Propagation of Error and the Reliability of Global Air Temperature Projections
Guest essay by Pat Frank
Regular readers at Anthony’s Watts Up With That will know that for several years, since July 2013 in fact, I have been trying to publish an analysis of climate model error.
The analysis propagates a lower limit calibration error of climate models through their air temperature projections. Anyone reading here can predict the result. Climate models are utterly unreliable. For a more extended discussion see my prior WUWT post on this topic (thank-you Anthony).
The bottom line is that when it comes to a CO2 effect on global climate, no one knows what they’re talking about.
Before continuing, I would like to extend a profoundly grateful thank-you! to Anthony for providing an uncensored voice to climate skeptics, over against those who would see them silenced. By “climate skeptics” I mean science-minded people who have assessed the case for anthropogenic global warming and have retained their critical integrity.
In any case, I recently received my sixth rejection; this time from Earth and Space Science, an AGU journal. The rejection followed the usual two rounds of uniformly negative but scientifically meritless reviews (more on that later).
After six tries over more than four years, I now despair of ever publishing the article in a climate journal. The stakes are just too great. It’s not the trillions of dollars that would be lost to sustainability troughers.
Nope. It’s that if the analysis were published, the career of every single climate modeler would go down the tubes, starting with James Hansen. Their competence comes into question. Grants disappear. Universities lose enormous income.
Given all that conflict of interest, what consensus climate scientist could possibly provide a dispassionate review? They will feel justifiably threatened. Why wouldn’t they look for some reason, any reason, to reject the paper?
Somehow climate science journal editors have seemed blind to this obvious conflict of interest as they chose their reviewers.
With the near hopelessness of publication, I have decided to make the manuscript widely available as samizdat literature.
The manuscript with its Supporting Information document is available without restriction here (13.4 MB pdf).
Please go ahead and download it, examine it, comment on it, and send it on to whomever you like. For myself, I have no doubt the analysis is correct.
Here’s the analytical core of it all:
Climate model air temperature projections are just linear extrapolations of greenhouse gas forcing. Therefore, they are subject to linear propagation of error.
Complicated, isn’t it. I have yet to encounter a consensus climate scientist able to grasp that concept.
Willis Eschenbach demonstrated that climate models are just linearity machines back in 2011, by the way, as did I in my 2008 Skeptic paper and at CA in 2006.
The manuscript shows that this linear equation …
… will emulate the air temperature projection of any climate model; fCO2 reflects climate sensitivity and “a” is an offset. Both coefficients vary with the model. The parenthetical term is just the fractional change in forcing. The air temperature projections of even the most advanced climate models are hardly more than y = mx+b.
The manuscript demonstrates dozens of successful emulations, such as these:
Legend: points are CMIP5 RCP4.5 and RCP8.5 projections. Panel ‘a’ is the GISS GCM Model-E2-H-p1. Panel ‘b’ is the Beijing Climate Center Climate System GCM Model 1-1 (BCC-CSM1-1). The PWM lines are emulations from the linear equation.
CMIP5 models display an inherent calibration error of Ā±4 Wm-2 in their simulations of long wave cloud forcing (LWCF). This is a systematic error that arises from incorrect physical theory. It propagates into every single iterative step of a climate simulation. A full discussion can be found in the manuscript.
The next figure shows what happens when this error is propagated through CMIP5 air temperature projections (starting at 2005).
Legend: Panel ‘a’ points are the CMIP5 multi-model mean anomaly projections of the 5AR RCP4.5 and RCP8.5 scenarios. The PWM lines are the linear emulations. In panel ‘b’, the colored lines are the same two RCP projections. The uncertainty envelopes are from propagated model LWCF calibration error.
For RCP4.5, the emulation departs from the mean near projection year 2050 because the GHG forcing has become constant.
As a monument to the extraordinary incompetence that reigns in the field of consensus climate science, I have made the 29 reviews and my responses for all six submissions available here for public examination (44.6 MB zip file, checked with Norton Antivirus).
When I say incompetence, here’s what I mean and here’s what you’ll find.
Consensus climate scientists:
1. Think that precision is accuracy
2. Think that a root-mean-square error is an energetic perturbation on the model
3. Think that climate models can be used to validate climate models
4. Do not understand calibration at all
5. Do not know that calibration error propagates into subsequent calculations
6. Do not know the difference between statistical uncertainty and physical error
7. Think that “Ā±” uncertainty means positive error offset
8. Think that fortuitously cancelling errors remove physical uncertainty
9. Think that projection anomalies are physically accurate (never demonstrated)
10. Think that projection variance about a mean is identical to propagated error
11. Think that a “Ā±K” uncertainty is a physically real temperature
12. Think that a “Ā±K” uncertainty bar means the climate model itself is oscillating violently between ice-house and hot-house climate states
Item 12 is especially indicative of the general incompetence of consensus climate scientists.
Not one of the PhDs making that supposition noticed that a “Ā±” uncertainty bar passes through, and cuts vertically across, every single simulated temperature point. Not one of them figured out that their “Ā±” vertical oscillations meant that the model must occupy the ice-house and hot-house climate states simultaneously!
If you download them, you will find these mistakes repeated and ramified throughout the reviews.
Nevertheless, my manuscript editors apparently accepted these obvious mistakes as valid criticisms. Several have the training to know the manuscript analysis is correct.
For that reason, I have decided their editorial acuity merits them our applause.
Here they are:
- Steven Ghan___________Journal of Geophysical Research-Atmospheres
- Radan Huth____________International Journal of Climatology
- Timothy Li____________Earth Science Reviews
- Timothy DelSole_______Journal of Climate
- Jorge E. Gonzalez-cruz__Advances in Meteorology
- Jonathan Jiang_________Earth and Space Science
Please don’t contact or bother any of these gentlemen. On the other hand, one can hope some publicity leads them to blush in shame.
After submitting my responses showing the reviews were scientifically meritless, I asked several of these editors to have the courage of a scientist, and publish over meritless objections. After all, in science analytical demonstrations are bullet proof against criticism. However none of them rose to the challenge.
If any journal editor or publisher out there wants to step up to the scientific plate after examining my manuscript, I’d be very grateful.
The above journals agreed to send the manuscript out for review. Determined readers might enjoy the few peculiar stories of non-review rejections in the appendix at the bottom.
Really weird: several reviewers inadvertently validated the manuscript while rejecting it.
For example, the third reviewer in JGR round 2 (JGR-A R2#3) wrote that,
“[emulation] is only successful in situations where the forcing is basically linear …” and “[emulations] only work with scenarios that have roughly linearly increasing forcings. Any stabilization or addition of large transients (such as volcanoes) will cause the mismatch between this emulator and the underlying GCM to be obvious.”
The manuscript directly demonstrated that every single climate model projection was linear in forcing. The reviewer’s admission of linearity is tantamount to a validation.
But the reviewer also set a criterion by which the analysis could be verified — emulate a projection with non-linear forcings. He apparently didn’t check his claim before making it (big oh, oh!) even though he had the emulation equation.
My response included this figure:
Legend: The points are Jim Hansen’s 1988 scenario A, B, and C. All three scenarios include volcanic forcings. The lines are the linear emulations.
The volcanic forcings are non-linear, but climate models extrapolate them linearly. The linear equation will successfully emulate linear extrapolations of non-linear forcings. Simple. The emulations of Jim Hansen’s GISS Model II simulations are as good as those of any climate model.
The editor was clearly unimpressed with the demonstration, and that the reviewer inadvertently validated the manuscript analysis.
The same incongruity of inadvertent validations occurred in five of the six submissions: AM R1#1 and R2#1; IJC R1#1 and R2#1; JoC, #2; ESS R1#6 and R2#2 and R2#5.
In his review, JGR R2 reviewer 3 immediately referenced information found only in the debate I had (and won) with Gavin Schmidt at Realclimate. He also used very Gavin-like language. So, I strongly suspect this JGR reviewer was indeed Gavin Schmidt. That’s just my opinion, though. I can’t be completely sure because the review was anonymous.
So, let’s call him Gavinoid Schmidt-like. Three of the editors recruited this reviewer. One expects they called in the big gun to dispose of the upstart.
The Gavinoid responded with three mostly identical reviews. They were among the most incompetent of the 29. Every one of the three included mistake #12.
Here’s Gavinoid’s deep thinking:
“For instance, even after forcings have stabilized, this analysis would predict that the models will swing ever more wildly between snowball and runaway greenhouse states.”
And there it is. Gavinoid thinks the increasingly large “Ā±K” projection uncertainty bars mean the climate model itself is oscillating increasingly wildly between ice-house and hot-house climate states. He thinks a statistic is a physically real temperature.
A naĆÆve freshman mistake, and the Gavinoid is undoubtedly a PhD-level climate modeler.
The majority of Gavinoid’s analytical mistakes include list items 2, 5, 6, 10, and 11. If you download the paper and Supporting Information, section 10.3 of the SI includes a discussion of the total hash Gavinoid made of a Stefan-Boltzmann analysis.
And if you’d like to see an extraordinarily bad review, check out ESS round 2 review #2. It apparently passed editorial muster.
I can’t finish without mentioning Dr. Patrick Brown’s video criticizing the youtube presentation of the manuscript analysis. This was my 2016 talk for the Doctors for Disaster Preparedness. Dr. Brown’s presentation was also cross-posted at “andthentheresphysics” (named with no appreciation of the irony) and on youtube.
Dr. Brown is a climate modeler and post-doctoral scholar working with Prof. Kenneth Caldiera at the Carnegie Institute, Stanford University. He kindly notified me after posting his critique. Our conversation about it is in the comments section below his video.
Dr. Brown’s objections were classic climate modeler, making list mistakes 2, 4, 5, 6, 7, and 11.
He also made the nearly unique mistake of confusing an root-sum-square average of calibration error statistics with an average of physical magnitudes; nearly unique because one of the ESS reviewers made the same mistake.
Mr. andthentheresphysics weighed in with his own mistaken views, both at Patrick Brown’s site and at his own. His blog commentators expressed fatuous insubstantialities and his moderator was tediously censorious.
That’s about it. Readers moved to mount analytical criticisms are urged to first consult the list and then the reviews. You’re likely to find your objections critically addressed there.
I made the reviews easy to apprise by starting them with a summary list of reviewer mistakes. That didn’t seem to help the editors, though.
Thanks for indulging me by reading this.
I felt a true need to go public, rather than submitting in silence to what I see as reflexive intellectual rejectionism and indeed a noxious betrayal of science by the very people charged with its protection.
Appendix of Also-Ran Journals with Editorial ABM* Responses
Risk Analysis. L. Anthony (Tony) Cox, chief editor; James Lambert, manuscript editor.
This was my first submission. I expected a positive result because they had no dog in the climate fight, their website boasts competence in mathematical modeling, and they had published papers on error analysis of numerical models. What could go wrong?
Reason for declining review: “the approach is quite narrow and there is little promise of interest and lessons that transfer across the several disciplines that are the audience of the RA journal.”
Chief editor Tony Cox agreed with that judgment.
A risk analysis audience not interested to discover there’s no knowable risk to CO2 emissions.
Right.
Asia-Pacific Journal of Atmospheric Sciences. Songyou Hong, chief editor; Sukyoung Lee, manuscript editor. Dr. Lee is a professor of atmospheric meteorology at Penn State, a colleague of Michael Mann, and altogether a wonderful prospect for unbiased judgment.
Reason for declining review: “model-simulated atmospheric states are far from being in a radiative convective equilibrium as in Manabe and Wetherald (1967), which your analysis is based upon.” and because the climate is complex and nonlinear.
Chief editor Songyou Hong supported that judgment.
The manuscript is about error analysis, not about climate. It uses data from Manabe and Wetherald but is very obviously not based upon it.
Dr. Lee’s rejection follows either a shallow analysis or a convenient pretext.
I hope she was rewarded with Mike’s appreciation, anyway.
Science Bulletin. Xiaoya Chen, chief editor, unsigned email communication from “zhixin.”
Reason for declining review: “We have given [the manuscript] serious attention and read it carefully. The criteria for Science Bulletin to evaluate manuscripts are the novelty and significance of the research, and whether it is interesting for a broad scientific audience. Unfortunately, your manuscript does not reach a priority sufficient for a full review in our journal. We regret to inform you that we will not consider it further for publication.”
An analysis that invalidates every single climate model study for the past 30 years, demonstrates that a global climate impact of CO2 emissions, if any, is presently unknowable, and that indisputably proves the scientific vacuity of the IPCC, does not reach a priority sufficient for a full review in Science Bulletin.
Right.
Science Bulletin then courageously went on to immediately block my email account.
*ABM = anyone but me; a syndrome widely apparent among journal editors.
Pat,
This has already been explained to you numerous, so it’s unlikely that this attempt will be any more successful than previous attempts. The error that you’re trying to propagate is not an error at every timestep, but an offset. It simply influences the background/equilibrium state, rather than suggesting that there is an increasing range of possible states at every step. For example, if we ran two simulations with different solar forcings (but everything else the same), this wouldn’t suddenly mean that they would/could diverge with time, it would mean that they would settle to different background/equilibrium states.
@ and Then There’s Physics
I’m a layman and no mathematician but having read the first few pages of the paper it seems to me that your points are answering the wrong question. (?)
The point made, or so it appears to me, is that where there is uncertainty in the assumptions being made within a model then – if, as they should be those uncertainties are expressed and included within the model, as the time-steps are calculated then the uncertainty grows into a wide band with a diverging top and bottom spread of values. In other words they diverge.
If the uncertainties are not included as part of the model then surely it is linear and unable to produce meaningful results?
If you have multiple uncertainties, as in climate, which are input into a model then the spread or divergence must become even greater with time.
Some of those would seem to be (but far from limited too) temperatures and effect on on atmospheric water vapour levels; cloud formation and cloud cover; solar activity; volcanic activity etc etc. each would have an effect on some of the others and with a amount of uncertainty which would need to be expressed.
As I said, I am a layman and would appreciate it if you could enlighten me.
Thanks
The point made, or so it appears to me, is that where there is uncertainty in the assumptions being made within a model then ā if, as they should be those uncertainties are expressed and included within the model, as the time-steps are calculated then the uncertainty grows into a wide band with a diverging top and bottom spread of values. In other words they diverge.
Except this is not correct. An uncertainty only propagates if it applies at every step (i.e., if there is some uncertainty in the expected value at every step). If, however, some value is “wrong” by some amount that is the same at all time steps, then this does not propagate (by “wrong” I mean potentially different to reality). In this case, it is quite possible that the cloud forcing is “wrong” by a few W/m^2. What this would mean is that the equilibrium state would also then be “wrong”. It doesn’t mean, however, that the range of possible equilibrium states will grow with time, since this error does not propagate.
As I mentioned in the first comment, imagine we could run a perfect model in which every parameter exactly matched reality. Now imagine running the same model, apart from the Solar forcing being different by a few W/m^2. What would happen is that this would change the equilibrium state (there would be a constant offset between the “perfect” model and this other model). It would not mean that the difference between the model with the different solar forcing, and the “perfect” model would grow with time.
“An uncertainty only propagates if it applies at every step”
Um… No. A climate model is essentially an attempt to integrate a bunch of co-dependent variables numerically. If you knew anything about numerical integration, you would know that errors propagate wildly. The tool is, fundamentally, unsuited to the purpose it is being put.
ATTP, Stop focusing on the output result and thinking it’s error within an acceptable range therefore ok to propagate.
The issue is that the uncertainty that is propagated at each time step isn’t seen in the output because the output has been constrained by design to be within “reasonable” values. This is seen to be evidence the model is “doing the right thing” but the real problem is that at every single time step the output is meaningless for a climate calculation because the climate signal is much smaller than the error and what we’re left with is a fitted result.
Now you’ll arc up and suggest CGMs aren’t fits and are based on physics but again you’re mistaken because there are components (eg clouds) that aren’t and they’re approximations, they’re fits and by including them the models themselves are reduced to fits.
The whole GCM enterprise further relies on the assumption that errors cancel at each step throughout and that’s a ridiculous assumption. Completely unjustified and most certainly incorrect. In fact there is a small (unintentionally) built in bias that results in an expected result.
“imagine we could run a perfect model in which every parameter exactly matched reality.”
Yet you are UNABLE to run one where ANY parameter matches reality.
The ONLY thing you have is hallucinogenic anti-science IMAGINATION and FAIRY-TAILS
” If you knew anything about numerical integration, you would know that errors propagate wildly.”
I know lots about numerical integration (so does ATTP). I have spent a large part of my professional life doing it, in computational fluid dynamics, a regular engineering activity of which GCMs are a subset. Your statement is nonsense.
There’s no reason that difference would be equal over time. That’s a sign you have created a linear model, and it’s a decidedly non-linear system you’re modeling.
If this is what you think you guys are lost.
Micro,
I totally agree. This is probably the most telling comment of this whole discussion.
rip
ATTP, I was thinking about this more, you totally do not get that WV acts as a regulating medium, it actively alters the out going radiation response based on cooling temperatures, and not the stupid SB 4th pwr decay, this is on top of that, it’s the bends in the clear sky cooling profile. And since this is decidedly non-linear, and it controls the response to Co2, you’re not accounting for it in your models.
Think about how much the atm column shrinks at night. when it’s calm, it can only cool by radiation, and radiation is omni directional. Also for every gram of water vapor, there is a 4.21 J exchange of IR for a condense – reevaporation cycle as let’s say a 3,000 meter tall stack cools.
Interestingly it cools really quickly till air temps near dew point, then it stops cooling. It’s just there’s about -50W/m^2 of radiation to space just through the optical window based on SB calculations, yet net radiation is less than -20W/m^2. There’s about 35W’m^2 of sensible heat keeping the surface temp from falling as quickly.
There’s a 90F differences in the middle of the spectrum, I’ve measured over 100F differences.
How much energy is about 1 psi change between morning min T and afternoon Max temps at the surface(plus enthalpy lost, water condensed)? Oh wait, without the pressure change, average of about 3,300W/m^3.
attp says:
Funny, your example uses the ONLY independent variable in the whole shabang. Use any other co-dependent variable, and your example is busted.
Only in linear systems
In chaotic systems a single butterfly flapping its wings once….
…and that is a huge point. Climate models treat the climate as a linear system, because we do not have computational tools that can address the uncertainty of non linear systems.
To accept chaotic behaviour is merely to affirm ‘we can’t predict where this is going at all’. Or to put it in the vernacular. Climate science is at that level just bunk.
Even those people here who look for ‘cycles’ in climate with the ardent passion of ‘chemtrail’ observers, may in the end be barking up only a slightly less egregious gum tree than the climate scientists. Chaotic behaviour produces quasi-periodic fluctuations: That is over short times spans it may look briefly like a cycle, but then as it moves towards new attractors, it will enter a different ‘cycle’ and those of us who have built electronic circuits utilising chaotic feedback (super regenerative radios) know that, absent of a forcing signal, what you get is NOISE pure and simple, with no detectable single spectral component.
Nothing is more infuriating than to have someone lecturing you on the characteristics of linear equations challenging you to disprove their finer points, when your whole position is predicated in a provable assertion that what is being modelled cannot be represented by linear equations in the first place.
“Except this is not correct. An uncertainty only propagates if it applies at every step (i.e., if there is some uncertainty in the expected value at every step). If, however, some value is āwrongā by some amount that is the same at all time steps”
incorrect, the value increases with each step over time, you are a completely anti scientific chappy, clueless
Typical attp tactic, start off with a lie then spin sophistry round your false strawman.
Micro thanks for exposing ATTP’s cut and paste knowledge. Once you get in depth with him, he vanishes every time and runs back to his echo chamber
this annual average Ā±4.0 Wm-2 year-1 uncertainty in simulated LWCF is approximately Ā±150% larger than all the forcing due to all the anthropogenic greenhouse gases put into the atmosphere since 1900 (~2.6 Wm-2), and approximately Ā±114Ć larger than the average annual ~0.035 Wm-2 year-1 increase in greenhouse gas forcing since 1979
The error DOES in my opinion propagate.
And Then There’s Physics says If, however, some value is āwrongā by some amount that is the same at all time steps, then this does not propagate.
If the correction for cloud fraction error was a simple linear adjustment to models to correct the error, we would never have known about it. The adjustment would have been applied, and the model prediction would have aligned with observed cloud fraction.
Since nobody can accurately predict how clouds respond to GHG forcing, the margin for error grows with every iteration step, The uncertainty of how clouds will respond to the GHG forcing applied in a single step has to be carried through to the next iteration.
When the margin for error drastically exceeds what is physically plausible, I think we can safely assume the predictions of the model are total nonsense.
Page 23 of Pat Frank’s paper, hindcast cloud fraction error of global climate models.
ATTP says,
This assumes a linear response. It assumes that climate (and thus, presumably, weather) is a linear function of forcings.
If the initial value is āwrongā by some amount – or inaccurate by some amount – then that will affect the next iteration in some way.
If the next iteration is affected by the same amount every single time then the response is always constant.
Once again we have pseudoscience pretending that clouds don’t exist. That phase changes (water vapour to water droplets, for example) are smooth.
Why does ATTP worry about a declining Arctic Icecap when he doesn’t believe in non-linear phase changes? Melting can’t exist in his understanding of climate!
Except he has no understanding. He’s just a climate fanatic. It’s faith, not science.
Youāve got the essence, Old England, āthat where there is uncertainty in the assumptions being made within a model then ā if, as they should be those uncertainties are expressed and included within the model, as the time-steps are calculated then the uncertainty grows into a wide bandā¦
You have grasped the central point that continually eludes ATTP and virtually every single climate modeler.
The error is systematic, resident in the model, and is introduced into a simulation by the model itself. It enters every simulation time-step, and necessarily produces an increasing uncertainty in the projection.
Look at ATTPās reply to you.. His āwrong by some amountā supposes a constant offset error and is a completely wrong description of the systematic error.
Look at manuscript Figure 5. Every single model makes has a different error profile with positive and negative excursions. I pointed this out to ATTP in prior conversations. He ignores it. Perhaps because he doesn’t understand the significance. Change the parameter set of any one model, and its error profile will be different.
But ATTP (and others) want to add up all the errors to get one number, and then assume that number is a constant offset error that will correct any model expectation value to be error-free. His (their) idea is beyond parody.
Then he goes on to suppose statistical uncertainty is physical error, i.e., ATTP: āthe range of possible equilibrium states will grow with time, since this error does not propagate,ā
ATTP makes a standard mistake of my reviewers, here specifically number 6, but he has already also made mistakes 4, 5, 7 and 8.
He makes those same mistakes over, and over again.
TimTheToolMan gets it right, as usual.
Tim, do you have any idea why uncertainty is so opaque to climate modelers?
Itās dead obvious to any experimental scientist or engineer.
In this post, Nick Stokes admitted that GCMs are engineering models. I.e., Nick: “a regular engineering activity of which GCMs are a subset.”
Engineering models are useless outside their calibration bounds. Nick has repudiated the entire global warming scary-2100 enterprise.
Yet another inadvertent validation in an attempted refutation. Thank-you, Nick.
Eric Worrall, your comment is right on.
Thanks for posting Figure 5. It shows that every model has a different error profile, with positive and negative excursions.
Mere inspection of the figure shows how ludicrous is ATTPās idea that all those errors should be merely added together into a number. And then subtracted away to make everything accurate. Only in consensus climate science.
“Engineering models are useless outside their calibration bounds.”
So what are the “calibration bounds” of, say, Nastran? Or Fluent, or Ansys? Pat, you don’t have a clue about engineering models.
well it’s obvious you don’t understand this.
Calibration isn’t defined by the simulator, but the models as applied to the design you’re evaluating. And it’s in comparison to the real circuit in operation.
Pat writes
I dont think it is. I think even Nick gets it and one day might even accept it (no Nick, it doesn’t mean your CFD work is dead, or weather models are wrong – models have their place still!) but no climate modeler can admit to it because it’d be well…a career limiting move. And as the GCMs are the cornerstone to so much of our science today, untangling the mess would be horrendous. Better to let sleeping dogs lie.
Pat Frank,
Look at manuscript Figure 5.
Ok, it shows latitudinal profiles of 25-year averaged model cloud fraction error versus cloud fraction observations averaged over a similar timescale. It demonstrates latitudinal error offsets between models and observations, as well as showing differences between models.
Every single model makes has a different error profile with positive and negative excursions.
Yes, this is well known and clearly understood by ATTP. Different models, different offsets.
Change the parameter set of any one model, and its error profile will be different.
Yes, this would obviously be true but how is it relevant to error propagation within a projection? Within an individual model projection run the parameter set will remain the same, thereby maintaining the same offset error.
Put in context of your Figure 5, your error propagation suggests that those error profiles should change quite dramatically over time. Why would that happen?
Nick Stokes, no matter your diversionary sneering, I’m clued in enough to know that engineering models are unreliable outside their calibration bounds.
Paulski0 āWithin an individual model projection run the parameter set will remain the same, thereby maintaining the same offset error.ā
Not correct, for two reasons. The parameters are not unique. They have large uncertainty widths. One can get the same apparent error with different suites of parameters. A given error is just representative. It does not transmit the true range of model errors. The uncertainty is made cryptic unless this is taken into account.
Second, even with unchanging parameter sets, any given projection simulation step is wrong, but to some unknown amount. Those wrong climate states are projected forward. Every step begins with initial value errors.
The projection error from step to step therefore varies, and in unknowable ways.
In a futures projection, one canāt know the errors. One only knows the uncertainty, by way to the propagated calibration error statistic. And uncertainty grows with each projection step because of increasing ignorance of the relative positions of the simulated state and the correct physical state in phase space.
Error propagation says nothing about error profiles in projection simulations. It addresses the reliability of the projection. Not its error.
I think you and the reviewers may be missing the point here attp. Millions spent on building climate models…and a simple linear model can recreate them very closely…..surely this is worth publishing and worth investigating further?
A simple linear model is what climate models are, stripped of decorative complexity, but whilst models may be represented by a model of that nature, reality it seems is just too complicated for that class of model to have a snowballs chance in hell of representing the vagaries of actual climate.
So I dont know what you are saying, but its not worth spending a copper nickel on.
I looked into cutting edge attempts by seriously bright mathematicians to even discern whether a given set of non linear partial derivatives led to a bounded set of solutions (broadly, a climate that never goes below snowball earth or boils the oceans dry) and we cant even do THAT. Observationally climate is amazingly stable.
But wobbles a lot as well.
And we have absolutely no idea whether it could one day wobble off to a whole new regime, just because a butterfly flapped its wings, let alone by injecting tons of CO2 into it. All we can say is that in times gone by, when CO2 was way greater than it is today, or is likely to be in the foreseeable future, the climate seems to have been stable enough for life to flourish.
The state of climate change science stripped down to the actual science, which is almost none, is simply stated
1/. We don’t know.
2/. Even if we did know the partial differentials governing it, we still wouldn’t know what the climate will do.
3/. We lack both the mathematics and the computational power to ever know better than that.
4/. Climate change is therefore not worth spending any grant money on.
5/. Even WUWT has no function beyond pointing out points 1, 2, 3 and 4.
6/. The IPCC is an organization without any purpose, since it exists to advise governments on situations that have no existence in reality.
7/. Renewable energy is therefore a crock of excrement, a pointless waste of money.
8/. Anyone who disagrees with any of the points above is like a holocaust denier.
9/. There is an urgent need to set up an international organisation to help whole swathes of the population come to terms with the facts that:
– the cheque isn’t in the post
– the tooth fairy doesn’t exist
– he/she wont love you in the morning.
– ‘man made climate change’ is as real as Tinkerbelle.
Great post, Leo Smith. Your number 1 has been the conclusion of my AGW assessment from the first. š
If only you were head of the US National Academy. Or Pres. Trump’s science advisor. š
Another point, that I think I’ve made to Pat before, is that if he is correct he should be able to easily demonstrate this. If you’re running computational models, one way to estimate the uncertainty is to simply run them many times with different initial conditions. If the uncertainty propagates as Pat suggests, then the range of results should reflect this. As I understand it, this has been done, and they do not.
This is clear even from the published CMIP5 simulations. Pat Frank claims that the error arising from cloud uncertainty alone should accumulate to an extent of Ā±16°C by 2100. And he seems to infer the cloud error from disagreement between the models. But the CMIP5 models clearly do not diverge by 16°C by 2100. Here is a plot
The spread is mainly due to the different scenarios; for an individual scenario it is maybe Ā±0.6°C.
Yes Nick
Hundreds of scam CO2-hatred “scenarios”.
NOT ONE anywhere near REALITY.
Thanks for drawing that to everybody’s attention.
Nick
And that’s your argument to establish the effectiveness of the models? Really
That’s not true Andy! Give Nick his due. ONE of those models is quite close to reality – the one at the very bottom. The rest of the models should clearly be fired. But that one should be given a prize, and it shows that temperatures in 2100 will be about the same as today. So according to the one believable model, there is no C in CAGW, and no real W either. Great! Can we all pack up and go home now? And stop wasting money on this nonsense?
@aTTP (1:24am) and Nick Stokes (2:36am)
It seems the issue is not in getting different results with different initial conditions but rather running slightly different models from the same initial condition.
The simplest setup would be to select a single tunable parameter (e.g. clouds), vary the value up or down to create 2 model formulations, and run them both from the same initial conditions. The different values may cause divergence or other feedbacks/interactions may dampen it to insignificant.
If I understand the source of Nick’s spaghetti graph, the graph demonstrates the differences between models, not the potential uncertainty inherent in any one model. Each spaghetti line has it’s own uncertainty band that it not displayed.
MJB,
Yes, you could also do what you suggest (i.e., run with the same initial conditions, but different parameters). If we consider clouds, then there is probably a range of a few W/m^2. This would correspond to potentially difference of about 1K; not even close to the +-15K suggested by Pat Frank.
As far as the spaghetti graph is concerned, I think it is a combination of individual models run more than once and different models, so you are correct that it isn’t a true uncertainty. However, it does illustrate that the range is unlikely to be as large as suggested by Pat Frank.
If constraints are being applied for each calculation then you are not getting modelled outputs but constrained outputs. Do the runs with no constraints to see the inherent validity of the underlying physics, not the hand-tailoring needed to sell a story.
But hey, if the need is to sell a story….
I find it hilarious that Nick and others still think the chart at this comment is relevant since it is pseudoscience crap:
https://wattsupwiththat.com/2017/10/23/propagation-of-error-and-the-reliability-of-global-air-temperature-projections/#comment-2643766
How can anyone think wild guesses to year 2100, be considered good science,when most of it UNVERIFIABLE! Models are a TOOL for research,not to create actual fact based science,since it lacks real data for the next 83 years. This is the what the AGW conjecture is based on,a puddle of unverifiable guesses,
Bwahahahahahahahahaha!!!
Imagine what real Meteorologists, who do short term modeling for weather prediction in the next few days know how quickly short term predictions can quickly spiral out of reality. I see them adjusting their forecasts daily,sometimes even in hours, as new information comes in,but can still be waaaaay off anyway,as they were in my city just yesterday.
Models are a TOOL for research,not a creator of data.
…and Then There’s Physics ,
You said, “However, it does illustrate that the range is unlikely to be as large as suggested by Pat Frank.” I’m not sure that you can justify that statement. The propagation of errors provides a probabilistic uncertainty range, which is an upper bound, not the most likely outcomes. That is, with numerous ensemble runs, they are most likely to cluster around the most probable values, but that doesn’t preclude them from sometimes reaching the maximum values if a large enough number of runs are made.
Extrapolating the apparent arc of the upper limit from the spaghetti plot of model runs, you reach a maximum divergence value of approximately 8.5K to 9.5K truly slightly more than half the 15K to 16K suggested
Clyde,
Normally what’s presented are 1, or 2, sigma uncertainties. This would mean that either about 66%(1 sigma), or 95% (2 sigma), of your results should lie within this range. Depending on what is presented, you would expect either 1/3 of your results (1 sigma), or 5% of your results (2 sigma), to lie outside the range. Therefore, if you ran a lot of simulations and the results never ended up outside the range, then the range would probably be too large.
Bryan,
Except, the range is mostly because of the range of emission scenarios, rather than scatter for a single scenario. Therefore, the overall range isn’t representative of some kind of model uncertainty.
sun: the RCPs aren’t “guesses,”
they’re assumptions.
Crackers,when they run to year 2100, they are indeed wild guesses,since there is ZERO evidence to support it, you are playing word game here. They are unverifiable,can’t run a hypothesis on it since most of it is far into the future,thus qualifies as wild guesses.
He writes,
“sun: the RCPs arenāt āguesses,ā
theyāre assumptions.”
Yawn, is this how low science literacy has fallen?
The difference between an assumption and a guess is basically the reputation of the person making them.
Fenchie77,
You are of course correct. The fact that the models require constraints is enough to invalidate them.
I doubt the is a Mechanical Engineer in the crowd that would trust his/her family’s safety to a 5th floor apartment deck that was designed with, or the design was verified by, a stress analysis (i.e., modelling) program that required constraints be placed within it to keep the calculations within reasonable ranges.
ATTP, “ one way to estimate the uncertainty is to simply run them many times with different initial conditions.”
No, it’s not. Your proposed method tells one nothing about physical uncertainty.
Mistakes 1, 3, 4, 6, and 10. Good job, ATTP.
Nick Stokes, “Pat Frank claims that the error arising from cloud uncertainty alone should accumulate to an extent of Ā±16Ā°C by 2100.”
No, I don’t Nick. You’re proposing that Ā±16Ā°C is a physically real temperature.
It’s an uncertainty statistic. An ignorance measure. It’s not physical error.
You’ve made mistakes 2, 6, 11 and, implicitly, 12.
You’ve many times now demonstrated knowing nothing about physical error analysis. Now it’s many times plus one more.
Pat,
Hold on. You’re suggesting the results from the models are far more uncertain than mainstream climate modellers suggest and yet you’re also suggesting that if you ran the models many times (with different initial conditions and using different parameter values) you would get an overall result that was not representative of the uncertainty. This doesn’t seem consistent.
Nick Stokes
October 23, 2017 at 2:36 am
My comment to you it is not actually about the particular point you trying to make there, but more in the aspect of contemplating the validity of the whole argument in question here about GCMs.
You see, you have a clear beautiful plot there, but really no much relevant, as it does not have the according ppm concentration trends also.
Last time I checked AGW is all about temps as per ppm…….and the correlation there…..
Ignoring this actually puts one in the position of misinterpreting the value of GCMs as an experiment…..either intentionally or not.
So when the nice plot you posted helps with your point, maybe, in its essence misleads towards a result of misinterpretation and confusion about the actual value of GCMs as an experiment, which by the way are not climate models anyhow, and very very expensive experimental tools at that.
Don’t you think that the plot you provide, the way it stands has no much support value about the RF or the fCO2 as contemplated by the AGW hypothesis one way or another?
cheers
“Thatās not true Andy! Give Nick his due. ONE of those models is quite close to reality ā the one at the very bottom.”
yeah, predict 1 2 3 4 5 6 and throw a dice, and one will be right.
Logic is not you nick or ATP
Idiots pretending to be scientists. Why not get English lit Mosher in on the act too
or some more of pseudo science sensitivity studies that are nothing but tuned junk driven by observations
ATTP
“Pat,
Hold on. Youāre suggesting the results from the models are far more uncertain than mainstream climate modellers suggest and yet youāre also suggesting that if you ran the models many times (with different initial conditions and using different parameter values) you would get an overall result that was not representative of the uncertainty. This doesnāt seem consistent.”
It’s not inconsistent.
The models are far more uncertain than claimed, because of 1 much comes from hindcast tuning, not physics (not incomplete and some much not understood physics) Unless you are going to be uber absurd and claim that is not true
The range outcomes is uncertainy (in model physics which leas to instability (not variability)), error, and different tunings.
as with Mosher, logical examination is not for you, as usual, add Nick in there
However if you replace the linear models by non linear ones, the behaviour is exactly as he describes.
It is not the coherence of linear models that is under criticism, it is their applicability at all.
It is of no use to refute the fact that your cat scratched my leg, by pointing out that dogs just dont do that.
ATTP, physical uncertainty is with respect to physical reality, not with respect to model spread.
You’re conflating model precision with model accuracy (mistake #1). You make this mistake repeatedly. So do climate modelers. You all seem unable to grasp the difference.
Running a model over and over, with different initial conditions, tells you nothing, nothing, about physical uncertainty (mistake #3).
Unless (BIG! unless here) your model is falsifiable and produces physically unique predictions.
Climate models violate both conditions.
Run them until you’re blue in the face, and you’ll have learned nothing except how they move around.
Nick Stokes October 23, 2017 at 2:36 am
Nick, can you extend your chart so we can see how high the projections go for RCP 8.5? The chart cuts them off at the year ~ 2080.
I suspect that if you increase the vertical axis and in addition, include the uncertainty surrounding each run, you will end up with roughly the range suggested by the author of this post.
Think!?
A never believable claim from confirmed liars or misdirection specialists.
If you believe your falsehood, write up a mathematical article and publish it.
Until then, your belief is just so much speculation.
Without proof or logic.
sun says ‘when they run to year 2100, they are indeed wild guesses,since there is ZERO evidence to support it’
no, they’re assumptions, not guesses.
there can be no evidence from
the future, only assumptions.
a model has to assume path of future emissions.
these
are the RCPs.
there are four of them for different
scenarios of future energy
use.
unless you can predict for us
that future path. go ahead and try.
So much for contributions from Nick.
What are the starting uncertainties in climate models, Nick?
Technically, adjusting a temperature record is an immediate admission of error and even roughly identifies the error range.
Yet, not one of the models initializes with that one uncertainty or propagates it through.
Gross assumptions regarding total lack of temperature equipment calibration or certification
Total lack of side by side measurements before swapping equipment.
Total lack of side by side measurements before moving the temperature station.
Total failure to track temperature station infestations or to identify errors caused.
Instead, Nick apparently espouses averaging temperatures repeatedly to accurize numbers and improve precision.
Run the models many times…
A solution that is far worse than claiming stopped clocks are correct twice a day.
The sample standard deviation (SD) in a statistical sense is only meaningful if the underlying population is normally distributed; The percentage of values claimed to fall within some error window depends on the shape of the population distribution. If instead you are talking about the standard deviation of a sampling distribution of a summary statistic (such as the mean), then the central limit theorem is invoked to adopt the assumption that the theoretical sampling distribution of that summary statistic (which you are sampling from) is distributed Normal. The standard error (SE) is the sample estimate of the standard deviation of that sampling distribution.
If the SE (or sometimes the SD though far less likely) is used to support a statement of confidence about the population parameter, such as the mean, then the correct confidence statement is that the error window has some x chance of encompassing the population parameter. Again, assuming a Normal distribution. The notion that the one confidence window you calculate will contain x percent of ‘the data’ or sample statistic should I run the process over and over again is incorrect. Each time you sample, both the mean and the SE vary, and as such so will any confidence statement drawn from the sample statistics.
The proper statement of interpretation of confidence (or uncertainty) is that, in the long run of N (very large) samples of size ‘n’, my ‘x level of confidence’ error windows will capture the population parameter x percentage lf times.
Error propagation is different altogether. Different formulae, and they also depend on what operations you are performing on your data.
Generating an error bar from a large collection of predictions from different models and even, within each model, varrying the initial conditions is an ad hoc method to generate error intervals. It seems supremely niave to believe that varrying these things will happen to capture the uncertainty in the accuracy of the coefficients and values of the model parameters for any coefficient or parameter value that itself posseses some none neglibeable and varied amount of uncertainty associated with them.
Even in a bivariate linear regression model, Y = B1X1 + B2, there is uncertainty in the prediction of Y (y’) and uncertainty in the estimate (b1) of the B1 coefficient and the estimate (b2) of the B2 intercept and, often times, even uncertainty in the observations (x) of X used to generate the model in the first place.
Suppose we sample from a linear system. We don’t know it but the X’s in our model are all appropriate in explaining Y. Good for us so far. But we don’t know what the exact values for X are. So, we sample Y, we also sample X1 to Xk, we then crunch the numbers (do the regression) and come up with the estimates (b1 to bk) of the coefficiemts (B1 to Bk). Thus, we now have a model. The accuracy of our measurements of Y and X1 to Xk (and, normally, the appropriateness of our X1 to Xk in explaining Y but, again, here we are assuming they are appropriate) will help determine how well this model actually does in explaining and predicting Y. Y is unknown as are (probably most) of the true values of X – presumably Time (year) would be one of them. Our measurements of Y and X1 to Xk (y and x1 to xk) are, for the most part, all we have, but we based our model off of the measurements. There are uncertainties in the measurements. We don’t know the direction of those errors or their magnitude (offsets as i think is being used above), because we don’t know the relation of the measurements to their true value. These errors will propagate as the model is run iteratively, being fed its own outputs as inputs at each iteration.
Tweaking the estimated values of X1 to Xk and b1 to bk to generate different estimates of Y (y’) is an ad hoc attempt to quantify this additional uncertainty in X and Y through ’empirical’ simulation.
Pat Frank well-approximates the model temperature outputs using a simplified linear equation. He then focuses on the effect of cloud coverage on solar insolence (if memory serves) and (presumably) uses error propagation formulae to quantify the effect of this uncertainty in the estimate of temperature.
There is either a theoretical/mathematical explanation for why error propagation does not apply or there isn’t and the modellers technique for evaluatiom is gravely misguided.
Pat Frank October 23, 2017 at 9:36 am
Nick Stokes, āPat Frank claims that the error arising from cloud uncertainty alone should accumulate to an extent of Ā±16Ā°C by 2100.ā
No, I donāt Nick. Youāre proposing that Ā±16Ā°C is a physically real temperature.
Itās an uncertainty statistic. An ignorance measure. Itās not physical error.
Youāve made mistakes 2, 6, 11 and, implicitly, 12.
Youāve many times now demonstrated knowing nothing about physical error analysis. Now itās many times plus one more.
_________________________
What is it they say Pat, a little knowledge is…. š
At least Nick might run off now and try understand physical error analysis, seems the sort that does not like understanding things š
* Does not like Not understanding things.. heh, wish I could edit my stupidity instead of posting again š
“the effect of cloud coverage on solar insolence”
Busy old fool, unruly sun
Much as i hate to chip in, in support of both Nick and aTTP. They are giving you accurate information. If the models were wrong in the ways described above…. they would be “more” wrong and it would be very obvious to even the most committed warmist modeller. All models are wrong, it’s inherent in modelling. Some are really, really wrong. But most of the ones in active use are not. I would agree that the current crop run hot and im not a massive fan of zekes recent work, trying to show that they dont. But we have apply healthy scepticism and critical thought to all of this. We cannot push that all to one side because we simply like the sound of what’s being said. Mosher, to his previously sceptical credit makes that point often. He sometimes at least recently doesn’t take his own advice. But i guess we are all guilty of that.
Depending on your nationality there’s always the PNAS route to publishing. Pal reviews can cut both ways.
Cracker,
Assumption
“a thing that is accepted as true or as certain to happen, without proof.”
Guess
“estimate or suppose (something) without sufficient information to be sure of being correct.”
Meanwhile you keep playing word games while I keep saying they are junk,you never disputed that they are junk.
I stated:
“Crackers,when they run to year 2100, they are indeed wild guesses,since there is ZERO evidence to support it, you are playing word game here. They are unverifiable,canāt run a hypothesis on it since most of it is far into the future,thus qualifies as wild guesses.”
and,
“How can anyone think wild guesses to year 2100, be considered good science,when most of it UNVERIFIABLE! Models are a TOOL for research,not to create actual fact based science,since it lacks real data for the next 83 years. This is the what the AGW conjecture is based on,a puddle of unverifiable guesses,”
You have NOTHING to sell here.
You are pathetic.
Nick by eyeball, the spread is eight + from the smudge at ~0C in 2100 to the topmost steeply exiting the top of the graph at about 2075. And these represent the models that survived the cut. You would still be wrong with a linear model but more difficult to criticize had you guys not been charged with the task by Grouchmarxist highschool drop out Maurice Strong (creator of both UNFCCC and IPCC) to find burning fossil fuels will destroy the planet, thereby justifying trashing economies and freedoms and having global governance by elites. Models vs observations to date show climate sensitivity to be at most ~1, but this takes the scare out of rising CO2.
I’m thinking we should crowd source a large fund and place a bet that with the collapse of the Paris agreement we will not achieve a rise of 1.5C going gangbusters with fracking oil and gas, burning coal, making concrete, etc. If we haven’t got over halfway there by 2050 we declare a win and make the fund available to third world economies for developing cheap reliable electricity generation. Honesty in temperature collection would need some resources and oversight.
“Nick by eyeball, the spread is eight + from the smudge at ~0C in 2100 to the topmost steeply exiting the top of the graph at about 2075.”
The spread for each scenario is much smaller. The fact that scientists don’t know what will be done about GHGs and have to cover the range of possibilities has nothing to do with error propagation. But there is a real test of PF’s ridiculous errors. Ā±15°C would be about Ā±9°C in the 30 years since Hansen’s prediction. Now we quibble about small fractions of a degree difference in scenarios, and another small fraction that might be a transient for El Nino, but there is nothing like a 9°C error.
Nick Stokes first thinks uncertainty is physical error (mistake #6), and then effortlessly moves on to suppose it’s a physical temperature instead (mistake 11).
Nick’s self-contradictory assignments also implicitly embrace mistakes 2, 4 and 12.
Clyde Spencer it’s even worse than that, because the cloud forcing error is inherent in the model and is systematic.
That means one never knows the most probable value.
ATTP supposes that model precision is a measure of reliability.
Mistakes 1 and 3.
blunder bunny wrote, “they would be āmoreā wrong and it would be very obvious to even the most committed warmist modeller.”
Not correct. GCMs are tuned to give a reasonable projection. That practice hides physical error and side-steps uncertainties.
Mark – Helsinki I can’t offer a rationale for it all. š
RW I can’t add anything to your thoughtful post, but can mention that,
Vasquez, V. R., and W. B. Whiting (2006), Accounting for Both Random Errors and Systematic Errors in Uncertainty Propagation Analysis of Computer Models Involving Experimental Measurements with Monte Carlo Methods, Risk Analysis, 25(6), 1669-1681, doi: 10.1111/j.1539-6924.2005.00704.x.
assess random and systematic errors in nonlinear numerical models and recommend propagating systematic model error as the root-sum-square.
The precedent of that paper, by the way, encouraged me to make Risk Analysis my first journal for submission. The rest is history. š
Nick Stokes “But there is a real test of PFās ridiculous errors. Ā±15Ā°C …”
That’s not physical error, Nick.
Mistakes 4, 5, 6 and 11, and probably 12 implicitly.
Well done. š
Why don’t people understand uncertainty? They taught us that in first year physics. I could easily make a model that only ever has one outcome, but if it propagates an uncertain value then the error bars will be huge by the end. That doesn’t mean my model will ever show that, thus conflating model precision with uncertainty. The error bar means that my model could be wrong by that much. Of course if your model is wrong it won’t tell you, that’s the whole point of error bars.
Pat Frank, thanks for the reference. I have downloaded it and will check it out.
From ..and Then There’s Physics October 23, 2017 at 9:36 am
“youāre also suggesting that if you ran the models many times (with different initial conditions and using different parameter values) you would get an overall result that was not representative of the uncertainty.”
But if the model would indeed propagate the suggested systematic, “throughout”, physical error, it *will* be noticed. The current models are not sufficiently taking into account non-linear effects of known modelling errors. This then causes the accuracy of the model to decrease rapidly with each time-step and explains perfectly the issues seen today comparing measurements with runs of 10-20 years ago.
Many climate scientists seem to make the same mistake simply because they continue to apply tools without allowing rigorous review of the validity of using those tools that way. This is a larger systematic *human* error in that particular field. And it’s not the first time in recent history but certainly becoming the most costly. And the cause of it lies within underlying role of politics, money and emotion, which has grown into something big to “fail”. The cure here is “back to basics”: re-examination of the toolbox itself.
Jarryd Beck, thank-you. š
It seems to me that training in climate modeling completely neglects physical error analysis. Not one climate modeler I’ve encountered has a clue about it. And they’re often hostile to it.
I like his self declared hero status after his sixth rejection, obviously due to the corrupt system and fear of what the analysis would unleash – no other explanation possible here.
The fact that the insiders circle the wagons when criticized is proof that the criticism is meritless.
Gotcha.
“Imagine what real Meteorologists, who do short term modeling for weather prediction in the next few days know how quickly short term predictions can quickly spiral out of reality. I see them adjusting their forecasts daily,sometimes even in hours, as new information comes in,but can still be waaaaay off anyway”
This is a big reason why us real operational meteorologists (for 35 years) have such a high % of skeptics vs in other sciences. We must constantly reconcile the forecast with realities. Quickly adjust based on models that also quickly dial in new/fresh data and come out with a new scenario that can sometimes look much different than the previous one………..with errors/changes often growing exponentially with time.
Individual ensemble members of the same model can look completely different beyond a week. Different models in week 2 can have very different outcomes, not just regionally but in the position of many large scale features that define the pattern.
However, despite this, climate models are much different and they are not as effected by the random, chaotic short term fluctuations in initial conditions that can never be captured perfectly and lead to exponentially growing errors with time.
For instance, if the amount of solar forcing in a climate model was too high/low, one would not expect it to result in output/projections that amplify exponentially over time. It would remain pretty much constant. There would also be potential negative/positive feedbacks but they would be limited and probably not greater than the error from the solar forcing being too high/low.
Another difference. With weather models, we change them/equations every several years or so to make potential slight improvements, with experimental models constantly being run and compared to the existing models……..with mixed results.
I am in not involved in modeling but it seems clear that certain models are superior than others, especially when it comes to handling particular atmospheric dynamics. However, the gatekeepers of all models seem committed to making improvements of their models vs justifying keeping the current one(s).
Skill scores for different time frames are constantly tracked and accountability/performance is well known and acknowledged based on the blatantly obvious, non adjusted statistic for all to see.
I don’t see this being the case for climate models. Adjustments have lagged well behind the reality of observations screaming out loud and clear that they are too warm. Anyone with a few objective brain cells can see that global temperatures are not increasing at the rate of model projections. If it takes an El Nino spike higher in global temperatures to get up close to the model projections ensemble mean for instance, instead of treading along the lower baseline of the range for a decade, then the models are too warm.
There can be no scientific justification to continue with those same models. They need to be adjusted. Wishing and hoping and having decades before needing to truly reconcile models with reality because you are convinced the equations are right and the atmosphere will come around is not authentic science……….it’s just a tool to be used for something other than authentic climate science.
Pat,
Thank you very much for this excellent article, the work and well thought out discussion. I may not agree entirely with everything but believe you make some great points and it deserves to be read/published………even if the gatekeepers don’t agree with all of it.
One wonders if they disagreed with just as much but it supported the CAGW narrative, if it would have been published.
WTF,
Pat referred to a nice post Willis Eschenbach made a few years ago,which YOU should visit,that materially support the main point Pat makes here,here is a useful quote from Willis:
” Willis Eschenbach
May 16, 2011 at 12:01 am
Steve McIntyre has posted up R code for the analysis Iāve done, at ClimateAudit.
The main issue for me is that the climate model isnāt adding anything. I mean, if you can forecast the future directly from the forcings, then thereās no value-added. A good model should give you something that you canāt get from a simple transformation of the inputs. It should add information to the mix.
But the GCMs donāt add anything new, they just spit the forcings out in a slightly different form.
Now, you could say that the model is valuable because it allows us to calculate the variables of lambda and tau ā¦ except that each model comes out with a different value of those two.
The main problem, however, is that we have nothing to show us that the underlying concept is true, that forcing actually controls temperature linearly. So that means that the different lambdas and taus we might get from the model may mean nothing at all ā¦
w.”
https://wattsupwiththat.com/2011/05/14/life-is-like-a-black-box-of-chocolates/#comment-661218
Imagine people trying to model chaos with linear functions……….,using ZERO data as real data, but not yet existing data of the future………
Ha ha ha ha ha…………..
Mike, I wasn’t trying to denigrate Meteorologists with their prediction being wrong in my city,just trying to point out that even with short term predictions based on REAL data can STILL be off from the forecast target.
You wrote,
“This is a big reason why us real operational meteorologists (for 35 years) have such a high % of skeptics vs in other sciences. We must constantly reconcile the forecast with realities. Quickly adjust based on models that also quickly dial in new/fresh data and come out with a new scenario that can sometimes look much different than the previous oneā¦ā¦ā¦..with errors/changes often growing exponentially with time.”
The big difference is that you usereal updated data regularly to adjust the forecast with, While IPCC create a spaghetti based climate model using a lot of assumptions on forcings we know little about and say we can make a forecast far into the future with significant confidence.,
The whole thing is absurd!
Sunset,
I never considered your comment as denigrating meteorologists. Just the opposite, a compliment with regards to how we are reality based in using models based on their usefulness.
I’ve busted at least hundreds of forecasts……..it part of the job. The best busted forecast is the one that gets updated the quickest. I was on television for 11 years and that means that thousands of people see the face and person who busts forecasts and you hear about it.
In the earliest years, I hesitated to update as quickly because of believing the models when I made the first forecast and sort of hoping they would revert to the previous solution when they diverted the wrong way.
I also showed over confidence because of too much trust in models.
The reality is that you can be the best model data analyst on the planet but if the model is wrong, it doesn’t matter………you will be wrong.
With experience, you learn to be more skeptical and certain model tendencies. With so many more models and ensembles available, it provides an enormous opportunty to consider potentially different scenarios.
In the 1980’s, most of us just used one (or 2) operational model and went with whatever it showed.
Reading the “climategate” emails, the “corruption” is well documented. I would not regard every single individual with bias as corrupt, since they also display expectation bias. Trenberth’s assertion that there must be something wrong with the data tells an entire story in one brief sentence. Other emails such as Jones indicating that papers critical of model results and methods need to be suppressed (not published) rather than addressed substantively are also revealing. The “corruption” may initially have been more due to “noble cause” fixation than to economic bias, but once economics and university and agency policy enters the picture, the result can be out right corruption. Any of the journals could have published Dr. Frank’s paper and then left the podium open for actual discussion and demonstration of any mistake he might have made. Not doing so looks unscientific, and outright faith-based rather than grounded in scientific argument.
WTF, ad hominem comment.
If you can’t appraise the manuscript and the reviews you have nothing worthwhile to offer.
So far, you’ve offered nothing more worthwhile than a view into your character.
Forrest,
I think that is pretty easy. Run a climate model many times with different initial conditions, and show that the range of outputs diverges as suggested by Pat’s proposed error propagation.
Forrest,
As far as I’m aware, they have. There is some uncertainty (i.e., running a model with different initial conditions does indeed produce a different path/output) but they do not show the output diverging as suggested by Pat Frank’s analysis. We expect the equilibrium state to be constrained by energy balance and so it is very hard to see how it could diverge, as suggested by Pat Frank, without violating energy conservation.
Forrest,
I wasn’t suggesting that the equilibrium state should be the same at all times, I’m pointing out that it should tend towards a state in which energy is in balance (i.e., energy coming in matches energy going out). The reason it has changed in the past is because things have happened to change the energy balance. The Sun’s output isn’t constant. Our orbit around the Sun can vary. Volcanoes can erupt. Ice sheets can retreat/advance (often due to orbital variations), greenhouse gases can be released/takenup etc. However, the state to which it will tend will be one in which the energy coming in matches the energy going out.
So, if someone wants to argue that the range of possible temperature is 30K (as appears to be suggested by Pat Frank’s error analysis) then one should try to explain how these states all satisfy the condition that they should be in approximate energy balance (or tending towards it).
Yes, that is the controversy. Pat is essentially arguing that something that would produce an offset should be propagated – at every timestep – as an error. This is not correct, which should be pretty clear from Nick Stokes’s recent comment with the output from climate models.
I see that “nophysics” has very little comprehension of error propagation.
Why is that not a surprise?
Little errors GROW to be big errors… that is the way the climate change mantra works !!
in response to ATTP,
This argument is seriously flawed.
Just like shorter range EPS global weather models the outturn is constrained to within realistic climatic values…otherwise they do indeed blow out into massive range of error. Climate models will be no different, but the constraint range means error propagation is limited with each time step.
Seems to me that since the models are blatantly wrong…the offsets, forcings, whatever, are cancelling each other out….either way, you end up linear that exactly matches CO2….something anyone could do with a ruler
First problem seems to be getting modelers to admit that…
but then they are handicapped from the get go…..they are having to back cast to a fake temp history in the first place
forcings add. (aerosol forcing is negative).
by 2016 the anthro
GHG forcings add to
a CO2-equivalent of 489 ppmv
https://www.esrl.noaa.gov/gmd/aggi/aggi.html
489 ppm CO2eq of anthropo forcing? …
current level of CO2 is ~400ppm, meaning, without human action Earth would “enjoy” -89 CO2eq GHG forcing. +2K per CO2 doubling is also -2K per CO2 /2, hence -2K for the effect of going from 400 to 200, another -2K for going from 200 to 100 etc. Let’s stop here, although the theory goes that we should keep going on.
So the theory says that without human GHG, Earth temp would be no less than 4K below current level. Remember that LIA was only 1K below current (so says IPCC), so imagine the effect
I say: LOL !!!
paqyfelyc claim – “THERE IS a line of code that says āthis much more CO2 give this much less heat loss (aka warming)ā
there is a line of code,
a well-honed equation with evidence
to back it up, that uses CO2’s
radiative forcing (which is not warming), at
the tropospause, not the
surface.
because it’s a fact that CO2
absorbs IR. and a fact that the
earth emits
IR. it’s not difficult
to understand, with a model
or equations, why that
means more CO2 means
more warming.
You’re ignoring the water vapor that has 10 or 20 times the energy content, with a temperature sensitivity at Sea level air pressure and temp. And it does what it wants.
micro: water vapor certainly isn’t ignored in
climate models.
but water vapor in the atmosphere only
changes when the temperature first changes;
then it’s a feedback.
Bzzzzzzzz! Wrong.
Do you live someplace that you get dew at night?
Oh’s it a feedback, about -35W/m^2
A lot more than Co2’s forcing.
“radiative forcing” is orwellian newspeak. Indeed CO2 radiates (as just any matter…), and that’s the real radiation that should appear in the equations, not some “forcing”.
itās not difficult to understand, even without a model or equations, why that means more CO2 means
more RADIATION in and out atmosphere and less radiation directly from Earth gets to space. If, and if then to what extend, this result in warming (or even cooling!), is much more questionable.
This is what my work addresses. Specifically cooling under clear calm skies. This is the only condition that really matters. But that’s another argument for another time.
What I found was surface cooling rates adjust themselves, as it get near dew point, water vapor condenses, and that sensible heat supplies a significant portion of the energy radiating to space, which at dusk, was cooling the surface at 3 or more degrees F/hr, but an hour or two later it can be near zero, and there’s still 5 hours of dark, and there is still a 100F (the other night here) temp difference that has to be radiating to space.
You can see this everywhere by just logging RH, Dew Point and Air temp, and you see under clear skies the temp stops falling some nights, and you can measure a 80-100F temperature differential with an IR thermometer, and it isn’t cloudy either.
Everyone assumed it was just reaching equilibrium, it is not. This is the biggest “discovery” in climate science in 100 years, because it shows us water vapor has been actively regulating temps, not ghg’s.
Oh, so CS is just the ratio of the the two cooling rates times the 1.1C/doubling Co2, so above location that got measured it’d be about 1.1C/3, so 0.33C/doubling.
Frankly I should get a Noble Prize for this.
@micro6500
Your work makes sense, so much so that i don’t see anything new in it. Of course atmospheric water is a major heat buffer, that prevent temperature to go down as long as there remain water vapor to turn into liquid water, and hence to compensate heat escaping away through radiation. I doubt very much this deserves a Nobel, or Captain Obvious would already had been awarded (but who knows ? Al gore and Obama got one, so with the right political connection …)
Even “climate scientists” know that, although I suspect they don’t care. The word “dew” doesn’t even appear in the description of the NCAR Community Atmosphere Model (CAM 3.0) : the only water movement they care about is evaporation and cloud formation.
But I figured out it was a negative feedback to Co2. Tell me anyone who has proof of that?
But you’re right, it was stupid obvious. But people assumed it was something else. I recognized it for what it was, the end of co2 panic.
paqyfelyc says – āradiative forcingā is orwellian newspeak. Indeed CO2 radiates (as just any matterā¦), and thatās the real radiation that should appear in the equations, not some āforcingā.”
RF comes from solving the two-stream
equations, which are obtained from
applying energy
conservation and the Planck law to
the atmosphere.
now i think it’s the two-stream equations
that appear in the models, and not the
RF relations. see, for example, equations
4.229 & 4.230 in this model description:
http://www.cesm.ucar.edu/models/atm-cam/docs/description/description.pdf
The problem then is they either are doing the wrong terms or they are leaving the big one out. The assumption that Co2 adds is incomplete, it adds, but water vapor drops nearly as much as was added, it is the negative feedback that is either unknown or ignored. And it only does so for part of the night, averaging a whole day hides the fact it varies.
@crackers345 October 25, 2017 at 8:10 am
No, RF comes from the difference between to virtual numbers: the modeled radiation with [anything], and the same without. This as the same value than a seller pretend you gained 30$ on a thing you paid 70$, because he pretends its price should have been 100$, or an official pretending he made 10M$ saving while he spend 110 M$ instead of 90 M$ previous year, because without the saving he would have had spend 120M$. Pure bovine outgoing matter, that i wouldn’t buy if i were you.
You’ll find numerous instance of the word “forcing” in the model.
Just after one of them i found this extract:
“the large warm bias in simulated July surface temperature over the Northern Hemisphere, the systematic over-prediction of precipitation over warm land areas, and a large component of the stationary-wave error in CCM2, were also reduced as a result of cloud-radiation improvements”
Which translates:
“the model can fit the elephant as need be, It has far enough parameters to “improve” ”
BTW you still didn’t react to my comment
https://wattsupwiththat.com/2017/10/23/propagation-of-error-and-the-reliability-of-global-air-temperature-projections/#comment-2644237
paqyfelyc claimed – “No, RF comes from the difference between to virtual numbers: the modeled radiation with [anything], and the same without.”
difference of what?
this paper calculates RF; this paper describes their methods:
>> We use the Spectral Mapping for Atmospheric Radiative Transfer code, written by David Crisp [Meadows
and Crisp, 1996], for our radiative transfer calculations. This code works at line-by-line resolution but
uses a spectral mapping algorithm to treat different wave number regions with similar optical properties
together, giving significant savings in computational cost. We evaluate the radiative transfer in the range
50ā100,000 cmā1 (0.1ā200 šm) as a combined solar and thermal calculation.
Line data for all radiatively active gases are taken from the HITRAN 2012 database. Cross sections are
taken from the NASA Astrobiology Institute Virtual Planetary Laboratory Spectral Database http://depts.
washington.edu/naivpl/content/molecular-database.<<
B. Byrne and C. Goldblatt
http://onlinelibrary.wiley.com/doi/10.1002/2013GL058456/pdf
It’s worthless, unless you don’t want to know what it’s doing. Now if they ran it over 24 hours and included H2O, you’d see H2O changing, negatively in response to the increase.
But they leave that out.
Funny how they all seem to leave that out.
ps – RF is calculated
at the
tropopause,
not the surface or TOA
crackers345
We are talking about radiative forcing, and you make a long (and boring) quotation about … “radiative transfer”. WTF ? Do you think these are the same?
And you still didn’t reply to my comment https://wattsupwiththat.com/2017/10/23/propagation-of-error-and-the-reliability-of-global-air-temperature-projections/comment-page-1/#comment-2644237
If the climate settles into an equilibrium state. And if the equilibrium state results in a constant temperature, but that is not necessarily how the equilibrium will look – why should it? Nothing in the climate ever stops changing because it cannot ever do so. One of the biggest problems with modeling the climate is knowing what the starting point is.Get one parameter wrong by a little, and your projections can be wildly wrong.
“Get one parameter wrong by a little, and your projections can be wildly wrong.”
climate models are
“spun up” so they start
in a equilibrium
state.
see
http://www.oc.nps.edu/nom/modeling/initial.html
And why would start start in equilibrium, when the Earth is supposed to be an out of equilibrium system (that’s what the Gaia hypothesis is all about, as you probably don’t know)?
@ aTTP
What are the mathematical condition for “if we ran two simulations with different [whatever source] perturbation [aka “forcings” in climate newspeak] (but everything else the same), this wouldnāt suddenly mean that they would/could diverge with time, it would mean that they would settle to different background/equilibrium states” ?
Answer: stable, non chaotic system, that can be treated through perturbation analysis.
You assume 1) equilibrium with null forcing, and 2) forcing will just offset equilibirum by some finite and calculable amount.
The first is obviously false regarding climate, since it wildly varies with zero forcing, as a true chaotic system it is. The second assumption is “not even wrong” when the first isn’t true.
So your objection just means the “climate” you are modelling is from some other world.
Your pseudo is usurped.
This is pure comic gold!
If the inputs to a function are uncertain, then the output of the function will at best be equally uncertain.
In reality, every time you perform an operation on uncertain data, you increase the uncertainty.
And since this is solving simultaneous differential equations by time step, where you have to allow all nodes to reach numerical stability prior to the next step. Each of these nodes carry the uncertainty into the next iteration. And they are modeling an abstraction of the real system.
I’ll point out I spent 15 years as a simulation subject matter expert, covered about a dozen simulators, and created models and circuits that got checked out and reviewed by engineers who had actually build the real thing and tested it extensively on a lab bench. Including simulators that operate like gcm’s operate. Also designed a chip for NASA GSFC, fastest design for them at the time.
MarkW
October 23, 2017 at 9:12 am
If the inputs to a function are uncertain, then the output of the function will at best be equally uncertain.
In reality, every time you perform an operation on uncertain data, you increase the uncertainty.
———–
Maybe I am wrong, and also missing your point, with this simplicity of mine, but just for the sake of it.
There is a “100% certainty” with this models… they all do a warming in a very significant correlation with CO2…SIGNIFICANT AND CLEAR CORRELATION BETWEEN THE WARMING TREND AND CO2 TREND in all of these simulations….. “100% certain”, as far as I am aware.
Also as far as I am aware, these models are not set up or made to do that, they just do it…….there is no any line of code that “says”: “you get this much CO2 give me this much warming”,
or some thing like that…..and besides as per my understanding of these models, they do not actually do any “detectable” quantity warming as caused by CO2……as strange as that may seem.
Correlation does not necessarily mean causation, still it needs a confirmation and some kind of validation even in the case of the GCMs, even when and where it may seem from the outset to be so obvious….but still needed though.
cheers
@whiten
THERE IS a line of code that says “this much more CO2 give this much less heat loss (aka warming)”. If there wasn’t , CO2 wouldn’t appear at all.
The truth is, this sort of code cannot prove the assumption, it can only prove it is wrong. And it does, fairly well.
Everyone, listen to MarkW.
paqyfelyc
October 23, 2017 at 1:18 pm
I have no much choice, but wholly to agree with you there…..in principle.
Fairly well in the prospect. š
considering that it could be proved at some point.
thanks.
cheers
…and Then There’s Physics,
You suggested, “Run a climate model many times with different initial conditions, and show that the range of outputs diverges as suggested by Patās proposed error propagation.”
Actually, that has been done: it is illustrated in the ‘spaghetti graph’ above supplied by Nick Stokes. One of the most critical input parameters is the assumed, and unknowable, RCP. I have not seen a similar presentation for other assumptions about all the input parameters that are known with imperfection even for their current values, let alone future values. I have rarely seen estimates of the ‘albedo’ with a precision greater than 2 significant figures. What would the outputs of ensembles look like if a reasonable range of albedos with different values were used as initial conditions? When we start varying ALL the inputs, one at a time, that will give us a better idea how they may influence the total output. They might even come close to Frank’s upper-bound uncertainty.
Right, ATTP. You say a plus/minus root-mean-square uncertainty statistic is a positive sign physical offset error.
It’s not. (+/-) does not equal “+”. I know it’s a hard concept, but do try.
You’ve made mistake number 7. And that’s over and over, for you.
You also show no understanding of the difference between physical error, which can be known, and statistical uncertainty, which is an ignorance metric.
The first requires the observation as a test against a prediction.
The second conditions a prediction where the observation is not known.
You don’t get that distinction here. You’ve never gotten it in any of our conversations.
I rather doubt you’ll ever figure it out.
Ā±1
very hard to see how it could diverge,
============
The result (future) only converges over a narrow range of conditions even if the energy is identical.
For example. Hot land and. Cold ocean vs cold land and hot ocean. The energy is the same but the climate is not. The nonlinearity of the system allows both possibilities to occur. Or at least to remain outside of current mathematics to calculate any more than we can predict the next roll of the dice.
I’m a super layman here — however TTP’s statement cried out to me. How is a solar forcing not applied at every step? If there is extra heat in step 1, then step 2 will proceed from that extra heat step. Of course, we know that extra heat will be radiated out to some extent. Is that a linear process? Do the propagation from those errors not become cumulative also? Every joule not released back into space also accumulates.
So while I agree with your basic premise — the errors have to be added at every step — I disagree with your disagreement — the errors *do* add at every step.
No… each step has physical uncertainties. There is no “background,” just a series of steps, each of which contributes the potential for a certain amount of error.
I don’t know why this is so hard to understand. Consider the operation of moving a wheel 1 mm. Each time you perform the operation, you miss by 1mm a little bit. Some of those errors cancel out, but over the 1000-step process of moving the wheek 1m the total possible error increases at each step.
If we ask “where is the wheel after 1000 steps?” we would have to qualify the answer with the total possible physical error in the process to give a true estimate of position. You can’t just run a bunch of simulations and say “Look! they converge near 1m!” That’s a different question.
I don’t know that this quite renders models totally useless, but it certainly demonstrates some important limitations.
Yes and also various functions have feedbacks. Thus some errors propagate in this manner. The feedbacks are not necessarily linear either. Each of Nick’s runs in his chart above showing the disparate rcp scenarios also has error bars, widening the top and bottom of the existing spread.
“This has already been explained to you numerous…”
I draw everyone’s attention to this common rhetorical trick of speech. The attempt is rhetorically to position the speaker as the expert and teacher, the addressee as ignorant and naive.
Examples of usage from other contexts:
It has already been explained to you repeatedly that there were no camps or penal colonies under Stalin, and it is unlikely that this attempt will be any more successful than previous attempts. The allegation was invented by right wing anti party conspirators.
It has already been explained to you repeatedly that there was no famine under Mao…..
It has already been explained to you repeatedly that eating cholesterol raises blood cholesterol…
It has already been explained to you repeatedly…..
No, it hasn’t. What has happened is that someone has asserted these things. They have not explained repeatedly.
When the activists in a field commonly resort to this sort of speech, as if by a collective agreement, we know, and have explained repeatedly, that this is a bunch who have abandoned any critical thought and just mouth and parrot the party line.
And sometimes itās not a rhetorical trick; itās just someone thatās frustrated because he really has explained it over and over.
attp, i wonder what the reason for all the messing around with aerosols was ? would models do what pat says if initial conditions are not constrained with variable parameters down the line after initiation of the model run ?
I believe the point Pat is trying to make, and nobody seems to get, is that when measurements are used as input into model equations the measurements have a physically known error range, which should be traceable to a National Institute of Standards reference. Once that physical error is accounted for it propagates throughout each iteration of the model. The reference errors are not a statistic of the measurement but an absolute value of the accuracy- i.e. a temperature could be any number that falls in the error range any time a measurement is made.
The equation y=ax + b, run once, generates and absolute error range of a*{AE} +b*(AE). if x=100 and b=10 and the absolute error is .0001 the result is 110*(.0001)+ 0.001=.0111. The next iteration, further extending the calculation would start with an Absolute error of 0.0111.test
It’s easy to see that the potential error in the calculation can easily balloon after a number of iterations. Based solely on the absolute error of the instrument the potential error can easily become much larger than the any statistical test would suggest.
Observations are not the same as statistics of observations.
aTTP
“Run a climate model many times with different initial conditions, and show that the range of outputs diverges as suggested by Patās proposed error propagation.”
This statement reflects a fundamental misunderstanding about what a model run is and what an uncertainty is. The uncertainty is an inherent property of a measurement, or in the case of clouds, an assumption. The uncertainty about a calculated value is not based on the variability of the result or of multiple runs of a model. It is an inherent property of the inputs and propagates through the calculations in a standard fashion according to strict rules. The output of a model might be exactly the same each time with different inputs! That doesn’t have any influence on the propagated uncertainty.
That this fact of mathematics escapes anyone in a position to affect public policy, I find concerning.
Because this math fundamental escapes so many in the modeling field, apparently, here is a primer from Wikipedia: https://en.wikipedia.org/wiki/Propagation_of_uncertainty
“When the variables are the values of experimental measurements they have uncertainties due to measurement limitations (e.g., instrument precision) which propagate to the combination of variables in the function.”
In the case of clouds, which are poorly characterised, having to choose a forcing without knowing its real effect to better than say, 4 W/m^2, (Ļ1) is the same as a measurement with an uncertainty of 4W. Picking ‘the wrong number’ does not reduce the uncertainty about what follows. It is not “30 W Ā±4W, therefore the true answer is between 26 and 34”. It is that any number selected has an uncertainty of 4W. It is 26Ā±4, 34Ā±4, 30Ā±4 or any other number like 10 or 50.
“For example, the 68% confidence limits for a one-dimensional variable belonging to a normal distribution are approximately Ā± one standard deviation Ļ from the central value x, which means that the region x Ā± Ļ will cover the true value in roughly 68% of cases.” (ibid)
The Resistance formula shows that it is the largest input uncertainties that cause the majority of the magnitude of a propagated uncertainty. Thus temperature, which has a relative low % uncertainty, is minor compared with forcing due to clouds where the uncertainty is large compared with its value.
I encourage everyone to read the Wiki entry and if it is too difficult, try putting in some numbers using the Resistance formula. It will show you that uncertainty never decreases through a calculation.
The Author is correct, and the rebuffs from several journals show that they accept his arguments, but excused themselves from publishing it on the grounds that the readers would not be interested in finding the true answers to this important question. It’s their call, but the rejection was not because the work is incorrect. Obviously many responses and reviews were inane. I am not surprised, I continue to be disappointed by the sorry state of climate science.
ATTP, āThe error that youāre trying to propagate is not an error at every timestep, but an offset.
Youāre wrong. I show in the manuscript that it long wave cloud forcing error is systematic and inherent in the models. It enters every single time step.
Iāve explained that to *you* several times, and and you never grasp the concept.
Just to elaborate further, adding up calibration errors of various models to get a final number does not make error a constant to be subtracted away from a prediction.
Model calibration errors vary with the model, with the forcings, and in each model with the choices of poorly constrained parameters. Your proposed subtraction is a meaningless exercise.
Yes.
ATTP, “Pat is essentially arguing that something that would produce an offset should be propagated ā at every timestep ā as an error.”
No, I’m not. I’m propagating a model calibration error statistic.
Calibration error statistics are not offset errors.
Model cloud error is not an offset error (mere inspection of ms Figure 5, or the figure in Eric Worrall,s comment, is enough to prove the case.
It’s explained in my manuscript.
I’ve explained it to you repeatedly.
You insistently make the same mindless mistake over and over again.
It was wrong the first time you supposed it. It’s wrong this time. It’ll always be wrong.
It will never be right no matter how often you repeat it.
But that won’t stop you, will it.
ATTP
It’s really this simple; the. Earth system cannot be accurately simulated unless all the climatic variables are precisely accounted for. The tiniest inaccuracy will garbage the run. And it wont be known where the mistake was created. Current models rely heavily on inference. They are all utterly unskilled in projecting.
If anyone is getting a “background equilibrium state” out of a climate model, the model is worthless. The boundary conditions for climate change continuously (TOA solar intensity changes +/- 47 W/m^2 every 180 days, the tilt of the earth changes every 18.6 years, the cloud cover – and hence albedo – changes over a tremendous range hourly, water vapor distribution in the atmosphere – the major climate driver – changes constantly in a manner that isn’t even known, etc., etc.), some in a semi-periodic manner and some randomly. We don’t even know what all of the variables are, but from what we do know, the climate can never reach a state of equilibrium.
Having said that, if you take the position that the models are based on calculating perturbations away from a background equilibrium (a common technique for analyzing non-linear systems), then I think you’ve made Mr. Frank’s case in part. In that case, you have linearized a highly non-linear system, and his error propagation analysis is perfectly correct.
palc wrote – “BTW you still didnāt react to my comment
https://wattsupwiththat.com/2017/10/23/propagation-of-error-and-the-reliability-of-global-air-temperature-projections/#comment-2644237”
it’s completely wrong, and
shows you don’t know
the science.
Trillion$. The simple number of reasons for rejection.
What is needed is the $trillions to publish as a supplement in the NY Times ? or ?
(following the money trail always leads to the edge of a cliff )
attp. Nice to be able to comment isn’t it. Under your rock … not so much.
No one is allowed to comment in “wrong” ways on said site because it is RIGHT. You know, omnipotent. It’s an interesting trait found in most climate change propagandist sites. It used to be that science was smart enough to explain itself and win an argument, but the collective understanding has dropped to where silencing the opposition is the only answer. You remember the Dark Ages, right?
Meet Kenny …
Wellington, if you have a point to make, please make it so that you add something to this discussion.
(It is important to me,since there is a possibility he is using at least two or more accounts here,which is a bannable offense) MOD
Mark: I could not recall at first what the references were about āallowed to commentā. Then I remembered Iāve read something way back but everyone must judge the veracity of the link for himself.
Mod: Iām sorry. I do not know more than what my quick āmemory refresherā search found.
Everyone: I care about the actual argument, not who is making it. However, I do consider circumstances like someone preventing an adversarial argument at oneās own site while engaging in it elsewhere (when that applies).
indeed, i live not too far from attp. i may have a word about the moderation on his own site in person.
Please, keep us informed
Thanks
JW
Hmm. You identify one reviewer as a Gavinoid. Are you sure you were not observing a Nickoid instead? The plumages are remarkably similar…
The two comments above by WO and FG are unhelpful and violate WUWT commenting policy: “those without manners that insult others or begin starting flame wars may find their posts deleted.”
@NeedleFactory – I was not aware that WUWT has a new moderator?
In any case, there is somewhat of a difference between pointing and laughing at an opponent, and viciously attacking them. Not much, but some. For examples, you can look at some of Nick’s comments elsewhere here, which, in between his ad hominems, simply prove the point that Forrest makes.
This’ll be interesting. I won’t understand a word but look forward to the debate.
Good response, Hotdog. You will have given some relief from guilt to thousands of skeptics like me who havn’t a clue about the subject nor the time to find out, but who nevertheless will be hoping that this is the definitive moment when the wall of pseudo-academic superiority behind all the modelling nonsense begins to be broken.
“Climate model air temperature projections are just linear extrapolations of greenhouse gas forcing. Therefore, they are subject to linear propagation of error.”
Err No.
The temperature outputs are the result of ALL THE INPUTS.
Those inputs include ALL KNOWN FORCING, not just GHGs, but solar, volcanic, land use, etc.
in addition there are feedbacks which cannot be predicted and which are emergent.
Your paper has not been accepted because you are wrong.
If there are feedbacks which cannot be predicted and which are emergent, there’s no reason to believe the models in the first place. They could be completely overturned tommorrow by a pesky emergent feedback.
The feedbacks emerge from the models. If you don’t believe the models, you don’t believe the feedbacks.
Are you trying to argue that nobody knew about feedbacks until the models discovered them?
Sheesh, you don’t need models to determine that feedbacks exist. Just think for yourself.
Nick, the key word employed by both Mosher and Sheri was “emergent” – “feedback.” That is unforseen “feedbacks” – I’m pretty sure you are quibbling over terminology, but pay attention to the intent instead. Those “emergent” conditions would create unexpected, unmodeled behaviour in the empirical data, and create unanticipated divergences between modeled results and measured empirical conditions. If those “emergent” influences tend to have a bias that cannot be accounted for, then the mean model results and empirical data will diverge over time – creating “hiatuses” or “pauses,” possibly even long term states like Little Ice Ages.
Sheri
October 23, 2017 at 5:34 am
But models do only one significant feedback…..temp to CO2, or maybe the other way around, where other feedback have no any potential or detectable effect, as actually is supposed to be under a RF warming ever increasing….the main standing point of AGW, for not saying the entire point of AGW….
So an RF warming can not actually be messed up by other feedback, especially when in fast up going trends….
So, the question: What actually ate all that supposed AGW RF expected warming!?
A “dog” feedback” perhaps!
cheers
Water vapor let’s it go to space until the surface cools off, then drains energy stored in atm column and as water vapor to slow cooling once air temps near dew points.
https://micro6500blog.wordpress.com/2016/12/01/observational-evidence-for-a-nonlinear-night-time-cooling-mechanism/
Duster said – “Those āemergentā conditions would create unexpected, unmodeled behaviour in the empirical data, and create unanticipated divergences between modeled results and measured empirical conditions”
no. the feedbacks emerge as a result of
the models’ underlying equations. viz
of the physics incorporated into the
model.
example: ice-albedo feedback. basic warming
from CO2 melts ice, so less
sunlight is reflected back to space
and so the ocean & air warm more.
this emerges from models, because they
continually calculate ice extents. they assume
ice has a certain albedo, and ocean another.
thus, when ice melts, more warming occurs,
beyond that of CO2 alone.
And if you implement it like you described, it’s wrong. Because that is wrong, once the incident angle gets under 20degrees or so, open water has nearly the same albedo as ice, so there’s only about 1/4 day under solar noon that is positive. The rest, if the sky is clear is a huge radiator to space.
micro, i notice you never get a response from nick or mosher to your posts . you should get your work written up and submitted .
I’m not very good at that kind of stuff. And I know I would jet get jerked around until either I got tired of it or they did. But it’d never be published while it matters.
So I published it at wordpress, and code and reports on sourceforge. I’m sure it’s been seen by more people through social media than some pay to play journal.
And sooner or later it’ll be the end of this mess.
I just want it called the “Crow Effect” lol!!!!
Crackers, what you are describing is certainly NOT linear.
I have seen Gavin tweets where he fully acknowledges the models do not model the feedbacks correctly if at all. One example, The ENSO pseudocycles are clearly chaotic responses that feed into GMST +/-, but the models are helpless on it.
I have seen Gavin tweets where he fully acknowledges the models do not model the feedbacks correctly if at all. One example, The ENSO pseudocycles are clearly chaotic responses that feed into GMST +/-, but the models are helpless on it.
Nick Stokes quibble over terminology? He lives for that sort of thing.
micro6500
October 23, 2017 at 2:06 pm
Thank you micro.
Appreciated a lot.
From my point, almost all comments of yours appreciated in my part.
But if I have not got this wrong…hopefully, as only a superficial pass at your work there…..it seems mostly, as far as I can tell, as a further detailed and very interesting at that, about the Trenberth “Iris”……which may explain how the earth and atmospheric response works in relation to RF forcing in short term.
Please do forgive me if I happen to have misunderstood your work…..but it seems to be very important in away to try and explain the non linearity of reality of climate versus the linearity propagated by the GCMs…
Please do let me know, if you would not mind, if I happen to have misunderstood your point……..no body is perfect..:)
Thanks.
Cheers
I’m not sure it operates like an iris, more like a turbo button.
I think what’s happening is the sensible heat from the cooling atm column, including all the water vapor that is condensing (and then re-evaporating), keeps the surface warm, near dew point temp until the Sun comes up to store up energy to do it again.
So, more like a bucket of water with a hole in it, and after it lowers air temp to dew point, opens a spigot that supplements the water level so it doesn’t drop much more than this until the Sun come up and fill them both back up(all the while the one is still draining).
micro6500 commented “Water vapor letās it go to space….”
what
the he11 does this
mean?
Means you do not understand how the surface cools at night
Except that they have been shown to not be. You can make assertions about how models are supposed to worj and how modelers think they do work, but unless you are a unique set of humans/modelers that never make errors, you are going to have prove what you say is right.
As for emergent feedbacks from models, please. The idea that your model is so brilliant that it is showing us things we didn’t know rather than being errors is the sot of arrogance that gets modelers a really, really bad name.
putz
Mosher contradicts Mosher
https://imgur.com/a/DEZYf
Steve Mosher, the linearity of GCM air temperature projections is demonstrated in dozens of examples right there in front of your eyes.
In the manuscript and the Supporting Information document.
I doubt you’ve even looked at either, though; much less read them, much less understood them.
Which might explain your denial of the demonstrated.
“Hereās the analytical core of it all:
Climate model air temperature projections are just linear extrapolations of greenhouse gas forcing. Therefore, they are subject to linear propagation of error.”
It’s the core of the nonsense. For a start, they aren’t “extrapolations” of forcing. You can find a curve fit, by fiddling parameters. So? That is true of many things. It doesn’t mean that the mechanism of the model is wrong or trivial, or even that its error propagation should follow the curve fit.
The statement that “therefore” they are subject to the linear propagation of error is just assertion. It has no basis.
“The volcanic forcings are non-linear, but climate models extrapolate them linearly.”
Gobbledook. What does non-linear here even mean? With respect to what? But again, climate models don’t “extrapolate” them. They admit them as a forcing in the set of equations, and give an approximately proportional response. Not unexpected.
From the figure captions
“The points are Jim Hansenās 1988 scenario A, B, and C. All three scenarios include volcanic forcings.”
Actually no. Scenario A did not include volcanics. Pat’s argument proceeds regardless.
Nick writes
Including say clouds in the models. Fitted but meaningless.
“It doesnāt mean that the mechanism of the model is wrong or trivial”
But you KNOW that it is wrong, don’t you NIck.
All that bluster to hide KNOWN errors.
So sad. !!
“And Jim Hansenās scenarios assumed NO volcanic forcings?”
You don’t read, and you don’t know anything. Scenario A assumed no volcanic forcings. B&C had forcings, clearly reflected in the featured figure.
I didn’t take Pat Frank’s argument to imply the first part of what you wrote. I took his curve fit (as you put it) to be a simplified model of what the model’s output. His model of the models does a pretty good job of doing that. Using the curve fit, he then propagates a specific error to generate the uncertaintany at each step of his model. The implication is that the more complex models are not properly propagating error.
The notion that Frank’s critique is innapplicable ‘because d-d-d-different models!’ is gibberish baloney.
Team up with other ‘error propogation is not applicable’ folk around here and write a rubuttal guest post.
RW wrote, “I didnāt take Pat Frankās argument to imply the first part of what you wrote. I took his curve fit (as you put it) to be a simplified model of what the modelās output.”
Thank-you RW. That’s a very succinct recapitulation.
Honestly, it’s a relief to read the remarks of folks here who get the analysis. Thank-you all. š
crackers345, so yes one could conceivably hard code the Earth’s orbital path into the code as an influence on solar insolation. I am not sure where the controversy lies here, and as I have said elsewhere I don’t know climate models.
^ wrong subthread
which scenario was closest to reality and how many volcanoes erupted while that reality played out ?
Nick Stokes wrote, “For a start, they arenāt āextrapolationsā of forcing.”
The emulations demonstrate GCMs do exactly that. In any case any “projection” is an extrapolation of conditions into future states. So, you’re wrong empirically and in priciple, Nick, and all in one sentence.
“You can find a curve fit, by fiddling parameters. So?”
So, it means that climate model air temperature projections linearly extrapolate forcing to project air temperature. That’s all they do.
The consequence? Linear propagation of error. And that’s QED.
“It doesnāt mean that the mechanism of the model is wrong or trivial, or even that its error propagation should follow the curve fit.”
The demonstration has nothing to do with “the mechanism of the model.” The model is a black box.
The demonstration has to do with model output. It’s shown to be linear. That’s the only thing necessary to show, to justify linear propagation of error.
The linear equation successfully emulates the air temperature projections of any GCM. That makes it completely appropriate to use for propagation of projection error.
“What does non-linear here even mean?”
It means what the Gavinoid implied it means: inflective departure of forcing from a smooth curve. Take a look at the graph. Forcing does that when volcanoes enter the picture.
“They admit them as a forcing in the set of equations, and give an approximately proportional response. Not unexpected.”
With the bolded phrase, you’ve inadvertantly validated my analysis, Nick. That’s twice now. Thanks again. š
“Scenario A did not include volcanics.”
Scenario A included volcanics prior to 1990. The historical set when viewed from 1988. It’s right there in the graph.
David Cosserat
I doubt it will be a definitive moment, that was supposed to be Climategate which was all to easily swept under the carpet.
However, there is always room for debate on any subject and having it in the open is beneficial to us all, sceptics and alarmists.
As for me not understanding the content, I’m not a scientists, nor even well educated, but I long ago learned that the climate debate is more than just science. Besides, after 60 years of observation, I don’t see any meaningful change in the planet’s climate other than my garden plants growing better than they ever have.
“the climate debate is more than just science” Egg zactly. Make into a poster and plaster on every wall.
As a system, the socio-political climate complex has shown high stability and resilience built on strong across-the-board negative feedback to any forcing.
They donāt even need a carpet.
I tried to download the paper and just got a complaint from the host about an ad-blocker, and promotional material.
The problem is in your side. I downloaded it without any problem, no pop-ups, pop-unders or anything. Here is a screenshoot:
http://oi64.tinypic.com/15q4rx4.jpg
There is a 97% chance you may probably have a virus in your computer. Have you visited any naughty website?
No, I tried to download it directly, like you would any other file. When I went through their website, resisting their $5 blandishments, it came through.
I modeled an attempt to download it and it worked every time.
I modeled an attempt to download it and it worked every time.
LOL
diddums !! do you need a hanky?
Interesting. I do adblock and had no such problemā¦..Maybe the site simply doesn’t like you?
Nick, you did notice that there are two download options. Only one asks for money. The other is labeled “Slow Download.” It really isn’t that slow unless you are using a 1990s modem for your connection.
PF gave a link, I right-clicked, said save link as, and got a whole lot of gibberish html.
Nick, it downloads fine. If you really don’t have a copy, I can put it in DropBox for you.
As I noted above, when I tried to save with “save link as”, I got that nonsense. When I got through to the page displayed by Atheok, I was able to download it, as I noted above. I have read it, quoted sections, and shown images of text from it.
To me it seems like an overall confusion between error and uncertainty? They are not the same according to the GUM standard (Guide on Uncertainty in Measurements). An error can be corrected (calibrated) if known. An uncertainty cannot.
You got a crux issue, Jan PC. š I have yet to encounter a climate modeler who understands that difference.
Another point raised in GUM is that random errors become systematic when they are propagated forward into a calculation. That’s another form of systematic error thoroughly ignored in climate modeling.
The computer simulations in question have hard coded in that an increase in CO2 causes warming. Hence these computer simulations beg the question as, does CO2 cause warming, and therefore are of no value. In terms of atmospheric physics there is plenty of reasoning to support the idea that the climate sensivity of CO2 is really zero.
“The computer simulations in question have hard coded in that an increase in CO2 causes warming.”
Evidence?
I’m asking for evidence that “have hard coded in that an increase in CO2 causes warming”. Do you have any?
“Iām asking for evidence that āhave hard coded in that an increase in CO2 causes warmingā”
Do you use GISS as a hind-cast?
There is your answer.
Or are you that naive? really ???
Nick writes
Well we know for a fact that adjustable parameters are changed to set the required radiative imbalance in the models. How’s that?
The scenarios with more GHG’s lead to warmer model projections…
This is either a coincidence or a pretty good clue that the models “have hard coded in that an increase in CO2 causes warming.”
‘This is either a coincidence or a pretty good clue that the models āhave hard coded in that an increase in CO2 causes warming.ā’
No. It suggests that the GHE physics means CO2 would cause warming, and that they correctly model the physics. But “hard-coded”. That is just made up.
On that logic you could say that computation could never reveal anything. Because if it predicts anything, then the result must have been hard-coded in.
“It suggests that the GHE physics means CO2 would cause warming,”
So you ADMIT that ERRONEOUS science is programmed into the models.
FINALLY you are waking up to reality !
WELL DONE , Nick.
Nick, you do realise that you just admitted to every word Forrest has said, don’t you ?
So FUNNY !!.
Try a new pair of socks.. those one don’t seem to be so tasty for you. !!
Evidence ( or dense? )
Unless Nick wrote it, it ant !
So if an economic model predicts 2% inflation, they must have hard-coded 2% in?
If CO2 isn’t hard coded in the models, then what is the point?
“If CO2 isnāt hard coded in the models, then what is the point?”
……we have a winner!
Vanna has some great parting gifts for the rest of you…….
Come on Nick, don’t be THAT stupid:
“Evidence?”
David made the OBVIOUS reply to your, …. he he….question.
?w=700
Nick Stokes October 23, 2017 at 2:15 am
āThe computer simulations in question have hard coded in that an increase in CO2 causes warming.ā
Evidence?
…
Nick Stokes October 23, 2017 at 3:38 am
So if an economic model predicts 2% inflation, they must have hard-coded 2% in?
I have friends and neighbors who still think that climate alarmists are arguing in good faith.
Unbelievable.
David Middleton
October 23, 2017 at 3:04 am
The scenarios with more GHGās lead to warmer model projectionsā¦
—————-
Not trying to be picky, but at the best the above still is no more than an assumption, even in the case of the GCMs, when it may seem to be so obvious and “certain”.
It still needs a kind of validation……otherwise it remains an assumption in principle.
Considering the strong correlation of CO2 ppm trend with temps in GCMs, and the connective relation, is no hard to consider that detectable distinction “who jumps first to increase” the temps or CO2 ppms in any GCMs scenario may clarify that is possible.
I never know of any such trial feat ever attempted or performed, as per a way of validating the assumption!
For as long as this point remains not clarified, in principle, the assumption remains so at its best, an assumption, no matter how strange it would seem to consider it that way, under the circumstances. .
cheers
“If CO2 isnāt hard coded in the models, then what is the point?”
Sometimes they code in everything else, and then infer CO2 or GHG as what’s left. It’s a valid technique as long as you’re completely omniscient on every other factor involved.
More obfuscation from Nick. A model IS hard coded. The result predetermined. A model produces different results because it is initialized differently by the user, provided different values for the parameters by the user, or provided different parameters by the user, or perhaps because the code, for whatever insane reason, uses a random number generator. Where you draw the line between one model and the next is arbitrary. Comparing predictions to observations are the only way to definitively test a model. A given model is refuted when its prediction does not match observation. When a model is based on thr observations it is used to predict, it is overfitted, liable to be modeling more noise than it should, and will underperform with new observations.
“A model IS hard coded. The result predetermined.”
This gets to silly quibbling. Of course you can say that any computer program is hard coded, and computers do what they are told. So when Deep Blue wins at chess against Kasparov, that was hard-coded. Gets a bit silly, but technically true. It doesn’t mean that the programmer put in the tricks that brought Kasparov undone.
There is a popular line of articles at WUWT about chaos and the unpredictability of GCMs (and CFD, and weather, for that matter). It’s true that GCM’s approach attractor solutions that can’t be worked out from initial conditions without that computation. As with CFD, you learn things from computation that the programmers couldn’t have told you.
Mosher
https://imgur.com/a/DEZYf
1sky1
I said: “Without water vapour and CO2 and the other minor radiative/absorptive gases, the surface would be a rocky waterless planet with a mean surface temperature around that of the Moon.”
You said: “That would be the case only if the Moon had an GHG-less atmosphere of equal density!”
What you said is only true if you subscribe to the minority view that the pressure (hence density) of the GHG-less components of an atmosphere have a warming effect. That is no longer a mainstream sceptical view because the physics involved in this so-called non-GHG warming effect have never been satisfactorily described.
Nick. So I think we agree that this is just weed territory. Having said that, it is a waste of comment space to nit pick at stuff like a claim that CO2 is hard coded into the model. Clearly it is at some level hard coded in the model to increase temperature with CO2, all other things being equal. Just because there are other factors that might swamp that influence out in the model doesn’t mean the comment was worthy of additional scrutiny. It’s borderline troll territory in my view to get into weeds like that. I’m willing to grant intellectual charity to the poster. I’m willing to believe that they are aware that C02 is not the only factor in these models and that, in fact, depending on some of the other factors, the model could predict reduced temperatures despite increasing levels of CO2.
Nick. No argument from me vis a vis the utility of modelling and running them to see what happens. I’m willing to believe though that there is a mind out there that is sharp enough to foresee what the model will do (or have a pretty good idea) without running it. But for sure, for the rest of us, we need to run the model to see what happens. The deterministic aspect does not hinge on our ability to work out what the model will output though. I don’t know climate models though, so if there is some built-in random number generation (simulated stochastic stuff) then obviously no one would be able to know what the model will output in advance.
RW commented – “A model IS hard coded”
Given the 1/R^2 law of
gravitation, is the orbital
path of the Earth
“hard coded?”
craclers345. If something is hard coded, it is baked into the architecture, the programmer’s code rather than being a parameter. So, altough levels of CO2 itself is undoubtedly a specifiable parameter, what is done wirh the value is hard coded – i.e. the complex function that outputs temperature among other things given rhe values of many other variables in the function. This is what I take willhaas to be saying when he wrote the bit Nick objected to. Not sure what your question has to do with programming a climate model, but it smells like more quibbling over arbitrary distinctions to me.
crackers345, so yes one could conceivably hard code the Earthās orbital path into the code as an influence on solar insolation. I am not sure where the controversy lies here, and as I have said elsewhere I donāt know climate models.
To within 1/4 watt/m^2 at TOA, this formula agrees with Lief Svaalberg’s daily recorded TOA values.
TOA_DOY =1362.36+46.142*(COS(0.0167299*(DOY)+0.03150896)) for DOY = 1 on Jan 1 to 365. (Excel format)
Willhaas,
You say: In terms of atmospheric physics there is plenty of reasoning to support the idea that the climate sensitivity of CO2 is really zero.
I strongly disagree. You are (I hope inadvertently) undermining the mainstream sceptical position.
It is certain that the presence of CO2 in the atmosphere since time immemorial contributes in a minor way to the current mean surface temperature of 15degC (288K), the main (invisible) contributor being water vapour.
Without water vapour and CO2 and the other minor radiative/absorptive gases, the surface would be a rocky waterless planet with a mean surface temperature around that of the Moon, namely -75degC (198K) as determined from the NASA Moon orbiter. In the absence of all such gases, the earth’s atmosphere would be transparent to the remaining atmospheric constituents – Nitrogen and Oxygen – which are not significantly radiative/absorptive (by several orders of magnitude) at earth atmospheric temperatures.
Therefore, since CO2 is a minor contributor to the warm world we currently experience, a doubling in CO2 must, in logic, cause some change in mean surface temperature.
The real debate is: how much of a change? Sceptics say, not a lot. Alarmists say, by a dangerous amount. But there is no reason in physics (atmospheric or otherwise) that says the climate sensitivity to changing CO2 is exactly zero. To assert that is to walk straight into the climate alarmist trap…
Your argument applies equally to homeopathics. Fortunately mainstream skeptic position is an oxymoron.
David: I tend to agree. At one time, it was forbidden to say CO2 had no effect because it made skeptics look unscientific. Now, it seems common and almost mainstream here. To say it has “no effect” makes the same assumptions saying it does have a great effect doesāthat we know everything there is to know about climate. I can’t see how a real scientist can make a statement to that effect.
‘argument applies equally to homeopathics’
Chutzpah you have but you are not a Jedi yet.
Actually, Earth radiates as if it were -18 C (255 K) for an average power of 240W, while the moon, which has only 0.13 albedo (Vs 0.3 Earth’s), receives and radiates ~295 W. So, this just doesn’t add up to a mean surface temperature of 198K for the moon. Or, rather, to have things add up, you have to consider that moon surface is so bumpy that it’s surface si much larger than 4Pi RĀ², but then you cannot compare this temperature from NASA moon orbiter to Earth’s.
jaakkokateenkorva, October 23, 2017 at 5:17 am.
You say: Your argument applies equally to homeopathics.
In relation to the totality of ‘greenhouse’ gases, CO2 is a small proportion of GHGs. Some say 25%, others say 5%, but either way, it is NOT a vanishing small proportion. (Yes, it is a vanishing proportion of the atmosphere as a whole including the non-radiative/absorptive gases, but these do NOTt contribute to warming the planet.)
So the warming effect of CO2 is not in any way comparable to the charlatanry of homeopathy. It is real.
I find your ‘oxymoronic’ comment completely incomprehensible…
Thanks Hugs, but yeah. I don’t have enough midi-chlorians to speak on behalf of all skeptics, 97% scientists etc and, thus, limit writing only my own opinions. Doesn’t prevent me standing by the gas law pV=nRT though. Meaning the relationship between pressure, volume, temperature and mass of gas are interrelated irrespective of the composition.
David. The concept of “mainstream skeptic position” is much like “scientific consensus”. In my opinion equally useful at best and oxymoronic at worst.
“It is certain”
It used to be certain that humans were causing the Earth’s atmosphere to cool, back in the 1970’s. Then the climate warmed up and we don’t hear that certainty anymore.
Being certain about something does not necessarily make it true. You are just guessing as to what CO2 is doing in the atmosphere. It may not be adding any net heat to the atmosphere at all. Prove it does.
This skeptic is skeptical you or anyone have any proof to the contrary. “Certain” is not good enough.
Am I hurting the skeptic’s cause by my assertions?
No, the skeptic’s cause is to demand proof of other’s assertions. If there is no proof, skeptics should say so. I’m saying so. Prove me wrong.
David Cosserat
October 23, 2017 at 9:10 am
… but these do NOTt contribute to warming the planet…
While I agree about the basic argument you offered, you do make a mistake. f for instance, you descend from Jerusalem to the Dead Sea, you experience sensible warming as you descend. Since the atmosphere has the same composition, the difference in temperature is not due to CO2, methane or water vapor.
The proponents of GHG planetary temperature say the surface temperature of a planet is mostly due to the atmospheric composition giving the greenhouse effect and irradiance, naysayers say it’s almost completely due to the density and molar mass of the atmosphere, as well as irradiance on the planet.
Of course there are models showing the later is correct whereas the former has never been empirically demonstrated in the real world.
Take a look at the Galilean Moons in order of descending atmospheric density: Io, Callisto, Ganymede, Europa. Now, I’ll list these in order of descending average temperature: Io, Callisto, Ganymede, Europa (coincidence?). Now, the irradiance is quite similar for all these moons and only one has a significant amount of greenhouse gases comprising its atmosphere — Callisto.
Now, can someone tell me why Io has a higher average temperature than Callisto, despite having an atmosphere almost entirely comprised of sulfur compounds whereas Callisto has an atmosphere of CO2? Hint: Io’s surface pressure is orders of magnitude higher than Callisto’s.
Why wouldn’t quantized molecular vibrations induced by back radiation have a significant impact on planetary surface temperature? It’s because heat transfer in an atmosphere is dominated by unquantized kinetic energy (molecular collisions that theoretically occur every 10^-7 s at Earth’s surface pressure) and convection. Furthermore, molecular vibrations are quantized, you simply can’t add more vibrational energy to a molecule if it is already in its energized state. Molecules already in their energized state are transparent to the radiation that already put it into that energized state. Trying to make planetary surface temperature about quantized molecular vibrations is like trying to flood the sea by spitting in the ocean.
of course it has an effect. whether that effect leads to x amount of warming over x amount of time is an entirely different subject . remember negative feedback , i don’t see the oceans boiling off any time soon . nor turning to acid,david attenborough what can i say, another childhood hero ruined.
That would be the case only if the Moon had an GHG-less atmosphere of equal density! What is overlooked here is the fact that moist convection–not LW radiation–is the principal mechanism of heat transfer from Earth’s surface. Were the Earth totally dry, its GHG-less atmosphere would still be warmed largely by convection, making the surface temperature problem far more thermodynamically complex than that of the Moon.
RW Turner
You said: The proponents of GHG planetary temperature say the surface temperature of a planet is mostly due to the atmospheric composition giving the greenhouse effect and irradiance, naysayers say itās almost completely due to the density and molar mass of the atmosphere, as well as irradiance on the planet.
You justify your position by citing data concerning the Gallelean Moons.
You say: Now, can someone tell me why Io has a higher average temperature than Callisto, despite having an atmosphere almost entirely comprised of sulfur compounds whereas Callisto has an atmosphere of CO2? Hint: Ioās surface pressure is orders of magnitude higher than Callistoās.
No I can’t. But how reliable is your data and how reliable your conclusions? If you have such a dramatic example of non-GHG warming, surely you have investigated the data in depth and could further enlighten us and give us some references? Please!
1sky1
I said: “Without water vapour and CO2 and the other minor radiative/absorptive gases, the surface would be a rocky waterless planet with a mean surface temperature around that of the Moon.”
You said: “That would be the case only if the Moon had an GHG-less atmosphere of equal density!”
What you said is only true if you subscribe to the minority view that the pressure (hence density) of the GHG-less components of an atmosphere have a warming effect. That is no longer a mainstream sceptical view because the physics involved in this so-called non-GHG warming effect have never been satisfactorily described.
“That is no longer a mainstream sceptical view because the physics involved in this so-called non-GHG warming effect have never been satisfactorily described.”
What are you talking about, never described? It’s been described for hundreds of years by the first chemists and expanded ever since. It’s such basic physics that I believe they start teaching it in primary school these days.
PV = nRT
https://www.researchgate.net/publication/317570648_New_Insights_on_the_Physical_Nature_of_the_Atmospheric_Greenhouse_Effect_Deduced_from_an_Empirical_Planetary_Temperature_Model
RW: that paper doesn’t use physics to
describe the claimed non-GHG heating,
it does curve fitting only.
it’s also very wrong, which is why
it couldn’t get published anywhere
except in
a
predatory journal.
jaakkokateenkorva,
Anyone who has used a hand pump to inflate a tire, and has been careless enough to touch the outlet valve, has received an immediate affirmation of adiabatic heating by compression, as specified by the Universal Gas Law. However, it will be safe and comfortable to touch the same spot an hour later.
The Earth has had a thick atmosphere for billions of years. Because a hot body radiates in proportion to the fourth-power of the absolute temperature, the atmosphere would have been radiating at a much higher rate when first formed. Clearly, both the original molten surface and the atmosphere have cooled during the intervening billions of years. We still see local heating of an air mass when it rises over a mountain range and plunges down the leeward side, or when an air mass dives off a plateau into a depressed basin. However, in both cases, if the air mass became stationary, it would radiate the excess heat and cool down. That is, what we observe today is largely potential and kinetic energy being converted locally into palpable heat energy. If any of the parameters in the equation T=pV/nR are CHANGED, one can expect a change in the temperature. However, the major changes took place far enough back in time that most of the initial temperature increase has leaked away. The major role of the atmosphere, with respect to heating, is to be a transport medium for water vapor and clouds.
RW Turner,
Concerning my challenge to you (October 24, 2017 at 12.19am), I have been eagerly awaiting your response which sadly has not so far been forthcoming. So I did some research on Io and Callisto. I Googled Wikipedia (mainly) and Universe Today and assembled the following data:
Io
1. Mean orbit radius: 421,700 km
2. Mean diameter: 3643 km
3. Surface area: 41,910,000Ā km2 (0.082 Earths)
4. Atmospheric composition: 90% SO2
5. Surface atmospheric pressure: 1 nanobar [this from Universe Today]
6. Mean surface temperature 110K.
7. Albedo: 0.63
8. Internal heat source: 0.6 to 1.6Ć10^14Ā W (global total)*
*Note: Io’s main source of internal heat comes from tidal dissipation rather than radioactive isotope decay, the result of Io’s orbital resonance with Europa and Ganymede. The figure given across the whole surface is equivalent to a significant heat flux of 0.41W/m^2.
Callisto
1. Mean orbit radius: 1,882,700Ā km
2. Mean diameter:4820km
3. Surface area: 73,0000,000Ā km2 (0.143 earths)
4. Atmospheric composition: 100% CO2
5. Surface atmospheric pressure: 7.5 picobar
6. Mean surface temperature: 134K
7. Albedo: 0.22
8. Internal heat source: Negligible
So, yes, as you said Io has an atmospheric pressure around 2 orders of magnitude greater than Callisto. But at a pressure of 1 nanobar this is still totally insignificant. In fact it is 9 orders of magnitude below earth’s surface atmospheric pressure! Even if a non-GHG such as SO2 were capable of doing the job that you claim it can do, it would not be enough to warm a single flying gnat.
It is obvious that the surface temperatures of Io and Callisto are influenced simply by their respective albedos and, in the case of Io by its significant source of internal energy.
In any case as a final blow to your theory, it appears that the surface of Io is NOT as you have claimed warmer than the surface of Callisto.
So I think your “Io versus Callisto” hypothesis is in total ruins. š
RW Turner,
Re. your comment (October 24, 2017 at 8.46am), the physics involved in the so-called non-GHC warming effect has indeed been described over and over and over ad nauseam. But what I had claimed to you was that it has not been satisfactorily described.
The two responses that follow on from yours, from Crackers and Clyde Spencer, say it all…
Crackers is correct – curve fitting alone will not persuade anybody unless the underlying physics is made clear. I am still hoping that my friends N&Z will be able to do this one day, but so far I have not been convinced.
Clyde Spencer is also correct – the tyre pumping analogy is complete crap because it does not represent a steady state flow of energy situation, as is the case with planetary surface temperature elevation.
David, I think you found the brightness temperatures, not the true temperatures. I have the surface temperature as estimated by Voyager data at 143K and Callisto at 134K.
https://books.google.com/books?id=SO48AAAAIAAJ&pg=PA331&lpg=PA331&dq=average+temperature+of+io+143+K&source=bl&ots=h95La-5A01&sig=CrraabuweItecqyQr3tDY7t7ivs&hl=en&sa=X&ved=0ahUKEwjZ6uPCkYrXAhUB-GMKHQL0A3gQ6AEIQjAD#v=onepage&q=average%20temperature%20of%20io%20143%20K&f=false
https://www.space.com/16419-io-facts-about-jupiters-volcanic-moon.html
Crackers, the only criticism I’ve ever seen of that paper is the journal it is in and that it’s “simply wrong.” Sounds like a cultist argument. Perhaps you would like to be specific in how the paper is incorrect in its conclusions. It’s like saying the models show a near perfect fit, but I’ll stick with the models that don’t work. You do that, I’ll go with the empirical based models that do work. Furthermore, the paper doesn’t do simple curve fitting, it used physics based models to estimate surface temperatures of rocky planets and then compared that to empirical observations and then investigated why the models did or didn’t work.
“A key entailment from the model is that the atmospheric āgreenhouse effectā currently viewed as a radiative phenomenon is in fact an adiabatic (pressure-induced) thermal enhancement analogous to compression heating and independent of atmospheric composition. Consequently, the global down-welling long-wave īux presently assumed to drive Earthās surface warming appears to be a product of the air temperature set by solar heating and atmospheric pressure. In other words, the so-called āgreenhouse back radiationā is globally a result of the atmospheric thermal effect rather than a cause for it ”
You have all the time in the world, now go ahead and actually demonstrate how this is wrong.
“So, yes, as you said Io has an atmospheric pressure around 2 orders of magnitude greater than Callisto. But at a pressure of 1 nanobar this is still totally insignificant.”
Totally insignificant? Yet Io has a significantly higher surface temperature than Ganymede and Europa has an even lower surface temperature corresponding with its atmospheric density being the lowest of the 4 moons.
This might help…
http://formulas.tutorvista.com/physics/work-done-by-gravity-formula.html
David Cosserat:
If the well-known ideal gas law and the attendant adiabatic heating, which gives rise to a ubiquitous atmospheric lapse rate, are not physical explanation enough for you, consider the issue of convection of heat into a non-existent atmosphere. Your projection of Moon-like surface temperatures on a GHG-less Earth is physical nonsense.
Not so! Unlike a pressurized tire cooling down to the surrounding ambient atmospheric temperature, adiabatic heating applies to ALL parcels of air at a given elevation. There simply is no cooler air surrounding any parcel (unless introduced by advection).
“It suggests that the GHE physics means CO2 would cause warming, and that they correctly model the physics.”
no…it means they were hindcast to a period of time when both CO2 and temps were rising
Micro6500 says: It never would have a temp like the moon, it has an atm. Enthalpy at daily Tmax / atm cubic meter of air at sea level is ~38.8kJ/kg/m^3, and drops to 24.9kJ/kg/m^3 at Tmin.
What is it about the thought experiment we are discussing, involving earth with a GHT-less atmosphere that you now suddenly don’t get? Earlier, I thanked you for supporting me in saying that all the radiation to space would be from the surface, which would be at a comparable mean temperature to that of the Moon. Now you are contradicting yourself. Yes its GHT-less atmosphere would have a heat content (enthalpy) but this atmosphere would not radiate. It would simply be at a comparable mean temperature to the surface, maintained by conduction/convection between the two.
I’m not sure it would be as cold as the moon, and having an atm, even here might be enough to start the water cycle, as with no GHG’s the atm would be hard to cool, and transport might easily allow excursions over 0C, and the equator isn’t average temp of the earth anyways.
So my comments were on topic. I believe you are wrong about water on an Earth without GHG’s being ice. At least sometimes at the equator it will be water.
paqyfelyc
You say: Actually, Earth radiates as if it were -18 C (255 K) for an average power of 240W, while the moon, which has only 0.13 albedo (Vs 0.3 Earthās), receives and radiates ~295 W. So, this just doesnāt add up to a mean surface temperature of 198K for the moon.
1. The earth without any GHGs would be a waterless rocky planet, with a similar albedo to the Moon.
2. Your calculations may not add up to a mean surface temperature of 198K but that is what the NASA Orbiter measured. So think again carefully about what is wrong with your reasoning…not what is wrong with the NASA data. š
I don’t think NASA measure are wrong. Nor is my calculation (which isn’t mine, anyway!).
I just say you cannot compare the 198K of moon surface to Earth surface temperature. You should use the temperature moon would enjoy if it were perfectly flat, or dulled by an atmosphere so that it behaved as a flat surface (as Earth does, apparently).
1. The earth without any GHGs would be a waterless rocky planet, with a similar albedo to the Moon.
Well, of course, it would, since to remove GHG you have to remove all water .
Now, is this would be true without GHG effect from water and other gas? No it wouldn’t. Earth would still radiate as a 255K body (240 W in and out) from somewhere around ~10km above surface, and because of lapse rate (which has nothing to do with , surface temp would still be so that liquid water would cover most of the planet.
paqyfelyc,
You are getting muddled. Without any GHGs in the earth’s atmosphere the earth would not “radiate as a 255K body from somewhere around ~10km above the surface” because non-GHG gases DONT RADIATE!
The only radiation would be from the surface. Just like happens on the Moon.
I don’t know where the myth of non-GHGs not emitting radiation originated, but I’m pretty sure that everything with a temperature a few degrees above absolute zero is going to emit IR. Besides, how do you propose that an atmosphere with no GHGs loses heat? Convection and collisions will transmit heat within the atmosphere, but how is heat going to exit the atmosphere without emission from these gases?
@David Cosserat
As Robert W Turner pointed out:
even an otherworldly non radiating atmosphere must have some way to lose energy to compensate for the energy it will get from the surface (condensing water of clouds, for instance). Or else, it’s energy and temperature will forever rise with no limit, which cannot happen.
Liquid water and ice of clouds, dust, etc. will do the job.
So one way or another, atmosphere would still emits energy according to it temperature, itself according to its altitude.
GHG probably play some role, so that the emitting apparent altitude would be lower without them, but it won’t be zero
Robert W Turner commented – “Besides, how do you propose that an atmosphere with no GHGs loses heat?”
it escapes out the top
of the atmosphere
What?
This is the problem, hardly anyone actually understands EM wave propagation.
Non-GHG’s don’t radiate. It would cool by IR radiation from the surface to space based on SB equations.
Robert W Turner,
You say: Iām pretty sure that everything with a temperature a few degrees above absolute zero is going to emit IR.
An oxygen/nitrogen-only atmosphere (the point of discussion here) would not radiate significantly at earth temperatures. Hence the standard use of the term “non-GHGs” to describe such gases.
You ask: …how do you propose that an atmosphere with no GHGs loses heat?
It doesn’t. As I explained previously to paqyfelyc, all the heat would be lost by radiation directly from the surface to space. Just like happens in the case of the Moon.
And the hypothetical nitrogen/oxygen-only atmosphere would be warm just like the surface but would have no way (or need) to loose that heat to space.
paqyfelyc,
You say: …even an otherworldly non radiating atmosphere must have some way to lose energy to compensate for the energy it will get from the surface (condensing water of clouds, for instance). Or else, itās energy and temperature will forever rise with no limit, which cannot happen.
You are muddled again. The subject of discussion is a non-GHG atmosphere. Such a case usually assumes a waterless earth.
But, to indulge you, if water were to be introduced into such a waterless world, it would be in the form of ice because the surface temperature would be very much less than the freezing point of water (just like the Moon’s). The ice would increase the albedo and therefore further reduce the surface temperature. š
Not necessarily, it would depend on the orbit and Sun.
I’m not sure if we are warm enough with a nitrogen oxygen atm if water would be ice or not, but there nothing stopping such planet, and it wouldn’t have to be ice, and once the water cycle started, would be a lot like earth.
Micro6500 says: …hardly anyone actually understands EM wave propagation. Non-GHGās donāt radiate. It would cool by IR radiation from the surface to space based on SB equations.
Thanks for your coherent support. It seems it is not possible to have a sensible scientific discussion with people who have read just enough physics to learn that, yes, “all gases at temperatures about absolute zero radiate” but who insist that oxygen and nitrogen radiate significantly when it is well established that they do not. Which is why it is standard practice, in the context of the earth’s current atmosphere, to refer to oxygen and nitrogen as “non-GHGs”.
Micro6500 says: Not necessarily, it would depend on the orbit and Sun.
Er…well, yes, true if you introduced water onto a non-GHG version of Mercury, for example!! But I think you slipped a cog… we were discussing a GHG-less earth having a surface temperature similar to that of the Moon. š
So was I. The big differences is we have enough gravity to hold an atm, the moon doesn’t. All we would need is enough solar to get the water cycle running. Without GHG’s, maybe we need a closer orbit, but nothing like Mercury, probably not even Venus. So there is a range where you could have a water cycle and no other GHG’s. Although that isn’t going to happen anyways, but have at it.
It never would have a temp like the moon, it has an atm.
Enthalpy at daily Tmax / atm cubic meter of air at sea level is ~38.8kJ/kg/m^3, and drops to 24.9kJ/kg/m^3 at Tmin.
Micro6500 says: The big differences is we have enough gravity to hold an atm, the moon doesnāt. All we would need is enough solar to get the water cycle running. Without GHGās, maybe we need a closer orbit, but nothing like Mercury, probably not even Venus.
This is way off topic which is about the earth in its current orbit.
Micro6500 says: It never would have a temp like the moon, it has an atm. Enthalpy at daily Tmax / atm cubic meter of air at sea level is ~38.8kJ/kg/m^3, and drops to 24.9kJ/kg/m^3 at Tmin.
What is it about the thought experiment we are discussing, involving earth with a GHT-less atmosphere that you now suddenly don’t get? Earlier, I thanked you for supporting me in saying that all the radiation to space would be from the surface, which would be at a comparable mean temperature to that of the Moon. Now you are contradicting yourself. Yes its GHT-less atmosphere would have a heat content (enthalpy) but this atmosphere would not radiate. It would simply be at a comparable temperature to the surface, maintained by conduction/convection between the two.
My sympathies and best wishes. The warmists really are hoof-deep in the trough and will use any malign tools to protect their grants.
Hey Pat,
you KNOW you are over the target when you start taking flack form the self-appointed small-guns of the AGW farce.
WELL DONE. š
Thanks, Andy. š
You’re right about ESS reviewer 2 round 2. However you were too focussed on his errors about what he said about your paper to notice this hilarious one:
“if the GMST of one year was solely a function of the GMST of the previous year, then GMST would be effectively constant over time.”
Maybe he should talk to Mandelbrot about Z[n+1] = a Z[n] (1 – Z[n])
I know the error is irrelevant in the context, but to think that someone claims to be able to understand modelling does not know about such sequences is … kind of frightening, given the power that has been given to these people.
Hi Rich — thanks and you’re right I was too focused to notice that.
I did notice that he left “function” undefined though. š
The discussion in the paper about error propagation is just nutty. It says
“In a climate projection of ānā steps, each time step āiā initializes with the climate variables delivered by the āi-1ā step.”
OK.
And then the linear fit in Sec 2.2 is described with annual steps. Well, OK, could be anything. It’s just a fit PF has chosen. You could have any step length.
So then he works out the accumulation of variance. It accumulates as sqrt(n), which he then takes to be n=100 (years). That’s where the Ā±15°C comes from.
But it’s supposed to be the error of the GCM, and they don’t have annual steps. They have steps of about 30min. That is about n=1.75e6 steps in a century, or Ā±419°C.
There are two things just wrong here:
1) The uncertainty claimed is not step to step. It’s a claimed error in the identification of TCF. That doesn’t switch between Ā±4 W/m2 every 30 mins, or even every year. But the error is calculated as if it did.
2) The model has physics. If it ever did wander by 15°C or 419°C, that would greatly change energy balance. i/o flux would change to restore.
“The model has ERRONEOUS physics.
” If it ever did wander by 15Ā°C or 419Ā°C,”
OMG, you just keep digging your hole deeper and deeper.
“i/o flux would change to restore.”
With that one little clause, you have destroyed the WHOLE AGW meme.
but do you even realise it ?????? š
Beautifully done, Nick !!
Who’s side are you really on. š
His own?
The following image is a combination of the ECMWF 15 day and 46 day forecast temperature for the UK. Note how the outcome is constrained within upper and lower bounds. This is done to restrict output to within climatic normal values. What it doesn’t show is that if these constraints were not there these individual ensemble members would blow out into a massive spread of solutions. I would imagine climate models are similarly constrained in output so as to give a more clear and desired outcome. This process restricts time step error propagation….but also allows modelers to steer the output towards a desired outcome.
“Note how the outcome is constrained within upper and lower bounds.”
And notice what they are. Between 14°C and 0°C. It isn’t an artificial computational limit. It’s just physics. There just isn’t enough heat coming in to sustain a temperature above 14°C. And too much to go below 0.
In reply to Nick Stokes:
It isnāt an artificial computational limit. Itās just physics. There just isnāt enough heat coming in to sustain a temperature above 14Ā°C. And too much to go below 0.</quote.
I am sorry but these constraints are mostly mathematical not physics. I have run basic models without these artificially applied constraints and they blow up into wild output in no time. Hell the American GFS weather model still does this sometimes.
For example, let's say ensemble member number one at time step 2 comes out with a forecast temp of 14 deg C. The model calculations then can make the next time step potentially go warmer or cooler because of the multitude of input variables and their associated errors….if the next time step is warmer and the next and the next..then you have blow out. They have built in climatic normal limits to make sure that these ensemble members do not venture outside of climatic normals. The physics of the models may act to some extent to constrain the range of outcomes, but mathematics don't care about these when it comes to error propagation so artificial climatic normal constraints must be applied.
Nick Stokes October 23, 2017 at 4:03 am
āNote how the outcome is constrained within upper and lower bounds.ā
And notice what they are. Between 14Ā°C and 0Ā°C. It isnāt an artificial computational limit. Itās just physics. There just isnāt enough heat coming in to sustain a temperature above 14Ā°C. And too much to go below 0.
That seems to imply that the model does not represent the modeled system realistically enough to constrain itself and has to have limits set to prevent it “blowing out” as pbweather describes?
Gavin
And yet the models are based on cumulative forcing and feedbacks, it seems to me total energy input and it’s resultant effect should limit such a model to within the bounds of what is possible with out artificial constraint.
Gavin
And yet the models are based on cumulative forcing and feedbacks, it seems to me total energy input and itās resultant effect should limit such a model to within the bounds of what is possible with out artificial constraint.
Not at all. You have to imagine that each new time step in the model is like the beginning of a new model and that the input variables into that model is the result of the previous time step calculations. If the result of the previous time step is an extreme outlier, then the input for the next time step starts with probably unrealistic input variables and from there could blow out into something physically/climatically not possible/likely. Hence why weather model models set restrictive bounds linked to climatology not just annually, but seasonally.
If this process was not included you may have a situation, for example, where the model forecasts 35 deg C in the middle of January for the UK. Physically this is within annual ranges of possibilities, but seasonally it is not possible or likely. To say that the physics of the models restricts this range is pure not completely true. It is an mathematical algorithm that checks input and output at each time steps to see if it is within normal climatic limits as the main limiting factor.
Just wondering if commenters here were to be on a mission to be placed in orbit around Saturn and the path to get there had been modeled to find the fastest and safest method but the models included programed constraints on what could happen, would any of you get on board that spacecraft?
If you need constraints to stay reasonable then your underlying physics are not well understood. for AGW we know that this can’t be the case since “the science is settled”.
So prove it, run all the models without constraints and let the data fall where they may…What, have you got something to hide???
It would be nice for once, to see just 1 of the traces from start to end, with the initial conditions (parameters)
I see that none of the trolls shows the slightest sign of having read the paper.
The overwhelming ridicule has already been achieved, at least from metrology perspective, with only the idea of measuring the average global outside air temperature anomalies with 0.1 Ā°C precision and accuracy today. Planetary gas composition and energy balance anomalies are even funnier. Modelling all these figures decades into the past & future takes it all into a new dimension. Might work in Star Trek, but doubtfully they’ve gone that far yet.
“I see that none of the trolls shows the slightest sign of having read the paper.” Translation: The speaker is omniscient and the conclusion self-evident, so if you disagree, you obviously did not read the paper.
Few things are self-evident, though in climate science, it seems to be the entire area is self-evident and therefore anyone disagreeing did not read the material. Sorry, life does not work that way. Nor does science.
@Nick Error accumulates as sqrt(n) if and only if the error is random with mean zero.
If the error is systematic, i.e. embedded in the program logic and function parameters, the error should accumulate > sqrt(n), possibly approaching n
Thanks, Stephen.
And when one is making a sequential series of calculations, each subsequent calculation depending on the last and each a prediction further out along some coordinate, the systematic uncertainty increases similarly.
When the prediction is a future time state, physical error is unknown and all one has to condition the prediction is accumulated uncertainty.
NS,
You said, “i/o flux would change to restore.” Are you suggesting that prophesies of a Tipping Point and runaway Venus-like conditions are invalid?
Nick’s point 1: Who said anything switches between +/- 4 W/m^2 at any time interval? Pat Frank didn’t. The error reflects a range of possibilities. Nick’2 point 2: not all that relevant to whether or not error propagation is applicable to the climate models.
Stephen Rasey,
“@Nick Error accumulates as sqrt(n) if and only if the error is random with mean zero.”
RW,
“Who said anything switches between +/- 4 W/m^2 at any time interval?”
He’s saying that successive time intervals contribute independent errors with sd 4. And yes, he’s accumulating them as sqrt(n), as the second figure in the article shows. Here is the section of the paper:
And why annual, when GCMs have steps of 30 minutes? It comes back to that averaging nonsense. He averaged 20 years of data, and said the average was 4 W/m2/year, not 4 W/m2. If he had expressed it as averaging 240 months of data, he would have taken monthly steps, and compounded the error at sqrt(12) times the rate.
Nick Stokes, āHeās saying that successive time intervals contribute independent errors with sd 4.ā
No, Iām not. Iām saying that uncertainty increases, not that error increases.
The error of a future time state is entirely unknown.
Youāve made mistakes 2, 4, and 6 again.
Nick, āHe averaged 20 years of data, and said the average was 4 W/m2/year, not 4 W/m2.ā
I did no such thing. Serious mistake, Nick. And you claimed to have read the paper.
I averaged no set of data. Lauer and Hamilton averaged data, not me.
They calculated the 20-year annual mean calibration error statistic made by their set of 27 CMIP5 models.
Their 20-year annual average ensemble mean bias (eqn. 1) was calculated as delta = (1/N)[sum over N of (model minus observed).
Guess the dimension of the 1/N divisor. Does that dimension enter the dividend?
I took their rms annual average long-wave cloud forcing error statistic. They reported that as Ā±4 Wm^-2 annual average = Ā±4 Wm^-2year^-1.
Itās a statistical average, Nick. Not a measurement average.
Statistical averages are of dimension (property average)/(unit averaged). The average height of people in a room is meters/person, not meters.
Twenty calibration error magnitudes averaged over 20 years is annual average error is error/year. Multiply by twenty, recover the sum.
Your assessment is utterly wrong (again).
wat
Benjamin ….. exactly
Back to middle school math for you, Benjamin Winchester, and take Toneb with you.
Nick Stokes, āannual steps. Well, OK, could be anything. Itās just a fit PF has chosen. You could have any step length.ā
Air temperature projections are published as annual time series. The CMIP5 LWCF calibration error statistic of Lauer and Hamilton is an annual rms average. Those realities completely justify my choices.
Nick, ā[GCMs] donāt have annual steps. They have steps of about 30min. That is about n=1.75e6 steps in a century, or Ā±419Ā°C. ā
Nick supposes the LWCF calibration error statistic for 30 minutes is the same as for one year. Pretty naĆÆve mistake, Nick.
Nick, āItās a claimed error in the identification of TCF.ā Itās a measured and published simulation error. No one claimed anything. Except you.
Nick, ādoesnāt switch between Ā±4 W/m2 every 30 mins, or even every year. But the error is calculated as if it did.ā
Itās not error, Nick. Itās an uncertainty statistic. Mistakes 2, 4, 5 and 6. Youāre out.
Nick, āIf it ever did wander by 15Ā°C or 419Ā°C, that would greatly change energy balanceā
Youāre treating an uncertainty as a physical temperature and a calibration uncertainty statistic as an energetic perturbation on the model. Mistakes 2, 11, and implicitly 12. Seriously fatal mistakes.
In fact each of your mistakes is fatal, Nick.
Your analysis is wrong throughout.
Indeed, your analysis is so clueless it doesn’t even rise to wrong.
“Air temperature projections are published as annual time series. The CMIP5 LWCF calibration error statistic of Lauer and Hamilton is an annual rms average. Those realities completely justify my choices.”
That is nonsensical reasoning. Your results are entirely dependent on the time step. If you have 100 steps, you multiply by sqrt(100). If you have 1000, you would multiply by sqrt(1000). Sp you are telling us that the whole magnitude of your estimated error is dependent on arbitrary publishing conventions.
You have committed errors 0, 7, 32 and 77. Average error score 29 sec^-1. You didn’t last long long.
Nick Stokes āYour results are entirely dependent on the time step.ā
No, theyāre not. Youāre assuming that the average annual Ā±4 Wm^2/year LWCF calibration error is invariant with the averaging interval.
I dealt with that in ESS Round 1 reviewer 4, item 4.
If the averaging interval changes, so does the calibration error value. Put those new values into the error propagation equation, and a completely comparable uncertainty comes out.
The centennial projection uncertainty is about Ā±15 K no matter what.
Iām horrified allegedly qualified scientists donāt know what error is.
Allegedly. Ah yes, only Pat Frank does.
You certainly don’t Nick. The evidence for that is fully in view above.
By the way, my perception of error is taken from the literature. One can read all about it in Sections 7 and 10 of my Supporting Information document. You did read that, too, didn’t you Nick.
They know. They just don’t care.
They know, which is why Pat Frank’s paper keeps getting rejected.
Excellent article.
This is precisely where we need to focus.
Iterative models exlode even the tiniest errors and faster computers just generate garbage faster.
I will definitely be downloading and propagating this.
You ask – at 2:52 am in the same time zone that Willis is in.
Some of us need a night’s sleep now and again.
Hello Pat,
I am no scientist, although I did love learning about Geology & as well as watching the stars.
Keep your head held high and hold to what you believe is true.
One of my favourite sayings, ‘To thine own self be true.
Sandra
Thanks, Sandra. š
If models are moving away from linear propagation of everything to more life-like nonlinear behaviour, involving a mix of positive and negative feedbacks – then that is a move in the right direction.
Haha Oh Pat Frank, you’ve lost the plot here.
“The stakes are just too great. Itās not the trillions of dollars that would be lost to sustainability troughers. Nope. Itās that if the analysis were published, the career of every single climate modeler would go down the tubes, starting with James Hansen. ”
I actually showed your work to a friend of mine who work on climate models. He was not impressed. Sloppy math. That’s all there is to it. No massive conspiracy that you could singlehandedly expose IF ONLY they’d publish it in the peer reviewed corner of the internet rather then the WUWT corner.
Cheers
Ben
So show us an example of the the “Sloppy Math” !! Or do you just do Drive By Whining ?
So why doesn’t your “friend” come and comment on it if you don’t understand the math? It seems, from the evidence, that anyone who works on models could use all the ideas for improvement they could get, seeing that not one of them can produce anything that looks like reality!
john, it’s hard to get imaginary friends to interact with real people in the real world.
A climate modeler rejects criticism of his models.
Color me surprised.
LOL!
I have a mate who knows a mate…who knows something about ……errr….what are we talking about again?
He knows a guy… Is this an “Ocean’s 11” reference?
There was one time, at band camp, when a climate modeller showed me his math…..oh boy, did all us girls have a laugh!
sloppy math guys, that’s all there is to it. Should be pretty clear from the text I quoted that the authors has completely and utterly lost perspective. Perhaps Pat Frank would do us all the favour of publishing the complete review comments of every time the paper was rejected, rather than selectively quoting and commenting? That would actually be interesting.
The full review comments are linked right there in my essay, benben. All 44.6 MB of zip file. Have at them. The link is just above the 12 points of reviewer scientific acuity.
And as John, Butch and bitchilly wrote, do bring your friend around for a conversation.
benben said “goo-goo” and “waah-waah”. !!
Many more people have access to WUWT than most climate pal-reviewed papers EVER get.
Heck , even benben could go ga-ga over it, if benben was capable..
” more people have access to WUWT than….”
…
Andy, please do not confuse quantity with quality.
Mark S
Are you saying something about the 97%?
Yes, I know YOU lack any sort of quality.
But many people here have more maths and science training than most so-called climate scientists.
You are just not intelligent enough to realise that.
You can catch with Willis here:
https://rosebyanyothernameblog.wordpress.com/
If someone tried to include a climate model as part of a nuclear power plant safety evaluation … such as “this climate model says the local temperature 40 years from now will be X and ocean levels will be Y so that’s what the nuclear plant design will be based on” … they would be laughed at by the nuclear regulators (whose employer funds climate models lol) and the nuclear plant investors.
Engineers and nuclear regulators understand climate models are primarily political constructs and not something that can be relied on for engineering. If climate models can’t be relied on for engineering purposes, then they are junk.
Just one example of the sheer nuttiness here. ESS reviewer 1 objected very reasonably:
“The dimension of +-4 W m-2 year-1 is not right, but should be +-4 W m-2, since it
is calculated from 20-yr means but not 20-yr trend. The author lacks basic knowledge on averaging.”
So for this he copped a tirade. But he’s obviously right. The PF deatiled response refers to Eq 6.2:
If you average 20 years of temperature, you get an average as T°C. It isn’t T°C/year, just because you had 20 years. You could average the same as over 240 months. That doesn’t make it T°C/month. And it isn’t just a typo. He has altered the text to make the claim clearer. The units are repeated many times, and the issue is central to the error calculation.
Eq 6.2 – sorry, Sec 6.2.
So, according to my understanding, this annual average of the cloud coverage as an input of a climatic model is decent. It looks different if one assumes that the average, which does not include a trend, is for the future. But as an input for a simple model, this is not objectionable for me. Except one searches for any grounds for a complete rejection of the model. There are many averages that are broken down in the year, in other scientific branches, why should this average be wrong: during the period mentioned, the heat input was so high. Now he is broken down annually.
For example, a company has been operating excellently over the last 20 years, with the exception of the last two years. Then came the proceeds. What was due to restrictive politics and data grief, especially the Middle Kingdom. If one now assumes that a newly elected President Trump would bring it to an end (which he has not normally done yet) to change this policy and end this data gag, the prospects for the company look the same much better. Therefore it is wrong to conclude from a relatively short period of the data on the further trends. In this case, an observation of the company’s healthy base would be more meaningful for the entire period.
Nick Stokes
L&H Table 4 is satellite measurement uncertainty, not model error. That is irrelevant to the argument.
L&H Figure 2 shows āthe ensemble mean bias of the CMIP3 and CMIP5 models., i.e., the difference errors between model simulations and observations.
Figure 2 does not reference one simulation made N times.
It references N separate and distinct simulations.
For each of N models: simulation minus observation = error/model.
For each model, twenty years of simulation error, averaged annually = [Sum of (error/model)/20 years] = error/model/year.
rms of N-fold error/model/year = Ā±(error/model/year)
L&H began at the grid-point level, meaning positive and negative errors were derived. The reported rms Ā±(calibration error) was global annual.
You’re making the same mistake Patrick Brown did in the discussion we had on his presentation page, supported by ATTP and reinvented by my ESS reviewer.
You’re forcing the units of a measurement average (magnitude) onto a statistical average (magnitude per unit). You’re wrong.
Still, on average you will see TĀ°C each(!) year. Would you calculate with an offset of TĀ°C to estimate the temperature in 20 years or would you calculate 20 x TĀ°C?
That’s basically what he is saying.
But what temperature do you want to take 20xT? the first, the last or the average yearly temperature of a twenty-year period? The calculation of the model author is not bad as input. And it also includes a trend, which I have judged negatively in my previous post. No, I think reviewers are overwhelmingly in a climate of rejection. If you think about what Climategate has shown, which reviewers have the say in which circles, is not surprising. Since then, nothing has changed, even if someone has been forced out of his age, he was replaced adequately.
So change the time scale to months. Would you calculate 240 x TĀ°C?
Everything is about the average / year. So once you define it as such you have to take it. To prove the author wrong you should debunk his calculation where he defines the +/- 4 W/mĀ² as a yearly value.
Regarding your example with temperature. I understood it this way: You have a 20-yr. trend (!) in temperature. Let’s assume the temperature is rising TĀ°C/year on average. Of course you have to apply TĀ°C every year to estimate the temp. after 20 yrs. If you just had varying temperatures without any trend, then T = 0Ā°C. Since the paper, the author was referring to, based its calculation on a 20-year trend I am assuming that
a: there is a trend 0
b: which implies there is a yearly contribution
it should be:
a: there is a trend greater or smaller than zero
[The mods note that, indeed, most trends are greater then, equal to, or less than zero. .mod]
I just checked Lauer and Hamilton, 2013. This is the paper, Pat Frank is referring to. As far as I can see they are talking about annual(!) means based on a 20-yr trend.
‘Figure 2 shows 20-yr annual means for liquid water path, total cloud amount, and ToA CF from satellite observations and the ensemble mean bias of the CMIP3 and CMIP5 models.’
It seems correct to use the yearly value, because Lauer & Hamilton are doing it as well.
http://journals.ametsoc.org/doi/pdf/10.1175/JCLI-D-12-00451.1
“As far as I can see they are talking about annual(!) means based on a 20-yr trend.”
There is nothing there about a trend. They are simply averaging 20 annual means, which does not change the units. And they give the result in Fig 2, eg for LCF, in W/m2, not W/m2/year which PF changes it into.
āConsider: (20-year sum of annual uncertainties)/20 years = uncertainty per year.ā
When you average, you donāt divide by 20 years. You divide by 20 units. It doesnāt change the dimension.
Erā¦ if the uncertainty doesnāt have a time dimension, over what period did the uncertainty accumulate? Very little happens over zero time in any physical process to generate uncertainty.
paulclim is correct, the correct step seems to be one year.
Nick, First, L&H Figure 2 shows annual differences. Annual differences, Nick.
Do they really have to put “per year” on those numbers for a reader (you) to figure out that “20-year annual averages” are magnitude per year?
Figure 2 does not show L&H’s root-mean-square LWCF annual average statistical error, which is what I used.
Pat,
“Nick, First, L&H Figure 2 shows annual differences. Annual differences, Nick.”
Complete nonsense. The caption to Fig 2 says:
“FIG. 2. Differences in (top to bottom) 20-yr annual averages of LWP, CA, SCF, and LCF from the (left) CMIP3 and (middle) CMIP5 multimodel means compared with (right) satellite observations. For details, see text.”
They show differences between the 20 year averages of multimodel means compared with 20 year annual averages of satellite measured. Nothing about annual differences. And they mark the result as W/m2. They are scientists; they get units right. Here is the bottom frame and caption:
They spell it out for Fig 1:
“Figure 1 shows the 20-yr annual mean liquid water path averaged over the years 1986ā2005 from 24 CMIP5 models that have LWP as an available variable.”
When they give error estimates, they express them generally as %. That is, % of W/m2. Here is Table 4, which does give a unit for LCF too. That unit is W/m2. There is not a unit W/m2/year anywhere in the paper.
I think the term error / year is misleading. If you take it word by word it leads to Nick’s interpretation which is correct because the error of the model is constant over time. But I think that’s not the dispute.
The question is how does this error behave in terms of uncertainties when modelling a non constant CO2 forcing. Let’s assume the time interval of the model is a year and the starting CO2 forcing is 4 W/mĀ² and the error also is +/-4 W/mĀ². The next year the CO2-forcing will be 4.035 W/mĀ² due to slightly risen CO2 concentration and the error will still be 4 W/mĀ². That delivers a band of statistically possible temperatures in the second year which means the uncertainty in the 3rd year will be even bigger because now you have to apply the model error to the results of the first loop which are inputs to the second. That’s the propagation of a constant error.
I don’t see a problem with computing a SD from 20 consecutive annual numbers each of which happens to be either averages or difference scores (which themselves were computed from averages and a reference point) from the same instrument.
If the numbers are from totally different sources (year to year changes in the underlying effect might qualify), one would have to weight the 20-year mean by incorporating the variability in each of the underlying annual averages; and the variance in the 20-year mean would be a function of the variance within each source (year to year) – each annual average should have a variance.
No need to worry about a trend, because 1) the aim is a conceptual 20 year average, and 2) the trend is conceptualized as diffferent effect sizes for each year.
Nick Stokes āThere is not a unit W/m2/year anywhere in the paper.ā
My bolding throughout.
Page 3831, column 1: āInterestingly, BCCCSM1.1, CCSM4, and NorESMl-M show similar biases in the simulated annual mean LWP,ā¦ā
Annual mean bias = error/year
Page 3831, column 2: āFigure 2 shows 20-yr annual means for liquid water path, total cloud amount, and ToA CF from satellite observations and the ensemble mean biasā¦ā
Ensemble annual mean bias = average error/model/year
Page 3833, column 1: Taylor diagram ācomparisons of the annual mean cloud properties with observationsā giving the āstandard deviation ā¦ of the total spatial variability calculated from 20-yr annual means.ā
In the Taylor diagram, āthe linear distance between the observations and each model is proportional to the root-mean-square error (rmse)ā
Thatās the root-mean-square error (rmse) of the 20-year annual means = error/year.
All the errors (biases) quoted are in reference to 20-year annual means. All of them are rmse.
Page 3833, āFor CMIP5, the correlation of the multimodel mean LCF is 0.93 (rmse = 4 W m^-2) and ranges between 0.70 and 0.92 (rmse = 4ā11 W m^-2) for the individual models.ā LCF is long wave cloud forcing, simulation of which produced the calibration error statistic used in my analysis.
Multimodel average LWCF rmse of 20-year annual means is Ā±4 Wm^-2/model/year.
L&H may not write it that way, but itās very clear from the usage in the text, and from the meaning of their equation 1, that that is exactly what they are reporting.
Pat,
“L&H may not write it that way, but itās very clear from the usage in the text, and from the meaning of their equation 1, that that is exactly what they are reporting.”
It is only clear to you. Normal folks see 4 W/m2 and take it that they mean 4 W/m2.
All the annual mean qualifiers that you bolded are just there to emphasise that seasonal effects are removed. Annual means are themselves made up of readings during the year, which are averaged. You described the process here
But the units are now simply cloud-cover, not /measure or /week or whatever. No change. Yet when you further average those annual averages in 6.4, you change the units to /year. It makes no sense.
Nick Stokes, eqn. 6-3 shows the 20-year mean observed cloud cover bringing in the average over the year and dimension year^-1.
Eqn. 6-4 does the same for simulated cloud cover.
As soon as you take a 20-year mean, the dimension of the dividend acquires year^-1.
Pat,
“As soon as you take a 20-year mean, the dimension of the dividend acquires year^-1.”
So why don’t units change in 6.2? Why is it different averaging over sub-year intervals?
If you take rms, do they acquire year^-0.5?
Nick is right. Annual means means average over a year in this context. It would be different, if there were a trend, i.e. a growing error over 20 years. In this case you could read it as error/year. But that is not the case because over 20 years the total error would be 80 W/m2 which is more than the observation. So most likely annual means has to be read as average over a year and a 20 year average of annual means most likely is the average of 240 consecutive months.
Nick Stokes, āSo why donāt units change in 6.2? Why is it different averaging over sub-year intervals?ā
Why should units change in eqn. 6-2? Change from what?
Equation 6-2 is step 1 applied to cloud cover expectation values. It averages them at a given grid point within a given year. It has the same dimensions as eqn. 6-1.
Nick, āIf you take rms, do they acquire year^-0.5?ā
Different time averages acquire their own dimensions and error values. I worked with those of L&H. The dimensions are consistent throughout.
paulclim/a>, āNick is right. Annual means means average over a year in this context. It would be different, if there were a trend, i.e. a growing error over 20 years. In this case you could read it as error/year. But that is not the case because over 20 years the total error would be 80 W/m2 which is more than the observation. ā
L&H report a calibration experiment. Calibration error statistics do not require a trend. They require consistent dimensional units. One could take the model LWCF error from multiple individual years and combine them into a calibration annual error statistic, and it would be of dimension W/m^2/year.
Calibration error is not a slope.
Further, your ā80 W/m^2ā is the result of treating the Ā±4 W/m^2 as a physical error, rather than a statistical uncertainty. Ā±4 W/m^2 is not the slope of a trend.
You have also dropped the āĀ±ā in your analysis. I.e., you should have written, ā total error would be Ā±80 W/m2.ā
How can a slope be āĀ±ā?
Is Ā±80W/m^2 more than the observation? When can a physical error value be simultaneously plus and minus?
Never, obviously. The āĀ±ā is a dead-obvious statistical give-away.
The L&H 20 year annual mean is across 1986-2005.
Nick is wrong.
Sorry for not closing the html link
Pat,
“Why should units change in eqn. 6-2? Change from what?”
I don’t think they should. But again, here is 6.2
You are averaging cloud cover over n periods in a year. Yet the result is still in cloud cover units, with nothing in the denominator. Yet when in the almost identical next stage
you average cloud cover over N years. Then you say it must be expressed as CC units/year.
What if you had just averaged the n*N readings in one sum? If n is the same for each year, you’d get the same numerical result. What would then be the units?
Nick Stokes, āWhat if you had just averaged the n*N readings in one sum? If n is the same for each year, youād get the same numerical result. What would then be the units? ā
Averaged ān*N readings in one sumā over what unit-time scale? Choose your scale. Do the dimensional analysis. Let us know.
OK, suppose the n readings per year were monthly. So you average over 240 months. Same numerical answer. Are the units CC/year or CC/month.
Or again, for the 20 annual averages, what if you averaged the first decade, then the second, and then averaged the two decades. Again, same numerical answer. Are the units CC/year or CC/decade?
Nick Stokes, āsuppose the n readings per year were monthly. So you average over 240 months. Same numerical answer. Are the units CC/year or CC/month.ā
Sorry for the delayed response. Just back from travel, and opportunities were limited.
But letās parse your question, shall we?
Start with a monthly average of CC across a year as (sum of all 12 monthly aggregated observations/12). We have average CC that year. The average CC for that year is CC/month isnāt it.
Every month of the annual average now has identical CC, from Jan. through Dec.; the uniformity of CC imposed by the averaging process.
But why stop there? Average 365 days of CC. Every day of the resulting yearly average also has the identical average CC value. And that daily average CC/day value is identical to the CC/month value of the previously monthly averaged year. Is the new average dimensioned CC/day? Even though itās identical in numerical value to the CC/month?
How did CC/month transmogrify into CC/day?
How about averaging the CC every minute of every day of every month for the entire year. The same identical average CC value. Is the average dimension CC/minute, or not? Surprise of surprises, CC/minute value = CC/month value.
Is it confusing that the yearly averaged CC has the same numerical value at 1 minute as it does at 1 year?
The way out of the conundrum is to realize that the uniformity of CC average value at all subordinate time scales is imposed by an averaging process that extends across the year.
An annual averaged CC has some physical meaning if one wanted to compare annual average CC year-by-year, to look for annually significant changes.
The CC/minute value obtained from an annual average has no physical meaning. Comparison of minute 1, January 1 of annually averaged year 1 with minute 1, January 1 of annually averaged year 2 is physical nonsense.
The CC average value of a minute has no meaning taken from an annual average of CC.
The same discriminatory logic applies to your 240 month average. The 240 month averaged CC is of uniform intensity across that entire time at all internal time scales. Across the 240 year avarage, CC/second = CC/month = CC/year, imposed by the averaging process.
What would be the physical meaning of the 240 month average CC at one second resolution, or one month?
So, the answer to your opening question is āyes.ā But taking an average imposes uniformity of value across the averaged metric. The smaller dimensionality denominators are numerically valid but physically meaningless.
Thereās a distinction that goes to the heart of your continual misperceptions about physical error analysis, Nick.
The CC/month of a 240 month average has no physical meaning, except on a 240 month scale, and perhaps in the context of comparing alternative 240 month averages. The significance is at 240 months, not one month. The uniformity of CC value at all internal subordinate time scales is a numerical artifact of the method; one that has no particular physical significance.
No one would be particularly interested in seeing how a 240 month average changed in terms of CC/month because over a 240 month time range, the imposed monthly denominator transmits a false claim of resolution. It has negligible physical meaning.
Now letās look again at the L&H error metric.
Let L&H calculate twenty annual errors as annual (CC_mod – CC_obs). Then sum up the 20 differences, and divide by 20. The unambiguous result is an annual average error, of denominator year^-1.
That error is identical to the one obtained by separately averaging 20 annual CC_mod and 20 annual CC_obs.This yields CC_mean-mod and CC_mean-obs. Their difference produces the ideintical annual average error. The same per annum (year^-1) is physically meaningful because the annual resolution is physically real.
The reason is that each annual CC_mod and each annual CC_obs was physically unique. Their 20-year annual average therefore has physical meaning as an annual average of denominator year^-1.
Following from the above, L&H would also have a physically meaningless but numerically identical per month average error and a per second average error.
Those error metrics are physically meaningless because the resolution of the average does not produce physically valid individual months or individual seconds.
That completes the resolution of your posed numerical conundrum.
Itās interesting to note that if the satellite observations were sufficiently numerous and regularly spaced across every month of the 20-year experimental range, a per month rms LWCF CMIP5 model simulation calibration error statistic could conceivably be calculated.
It would probably not be Ā±4 W/m^2, but whatever it turned out to be could be propagated in monthly steps rather than annual steps, across an air temperature projection.
Whatever the centennial temperature uncertainty turned out to be, I have no doubt but that its message would be identical to that conveyed by the annual propagation. The one that yielded about Ā±15 K. The meaning is that air temperature projections are entirely unreliable right up through the CMIP5 versions of them.
Pat,
“Is it confusing that the yearly averaged CC has the same numerical value at 1 minute as it does at 1 year?”
Well, very little of what you said makes sense. But let’s make it really simple. Suppose you measure the temperature of a location every day for 20 years. And it turns out, unexpectedly, that it’s a location with constant temperature, 24°C.
So you average the days of every year, then average the 20 years. 24°C/year. Yes?
Or you average the days of every month, then average the 240 months. 24°C/month. Yes?
You and Forrest spoke learnedly earlier about dimensional analysis. In any such system, 24°C/year is 2°C/month. It has to be, if dimensions mean anything.
So what gives?
If it was a vector, you’d be right, its not, its a scalar.
Vector/scalar has nothing to do with dimensions. The magnitude of a vector is a scalar, but has the same dimensions.
It’s the differences between a rate and a scalar. If it was 24Ā°/yr as a rate change, it would be 2Ā°/month, but it’s not a rate! How hard is that for you to understand? And while they have the same dimension(but hold on), its not the same units. Ie 24Ā°/yr change, vs 24Ā° average/year.
So really they don’t have the same dimensions.
“differences between a rate and a scalar”
This is getting very muddled. rate and scalar are not opposites. 24°C/year could be a rate and is a scalar. But units are units. Dimensions are dimensions. They don’t embody any extra meanings like whether rate or not.
But that’s just it, they are not the same dimension, you’re describing delta T/time period vs Avg T/time period, they are not the same thing!
Nick Stokes, “Well, very little of what you said makes sense.”
Meaning only that you understood very little of what I wrote.
I pointed out that averaging across long times imposes the numerical uniformity that confuses you.
However, your present example is even more confused. You begin by proposing a site of constant 24Ā°. Constant 24Ā° means a reading every second will register 24Ā°. Averaging per second then yields 24Ā°/second, too, Nick.
This 24Ā°-no-matter-the-unit-dimension is identical to the conundrum I resolved in my prior post. The one you didn’t understand.
What makes your current example more confused than your previous one is that the constant 24Ā°/(dimension of your choice) is a direct consequence of your built-in assumption of constant 24Ā°.
Bottom line is that your system of constant 24Ā° yields 24Ā°/year, 24Ā°/month, 24Ā°/second, and 24Ā°/(internal duration of your choice). That’s the physical meaning derived from your own assumption. Your numerical playing about produces physical nonsense.
For example, for what physical reason would one divide a constant 24Ā°/year by 12 months, when it is known that the temperature during every one of those months was 24Ā°? The answer is: no physical reason at all.
Your division by 12 is physically meaningless, making your 2Ā°/month physically meaningless. What is the dimensional meaning of a physically meaningless construct?
Dimensional analysis doesn’t mean making physically nonsensical algebraic ploys. It means tracking dimensions through a calculation to be sure the final dimension is correct with respect to the equations used.
Nick, this quote from Einstein, from the Stanford University Online Encyclopedia may help you understand that the way to think within the physical sciences is different from your numerical consistency.
Physical meaning must, and always does, take precedence over philosophical (numerological) formalities.
“The reciprocal relationship of epistemology and science is of noteworthy kind. They are dependent upon each other. Epistemology without contact with science becomes an empty scheme. Science without epistemology isāinsofar as it is thinkable at allāprimitive and muddled.
“However, no sooner has the epistemologist, who is seeking a clear system, fought his way through to such a system, than he is inclined to interpret the thought-content of science in the sense of his system and to reject whatever does not fit into his system. The scientist, however, cannot afford to carry his striving for epistemological systematic that far.
“He accepts gratefully the epistemological conceptual analysis; but the external conditions, which are set for him by the facts of experience, do not permit him to let himself be too much restricted in the construction of his conceptual world by the adherence to an epistemological system.
“He therefore must appear to the systematic epistemologist as a type of unscrupulous opportunist: he appears as realist insofar as he seeks to describe a world independent of the acts of perception; as idealist insofar as he looks upon the concepts and theories as free inventions of the human spirit (not logically derivable from what is empirically given); as positivist insofar as he considers his concepts and theories justified only to the extent to which they furnish a logical representation of relations among sensory experiences (my bold).
“He may even appear as Platonist or Pythagorean insofar as he considers the viewpoint of logical simplicity as an indispensable and effective tool of his research.
(Einstein 1949, 683ā684)
Note the sentence, “Epistemology without contact with science becomes an empty scheme.”
That empty scheme produced your 2Ā°/month.
Contact with science shows your 2Ā°/month is physically meaningless.
“But thatās just it, they are not the same dimension, youāre describing delta T/time period vs Avg T/time period, they are not the same thing!”
This, and Pat’s stuff, is just nuts, at a basic level. Quantities have dimensions and corresponding units. It’s as simple as that. The dimensions don’t depend on why you are calculating it, whether someone thinks you should have been calculating it, whether it is constant, or whatever.
And there are simple rules relating dimensions and units. If X is 24 units/year, then it is 2 units/month. It doesn’t matter what kind of thing X is or what “units” are. 1 m/sec is 1000 mm/sec is 3.6 km/hr. You don’t have to ask further questions in making those conversions. You can look them up from tables. Or Google.
The problem isn’t that it accumulated 24 units/year, it’s an average quantity of 24units/year, not the same thing.
The difference of standing in a creek that’s 24″ deep, and one that’s normally empty, until it storms and swells to 24″, and then drains.
Nick Stokes, “This, and Patās stuff, is just nuts, at a basic level.”
No it’s not. micro6500 is right. Your October 31, 2017 at 9:30 pm post and prior treat 24Ā°C/month as the slope of a trend, when it’s the average of quantities (in your example, of a constant over time).
“Quantities have dimensions and corresponding units. Itās as simple as that. The dimensions donāt depend on why you are calculating it,…”
The physical meaning is the crux of the issue. That meaning depends on the physics of the system, Nick. Your usage is physically meaningless.
Slopes and statistical averages have the same dimensional form: magnitude/unit. But the one is not the other, physically. You have consistently and incorrectly conflated them. So did my reviewer.
When you first began this line of questioning, I thought you might be trying to pose a trick question. Now I see that I was wrong. You plain don’t understand what you’re doing, Nick.
Just as in your confused view that instruments of measure have infinite resolution.
Nick is correct, though. The 4 W/m^2 is the multi-model mean spatial root-mean-square-error of longwave cloud forcing between models and observations calculated over 20 years (I think the citation is Lauer & Hamilton, 2012). It is not 4W/m^2/yr, it is simply 4W/m^2.
I showed that revised Sec 6.2 from the SI.
“Consider: (20-year sum of annual uncertainties)/20 years = uncertainty per year.”
When you average, you don’t divide by 20 years. You divide by 20 units. It doesn’t change the dimension. If you average your weight over a year, it doesn’t come out as kg/year. If you average over 12 months, it isn’t kg/month. That’s elementary. Yet PF makes a big issue of it.
āConsider: (20-year sum of annual uncertainties)/20 years = uncertainty per year.ā
When you average, you donāt divide by 20 years. You divide by 20 units. It doesnāt change the dimension.
Er… if the uncertainty doesn’t have a time dimension, when did the uncertainly accumulate? Very little happens over zero time in any physical process to generate uncertainty.
paulclim is correct, the correct step seems to be one year.
(sorry, wrong subthread)
“Instead you went for the smear.”
Wow. I said his misuse of averaging was nuts. It is.
So what do we get from Pat Frank
“Itās not the trillions of dollars that would be lost to sustainability troughers.”
“Given all that conflict of interest, what consensus climate scientist could possibly provide a dispassionate review?”
“As a monument to the extraordinary incompetence that reigns in the field of consensus climate science”
” a noxious betrayal of science by the very people charged with its protection.”
And then to the editor of Journal of Climate
“So how does it work, Tim ā are you truly so dishonest, or merely stampeded into cowardice?
You must know very well manuscript JCLI-D-15-0797 is correct.
And that the reviews are incompetent.
When did science become the willful suppression of valid criticism?
When did science countenance a conscious defense of the false?
When did wrong become privileged?
When you sit at your desk, wouldnāt professional dishonesty rob your life of meaning and purpose?
Yours sincerely,
Pat Frank”
Nick,
Your citation (sec. 6.2) notes that that Cug has dimensions of āmean cloud cover unit per year yet you insist that the dimension does not include āper yearā.
Also, you have cited examples of averages where the āper unitā (year, month, day) dimension does not apply yet your citation is a mathematical average and it is defined having dimensions of āper yearā. Please explain.
ATTP, you describe the variable in question as being a āmulti-model mean spatial root-mean-square-errorā. The equation cited by Nick, sec 6.2, is a simple average; giving it a fancy name doesn’t change that but it may intimidate those who donāt recognize the form of the equation. Nice try but still a FAIL.
Ray,
“Your citation (sec. 6.2) notes that that Cug has dimensions of āmean cloud cover unit per year yet you insist that the dimension does not include āper yearā.”
I’ll show the original citation again
Says the result of simple averaging, Cug, has units cloud cover/year. It doesn’t spell out the units of the thing averaged, Ckg, but it refers to Eq 6.2 as the origin. Here is eq 6.2:
Dimension cloud cover unit, not per year. Bizarrely, that expression is also an average over sub-units of time in a year, but does not acquire a per time unit.
“The equation cited by Nick, sec 6.2, is a simple average; “
Yes. It isn’t what ATTP is talking about, but that is all it is. And yet, as the ref said, he got it wrong. And it matters because it is where the notion that errors have to be compounded annually came from.
Nick Stokes, āBizarrely, that expression is also an average over sub-units of time in a year, but does not acquire a per time unit.ā
Itās pretty simple Nick. Eqn. S6-1 and S6-2 provide the average of cloud cover per grid-point in each given year.
The number of simulated cloud cover values per given year need not match the number of observations per given year.
Eqns. S6-1 and S6-2 sum up the observed or simulated values, respectively, in each given year and take the average magnitude for that year.
Simulated and observed cloud cover units must eventually have identical time scales, in order to make the difference in eqn. S6-5. Hence conversion to the uniformly annual scale.
The equations are not time averages. They are number averages. Any time sub-unit (simulation time-step; observational time interval) is lost because the number average scales to the annum in each given year.
It really is that simple, despite the complicated hash youāre trying to make of it.
I think the real source of the problem youāre having is in dealing with gritty Galilean physics while being trained to Socratic idealism.
Adding: Nick Stokes, āAnd yet, as the ref said, he got it wrong. ā
No, Nick. The ref got it wrong, and so have you (and ATTP).
The metrics are key to at least part of the discussion. Get the metrics right first.
Jan says it well. There are different kinds of error: experimental errors are inherent in the result and have to be propagated through subsequent calculations. Systematic errors (using an incorrect forcing value) can be identified which means corrected. The ISO has a lot on GUM and standard reporting structures for it. There is no way climate model outputs are ISO compliant because they do not properly report the propagated uncertainties. They treat the calculated number as golden perfection.
Nick has persistently indicated he doesn’t think uncertainties propagate.
The question of an uncertainty per year is interesting. If at the beginning there is no uncertainty at all one has perfect measurements or perfect assumptions. Fine, let’s presume that is the case. After 1 year there is an uncertainty of 4W/m^2 it can stand alone. If that model is run for another year carrying an uncertainty about the ‘inputs that arises from Year 1, that uncertainty has to be propagated through year two.
Suppose the uncertainty were assessed after 30 minutes of model run, (starting with perfect inputs). Whatever uncertainty arises comes from what exactly? Doesn’t matter. Black box it.
The big problem is the misunderstanding that a range of modeled results does not represent the uncertainty about the answers. The propagation of error (uncertainty – look it up) goes with each calculated output. So what I see above is aTTP trying to sell variability of model outputs as the uncertainty attached to each result. I am not sure who is convinced by such a misrepresentation. One might get 1000 different model results that were 100% certain as to their value. There is a fundamental difference between the range of results and the uncertainty of each result, which depends on the uncertainty of the measurements/inputs.
Michael Mann produced a mathematical calculation that assembled a lot of tree ring data to produce a proxy temperature series. Whatever date was fed in from any set of trees, the result was always a (rather convenient) hockey stick shape. This is an example of a mathematical process in which large differences are observed in the inputs, but the result was always the same. In such a case the variability in the result is small. Mann’s results were quite non-varying. That in no way means his results were ‘certain’.
Crispin,
“They treat the calculated number as golden perfection.”
As they should. They are reporting a calculation, and the only relevant error would be machine rounding. The point of an error estimate is that it tells you the range of results you might have got if different circumstances had affected the measurement. Different instruments, different temperature, different observers etc. But with a model calculation, you don’t have to speculate. You can do it again and again with those changes, and actually see the range of results you get. That is the point of an ensemble.
“There is a fundamental difference between the range of results and the uncertainty of each result, which depends on the uncertainty of the measurements/inputs.”
No, again the uncertainty is actually just an estimate of the actual range of results you might get if the inputs varied over a range of uncertainty. That’s why an ensemble is the proper way to work it out. It’s always possible that not everything that should be varied will be, but the same is true of any error estimate.
What you won’t see is the IPCC or similar presenting just one model result and saying that this is golden perfection. They always present a range. OK, Hansen 1988 was an exception, but that was because they only had one. You have to start somewhere.
If you average 20 years of model temperature error, you get error/year.
Magnitude average vs. statistical average, Nick. You’ve got no idea. Neither did that reviewer.
As Forrest Gardener noted, you quoted the reviewer as though his view were definitive, without addressing my reply. That’s to lie by omission, isn’t it.
That reviewer also did not realize that one need not have a trend of calibration measurements to combine them into a calibration error statistic. The calibrations need only reference the same instrument, method, and conditions.
That reviewer implicitly assumed the statistic dimension, property/unit, is a slope. it’s not.
Pat,
“Magnitude average vs. statistical average, Nick.”
Gobbledygook. There is only one kind of average, and you spelt it out here
And it does not somehow acquire new units. If you average your height over 20 years, you get height in feet. Not feet/year. If you average the annual averages of your height, you get the same answer. And in feet.
The mean of 20 annual-average-errors is both the mean annual-average-error and the average-error per year. They are one and the same.
RW: i see
you didn’t respond
to the comments
I made today in reply
to yours.
crackers345: I didn’t see them until now.
If I average the change in my height over 20 years it’s inches per year. Nick.
And the errors recorded in L&H are likewise differences over time.
Thank-you for quoting my final email, to Tim DelSole Nick.
I invite everyone to read the two J. Clim. reviews and my responses. Between them, the two reviewers made pretty much every one of the 12 mistakes listed in the head post.
They are utterly incompetent.
Review #1 was so poor as to suggest the reviewer never critically read the paper at all.
Reviewer #2, among other fatal mistakes, also inadvertently validated the analysis (giving you something in common with him, Nick).
Tim DelSole is a trained physicist. He should never have accepted those reviews. But he did.
He wrote, āOn the basis of these reviews and my own evaluation, I am sorry to inform you that this manuscript is rejected.. (my bold)ā
He rendered an independent review judgment but provided no reasoning. In view of the incompetent reviews, his rejection stands without justification.
His decision amounts to a star chamber proceeding.
Youāre right. I was angry. The whole thing is a profound betrayal of science.
Consensus climate science is a parody. And its professional integrity is trashed.
And you support all of that, Nick, without knowing what youāre talking about.
Someone else who doesn’t know the difference between a magnitude average and a statistical average.
And who considers himself a scientist. Too funny. š
Crispin, “They treat the calculated number as golden perfection.ā
Nick, “As they should. They are reporting a calculation, and the only relevant error would be machine rounding.”
Thus spake the numerical modeler.
Nick’s comment ought to settle the case for every single physical scientist and engineer here.
Nick shows no awareness that the accuracy of calculations from physical models are conditioned by comparison with physical data.
But he wants us to know that he can critically evaluate physical error analysis. And ATTP supports him.
Incredible.
Yep, I have been watching. Its been quite funny to watch them both making exactly the same mistakes as the reviewer, all because of their complete lack of understanding about error propagation etc.
“Reviewer #2, among other fatal mistakes,”
The reviewer (J Clim) began thus:
“This manuscript is an impostor. It pretends to be a research paper, but it is not. A research paper makes a good-faith effort to provide relevant context as defined by prior work and then goes on to show how the prior work is extended (perhaps by showing flaws in it). By contrast, the present manuscript uses scientific-looking reasoning and referencing but blatantly ignores or misrepresents prior work. Furthermore, the manuscript contains fundamental errors in the technical development it claims to present as novel. Hence I cannot but recommend rejection.”
Sounds accurate to me.
” know the difference between a magnitude average and a statistical average.”
You just make this stuff up. So what is the difference? With references, please.
Using the weight example. So i weigh myself periodically, different scales, the numbers bounce around week to week. Maybe i got fat then slimmed down again over a monthly time frame and the measurements reflect that trend to some degree… I calculate the standard deviation over the year. In the absence of any other information, there isn’t a big problem applying the standard deviation to any single time point in the year. I do rely on the assumption that the errors are random and distributed Normal. I also rely on the assumption of homoscedasticity. That is, that the error, whatever shape it assumes, remains that shape across the trend; put yet snother way, that the errors are independent of the trend.
RW: you’re
confusing two
different
uncertainties — that of
the measurements, and
the statistical uncertainty.
crackers345. I rebutted the notion that an SD calculated across a time series cannot be expressed in the temporal units of the data. I might have posted in the wrong subthread though.
Nick Stokes āYou just make this stuff up. So what is the difference? With references, please.ā
Who needs references when a simple demonstration is available?
Several measurements of the height of one person: meters. A measurement average.
Average height of people in a room: meters per person. A statistical average.
Middle school math, and a numerical methods PhD stumbles over it.
Pat,
“meters. A measurement average.”
This was the distinction in question, quoting your words:
ā know the difference between a magnitude average and a statistical average.ā
Still wondering.
Although this was one for the ages:
“Average height of people in a room: meters per person. A statistical average.”
So if you have three thermometers in a room measuring temperature, and you average, does that make the result °C/thermometer?
Nick Stokes
Your example misses the crux.
Three thermometers in one room ā presuming uniform air temperature throughout ā are measuring a single variable. Average result in K. Magnitude average.
Three thermometers, one each in a separate beaker of water, each measuring a separate variable. Average result in K/beaker. Statistical average.
Same deal with the L&H simulations. They are not N versions of one simulation. They are N separate and distinct simulations. The average is a statistical average, not a magnitude average.
Pat,
“presuming uniform air temperature throughout ā are measuring a single variable”
So the units depend on what you presume? What if you don’t know?
Should the average temperature of the US be in °C/thermometer?
Nick Stokes I was explicating the assumption hidden in your example.
If the temperatures in your example are not uniform across the room, then the average measurement would be K/measurement.
The differences from the average would be a Ā±K uncertainty in the temperature of the room.
So, without your hidden assumption, your example disproves your case.
In your US temperature case, the thermometers are not measuring a uniform variable. Theyāre measuring different variables ā different local temperatures in different environments.
The uncertainty that applies in their average, apart from state variability, is the uncertainty due to field calibration error. Something I published about (870 KB pdf).
What is so hard about the distinction between combining repeat measurements of one thing, and combining many measurements of many things?
Pat
“In your US temperature case, the thermometers are not measuring a uniform variable.”
But the question is, what are the units of the US average T. Never mind he uncertainty at this stage. It sounds like on your definition they should be °F/thermometer. But what if they averaged states first. Is it °F/state?
Nick Stokes, it would all depend on scale, wouldn’t it.
L&H started out with error/grid-point for each model, didn’t they. They combined the errors to finally scale up to global average annual error, i.e., average annual error/globe.
You can choose any scale you like. Your final scale could be (average annual ā°F)/(continental US), i.e., US annual average temperature in ā°F.
Pat,
“Nick Stokes, it would all depend on scale, wouldnāt it.”
So it depends on uniformity (of temperature) and then scale. Sounds a lot to teach those middle-schoolers. But there is no consistency here. With heights in a room, you wanted to say the average was ft/person. Now you seem to want ft/room.
“L&H started out with error/grid-point for each model, didnāt they.”
Actually, no. They describe their model aggregation here:
Each x is an average of a quantity (not error) over 20 years. They then form the mean differences, which is their measure of discrepancy.
Forrest Gardener, your statements are equivalent. But in your first, the āper personā is implicit rather than explicit.
If the average height of the people is 1.8 m, then the average height of all the people is 1.8 m/person.
More quantitatively, (sum of N heights)/(N people) = average height/person.
When doing some analysis quantitatively, one has to be explicit about the dimensional analysis of the equations used.
If intermediate dimensions do not cancel to leave only the proper dimensions of the final unit, then the equation is wrong.
Nick Stokes, āSo it depends on uniformity (of temperature) and then scale.ā
No. The central issue of your example was not uniformity of temperature, as you suggest, but whether the measurement is of a uniform system. Your āuniformity of temperatureā conflates a single system at a single temperature with diverse systems at the same temperature.
They are not physically identical at all. Their averages are not the same either. The latter produces a statistical average.
Obviously the dimensional unit changes with the scale.
This isnāt āa lot to teach those middle-schoolers.ā but it does seem to escape certain numerical modelers.
Nick, āBut there is no consistency here. With heights in a room, you wanted to say the average was ft/person. Now you seem to want ft/room.ā
Youāre just pretending to be diversionary, arenāt you. Suppose you have 12 rooms, with 20 people per room. If you wanted an over all height average, why would it not be ht/person/room? You remember ā as in W/m^2/year.
Nick, āEach x is an average of a quantity (not error) over 20 years. They then form the mean differences, which is their measure of discrepancy.ā
Now youāve stopped playing at it. I wrote that L&H began with the error per grid point. Their [x_i(mod) minus x(obs)] is an error metric.
You dispute that and now rename error as a āmeasure of discrepancyā, as though that were something different.
Get serious.
“Itās a long time since I did any dimensional analysis.”
Obviously. If you brush up on it, you’ll be reminded that
” dimensional analysis is the analysis of the relationships between different physical quantities by identifying their base quantities (such as length, mass, time, and electric charge) and units of measure (such as miles vs. kilometers, or pounds vs. kilograms vs. grams) and tracking these dimensions as calculations or comparisons are performed.”
“person” is not any kind of base quantity.
well maybe it matters when trying to compare against rooms full of virtual people who are getting added to some because they didn’t have enough with real people to measure.
When half your measurements average doesn’t even exist, maybe /thermometer should be added.
Nick Stokes ““personā is not any kind of base quantity.ā
āpersonā does not designate an individual human being? Is Australian English (Strine) so degraded as to have lost the ability to assign simple meaning?
Forrest, you’re right. Apologies to you and Australia.
Nick, however, does grave disservice to your (his) nation.
Follow Willis on Twitter https://twitter.com/WEschenbach
In the computer age it would not matter which time scales you calculate. Even minutes went, one would have these data available. It depends on whether (as it is obviously the CO2 believers do) see CO2 * H2O in every temperature change of the past and then project it to the future. I’m not saying the model is fault-free. No model is this. However, many made it through the review, which are obviously flawless. Even though the average of models in an El Nino Uptick was recently presented by Gavin as evidence of the longer-term accuracy of the models.
However, in the longer term it looks as follows:
And extending this lower level extrapolation backward into the LIA indicates no attribution to CO2. It is a fabricated non-catastrophe! Utterly!
the LIA wasn’t caused
by CO2. its end was due to
three main factors – decline in volcanic
aerosols with decline in ice-albedo
feedback; some solar in first half
of 1900s; and GHGs — CO2’s forcing
changes fastest in the beginning.
crackers345
Name the volcanoes and their eruption dates that were “high” during the Medieval Warming Period (you know, that hot period around the world that cooled off into the LIA) that caused the LIA to “end”. See, when volcano activity is “very high” the extra soot and ash and gasses n the sky COOL the average global temperatures. So, the “lack of atmosphere contaminates cooled the world after they caused it to warm? And you need to name the volcanoes and the eruption dates that cooled things between the Minoean Warm and the Roman Warming Period and then again between the Roman era and the MWP.
The LIA was at its deepest (lowest average global temperature) in 1650, then temperatures have been gradually raising ever since with a characteristic 60-70 short term cycle. So, what exactly does “solar” changes in “the first half of the 1900’s” have to do with a temperature change from 1650 to 2000? See, there was no substantial CO2 change globally between 1650 and the 1930’s and 40’s.
And, in fact, most of the CO2 change has been DURING the period when the earth’s global average temperatures have changed the least!
crackers.. can you give some data showing the rise in volcanic aerosols which created the LIA in the first place, say 1300 to 1700?
There seem to have been quite a few equatorial volcanic eruptions during the period of the LIA – starting with the 1257 eruption of Mt Samalas in Indonesia….
https://www.nature.com/articles/srep34868
http://notrickszone.com/wp-content/uploads/2016/04/Volcanoes_4.jpg
RACook – the volcanic eruptions were
1250-1275 AD. they’re shown in this paper:
Gifford H. Miller, Geirsdottir, A., Zhong, Y., Darren Larsen, Otto-Bliesner, B. L., Holland, M. M., Bailey, D. A., Refsnider, K. A., Scott J. Lehman, Southon, J. R., Anderson, C., Bjornsson, H., Thordarson, T. 2012: Abrupt onset of the Little Ice Age triggered by volcanism and sustained by sea-ice/ocean feedbacks. Geophysical Research Letters, 39: L02708. DOI: 10.1029/2011GL05016
It may not be that the money and the fame keep such papers out, it may be the faith in a belief that results in defending it no matter what. Ask what would convince a true believer that he/she is wrong. My guess is, there is nothing that would. Meaning it’s not about science. So not allowing the publication of material is merely an expected reaction to challenging that which the editor cannot fathom ever to be false. Extreme faith in the belief means protecting it at all costs.
Until a few years ago (until the climate hype), there was a publicly accepted (even from Hollywood) law in the whole of science. Every outsider opinion was at first pitted mercilessly and may it still be right on closer examination. This was publicly known and accepted, so that in the public no “final and all-knowing” science was ever perceived. It was only with the advent of climate science as a media Darling and a vehicle for re-education and social change that this changed. Now, no more public perceptions were perceived and accentuated. One lives in a bubble of self-reinforcing reflection. But every evil has its good, now people turn away from faith. Not because they think that science is not mature (they are already being re-educated), but because they have to solve other more urgent problems. Thus, the 97 percent mentality has developed as a poison for the more advanced goals of the vehicle climate science and for those who are behind it.
I suggest you all see Patrick Frank, PhD presentation at the 34th Annual DDP meeting, July 10, 2016, Omaha, Nebraska.
And I suggest you watch ….
“https://www.youtube.com/watch?time_continue=2&v=rmTuPumcYkI”
Toneb,
Dr. Brown is very sloppy in his attention to significant figures in his calculations. Demonstrating that he is not the sort of person who pays attention to detail, I would be inclined to check all his calculations.
In his last example, where he re-visits the problem of how far the person has walked, in order to address the issue of base-error, if the question is re-framed as to where the person ends up, then the calculation used by Frank is appropriate. And, in the context of this blog, it really is more important to consider what the final future temperature will be (and the uncertainty of it) than how much the temperature will increase over a period of time.
You are kind to Mr Brown, Clyde Spencer.
He says that +/-4 W/mĀ² is a time invariant, while it is NOT. A +/-4 W/mĀ² on 20 years means a +/-1 W/mĀ² on 5 years or +/-0.2 W/mĀ² on 1 year, or +/-20 W/mĀ² per century
And this way things add up finely: wheter you use a 20 step model with +/-0.2 W/mĀ² uncertainty per step, or a 1 step model with +/-4 W/mĀ², in both case you have a +/-4 W/mĀ² uncertainty.
Provided, however, you can reduce uncertainty when you reduce step size.
Trouble is, for climate modeler, you cannot. Uncertainty is certainly no less than +/-0.1 W/mĀ² on a single day, and that’s already to big a step for GCM do work
Toneb, you pointed Dr. Brown’s youtube version.
Try looking here, the original video at his personal site, and the discussion we had below it.
His argument does not withstand critical scrutiny.
Who knew that plant food appears to be the boogeyman of the century in that it merits such attention? Like saying bunnies are horrendous and we must study each soft fiber in their fur.
Good discussion.
Great work Pat.
Good luck! Never before has so much policy been driven by mere curve-fitting.
The computer games are obviously a joke…..and yet, people not only defend them but the science behind them…..and they get away with it!
Pat ….. just giving 2 cents worth. It seems your assertion of propagated linear error is in connection to the article published here about how the temperature is just the result of a random walk. As such, the climate scientist canāt acknowledge your paper, as in their mind, CO2 is actually forcing the climate. I tend to agree with the random walk theory for the small changes in temperature noted over short time periods, but not for the changes that clearly appear as cycles, such as ice ages. But again, CO2 is not relevant to the long term discussion, nor is modeling. This whole model BS exercise is just part of an agenda against fossil fuels. It doesnāt need to be accurate or precise, nor does it have apply to reality.
Good luck.
I don’t know whether it was all this, I suspect a few scienctist adopted the incorrect view of the climate, and thought they were saving the world in the best of the Saturday afternoon B flick, but I know the FF haters jumped on this like as soon as they figured out the scientists didn’t might being used for propaganda, Which they didn’t, because they thought they were trying to save the world.
it is nothing more than virtue signalling of the highest order. the entire community is too far down the rabbit hole to even consider any other view point. this nonsense has and is being taught to pre teen children in schools ,what hope have the next generation of climate scientists got of being objective at the outset.
Yes, it is a campaign, served from Billionaires with interest in new “renewable” energy. Follow the money. CNN international yesterday from Tom Steyer: “In the past there are presidents impeached for far less”. Guys, that’s it. Pres. Trump is dangerous to the shops at this level and now an impeachment is to be created for intimidation. It was not said, however, what the “far less” or even the reason should be. Does not this remind you of the per reviewer of Pat?
Thanks, Dr. Dean – though my error analysis concerns how climate models project temperature. It has no bearing on what the physically real temperature does.
of course, every author thinks
his/her article is 100% correct and vital. why
don’t you
publish the full reviews
so we can
see what the reviewers wrote in
detail?
Gee Crackers and Mark, Nature published an error filled Dr. Mann “hockey Stick” paper,way back in 1998. It is now known as one of the most critically smashed paper in climate research history.
Carry on with your hypocrisy.
it’s the exact opposite — many studies have
replicated a HS.
indeed, crackers345, “many studies have replicated a HS” out of data known to have no trend at all, just by applying the faulty statistical novation of gavin. That the point that destroyed the HS even for IPCC.
He published the full review. It is in the zip file, link in the article.
To me it seems the reviewers brought up arguments without thinking about it and without checking the sources, like the one discussed above with the Cloud Forcing uncertainty. In all reviews I cross checked I found the same thing. Weak arguments in the reviews, easily and clearly debunked by Pat Frank. It looks like the result of the reviews was clear from the beginning on. The procedure just needed to be followed.
In my eyes that conspiracy is even more fascinating than the uncertainty monster which everyone assumed anyways.
Pat there are a lot of boutique journals that will publish your work for a fee. Just pony up and PRESTO! It will solve all your problems with rejection.
So it doesn’t matter that these journals give bogus reasons for rejecting a paper, because there are boutique journals.
Your assertion that the reasons are bogus is “hand waving.”
does it take a hand waver to know one i wonder ?
Mark S. Johnson, take a look at the incompetent reviews, link provided in the text, before making facile dismissals of bogosity.
@ATTP. Your quote “So, if someone wants to argue that the range of possible temperature is 30K (as appears to be suggested by Pat Frankās error analysis) then one should try to explain how these states all satisfy the condition that they should be in approximate energy balance (or tending towards it).”. No one is remotely arguing that. The point is that the error propagating through the modelling process results in a range of possible temperatures that is very large because of error propagation, not because the actual physical range of possible temperatures is that large.
The point to remember about climate models is that they cannot be used as proof of the CAGW hypothesis because the assumptions of the CAGW hypothesis are programmed into the models. They are merely another representation of the theory, they are not proof of anything.
dbakerber,
You said, “They are merely another representation of the theory, they are not proof of anything.” To expand on that, they are actually very complex working hypotheses that have been formalized by coding the assumed mathematical relationships. In an ideal world, where the Scientific Method was followed, the models should be subject to comparison and validation by empirical evidence. If necessary, the models should be adjusted (revised working hypothesis) to achieve better agreement with reality. Instead, we are basically told that the “science is settled,” and that the models are reliable. Yet, even today, Einstein’s theories are still being tested by replication.
Einsteinās theories are still being tested by replication.
===
To date not a single prediction of GR has proven to be wrong.
In contrast climate science doesn’t make predictions. They make projections. Thus the need to add “science” to their name. Like political science and christian science.
Science Bulletin rejected your paper for insufficient “novelty and significance”. It is clearly significant.
One can only conclude they thought it was not novel.
Absolutely!! His findings are not novel to them because they understand that their models are not correct, that is why they need perpetual funding to keep fixing them. What they should have just written in their rejection is “You’re not helping.”
It’s wrong, mostly, for reasons explained over and over and over again.
Frank’s model for the error propagation would mean that the temperatures should go below 0K in a few centuries. Amongst the much more detailed criticisms provided by others, this provides a basic sanity check showing that his understanding is incorrect.
What happens in the real world (or the models) when the cloud forcing is 4 W/m2 lower? The temperature drops a little, and then the outgoing radiation decreases, and the temperature stabilizes. It doesn’t drop to 0K. It has a static effect on temperatures.
Pat Frank basically treats the static uncertainty of 4 W/m2 as an expanding uncertainty, as if the cloud forcing could keep changing each year by another 4 W/m2 until the Earth froze over or boiled off. This is not realistic, neither in the models or reality. Basically: he’s treating the uncertainty as W/m2/year, rather than W/m2.
And that’s why his paper was rejected. Because it’s wrong.
Forrest, I’m explaining how a static uncertainty in the cloud forcing does not propagate through the models, nor through the real world. And I’m explaining how Pat Frank made his mathematical mistake; essentially by changing the units from W/m2 to W/m2/year.
Stop and think about it. Do you think that a static uncertainty in the cloud forcing can mean that the temperature could be anything? Does that make sense in reality, or in the models? What about the restorative forcings?
OMG, Mr. Winchester. “Frank’s model … would mean that temperatures should go below 0k….” No, Frank’s “model” (I don’t think it’s a model at all) has nothing to do with predicting what temps will do. If I understand correctly it is the flawed model, not temps, MODEL TEMPS, that would go to 0k. IF there were proper error bars showing uncertainty (as distinguished from known errors) with these models, the ERROR BARS (THUS THE MODELS) would go to the impossible realm of heat/cold, not actual temperatures. If the model does that, then it is flawed and cannot be corrected by “adjusting parameters” (newspeak for “tuning”). I’m not a scientist or mathematician, but it’s clear to me. Are you really as thick as ATTP?
Treating it annually is correct. It is uncertainty per year. How far it drags temp off in one direction or the other after x many years depends, mathematically, on the model – the variables in the function – and the values of the parameters – the variables in the function. I think Pat Frank is doing error propagation for functions.
If, after doing this, the range of temperature projections goes ‘off the charts’ into historically unprecedented teritory, a real scientist concludes that there is something about their model that is probably REALLY EFFING WRONG!!!
Since Pat Frank’s model emulates the temp outputs of climate models, and the climate models do not properly propagate the error, the direct implication is that the climate models, when properly propagating error, would show equally as unacceptable error bars, requiring one to conclude that the climate models are also really effing wrong.
If it were me, I’d actually use the historically ridiculous outcomes of my model to help guide the selection of parameters and the values of the parameters. Again, error propagation depends on the function itself.
Doing a proper error propgation could actually be used to help fine tune the climate model. It could turn out that the fine tuning (which results in non-off-the-chart error bars) conicides with models that are far less sensitive to changes in GHGs.
Benjamin Winchester, “Frankās model for the error propagation would mean that the temperatures should go below 0K in a few centuries.”
It means nothing of the kind. You’re supposing an uncertainty statistic is a physically real temperature. Fatal mistake.
BW “as if the cloud forcing could keep changing each year by another 4 W/m2 until the Earth froze over or boiled off.”
Wrong again. You’re treating a calibration error statistic as though it were an energetic perturbation. Fatal mistake #2.
Take comfort, though, because many PhD-level climate modeler reviewers did the same thing. You’re qualified to be a consensus climatologist, Benjamin, but not to be a scientist.
I always enjoyed dialogue between RPS Gavin.
https://pielkeclimatesci.wordpress.com/?s=Gavin+Schmidt&submit=Search
“I always enjoyed dialogue between RPS Gavin.”
There will be less of it now.
https://twitter.com/ClimateOfGavin/status/921894302283894784
And how did the dialog go? Why was he blocked? And why would that lead to the dishonesty of not citing them? On that single statement alone, he destroyed an “ocean” of trust.
Gavin uses the sneaky approach, he uses the “mute” feature on his Twitter account. At one point, Gavin had blocked me, but now I’ve determined he’s used “mute”.
BTW Gavin is not without his own set of sins, as this video shows his cowardice on full display, when asked to appear on stage with Dr. Roy Spencer.
Anthony Watts
October 23, 2017 at 4:14 pm
A bit to much to be forced and subjected to the “science” of a NASA scientist that requires indisputably the acceptance that actually science and “total” truth are compatible, in expression and method.
But then that is Gavin and his “science” of total truth, and in the same time the contemplation of some uncertainty there still considered by Gav !!!!!
Gavin the sciencie drama queen, maybe……
volcanic models suffer from the pneumonoultramicroscopicsilicovolcanoconiosis
Vuk,
Well isn’t that just supercalifragilisticexpialidocious!?
On this discussion of error propagation through iterative numerical analysis I am not qualified to critique or assess.
But I do know without a doubt this:
the implementation of all of the GCMs with tunable parameters (for real physical processes that are smaller than their grid scale, like convection, precipitation, cloud formation, you know, those pesky trivial things) makes them nothing more than tuned to expectation. They are simply confirmation bias by the modellers who tune them at every run. If they get a wild, bad run (too hot, too cold) happening, they reinitialize, tweak and start again. All of that is by their own admission on the tuned parameterization.
And then they combine them in an āensemble meanā to give them some false patina of validity and confirmation.
GCM tuning is junk science and their GroupThink completely blinds the model community to this reality that surrounds them like a fetid swamp.
Dear Anthony,
Try a Chinese journal, they are openly amenable to climate skepticism.
And before anyone says āChinese…!!ā, as if we are talking about Amazonian nomads, please remember that it is the Chinese who manufacture all of your computer gagetry. In addition, to allay western prejudice, they always have two western reviewers, in addition to their own.
Ralph
It is not Anthony. It is Pat. He must try no “chinese journal” but instead he worked out the pour per review process. “Pfui Deibel” on german.
I try to keep this as simple as possible but not simpler than that.
The IPCC’s climate model is the same as eq. (1) when its parameters are based on the choices made by the IPCC
dTs = Ī»* RF (1)
where Ī» is 0.5 K/(W/m-2) and RF as specified by Myhre et al. This is a linear dependency.
Eq (1) gives the climate sensitivity of 1.85ā°C, the IPCC’s official value is 1.9 ā°C Ā± 0.15 ā°C; In the AR5 has been tabulated the TCR mean of 30 GCMs to be 1.8 ā°C (Ā±0.6 ā°). The high variation in the TCR value of GCMs is due to various Ī»-values applied in the models and they are due to various feedbacks applied.
The most complicated computer models and the simplest model give the same results from the CO2 concentration of 280 to 560 ppm. This means that Pat Frank is right: GCMs are built on the linear relationship between the GH gas forcings and the surface temperature.
Dr. Antero Ollila
P.S. When I sent my manuscript of reproducing the radiative forcing of CO2 to Science and Nature, the editors rejected it immediately by saying that this is no interest for our readers. Actually it is one of the cornerstones of the present global warming theory.
Link: https://wattsupwiththat.com/2017/03/17/on-the-reproducibility-of-the-ipccs-climate-sensitivity/
It may be a corner stone, but can you claim that your result was both new and important?
There are thousands of papers which are published on climate change that also are neither new or important. As long as they repeat the mantra of death to earth then they get published.
“New and Important” is just the excuse rather than being honest and saying “Not conforming to the narrative.”
The errors associated with “fudge factor assumptions” are far greater than any propagated statistical errors.
That is what I thought, but really, really can’t decide based on this.
I would agree with the poor understanding of reviewers.
Two of my reviewers stated that CO2 (ie atmospheric pressure) did not reduce by 40% at 4,000 m. And one stated that āthe concentration of CO2 at altitude is the same as at sea level, and so plants cannot be starved of CO2 at altitudeā.
Both have misunderstood the difference between concentration and partial pressure, even though the figures were clearly marked in micro-bars. I did ask if they would be short of oxygen at the top of Mt Everest, the same as plants would be short of CO2 there, but got no reply.
The review process is certainly broken, because there is no discussion and so reviewers can remain in their bubble of misunderstanding. At least a blog review will throw up plenty of discussion, from many views, and deliver a better understanding.
Ralph
The internet has the real prospect of destroying the scientific journals’ grip on peer-review and thus their business model. The internet as it has already done to the main-stream Newspapers and News magazines precipitous declines in circulation. They cannot control the information flow anymore. Authoritarian regimes (of the Left and Right politically) are desperate to control the internet (as China, Russian, Turkey governments are doing).
Pat Frank’s manuscript will now receive a wider readership (outside of the GroupThink-controlled GCM community) here at WUWT and on other reposted blogs than if the GCM community had simply let it be published in a small, low impact, low circulation journal.
arXiv.org is a place where many physicists place their initial manuscripts for critical review and author replies before submission to a journal.
https://confluence.cornell.edu/display/arxivpub/arXiv+Governance+Model
Maybe Pat Frank could get it to arXiv.org to force an open review by his anonymous reviewers?
Joel O’Bryan, PhD
Only to the point, if people will accept it. They just say well it’s not been reviewed, no attempt to review the physics to see if there is a flaw in that logic, just rebuff.
I know https://micro6500blog.wordpress.com/2016/12/01/observational-evidence-for-a-nonlinear-night-time-cooling-mechanism/
This explains the surface temperature regulating mechanism of Water vapor, and how that is what’s controlling surface temps. And co2 has little to no role.
“Pat Frankās manuscript will now receive a wider readership “
There is no evidence from the discussion here that anyone (except attp and me) has actually read it, let alone critically. Just the usual knee-jerk, all-purpose stuff.
Nick,
There’s tonnes of peer-reviewed papers over the years that have turned out to be crapola. No malfeasance, nefarious intent, nor bad ethics needed. Peer-review science papers is not the Gold Standard the naive public thinks it is.
What is paramount to the Climate Alarmist establishment though is the suppression of ideas and analyses that run counter to or interfere with the public’s acceptance of economic control goals embodied UNFCCC COP agreements. (Kyoto, Rio, and now Paris agreements).
The GCMs declaring an inflated CO2 sensitivity are the central pillar of the Alarmist’s Big Socialism tent. Without that everything they’ve attempted for the last 30 years toward economic control via energy resources unravels.
…and Pat’s paper is blasphemy. Modellers hate it because it shows them to be mathematical fools, whereby even if his error propagation idea is wrong, the realization by a wider audience that supercomputer-run GCMs are simply massive, expensive Rube-Goldberg implementations of y=mx+b cannot be avoided.
“There is no evidence from the discussion here that anyone (except attp and me) has actually read it, let alone critically”
There is NO evidence that you actually understood it.
Your comment say.. nope.!!
You make all the same mistakes that Pat has listed.
Nick,
I read the paper and the large zip with reviewer comments and the PF replies to them.
I have not posted here yet because that reading takes a lot of time.
Thus, your assertion that only 2 people have read the paper is wrong. Absolutely completely, demonstratively wrong.
What do you do when you are found to be wrong?
As you know from blogs over the years, when I am wrong I acknowledge and correct. Geoff.
Geoff,
I said there was no evidence from the discussion here that anyone had read it. You had not contributed to the discussion. Would you like to? Maybe explain how that cloud cover acquired a year^-1 dimension which seems to be the basis for annual growth?
***Geoff,
I said there was no evidence from the discussion here that anyone had read it. You had not contributed to the discussion. Would you like to? Maybe explain how that cloud cover acquired a year^-1 dimension which seems to be the basis for annual growth?**
AS ALWAYS, NICK DOES NOT ADMIT HE WAS WRONG, BUT CHANGES THE SUBJECT, AS HE HAS BEEN DOING IN HIS SO-CALLED ‘DISCUSSION ON THIS POST.
I tried ArXiv, Joel. They rejected the upload.
Mixing up pressure and concentration is an elementary mistake indeed. And when your reviewer or reader makes such an error, you really need to think what lead to that mistake.
BTW, always add cite when referring to something reviewed, Ellis 2017 (in preparation) would do if it is not out yet…
Well, I did give both āmicro-barā and āequivalent surface ppm concentrationā, just to try and make things clear. But whatever the notation, I still cannot see how any scientist can think that plants (or indeed animals) would not be starved of CO2 (or oxygen) with altitude. How could anyone think that?
The reference is:
Modulation of Ice Ages via Dust and Abedo. Ellis and Palmer 2016
http://www.sciencedirect.com/science/article/pii/S1674987116300305
Ralph
Two of the comments were…
Quote:
Finally the authors calculate that CO2 drops by 40% at 4000m altitude relative to sea-level. However, this is simply incorrect physics. This means that the calculations of elevation-CO2 deserts are also incorrect.
I strongly suspect that the calculations showing a 120ppm drop in CO2 at 4000m in the tropics and 65ppmb drop at 2000m in the extra-tropics are completely wrong. Observations of CO2 with altitude show tiny variations (less that 5ppmv, e.g. Olsen & Randerson, 2004, figure 3). So this calculation in table 4 is very likely wrong. This means that the emergence of deserts at sub-190ppmv is not realistic because it is based on flawed calculations. This undermines much of the rest of the manuscript.
Endquote.
With pressures given as both āmicro-barā and āsurface concentration ppm equivalentā, for ease of clarification.
The emergence of CO2 deprivation deserts during the ice age, was primarily at high altitude in already arid locations – like Pategonia and the Gobi region. And we know that new deserts emerged in northern China, because of the huge increase in dust deposition upon the Loess Plateau, which records deposits from nearly all the ice ages. Indeed, the Loess Plateau gives an ice age record that is as valuable as the ice core records from Antarctica.
But the paper was rejected because of that āmistakeā.
R
And to cap it all, all references to the paper have been deleted from Wiki because I am apparently a āclimate denierā – whatever that means…
R
ralf is connolley the clown still at his wiki capers ?
in Manabe and Wetherald (1967), which your analysis is based upon.ā and because the climate is complex and nonlinear.
==========
Are climate models “complex and nonlinear”? Is this error analysis looking at climate or climate models?
The climate models oscillate between hot and cold on different runs. The modellers consider this to be prediction error from the true result but it is not, because the future does not exist. It is an imaginary concept.
As a result the future cannot be calculated beyond a probability. It most certainly does not exist as an average of that probability as climate science would have us believe.
The difference between the probability envelope and the average is uncertainty that is a physical property of mistaking the future for something that exists, ignoring that the future is not yet written.
As such this is separate from and independent of the error measurement.
āThe future doe not existā
Agree. It fascinates me that science, in collecting data, always is looking to the past, at traces and leftovers of things āgone byā. The future doesnāt exist by definition, and a confirmation of a prediction can only be done with data from past phenomena. All the phenomena happening in the ānowā have continuous and complicated if not chaotic interactions – a dynamic process that may leave some traces that on longer time-scales have some stability (ādataā). But the overwhelming majority of the effects of these interactions are lost on us, as they are not stable and static enough to be registered.
So it seems science is always a prisoner of leftovers in the past. Is that maybe the reason why would-be scientists and charlatans often try to escape to a perceived future and pretend to know about it?
doe=does
Errm , you do realise that Hansen retired some time ago, right?
I think careers can go rectum even post mortem. I certainly hope so.
“reputation” would have been a better word choice. With reputation goes career for those still practicing the career choice profession of science.
yet his modeling was a parody of science and maths
ATTP and Nick cannot comprehend the difference between model parameterization uncertainty and real world physical uncertainty.
The uncertainty is in the models, it has nothing to do with the real world
All uncertainty has no bearing on the real world
x amount of output is hindcast tuning not model physics.
it is absolutely absurd that Nick and ATTP cannot tell the differnence between a model and the physical world
it is even more alarming given how much we do not inderstand at all, including interaction of ocean, surface and atmosphere.
throw in lack of resolution, and a heap of parameterizations..
My work requires accuracy, if I performed my function as climate modelers do, I would be fired.
ATTP, you wouldn’t last 5 minutes in my world, neither would Nick
I’d like to see these models run for 5000 years to see what they produce. I bet the outcome is quite unvelievable š
see the work of David Archer, U Chicago.
he’s done great work on how CO2 will vary
in the next 100,000 years from our
current emission pulse. some of the
CO2 pulse stays around for basically
forever.
Oh ye. In 100,000 years we will know he was right. Or not. But, for sure, anythings that happen in the next 100,000 years (supervolcano exploding, large enough object hitting Earth, greening of Sahara, …) can be used as a excuse, so we basically never know. “great work”, you said?
Oh also, any output that is a range, is not a prediction. A prediction is a unique guess, a specific value.
A range is a cast net, not uncertainty, not variability, not prediction, not probability.
Who taught these clowns maths?
You might consider that your error analysis is of the model mean of the individual models. Not of the models themselves.
What I would like to see is the actual individual runs for each model. Do they all converge in the future or do they diverge?
I suspect we never see these individual runs for individual models because it would quickly reveal that the models themselves show that an infinite number of different climates can result from just 1 set of forcings.
In other words. The individual model runs will show that the future may be hot or it may be cold regardless of what we do with CO2.
And for that reason only the model means are published. Because they make it appear that we can dictate future climate by controlling CO2.
If you do 1000 model runs and they provide 1000 different outcomes and one is tracking temps, that means you have no idea what you are doing, and got lucky, as in, that is a fact, because if you did know what you were doing, the mean and extremes would cluster around the observed values.
because model runs do not, even come close to clustering around observations, neither the mean now extremes, it is clear models are useless.
Unless you are Zeke hausfather and you tilt the model runs down towards observations, which in my opinion is epic self delusion or outright intentionally misleading
There is 0 accuracy in climate modeling, they cannot even hindcast, they are tuned to get past values
Climate modelers would not make it in any other field of science, they would be eaten alive
This isn’t correct.
These models are deterministic, you give them the exact same inputs, they will generate the same outputs. It’s changing either parameters, or initial conditions where it changes output.
What you’re describing are different but equal initial conditions.
For instance, you should be able to build an initial set of data for 6 months ago, and now. Run the same model parameters, one starting 6 months ago, and one starting now. Simulate out 50 years to the same point, one would expect they would have the same results. I strongly suspect they would be different.
“This isnāt correct.
These models are deterministic, you give them the exact same inputs, they will generate the same outputs. Itās changing either parameters, or initial conditions where it changes output.
What youāre describing are different but equal initial conditions.
For instance, you should be able to build an initial set of data for 6 months ago, and now. Run the same model parameters, one starting 6 months ago, and one starting now. Simulate out 50 years to the same point, one would expect they would have the same results. I strongly suspect they would be different.”
I don’t think I agree with that.
I wasn’t referring to initial starting conditions, I was referring to the instability in “weather” š generation algorithms, they will not produce exactly the same on different runs.
To produce exactly the same outcome, everything coded into that model, and everything tuned would have to be precise and have built on correction (as it is difficult to get the exact same output using physics theory mathematics. I don’t think, but can’t say, that the models are built like that.
Do you have an example of models producing identical output on different runs?
Cheers
The “Weather Generation Algorithms”, just shuffle up the initialization inputs, or parameter settings, or both, a Monte Carlo Analysis. Computers are numerical calculators, When they do the same equation, and get different results a second time, it’s broken.
Mark says – “For instance, you should be able to build an initial set of data for 6 months ago, and now. Run the same model parameters, one starting 6 months ago, and one starting now.”
this can’t be done, because two important
sources of data are
missing: gridded deep ocean heat content,
and aerosol loading (it’s
needed as a function of
latitude).
climate models are about solving
a boundary value problem.
weather models solve an initial value problem.
Depends what you mean. Can we initialize a realistic model to represent reality, no. But that isn’t what a gcm is. It’s a big fricking circuit, where there’s a bunch of nodes that share data every time step, and between time steps they recalculate the new input data.
Every one of those nodes can be set during initialization, and the numeric solver will generate the same output with the same input.
Diff eq solvers are deterministic.
Now they give you somthing thing you didn’t expect, that you have to figure out. But if you do the same thing, you get the same output or it’s broken.
The use of the Ensemble Mean ensures loyalty to the Group. Any group that wants to be included in the group (think Club membership), must adhere to the rules of the Group set by the gatekeepers. Violate the rules and not only not get included in the Ensemble, but your group’s manuscripts (and your post-docs’ and your grad students’ manuscripts) will get harsh anonymous peer reviews forcing editors’ rejection. There goes your publication, there goes your grants.
This is how a few gatekeepers of the GroupThink enforce conformity on the members and dissent is silenced.
Unless climate science has somehow been endowed with naturally occurring models, all computational results are “hard-coded in.”
If the climate models correctly handled the physics, this wouldn’t keep happening…
and tuning for the past is tuning for the future too, cannot be one withou the other David
The output is all the evidence you need to show they have no idea what they are doing.
All they know from the past is temps are not going to swing outside of those higher values and of couse a few low runs to catch lower trends so they can claim “one model was right” when that is one of the biggest logical fallacies I have ever heard
It’s programmed into the code that parameterizes evaporation. CMIP it’s done with mass conservation.
http://www.cesm.ucar.edu/models/atm-cam/docs/description/node13.html#SECTION00736000000000000000
“If the climate models correctly handled the physics, this wouldnāt keep happeningā¦”
But it would…..
That graph is an ensemble of individual runs which have representation of natural variation within the climate system.
There are 95% confidence limits derived from that spaghetti of runs to encompass that natural variation.
You seem to expect that the averaging all those individual runs should have the GMT running up the middle of them, whereas (as we know) there the are dips/bumps of natural variation involved.
The actual GMT can only be expected to lie within those confidence limits.
Barely tracking the lower 95% confidence band (P97.5) is not a demonstration of predictive skill.
If the model is a reasonable depiction of reality, the observed temperatures should track around P50 and deflect toward the top or bottom of the 95% band during strong ENSO episodes.
“Barely tracking the lower 95% confidence band (P97.5) is not a demonstration of predictive skill.”
Yes it is … observations are within the constraints of the variation of individual runs, which is all the ensemble model runs can be asked to fairly do.
?w=700
Also, you do agree we have to compare against the forcings that actually occurred and not the ones used in the model runs? (should be rhetorical).
And let’s see where the GMT tracks during a prolonged +PDO cycle, as compared to a -PDO.
http://research.jisao.washington.edu/pdo/pdo_tsplot_jan2017.png
If you make a barn big enough even a child can hit it !!
David Middleton October 23, 2017 at 12:09 pm
Barely tracking the lower 95% confidence band (P97.5) is not a demonstration of predictive skill.
And yet when the Arctic sea ice extent does this we’re told it’s normal!
http://nsidc.org/arcticseaicenews/charctic-interactive-sea-ice-graph/
Tonyb,
The variation described by multiple runs of a single model is what should be compared to measured reality, not an ensemble of runs of different models. A y model where reality does not fall within that models “uncertainty envelope” at some specified level (e.g. two sigma) should be rejected as not an accurate representation of reality. The pooled variation across many different models means nothing… each model is a logical construct which stands, or falls, on its own… most fall. Those should be modified or abandoned.
Tonyb,
The variation described by multiple runs of a single model is what should be compared to measured reality, not an ensemble of runs of different models. A y model where reality does not fall within that models “uncertainty envelope” at some specified level (e.g. two sigma) should be rejected as not an accurate representation of reality. The pooled variation across many different models means nothing… each model is a logical construct which stands, or falls, on its own… most fall. Those should be modified or abandoned.
Red herring fallacy.
‘all computational results are āhard-coded in.ā’
If that’s all you mean by the phrase, it’s a pointless tautology. But the original commenter said:
“The computer simulations in question have hard coded in that an increase in CO2 causes warming. Hence these computer simulations beg the question as, does CO2 cause warming, and therefore are of no value.”
If it’s calculated by a computer is means the result was hard coded in, and therefore of no value. Maybe he meant that all calculated results are of no value – you can’t be sure here – but I hope not.
The physics have to be hard-coded in. However, the model is of no or little value it never demonstrates predictive skill.
The model is basically hard-coded to yield more warming with more CO2. It has to be. However, more warming with more CO2 is not a demonstration of predictive skill.
X (+/- y) with each doubling of CO2 would be predictive skill… if the observations approximated X (+/- y). However the observations consistently approximate X-y ((+/- y)).
DM – climate models are, of course,
not hard coded in. they’re the result of
numerically solving a large set of PDEs.
look at the description
for any model
for the mathematics.
GCM’s are nothing more than engineering code. If they were based on pure physics there would no difference between the dozens of models presented by the IPCC and various modelling groups. There should be be no need for “ensembles”.
i don’t know what “just engineering code” means.
the models give different results because they make
different choices, especially about parametrizations.
this is actually the main use of climate models by
scientists — not to project the temp
to 2100, but to compare results when a particular
component of a model is altered
because climate warming is GHG driven in models, warming in models is utterly dependent on GHG extrapolation, higher values more warming, they cannot ever go the other way, and history shows that is incorrect, unless you believe in hockeysticks.
To show this, run a climate model with increasing GHGs at current rates and run it 10000 years, and watch the earth turn into a melting ball. After such long runs the models would produce complete nonsense, but we are meant to believe them over 100 years?
ugh
Climate science is nothing but VERY IMPRECISE NUMBERS and you cannot create a precise theory out of very imprecise numbers
You would think that climate scientists would see the outputs as nonsense, but there really are buffoons out there that believe the Earth will become like Venus from our CO2 emissions.
They wont admit it, but the incorrect theory about Venus is exactly what has led to this junk science
Pretty sure that’s exactly where Hanson came from, NASA Planetary group iirc studying Venus.
you don’t say Micro š Thanks for that, it was a suspicion, until I read your post. Now I am convinced.
To sum this up, your paper was not published because it was not deemed ‘popular’ enough. We’re living the Dark Age of science, but the science renaissance is coming soon.
+1 Book burning
I’ll trust climate models more when they can 1 Hindcast without tuning and 2 cluster around observations to a reasonable degree and show some accuacy in tracking OBSERVED variability.
Juggling models (like hausfather) to come down towards obs is scientifically criminal. And Nye says lock of deniers š
I do not have detailed knowledge of the spaghetti graphs or the details of model ensembles, however the range of outputs is huge.
Our climate has only one set of input parameters and only one output. Leaving aside chaos theory, the chance of simulating our climate is zilch, given our lack of knowledge and inability to model clouds and other important influences.
How do ensembles of models have any meaning? They contain values or assumptions that initialise the model and exclude alternatives. These alternatives are then given equal probability in parallel calculations in order to give a spread of less likely scenarios. Then the mean of these scenarios is selected if it gives a better match with reality, though it simply shows that the original assumptions were not optimum. How can a mean of wrong values have more credibility than the value closest to the mean?
For years, model tuning involved the warming effect of CO2 reduced by the aerosol cooling factor, ignoring the fact that for years, both values were far beyond credible, observational values. It seems that today, reality is still an inconvenient factor.
cat – where are the observed forcings for CO2
and aerosols
published?
& why were the RFs used not
“credible?”
I think that spaghetti graphs are results of different feedbacks applied in models.
Pat,
If you’re looking for somewhere to post your paper, you could always try viXra.
ATTP, see Micro’s reply to your incorrect post above, to indepth for you? š
Ken: Why not at your site, where you could attack it and mod away inconvenient truths from Micro?
Nick Stokes October 23, 2017 at 3:12 am
“On that logic you could say that computation could never reveal anything.”
Doesn’t “revealing” necessarily imply proof or the accuracy of what is being revealed? Nothing has been revealed if what is “revealed” is not so. Computation may suggest something, however it can’t reveal, let alone prove anything.
This isn’t rocket science. I have been writing (or building analog) simulations for more than 30 years. If the dominant dynamics in your simulation are divergent, parameter variation will cause divergent results. If the dominant dynamics are convergent, they won’t. We have always been told by warmists that global warming dynamics are divergent. CO2 causes a temperature rise that causes more water vapor that causes more temperature rise that causes both ice to melt and reflect less energy and a warming ocean to give up more CO2. Rinse and repeat. If the models are not diverging from different parameters, they are admitting that the dominant dynamics are convergent. If that is the case, we can cancel the panic. Warmists, which is it?
actual knowledge and experience makes ATTP dissolve like CO2 in low O2 content water.
This is very disappointing, Pat.
Following the link to the book’s download site brings me to:
?dl=0
Two large computer disaster hot spots and all other possible download links do not show alt text or any other hints.
A very untrustworthy method and place for a simple download.
If I remember correctly, those are fake ads ….
You are correct.
Clicking hot buttons that harbor malicious or seriously greedy content is never worth the effort.
Just click slow download, and down it comes, AtheoK
Pat Frank, I will not!
What you suggest is not safe or rational computing.
I have practiced safe computing since the early 1980s.
If I do not get to read the URL address of a link, I will not click on it.
Clicking unlabeled hot spots is begging for computer/network infection.
When you provide a link that is open and clear; I’ll consider downloading.
Another caveat, vague/hidden/questionable download hotspots surrounded by obvious phishing hot spots is a sure way to keep people with common sense from touching anything on that page.
Vague – The hotspot fails to indicate what will download.
Hidden – The hotspot URL does not show anywhere.
Questionable – Hotspots surrounded by phishing hotspots.
Your claimed download link taps all three danger warnings.
I didn’t get any of that extra stuff when I clicked on the link, not even the choice between slow and turbo download.
I’ve used that site several times to transmit large files to various people. There’s never been any trouble.
Tell you what, though. If you send me the link to a free upload site that you trust, I’ll put the paper there for you. You can email me at pfrank_eight_three_zero_AT_earth_link_dot_net.
Interesting topic, but lots of wasted time. We know that we do not have the data to correctly initialize the models. We also know that we do not have the physics to correctly model atmospheric flows in a free convective atmosphere. As a consequence, we know that all of the models will contain errors. Unfortunately, we have no idea of the size of the errors or how they get promulgated through time.
Models generate hypotheses. Hypotheses need testing. How do we test the model output? Given the tuning, we clearly need time to pass in order to benchmark model runs against reality. We do not yet have enough actual data to test the CMIP5 hypotheses. Suppose as the AMO rolls over into a negative phase that we observe a cooling similar to what the unadjusted data shows for the 1950-1970s period. I think it is safe to say that most of the CMIP5 runs will be rejected and we won’t care about error promulgation. Unfortunately, we are currently in a situation where all we have are the hypotheses and not enough data to test them.
As the saying goes, all models are wrong but some are useful. We may come to find out that all the models are wrong and that those pimping the results have done significant damage to both the credibility of science in general and the welfare of the human population.
We may come to find?
Too late, it’s wrong, AGW is wrong, the modern warm period is almost all natural. There is RF from CO2, just that there is also a great big negative water vapor response that cancels it at night.
Kudos to WUWT for giving Pat Frank another kick at the cat in showing us his paper. The one that if published, would overturn the status quo and we could all go home and not have to worry about whether CAGW was really an issue. No one can accuse WUWT of not giving fair time to every unpublished skeptic.
But this was just too much to accept:
āThe stakes are just too great. Itās not the trillions of dollars that would be lost to sustainability troughers. Nope. Itās that if the analysis were published, the career of every single climate modeler would go down the tubes, starting with James Hansen. ā
Hopefully, his above quote was not his Introduction to the folks that were considering publishing his analysis.
If someone else had written that statement, then could be worth taking the time to really investigate his paper further. The hubris of his own work in his own mind and resentment towards the Journal’s for not being taken seriously, even here by many commenters, is proof enough (at least for me) that his analysis is not the the dragon slaying that he thinks it is.
spot on.
Errr…No!
Is that a pimple you see in your mirror benbenben
Earthling2, thanks for your self-serving, subjectivist, and critically meaningless contribution.
@Mosher: “in addition there are feedbacks which cannot be predicted”
Who knew that you could program a mathematical model into a computer when you did not understand all of the interactions!!!
I am in awe /not
@Nelson: “Models generate hypotheses.” Do they?
I thought computer models consisted of a set of formulas
ie they did what they were written to do.
It strikes me that the GCM linearity is perfect for task, validated by possessing a linear relationship to funding designed to promote the implementation of ideology.
Here is part of Section 1 of the blog version of my 2017 paper in Energy & Environment.
https://climatesense-norpag.blogspot.com/2017/02/the-coming-cooling-usefully-accurate_17.html
“For the atmosphere as a whole therefore cloud processes, including convection and its interaction with boundary layer and larger-scale circulation, remain major sources of uncertainty, which propagate through the coupled climate system. Various approaches to improve the precision of multi-model projections have been explored, but there is still no agreed strategy for weighting the projections from different models based on their historical performance so that there is no direct means of translating quantitative measures of past performance into confident statements about fidelity of future climate projections.The use of a multi-model ensemble in the IPCC assessment reports is an attempt to characterize the impact of parameterization uncertainty on climate change predictions. The shortcomings in the modeling methods, and in the resulting estimates of confidence levels, make no allowance for these uncertainties in the models. In fact, the average of a multi-model ensemble has no physical correlate in the real world.
The IPCC AR4 SPM report section 8.6 deals with forcing, feedbacks and climate sensitivity. It recognizes the shortcomings of the models. Section 8.6.4 concludes in paragraph 4 (4): āMoreover it is not yet clear which tests are critical for constraining the future projections, consequently a set of model metrics that might be used to narrow the range of plausible climate change feedbacks and climate sensitivity has yet to be developedā
What could be clearer? The IPCC itself said in 2007 that it doesnāt even know what metrics to put into the models to test their reliability. That is, it doesnāt know what future temperatures will be and therefore canāt calculate the climate sensitivity to CO2. This also begs a further question of what erroneous assumptions (e.g., that CO2 is the main climate driver) went into the āplausibleā models to be tested any way. The IPCC itself has now recognized this uncertainty in estimating CS ā the AR5 SPM says in Footnote 16 page 16 (5): āNo best estimate for equilibrium climate sensitivity can now be given because of a lack of agreement on values across assessed lines of evidence and studies.ā Paradoxically the claim is still made that the UNFCCC Agenda 21 actions can dial up a desired temperature by controlling CO2 levels. This is cognitive dissonance so extreme as to be irrational. There is no empirical evidence which requires that anthropogenic CO2 has any significant effect on global temperatures. ”
However establishment scientists go on to make another schoolboy catastrophic error of judgement by making straight line projections.
“The climate model forecasts, on which the entire Catastrophic Anthropogenic Global Warming meme rests, are structured with no regard to the natural 60+/- year and, more importantly, 1,000 year periodicities that are so obvious in the temperature record. The modelers approach is simply a scientific disaster and lacks even average commonsense. It is exactly like taking the temperature trend from, say, February to July and projecting it ahead linearly for 20 years beyond an inversion point. The models are generally back-tuned for less than 150 years when the relevant time scale is millennial. The radiative forcings shown in Fig. 1 reflect the past assumptions. The IPCC future temperature projections depend in addition on the Representative Concentration Pathways (RCPs) chosen for analysis. The RCPs depend on highly speculative scenarios, principally population and energy source and price forecasts, dreamt up by sundry sources. The cost/benefit analysis of actions taken to limit CO2 levels depends on the discount rate used and allowances made, if any, for the positive future positive economic effects of CO2 production on agriculture and of fossil fuel based energy production. The structural uncertainties inherent in this phase of the temperature projections are clearly so large, especially when added to the uncertainties of the science already discussed, that the outcomes provide no basis for action or even rational discussion by government policymakers. The IPCC range of ECS estimates reflects merely the predilections of the modellers – a classic case of āWeapons of Math Destructionā (6).
Harrison and Stainforth 2009 say (7): āReductionism argues that deterministic approaches to science and positivist views of causation are the appropriate methodologies for exploring complex, multivariate systems where the behavior of a complex system can be deduced from the fundamental reductionist understanding. Rather, large complex systems may be better understood, and perhaps only understood, in terms of observed, emergent behavior. The practical implication is that there exist system behaviors and structures that are not amenable to explanation or prediction by reductionist methodologies. The search for objective constraints with which to reduce the uncertainty in regional predictions has proven elusive. The problem of equifinality ā¦ā¦. that different model structures and different parameter sets of a model can produce similar observed behavior of the system under study – has rarely been addressed.ā A new forecasting paradigm is required”
Doc:
What could be clearer? The IPCC itself said in 2007 that it doesnāt even know what metrics to put into the models to test their reliability. That is, it doesnāt know what future temperatures will be and therefore canāt calculate the climate sensitivity to CO2.
They’ve never stopped saying it in one way or another:
“In sum, a strategy must recognise what is possible. In climate research and modelling, we should recognise that we are dealing with a coupled non-linear chaotic system, and therefore that the long-term prediction of future climate states is not possible. The most we can expect to achieve is the prediction of the probability distribution of the system’s future possible states by the generation of ensembles of model solutions. This reduces climate change to the discernment of significant differences in the statistics of such ensembles. The generation of such model ensembles will require the dedication of greatly increased computer resources and the application of new methods of model diagnosis. Addressing adequately the statistical nature of climate is computationally intensive, but such statistical information is essential.”
http://ipcc.ch/ipccreports/tar/wg1/505.htm
Keep shouting this from the rooftops. but don’t expect too many to hear you. There’s too much at stake for logic to win the day amongst the academics.
Very random question. Are the oscillations in the model output you describe in anyway similar to the phenomenoa āhunting oscillationsā?
Pat
Perhaps you could provide examples of error propagation that will illustrate your position. Being an old land surveyor I am very familiar with the concept as it effects all sequential measurements that rely on a previous measurement. Think of surveying a rectangular tract of land using a series of measurements each containing an angle and a distance. Assume all reasonable care is used to constrain errors. There is an uncertainty in each measured angle and each distance. These errors propagate as the traverse continues so the error elipse at each new measured point grows and changes shape. They soon get larger than expected error but you cannot reject any coordinate pair that falls in the elipse as a blunder or obviously erroneous. When the traverse closes out on the beginning point. a proportion of the closing error is distributed to each traverse point. If this proportioning process is skipped you have large possible positional error in each corner of the tract. The error elipse at each corner is not the variance you would expect if you reran the traverse lots of times. It is in the dimensions of the coordinate system but is a statistical number, not a possible dimension. It is the size of your propagated uncertainty and you cannot reject any coordinate that falls in it without further analysis. Depending on the your methods the error elipse may be far outside any normal set of measurements.
It seems reasonable to me that the iterative process used to generate future climate scenarios is similar to the survey process and subject to the same error propagation analysis.
It seems reasonable to me that the iterative process used to generate future climate scenarios is similar to the survey process and subject to the same error propagation analysis.
==============
a good analogy. The problem with the future is that there is no way to “tie in” the end point and redistribute the error. The best that happens is that on each run of the model, it gains or loses energy due to loss of precision errors. This “missing” energy is then divided up across the model and added or subtracted back it as may be required to maintain the fictional accuracy of the model.
However, for a model to be accurate, you want to keep your time slices quite small, which means there will be many trillions upon trillions of iterations. And even if each iteration is 99.9999% accurate, if you iterate enough times it doesn’t take long before your actual accuracy is approximately 0%. (99.9999% ^ n tend to zero as n increases).
Ferd
Thanks for your comment. Can you think of other processes that are subject to uncertainty growing due to iterative processing? I’m guessing here but maybe some of the financial models are in this boat. The point that seems to mystify so many is the concept of an error uncertainty with the units of the process that can grow to nonphysical dimensions. The uncertainty tells you about the reliability of the solution not its likelihood.
DMA, I struggled very hard to get the concept across. Honestly, I went in to the process expecting they’d all be familiar with error propagation. After all, what physical scientist is not? But I drew a total blank with them.
It’s been as though they’e never heard of the topic. And they’ve been completely refractory to any explanation I could devise, including extracts from published literature.
Pat Frank – Many thanks for this article. Unfortunately, as in all of climate science, the subject is sufficiently far from primary school mathematics for it to be easily countered by obfuscation. It is truly disturbing how long it can take to get past determined gatekeepers. I was pleased that you picked up that Willis Eschenbach had come to the same conclusion as you, via a different path, back in 2011.
The article I wrote for WUWT in 2015 – https://wattsupwiththat.com/2015/09/17/how-reliable-are-the-climate-models/ – came at it from a third direction: pointing out that CO2 was the only factor in the climate models which was used to determine future temperatures. I know perfectly well that I can never get any paper past the gatekeepers, no matter how much extra detail and precision I put into them, so I echo your praise of WUWT for providing an uncensored voice to climate skeptics. I have however sent some of my articles separately to an influential person in a reputable scientific organisation, and had quite reasonable replies until I suggested that their organisation could emulate Donald Trumpās proposed Red Team ā Blue Team debate. I received the terse one-line reply āThere will be no debate supported by [the organisation].ā. [It was a private conversation, so no names.].
I wonder whether there might be scope for a spin-off site from WUWT, that would effectively be a journal. Antony would be the publisher. There would be an editorial committee. When papers are submitted, Antony would allocate them to the most appropriate editor. That editor would then solicit peer review, in the same way as normal, but possibly using a broader range of reviewers. The main difference would be that the editor would have the knowledge and discretion to discount negative reviews if the criticisms have little merit.
Sounds like pal review.
@crackers345
By “a broader range of reviewers” I intended that the reviews would predominantly be solicited from the same consensus scientists as a present, but possibly with some luke-warmers / skeptics included. So I don’t think it would be fair to characterise that as pal review. The difference would be in having a knowledgeable editor. In my field, the editors know very little and don’t feel qualified to overrule a critical review. If you accuse my idea of anything, you should accuse it of having a pal editor, not pal reviewers.
Mike Jonas, it’s a culture of studied blindness, isn’t it.
Hundreds of Billion$ of dollars and hundreds of thousands of hours of computer time wasted to create ‘ensembles’ (?? What the fashionable CAGW alarmists are wearing this year?) of computer models that produce pathetically simple freshman-in-high-school y = mx + b results?
Linear Thinking In A Cyclical World…..
http://www.maxphoton.com/wp-content/uploads/2014/01/linear-thinking.png
climate models are about solving
a boundary value problem.
weather models solve an initial value problem.
===============
keep in mind that the famous ensemble mean spaghetti graph is in actual fact a mean of means. The individual model runs themselves are in point of fact not shown. Rather the spaghetti graph shows only the means of the actual model runs.
The simple fact remains. For 1 set of forcings, there are an infinite number of different future climates possible. Thus, any attempt to control climate by varying CO2 is a fools errand.
http://tutorial.math.lamar.edu/Classes/DE/BoundaryValueProblem.aspx
The biggest change that weāre going to see here comes when we go to solve the boundary value problem. When solving linear initial value problems a unique solution will be guaranteed under very mild conditions. We only looked at this idea for first order IVPās but the idea does extend to higher order IVPās. In that section we saw that all we needed to guarantee a unique solution was some basic continuity conditions. With boundary value problems we will often have no solution or infinitely many solutions even for very nice differential equations that would yield a unique solution if we had initial conditions instead of boundary conditions.
Here is the problem. Climate models are presented to the public as deterministic models with all the authority of a Newtonian two-body problem. As in any model, there are measurement and other errors. It is certainly true that the errors of one period must be propagated as errors in the initial conditions of the next period and so the cumulative errors grow. It may also be true that there are physical reasons why the expanding errors are physically impossible. But here is the point, you can bound the cumulative errors with a physical constraint if you like, but the fact is that at that point, the errors have swamped the deterministic effects of the model. The model is now just a statistical fit wrapped around some physics equations. There is no deterministic authority to the model. The only way to justify such a model is through the usual regression model validation processes using a calibration data set and then testing on data held in reserve. It’s just a statistical fit of two trending series: CO2 concentrations rising and temperatures rising. You could just as well fit rising government debt to rising temperature values. Trending data series have approximately one degree of freedom to explain their variation, so you fitting statistics will be terrible.
“Climate models are presented to the public as deterministic models with all the authority of a Newtonian two-body problem. “
No, they aren’t. People here love to quote just the first sentence of what the AR3 says about models:
“The climate system is a coupled non-linear chaotic system, and therefore the long-term prediction of future climate states is not possible. Rather the focus must be upon the prediction of the probability distribution of the system’s future possible states by the generation of ensembles of model solutions. Addressing adequately the statistical nature of climate is computationally intensive and requires the application of new methods of model diagnosis, but such statistical information is essential. “
You don’t find this caveat in the summary for executives, what you find is “extremely likely” BS.
And anyway this only translate into “let’s pretend that ensemble model solutions are representative of probability distribution of the systemās future possible states”. Which is utter BS, too.
Really, Nick?
According to you, Nick, people here “love to quote”. How absurd. If so, you should provide proof of your assertion.
A quick WUWT search of the opening phrase in that sentence reveals five examples with three of those examples components of a linked series of articles by Kip Hansen.
When quoted, source links were provided; for anyone desiring to read all of the context.
Within the two remaining search examples, the phrase, or a distant paraphrase is discussed ancillary to other topics.
Ergo; you claim is blatantly false.
Just another fake thread bomb by Nick Stokes.
Thank you for your reply. Regardless of who made what representations, the errors (or stochastic variables, if you like) swamp the deterministic variables, so the models are complicated regressions, not initial value solutions to a deterministic process. Please address the substance of my comment above.
Atheok
“A quick WUWT search of the opening phrase in that sentence reveals five example”
Too quick. This search returned 36 results for me – a few duplicates, but most different quotes. And if you ask for the “similar results”, you get 49. But the telling statistic is that if you search for the very much linked second sentence, ” Rather the focus must be upon the prediction…” you get only 7. And a few of those are from me, just pointing out the context from which it was ripped.
Nick, You are being a little disingenuous here with your wording of “a few duplicates”. Of your 36 results, more than 20 are from Kip Hansen, mostly from a single multi-part posting. Other results also contain duplicates. You may want to revisit your nonsense (to use one of your own favorite words) claim of “people here love to quote…”
Due to my slow DSL connection I have not read Mr. Frankās paper and other documentation. However, from the OP and comments, it is clear to me that he is correct in saying that most all the climate modelers and alarmists do not understand that āerrorā and āuncertaintyā are very different things.
Error is the difference between a measured (or assumed) value and the true unknown and unknowable value. Uncertainty is an estimate of the range of the probable range of the error. So if I state that the length of an object I measure is 1.00 meter and my measurement uncertainty is +/- 4 mm at 95% confidence, it means that there is less than a 2.5% chance that the true length is greater a than 1.004 m and less than a 2.5% chance it is less than 0.996 m. It is possible that the true length of the measured object is 1.00 m in which case the error is zero, but the uncertainty of the measured value is still +/- 4 mm.
Uncertainty can result from either systematic or random sources. E.g. an error in the value of a calibration reference will produce a systematic uncertainty in the calibrated instrument. A gauge block with a stated dimension of 10 cm that has a true dimension of 1.01 cm will result in a systematic uncertainty of + 0.01 cm in the calibrated instrument. Random uncertainty components are also present such as those that arise from instrument resolution. Only the random components of uncertainty can be reduced by statistical analysis of multiple measurements.
With respect to the subject of the post, propagation of Uncertainty may be illustrated by a simple example. Suppose I use a 1 cm gauge block to set a divider (i.e. a compass with two sharp points) which I then use to lay out a 1 meter measuring stick. I start at one end, lay off 1 cm with the divider, make a scribe mark, lay off another cm from this mark and so forth 100 times. Now suppose that my gauge block was actually 1.01 cm. Can I claim that my 1 meter stick is accurate to 0.01 cm ā of course not ā my process included a systematic error that compounded in each iteration. As a result my meter stick will be off by + 1 cm (100 x 0.01 cm). In reality, of course, the process described will result in additional errors due to inexact placement of divider points and scribe marks, etc. These factors could either increase or decrease the overall error.
Now suppose I made ten meter sticks by this method. Would I gain any confidence in their accuracy by comparing them to each other? I would likely see a small variation due to the random nature of errors in aligning divider points and making marks, but the large systematic error would remain unapparent.
These days it is not considered proper science to discuss āerrorā since it is by strict definition unknowable. The analysis of propagation of error thus requires one to start with a fallacious claim that error can be known. In reality, only uncertainty can be reasonably estimated and then only in probabilistic terms. The analysis of propagation of uncertainty is not trivial even in relatively simple systems when there are several variables with confounding interactions. It can become intractable in complex systems. This is at least one reason that in engineering the issue of uncertainty is often addressed by application of very large safety factors. For example steel members in a bridge may be designed on paper to carry loads five times greater than the maximum expected. In other situations application of mathematical models may simply have too much uncertainty and require full scale physical testing to evaluate fitness for purpose.
Finally, with respect to Mr. Frankās publication problems ā why would anyone be surprised that a paper that effectively says the results produced by climate models are worthless expect anything but negative reviews and recommendations against publishing when the editors select climate modelers and scientists who rely on model outputs as peer reviewers? The conflict of interest is obvious and monumental.
Good summary of error and uncertainty, Rick.
+1 .
Someone please show me a climate model that starts in 1960 and replicates the climate changes since then using only data known at that point in time.
@Nick Stokes
Would you please list the basic physics used in climate models? It is my understanding only advection, the pressure gradient force and gravity are the core dynamics (fundamental physics) of a GCM. The rest are tunable parameters; merely engineering code.
The core dynamics are the Navier-Stokes equations, momentum balance (F=ma), mass balance and an equation of state – basically ideal gas law. It is called that because those are govern the fastest processes and are tcritical for time performance. But then follow all the transport equations; energy (heat), mass of constituents, especially water vapor, with its phase changes, including surface evaporation. Then there are the radiative transfer equations; SW incoming and IR transfer. This is where CO2 is allegedly “hard coded” because it affects IR transfer. Then of course there the equations of ocean transport of heat especially, but also dissolved constituents.
Here is GFDL model of that ocean transport
That’s what they are telling you was all hard-coded. And maybe just picking a few parameters.
Mr Stokes, I still cant decide whether you are deliberately obtuse or are really ‘as thick as a plank’.
You can substitute every wonderfully exotic variable in the models with the timetables of London buses and as long as you keep the CO2 forcing variable intact you would get the same results.
Its complete rubbish. GIGO but with the proviso that the output is pre-designed.
Nick neglected to mention that ocean models don’t converge.
See Wunsch, C. (2002), Ocean Observations and the Climate Forecast Problem, International Geophysics, 83, 233-245 for a very insightful discussion of the ocean and its modeling.
DR: here’s the description of some
of the NCAR CAM
model:
http://www.cesm.ucar.edu/models/atm-cam/docs/description/description.pdf
equations and all
“With four parameters I can fit an elephant, and with five I can make him wiggle his trunk.”
Just for fun, i counted the number of time the locution “assumed to be” is used. I stopped counting at 25. I check an few of them, they weren’t trivial. I didn’t count the number of other way assumption can be expressed…
So
hundreds of assumptions ; hundreds of parameters.
No wonder the trunk wiggle as requested, as a serpent before his charmer.
I am pretty sure that with the very same model and a few fiddling of parameters, i could easily have CO2 reduce temperature and induces a new ice age, without you being aware of the trick
Let me point this out again, the radiation model is probably correct, more co2 does cause more radiation. It’s what happens from there that matters in the models.
Let make sure in the code, more water vapor is created. In model E they called it supersaturation. They allowed humidity to go over 100% to get water amplification out of the boundary layer into the atm. Without this hack, the models run cold. Most of these ran hot, so they adjusted aerosols to make them match measurements.
In CMIP they do the same, but in a section called mass conservation. They make sure enough water vapor mass makes it out of the boundary layer.
http://www.cesm.ucar.edu/models/atm-cam/docs/description/node13.html#SECTION00736000000000000000
The problem is temps follow dew points. remember it represents an energy barrier, to go lower it has to shed water, which requires the air to cool. As it does condense, it’s emitting sensible heat, IR, and if we were not there, it would likely hit another water molecule, and help evaporate it. At night, this layer radiates to space, but towards the surface as well, it slows the drop in temp, reduces how cold it get before morning.
I was out imaging till 5 am, the temp stopped dropping after midnight. Air temp was low 40’s, IR temp of the sky was -60F. That’s about 50W/m^2 of radiation out the optical window. All night, but temps didn’t really drop.
You can see how in the same situation, net radiation drops.
Also, when you look at temps dropping, an exp decay will be a smooth curve, the radiaus changes, but it’s smooth, when you look at them, and there’s a slight bend in the curve, the rate of energy flow (flux) changed. and you can see that in both of these.
If you understand how semiconductor carriers are manipulated, and controlled in a mosfet into operating like a switch, and then consider how the atm is actively controlling out going flux at night, there are a lot of similarities.
micro ,i would love to see your work submitted as a guest post here. i understand why you wouldn’t submit it to a journal for review.
But don’t ask about the fudge factors they use to parameterize what they don’t/can’t calculate.
Those fudge factors are the parameter tables they use to tune the model’s output projection to meet expectation.
Start with solid science as the first ingredients, then finish with large doses of crap, and you have a cake that tastes like crap.
I’d say you have a cake that basically IS crap. But then when you dig deep enough, even the IPCC admits this (quoted elsewhere in this discussion).
Pat Frank, you must be on to something. Anything that gets ATTP and Nick Stokes bowels in such an uproar has got to be worth pursuing.
Mosher……………not so much.
The climate models do not meet the important scientific concept of falsifiability, and are therefore unscientific.
Falsifiability is the principle that in hypothesis testing a proposition cannot be considered scientific if it does not admit the possibility of being shown to be false. For a proposition to be falsifiable, it must – at least in principle – be possible to make an observation that would show the proposition to be false, even if that observation has not actually been made [//psychology.wikia.com/wiki/Falsifiability]
No one in our lifetime can make an observation that would show the model to be false, since model projections extend to the year 2100. Therefore the models are unscientific. Just voodoo pseudo-science.
If observations fell outside the full model range at any point from the start of the forecast period (2006) to 2100 they would be falsified.
But by the ‘magic of climate science’ and the application ‘heads you lose , tails I win ‘ this is not an issue .
After all the first rule of climate ‘science’ , is when the models and reality differ in value, it is reality which is in error .
The bigger picture is not why Climate Models have been so spectacularly wrong (they have been, 97% of the time : ); it is why they have been so wrong, and will always be so, because of political, not scientific reasons.
I used to view Science as a noble profession, one that lifted an ordinary person like me to incredible heights, not imagined by my hard-working ancestors.
Sad that a once noble profession has been subjugated to political propaganda.
http://blogs.reading.ac.uk/climate-lab-book/files/2014/01/fig-nearterm_all_UPDATE_2017-panela-1.png
Crackers, bad boy! You are using what looks like GISSTEMP to validate models. Try defending the cooking by GISS first, then claim them a support. A complex question fallacy?
Tom Halla
HadCRUT4 shows similar results. Annual observations in 2016 were very close to the multi-model average (fig. from Ed Hawkins).
http://www.climate-lab-book.ac.uk/files/2016/01/fig-nearterm_all_UPDATE_2016.png
They are both FUBAR. Showing 1998 as notably cooler than 2017, or either as substantially warmer than 1938, are examples of creative writing.
?
http://blogs.reading.ac.uk/climate-lab-book/files/2014/01/fig-nearterm_all_UPDATE_2017-panela-1.png
crackers, come back and post that crap again when la nina is in full swing .the response from you and dwr to tom halla is laughable.
many ENSOs have happened, yet
the temperature has reached this point.
how will the next la nina compare to
previous la nina years? they’ve each
been getting warmer. why do el nino
years keep getting warming?
billy: for both NOAA surface
and UAH LT, the latest
el nino season (2015-16), la nina season (2016-17; weak) and neutral season (2014-15)
all set record highs
for their ENSO classification.
How accurate are various GCMs in any case?
Here are some figures from the earlier CMIP3 exercise.
http://www.geoffstuff.com/DOUGLASS%20MODEL%20JOC1651.pdf Please refer to Table II. Your attention is drawn to the performance of the CSIRO Mark 3 model, coded 15, against the ensemble means at various altitudes. Trends are in millidegrees C per decade.
Surface 1000 925 850 700 600 500 400 300 250 200 150 100
163 213 174 181 199 204 226 271 307 299 255 166 53
156 198 166 177 191 203 227 272 314 320 307 268 78
____________
Boiled down to its essence, we have our Australian CSIRO publishing that they have calculated model temperatures whose least significant figure of 0.001 Ā°C per decade.
At three altitudes, the model result is within +/- 0.001 Ā°C per decade of the average of many simulations by others.
Outstanding!!
Yes, the models provide unbelievable precision, absolutely unbelievable.
“Think that fortuitously cancelling errors remove physical uncertainty”. As in if you’re lucky the sum of all the positive differences between measurements and the real value will exactly equal the negative. Somehow that has become a law that if you do take enough measurements then it will happen.
I used to use an example of a sniper to teach the difference between precision and accuracy. A good shooter will be precise ie. a small spread of holes in the target, even if the sight is off (not accurate).
If 10 000 shots are spread randomly around the bullseye over 10cm, it would be quite fortuitous for the mean to be within 1mm ie. perfectly random. For a climate scientist, its law that it will happen, never mind that many shots after calibration will not make it better.
PS You’ve hit a nerve. This has been painful to type on this site.
Robert B,
We can use your example to illustrate propagation of errors. It is not so easy to invent a good analogy for propagation of error.
Suppose that there is a telescopic sight that can move. If it moves once, then steadies, it might send all shots to one side of the target. This is an offset, a term used above.
If, however, the sight was forever loose, it could move to left or right or up or down, this being an error that can be negative or positive but without the ability of negative excursions to be wiped out by positive ones – the shots with their errors are already in the wall. The +/- case.
We can take each new magazine as the equivalent of starting a new round of model computation. The errors of the sights will continue to be present with each magazine change. They do not go back to a magic zero error when the magazine is changed. Because there are more shots accumulating all the time, there is a probability that wider and wider errors will happen. The errors propagate. The uncertainty bounds look like Pat’s illustration with +/- 18 degrees.
Some who posted above seem to think only in terms of the sights being firm, but offset. This produces what they argue for, but it is a precision concept that they are left with, once the offset error is corrected. Theirs is a wrong, naive analogy. It is not the +/- case.
But Pat is dealing with the analogy of the loose sights that can end up with bullets anywhere, any time. Unconstrained except going in the general direction of the barn wall and not back to hit the shooter. Hi Pat, please correct if I am wrong.
Pat, I think you can add another climate science common error to your list. It is an a priori assumption that if variable A increases, variable B will be more likely to increase than decease. It comes from thinking too often that if CO2 increases, then temperature will increase – almost by immutable law. Geoff.
Geoff S. and Robert B.
The sniper analogy can be improved by adding a shooter that knows how to adjust the sights but doesn’t know enough to tighten the mounts. He sees each shot and adjusts the sight as if everything on that shot was good. He then aims at the center of the target and repeats the process. Not only is the loose sight causing error (mostly random) the next shot is based on the position of the last. Thus the error propagates and the group spreads far beyond the accuracy capability of the rifle.
I believe the iterative process of the climate models works the same way with uncertainty growing with each iteration to the point that they quickly get into the realm of meaningless results even if they are constrained to give realistically possible results.
Hi Geoff, it’s not that the gun can send bullets anywhere.
It’s that it systematically sends them somewhere but we don’t know where.
And every different gun send them systematically somewhere else, and again we don’t know where.
All we know is that when we set up a nearby target and test all the different guns, the bullets get spattered in some way about the target with an average scatter of, say, Ā±2m.
And the problem is somewhere in the gun, or somewhere in the bullet, or both, but we don’t know where.
And when we shoot at the real target, we know it’s 1000 meters away but we have no idea where it’s located. Also, we can’t see the bullets, can’t find the bullets, and have no idea where they are with respect to the real target (which remains invisible).
But we do know that the bullets get to within Ā±2m of a target at 10 meters. š
That’s climate models and their long wave cloud forcing calibration error.
On another topic, Geoff, you’ve probably noticed that ATTP thinks one can add up all the individual measurement errors in a calibration experiment, combine them into one number (so positive and negative calibration error values cancel), and then subtract that one final number from real measurement data and decide those data are now error-free.
As a seriously well-trained and experienced analytical chemist, do you think you can find a way to explain to him that he’s got a really bad idea?
I’ve tried many times, and he’s completely refractory.
Best wishes to you, Geoff. š
Geoff, the sniper analogy can be improved by assuming the sniper is letting go of bloated balloons at his/her target. The balloons flazzzpt randomly directional. Certain physicals laws allow for a spead if sorts and so we take a mean of the spread and call it projection.
Some noted Climate Scientists think that Nyquist only applies to time and not space. He would be turning in his grave. Sampling theorem and its observations about all samples/measurement we rely on is not to be ignored. IMHO of course.
Analogue is always accurate but never precise. Digital is always precise but never accurate.
That goes for writing down the figures as well is in the computer/instrument.
T(n+1) = T(n)+Ī» āF(n+1) / Ļ + ĪT(n) exp( -1 / Ļ )
The “black box” equation is linear on each time slice. However the result is matching the model mean not individual runs. Whether the error term converges or diverges would seem to me to be an important question.
My best guess is that climate due not have a fixed mean or variance and the law of large numbers does not apply and the error term does not converge. Rather climate is a fractal. A 1/f distribution and scale invariant at all time scales.
But keep in mind our actual climate is not a model mean. The climate we experience is like a single run of a climate model. We end up with just 1 outcome from all possible outcomes.
The climate models however are. Showing us the average of all outcomes. Which is a very different beast statistically.
Fundamentally the climate models are wrong because they project future climate to be the mean of all possible climates.
In actual fact all the climate models are actually showing us is that there are an infinite number of future climates possible from a single forcing. And the spaghetti graph shows the boundaries.
The ensemble mean is simply a projection without predictive power. Climate models have been given a bad reputation because the model mean has been misrepresented to the public as having predictive power. Which it does not because this is a boundary value problem.
Similar to a simulation of a roll of the dice we know the boundary is 2 and 12. But the average of 7 has no predictive power for what will actually be rolled.
ferdberple,
The ‘Spaghetti Graph’ shows us the sensitivity of the models to initialization perturbations and differences in assumptions about future CO2 emissions. It shows us a range of possible outputs from the models. However, without some way of calibrating or validating the models, there is no reason to believe that the mean or median, or even the binned-mode, is the best estimate of the future state of temperatures. Even if the models had expertise in predicting temperatures, without a reliable prediction of future increases in CO2 (Assuming it actually is the control knob!) there is no hope of the ensembles having predictive value. At best, all the modelers can say honestly, is “Assuming that one of the RCPs is close to what the future emissions will be like, we think that the future temperatures will be within a range demonstrated for the ensemble for that particular RCP.”
At this point in time, it appears that only the Russian model (near the low extreme) is tracking the actual average global mean temperature. Logically, there can only be one best prediction. Averaging it with all the other predictions only reduces the quality of the prediction.
And yet again, this is just embarrassing. People who actually understand the math vs those who don’t. Well, it’s definitely one way to progress. The hard way.
Joe Crawford October 23 @9:28AM:
“I doubt there is a Mechanical Engineer in the crowd that would trust his/her familyās safety to a 5th floor apartment deck that was designed with, or the design was verified by, a stress analysis (i.e., modelling) program that required constraints be placed within it to keep the calculations within reasonable ranges.”
You have forgotten that there is a Mechanical Engineer in the crowd, Bill Nye ‘The Science Guy’.
We don’t claim him, Tom. There is at least one exception to every rule.
Let Eli make this simple. Take some parameter B. Nick Stokes is saying that three values used for annual runs are
1. 2.0
2. 2.1
3. 1.9
Pat Frank is saying the three values must be
1. 1.0
2. 2.1
3. 2.9
In both cases the average is 2.0. Nick says this is an average of 2.0. Pat says this is an average of 2.0/yr
Now you would think that if Pat Frank were correct, running a model without changing the atmospheric composition would give wildly diverging results as the number of years increased. But GCMs don’t behave that way, and indeed doing such runs is a basic test of the model and tells something about the unforced variability in the model on different time scales which can be compared to the observed natural variability on those time scales.
Running the models with no changes to gas mixture, should replicate the range in temps due to the ocean cycles, and el ninos.
These alone should make a wide range of run result.
If you’re saying they don’t, just more proof of how flawed they really are.
Oceans are thermal storage in a model, they have to have delayed thermal processes associated to them or they are not modeled correctly.
Eli “Now you would think that if Pat Frank were correct, running a model without changing the atmospheric composition would give wildly diverging results as the number of years increased.”
Eli would think that. So would my climate modeler reviewers. No trained physical scientist would think that, though, because they’d all know the difference between physical error and an uncertainty statistic.
They’d also have in mind that models are tuned to produce “reasonable” values. They’d know that tuning models does nothing to reduce uncertainty.
Eli doesn’t know any of that. Eli is not a physical scientist. He’s a member of this caste.
Eli is in fact a Professor of Physical Chemistry.
In that case, Nick, he’s no credit to his profession. He’s a volunteer member of that caste.
He is a trained physical scientist who thinks you are wrong. A very large number of scientists think you are wrong. You have not produced any who think you are right. And you give no references to support your nutty ideas on averaging.
Pat Frank, if you were even close to being right, you wouldn’t have such a problem getting published.
Nick Stokes, argument from authority.
Eli is required to give a quantitative reason. Heās not done that. His āwildly divergingā was so wrong analytically as to imply complete ignorance.
Three of my reviews expressed agreement with the analysis and recommended publication. Every single physical scientist Iāve spoken to directly has understood and agreed with the analysis.
The only rejectionaires have been climate scientists, all of whom worked from a huge professional conflict of interest. And their arguments are demonstrated wrong, or like Eliās, to be candid expressions of utter ignorance.
Mark S. Johnson thank-you for your outstandingly naĆÆve comment.
Good work. Not only do you show that mankind is risking vast amounts of resources on quasi-science but, you also show how they get away with the fraud by shutting out any opposing views. A class action lawsuit has to be launched.
As of this post and to my best knowledge, I have resolved all the objections on this thread.
If I am wrong and any remain unresolved, please point them out in reply below.