is its evaluation of when the Bayes estimate is likely to be better than the mle and fiducial distribution in practice: basically, iff the prior distribution is accurate enough, what he calls passing the “threshold”. Thus, he supports your idea that the prior is the Achilles heel. Lots of us agree anyway, but it is good to have a scholarly work on the subject.
@Matthew R Marler at 10:34 am
Is it possible that there are more thermometers in the regions that have the greatest change?
“greatest change?”
Change in what?
The greatest change in temperature. You said that there are more breakpoints in the regions that have the most thermometers, and I was wondering if those are the regions with the largest number of industrial developments and such in that period of time. If so, the larger number of breakpoints would be an expected outcome

mse is mean standard error in my example. I was explicit in the formula, name and abbreviation.
I’ve never heard of mean squared error,

Ok, I was wrong and learned something.
mse is mean squared error.
mean standard error (= StDev of sample / SQRT (number of samples) )
and might be abbreviated as sey, sem, or RMSE
http://epitools.ausvet.com.au/content.php?page=UserGuide15

@Willis Eschenbach at 7/2 3:00 pm

have long held that the reason that the standard error of the ocean heat content (OHC) values is incorrectly claimed to be so small is that the error in not carried forwards in the step where they remove the “climatology”. Instead, the “anomalies” with the monthly average values removed are treated as observational points with no associated error … when in fact they have an error equal to the standard error of the mean of the monthly measurements.

I think the claimed error bars on OHC for the years 1955 to 2005 have a much simpler and basic explanation.

Furthermore, the real uncertainty is almost always greater than or equal the sampled data. For how do you know the uncertainty in the data you did not sample? I can plunge a thermometer 300 m at 50 deg N 20 deg W on July 1, 1972. And do it again on Sept 1, 1973. Those two readings do not remotely define the uncertainty in temperatures for the entire North Atlantic for the decade of the 1970s. But to read Levitus-2009, the uncertainty in the Ocean Heat Content prior to 2003 is based precisely on such poor spatial and temporal sampling of ocean temperatures with unrealistically narrow uncertainty bands. Prior 1996, Ocean Temperatures profiles were primarily done for antisubmarine warfare research, and thus concentrated around submarine petrol areas. See Figure 1 Ocean Temperature data coverage: maps b=1960, c=1985. from Abraham, J. P., et al. (2013) (pdf) (from Rasey comment in “Standard Deviation, the overlooked….” WUWT June 15, 2014)

Suppose we want to study the changes in heights and weights for the adult population of the United States. Since 2004 we have been measuring every 1000th adult that walked into every doctor’s office around the country. Before long you’d have a good tight estimate on the mean with a narrow mean standard error. Of course there are caveats. We are not measuring children, just adults. A little deeper in the fine print, we see that we measure people who “walked” into Doctors offices. If they are in a wheel chair, or on crutches, they don’t get measured. Deeper down, we realize that people who cannot get to Doctor’s offices don’t get measured either.
But It is still a good dataset! (If you remember the caveats).
Is there any way we can find some data from before this well designed study started to find any trend in heights of adults back to 1950? “Wow, look what I found! Here is some data from a well designed study, high quality control, that gives me heights and weights of adults. They’ve been doing it for years. It’s just what we need!”
Oh, who conducted the study?
“The U.S. Army medical corps.”
Where did they collect the data?
“Fort Benning and Fort Jackson Basic Combat Training centers”
So…. you have a good handle on mean and standard deviation on adults…. who are fit and young enough to join (or be drafted) into the Army — at boot camp. That’s what you are going to splice onto your new doctor visit dataset?
Levitus 2009 did something very similar.
Today we have a great ARGO data collection effort. It only measures the top 2000 m of ocean. It only reports from open ocean, so forget the arctic regions. Continental shelves are not deep enough for the bouy hibernation. Restricted seas, like the sea of Japan may be over or under sampled. But it is a good dataset.
Levitus looked for a well collected dataset at times earlier than 2003 when ARGO was started. He found such a dataset that was originally collected under contract with the Office of Naval Research who were keenly interested in ocean temperature profiles as an element for the hiding and detection of submarines (our’s and their’s). Very well collected data. The problem is they focused measurements where submarines patrol: NW Pacific, NE Pacific, NW Atlantic, NE Atlantic, Barrents Sea. Within 1500 miles of the coast. (See the Abraham maps in the links above). It is sparse sampling, but high quality — where</b. it was done. But just like the apocryphal boot camp study above, they have nice tight narrow standard deviations on an artificially restricted dataset they pass off as representative of the data they collect today.
It is one thing for Levitus 2009 to have been published. It is quite another that its conclusions are repeated today with a straight face.

Stephen Rasey: Why would it be the expected outcome? I do not disagree. But the reason there should be more breakpoints in places with a large number of thermometers is key to grokking the problem.
On that we agree.

@Willis Eschenbach at 3:00 pm
I have long held that the reason that the standard error of the ocean heat content (OHC) values is incorrectly claimed to be so small is that the error in not carried forwards in the step where they remove the “climatology”. Instead, the “anomalies” with the monthly average values removed are treated as observational points with no associated error … when in fact they have an error equal to the standard error of the mean of the monthly measurements.
Willis, I agree with you completely. I believe with almost ALL work with anomalies, a poor job is being done with the accounting of uncertainty in the statistics.
In my rather lengthy reply of July 3, 9:29 am I did not mean to disagree or discount your observation. I chose instead to pick upon a separate, easier to identify, and more basic error in the OHC (Ocean Heat Content) dataset from measurements made prior to 2003 — unrealistic measurement uncertainty coming from a highly biased and sparse sampling (in time and space) of Temperatures below 300 m.
Levitus-2009, reports a 1.8 x 10^22 Joules, increase from 1955 to 2008. Keeping in mind that it takes 27.5 ZJ (= 2.7 x10^21 Joules) to raise the 0-2000 m interval of the ocean 0.01 deg C, that means Levitus total temperature change is less than 0.08 deg C. As in you Decimals of Precision post, if 100,000 measurments per year from ARGO might justify a 0.005 deg C accuracy, 1000 measurements per year can do no better than 0.05 deg C and 10 measurement/year can do no better than 0.5 deg C.
On top of all that sample bias, I too believe that they are disposing of the sample uncertainty when they are taking their anomalies. But we have to get deeper in the weeds to prove it. Such as how are they binning in time and space the sub 300 meter readings? What do they do with cells that have zero, 1, or two readings? What happens if they have 5 measurements in a vol-time grid cell for May 1965, but no more until July 1969? What happens if a sub 1000m vol-grid cell is never sampled in January until 2004 with the first ARGO to visit the cell?

Addendum to my July 5, 2:58 pm
Answer #1 is correct if and only if the adjustment is the same for all months.
The bulk shift of a station-month temperature record to an Temperature Anomaly record is the addition of a constant with no uncertainty IF and ONLY IF the shift is the same for all months AND ALL STATIONS that it will be compared against.
The head post is about “Problems with the Scalpel Method” of BEST.
BEST uses it’s scalpel by comparing the Temperature Anomaly record of a station with a krigged regional field derived from the Temperature Anomaly records of “nearby” stations. This comparison means that the Trmse, the mean standard error of mean Temperature must be included in the uncertainty analysis when comparing between stations of the same month.
I have show above that the individual station error bars of the mean Temperature of any month for any forstation, derived from the only observed measurements, the daily Tmin and Tmax, is greater than 0.6 deg C if the daily min and max range is only 10 deg. C. I have also shown that with a 0.6 deg uncertainty per month, then a 30 year average for the month will also have an uncertainty of at least 0.1 deg C. Every station in the regional homogenization grid experiences these same uncertainties.
I submit that with these real uncertainties in means, as well as in average maxes and average mins that derive from the raw recorded station data, that it is impossible for BEST or anyone else to determine an empirical breakpoint of any station based upon fit to a regional trend.
The regional trend, which we have no reason to believe is a monotonic surface, has significant fuzzy thickness from error bars. It is a surface where each control point has a Trmse of over 0.6 deg C, an anomaly adjustment of 0.1 deg C. The derived krigged surface has significant uncertainty thickness is then applied to a subject station, who also possesses a 0.6 deg C Temp uncertainty and 0.1 deg uncertainty to it anomaly. Under such circumstances, it is unlikely that any station will exceed the errors present to earn an empirical breakpoint, much less an average of over 5 breakpoints per station.
There are many reasons I reject the BEST scalpel and trend reliance to tease out a climate signal. From day one I had objections based upon information theory and the loss of low frequency (Climate) signal caused by the scalpel and emphasis on temperature trends. The results of BEST’s work on individual stations don’t make sense: It can find 20 station adjustments at Lulling, TX, 8 stations adjustments at DENVER STAPLETON AIRPORT (but misses the opening and closing of the airport), yet it misses the fire-bombing of Tokyo in March 1945.
Even the greatest supporters of BEST must acknowledge the BEST Scalpel requires precision in the data to justify 0.1 degree breakpoint shifts — a precision that does not exist when the raw data are daily mins and maxes.

More on Zeke’s post at Curry on July 7,
of the 547 comments so far, 98 of them are Steven Mosher with sentences so short, terse, and abbreviated of meaning, his word processor must use the BEST scalpel as a plug-in.
Paul Matthews has a short comments that sums up a good deal of the thread.

Paul Matthews | July 7, 2014 at 9:51 am | Reply
Congratulations, you’ve written a long post, managing to avoid mentioning all the main issues of current interest.
“Having worked with many of the scientists in question”
In that case, you are in no position to evaluate their work objectively.
“start out from a position of assuming good faith”
I did that. Two and a half years ago I wrote to the NCDC people about the erroneous adjustments in Iceland (the Iceland Met Office confirmed there was no validity to the adjustments) and the apparently missing data that was in fact available. I was told they would look into it and to “stay tuned for further updates” but heard nothing. The erroneous adjustments (a consistent cooling in the 1960s is deleted) and bogus missing data are still there.
So I’m afraid good faith has been lost and it’s going to be very hard to regain it.

Had to repost this gem of an observation from Patrick B, with Mosher’s typical “read the literature” (where) retort.
Patrick B | July 7, 2014 at 9:48 am | Reply

How could you have written this article without once mentioning error analysis?
Data, real original data, has some margin of error associated with it. Every adjustment to that data adds to that margin of error. Without proper error analysis and reporting that margin of error with the adjusted data, it is all useless. What the hell do they teach hard science majors these days?
Steven Mosher | July 7, 2014 at 11:24 am | Reply
the error analysis for TOBS for example is fully documented in the underlying papers referenced here.

It’s true. Patrick B’ comment is the first time “error” appears in the time stream. It is not in the head post.