UPDATE (January 31, 2011): Refer also to Part 2 of this post here.
My blog post October to December 2011 NODC Ocean Heat Content Anomalies (0-700Meters) Update and Comments was cross posted by at WattsUpWithThat here. (As always, thanks, Anthony.) Starting at the January 27, 2012 at 8:44 am comment by J Bowers, I was informed of a critique of my ARGO-era model-data comparison graph by the Tamino titled Fake Predictions for Fake Skeptics. Oddly, Tamino does not provide a link to my post or the cross post at WattsUpWithThat, nor does Tamino’s post refer to me by name. But I believe it would be safe to say that Tamino was once again commenting on my graph that compares ARGO-era NODC Ocean Heat Content data to the climate model projection from Hansen et al (2005), “Earth’s energy imbalance: Confirmation and implications”. If he wasn’t, then Tamino fooled three persons who provided the initial comments about Tamino’s post on the WUWT thread and the few more who referred to Tamino’s post afterwards. The ARGO-era graph in question is shown again here as Figure 1. Thanks for the opportunity to post it once again up front in this post, Tamino.
Figure 1
Take a few seconds to read Tamino’s post, it’s not very long, and look at the graphs he presented in it.
My graph in Figure 1 is clearly labeled “ARGO-Era Global Ocean Heat Content Model-Data Comparison”. The title block lists the two datasets as NODC/Levitus et al (2009) and Hansen et al (2005) Model Mean Trend. It states that the Model Mean Trend and Observations had been Zeroed At Jan 2003. Last, the title block lists the time period of the monthly data as January 2003 to December 2011.
In other words, my graph in Figure 1 pertains to the ARGO-based OHC data and the GISS model projection starting in 2003. It does not represent the OHC data or the GISS model hindcast for the periods of 1993-2011 or 1955-2011, which are the periods Tamino chose to discuss.
Do any of the graphs in Tamino’s post list the same information in their title blocks? No. Do any of Tamino’s graphs compare the climate model projection from Hansen et al (2005) to the NODC OHC observations? No. Does Tamino refer to Hansen et al (2005) in his post? No. Do any of Tamino’s graphs present the NODC OHC data or the Hansen et al model projection during the ARGO-era, starting in January 2003 and ending in December 2011? No.
In other words, Tamino redirected the discussion from the ARGO-era period of 2003-2011 to other periods starting in 1955 and 1993. ARGO floats were not in use in 1955 and they were still not in use in 1993. He also redirected the discussion from the projection of the GISS model mean to the linear trend of the data itself. Yet Tamino’s followers fail to grasp the obvious differences between his post and my ARGO-era graph.
In my post, I explained quite clearly why I presented the ARGO-era model-data comparison with the data zeroed at 2003. Refer to the discussion under the heading of STANDARD DISCUSSION ABOUT ARGO-ERA MODEL-DATA COMPARISON. Basically, Hansen et al (2005) apparently zeroed their model mean and the NODC OHC data in 1993 to show how well their model matched the OHC data from 1993 to 2003. Hansen et al explained why they excluded the almost 40 years of OHC and hindcast data. The primary reason was their model could not reproduce the hump in the older version of the Levitus et al OHC data. Refer to Figure 2, which is a graph from a 2008 presentation by Gavin Schmidt of GISS. (See page 8 of GISS ModelE: MAP Objectives and Results.) I presented that same graph graph by Gavin in my post GISS OHC Model Trends: One Question Answered, Another Uncovered, which was linked in my OHC update from a few days ago.
Figure 2
I zeroed the data for my graph in 2003, which is the end year of the Hansen et al (2005) graph, to show how poorly the model projection matched the data during the ARGO-era, from 2003 to present. In other words, to show that the ARGO-era OHC data was diverging from the model projection. Hansen et al (2005), as the authors of their paper, chose the year they apparently zeroed the data for one reason; I, as the author of my post, chose another year for another reason. I’m not sure why that’s so hard for Tamino to understand.
And then there’s Tamino’s post, which does not present the same comparison. He did, however, present data the way he wanted to present it. It’s pretty simple when you think about it. We all presented data the way we wanted to present it.
Some of the readers might wonder why Tamino failed to provide a similar ARGO-era comparison in his post. Could it be because he was steering clear of the fact that it doesn’t make any difference where the model projection intersects with the data when the trends are compared for the ARGO-era period of 2003 to 2011? See Figure 3. The trend of the model projection is still 3.5 times higher than the ARGO-era OHC trend. Note that I provided a similar graph to Figure 3 in my first response to his complaints about that ARGO-era OHC model-data comparison. See Figure 8 in my May 13, 2011 post On Tamino’s Post “Favorite Denier Tricks Or How To Hide The Incline”.
Figure 3
Figure 3 shows the ARGO-era OHC data diverging from the model projection. But the visual effect of that divergence is not a clear as it is in Figure 1. I presented the data in Figure 1 so that it provided the clearest picture of what I wanted to show, the divergence. That really should be obvious to anyone who looks at Figure 1.
Some might think Figure 1 is misleading. The reality is, those illustrating data present it so that it provides the best visual expression of the statement they are trying to make. The climate model-based paper Hansen et al (2005) deleted almost 40 years of data and appear to have zeroed their data at 1993 so that they could present their models in the best possible light. Base years are also chosen for other visual effects. The IPCC’s Figure 9.5 from AR4 (presented here as Figure 4) is a prime example. Refer to the IPCC’s discussion of it here.
Figure 4
The Hadley Centre presents their anomalies with the base years of 1961-1990. Why did the IPCC use 1901-1950? The answer is obvious. The earlier years were cooler and using 1901-1950 instead of 1961-1990 shifts the HADCRUT3 data up more than 0.2 deg C. In other words, the early base years make the HADCRUT anomaly data APPEAR warmer. It also brings the first HADCRUT3 data point close to a zero deg C anomaly, and that provides another visual effect: the normalcy of the early data.
Base years for anomalies are the choice of the person or organization presenting the data. Climate modelers choose to present their models in the best light; I do not.
Those familiar with the history of Tamino’s complaints about my posts understand they are simply attempts by him to mislead or misdirect his readers. And sometimes he makes blatantly obvious errors like using the wrong sea surface temperature dataset in a comparison with GISS LOTI data. His post Fake Predictions for Fake Skepticsis just another failed critique to add to the list.
ABOUT: Bob Tisdale – Climate Observations
SOURCE
The NODC OHC data used in this post is available through the KNMI Climate Explorer:
http://climexp.knmi.nl/selectfield_obs.cgi?someone@somewhere
Bob Tisdale - Climate Observations 
Great Post!
Thank you.
Bob,
Your research and analysis are appreciated, and enhance the knowledge of many. Tamino cannot accept the conclusions of your work for the simple reason it contradicts or questions the very premise of “the cause”. His analysis and commentary is not about science, but a means to an political end. Justifying your work with Tamino is nothing more than “pearls before swine”. Thanks for the extensive work you do and sharing it with us all.
Bob.
If this isn’t a practice run for April 1 I do not know what is. have a giggle.
http://www.theregister.co.uk/2012/01/30/ocean_currents_warming_and_moving/
regards from CO2 warmed Thailand
The 1993 – 2003 period was “chosen” because that was the extent of a new, purportedly more accurate data set (Willis 2004), based on a combination of sea level data with in situ temp measurements. This is why people are explaining the data selection in terms of satellite altimetry coming on line in 1993, Bob. you seem to be unaware of the reference material Hansen et al 05.
I have two questions at WUWT for you. Let me put them here.
The best way to test Hansen’s trend prediction is to simply extend his trend line from where it ended in an unbroken line, like this.
1. Do you agree?
if not
2. Why?
Barry: “The 1993 – 2003 period was ‘chosen’ because that was the extent of a new, purportedly more accurate data set (Willis 2004), based on a combination of sea level data with in situ temp measurements.”
You’re right on one count. Hansen et al (2005) compared their OHC simulations against Willis et al (2004) data. They are very clear about that in the text of their Figure 2. But if you haven’t noticed, I’m not presenting Willis et al (2004) OHC data. I’m presenting NODC OHC data. Hansen et al also explain why they don’t use a longer-term dataset. Read the first paragraph of Hensen et al under the heading of Ocean Heat Content. Basically, the models could not reproduce the decadal variability of the now obsolete NODC OHC data shown here:
That graph is Figure 2 from my post above.
But here’s an observation for you. Refer to Figure 3 in Willis et al (2004).
Click to access Wills_2004.pdf
And refer to Figure 2 from Hansen et al (2005):
Click to access 2005_Hansen_etal_1.pdf
Do you notice any difference between the presentation of the OHC data in Willis et al (2004) and in Hansen et al (2005)? Sure does look like Hansen et al shifted the Willis data, doesn’t it? In the Hansen et al Figure 2, the data starts at zero anomaly in 1993, but in Willis et al (2004) the OHC data in 1993 starts around -1 J/m^2. So it’s okay for Hansen et al to shift data around, but it’s not okay for Bob to do it.
Posts that discuss the annual RealClimate model-data comparisons are linked in my NODC OHC updates. The RealClimate posts are also discussed in my first debate with Tamino. RealClimate shifts the base years around to accommodate the changes in the OHC data. And RealClimate changes the term of the presentation from one year to the next. In the 2009 update…
http://www.realclimate.org/index.php/archives/2009/12/updates-to-model-data-comparisons/
…RealClimate included data from the start of the dataset 1955 with one set of base years, but in the 2010 update, they start the comparison in 1970 with another set of base years for anomalies:
http://www.realclimate.org/index.php/archives/2011/01/2010-updates-to-model-data-comparisons/
In short, it’s also okay for Hansen’s associates at RealClimate to shift data around, but it’s not okay for Bob to do it.
Barry says: “This is why people are explaining the data selection in terms of satellite altimetry coming on line in 1993, Bob. you seem to be unaware of the reference material Hansen et al 05.”
I’m quite aware of the reference material as discussed above. My argument with Jason was that he was insisting that Sea Level data and OHC data were the same thing, and they are not. Willis et al (2004) used Sea Level data to supplement and refine the in situ data-based estimate of OHC.
Barry says: “The best way to test Hansen’s trend prediction is to simply extend his trend line from where it ended in an unbroken line, like this.”
Disagree. First, Tamino says that he included 2003 data in his “1993-2003 fit”, but Tamino’s trend uses data through 2002, not 2003. The NODC clearly marks their quarterly data in the spreadsheet they provide:
ftp://ftp.nodc.noaa.gov/pub/data.nodc/woa/DATA_ANALYSIS/3M_HEAT_CONTENT/DATA/basin/3month/ohc_levitus_climdash_seasonal.csv
The Global OHC value at 2002-12 is 6.368768*10^22 Joules, and the value at 2003-12 is clearly marked 11.6546*10^22 Joules. But the data included in Tamino’s trend stops at 2002-12. Tamino even makes the following observation after the graph of his you linked:
“We can see that observations don’t follow the prediction exactly — of course! The main difference is that during 2003, the observations were hotter than the prediction. For that time span at least, the oceans had more heat than predicted.”
Yet his 1993 to 2003 trend does not include the “hotter than the prediction” 2003 data. Yet his graphs and his text say that it is included. If you’re having trouble seeing that, here’s graph similar to Tamino’s graph with the hash marks included. It shows the trend based on the period of 1993 to 2002:
And here’s what the graph would look like if had actually included 2003 data in his trend:
Second, Tamino did not present the model data. He only presented a trend of the NODC OHC data for 1993-2002 and extended it to 2011.
Third, as also noted in this post, Tamino did not present the 2003 to 2011 graph, which is what this discussion is about.
Fourth, the best way to present the GISS Model-ER projection would be to use the actual GISS Model-ER data. But it’s not available in an easy-to-use format like one would find at the KNMI Climate Explorer. If it was available, all of this nonsense about my shifting data, my misrepresenting data, blah-blah-blah, would disappear.
For example, Figure 2 in the post above is a graph from a 2008 presentation from Gavin Schmidt of GISS. It includes the OHC simulations of the Model-ER for the period of 1955 to 2010, which is the model data shown in the RealClimate model-data posts. It also includes the older version of the global NODC OHC data. So if we replicate the ensemble mean data of the GISS Model-ER, replace the older NODC OHC data with the current version, and then use the base years of 1955-2010 so that no one can complain about cherry-picked base years, this would be a reasonable facsimile of the long-term comparison from 1955 to 2010:
Notice where the ensemble mean of the GISS Model-ER would intersect with the data for the ARGO-era graph? Sure looks like 2003 to me. Let’s check.
That graph looks familiar, doesn’t?
Pingback: Part 2 of Tamino Once Again Misleads His Disciples | Bob Tisdale – Climate Observations
Pingback: Part 2 of Tamino Once Again Misleads His Disciples | Watts Up With That?
Bob,
If you think Hansen’s fig 2 comparison is spurious because the zeroing is an unjustified choice, doesn’t that make your own effort at WUWT spurious, too? If you are trying to demonstrate the arbitrary nature of the original choice, then you may have succeeded, but it seems you are taking your own choice of zeroing seriously, but justifying it by saying you should be allowed to be as cavalier as Hansen.
Willis 04 data is not once mentioned in your articles, and you have instead strongly given the impression that Hansen’s Fig 2 relied on the NODC data, deliberately omitted years of that data and selected the period for convenience. Let me quote you:
Omitting mention of the Willis 2004 data and inferring that Hansen used deliberately truncated NODC data is a strong misrepresentation.
Hansen does not say that the inability of models to reproduce large decadal fluctuations is the reason why the period 1993 – 2003 is selected. He only notes the fact. You are free to speculate what Hansen meant, of course, but you are presenting speculation as fact.
Hansen’s choice of Willis’ data is given in the OHC section:
The Willis data is chosen because (i) it is the then best estimate of OHC content, (ii) the warming signal is clearer, being less perturbed by internal variation (according to Hansen).
One can argue the merits of choosing to work with the Willis data over Levitus/NODC. But you have created a false narrative about the data Hansen used by omitting any mention of Willis 2004 in your posts. Within the context of this false narrative you have presented derogatory speculation on Hansen’s motives as fact.
Personally, I think the 11 year period is too short for a meaningful comparison with models. I agree that the model/obs match is potentially too convenient for this particular graph. the choice for ‘zeroing’ is not explained satisfactorily, either in the paper or elsewhere.
However, according to NODC data, 1993 is not an outlying anomaly. It is on the low side, but not a huge departure. OTOH, your choice of start time for comparison is in the middle of a HUGE upwards spike in OHC. It is virtually certain that such a departure will be followed by a lower trend as the system equilibrates. It’s like running short term trends from 1998 in the surface record. Of course short term results will be spurious.
Hansen’s fig 3 appears not to be creditable, but not for the reasons you laid out. Your 2003 – 2011 graph is much less so.
It is difficult to find descriptions of methodology on baselining/zeroing model comparisons with OHC obs (methods WRT surface temperatures are easier to find and well-defined). I seems that you have no rationale other than you should have the same freedom as Hansen, Schmidt etc. Hardly a scientific rationale.
Barry: You started with, “If you think Hansen’s fig 2 comparison is spurious because the zeroing is an unjustified choice…”
Please advise where I have ever stated or implied that Hansen et al’s zeroing of the data in their Figure 2 was wrong in any way. I’ve used it as a reference in a what’s-good-for-the-gander thing, as you later note.
You wrote, “Willis 04 data is not once mentioned in your articles, and you have instead strongly given the impression that Hansen’s Fig 2 relied on the NODC data, deliberately omitted years of that data and selected the period for convenience.”
Hansen et al clearly state that their models cannot reproduce the decadal variability of the older OHC datasets. Based on their inability to reproduce those older variations, 1993-2003 IS a period chosen for convenience. Regarding my not mentioning Willis, look at the text for Hansen et al’s Figure 2. There’s a referral to reference 20. Reference 20 was Willis et al (2004). I’ve provided links to Hansen et al in the posts. Would you like me to mention and link all papers referenced in Hansen et al?
Regarding my description of the reasons why Hansen et al selected the period of 1993-2003, I’ll look again at my description and determine if I need to change the language.
You wrote, “OTOH, your choice of start time for comparison is in the middle of a HUGE upwards spike in OHC.”
I have explained the multiple reasons for the use of 2003 numerous times recently and in the first go-round with Tamino.
https://bobtisdale.wordpress.com/2011/05/13/on-taminos-post-favorite-denier-tricks-or-how-to-hide-the-incline/
You concluded, “It is difficult to find descriptions of methodology on baselining/zeroing model comparisons with OHC obs (methods WRT surface temperatures are easier to find and well-defined). I seems that you have no rationale other than you should have the same freedom as Hansen, Schmidt etc. Hardly a scientific rationale.”
Data selection and presentation is the choice of the author of a scientific paper and a blog post. I can’t fathom why people don’t understand that. Hansen et al and RealClimate choose to present the models in the best possible light. I do not.
Let me be clear I am not interested in a game of sides or defending anyone.
If your argument was simply that anyone can play with numbers to get the results they want I would have little issue with your article. But you are presenting the results of your 2003+ graph as if they are based on a solid foundation (statistically sound zeroing) and are meaningful in some way. Saying it’s good for the goose cuts no mustard scientifically speaking.
You mislead readers who may not follow the references as assiduously as I into thinking that Hansen deliberately truncated NODC data to get the results he wanted. I don’t think that is a responsible way to describe the provenance of the 1993 to 2003 data period.
Bob says: “Please advise where I have ever stated or implied that Hansen et al’s zeroing of the data in their Figure 2 was wrong in any way.”
I quote you:
It’s hard to figure why you think that manipulating baselines to get a good fit is not “wrong in any way.”
But if you think that this is ok, then I still don’t understand how your choice of zeroing from 2003 is superior to simply extending Hansen’s trend line – other than, as you’ve said, it gives you the results you want. If you want to test the trend prediction from Hansen et al, then use their model trend line, match baselines for Willis 04 and NODC data and see how it pans out.
If you don’t do this, then you are not analysing Hansen’s trend comparison – you are doing another one, and you are doing it for a shorter period, with no modelling to account for forcing variability post 2003. The model hindcast runs for Hansen et al (described in their 2007 paper) account for fluctuations in GHGs, aerosols, solar throughout the period. This is not the case for your analysis. I asked if you were assuming a continued trend, you replied yes – but this is not a good assumption, as various forcing changes (aerosols/solar for example) may affect modelled results for the post 2003 period (perhaps we’ll see this in AR5). You ask how long before the models are invalidated – well, first do the modelling (in this case hindcasting, a different experiment to forecasting).
Bob says: “Hansen et al clearly state that their models cannot reproduce the decadal variability of the older OHC datasets.”
Yes.
Bob says: “Based on their inability to reproduce those older variations, 1993-2003 IS a period chosen for convenience.”
That is your interpretation, NOT an assertion of the paper. As you say, they “clearly state that their models cannot reproduce the decadal variability of the older OHC datasets,” so they are not ignoring inconvenient truths. Willis 2004 offers a new improved data set. That is all the rationale that is directly given in the paper for using it, and you correctly give that as reason #2 in your prior post. The other rationales you have given are at best implied in Hansen et al, and a proper interpretation of that paper would caveat those suppositions accordingly. A purely scientific approach would leave out such supposition altogether. For instance, you could focus on the statistical significance on the period chosen by Hansen et al for OHC. Have you calculated that and compared with your own results? This gets us into science and away from speculation about motives.
Bob says: “I have explained the multiple reasons for the use of 2003 numerous times recently and in the first go-round with Tamino.”
Indeed you have, but I don’t believe you have responded to my criticism of that choice. Perhaps you could address that in your next reply. I’ll repeat it for clarity.
“…according to NODC data, 1993 is not an outlying anomaly. It is on the low side, but not a huge departure. OTOH, your choice of start time for comparison is in the middle of a HUGE upwards spike in OHC. It is virtually certain that such a departure will be followed by a lower trend as the system equilibrates. It’s like running short term trends from 1998 in the surface record. Of course short term results will be spurious. “
Regarding your question on how long it would take to invalidate the models, there are suggestions given in this paper (h/t MapleLeaf);
Click to access sst-et-temp-mondiales.pdf
Model runs in this paper came with a small number of decadal trends that were negative in an overall warming scenario (6 – 8% of model runs with 4000 meter depth). A slow down in OHC warming at 750 meters for a dozen or so years would not invalidate these models.
It is a familiar story with global surface trends. While model ensemble means show overall warming and little deviation at decadal level, individual model runs have cooling periods for 10 and even 20 years with an overall warming scenario (A1B projections, AR4). A slow down in warming from 2003 to 2011 in OHC at 750 meters is a long way from breaking the models.
Barry: Thanks for the link to Palmer et al. According to their model-based study, and based on the depths of 0-700m for the NODC OHC data, to invalidate the models, we’re looking at somewhere between the 15-20 years for SST and the 6-7 years for OHC for depths of 4000m. Since most of the variability takes place in the top 700m, we’re looking more towards the 15 years, again according to the paper. But then we’d have to assume the models used by Palmer et al have bases in reality.
The use of individual model runs to explain slowdowns is a valid practice only if the authors can show that the noise internal to the models that are creating those slowdowns result from the same processes that have caused the slowdown in nature. The conflict I see: modelers need ensembles for hindcasts and projections to overcome noise that is caused by model imperfections. And now they’re using the noise based on those imperfections as means to explain away the possible failure of the models.
In your earlier January 31, 2012 at 9:42 pm comment, after quoting me, you wrote, “It’s hard to figure why you think that manipulating baselines to get a good fit is not ’wrong in any way.’”
It’s not. Why would you think it was? If your intent as a modeler was to show a good model fit to the Atlantic Multidecadal Oscillation since 1975, for example, you wouldn’t use 1900 to 1925 as your base years, you’d use 1975 to present. That’s simply common sense.
For the rest of your January 31, 2012 at 9:42 pm comment, you’re belaboring points you’ve already expressed and to which I’ve already replied.
Regards
I appreciate you taking my comments on.
Bob says: “The use of individual model runs to explain slowdowns is a valid practice only if the authors can show that the noise internal to the models that are creating those slowdowns result from the same processes that have caused the slowdown in nature.”
How much confidence should we put in analyses that don’t discuss physical processes at all?
GCMs attempt to model the physical behaviour of the systems that make up climate and weather. Most blog-critic attempts to compare model and obs are purely statistical, the vast majority being simple linear regressions. I agree with the theme of what you’ve said here, but I think you’ve aimed it at the wrong milieu.
Bob says: “The conflict I see: modelers need ensembles for hindcasts and projections to overcome noise that is caused by model imperfections. And now they’re using the noise based on those imperfections as means to explain away the possible failure of the models.”
Noise is not caused by ‘imperfections’. Noise (as in stochastic variation) is a natural feature of the real climate system. If individual models replicate the character of the noise in the observations, this is a good sign of their fidelity (but by no means a guarantee). Ensembles are done to get mean estimates for trends, or probability distribution for a given scenario in forecasting. Ensembles damp out the variability – yes – but not because they are trying to hide it!
AR4 (physical) models were finished by 2006. Many showed 10 and even a couple of 20 year cooling periods with overall warming (A1B scenarios), and soon after AR4 came out people were claiming that the models had failed because of a flat trend in surface temperatures since 1998. There is a new voice in the blogosphere saying that every other week. The AR4 modellers weren’t fudging to anticipate this criticism. The cool periods in individual runs were a product of the physics the models are based on. They were not tricked up to “explain away model failure.”
(One entrenched critic tried to tell me the models were busted because monthly surface temperature data fell below the range of an ensemble in an IPCC graph in Ch 10 that had 3-year smoothing.)
It’s just as likely that modellers are explaining short-term variation better because they are understanding it better. At least – there is as much substance for my suggestion here as there is for yours that they are trying to perpetuate some meme about global warming. If we’re going to indulge in guessing motivation,let’s at least admit that we’re speculating.
Having read your entire post more carefully (rather than focussing on Hansen fig 2), I see now that your theme is general. To whit: the modellers are fiddling their graphs and tweaking their models to maintain a certain frame on the subject of global warming – to make it more alarming I suppose you mean. You are implying scientific dishonesty and conspiracy to mislead (the IPCC is a committee).
Regarding your criticisms on the IPCC figure 9.5, the original model/obs comparison (Stott et al 2006) baselined the model data at 1961 – 1990, then offset the model values so that the mean for the entire period (1870 – 2004) was zero. If both sets had been aligned at the earliest period, when CO2 warming was most minimal (CO2 warming was swamped by natural variation before 20th century), the result would be very similar to the IPCC graphs. But the criticism is unwarranted in the first place, because the objective is clearly to compare trends over the 20th century, and the clearest visual representation is to centre model and obs around zero for the first year. Because the time frame is a century there is no fear of spurious trend results. The difference between the final values is not the point (nor is it highlighted or even discussed), and the resulting change due to baseline choices is (visually) minimal anyway. It’s not going to make a difference to policy-makers (I speculate!). But once again, this is a digression away from science and into politics.
Scientists try to avoid having to make choices. But when they must it is better if they are well justified. The IPCC choice above is defensible. I’m not so sure about Hansen et al (although they do caveat and give error bars). Your choice for 2003+ model/obs comparison is not well-justified, for reasons I have already belaboured, as you say.
Barry says: “Noise is not caused by ‘imperfections’. Noise (as in stochastic variation) is a natural feature of the real climate system. If individual models replicate the character of the noise in the observations, this is a good sign of their fidelity (but by no means a guarantee).”
The greatest cause of year-to-year variations in global surface temperature and in ocean heat content is ENSO. There are no coupled ocean-atmosphere general circulation models that model ENSO well. All fail in numerous areas, though some model it better than others. Some don’t bother, like the GISS Model-E. The Model-E not referred to in many papers that analyze how models simulate ENSO. If the models cannot model ENSO, then their noise has nothing to do with ENSO. And that ties back to my original comment that you quoted.
Barry say: “AR4 (physical) models were finished by 2006. Many showed 10 and even a couple of 20 year cooling periods with overall warming (A1B scenarios)…”
Surface temperatures cooled for 30 years between 1945 and 1975. By your own description, the models cannot capture those variations. I have been through each CMIP3 model and ensemble member that presents sea surface temperature, and I could find no CMIP3 models that capture the multidecadal sea surface temperature variations of the North Pacific, north of 20N. I’m not talking about the PDO; I’m referring to sea surface temperature anomalies. So please point me to a paper that describes which of the CMIP3 models capture those multidecadal variations.
Barry says: “But the criticism is unwarranted in the first place, because the objective is clearly to compare trends over the 20th century, and the clearest visual representation is to centre model and obs around zero for the first year.”
I wasn’t criticizing; I was pointing out the obvious. And the primary objective of Figure 9.5 was to show that, according to the models, only anthropogenic forcings could be the cause of the rise in surface temperatures during the late warming period. It had nothing to do with trends. Trends are discussed under the heading of “9.4.1.2 Simulations of the 20th Century” but they are not referring to Figure 9.5.
Barry says: “Your choice for 2003+ model/obs comparison is not well-justified, for reasons I have already belaboured, as you say.”
And I’ve already discussed the numerous reasons why I present the 2003+ comparison, so we will agree to disagree.
Pingback: Part 2 of Tamino Once Again Misleads His Disciples | My Blog
So why are you making serious comparisons of short-term observations with models when models don’t simulate short-term variability well?
barry says: “So why are you making serious comparisons of short-term observations with models when models don’t simulate short-term variability well?”
The answer to that is obvious. Also, models don’t perform well on a multidecadal basis either.
The answer isn’t obvious to me, in the same way as I can’t figure out what prompts people to talk in absolute terms about short-term trends in surface temperature that fail statistical significance testing.
If I said that your purpose is to prove the models wrong, would that be a fair description of your intent?
Barry, I’ve explained the reason for the graph.
Good bye
It will be right that Noise figure is expressed in dB. Smaller numbers correspond to lower noise and a greater ability to detect weak signals.
Pingback: Corrections to the RealClimate Presentation of Modeled Global Ocean Heat Content | Bob Tisdale – Climate Observations
Pingback: Gavin Schmidt issues corrections to the RealClimate Presentation of Modeled Global Ocean Heat Content | Watts Up With That?
Pingback: Part 2 of “On Sallenger et al (2012) – Hotspot of Accelerated Sea Level Rise on the Atlantic Coast of North America” | Bob Tisdale – Climate Observations
Pingback: Argo, profiled … Tisdale corrects | pindanpost
Pingback: Is Ocean Heat Content Data All It’s Stacked Up to Be? | Bob Tisdale – Climate Observations
Pingback: Is Ocean Heat Content Data All It’s Stacked Up to Be? | Watts Up With That?
Pingback: On Sallenger et al (2012) – Hotspot of Accelerated Sea Level Rise on the Atlantic Coast of North America | Bob Tisdale – Climate Observations
Pingback: Tamino Resorts to Childish Attempts at Humor But Offers Nothing of Value | Bob Tisdale – Climate Observations
Pingback: Tamino Resorts to Childish Attempts at Humor But Offers Nothing of Value | Watts Up With That?