There has been some discussion of figure 1.4 in the AR5.
There are a number of odd features about that graph. Today I want to make an observation limitingmy comments to how the authors of the AR5 shifted the multi-model mean relative to the value in the AR5. I can say these three things: The AR5 authors shifted. Their decision to shift was entirely unnecessary and injects confusion for no good reason. However, quantitatively, the the magnitude of the shift is small. So, quantitatively any problems in the figure lie elsewhere.
To begin, I would like readers to focus on a specific feature in the figure above: The ‘projections’ from the AR4 are are aligned to match the observed temperature in 1990. The text below the figure in the final draft reads
“Values are aligned to match the average observed value at 1990. Observed global annual temperature change, relative to 1961–1990, is shown as black squares 7 (NASA (updated from Hansen et al., 2010; data available at http://data.giss.nasa.gov/gistemp/); NOAA (updated from 8 Smith et al., 2008; data available at http://www.ncdc.noaa.gov/cmb-faq/anomalies.html#grid); and the UK Hadley 9 Centre (Morice et al., 2012; data available at http://www.metoffice.gov.uk/hadobs/hadcrut4/) reanalyses)”)
I’m guessing the values were aligned to match the average of the three observed values in 1990.
Key Principle For Comparison
In my opinion, one key principle in fairly comparing data to a published projection is that if the projection clearly and specifically defines any feature of the projection then that feature should be retained when comparing published projections. A second principle is the “apples to apples” principle, which requires that, as much as possible, observations and projections should be compared on the same basis. There are a few other principles, but these two are sufficient to discuss ‘the shift’.
To give an example of the first principle: In the AR4, the authors specifically and repeatedly stated that their projections were all relative to the 20 year baseline from 1980–1999. These can for example be seen in the SPM where among other places, we read
“Figure SPM.5. Solid lines are multi-model global averages of surface warming (relative to 1980–1999) for the scenarios A2, A1B and B1, shown as continuations of the 20th century simulations. Shading denotes the ±1 standard deviation range of individual model annual averages.”
In the heading for Table SPM.3. we also see:
Temperature Change (°C at 2090-2099 relative to 1980-1999)a
Note there is nothing remotely vague about baseline used for projections: The projections in the AR4 were not based temperature in 1990 specifically. The model anomalies are expressed relative to the model mean temperature over the 20 year periods from Jan 1980 to Dec 1999. Applying the 2nd principle: even if the AR4 authors did not specifically say, “When comparing model projections to observations, we should use the same baseline for models and projections”, that ought to be done. So, we can assume that the authors intended that projections based on ‘1980-1999’ should be compared to observations based on ‘1980-1999’ as opposed to projections based on ‘1980-1999’ should be compared to observations relative to the specific year 1990.
Already we see a problem with the figure in the AR5: The comparison between AR4 projections which were described as based on ‘1980-1999’ are compared to observations relative to the observed temperature in 1990. Or something. The questions one might ask then include:
- Does the shift in baseline the AR5 authors elected to make have any noticeable effect on the comparison?
- If it does, does the shift in baseline make the AR4 projections seem better or worse than they actually were?
A feature of the AR4 Projections
Below, I have plotted the 13 month average multi-model mean AR4 Projections (black) along with a least squares line (purple) fit through the model-mean monthly temperature anomalies based on the baseline stated in the AR4, i.e. 1980-1999)
In my dating scheme, January 1980 is date “1980.042”. So, the mean ‘date’ over the baseline is 1990. The mean temperature anomaly for monthly values during the baseline is 0 (by definition). These two means are shown with dashed black lines. Notice that through the mathemagic of least squares fitting, and use of baselineing the trendline value corresponds to an anomaly of 0C on 1990.0.
For now, notice the 13 month average temperature for centered on 1990 is higher than the trend line fit through the data. Already, anyone capable of doing the mental shift might notice that it would be possible to make the projections appear “cooler” than they were when published by deciding to shift the baseline from the stated model mean over “1980-2000” to relative to the model value from 1990 specifically. That is: If the projections had been stated relative to the model value averaged over Jan 1990-Dec 1990 specifically the projections would have appeared in in the dark red line below which is shifted by subtracting the model mean value over the months from Jan 1990-Dec 1990 from the multi-model mean, thus forcing the projection through zero averaged over that set of 12 months. (The shift happens to correspond to 0.085 C).
As you can see, had the authors published projections relative to the average over 1990, would have looked qualitatively similar, but the numerical values would have been relative to 1990. In particular the numerical values in the shifted projections would be 0.085C “cooler” because 1990 was a warm year in the models.
Right now, some of you are thinking “Wow! That looks bad! They shifted the projections down 0.085C”. I also admit that initially, that’s what I thought they did. (Update Oct. 4: Tamino’s rebaselineing to explain models are pretty good is is somewhat similar to this though. See Oct 4 post)
However, that’s not what they did. The example above merely illustrates how much one can change projections if we change baselines in ways that might “sound” reasonable on quick reading.
What the AR5 authors actually did is not quite so bad.
In fact, turns out that the effective shift imposed by the AR5 authors is tiny. At least when they changed the baseline used by the AR4 models for projections, they also changed the baseline for the observations. That is: they made an “apples to apples” comparisons– just using a different baseline from that in the AR4. This particular shift turns out to not be so bad because both the models and the observations were “warm” relative to the AR4 baseline. So, the AR5 authors ‘shift’ was to move the projections to a point where observed 1990 matches model mean 1990. (The also then did a bulk shift to make the baseline correspond to an even earlier period. That is of little consequence). Below, I show the model mean (black), the model mean for 1990 (dark red), and the observed value averaged over GISTemp, HadCrut4 and NOAA/NCDC for 1990 (green).
When aligning the projections to data, the authors of the AR5 essentially shifted multi-model mean projections (black line) such that the model mean 1990 (red dot) was equal to the the observation (green square). The result illustrated relative to the 1980-1999 baseline is shown in the green trace:
Notice that the shift is negligible and more over the projections “warm up” relative to the data.
So why does the new graph seem to indicate the models are ‘ok’?
The difficulty here is that the AR5 graph shows the large grey ‘uncertainty bands never shown in the AR4. The AR4 itself showed uncertainty bands equal to 1 standard deviation of the multi-model mean; for the A1B scenarios, the width of those bands are shown with the bright yellow arrows . If you examine the graph below you will see the data are firmly outside that region:
That is, of course, exactly as I have been noting for some time. If we examine where observations fall relative to the projections as made in the AR4 the observations have been dancing in and out of the range of the models. This is true in both temperature anomaly and more distinctly in terms of trends. (The later are not sensitive to this shifting of model baselines and in many ways better for that reason.)
On the other hand, the observations would appear to be well inside the extremely large grey bands which did not appear in the AR4 and which have the rather interesting feature of showing zero uncertainty in 1990. So, the question is: What are those bands? I’ll look into that more specifically later on (mostly because I haven’t looked into it in detail). But it is intriguing that they show zero uncertainty in 1990, which tends to suggest there is something a bit odd about them. 🙂
Why is SAR so small?
Actually, my own calculations, using AR4 projected temperature data, downloaded from the IPCC website, suggests that the projected anomalies are relative to 1980-2000, not 1980-1999.
This makes the official multi-model means in, for example, scenario A2 approximately 0.013c too low, with similar differences for other scenarios.
I could never get those responsible for calculating the anomalies to accept this but I suspect it was simply a miscalculation, even as basic as using too many rows in a spreadsheet.
In my opinion, another significant fact, is that while the observed temperatures are often within the lower range of the grey bands, they are NEVER within the upper range, as one might expect in unbiased models.
Is the grey bands the merge of all spaghetti?
The grey band appears to extend the combined range of all the colored projections, by 0.2 degrees in each direction. I think the intent of this was to provide a “weather” band, as the projections are generally of the forced response only. One might reasonably expect that, even if the models have that correct, observations in El Nino / La Nina years would deviate from the forced response but stay within the 0.2-degree grey area. For 1990, there’s no need for a grey band because the projections are constrained to the observed value at that point.
Not a great idea to try to depict this “weather allowance” on the graph, though, in my opinion.
HaroldW, IIRC .2C is how much they claimed was “explained” in Ch 8, but I don’t think it was just weather. I could be wrong because it is hard to say just what they were saying since they spent so much time explaining what they were not saying wrt what they were saying.
Mike: I suspect the grey band is the spread of all spaghetti forced to match over 1990. We’ve discussed the effect of that before when discussing paleo reconstructions. matching over one year rather than over a long baseline has the effect of:
1) making the ‘uncertainties’ appear to be zero in that year. This is because all runs are baselined to that year and so all “agree”. The Agreement is achieved by a mathemagical operator called “subtraction”.
2) drastically widening apparent uncertainty in years far away from that 1 year. If the only reason for the ‘uncertainty’ was weather, and weater was just gaussian white noise with with noise having a standard deviation of σ then the “uncertainty” would be zero in the ‘baseline’ year and ~1.4 σ in all other years. That is: if we had a simple system, this method essentially makes “noise” look 40% larger that it would look if we used a sane method of baselining. (That is: sane means using a long time– as the AR4 actually did and ‘not sane’ being baseline to 1 year as in the AR5 comparison.)
I’m not certain that’s what they did, but it looks suspiciously like the did this. If so, this is nutty. But I’m going to make graphs later.
Note by the way: When making their own projections, the authors of the AR5 are using a long baseline, not a one year one. So, it would be odd for them to interpret the AR4 “baseline” as a one year baseline for the purpose of comparison when:
1) the AR4 uses a long (20 year) baseline to make projections.
2) the authors of the AR5 use a long baseline to make projections and
3) It’s nutty to use a short one (because it unnecessarily inflates the appearance of uncertainty when you could have “less” by merely doing the comparison in a rational way.
Lucia, take a look at the Technical Summary in AR4 as it contains highly relevant information on projections, both for AR4 and the earlier reports.
Ray
They say they use 1980-1999. I’m sticking with 1980-1999. Moreover, if there is a mistake in that regard, it makes very little difference in projections or anything else.
Lucia, also note that Figure 1.5 in the Second Order Draft is limited to the AR4 projections. This figure was disappeared in the Government version.
For whatever this is worth:
Using the original Figure 1.4 with the gray shaded areas, take the intercept of the lower boundary of the TAR in 2013 and follow a horizontal line backwards in time (i.e. at 0 degrees slope) until it intercepts the upper boundary of the FAR.
The intercept with the FAR occurs in about 1997 just above the upper confidence boundary of 1997’s observed temperature range.
This line covers a period of about 16 years. By definition, no part of the flat line between the FAR Top in 1997 and the TAR Bottom in 2013 crosses into the gray shaded area.
This means that even if the observed temperature trend between 1997 and 2013 had been completely flat, defenders of the IPCC’s past work could still claim in 2013 that observed trends are “consistent with past projections” if they have to include all past projections as their criteria, not just AR4’s projections.
Unfortunately, making that kind of claim as a public relations gambit wouldn’t make a lot of sense given the climate’s supposed high sensitivity to ever-increasing concentrations of GHG’s.
If ocean heat sequestration isn’t credible as an explanation for The Pause, and if volcanoes aren’t credible as an explanation for The Pause, then another means of hiding this obvious inconsistency is to make a new Figure 1.4 which looks robustly scientific while at the same time obscuring, with a lot of intricate detail, the obvious questions raised by a 16 year pause.
Hence a new Figure 1.4 appears, and it is a real work of art — everything its creators should have expected of it.
The GCM results for the GAST reported in AR5 are consistent with projections made in the peer-reviewed literature in 2001.
Long-range correlations and trends in global climate models: Comparison with real data
Abstract
We study trends and temporal correlations in the monthly mean temperature data of Prague and Melbourne derived from four state-of-the-art general circulation models that are currently used in studies of anthropogenic effects on the atmosphere: GFDL-R15-a, CSIRO-Mk2, ECHAM4/OPYC3 and HADCM3. In all models, the atmosphere is coupled to the ocean dynamics. We apply fluctuation analysis, and detrended fluctuation analysis which can systematically overcome nonstationarities in the data, to evaluate the models accordingto their ability to reproduce the proper fluctuations and trends in the past and compare the results with the future prediction.
Lucia,
“the observations would appear to be well inside the extremely large grey bands which did not appear in the AR4 and which have the rather interesting feature of showing zero uncertainty in 1990.”
.
Maybe the broad grey band describes the “indicative likely range†shown on figure TS.14.
.
The breath of the grey band is such that no plausible projection of warming (or cooling!) could fall outside that band. This is a meaningful projection? Such graphics are a bit like a bad joke: not funny at all, and more than a bit sad.
SteveF
Re: “The breath of the grey band is such that no plausible projection of warming (or cooling!) could fall outside that band.”
That may be the most realistic depiction yet of the state of the art in “climate science!”
I thought the issue was over the authors replacing that graph with a new graph in the final edition where the starting point is shown approximately 1.5 degrees lower. I don’t think many people would have had an issue with this graph.
Shenanigans24
It’s best not to jump to the conclusion that there is only one issue. There can be many. But this graph has been discussed, and I’m going to discuss the uncertainty bounds later. It appears there have been several ‘shots’ at making a decent figure 1.5, and possibly all are flawed, but in different ways.
I argue that the noisy fluctuations are causing problems with the interpretation. In that case, remove the noise and look at the underlying trend:
http://contextearth.com/2013/10/04/climate-variability-and-inferring-global-warming/
Re:WebHubTelescope (Comment #120003)
October 4th, 2013 at 12:16 am
WHT,
Not exactly on topic, so a very short response – your methodology for abstracting the long wavelength trajectory contains two important flaws. The first of these is the problem of aliasing of the long wavelength signal in the process of fitting your noise terms. Troy (Masters) wrote an excellent three part series highlighting the problem. The third part is here, and provides links back to the first two parts:-
http://troyca.wordpress.com/2013/02/
SteveF also examined the question in a two part series which adds a lot to the discussion:-
http://rankexploits.com/musings/2013/estimating-the-underlying-trend-in-recent-warming/
http://rankexploits.com/musings/2013/more-on-estimating-the-underlying-trend-in-recent-warming/
The second flaw is the assumption that you can take a limited subset of the total change in forcings to do what you do.
Your conversion to a climate sensitivity via correlation is also quite fundamentally flawed (as I thought I had already demonstrated to you w.r.t the BEST paper).
You might try this very quick and simple experiment to test the validity of your abstraction methodology. The longest period cycle evident in the temperature series is about 60 years. First, try plotting the derivative of temperature series calculated on a moving 31-year basis i.e. fit an OLS gradient to each successive group of 31 points and plot the result against mid-year-date. What do you see? Secondly, try replacing the temperature series with a 65-year moving average, and overplot on your abstracted trajectory. What do you see? I suspect that these two simple exercises might give you some added insight into your abstraction methodology.
paul_k,
I did what you said and I’m not surprised at what I see. I don’t have a SP background but the 65 year MA completely obscures the fluctuations in the 31 year trend. doesn’t the addition of the AMO index from WHT’s latest link help with that? but
WHT,
What happened to the diagnosed TCR when you added the AMO index?
Oh I see (@WHT). Adding the AMO gets better agreement at mid-century but starts to lose fidelity in replicating the pause.
To which the response is hmmmmmmmm
@BillC,
Exactly. Have a look at WHT’s Figure 5 which is perhaps more revealing. The residuals from the fit are enforcedly mean-centred by the LS methodology, but they exhibit some first and second order character, typical of a mis-specified model; and equally importantly they show a clear cyclic signal which has not been explained by the various datasets which WHT has added in. (High at 1880s, low at 1910, high at 1940s low at 1970s, high in 2000s.) His late time trajectory aliases some of this cyclic signal and converts it into long-wavelength trajectory – which is why it heads in the wrong direction relative to the pause.
You can compare his residuals and his late-time abstracted gradient with this analysis here (see Figure 6):-
http://rankexploits.com/musings/2011/noisy-blue-ocean-blue-suede-shoes-and-agw-attribution/
with follow-up here:-
http://rankexploits.com/musings/2012/more-blue-suede-shoes-and-the-definitive-source-of-the-unit-root/
Unlike WHT’s analysis, the residuals from the above analyses have no first or second order character (zero coefficients for a quadratic fit) and no residual cyclic character above a periodicity of 22 years. The late-time gradient is close to 0.1 deg C/decade – identical to more sophisticated published approaches to this problem (like Empirical Mode Decomposition applied by Wu et al).
Lucia,
Excellent and meticulous post. I await the second installment with popcorn and impatience.
You probably noticed that Richard Betts stepped in to offer his explanation of the SOD and final figures on Climate Audit. In the same thread there are two very powerful posts by Dr Brown of Duke (posted under “rgbatduke”) that are a “must-read”.
http://climateaudit.org/2013/09/30/ipcc-disappears-the-discrepancy/#more-18425
“On the other hand, the observations would appear to be well inside the extremely large grey bands which did not appear in the AR4 and which have the rather interesting feature of showing zero uncertainty in 1990. So, the question is: What are those bands?”
Is that the question? It strikes me that a rather better question is along the lines of ‘hang on a minute, don’t those error bars mean that the difference between ~0.1 degrees of warming and ~1.4 degrees, by 2015, will not be reliably measurable?
When I initially examined the first Figure I thought that what the authors had done was to normalize all the model data-points in 1990 to the measured temperature; this gives you a zero CI, and gives one a common place for all the models to run. This will also flatter model runs which are going astray.
Having read your post and SteveM’s, I still think this is what was done in the original figure.
In the update, they used an average period to zero the models, which gives a wider set of IC’s at the start and throughout the run. Allowing wider CI’s gives them the ability to state that recorded temperatures match model returns.
“The difficulty here is that the AR5 graph shows the large grey ‘uncertainty bands never shown in the AR4.”
Umm the newer figure doesn’t feature the grey uncertainty bands?
You didn’t want to discuss the new version?
“It appears there have been several ‘shots’ at making a decent figure 1.5, and possibly all are flawed, but in different ways.”
Yes, which is generally the reason for ‘drafts’
Dave,
Only if those error bars are useful. I doubt they are.
Nathan
Yes. I will be doing so later on. All versions are ‘out there’ so I think some people want to understand what was done in each and which analytical choice resulted in each feature. Of course, I get you may not wish to know this, but in that case, you can chose to not read posts discussing things that do not interest you.
Sure. And when I discuss the final figure, we will see that it’s got issues too. FWIW: Tamino and Skeptical science discussed the fitugure I have discussed. Tamino said some really stooopid things about it. So, I will likely write another post discussing this figure. If that post does not interesting you, feel free to not read it. If you read it and your only comment is to suggest that you are not interested in the post and I should be writing a different post that interests you more, I will moderate you– because I’m tired of that sort of silly, childish behavior on your part.
The SOI has a strong reversion to the mean and that mean is zero. Subtract that out of the global temperature signal and the pause or hiatus is removed completely. Knock yourself out if you want to add in the volcanic disturbances and/or the AMO signal — these reduce the variability but it isn’t a make-or-break situation. The TSI fluctuations are inconsequential.
It changes the TCR from 2.1C to 2C, I will alert the media.
http://imageshack.us/a/img163/831/d62g.gif
http://imageshack.us/a/img571/5131/kq0.gif
If it weren’t for two narrow warming spikes during the WWII years centered at 1939 and 1945 where would this discussion be?
My main point is that you all have this strong principle component called the SOI, which has nice properties of reversion-to-the-mean boundedness and zero-bias, but no one wants to apply it. Very unscientific of you all.
OT. Midst all the excitement of the AR5 discussion Roy has reported a big(gish) jump in UAH temperature.
http://www.drroyspencer.com/2013/10/uah-v5-6-global-temperature-update-for-september-2013-0-37-deg-c/
GL
Re: WebHubTelescope (Oct 4 07:30),
And you continue to ignore the fact that there are other greenhouse gases than CO2, so even if your method for extracting climate sensitivity were correct, the result would be too high.
It’s not us that’s ignoring long period cycles in the climate like the SOI and the AMO. You can find a number of posts on that sort of thing here. It’s the climate modelers and by extension the IPCC. Because of this, the modelers are forced to use the aerosol kludge to get their models to at least sort of look like the historic record.
DeWitt,
The other GHG’s are included in the definition of the CO2 sensitivity. It is well known that CO2 alone gives a climate sensitivity of 1.2C per doubling, and the control knob of CO2 (carrying along water vapor) and the industrial pollutants that scale with this value give an ECS of the estimated 3C. It is all wrapped up into one metric for convenience. Hope that helps your understanding.
As per your second paragraph, I do hope that analysts on both sides of the fence apply the SOI routinely. I don’t think Tamino has used it, but after reading Kosaka and Xie’s work, he appears quite excited about it. There are other commenters on SkS that are applying the SOI correction, such as Kevin C and Icarus, which is where I got my motivation to do this analysis from.
Having a conversation with the skeptic Clive Best, he estimates a TCR of 1.7C, while I get 2C. There isn’t a whole lot of difference that separates the skeptical camp from the “warmista” camp. I am only suggesting that how to interpret the fluctuating SOI contribution gives us the gap of 0.3C in TCR and potentially 0.5C in ECS.
HT–
I know. discussing the AR5 is interfering with my responsiveness to betting. I saw some comment on that yesterday afternoon, I need to get that post up this afternoon. (And decree winners on the ice!)
WHT,
“Subtract that out of the global temperature signal and the pause or hiatus is removed completely.”
.
I have seen you make that same claim elsewhere, and it is simply not correct. The exact number for the recent trend depends on some assumptions (eg. are sensitivities to solar and volcanic forcings approximately the same?), but if ENSO is accounted for, warming since 1998 is more like 0.072C – 0.09C per decade, and warming from 1978 to 1997 about 0.16C – 0.165C per decade (see the second of Paul_K’s links to my earlier posts on this subject). So while ENSO (and volcanoes) have certainly caused a significant part of “the pause” there is plenty of evidence that the underlying trend has indeed slowed. The cause for that slowing is for certain not clear, but longer term cyclical behavior (a ~60 year long pseudo-oscillation of magnitude +/- 0.1C to +/- 0.15C) seems consistent with the historical record.
.
WRT your two linked graphs: Please explain why you think that a plot of CO2 versus warming is a reasonable way to determine TCS or ECS, and especially how you can justify ignoring all the non-CO2 GHG forcings, all aerosol forcings (positive and negative), and all non-GHG forcings like land-use changes, carbon black on snow, etc., as well as the rate of ocean heat uptake?
.
BTW, do yourself a favor and stop telling a bunch of experienced scientists and engineers that they are ‘unscientific’; it reflects rather poorly on your understanding of the technical issues.
The super-sized grey error bands are code for “We don’t really know.” This language was noticeably absent from the Summary report and instead they used “We are more certain than ever… ”
Hmmmm
SteveF
I think you could have added that “BTW”
appears to be an attempt at a threadjack explanations for the pause or deviation are not relevant to the question of how to compare raw untortured observations of GMST to raw, untortured projections form the AR4.
Whether WebHubTelescope likes it or not, the AR4 did not create projectinos of “weather, solar and volcanic noise corrected temperature anomalies” and the observational groups do not publish “weather, solar and volcanic noise corrected temperature anomalies”. So when authors are comparing observations to to AR4 projections they should be comparing those projections to observations of the things the AR4 authors projected: i.e. uncorrected GMST. That “thing” is the observations of uncorrected GMST.
If WebHubTelescope has an explanation for why deviations occur: Dandy. But that is not relevant to the issue of this post which is: Is the level of agreement or disagreement between the models and observations correctly displayed in this particular figure.
I know WebHub is probably eager to share his theories– as many others are. And so he wants to share them in comments here. But it is a thread jack. 🙂
Lucia,
Sorry, I didn’t mean to encourage a thread-jack. It is just that I grow as tired of WHT’s unsubstantiated claims of continued rapid warming as I do of the Skydragon slayer claims of GHG warming violating thermodynamics… nonsensical motivated reasoning one and all. WHT makes the same rubbish claims on every thread where he comments; since The Oil Drum blog closed, I guess he needs an outlet for all his pent up enviro-passion.
Dave: “don’t those error bars mean that the difference between ~0.1 degrees of warming and ~1.4 degrees, by 2015, will not be reliably measurable?”
The difference between ~0.1 degrees of warming and ~1.4 degrees, is easily measurable. In the first case, we’ll hear “not inconsistent with models”, in the second case, we’ll hear “the IPCC was super-conservative in its forecast.”
SteveF–
Not your fault.
People will often respond when someone else introduces a new topic. I generally don’t mind so much. But I don’t quite know why WebHub decided to introduce the issue of “WebHubs explanation of why the disagreement should be disregarded” into a post on “Does the figure accurately show the level of disagreement”. But I was reading the exchange and realized: WebHub did a threadjack. (He may not have done it on purpose. But it’s a threadjack nonetheless!)
Of course different people have different explanations for the disagreement. But even if their explanations are correct that would not grant others permission to make graphs that do not fairly represent the relationship between projections and observations. In this post really is about whether the projections: which are of uncorrected GMST are fairly compared to observations of GMST.
Soom Mosh will arrive and point out the difficult with the comparison: Failure to use consistent mask. However, this is difficult to do on a single graph since the masks for GISTemp, NOAA and HadCrut are different. But that sort of comment would at least be about the graph and of course, Mosh would be right. (He must be busy because he often points out the masking issue.)
Lucia,
I did read SM’s account of the same graph but came away sorta scratching my head over the post – your explanation is a little more clear (to me at least – but I’m always at least the 9th one to admit I’m slow on the uptake)
Concerning the shifting the observational baseline – you say
” At least when they changed the baseline used by the AR4 models for projections, they also changed the baseline for the observations. That is: they made an “apples to apples†comparisons– just using a different baseline from that in the AR4.”
I don’t see where/how the observational baseline was changed or did I miss something?
Also I wish I had a filter at Climate Etc for WHT and – to be fair – all the other one trick ponies over there so finding relevant comments wouldn’t be so tedious.
ChrisinGa–
I have not yet engaged the issue of the new version of the graph. I’ll be creating figures baselined in 1961-1990 vs. ones based on 1980-1999 and show what happens. I’ll also explain what I think is “right” based on the principles of:
“Use the baseline the AR4 authors chose when comparing observations to the AR4 projections”
Squinting it appears to me the AR5 authors did not use the AR4 baseline– and they could have done so easily.
However, because there are arguments over the various figures, it’s necessary to engage all the different ‘mysteries’. Obviously, the final figure is the most important. But it’s worth discussing whether the final figure was “an improvement” over the earlier ones, or if different mistakes were made. And whether or not SkS will accuse people of being conspiracy theories, it is nevertheless interesting to see the direction of ‘errors’.
For example: You’ll see in the next post that with regard to commenting on this SOD-AR5 figures, both SkS and Tamino committed one humongonourmous boner and — get this– it just so happens the boner of an error is in the direction that we all know SkS and Tamino “like”. I would suggest that neither would have made the same error if IPCC figure 1.4 had happened to use 1993 as the year to “pin” models and observations. 🙂
Lucia,
I think you have missed the point in having ‘drafts’.
If you want to point out errors in draft versions, then you should make it clear you are discussing a draft version – something you make unclear above. You keep calling it the figure from AR5. It’s not a figure in AR5.
It’s also confusing as you talk about the text below the figure in AR5, is that the draft text associated with the draft figure? Or is it the final draft text associated with the final draft figure.
As to being ‘childish’. Meh. Knock yourself out.
“However, because there are arguments over the various figures, it’s necessary to engage all the different ‘mysteries’. ”
This is not necessarily so. Remember this is from a draft version. There are many reasons a figure may not appropriately match descriptions in text or previous versions. The most common is the author puts the wrong figure in by mistake. This particular figure may have been drafted for some other purpose and mistakenly added. It happens frequently, especially when the figure looks similar to the one you put in.
“For example: You’ll see in the next post that with regard to commenting on this SOD-AR5 figures, both SkS and Tamino committed one humongonourmous boner and — get this– it just so happens the boner of an error is in the direction that we all know SkS and Tamino “likeâ€. I would suggest that neither would have made the same error if IPCC figure 1.4 had happened to use 1993 as the year to “pin†models and observations. :)”
Oh god is this part of your ongoing war with Tamino and SKS?
Ok fine, you’re right… Not interested.
Lucia,
Sorry for feeding the threadjack. My bad.
Paul_K,
Not a problem. I only realized it when I almost got sucked myself. Then I thought… “whoa…. I actually do want to discuss the figure on this post.” The issue of choices in displaying agreement with models and data is important in itself. This is true even if ‘explanations’ of the disagreement are sort of more alluring (to everyone!)
You’re not the only one querying the gray bands.
If you aggregate different studies, you reduce the error bounds. It’s a meta study, or a consensus.
Why should things get more uncertain when you create a consensus result?
Actually, its quite easy to understand why this has been done. Plot the consensus with the consensus uncertainty, and compare to the actual, and the game is up. All the predictions have failed.