Data Comparisons

Comparison of TAR to AR4 projections (using TAR Baseline.)

31 Comments

Are the TAR and AR “projections” really different? This question seems to be asked constantly at blogs, so I wanted to created graphs that I think permit a fair comparisons.

In a recent post I tried to compare the AR4 projections to the TAR projections using the “crayon” method. Boris asked a few good questions which I’ll try to answer. Thanks to JeanS who digitized the TAR projections and also sent some data that has been smoothed using 11 year embedded with minimum roughness criterion at the end and can make “non-crayon” graphs. (Hence forth, in this article, all “11- year smoothed” results correspond to this particular type of smoothing.)

For today’s posts I first :

  1. Used the model-mean temperature projections from models used in the AR4 to create pseudo-AR4 projections. (I say pseudo because I will be changing the baseline in a way not done in the IPCC AR4 by its authors. I made the baseline change to set the multi-model mean of the AR4 pseudo-projections equal to the TAR baseline and also create uncertainty intervals that share some of the properties of those in the TAR. Model runs are downloaded from “The Climate Explorer.” )
  2. Rebaselined each model mean temperature anomaly to force the 11- year smoothed temperature anomaly in Jan 1990 to zero.
  3. Computed the multi-model mean temperature anomaly based on the individual smoothed model means. (The multi-model mean temperature of the multi-model mean will be zero in Jan 1990.)
  4. Computed the standard deviation of the multi-model mean of the 11 year smoothed temperature anomalies. Because all 11-year smoothed means equal zero in Jan 1990, this will also be zero in 1990.
  5. Super-impose the 11-year smoothed multi-model mean and it’s ±1 SD range and underlying unsmoothed multi-model mean on the TAR projections as digitized by Jean S.

This is shown below (I have TAR simulations output beginning in 1990 and AR4 simulation output since 1900):

Figure 1: Projections from TAR and AR simulations baselined with 11 year smooth set to 0 in 1990.
Figure 1: Projections from TAR and AR simulations baselined with 11 year smooth set to 0 in 1990.

As you can see, using this baselining the 11-year smooth multi-model mean of the AR4 simulations match the 11-year smooth initial point anchoring the TAR projections. So, that makes them consistent.

This uncertainty intervals for 11-year smoothed temperatures show zero standard deviation in January 1990 similar to the zero uncertainty intervals shown on the TAR. This makes sense if we interpret these particular uncertainty intervals for the AR4 to indicate the possible range for 11-year smooth temperature anomalies forced through zero in January 1990.

Comparison shows that the 11-year smooth multi-model mean of the AR4 simulations (dark blue) is roughly equal to the upper range of the TAR “projection/hindcast” (grey) from 1990-2004. The TAR was published in 2001, the AR4 was published in 2007.

It is rather evident that the 11- year smoothed AR4 simulations correspond to greater warming than indicated by the TAR during the period since 1990.

(For Boris: You will note that my AR projections are different from the previous ones. In the previous ones, I used “volcano only” screening, but did not say so. This time I’m not screening for volcanoes because the TAR doesn’t seem to indicate those. In the previous post also made a mistake crayoning in the 1995 temperature anomaly too low (contributing to part, but not all, of the excess “jump” you saw between 1995 and 2000 and didn’t thoroughly think through the rebaselining. Rebaselining to match the 11-year smooth baselines consistently increases the spread between the TAR and the AR4 compared to not doing so.)

Can we see that with data?

Below, I added the 11-year smooth Hadley and GISS surface temperatures and the annual average temperature anomalies, all baselined force the 11-year smooth through zero in January 1990:

Figure 2: Projections with observations superimposed.
Figure 2: Projections with observations superimposed.

Inspection shows the 11-year smoothed temperature anomalies just touch the lower 1SD range for 11 year smooth anomalies from multi-model means in the AR4. So, temperature are running below the best estimate from the AR4, though the degree to which this metric is low is not unexpected.

We should expect to see the 11-year smoothed “data” for years later than 2003 to move when next years data arrive. This will happen because the 11-year smoothed data close to the end points will be computed differentlywhen future data are available. Because of this, in some sense, when using the sort of 11-year smooth data shown here, the true (i.e. stable, robust) comparison is restricted to years up to 2003, that is 2 years after publication of the TAR. This could be moved a bit earlier by examining when the methodology for the TAR was “frozen”, but logic dictates the earliest possible date for “freezing” the methodology for the TAR is publication of the SAR which is officially dated 1995, but formally released in June of 1996. In my opinion, the date when models for the TAR were frozen, is “sometime in 2000”; I discuss my reasoning here.)

Some of you will certainty notice the annual average data lie outside those uncertainty intervals shown for 11-year smoothed anomalies for the AR4; however, that shouldn’t be interpreted to mean much because the uncertainty intervals for annual average temperature anomalies are much larger than those for 11-year smoothed temperatures which have reduced amounts of “weather noise” relative to annual average temperatures.

Wrap up

  1. After running near the upper range of the TAR during the “early hindcast” portion of the TAR “projections”, the 11-year smoothed data is now running more or less in the center of the projections particularly in the “robust” comparison region.
  2. If rebaselined to force the 11-year smoothed temperature anomalies to zero in 1990, the end point of the smoothed observations are now near the lower range of the AR4 projections. However, even if these results hold when next year’s data trickles in, the 11-year smoothed observations would not be sufficiently low to decree the AR4 projections off on this basis. It is accurate to say the 11-year smoothed observations computed with minimum roughness at the ends are currently low relative to the models. (Readers should recall that comparison of smoothed anomalies has lower statistical power than t-tests on trends. You also need to wait a long time before you get your first data point outside the “hindcast” region or a “hindcast/projection”. )
  3. No one should get too excited about the location of the end points either now nor back in 2007 when Rahmstorf et al got their knickers in a twist and decreed temperatures were running warmer than expected because the 11 year smooth trend was approaching the upper bound of the TAR range. The end points will move.

I hope that answers any questions for Boris, and lets people see how the data appear when run through an 11-year smooth filter with minimum roughness endpoints.

Update

: I inserted a few sentences liking back to a previous blog post When were the models used in the TAR frozen”

31 thoughts on “Comparison of TAR to AR4 projections (using TAR Baseline.)”

  2. Thanks bob. Oddly, that paragraph originally did have an ending…. (I must have cut and pasted and then not noticed.

  4. Scooter– I think the SAR predicted more warming than the TAR in terms of trend. But… I don’t happen to know the baseline.

  5. I think it’s the other way around.

    SAR was less than TAR (with SAR at about 0.15 deg/decade for 1990 on, IIRC) and FAR was considerably more than TAR.

  7. Deep–
    Yes. I commented on Gavin’s double standard vis-a-vis error bars. Other people need to use them; Gavin doesn’t need to when he doesn’t want to.

    I still don’t know if the TAR or the SAR projected higher. Did anyone ever find a citation that provided the specific projection for temperature rise from 1990-2010 in the SAR? The TAR has a discussion of some things the SAR would have projected if their SRES scenarios has been used in the Raper upwelling diffusion model used in the TAR rather than the mim-cam model used in the SAR. But I haven’t found the SAR on line. What did they actually say?

    Roger linked to a summary which only describes the rise over the entire century and says nothing about whether the trend is linear, concave upwared, concave downward etc.

    Do you have a reference with information? If yes I can add that to the graph.

  9. Figure 1.1 in the AR4 shows the FAR/SAR/TAR projections. The FAR was the high one. The SAR was the lowest and the TAR was in between. So, yes, Deep is right, the SAR had slower warming than the TAR. I’d still like to find text of the SAR and FAR if possible.

  10. Roger’s FAR and SAR projections were for 1990-2007, and so was the graph he referenced (AR4 TS-26) .

    http://sciencepolicy.colorado.edu/prometheus/archives/IPCC%20Check.png

    I don’t think he interpreted it correctly though (which was Gavin’s point). I get about 0.28 or 0.29 for FAR and 0.14 or 0.15 for SAR, from the mid-lines. I don’t know what Roger’s dots are, as they appear to be on the upper bounds which makes no sense to me.

    The original graph is on p. 69 of the AR4 TS, in section 5.1 “Understanding Near-Term Climate Change”.

    “Projections for 1990 to 2005 carried out for the FAR and the SAR suggested global mean temperature increases of about 0.3°C
    and 0.15°C per decade, respectively.” [TS, p.68]

    So there you have it …

  11. This is the bit I wanted:

    “Projections for 1990 to 2005 carried out for the FAR and the SAR suggested global mean temperature increases of about 0.3°C
    and 0.15°C per decade, respectively.” [TS, p.68]

    So, the FAR was well over the observations; the SAR wasn’t too far off. Thanks.

    I still wish I could information on the sort of curvature (as Rahmstorf captured in his 2007 paper. ) But I think this is all there is.

  12. Nice post.

    Does this mean that the next time someone says “it is worse than we thought”, they are speaking in direct contradiction to the IPCC findings, which show that earlier, less dramatic projections (TAR) have proven over time to be more robust than the later (AR4) more alarmist ones?

  13. Geckko–
    If they means surface temperatures, those have not been warming faster than predicted in the two most recent reports. Maybe the rate exceeds the Second report, but certainly not the first!

  15. The point is this: the decadal rate for FAR “prediction” shown in Roger’s graphs is 0.34 or 0.35 deg per decade (admittedly lower than I thought when I first glanced at it).

    But Gavin’s point (and mine) was and is that the slope was wrong – it should have been no higher than 0.3 deg C per decade, and probably slightly lower (looking at graph fig TS.26). As I said the correct value appears to be 0.28 or 0.29.

    Am I missing something? I’ll be happy to retract if someone can point out my error.

    http://sciencepolicy.colorado.edu/prometheus/archives/IPCC%201990%20verification.png

  16. Deep–
    Long before I joined that thread in that post, Roger seemed to explain what he did to Gavin:

    I have not “corrected” it because I have a different interpretation than you do on the 1992 scenarios. We have a difference of opinion. I won’t be so arrogant to suggest that you “correct” your thinking because reasonable people can disagree on such things.

    As I pointed out to you, Figure A3.1 in that same report shows very large differences across scenarios prior to 2000, so I am not at all convinced that choice of scenario is irrelevant.

    But even though I see things differently, perhaps you are right, and instead of 0.6 in 2007, it is something like 0.45. But as you imply, your point doesn’t alter the overall conclusions of the analysis — the IPCC dramatically reduced its predictions in 1995 (which it acknowledges explicitly). (And a point I do not raise, the 1992 FAR Supplement reduced the original predictions from 1990.)

    So, a) he and Gavin seem to be discussing different scenarios and b) the choice doesn’t make a difference to the point Roger is making.

    I would probably have picked the dark green line in the center of the FAR scenarios, but Roger explained which he picked in that comment thread and illustrated how he got the number. He pulled it off the top line, not the center line. He mentions other numbers could be picked in the comments– and those correspond to the ones you are mentioning.

  17. The graph was and is wrong.

    Roger says “You’ll see that the values square with what was said in comments 15265 and 15266.”

    The IPCC statement I cited in 15265 was:
    “Projections for 1990 to 2005 carried out for the FAR and the SAR suggested global mean temperature increases of about 0.3°C
    and 0.15°C per decade, respectively.”

    From an examination of TS.26 I cited a value of 0.28 or 0.29 deg per decade. Or as Gavin put it, 0.45 in 2007, not 0.6.

    The graph shows 0.34-0.35 for FAR so it does not “square with” my comment.

  18. Deep–

    Look at the dots and letters. That explains where he got it. He said so in comments.

    This squares with the part of your comment you did not re-cite:

    I don’t know what Roger’s dots are, as they appear to be on the upper bounds which makes no sense to me.

    The original graph is on p. 69 of the AR4 TS, in section 5.1 “Understanding Near-Term Climate Change”.

  19. Roger said the “values square” with my comment, so I quoted the values I gave, which do not match those in the graph. I feel Gavin’s pain. I give up.

  20. Deep,
    In that comment, you said Roger said he got the values from a particular figure, and supplied the figure he said he got the figures from. Roger showed that, he did indeed get the values he used from the figure– confirming that he did what he said he did.

    I think that’s what Roger means when he says what he did squares with what you report he said he did. He used the values associated with dots on the graph.

    Meanwhile, the specfic value you quote are middle line– which Roger does not indicate with dots.

    What’s confusing about Roger’s use of “square with”?

    So, as I perviously said

    I would probably have picked the dark green line in the center of the FAR scenarios, but Roger explained which he picked in that comment thread and illustrated how he got the number. He pulled it off the top line, not the center line. He mentions other numbers could be picked in the comments– and those correspond to the ones you are mentioning.

    So, in short: Unless there is other information in the FAR itself, I would probably illustrate 1990 projections using the middle green line, you would have used the middle green line. Gavin would have used the middle green line. Roger picked the upper green line, said that’s what he did, and described that as being what he did and explained why he chose that.

  22. Roger picked the top end of the range in the figure. Do you have any additional information directly from the FAR? I don’t. That’s what I would use to decide whether a choice from that graph is categorically wrong.

    I already said that, if the only thing I had was that graph, I picked numbers off that graph, I’d pick the centerline and I’d explain that as my basis.

  23. The choice of “line” in FAR is a function of climate sensitivity.

  24. Thanks, Lucia. This was interesting. I’ll have to process it more when I’ve had more sleep.

    As for the FAR, it didn’t include sulphate aerosols, right?

  25. Boris–
    As I told Deep Climate, I don’t have a copy of the TAR, so I don’t know what the TAR accounted for.

  26. Lucia,

    I think you mean FAR, not TAR. In general, climate science needs to be a bit less fond of acronyms. I once had a friend burst out laughing when I started discussing the IPCC AR4 WGI SPM.

  27. Zeke–
    Yes, I mean the TAR. All fields have acronyms. The thing with climate science is they end up busting out into general use.

    I still remember back at PNNL, one of my coworkers was working on a project that has a euphemistic acronym. Engineers and scientists had been referring to something as an “eruption”. The powers that be had gotten together with the editors, and dubbed it the “Glass Displacement Event” (GDE).

    During a summer talk a summer student kept calling it the “GDE”. At the end of he talk, someone asked him what “GDE” stood for. The nervous student paused. Then someone in the back of the room said “The GOD DAMN ERUPTION!”

    The “event” was soon renamed. . .

    (BTW the event was this: A proposed process involved inserting electrical conductive rods into soil, heating the soil until it became molten, removing the rods and then letting the molten soil solidify. During one event a gas bubble formed, and violently spewed forth setting the protective tent on fire. Some carefully trained editors tried to convince people the word “eruption” did not apply to this because that only applied to volcanoes; the lined out every instance of “eruption” used in reports Showing examples like “pimple erupts” etc. left them unmoved. Glass Displacement Event, or the shorter “event” were dubbed “correct”.)

  28. Boris–
    By the way, I could rebaseline everything to the AR4 baseline too. The notion that the “real” observed temperature is the 11 year smooth value was not part of the TAR but imposed by Rahmstorf after data were measured. So, using the AR4 baseline shows the sensitivity to this choice.

  29. The choice of “line” in FAR is a function of climate sensitivity.

    Really? Why are the FAR predictions so much higher than the SAR and onward when CS estimates have been essentially the same since the Charney report?

  30. Boris–
    I suspect what Roger means is each ‘line’ in the FAR corresponds to a particular choice of climate sensitivity. This would make sense if they use a tuned upwelling diffusion model (like MAGIC), which is what was used in the SAR and the TAR. It would also make sense if they used something even simpler. The “simpler” models used to create projections prior to the AR4 have no “weather noise” and were tuned. (FWIW: The magnitude of the tuning parameters for the simple model used in the TAR were cited as coming from Raper et al (2001). The tuning parameter values come from results of AOGCMs existing at the time. However, some judgment was used as the full range possible from the all AOGCMs available at the time were not used. )

    In contrast to the sorts of results from “simpler models”, the spread around the multi-model mean in the AR4 represents the difference in means from individual AOGCMs and some internal variability that was not averaged out because there aren’t enough runs.

  31. Pingback: The Blackboard » Source of fishy odor confirmed: Rahmstorf did change smoothing.

Comments are closed.