Since he is a chemistry professor and has spent far more time on this than my back-of–the envelope calculations, so I am willing to defer to his expertise on this subject. Should he wish to emerge from anonymity, his substantial efforts are likely worth a peer-reviewed publication.

His more accurate conclusions on the data are quite similar to mine (though I anticipate he may find some differences on N2O) and those published by Hansen in 1998, namely (a) Hansen’s 1988 scenarios generally overshot actual emissions, and (b) some errors in projecting individual emissions rates may have cancelled out, thereby making the total forcing projected more accurate.

I fully appreciate that different people can interpret the significance of these data differently (what constitutes a predictive error?), and such different interpretations need not reflect malice, fraud, or ill-intent, but simply the fact that people can agree to disagree. Again to be clear, my post was not about Pat Michaels and should not be construed as a defense of him, Michaels’s decision to focus on Scenario A in Congressional testimony was a mistake. At the same time, it is also appropriate to conclude that Hansen’s claim that the real world has evolved most like Scenario B seems to be an overstretch.

On the various gases “Eli Rabett” concludes:

“It looks increasingly like Scenerio C from Hansen et al, is a much the best match (except for CO2 after 2000) to the actual greenhouse gas mixing ratios up till now. If you look at the prediction, there is not much difference between those for Scenerios B and C.”

You can see his thoughtful analysis here (with particularly useful information on CH4, he promises more info on N2O): http://rabett.blogspot.com/

He is welcome to resume commenting here under his own name and if he focused on substance and not insults or threats.

Thanks.

FYI, If you want your comments to appear you’ll have to register with TypeKey. Sorry about that, we’re working on alternative. Otherwise they will only appear if I stumble on them in the sea of spam as I did in this case.

Thanks.

It would be interesting to calculate a “skill score” for each scenario, which if I ever have time might be worth doing.

Nonetheless, CO2 emissions are only part of the story here. The overprediction of CH4 and CFC emissions are dramatic. Adding these to the mix makes it difficult to claim, as Hansen’s did, that the real world has evolved like B88 (which is of course why he revised them in 1998! Had they be on target he would have not changed them.)

Thanks again for the exchange.

I neglected to say that these are the averages for the periods 1989-2000(inclusive)
B88 .224
C88 .178

and 1988-1999(inclusive)
B88 .128
C88 .108

They are NOT the averages for the full periods ending 2005. The above averages were intended specifically to compare B88 and C88 over the shorter peiod that Roger had calculated things for.

You can see this by a previous comment above, but I should have made it clear below the data as well because it looked like it was an average over all the years.

Here’s the relevant cooment repeated:
“By the way, I calculated the average deviation as a fraction of the total for the first 12 years for each of the scenarios. I did it for two intervals after discovering the above “issue” with you analyis. I found that the result actually depended on starting point of the actual interval. For 1989-2000(inclusive), the average fractional deviation of B88 from ACTUAL was 0.224. For C88, it was 0.178. In that regard scenarios are with 4% of one another. For 1988-1999(inclusive), average for B88 was 0.128, (ie, within 13% of actual on average), for C88, it was 0.108. Again within 4%. The fact that the results are different for the two intervals again points to the problem with using an interval that is shorter than the maximum possible.”

My comment about a comaprison you made should read thus:

“I DO have an argument with using 12 years for the ACTUAL CO2 growth and 13 for the B88 scenario and then trying to compare the results!”

In other words, you used different time spans and then compared them (12 years for the ACTUAL and 13 for the scenario calculation)

While I agree that the C88 scenario gives results closer to the actual for the first 12 years (I was never even arguing that point with you), the key issue is how the scenario performs over the longest haul.

In fact, given all inherent variability in natural processes (essentially randomness) even with perfect foreknowledge of future emissions and perfect modeling of all the processes, I would not expect ANY scenario to play out precisely, especially not over the first few years when a deviation in the yearly increment of just 1 or 2ppm can throw the actual path away from the predicted significantly. With a good scenario, this difference will “made up” over time.

The key question is the following: does the scenario converge (and stays converged on) the ACTUAL emissions development? Or does it only diverege or perhaps first converge and then diverge? (as C88 seems to do if you run it out to 2005).

It appears that you and I have a basic (probably unresolvable) disagreement about how to evaluate scenarios, but I would argue that letting the secnario run for the longest time period is the most reliable gage of its performance (and hence validity). This is a direct result of the cumulative nature of emissions. Every subsequent data point has all previous data points rolled up in it (in effect).

Finally, just let me note that I never intended to argue that Hansen’s scenario B88 got it exactly right.

In fact, I’d have to wait another 5 or ten years, during which time it too may start to diverge significantly from the actual (as a fraction of the total to date). Even 16 years is a relatively short time period for judging these things.

Specific coments with regard to your last post (iimediately above)
With regard to your characterization of the deviation between B88 and C88: I would merely note that the differences between B88 and C88 and between B88 and the ACTUAL accumulated CO2 over that period are NOT as large as they might seem based on what you provided.

1) The total you gave for actual CO2 was 18.05 (as you indicated for the years 1989-2000 inclusive). I have no argument with that.
http://www.cmdl.noaa.gov/ccgg/trends/

But the 20.81 you gave for B88 accumulated was for 13 (!) years (between 1988-2000(inclusive)).

I DO have an argument with using 13 years for the ACTUAL CO2 growth and 13 for the B88 scenario and then trying to compare the results!

It’s easy to see that the latter is for 13 years just by dividing 20.81 by 12 or a value of 1.73ppm, an AVERAGE value for the interval that is greater than the value you get (based on Hansen’s assumptions) at the the very end! I don’t think so.

You must use the same number of years (and time interval) for PREDICTED and ACTUAL if you are going to compare them, of course.

If you subtract 1.5, I come up with the CORRECT value of 19.31 for the total B88 growth in CO2 (over the period 1989-2000 inclusive). That’s for the same years over which ACTUAL CO2 growth was 18.05.

This means the deviation of B88 from ACTUAL is NOT 2.76ppm as you said, but instead 19.31-18.05= 1.26ppm (which amounts to 7% of the total CO2 (18.05ppm) growth for the period.

Is 7% a “drastic” over-prediction” as you implied in your original post??

I’ll let others decide that.

By the way, I calculated the average deviation as a fraction of the total for the first 12 years for each of the scenarios. I did it for two intervals after discovering the above “issue” with you analyis. I found that the result actually depended on starting point of the actual interval. For 1989-2000(inclusive), the average fractional deviation of B88 from ACTUAL was 0.224. For C88, it was 0.178. In that regard scenarios are with 4% of one another. For 1988-1999(inclusive), average for B88 was 0.128, (ie, within 13% of actual on average), for C88, it was 0.108. Again within 4%. The fact that the results are different for the two intervals again points to the problem with using an interval that is shorter than the maximum possible.

The most valid results from a scientific (ie, statistical) standpoint are obtained with the interval from 1988-2005(inclusive).

Roger, if you intended the interval to be 1988-2000(inclusive), you will have to add an additional 2.11 (for the actual CO2 growth in 1988) to that total to give 20.16ppm.

If you do that, however, your C88 value is no longer “dead on”. It’s off by .66 and the average is no longer 1.504 but 1.55. My point is not to say this makes a big difference (C88 is still within 3% of 20.16), but merely to point out that the fact that C88 gave precisely the right result over the period was basically a fluke of the beginning and ending points. Early on, the actual time interval makes its biggest difference. In fact, if one stopped at 1998, B88 results would be within 4% of the actual (and C88 would give about 2%.)

As one goes further out in time, the actual interval makes less and less difference – and it is best to use the ENTIRE interval 1988-2005 at any rate. Because of the cumulative nature of the emissions, going out further in time becomes a more accurate gage of the validity of a scenario.

If a scenario is very close in the first few years, that may actually NOT mean much. What is MOST important is whether it remains close as more years go by. In this regard – OVER THE LONGER PERIOD (AT THE END OF 2005) (which is the most important measure of a scenario’s performance), B88 IS performing better than C88 as I indicated (and you agreed, I think). It is very unlikely that a scenario would just HAPPEN to be right (by pure luck) after 15 or 16 years. In fact, that becomes less and less likely as time progresses. In effect, each time data point for accumulate CO2 to that point is comprised of all the points before it, so, as time moves on, data points become more and more heavily weighted with regard to judging a scenarios performance.

THIS IS THE ESSENTIAL GRIEVANCE I HAD WITH YOUR ORIGINAL (“ right answer for the wrong reason”) POST, Roger.

Though C88 is a little better than B88 in the early years, the fact is, B88 gets closer and to the ACTUAL as time progresses (it is within 5% of the total accumulated CO2 in 2005) which is precisely the behavior that one would expect from a scenario that was doing what it is supposed to do. One other hand, C88 starts to diverge (andin the later years, a rather large fractional divergence is even more significant becasue of all the data points averaged in), so that it deviates by about 11% in 2005, having reached its minimum deviation around 2000, which one would expect from a scenario that is NOT the most accurate.

These scenarios were meant to run over a LONG time period, not just a few short years.

As I indicate above, to gage the deviations from the actual at any given point in time, it is best to use a “fractional deviation”, since a 1ppm difference after 10 years means much less than a 1ppm difference after 1 year because of the cumulative nature of the beast. Actually, part of the reason that BOTH b88 and C88 deviate from the actual is that the ACTUAL values for the first 5 years deviated significantly from the 1.5 that was assumed to be the best starting value. So, right out of the gate, both scenarios were down by almost 2ppm and had to play catch up to make up the difference (which they did over time, as one would expect).

Because of the cumulative nature (and a certain amount of “random errors”, it is a little like flipping a fair coin.

Early on, you might get a “run” of heads (or tails) that makes the fractional deviation from 1/2 (heads vs tails) greater than you expect. But if the assumption about their being equally likley is corrrect, the fractional deviation will tend toward zero as time increases, even though the ACTUAL deviation (difference between number of heads and number of tails) does not.

The following table shows “Fractional deviation as a function of year” (deviation of the scenario prediction given as fraction of a total accumulated CO2 to that point)

I used two different intervals (1988-2005inclusive and 1989-2005, inclusive) – labeled (a) and (b) because of the issue I found above with Roger’s analysis and to illustrate how the starting year can affect the outcome (particularly early on)

In order to compare a scenario developing with constant yearly increase (like C88) to that of B88 in 2000 and beyond, I made an artificial assumption.

For scenario C, I assumed it continued to increase in 2000 and beyond (by 1.5ppm per year), — while Hansen set the increase to zero at the BEGINNING of 2000.

Look at Hansen’s graph of the emissions here:

http://scienceblogs.com/deltoid/2006/06/hansen_et_al_global_climate_ch.php#more

The CO2 increment CEASES on Jan 1, 2000!

In other words, there should be NO increase in the year 2000 (Roger!) and beyond for the C88 scenario, (though in the above analysis, I let Roger’s assumption that there was an increase for that year slide). Also, the interpretation that one should merely “stop considering C88 beyond 1999” is not the way Hansen intended it to be interpreted. He clearly showed data points and wished to illustrate how the model played out with no increase (ie, how natural mechanisms affected the CO2 that was already within th atmosphere). I’ll let the reader look at the paper and decide what they think he meant:

I will let the reader draw their OWN conclusions about which one is better over the longer haul but I would point out that the most important gage of a scenario’s accuracy and usefulness is precisely how closely it performs over the long run (as opposed to the short run).

1988-2005inclusive
Year B88(a) C88(a)

88 0.29 0.29
89 0.11 0.12
90 0.03 0.04
91 .08 .05
92 .25 .22
93 .25 .21
94 .17 .12
95 .10 .06
96 .13 .08
97 .09 .03
98 .0006 .06
99 .04 .02
00 .03 .03
01 .04 .03
02 .0025 .07
03 .02 .10
04 .02 .09
05 .04 .12

averageof fractional deviations:
B88 .128
C88 .108

1989-2005inclusive
Year B88(b) C88(b)
89 0.19 0.17
90 0.18 0.16
91 .29 .26
92 .54 .49
93 .45 .40
94 .29 .24
95 .19 .14
96 .21 .15
97 .15 .09
98 .04 .02
99 .09 .014
00 .07 .003
01 .07 .008
02 .02 .05
03 .007 .08
04 .0001 .08
05 .03 .11
averageof fractional deviations:
B88 .224
C88 .178

Finally a free few moments … (and my toddler is on the upswing, thanks for the well wishes;-)

Total actual ppm growth in CO2 from 1989-2000 was 18.05, or an average of 1.504 per year.

http://www.cmdl.noaa.gov/ccgg/trends/

Hansen’s Scenario C88 fixed per year growth at 1.5 ppm. This is exactly what happened over this time period.

Your analysis starts with 1990 as the base year. I accept your point that linear vs. exponential growth does not matter on this time scale. However your analysis should properly start in 1988 as Hansen’s analysis actually did using the 1.5 ppm as the base growth rate.

So for 1989-2000, starting in 1988 with the annual CO2 increment at 1.5 ppm (growth of 1.5% in 1989 and 1% in 1990-2000) results in a total growth of 20.81 ppm, or a difference of 2.76 ppm from the actual amount of 18.05.

Thus my conclusion about the accuracy of C88 over B88 stands. Through 2000 Scenario C88 was more accurate than B88, whether measured by average growth rate or total PPM (they had better lead to the same conclusion!).

I have already stated that I accept your point that after 2000 B88 is more accurate than C88. We disagree however on how to interpret this, since C88 stopped the annual increase in CO2. I understand and respect that your perspective on this is different than mine. No worries.

I also will accept that different people can evaluate how significant the differences in the scenarios in different ways. Is a difference of 2.8 ppm between the scenarios over 12 years a big deal? This answer, in part, probably depends upon how one evaluates Hansen’s claims in 1988 that A88 was “business as usual” and B88 was “most likely”.

Bottom line to all this — when you factor in that Hansen himself has accepted that the total radiative forcing from the various gases (not just CO2) undershot his lowest scenario, I think it is quite fair for me to make the exact same point.

I hope that this clarifies my perspective, and I am willing to accept that in interpreting this we may agree to disagree. I appreciate the exchange and the chance to clarify my perspectives.

Thanks!

Though I already wondered — in no uncertain terms on this blog — about the fact that you seemed to be comparing growth rates (instead of explicitly “comparing the ppm data” on this blog as I have done), I nonetheless had to ASSUME that you had actually “compared the ppm data” off-line and had simply not included it on your blog.

Now, I find that you have actually NOT “compared the ppm data directly”?!

How then could you feel confident in making a statement like the following?

“Any conclusion that Hansen’s 1988 prediction got things right, necessarily must conclude that it got things right for the wrong reasons.”

/////////

Having said that, I would next make a few comments about the “time interval” considered.

There is NO valid scientific reason NOT to consider the longer time interval in this case. Statistically speaking, more data points is better, particularly when we are talking about a relatively short time interval and when there is “scatter” in the measured yearly CO2 increments to begin with (and undoubtedly also in the total emissions and natural uptake of emissions from year to year).

Your comment that
“That scenario [C88] froze time in 2000″

is not an accurate characterization. Scenario C88 did not “freeze time”. It imposed a cut-off on emissions. From the scenario standpoint, there is a difference. Every Hansen graph I have seen (including the one you provided on your first post about this matter) comparing temperatures going forward for the three scenarios — and the actual temperatures has clearly shown scenario C88 data points beyond 2000.

You also continued:
“meaning that going forward there are two evolving scenarios which both have dramatically overestimated emissions. The lower of the two is thus “more accurate” than the other.

This conclusion is not warranted based on the data, as I showed above.

Such a statement that B88 “dramatically overestimated” emissions (including CO2 emissions) is simply not warranted as I showed above with the example for CO2.

B88 prediction of emissions was within 1.66ppm of the ACTUAL over 1990-2005.

If we assumed (quite artifically) that emissions did not CEASE after 1999 (as Hansen assumed) under scenario C88 and that the yearly increment of 1.5ppm per year continued, this “artifial C88″ scenario would STILL provide a less accurate predicter of total accumulated CO2 added to the atmosphere (ppm) than scenario B88.

You continued: “Neither is particularly accurate or realistic. Any conclusion that Hansen’s 1988 prediction got things right, necessarily must conclude that it got things right for the wrong reasons.”

THIS STATEMENT BY YOU IS REALLY THE CRUX OF YOUR ENTIRE POST and is simply UNWARRANTED — shown by the data that I have presented above.

On the contrary, Hansen’s B88 predictions about total CO2 increase were within 1.66ppm (6% of actual 27.3) for the period 1990-2005.

“2. With respect to Hansen’s 1998 predictions C98 has been the most accurate.”

I won’t comment on number 2 because I was restricting my attention to Hansen’s 88 paper.

“3. In two sets of predictions compared with experience, Jim Hansen’s predictions of emissions have proved to be overly aggressive both times with respect to rates of emissions with his lower estimate proving most accurate.”

This conclusion (covering both papers, the other of which I did not consider here) is NOT supported by the data — since it is not supported by the data for the B88 case (which showed the predicted accumulated total CO2 within 1.66ppm of actual for 1990-2005)

I will have to confirm what value Hansen used as the “base” — 1.5 ppm seems right, but I will confirm.

Also, more rapid increases in CO2 since 2000 may make a difference here as well. Give me a few days, I’ll catch up to you. Thanks.