Crowdsourcing A Full Kernel Cascaded Triple Running Mean Low Pass Filter, No Seriously…

“Now the question is how I can improve it. Do you see any flaws in the methodology or tool I’ve developed?”
Because HADCRUT4 data by CRU is greatly incorrect (such as within parts of the 1930-1980 period), as is that from Hansen’s GISS, unfortunately any analysis based on HADCRUT4 is also greatly incorrect. While such has the global cooling scare of the 1960s-1970s occur without substantial cooling beforehand in the global or Northern Hemisphere average, as if it just happened with little basis, I would challenge you or anyone (if inclined to defend it, though you might not be) to find any publication made prior to the CAGW movement of the 1980s onward which shows the 1960s-1970s without showing far more cooling relative to the warmer period of the 1930s-1950s.
For instance, a 1976 National Geographic’s publication of the temperature record of scientists of the time, in the following link, is one I have literally seen in paper form in a library, unlike those repeatedly rewritten-later electronic versions which by strange coincidence (not!) happen to be respectively by a group infamous in Climategroup and by a department which has been under the direction of someone so much an activist as to have been arrested repeatedly in protests:
See that and other illustrations of original temperature data around 40% of the way down in http://tinyurl.com/nbnh7hq
“Are there any particular combinations of data sets that you would like to see?”
While it would be a significant amount of work, digging up original non-rewritten temperature data for history up through the 1970s (prior to the political era, back when there wasn’t motivation to fudge it), digitalizing it, and then carefully joining it onto later temperature data (from more questionable sources but when no alternative) would be a better starting point. Joining together two datasets like that isn’t in theory ideal but the best option in practice; several years of overlap could be used to help check the method. The Northern Hemisphere average, the Southern Hemisphere average, and the global average could best be each generated, as there are reasons, for instance, Antarctic temperatures follow different patterns than the arctic (as the prior link implies).
Of course, like me, you might not have time to do so in the near future even if you desired. But that is something needed yet utterly lacking, as browsing enough skeptic websites indirectly illustrates. The ideal for convenience of other analysis would be both producing plots and producing a list of the values by year (e.g. a data spreadsheet).

To add, regarding this:
I would challenge you or anyone (if inclined to defend it, though you might not be) to find any publication made prior to the CAGW movement of the 1980s onward which shows the 1960s-1970s without showing far more cooling relative to the warmer period of the 1930s-1950s.
The Berkeley BEST set made in the modern day by someone who pretended to be a skeptic without pro-CAGW-movement bias but was an environmentalist (as found out in looking at some of his writing beforehand IIRC) does not even remotely meet that challenge. Only original paper publications (or a clear scan not looking rewritten in any way) made prior to the existence of the global warming movement would count.

RichardLH says:
March 17, 2014 at 6:55 am
“Well they all draw from the same set of physical thermometers.”
So did the data published prior to the 1980s, but that can be seen to be drastically different. When depicting average changes of a tiny fraction of 1% in absolute temperature, of tenths of degree, they depend utterly on the interpolation between widely spaced apart stations, the choice of specific stations used, and, when applicable, hidden adjustments. The practical, convenient way to be certain of no bias in favor of the CAGW movement, without spending thousands of hours personally examining everything, is just to use data published prior to its existence.
In addition to the examples in my prior link, there are others, such as the Northern Hemisphere average history of the National Academy of Sciences, illustrated at http://stevengoddard.files.wordpress.com/2014/03/screenhunter_637-mar-15-11-33.gif in http://stevengoddard.wordpress.com/2014/03/15/yet-another-smoking-gun-of-data-fraud-at-nasa/ , which is utterly different from the CRU/BEST rewritten versions of NH average as well as global average history.
“The head figure is one where I have aligned the 4 major series over the 1979 onwards era and shown how they agree and disagree.”
Obviously, and similar has been seen before, like the undercover CAGW movement supporters on Wikipedia publish a similar plot. However, assuming that RSS or UAH are relatively accurate for the sake of argument, correspondence with them 1979-onwards has absolutely jack to do with disproving rewriting of the pre-1979 section easier to get away with.
If you wish to argue this, then try to meet my challenge:
Find (and link or upload) any publication made prior to the CAGW movement of the 1980s onward which shows the 1960s-1970s without showing far more cooling relative to the warmer period of the 1930s-1950s.
Since you’re arguing this is merely a matter of them all using the same thermometers, that should be easy rather than impossible. Again, what is shown must have been published back then, not a BEST publication of decades later for instance.

Greg Goodman says:
March 17, 2014 at 12:34 am
“Do you have a similar graph that goes back to say 1850?”
I only really trust the CO2 data to 1958. Proxies are… proxies, i.e., not direct measurements. There are no means truly to verify them.
But, this was a hangup with Ferdinand Englebeen. He was keen to point out that, if you accept the ice core measurements as accurate, the relationship should predict too low a CO2 level before 1958, as here.
Besides the fact that I do not trust the ice core data to deliver a faithful reproduction of CO2 levels, I pointed out it was moot anyway, because knowing what happened since 1958 is enough to discount the influence of human inputs during the era of most significant modern rise.
But, if one really must insist on matching a dodgy set of data farther back, there is no reason that the relationship
dCO2/dt = k*(T – To)
must be steady. The fact that it has been since 1958 is actually quite remarkable. But, there could easily have been a regime change in, say, about 1945 which altered the parameters k and To. I showed the effect of such a regime change to To could make the records consistent here. What could that signify? Possibly an alteration in the CO2 content of upwelling waters at about that time. Maybe Godzilla emerging from hibernation stirred up a cache of latent CO2 at the bottom of the ocean. Who knows?
But, in any case, it is moot. Knowing what happened since 1958 is enough to discount the influence of human inputs during the era of most significant modern rise.

Does throw up another question though. Why is it that GISS and HadCRUT are so far apart in the middle? They are close together at the start and the finish, why so different in the middle? I am not sure GISS (or HadCRUT) will thank me that much.
GISS has a (broken, often backwards) UHI correction. HADCRUT does not. Probably other reasons as well. The UHI correction used by GISS maximally affects the ends.
You cannot really say that they are close at the start and finish and different in the middle, because they are anomalies with separately computed absolute bases. That is, there is no guarantee that the mean global temperature computed from the corrected GISS data will correspond to that computed from the uncorrected HADCRUT4 data. Consequently, you can shift the curves up or down by as much as 1C. So perhaps they should be adjusted to be close in the middle, and maximally different at the start and finish.
There are multiple data adjustments in the two series, and while there is substantial overlap in data sources, the data sources are not identical. And then, as Steve M. notes, there are other long running series with their OWN adjustments and (still overlapping) data sources. It is amusing to see how different they are when plotted on the same axes via e.g. W4T, and then to imagine how their similarities and differences would change if one moved them up or down as anomalies around their distinctly computed global average temperatures on an absolute scale. It is even more amusing to consider how they would change if one actually used e.g. personal weather station data that is now readily available on a fairly impressive if somewhat random grid at least in the United States to compute an actual topographical UHI correction with something vaguely approximating a valid statistical/numerical basis for all of the series. It is more amusing still to consider what the series would look like with error bars, but that, at least, we will never see because if it were ever honestly plotted, the game would be over.
This is a partial entre towards not exactly an explicit correction to your approach, but rather to a suggestion for future consideration.
If one does look at your figure 2 above (HADCRUT4 vs smoothed), you will note that the removed noise scales from the right (most recent) to the left (oldest). This is, of course, a reflection of the increasing underlying uncertainty in the data as one goes into the more remote past. In 1850 we knew diddly about the temperature of the world’s oceans (which cover 70% of the “surface” in the global average surface temperature) and there were whole continental-sized tracts of the planet that were virtually unexplored, let alone reliably sampled for their temperature.
This leaves us with several chicken-and-egg problems that your filter may not be able to compensate/correct for. The most important one is systematic bias like the aforementioned UHI, or systematic bias because places like the Arctic and Antarctic and central Africa and central Australia and Siberia and most of the Americas or the world’s oceans were pitifully poorly represented in the data, and some of those surface areas are make dominant contributions to the perceived “anomaly” today (as maps that plot relative anomaly clearly show). Your smoothed curve may smooth the noisy data to reveal the underlying “simple” structure more clearly, but it obviously cannot fix systematic biases, only statistically neutral noise. I know you make no such claim, but it is important to maintain this disclaimer as the differences between different global average temperature anomalies are in part direct measures of these biases.
The second is that your smoothed curve still comes with an implicit error. Some fraction of the removed/filtered noise isn’t just statistical noise, it is actual measurement error, method error, statistical error that may or may not be zero sum. It might be worthwhile to do some sort of secondary computation on the removed noise — perhaps the simplest one, create a smoothed mean-square of the deviation of the data from the smoothed curve — and use it to add some sort of quasi-gaussian error bar around the smoothed curve. Indeed, plotting the absolute and signed difference between the smoothed curve and the data would itself be rather revealing, I think.
rgb