BoM's bomb on station temperature trend fiddling

I await detailed comments from Mosher, Zeke and their ilk explaining that these types of adjustments do not apply to USA institutions such as GISS and USHCN and furthermore Steven Goddard is totally wrong with his discoveries of “adjustments” made to US surface temperature records to “improve” on current global warming.

If the fiddles were not all to push a “Warming Agenda” you might give them some leeway – but this is plain old FOLLOW THE MONEY – FRAUD
Go get ’em Batman…

Reblogged this on Norah4you's Weblog and commented:
Fiddling on violin is one thing. Fiddling with facts such as correct temperatures an other.

Unless a warm bias is proven to be introduced overall, this sort of implication of fraud is counterproductive since it makes popular exposure of proven cases of fraud much more difficult. The usual suspects of the Hockey Stick Team and the Tree House Club will use it against skepticism by merely calling the plots cherry picking, since they lack inclusion of a statistical breakdown of actual impact on the final country wide trend. Chance alone offers a slight bias in either direction, overall. Homogenization itself is here being criticized for exactly what in most science circles justifies its use: lack of data about instrument changes. Twisting this alone into innuendo about fraud or extreme bias amounts to mere motivated and paranoid sounding word games.
Steven Goddard at least has a warm bias to report, but he too simply mixes in widely accepted (to the likes of Steve McIntyre and nearly every scientist you could find) measurement time of day adjustments to his overall stark claims of outright fraud, again subjecting skeptics to ridicule by seasoned and PR firm coached activists many of whom have great technical savvy too.
All of these type of mere smoking gun claims stop the real message of fraud from being successfully exposed, namely fake hockey sticks. Now that the latest one was exposed as having no blade in *any* of the input data thrusting suddenly upwards out of noise, the gig should be up, the Enron nature of peer reviewed climate “science” in top journals having been proven finally outside of black box confusion.
You don’t even have a missing person here yet you are loudly claiming an organization to be murderers, essentially crying wolf as far as that crucial demographic is concerned, the one that still figures skepticism is indeed just another right wing attack on science.

“Nero fiddles while Rome burns”

Nik, you are one track obsessed, reposting without following. TOB is not a part of the BOM explanation. Weasel worded might haves are, and they equate to poor science. The post is accurate. There are many logical reasons to conclude the adjustments are FUBAR. One thing is recognizable however, the adjustments outside of TOB, also cool the past and warm the present.
This, in combination with dozens of other climate science scandals, amounts to quack science. In pursuit of an explanation from the curious, many times the BOM ducked like a quack.

@John from Au……….How do I load your data into Excel?

Hey Mosh,
You’ve mentioned that BEST does spot checking between the GAT field and test station measurements. Since you normalize for latitude, altitude, and whatever else you adjust for, what is the process to prepare for the test?
Do you (BEST), un-normalize the field value at that point, or do you take the station values (min and max), and run the same normalization process to convert it into a single point field value?

Whenever one has many data points with a certain error in measurement, then one takes the average in order for the too high measurements to offset the too low measurements.
What one CANNOT do is take the too low measurements, move them upwards to better match the average, then recalculate the average.
Latitude is correct. If the microclimate data does not change the trend in the overall average, then the data needs no adjusting, res ipsa loquitor.
Remember that we are talking about 0.1 degree or less per decade. It has been demonstrated that in most cases past tmeps were moved downward and more recent temps moved upwards. That will indeed move the average trend enough to see at a resolution of 0.05 degrees.

In what other area of science can are you allow to make up data if your actual data is no good? Wouldn’t it be scientifically more rigorous to discard bad data? Isn’t it better practice to start tracking a site as new if it’s relocated instead of the fiction that the data is continuous?

Well Nick is great for entertainment value at least hahaha

Oh, my. there is a serious problem with Nick’s explanation of Ambley.
If the site was moved in 1980, then only the data before 1980 would need to be adjusted so that you could get a trend line. If pre-1980 the thermometer reading was .7 too high, then you, Nick, are saying that it was .7 too high compared to the post 1980 readings after the thermometer was moved. But that is not what you are saying. You are saying it was originally in a spot that was .7 too warm, and then in 1980 it was moved to a spot that was .7 too cold.
If you adjust the pre and post 1980 numbers, you are saying BOTH locations were inaccurate compared to some imaginary midpoint. Mathematically that seems OK, but in reality, it doesn’t work unless by some improbable chance the thermometer was moved from one location that was exactly .7 too warm, to a location that was EXACTLY .7 too warm at exactly August 1980. Possible, but unlikely.

Sometime they move without actually moving:
Analysis of the 100-year ­record at the station shows a cooling of 0.35C in the raw data had become a 1.73C warming after “homogenisation” by BOM.
A review of the data by independent scientist Jennifer Marohasy shows the warming trend had been achieved by progressively dropping temperatures from 1973 back to 1913.
For 1913 the difference ­between the raw temperature and the BOM homogenised figure was 1.8C.
BOM said the discrepancy in the data was consistent with the thermometer site moving from a farm building on a small hill outside the town to its current ­location on low-lying flat ground. Minimum temperatures are normally higher on slopes than on flat ground or in valleys.
However, the official catalogue of all stations used to make up the national temperature record says the Rutherglen thermometer is an automatic weather station in the grounds of a research farm, 7km southeast of Rutherglen. Not only has the station not moved since being ­established in 1913, it’s “well outside the town area, on flat ground over grass but with low hills a few hundred metres to the north”.
“There have been no documented site moves during the site’s history,” it says.
BOM has so far been unable to explain that discrepancy.
Retired scientist Bill Johnston, who has worked at Rutherglen, said a temporary thermometer had been put on higher ground near the office of the farm but it never provided temperatures to the bureau.
http://www.theaustralian.com.au/national-affairs/climate/climate-records-contradict-bureau-of-meteorology/story-e6frg6xf-1227037936046

Mario Lento
August 26, 2014 at 12:32 am
Mosher clearly has his BEST interests in mind…

In fact BEST’s results were used as a defense for the BOM corrections. What bugs me isn’t the homogenization so much as the apparent alteration of individual station records. Homogenized data should only be used to create a separate table of data entirely, area weighted, and not linked to any specific station(s). The resulting records can be tied to a centroid defined by the polygon delimited by the locations of the various stations used in the local homogenization. Also, the entire correction debacle seems to ignore the fact that if there is a global trend in a phenomenon, then that trend is going to be present in any series of measurements of that phenomenon regardless of whether raw data is corrected for measurement biases like TOBs and step changes or not. The only measurement problem that could seriously affect a trend estimate done properly would be a changing instrument sensitivity to the phenomenon being measured over time. Step changes would only be a problem if the trends were calculated over the step rather than treating the step as a discontinuity. In fact the timing, direction and extent of step changes should be analyzed separately, since they can offer distinct and independent data on the phenomenon

Steven Mosher
August 26, 2014 at 8:25 am
They are called un documented station moves. Happens all the time.
Some of the most un reliable data folks have is the meta data.
There are also metadata errors that are the result of “apparent” location changes that are pure humbug. Not long ago there was a discussion regarding a station on Long Island that in reality moved about 10 meters max. The metadata has it hopping all over a half-mile square area. The only real change was the installation of an automatic system in the same yard, a few meters away from the original Stevenson screen. The move would have been (barely) detectable with a consumer grade gps unit. No automatic (computerized) system on the planet will be able to determine this.
….
Bottom line: you get the same global answer whether you include cooling stations or not.
whether you adjust them or not. Given the absence of any physical mechanism to explain how
one patch of earth can cool while the rest warms, Given that the metadata record is not
gods word, given that the global answer doesnt change, it is reasonable and justifiable to apply occams razor and assume that the metadata missed a station move. It’s the most simple explanation.
The real bottom line from a methodological perspective is that if what you say above s true, there is no justification for any “adjustment” of the data.

Have you looked at the graphs that I showed? The need for the adjustment is obvious.

Did you read anything at all that I wrote?
And clearly, we have very different meanings for the word “obvious”.
There are three issues here: a) identifying a “rejectable” data outlier on the basis of some objective statistical criterion; and b) instead of rejecting it as an outlier, claiming that you can fix it; and c) once you’ve fixed it, including it in the data averages and error estimates with the same weight.
a) is basically impossible. As the wikipedia page on data outliers points out, identification of outliers is an essentially subjective process (by which they mean that one cannot justify it with the a priori application of statistical principles, one has to at some point make a subjective choice as an implicit Bayesian prior). If one makes the usual assumption of a smooth unimodal normal distribution, one basically cannot possibly argue that given four samples as in your graphs above, one of those samples qualifies as an “outlier” simply because the trends do not agree. You don’t a priori know what the correct trend is, or what the correct variance should be, for the four sites you select. Consequently, you cannot make a quantitative statement for how likely it is that your correction is correct, or what the distribution of reasonable corrections might be.
b) OK, so according to your subjective beliefs, it is an outlier even with only four samples. So reject it. Don’t claim that you can fix it, and acknowledge that in doing so, you cannot really shrink the variance as much as you would like. The existence of the outlier and the lack of metadata means that you cannot be certain that you understand either why it is different or why it appears perfectly reasonable and well-formed and yet cannot be right. Your evidence that it isn’t right is weak at best and can equally easily be interpreted as evidence that the other three sites are deviating the other way systematically from a “true” behavior somewhere in between. In this case just going from four samples to three is going to substantially lower N-1 (from 3 to 2) but again, the resulting sample standard error is going to be too large as it fails to account for the Bayesian prior probability that your rejection is in fact justified. It might not be, and you cannot be certain that it is.
c) But whatever you do, don’t try to fix it! This adds several more degrees of freedom — the “fit parameters” of your fix, in this case two independent numbers. You now have several Bayesian (subjective) assumptions — that the rejected data is in fact an outlier, that the other data is in fact accurate, that the rejected data is an outlier for a specific, modellable cause (less likely than it is an outlier for any possible cause) and that the rejected data can be “fixed” by optimizing the model-adjusted data against the remaining unrejected data. This process by definition is going to (on average) preserve the mean trend of the unrejected data, since that (or something extremely similar) is the criterion you optimize against. If you then use the fixed data to form the mean/determine the trend it is a self-fulfilling prophecy from the unrejected data — you merely affirm your subjective beliefs by optimally fitting the rejected data to conform to them. This is fine, but you can hardly blame people for not agreeing with your subjective beliefs.
The problem arises when one tries to form the standard error from the data including the fixed data. That data came with a triple price tag of Bayesian assumptions, each one of which should have reduced your certainty that your fix was correct. It certainly no longer counts as an “independent and identically distributed” sample from the point of view of evaluating standard error. You have gained nothing in the way of certainty along the way. You’ve simply substituted a subjective decision to deliberately reduce the variance of your samples around the original mean, the one that conforms to your subjective expectations, for the objective variance of the actual samples around an intermediate mean, or the objective variance of the reduced number of unfixed samples around their even less reliable mean.
The point being that your corrections could be correct. It might even be the case that they are probably correct, although I think you’d be hard pressed indeed to make that a quantitatively defensible assertion (which all by itself should give you pause, by the way — perhaps one can, as Mosh asserts, use global metadata from many other sites to justify a posterior model used to “fix” the data, but that seems very dicey to me as it makes a lot of assumptions about local spatial homogeneity of temperature trends that I’d be very skeptical about just based on looking at temperature measurements sampled at different sites in my own back yard over time, let alone sampled in different yards tens of miles or more apart).
rgb

I agree with RGB, and even further, this needs the KISS principle applied, the record is what it is and no need to artificially change anything especially using an algorithm that assumes and applies its assumptions, so that each specific period keeps changing. the simple thing is most agree that over a century we have warmed slightly under a degree C, we also know in a simple observation that the bigger cities have a UHI that can be three to four degrees or more above rural temperatures.
Those of us that have lived through many really hot summers in Australia know the difference between those hot summers and the last few years that have been very mild indeed, apart from the BOM and CSIRO clamour when averaging or smearing desert heat to urban areas just to give them a half degree above a past historical “record”!! But then we find the historical records have been adjusted down and the modern temperature record subject to automatic adjustment.
Seems to me that in the cities, the UHI SHOULD boost any record by at least four or more degrees, but that is just not happening, they struggle by artificial adjustments to concoct these claimed half degree “record” hot temperatures and ignore any acknowledgement of colder temperature “records” and that in itself exposes the confirmation bias of those involved
Then when you take a simple approach to Global temperatures, it is apparent that world surface temperatures have not risen for 15 to 18 years depending upon the global temperature set used and the stupid thing is that the actual previous warming period is shorter than the present hiatus that is likely to become a downward temperature trend.
Suggestion for Mosher and Stokes that p**ing into a gale force wind on the good ship warming is now reaching a ludicrous stage and time to pack it in and admit the obvious, the public don’t buy stupid propaganda even if it is dressed up, mixed up and homogenized by algorithm trickery!