Answering comments yesterday, I posted a graphical representation of IPCC projections for the 21st century. This is taken from the AR4:
It occurred to me that some readers might notice that, in the image, the projected line matches the hind-cast temperature at 2000. So, they might be interested in seeing how data might compare to projections if I created a diagram that looked more like this:
In the image above, the trend lines FAR, SAR and TAR projections are forced to pass through the decadal average temperature for 1990. (As far as I can determine, that is the centered average around 1990; I determined this by making similar plots and finding the centered average results in plots similar to the one above.)
So, having done that, I created two plots for the AR4 projections. There are:
- A plot of the average for NOAA, GISS Land/Ocean, HadCrut data merge, corrected using the ENSO correction Gavin described in his real climate post and
- A plot of the average of the NOAA, GISS Land/Ocean, HadCrut, UAH RSS merge with no ENSO correction.
In both plots I calculated the 11 year centered average around 2000, (i.e. based on Jan 1995-Dec 2005 inclusive), and called that the ‘average’ for 2000. I defined this as this temperature as the baseline for the anomaly. Then, using data starting from Jan 2001, I determined Ordinary Least Squares (OLS) and Cochrane-Orcutt (CO) trends forcing the trend through the origin. (So, this analysis differs from yesterday’s and previous ones primarily by forcing the trend through the origin.)
Results using land only data and correcting for ENSO
Figure 1: Solid yellow and red lines indicate the best fit trends. The purple line indicates the 2C/century central tendency projected by the IPCC. The dashed yellow and red lines indicate the ±95% confidence intervals using these methods.
Readers might note that the solid lines indicating best fit trends are drawn with thinner, lighter lines. This is because the important issue here is not the magnitude of the trend. The important issue is whether or not the purple line representing the central tendency of the IPCC projections lies inside the ±95% confidence intervals.
When I use the land only measurements and correct for ENSO, the IPCC projections fall outside the ±95% confidence intervals. So, based on this particular analysis, the IPCC projections are found inconsistent with the data.
What if I don’t correct for ENSO?
However, as I have often told readers, I prefer not to correct for solar cycles, ENSO etc. The reason is I have little confidence in the corrections, which might introduce more uncertainty. I also prefer to draw from all five data sources. So, using all five data sources the result looks like this:
In this figure, we see that based on OLS, the IPCC trend falls just inside the 95% confidence intervals. So, using classical statistics, that trend is not inconsistent with the data. If “IPCC trends” are assumed true in the first place, we continue to assume they are true. However, using CO, the IPCC 2C/century trend falls outside the 95% confidence intervals. So, using Cochrane-Orcutt says the IPCC 2 C/Century projection is false.
Note however, the IPCC does include some uncertainty when projecting 2C/century. The uncertainty intervals are not communicated clearly, but could, in principle be picked off the fuzzy graph. It does appear that lower range of the IPCC uncertainty interval may fall inside the uncertainty intervals for all four tests.
Do I read the second chart correctly? The first IPCC report predicted too high, but the actual temp was just within the low range of what was predicted? The third predicted too low? Or is the light blue area below the actual temperature the second? My poor mildly defective colour eyesight is getting lost….
Michael–
The first IPCC predicted 3C/century. It was much too high. However, when one compares climate projections to weather (as the IPCC seems to prefer) the weather scatter is all over the place.
For some reason, the IPCC doesn’t like to fit trend to data and find the uncertainty in the trend.
I don’t have a copy of the FAR– so all I have is that graph from the AR4. In that graph, even though the FAR trend of 3.0 C/century is way to high, some “weather” falls inside the uncertainty bands.
Lucia,
I think your 11-year average centred on 2000 begins in January 1995 (not January 1985)?
I hope you’ll be able to relate this new analysis to the study by two ANU experts which you’ve already discussed briefly on another thread. That study was commissioned by the Garnaut Review and its key conclusion (which, as you noted, is not inconsistent with anything that you’ve said) was reported on p. 1 of the Review’s 550-page Draft Report.
Well, then expect your results to be met with huge caveats and to be, in effect, meaningless. You might as well say, “I like numbers and I’m not going to consider the physical nature of the climate system at all.” What did WMC say? Statistics with no physical understanding?
Boris, first you detail the observational trends and provide solid baselines. Then you leave it to speculators like yourself to come up with the excuses for why the temperature changes the way it does. Without something such as she is doing in showing what it is, all the speculation about why it is as meaningful as conjectures about how many angels can sit atop a pin.
Boris–
I didn’t say I’m ignoring the physics. I’m saying I prefer to use weather and climate data that is minimally processed, rather than “correct” it for things like ENSO using corrections that are far from universally agreed upon.
When performing hypothesis tests, the uncertainty intervals are designed to acommodate the uncertainty in the data. It’s fair to suggest the uncertainty bounds I use may be too small, or the methods flawed. In which case, it’s legitimate to suggest reasons.
But that’s different from insisting that one “correct” data according to some scheme that no one can agree on, and then claim the “correction” reduced the uncertainty. The corrections rarely reduce uncertainty as much as those suggesting the corrections imagine.
You know perfectly well, that I show what the effects of corrections might be. I just did in the post shown above. So, it isn’t as though I ignore them.
When we look at the data this way, and apply the ENSO correction Gavin discussed recently at Real Climate, we conclude the IPCC projections are falsified at p=95% based on both the Lee&Lund method of testing hypothesis and iterative Cochrane Orcutt. If we don’t correct, it’s not falsified.
The fact is, when we ENSO correct, the falsifications of 2C/century are getting more robust rather than less so. That is to say: Using the same data analysis techniques results in an ever increasing fraction of “falsification” results.
So, if you like these corrections, and think they increase confidence in a result you should be leaning toward seeing the IPCC projections are falsified more strongly that those who dispute the corretions! The contrary appears to be true. You seem to simply wish to insist the corrections are “good” because I’m not persuaded the greater number of IPCC falsifications we get after “correcting” is meaninful!
People are always permitted to doubt, insist on more data etc. But, in my opinion, “correcting” the for things like ENSO rarely resolves the problem.
Lucia – As always, a thought provoking post. One minor semantic quibble. Although the word “correction” is commonly used for things such as trying to remove the effects of ENSO from the temperature trend, I think the more proper term to use is “adjustment” and suggest that you use that term moving forward. This makes it clear that the data are being “adjusted” in an attempt to remove the effects of well-known phenomenom but does not imply that the adjusted data are anymore “correct” (i.e. closer to truth) than the unadjusted data. In my opinion, all of the various “corrections”, whether they be for site location moves, buckets vs. intakes, UHI, etc. should deemed “adjustments” to properly reflect the fact that there is some degree of uncertainty with the adjustments and that the record is less than perfect.
lucia, you say things so much better than I do. Except on carbon taxes, of course.;-)
You still use temp measurements that measure broad swaths of the atmosphere and pretend that they are the same as surface measurements. That’s a fairly glaring error, imo. So why should broad sections of the atmosphere be treated as surface measurements? Note this has zero to do with corrections.
Boris–
I’ve posted results averageing over all, and over each individually. The IPCC projections falsify either way.
That is to say: The falsification is robust to choice of measurement system. 🙂
Lucia, how do these charts look if you use land/ocean or satellite temperature data?
Keith– Do you mean looking at each individually? Hmmm… sort of similar. I think I remember that if we don’t adjust for ENSO, NOAA and GISS failed to falsify individually. But the others falsified. I’ll check, and when June’s all come in, I’ll use this method for a change. The data doesn’t change much, so I find it more interesting to look at the data a variety of ways.