That there has been a rise in the trend of between .15 and .20 every decade for the last 4 decades is also a fair thing to say. Perhaps a caveat that has to be done in groups of longer than 20 years averaged?

So it is also fair to say that a period of less than 20 years that is not on track for a .15 to .20 trend is not inconsistent (not yet proven consistent or inconsistent) with that rise.

And it is also fair to say that it is consistent –with past behavior — for a period of less than 20 years to not be on track for a .15 to .20 trend.

But it is not fair to say the *last 10 years* is *consistent* with a .15 to .20 rise per decade, is it?

It has to be qualified, right? Throw in a “It’s not consistent because less than 20 years is an unreliable time period” or that the last 10 years is not known to be consistent or not because the models or what have you can’t deal with a time period that short because it is too variable. Is that fair?

It is “not inconsistent”

Starting at 1995, we only have to wait until 2014 to see if it was consistent or not, don’t we?

To answer the question, periods of 8 years with a negative sloping trend start in 1988, 1987, 1986, 1980, 1979, 1970, 1969, 1968 and 1967. Then I stopped. So an 8 year negative trend is not unheard of, but there hasn’t been one in 20 years. You see they used to happen every 10 years or so? So it is rare now. Volcanos in those years, yes? What about our volcano this year what will it do.

Jan 2000 to Apr 2008 has a trend of .02 so far with YTD at .35 Will 2008 end up like 2000 or 1999 or 1996? That would change an 8 year trend of .2 to basically zero, wouldn’t it?

8 years from 2007 back is .18 about the same as 9-13 years. Seems the trend stays the same at ~.2 So 6 flat years? But it appears the trend so far is about .15 per decade, yes. And sometime in a decade lower than usual yearly anomaly grabs the trend to that figure, such as in 1992-1994

Number of years since 2007 trend backwards in time
7=.08
8=.18
9=.22
10=.18
11=.18
12=.22
13=.22
14=.30<—1994
15=.35<—1993
16=.42<—1992
17=.40
18=.38
19=.40
20=.40
30=.55
40=.70
50=.60
60=.65

Seems like a forecast of .15 or so C per decade is an easy forecast to make these days.

1. 1998-2017. (I am not sure that it is 20 years, it might be more. IPCC uses 20 or 30 year means. In AR4 it mainly was 20)

2. The time frame depends on the signal to noise ratio. I do not know, if there are common definitions or rules. In principle, if you want to compare to 20 year trends, it is of course best to use 20 year trends. However, a time frame of about 12 to 15 years might be sufficient (15 year trends do not show much variability anymore).

3. You misunderstood. I say two things:
- the difference between two subsequent 5-year periods has less variability than 5-year trends
- subsequent 5-year means have the problem with noise, they are comparable to 10 year trends (8 year trends are not that different, that is true).
- the differences I mentioned (starting with 1971-75 period) are always positive. However, you find a few negative differences if you take other sequences.
I agree that five year means are not suitable for statistical “proof” of long-term trends. What you can say is that the observations show what is most probable if you have a long-term warming trend (positive differences in 90% of the cases). That’s what IPCC meant. It is, however, not a very strong argument, I agree. It does not confirm model projections, but IPCC didn’t claim that.

4. Are you sure, that without volcanoes there would have been no negative 7 year trends in the 80ies and 90ies? On that scale, you also have ENSO variability (and other). It is difficult to eliminate the volcano influence in the data, one could use model data, but since you do not trust the models, this does not help. However, I do not expect 7 year trends to be always positive. But for 15 year trends, I do.

Thanks for your continuing comments. A few more reactions follow:

1. What 20-year period in Figure 10.26 is meant by the data shown for, say, 2008?

2. I’d be curious as to what you view as the appropriate time frame for which signal rises above noise? It is not 8 years, so how long?

3. Are yo really saying that comparing two 5-year means is not subject to the same issues of noise that you claim make 8-year trends useless? Really?

With regards to the following: “If you compare 5-year periods (1971-1975, 1976-1980, etc.) you see always a rise from one to the next period.”

Are you sure about this?

How do I reconcile the claim that 8-year cooling trends are consistent with longer term (20 year) trends, but every 5-year period will “always rise”? Logically this cannot be so (since a year period of warming followed by 8 years with negative slope, won’t show the average of the latter to be warmer than the earlier period).

How about this: The IPCC probably overstretched in this claim?

4. There have been no volcanoes over the past 8 years significant enough to affect forcing (so I am told), but the fourth figure here shows a negative trend (but only for 7.25 years):

http://sciencepolicy.colorado.edu/prometheus/archives/climate_change/001433do_ipcc_temperature_.html

1. Figure 10.26 does not show in any way short term forecasts. Presented are at least 20 year means. An 8 year period of a 20- or 30 year running mean is not equal to an 8 year short term forecast.

2. No. If you have a curve with much interannual noise and a long-term trend, it is not a matter of probability, but a matter of the signal-to-noise-ratio. The test of consistency to the long-term trend is only reasonnable for a period where you have a reasonnable signal-to-noise ratio. The winter-summer example is a perfect comparison, since you have on the one hand a (relative) long-term trend due to external forcing (insolation change due to earth axis, and greenhouse forcing, respectively), and short term variability (noise) due to internal chaotic behaviour of the system (weather, and internal oscillations, respectively) on the other hand.
What you are doing is testing, if the noise is consistent with the signal. This makes no sense at all. You compare completely different things.

Climate models are certainly not as certain as the onset of summer, I used the analogy to show the difference between an externally forced long-term signal and internal noise (see above).

3. Note the difference: IPCC is talking about 5-year mean temperatures (2001-2005), not 5-year trends. This is a huge difference. If you compare 5-year periods (1971-1975, 1976-1980, etc.) you see always a rise from one to the next period. That’s what IPCC meant. Let’s see what 2006-2010 will be. But I agree, that a 5-year mean still contains some noise (but is much better than an 8-year trend, because there is less noise) and a 10-year mean would be better.
2003-2007 is warmer than 1998-2002 by between +0.05 and +0.1 K (depending on the dataset), so the statement would be the same.

4. You are right, that there is an influence of volcanoes. But that was not my point. I was just saying, that in principal negative 8-year trends (whatever the reason would be) are consistent with contemporaneous unambigously positive 20-year trends. You agree? BTW, there were no volcanoes and no negative trends during the last 8 years, you agree?

Thanks again for your substantive comments. Some reactions and answers follow:

1. On whether on not the IPCC made short term predictions, it certainly did. Figure 10.26 for example shows temperature increases and uncertainty ranges (for the forced component) beginning in 2000. Over at Real Climate they go into the model output to show spread of IPCC predictions for 7, 8, 10, and 20 years into the future. No where have I seen the IPCC say anything about the lack of skill, accuracy, or consistency in short-term forecasts (<20 years). perhaps they did, and if so, a pointer would be welcomed.

2. Whether or not a short-term trend (say the first 8 years of 20 years) is consistent with the 20-year trend is a matter of probability. It is perfectly fair to ask the question, and it has multiple answers, since we are not dealing with the onset of summer or playing cards or rolling dice. Sorting through those questions and answers is a perfectly reasonable thing to do.

We are of course dealing with an open system which is represented in a variety of ways (models) with a single realization that matters (Earth).

You analogy suggesting that climate model predictions are as certain as the onset of summer is noted. However, given that humans have observed the onset of summer tens of thousands of times, and climate model predictions of the 21st century exactly zero times, perhaps we need to find a better analogy.

3. Citation: Chapter 9, WG I, p. 683, in reference to the period 2001-2006:

“Six additional years of observations since the TAR (Chapter 3) show that temperatures are continuing to warm near the surface of the planet.”

If that sentence were re-written today would it say something like the following?

“Six additional years of observations since the TAR (Chapter 3) show that temperatures are NOT continuing to warm near the surface of the planet, WHICH IS OF COURSE CONSISTENT WITH PREDICTIONS OF LONG-TERM WARMING.”

4. You state: “BTW, the reality of the last 30 years clearly has shown that negative 8 year trends are perfectly consistent with contemporaneous 0.15K/per-decade-trends over 20 years”

Please show me eight-year periods in the historical record with a negative 8-year trend, which did not occur immediately after a major volcanic eruption. I am not saying that there are no such periods, but I do know that there are none in the past 30 years unaffected by volcanoes, which makes your statement a bit misleading. Just let me know those dates.

Thanks!

It is a third thing to say, and that is what I really said, that (a) the IPCC did not make short-term predictions and (b) that observations are consistent with the short-term variability in the models (but not short-term predictions, because there aren’t any!).

The IPCC projects a mean(!) trend of 0.2K per decade over the next two decades. This does not mean a 0.2K trend over the next few years.
In the same way as the prediction that the temperature trend from March to July will be positive in any case (for all 95% range) does not mean, that temperature will rise during the next week. Temperature trends over a week have much more variability than the trends from March to July.

Would you pretend that a week with a negative temperature trend is inconsistent with the prediction that the trend from March to July (in the northern hemisphere) is always positive?

Why do you persistently ignore the difference between variability of a variable and a specific value a variable has taken?

Could you please tell me where exactly the IPCC compares a six year trend to model results (chapter, page, citation)?

BTW, the reality of the last 30 years clearly has shown that negative 8 year trends are perfectly consistent with contemporaneous 0.15K/per-decade-trends over 20 years

Let me correct one misinterpretation that you have. My effort to compare observations with forecasts was motivated by the IPCC doing the exact same thing, including highlighting a 6 year (!) post-TAR trend. I don’t recall you or Real Climate finding fault with those efforts, but perhaps I missed your critique?

You write:

“you have to compare the variability in GCMs to the variability of the measured sample of 8 year trends over e.g. the last 50 years and then test if the distributions are similar or different (and you would have to compensate for the long-term trend and volcanic eruptions)”

I indeed posed exactly this option to James and he rejected it in favor of arguing that the most recent 8 years are consistent with model predictions.

It is one thing to say that short-term predictions of the IPCC are (a) not really made and (b) not to be expected to be skillful.

It is quite another to say that such short-term predictions of the IPCC are (a) not really made and (b) but consistent with observations.

The latter is a bit of eating your cake and having it also. It also sets the stage for overselling what the models are designed for.

There is no crime in admitting that IPCC predictions for the short term were presented with misleading (too small) error bars for comparison with what actually will evolve on plant Earth.

I am sure, James, if he really did (where?), only rejected part of your proposition, namely that it makes no sense to compare short term variability. Of course, it can make sense to compare 8 year trends in reality and in GCMs to test if GCMs get the short-term variability in general right (but not an individual one, because they are not designed to). However, you have to compare like with like. That means, you have to compare the variability in GCMs to the variability of the measured sample of 8 year trends over e.g. the last 50 years and then test if the distributions are similar or different (and you would have to compensate for the long-term trend and volcanic eruptions). But that is not what you have done in this post. You compared the variability in GCMs to the measurement uncertainty in one single realisation. The comparison of these two distributions is nonsense, because they describe totally different things.

The recent Realclimate post discusses the short term variability of individual GCM runs and not the validity of long-term trends.

My comments are in perfect agreement with the Realclimate posts and certainly with James Annans comments (just ask him if he disagrees on any point).