The 200 months of 'the pause'

I think I need a primer on temperature. Is the average temperature the average of the daily Tmax and Tmin, or is it the average of more data points through the day? My recollection is that it is the average Tmax and Tmin, and when compared with the hour or minute frequency data, there is quite a large difference in what the average is, much larger than the fractions of degrees for the one sigma stated earlier. Has the global average daily Tmax gone up with time, or the average Tmin, or both? If all of the warming is on the low end, is it less of a problem than on the high end?

http://wattsupwiththat.com/2013/08/27/the-200-months-of-the-pause/

Sea level change is plotted by most publicizing institutions in an extremely misleading manner via showing only total cumulative gain rather than variation in the rate of change. However, doing the latter leads to a striking pattern in sea level rise, cloud cover, humidity at appropriate altitude, and temperature:
http://s24.postimg.org/rbbws9o85/overview.gif
As can be seen, there is quite a reason that the *derivative* of sea level change is almost never, ever, ever plotted in graphs distributed by those favoring the CAGW movement.
Wow, really interesting. I wonder what Lief would say regarding this — usually he jumps all over arguments linking solar state to secular climate trends, but most of the data above seems linked to magnetic field variation and directly measured or proxy-inferrable variations in cosmic ray flux. Given a proposed, physically plausible causal mechanism, the correlations become something more than numerology and are rather compelling especially given that they appear on multiple time scales (including geologic time scales). While I wouldn’t say that they “prove” a solar-climate link beyond mere insolation variation or demonstrate that solar influences on climate suffice to completely trump possible GHE variations, they suggest that if nothing else, GCMs are omitting a possibly important variable. I was already aware of the fact that stratospheric H2O has plummeted in the last five or six years, that so far NASA is not attributing a cause, and that even NASA’s articles on this are pointing out that it should have a -0.5 C or thereabouts effect on climate if sustained, more if further amplified.
Why not break this out and turn it into a top article? A single streaming GIF isn’t the best possible way to present either textual or graphical information, and I hesitate to add the link to my collection of climate-related links because the source link does not have the feel of permanence (where WUWT is archived AFAICT indefinitely).
rgb

I think I need a primer on temperature. Is the average temperature the average of the daily Tmax and Tmin, or is it the average of more data points through the day? My recollection is that it is the average Tmax and Tmin, and when compared with the hour or minute frequency data, there is quite a large difference in what the average is, much larger than the fractions of degrees for the one sigma stated earlier. Has the global average daily Tmax gone up with time, or the average Tmin, or both? If all of the warming is on the low end, is it less of a problem than on the high end?
It’s far worse than you can imagine. Here is NASA/GISS’s direct statement on mean surface air temperature and its use of “anomalies” instead of the absolute temperature. Basically, the absolute mean surface air temperature is effectively not computable.
http://data.giss.nasa.gov/gistemp/abs_temp.html
The assertion is made that the “anomaly” — deviation from some sort of local average — is, though, and the further assertion is made that this anomaly can be extrapolated spatially to cover an area far, far larger than the locally sited surface station producing the data from which it is derived. The data from any given station is also full of holes, and these holes are “infilled”, that is, assigned a value based on averages of surrounding stations (to avoid breaking the code). With few exceptions, this extrapolated, infilled, anomaly result is then spatiotemporally averaged and turned into the “anomaly” you see in the surface data sets, usually without any explicit statement about probable error and whether or not the probable error is symmetric/normal or potentially skewed or biased by the computational methodology used.
The station data is itself transformed in various ways before the anomaly is computed, some of the transformations amount to adding or subtracting a constant amount from the baseline and/or anomaly itself (e.g. to correct for the “urban heat island” (UHI) effect that corrupts almost all of the station data because of how the stations are cited. The data transformations used are different for older data and because they are biased and in some sense completely arbitrary (very difficult to justify on any sort of formal basis) the transformations alone can produce “warming” where it is not visible in the raw data, or (by increasing the presumed UHI) it could conceivably produce relative cooling as well.
It is historically interesting to note that data adjustments in basically all of the major surface temperature records have AFAIK invariably increased relative warming from the beginning of the century to the present, never decreased it. A simple hypothesis test applied to this result, assuming as the null hypothesis that an actual statistical error is as likely to be positive (warming) as negative (cooling) produces a p-value that suggests HUMAN bias in the selection of the corrections with over 99% confidence, because the corrections are the equivalent of many coin flips in a row that all turned out heads. It doesn’t prove that the coin is biased, but it does not support the null hypothesis that it is unbiased and trusting the coin, is in the language of the con, a mug’s game.
The different SAT anomaly data sets primarily differ in the particular set of corrections and the particular algorithms they use for infilling and extrapolating. They also differ in their included data, but of course with a finite data set to select from (especially in the more distant past) there is enormous overlap in their input data and the data sets cannot be considered “independent” in any sense of the word. Nevertheless, the output results from the different algorithms differ significantly — by tenths of a degree or more — every month. I’m a bit of a statistician, and have some knowledge of things like autocorrelation and covariance, and if three different methods purporting to compute the same quantity with (say) 80% overlap in their input data produce a spread of results an average of (say) 0.2 C apart, it is absolutely certain that the standard error in the means of these methods is strictly greater than 0.2 C. Quite a bit greater, in fact. Roughly twice as great (even more if the overlap is greater than 80%).
This same problem plagues the estimation of the number of “independent samples” in Monte Carlo methods based on a Markov Chain of transitions where each succeeding state produced by the chain has a large autocorrelation with the previous state — if you use the raw number of “Monte Carlo” steps as if each step is an independent and identically distributed sampe, you will significantly underestimate the probable error. I have a peer reviewed publication in Physical Review, by the way, that proves this — I’m not making it up.
This is why I brought up the question of error in my comments to Mr. Monckton above — it is a critical omission in the entire discussion of AGW. In figure 1.4 of the leaked AR5 report, for example, a uniform error of 0.1 C is applied to the mean surface data points without discussion or justification. Yet if you go to WFT and plot the actual data from the surface stations, they visibly differ by more than this and are based on overlapping data and partially similar methodology — even the variations in method can hardly be considered to be pulled out of an “independent and identically distributed” hat!
Personally, I think somebody armed with a mouse or spreadsheet just made these error bars up. I think they said to themselves “Hmm, we need error bars or this won’t look like science. The independent coarse grained averages of the surface data have some visible spread, I’ll make up an error that isn’t too big — it might look like we don’t know what we’re doing — and that contains them to make them look reliable. 0.1 C is about right, yeah, that will do it.”
Of course, if I’m wrong I’d love to hear how they arrived at them, because given the overlap in the input data and the visible spread in the result, I think they are underestimating the error by a factor of 2 to 4 (where the 2 generously assumes that there is SOME independence in the computational algorithms used to add to the partial independence in the data).
No matter what transformations and biases that are applied to the surface data at this point, they are pretty much finished. HADCRUT4 is likely to be the last major change that squeezes just a bit more warming out of the very same data, because at this point the surface data is strongly constrained by lower troposphere measurements that are ALREADY diverging somewhat from the surface temperature and by sea surface temperature measurements that do not have UHI type corrections that can be arbitrarily manipulated (although both satellite and buoy data can be arbitrarily manipulated in other ways, it is more difficult to justify any correction that produces systematic time dependent warming, and more obvious that if one is introduced it comes at the direct expense of precision).
The point of including the error is that it adds a substantial range to claims of warming over the thermometric era. Instead of making the bald statement that “temperatures have increased 1 C over 140 years” or the like, one has to make a statement more like “temperatures have increased 1C plus or minus 0.5 C over the last 140 years”. Statements like “X is the hottest year on record” have to be blurred, because e.g. 1930 was well within an uncertainty of 0.5 C of the year X — dozens of years are — so the correct statement is: “We have no idea what the comparative ranking of year X is, but we can say with some confidence that it is the upper third of temperatures in the last 140 years.”
Nobody makes headlines or gets enormous grants from the latter, but the latter is — once one adds in the all important error bars — quite true, depending on what those error bars, honestly and properly computed, really are.
Precisely the same issues arise when trying to compute “anomalies” — the assumption is made that the secular trend of the anomaly can somehow be isolated from the secular trend in the average of the absolute temperature. While this might be approximately true in models with a nearly constant absolute temperature, it requires that the factors that move the average absolute temperature are in some sense orthogonal to the factors that move the anomaly. This is the unwritten assumption in all claims of AGW — that CO_2 is a separable factor that is the primary cause of secular trend in the anomaly but is not a factor in the secular trend in the underlying average temperature and that the two can somehow be independently computed even across very long time bases where many other conditions (including the sparsity and quality of the data) change. True or not — and I very much doubt that it is even crudely true, I think it is just plain wrong — the computation of the separability has to account for independent errors in the two — the error in the underlying secular trend in the average that is NOT due to CO_2 has to ADD to the error in the secular trend in the anomaly (all these silly linear fits to intervals in the surface temperature anomaly). Hence even the error in the anomaly itself properly estimated from a consideration of the independence of the various methods used to compute it and the spread of results is an underestimate of the true error. There is almost certainly a nearly equally large error in the baseline, one that increases into the past. This is evident in the fact that the spread in model computed “average surface temperature” — the baseline to which one must add the anomaly to get a supposed absolute surface temperature as a function of time — itself varies by almost 1 C around 14C. Some places in the literature you will see it is 15 C, for example, not 14 C. Some places it might be given as less than 14C.
We thus have the rather humorous implicit assertion that while we do not know the absolute mean surface temperature of the Earth within 1 C, and while if we added the various anomalies computed in their various ways from overlapping data TO mean temperatures selected within this range we’d end up with an envelope of estimates for the absolute means surface temperature as a function of time that spans at least 1.2 to 1.5 C, and that if we added the uncertainties in the actual anomalies to THAT the range would get even LARGER, we nevertheless do, indeed, know the mean surface anomaly within (say) the 0.1 C illustrated in AR5 and that anomaly can directly and fairly be compared to the similar anomaly computed in 1930, or 1900, or 1870, without any need to concern ourselves with error when positively asserting catastrophic anthropogenic global warming over the latter half (only) of that interval.
In the end, none of these considerations disprove AGW, or CAGW — only the future will prove or disprove the latter, and it is perfectly plausible that humans have had some effect on the climate, although that effect as a secular trend is essentially impossible to separate from non-CO_2 linked secular trends (that is, the “natural variability” of the climate). They do substantially weaken the claim that the historical climate record itself is evidence for CAGW. As independently discussed elsewhere, the rest of the evidence is the global circulation models, and these models independently fail any sort of reasonable hypothesis test when compared to the last 20 years of data and are almost certainly incorrect. I cannot even tell you HOW they are incorrect (and I doubt anyone else can, for all of the “certainty” that it is really this not that), only that one can pretty reliably reject the null hypothesis “this model is correct and predicts the climate future within the statistical spread of results this model produces when conditions are perturbed”, one model at a time. The average of 30+ plus such failed models is then a model that, on average, fails as Mr. Monckton illustrates above, although I think he would do just as well to reproduce figure 1.4 from AR5 is it does an admirable job of this all by itself (as do other figures from the body of AR5, e.g. the spread of model results from selected specific model compared to the actual surface temperatures, however accurate or inaccurate their computation might be).
rgb
[“The different SAT anomaly data sets” … And SAT means? Mod]

What percentage of years in recent centuries has the statement “the last decade was the warmest in the last 1000 years” been true?

Most of the climate models look great… if they only look back in time from when they were built. Happily most of the graphs the warmists push to the media include their post-dictions to give them the credibility they deserve. This makes them seem as though they made accurate predictions for many years before inexplicably falling off. The falling off just happens to begin with the year of the models creation. By the way my model of Superbowl winner predictions work the same way if anybody is looking for a system. I established it in 2005 and it accurately predicted the previous 20 winners. Since 2005 it hasn’t done very well, but there is probably some innocuous factor out there dampening the results, but it will surely clear up at any time and come roaring back to success.