The Blackboard

steven mosher (Comment#1241)
March 21st, 2008 at 11:02 am

no clue.

But here is thing to read

http://www.grida.no/climate/ipcc_tar/wg1/371.htm

steven mosher (Comment#1243)
March 21st, 2008 at 11:12 am

Here is an interesting plot

http://www.grida.no/climate/ip.....ig9-15.htm

Pick a scenario. A1F1 ( I can explin this if you like, but its a vision of the future where everyone
grows emmissions like crazy)

From 1990 to 2030. One GMV predicts a bit less than .8C in warming. Another model ( based on the same inputs)
estimates 1.25C warmin. you can go translate that into C per decade error

lucia (Comment#1245)
March 21st, 2008 at 12:16 pm

I sort of wish they’d put the scenarios on the axis and used symbols for the models. That would let us pick out the spread in the models.

steven mosher (Comment#1247)
March 21st, 2008 at 12:54 pm

Ha, that would be too easy to check there dang work.

Possums.

For example, Lets focus on B1 ( I’ll explain this later if you like but the nomeclature is a secret decoder ring )

B1 is a Scenario of future emissions is that is based on certain assumptions. They call it a “story line”
B1 would have populations, energy mixes, land use aassumptions, GDP assumptions, etc etc etc.
A fairy tale of sorts. This scenario is cashed out at a set of emissions ( forcing for lumpy, for example)

These forcings are fed to GCM. So this is your garbage in.
( no disrepsect intended)

So. B1 scenario is run by everybody ( mostly)

At one extreme. GFDL says : IF YOU INPUT B1 Scenario GHG into our model and run our model for 40 years, you will see
about a .95C increase in temperatures at 2030

At the other extreme the US DOE says: if you INPUT the B1 GHG scenario into Our model for 40 years, you will see
about a .575 increase overby 2030.

It’s very obscure stuff

Ian Castles (Comment#1250)
March 21st, 2008 at 2:29 pm

An interesting detail: the above chart shpws that the highest projected temperature increases among the SRES scenarios, in all models, are for the A1T scenario (the red squares). This scenario assumes much LOWER fossil CO2 emissions than A1B, A1FI and A2, but finds that the warming up to 2030 is GREATER because it also assumes substantial reductions in emissions of (cooling) aerosols. In October 2006 I took this up with Roger Jones of CSIRO, an IPCC Coordinating Lead Author, on John Quiggin’s blog:

“Roger, I’m hoping that you will be able to answer my question about the implications for modelled temperature of the steep decline in global sulphur dioxide emissions in the 1990s which has been identified in several studies published since the last IPCC Assessment Report…

“In your posting above, you said that the highest temperature of the range I’d quoted for 2030 “is attached to the A1T scenario that has more aggressive sulphate emission reduction than the other SRES scenarios.” If you assume, in accordance with the results of the studies cited above, that “more aggressive sulphate reductions” in line with the A1T projections for 2030 had already occurred by the mid-1990s, what does that imply for the increase in temperature that would occur from now on?

“Specifically, if sulphur emissions as estimated in Stern D. I. (2005) ‘Global sulfur emissions from 1850 to 2000′, Chemosphere 58, 163-175 and the database supporting that paper are substituted for those that were used to produce the SRES … projections, what is the effect on the global mean temperature up to now, and the projected increase between now and 2030?…

Roger replied as follows:

“Ian, re your last post. I can try this (substituting alternative SO2 pathways) using the MAGICC model, and it will be an interesting little test but it will have to wait until some of my contractual obligations are exhausted (that would be a fair while, if I was strict about putting such matters first all the time).”

That’s the last I heard about it. Maybe an expert familiar with MAGICC can redo the IPCC’s projections for 1990-2000, substituting David Stern’s estimates of actual sulphur emissions for those tabulated in Appendix II). It’s a telling indication of the uncritical acceptance of IPCC assessments, including by 200 governments, that no one seems to have attempted this policy-relevant exercise (I should acknowledge that I don’t know how to do it - but I’m sure that wouldn’t be difficult for any competent modeller).

Geoff Larsen (Comment#1251)
March 21st, 2008 at 2:57 pm

It all seems very obscure.

Ian Castles appears to have a good grasp of the source of this. See his comment Mar 20, 2008, 6.36 PM on Prometheus, “Update on Falsification of Climate Predictions, Mar 15, 2008 (your perspective). (I see that Roger has posted while I wrote this- I will post it anyway; I hope he doesn’t mind).

He starts “The authors of this article misunderstood the basis of the ‘projections’ for 1990-2000 in the IPPC’s Third Assessment Report (2000).

And concludes ” So when Rahmstorf et al concluded that ALL OF the IPCC projections underestimated the temperature rise , they were not in fact evaluating the performance of models against observations- at least as far as the 1990- 2000 period was concerned (the greater part of the comparison). For this period, the output of the individual models were discarded in favour of an average of a subset of the models.

Contrary to the claims in Rahmtorf et al (2007), the analysis does not in fact represent an evaluation of the IPPC models for the 1990-2000 decade. For the period 2000-2020 the IPCC models projected an increase of above 0.2 C per decade as shown in Lucia’s graphic above.”

I first became aware of this graphic when it appeared a few weeks ago in the Australian Financial Review. It was copied from Ross Garnaut’s interim climate review (he is an advisor to the Australian Government on CO2 emission reductions). From the review I traced it to the Rahmstorf et al 2007 article.

My first impression was that it was nonsense. How could this IPPC temperature scenario, relative to actual temperature trends, be correct when it was last updated in 2001 and temperatures have flat-lined since then.

I appreciate your endeavour to quantify this.

Geoff Larsen (Comment#1252)
March 21st, 2008 at 3:11 pm

(I see that Roger has posted while I wrote this- I will post it anyway; I hope he doesn’t mind).

In my comment above I referred to Roger; sorry I meant Ian (Castles)

Ian Castles (Comment#1253)
March 21st, 2008 at 3:26 pm

Thanks Geoff Larsen. I agree that it’s not clear where the figure in Rahmstorf et al 2007 came from, and wonder why the reviewers for ‘Science’ didn’t ask for a precise citation.

Please note that the concluding sentence in my posting to Prometheus said that ‘For the period 2000-2020, the IPCC models projected an increase of about [not 'above'] 0.2 C per decade as shown in Lucia’s graphic above’.

steven mosher (Comment#1255)
March 21st, 2008 at 4:20 pm

Ian, in AR4 isnt A1T the high tech , low emissions scenario?

Ian Castles (Comment#1256)
March 21st, 2008 at 5:16 pm

Yes Steven Mosher, as I understand it the six emissions scenarios used in AR4 were the same SRES scenarios, with the same projections of emissions, as those that were used in the TAR - e.g. in Figure 9.14 reproduced above. And yes, the A1T scenario is the ‘high tech’ variant of the A1 scenario family, with fossil fuel CO2 emissions projected to be well below 1990 level by 2100. (Note however that under the SRES Terms of Reference the writing team were precluded from considering the effects of ‘any future policies that explicitly address climate change initiatives’ (e.g. emissions targets under the Kyoto Protocol).

After I had criticised some of the assumptions underlying these scenarios in letters to the Chair of the IPCC, Dr. Pachauri, in 2002, Dr John Mitchell of the UK Met Office) wrote to Dr. Pachauri on behalf of the WMO Working Group on Coupled Models to urge that the SRES scenarios be retained without change for the AR4. In the light of current controversies, the reasoning underlying this recommendation is worth quoting:

‘For each emission scenario, it is necessary to run an ensemble of simulations to define the uncertainty due to natural variability, and to do this with as many models as possible to define the range of uncertainty in modelling the earth system. This uncertainty means there is little scientific justification in running new scenarios since the resulting climate change outcome is unlikely to be indistinguishable (sic) from existing scenarios with similar radiative forcing. Hence, the WGCM unanimously urge IPCC to retain the current SRES scenarios without change… This will allow a better definition of the range of uncertainty in projected changes due to model uncertainty and natural variability, which are likely to dwarf any difference due to tweaking the existing emission scenarios’ (Letter of 30 October 2002 from Dr. John Mitchell to Dr. Rajendra Pachauri, tabled with other papers in B-28/Doc. 6 at 28th Session of IPCC Bureau, Geneva, 10-111 December 2002).

steven mosher (Comment#1261)
March 22nd, 2008 at 10:42 am

Lucia,

Consider this. The IPCC “forecast” “prediction” “projection” of .2C per decade was the MEAN of
several models. For example, 20 models. So the .2C per decade is an estimate of the
TREND MEAN of several models. And the 1 sig bars should then be 1 sig of the estimated TREND.

William and others sem to suggest that these trends do not include weather noise. Ok.
If one superimposed typcial weather noise on GCM trend estimates what you you get.

That is, At year 10. IPCC says the AVERAGE trend of all GCM is .2C ( -+ some TREND ERROR)

This error does not, if we believe william, include a weather error.

Fundamentally, what you see is that the confidence of the IPCC projection is being oversold

steven mosher (Comment#1262)
March 22nd, 2008 at 11:44 am

when anyone talks to you about these models that predict future “climate” whatever that is, you do this.

Link them to this page

http://www-pcmdi.llnl.gov/ipcc.....ummary.htm

Read the comments for every model

Here’s one:

Both Run 1 and Run 2 of PIcntrl were initialised from the end of a 120 year so-called adjustment control run with the full coupled model. Model year 121 renamed as 1871 to make it easier to match years to scenario runs

b30.026.ES01 ( yrs 410-549), b30.026b(yrs(550-699); Note: this 1%to4x run branched from the PDcntrl run at year 400, but CO2 was held fixed until year 410, when it began to increase.

here is a comment on changes made. How many engineers could get away with this scant description:

This run was initiated from another spun-up control (Picntrl-FUB, not available at PCMDI). It was initiated at year 199 of Picntrl-FUB and run for 6 years. The beginning of year 7 was then renamed 1860. PIcntrl differs from PIcntrl-FUB in that aerosol emissions were added, fixed at 1860 values, CO2 was changed from 277.28 to 286.20 ppm, CH4 changed from 722.82 to 805.60 ppb, and N2O changed from 310 to 276.69 ppb. These changes caused very little subsequent drift in the global mean surface temperature of this run (-0.0024 K per century).

GUESS which agency made this simple data storage error. GUESS.

Three additional years from this run (1/2000-12/2002) are available as the first 3 years of the SRESA1B simulation (where they were mistakenly stored).

THE CLIMATE SCIENCE VERSION OF VERIFICATION: here is a comment from the IPCC database of “consensus science”

Run 0 (not sent to PCMDI) is exactly the same as RUN 1 until year 1970, then it differs because of a bug reading the file of sulfate from year 1970 to 1975. We verify that the climate of year 2000 of RUN 0 is very close to climate of year 2000 of run 1.

HA. very close!!! But hey, it’s climate science.

From NCAR

B06.57 this run din not branch directly from the control run, but from January 1, 1890 of a ghg+sulfate+ozone run, which began in 1870 and branched from year 130 of Picntrl.

SO, The IPCC estimate of CLIMATE TREND ( warming per decade, I take it) Is a combination of the following

1. Modelling ERROR ( getting the physics slightly wrong)
2. Programming ERROR ( err, we didnt make any right???)
3. Numerical Error (hey, the chip is always correct, just dont ask it to divide)
4. Input Error ( somebody handed me an emissions scenario and I ran it through the program.)
5. Screw Ups ( err we didnt run that test, but this other test is good enough)
6. Weather: ( we dont do weather, we do climate)

I’m going to see if I can actually find some actual realizations of climate predictions in the database of the IPCC

steven mosher (Comment#1263)
March 22nd, 2008 at 6:28 pm

Lucia

http://www.cgd.ucar.edu/ccr/st.....ments.html

Judith Curry works on this model. I cannot make any sense of their data. Perhaps she can help.

Click on A1B line. This will take you to the model outputs for that scenario

steven mosher (Comment#1264)
March 22nd, 2008 at 6:29 pm

nice plots

http://www.cgd.ucar.edu/cms/st.....36a.ts.png

steven mosher (Comment#1265)
March 22nd, 2008 at 6:32 pm

http://www.cgd.ucar.edu/cms/st......IPCC.html

steven mosher (Comment#1266)
March 22nd, 2008 at 6:35 pm

Sorry, but the NCAR GCM ( part of IPCC) appears to do the PDO thing.

http://www.cgd.ucar.edu/cms/st.....0-2100.png

So maybe weather noise IS in the IPCC “projection”

Philip_B (Comment#1268)
March 22nd, 2008 at 10:14 pm

Steven, between 1 and 2 there is Design error, where the model (theory) is incorrectly transformed into a programmable algorithm.

There is also Test error, where the test data is in error and the code is modified (incorrectly) to produce outputs consistent with the erroneous test data. Its a common problem in commercial systems where the right answer is in most cases known (although not everyone knows the right answer). It must be a huge problem with climate models.

I am familiar with developing and testing software of the size of the climate models and know the climate models must contain large numbers of errors of various sorts, because there is no definitive data to test them against. Quite simply, it is impossible to say that any output of a climate model is correct in the narrow sense of correctly conforming to the intentions of the designers, never mind correct in predicting the temperature.

Ian Castles (Comment#1270)
March 23rd, 2008 at 9:16 am

SteveM, Your first url above (comment 1262) is to the ‘catalogue’ of what is officially known as the ‘WCRP CMIP3 multi-model dataset” - or, spelling out the acronyms, it is the product of phase 3 of the Coupled Model Intercomparison Project of the World Climate Research Programme. This is an activity of the Programme’s Working Group on Climate Modelling (WGCM), which successully pressed the IPCC to retain the SRES projections for the AR4 (see my comment 1256 above). The collection of model output for the project is coordinated by the PCMDI (Program for Climate Model Diagnosis and Intercomparison), which was established in 1989 at the Lawrence Livermore National Laboratory (LLNL).

The PCMDI’s homepage says that they ‘are applying [their] collective expertise to support modeling studies initiated by the … IPCC [by] providing facilities for the storage and distribution of terrascale data sets from multiple coupled ocean-atmosphere GCM simulations of present-day climate …’ It also states that ‘In part the WGCM organized this activity to enable those outside the major modeling centers to perform research of relevance to climate scientists preparing the Fourth Assessment Report … of the … IPCC.’

I infer that THIS is effectively the database that supports the predictions (including of global mean temperature) in the AR4.

The top half of the list consists of simulations of hypothetical climates assuming specified profiles of future CO2-equivalent concentrations, or holding concentrations unchanged at specified levels (eg those of the year 2000). The bottom half lists realisations of one or other of three of the SRES scenarios (A1B. A2 and B1). As the AR4 said that projected temperatures in the short term are not sensitive to differences in emissions, all of these realisations presumably provide alternative projections of the future real world.

On my count there are over 100 model runs (including 33 from NCAR and 17 from the Canadian Center for Climate Modeling and Analysis) in which the projection period begins in 2000 or 2001. I think the 0.2 C/decade increase in GMST projected for the early decades of the 21st century is derived as some sort of weighted mean of the model runs.

The purpose of running many models models was to allow a better definition of the uncertainty in projected conditions arising from (amongst other things) natural variability (see the Mitchell letter cited in comment 1256). I still don’t understand how Rahmstorf et al can conclude that the observations over recent years are at the upper end of the IPCC predictions. I’d draw the opposite conclusion.

lucia (Comment#1271)
March 23rd, 2008 at 9:46 am

Ian,
I keep looking at Rahmstorf and the TAR, and I find myself simply perplexed.

Reading Rahmstorf, the text appear to suggest they are citing IPCC Climate Change 2001: The Scientific Basis. Yet, dashed line “results” they describe in their paper do not seem to correspond to the dashed lines in the TAR. Rahmstrof is a brief paper, and so one would suppose they did not do anything very complicated. And yet, Rahmstorf seemed to be comparing observations to projections from the “tuned mini- models” tuned to a sensitivity of 3.0 C. In contrast, the TAR says:

To be consistent with Chapter 3, climate feedbacks are included and the model has been tuned to give results that are similar to those of the Bern-CC and ISAM models for a climate sensitivity of 2.5°C (Chapter 3, Figure 3.12).

This difference in wording would suggest that Rahmstorf is not comparing to the actual projections in the TAR. While comparisons to other things might be interesting, with regard to

Also, when I blow up the Rahmstrof figure and compare to the ones actually in the TAR I find the Rahmstrof figure seems to underpredict the TAR itself.

So, is Rahmstrof even commenting on the TAR? Or are they commenting on predictions of models that might have been chosen by the authors of the TAR, but which those authors elected not to use?

There are other pesky issues having to do with considering uncertainty in means due to ‘weather’ or ‘instrument’ noise.

It’s all quite odd.

steven mosher (Comment#1272)
March 23rd, 2008 at 10:45 am

yes, Lucia, as you push deeper and deeper into this bog it get’s quite smelly.

steven mosher (Comment#1273)
March 23rd, 2008 at 10:51 am

Ian, Yes you are correct in everything you say. A while back I tried to get access to the Database but was denied
perhaps I’ll try again.

One interesting thingto note about chapter 10 of AR4 is that there is is no simply available data back up
for the charts. It’s a very simple matter, but they refuse to do this

Roger W. Cohen (Comment#1276)
March 23rd, 2008 at 1:46 pm

Now you can see the intellectual dishonesty of the whole modeling business. Here are the symptoms, and I do mean disrespect

– Hindsight back-adjustments are combined with projections, with vertical adjustments for optimum spotting
– Never give a simple clean best estimate and a range of uncertainty for a given scenario
– Never give a clear stewardship review of how well models have done to date
– Never give a scientific basis for estimating model accuracy
– More runs from more models is the measure of progress (AR4, Summary p.12: “A major advance of this assessment of climate change
projections compared to TAR is the large number of simulations available from a broader range of models”)
– Projections are not forecasts nor are they predictions. So what are they? Like Marley’s ghost in a “A Christmas Carol” it is not
science, but things that might be if we do not change our ways.

Result: Anything that happens in climate is consistent with models.

rxc (Comment#1306)
March 24th, 2008 at 11:10 am

Mr. Cohen has it exactly correct. If these climate “models” are to be believed, they should be accompanied by detailed documents that (1) describe all of the important physical phenomena, ranked by importance by a peer-reviewed process, (2) present explicit explanations of the mathematical models (these are called “separate effects models”) used to represent those physical phenomena, also peer-reviewed, (3) show comparisons of the performance/prediction/forecasts made by those models, compared to actual physical data, including a description of whether the modeler knew the results of the experiments in advance, (4) explain how each of the individual physical models are combined, (5) show the results of experiments that have been predicted using the “integral model”. All of this must be described in the code documentation, and it must be completely open to scrutiny. And this list says nothing about the computer programming issues, related to whether the models converge properly, or whether the programming was done correctly. Verification and validation - did they program what they said they would program(verification), and did they program what they intended to program (validation)?

As I understand it, the climate models fall down on a number of these factors, and should be considered nothing but toys. Something that undergraduate students might come up with.

avfuktare Vind (Comment#1316)
March 24th, 2008 at 1:56 pm

I suspect that, apart from the input emission senarios, parametrizations are the major source of error. When doing the parametrizations one must assume what is feedback and what is cause. Make a mistake here and your parametre gets the wrong sign. As model output is primarily dependent on the parametres and not pure physics, and so little is known about how clouds works, how sea ice changes and how the biosphere works (etc), parametrizations is one of the major weaknesses in the AOGCMs. The other possible errors is dwarfed (or should be dwarfed if at least the models were verified…) by the order-of-magnitude bigger error margins in these two areas.

Raven (Comment#1361)
March 26th, 2008 at 2:10 pm

Tammy dedicated a post to this topic.

He raised an issue that you did awhile ago wrt the uncertainty in the autocorrelation parameter. It sounds legimate but you would eventually get to the point where any trend is consistent with the IPCC projections.

lucia (Comment#1362)
March 26th, 2008 at 2:28 pm

Raven:

Did Tamino compare the figure in Rahmstorf to the figure in the TAR? Do the grey regions match? Do the slopes match?