26 October 2020
by Pat Frank
This essay extends the previously published evaluation of CMIP5 climate models to the predictive and physical reliability of CMIP6 global average air temperature projections.
Before proceeding, a heartfelt thank-you to Anthony and Charles the Moderator for providing such an excellent forum for the open communication of ideas, and for publishing my work. Having a voice is so very important. Especially these days when so many work to silence it.
I’ve previously posted about the predictive reliability of climate models on Watts Up With That (WUWT), here, here, here, and here. Those preferring a video presentation of the work can find it here. Full transparency requires noting Dr. Patrick Brown’s (now Prof. Brown at San Jose State University) video critique posted here, which was rebutted in the comments section below that video starting here.
Those reading through those comments will see that Dr. Brown displays no evident training in physical error analysis. He made the same freshman-level mistakes common to climate modelers, which are discussed in some detail here and here.
In our debate Dr. Brown was very civil and polite. He came across as a nice guy, and well-meaning. But in leaving him with no way to evaluate the accuracy and quality of data, his teachers and mentors betrayed him.
Lack of training in the evaluation of data quality is apparently an educational lacuna of most, if not all, AGW consensus climate scientists. They find no meaning in the critically central distinction between precision and accuracy. There can be no possible progress in science at all, when workers are not trained to critically evaluate the quality of their own data.
The best overall description of climate model errors is still Willie Soon, et al., 2001 Modeling climatic effects of anthropogenic carbon dioxide emissions: unknowns and uncertainties. Pretty much all the described simulation errors and short-coming remain true today.
Jerry Browning recently published some rigorous mathematical physics that exposes at their source the simulation errors Willie et al., described. He showed that the incorrectly formulated physical theory in climate models produces discontinuous heating/cooling terms that induce an “orders of magnitude” reduction in simulation accuracy.
These discontinuities would cause climate simulations to rapidly diverge, except that climate modelers suppress them with a hyper-viscous (molasses) atmosphere. Jerry’s paper provides the way out. Nevertheless, discontinuities and molasses atmospheres remain features in the new improved CMIP6 models.
In the 2013 Fifth Assessment Report (5AR), the IPCC used CMIP5 models to predict the future of global air temperatures. The up-coming 6AR will employ the up-graded CMIP6 models to forecast the thermal future awaiting us, should we continue to use fossil fuels.
CMIP6 cloud error and detection limits: Figure 1 compares the CMIP6-simulated global average annual cloud fraction with the measured cloud fraction, and displays their difference, between 65 degrees north and south latitude. The average annual root-mean-squared (rms) cloud fraction error is ±7.0%.
This error calibrates the average accuracy of CMIP6 models versus a known cloud fraction observable. Average annual CMIP5 cloud fraction rms error over the same latitudinal range is ±9.6%, indicating a CMIP6 27% improvement. Nonetheless, CMIP6 models still make significant simulation errors in global cloud fraction.
Figure 1 lines: red, MODIS + ISCCP2 annual average measured cloud fraction; blue, CMIP6 simulation (9 model average); green, (measured minus CMIP6) annual average calibration error (latitudinal rms error = ±7.0%).
The analysis to follow is a straight-forward extension to CMIP6 models, of the previous propagation of error applied to the air temperature projections of CMIP5 climate models.
Errors in simulating global cloud fraction produce downstream errors in the long-wave cloud forcing (LWCF) of the simulated climate. LWCF is a source of thermal energy flux in the troposphere.
Tropospheric thermal energy flux is the determinant of tropospheric air temperature. Simulation errors in LWCF produce uncertainties in the thermal flux of the simulated troposphere. These in turn inject uncertainty into projected air temperatures.
For further discussion, see here — Figure 2 and the surrounding text. The propagation of error paper linked above also provides an extensive discussion of this point.
The global annual average long-wave top-of-the-atmosphere (TOA) LWCF rms calibration error of CMIP6 models is ±2.7 Wm⁻² (28 model average obtained from Figure 18 here).
I was able to check the validity of that number, because the same source also provided the average annual LWCF error for the 27 CMIP5 models evaluated by Lauer and Hamilton. The Lauer and Hamilton CMIP5 rms annual average LWCF error is ±4 Wm⁻². Independent re-determination gave ±3.9 Wm⁻²; the same within round-off error.
The small matter of resolution: In comparison with CMIP6 LWCF calibration error (±2.7 Wm⁻²), the annual average increase in CO2 forcing between 1979 and 2015, data available from the EPA, is 0.025 Wm⁻². The annual average increase in the sum of all the forcings for all major GHGs over 1979-2015 is 0.035 Wm⁻².
So, the annual average CMIP6 LWCF calibration error (±2.7 Wm⁻²) is ±108 times larger than the annual average increase in forcing from CO2 emissions alone, and ±77 times larger than the annual average increase in forcing from all GHG emissions.
That is, a lower limit of CMIP6 resolution is ±77 times larger than the perturbation to be detected. This is a bit of an improvement over CMIP5 models, which exhibited a lower limit resolution ±114 times too large.
Analytical rigor typically requires the instrumental detection limit (resolution) to be 10 times smaller than the expected measurement magnitude. So, to fully detect a signal from CO2 or GHG emissions, current climate models will have to improve their resolution by nearly 1000-fold.
Another way to put the case is that CMIP6 climate models cannot possibly detect the impact, if any, of CO2 emissions or of GHG emissions on the terrestrial climate or on global air temperature.
This fact is destined to be ignored in the consensus climatology community.
Emulation validity: Papalexiou et al., 2020 observed that, the “credibility of climate projections is typically defined by how accurately climate models represent the historical variability and trends.” Figure 2 shows how well the linear equation previously used to emulate CMIP5 air temperature projections, reproduces GISS Temp anomalies.
The emulation passes through the middle of the trend, and is especially good in the post-1950 region where air temperatures are purportedly driven by greenhouse gas (GHG) emissions. The non-linear temperature drops due to volcanic aerosols are successfully reproduced at 1902 (Mt. Pelée), 1963 (Mt. Agung), 1982 (El Chichón), and 1991 (Mt. Pinatubo). We can proceed, having demonstrated credibility to the published standard.
CMIP6 World: The new CMIP6 projections have new scenarios, the Shared Socioeconomic Pathways (SSPs).
These scenarios combine the Representative Concentration Pathways (RCPs) of the 5AR, with “quantitative and qualitative elements, based on worlds with various levels of challenges to mitigation and adaptation [with] new scenario storylines [that include] quantifications of associated population and income development … for use by the climate change research community.“
Emulation of CMIP6 air temperature projections below follows the identical method detailed in the propagation of error paper linked above.
The analysis here focuses on projections made using the CMIP6 IMAGE 3.0 earth system model. IMAGE 3.0 was constructed to incorporate all the extended information provided in the new SSPs. The IMAGE 3.0 simulations were chosen merely as a matter of convenience. The paper published in 2020 by van Vuulen, et al conveniently included both the SSP forcings and the resulting air temperature projections in its Figure 11. The published data were converted to points using DigitizeIt, a tool that has served me well.
Here’s a short descriptive quote for IMAGE 3.0: “IMAGE is an integrated assessment model framework that simulates global and regional environmental consequences of changes in human activities. The model is a simulation model, i.e. changes in model variables are calculated on the basis of the information from the previous time-step.
“[IMAGE simulations are driven by] two main systems: 1) the human or socio-economic system that describes the long-term development of human activities relevant for sustainable development; and 2) the earth system that describes changes in natural systems, such as the carbon and hydrological cycle and climate. The two systems are linked through emissions, land-use, climate feedbacks and potential human policy responses. (my bold)”
On Error-ridden Iterations: The sentence bolded above describes the step-wise simulation of a climate, in which each prior simulated climate state in the iterative calculation provides the initial conditions for subsequent climate state simulation, up through to the final simulated state. Simulation as a stepwise iteration is standard.
When the physical theory used in the simulation is wrong or incomplete, each new iterative initial state transmits its error into the subsequent state. Each subsequent state is then additionally subject to further-induced error from the operation of the incorrect physical theory on the error-ridden initial state.
Critically, and as a consequence of the step-wise iteration, systematic errors in each intermediate climate state are propagated into each subsequent climate state. The uncertainties from systematic errors then propagate forward through the simulation as the root-sum-square (rss).
Pertinently here, Jerry Browning’s paper analytically and rigorously demonstrated that climate models deploy an incorrect physical theory. Figure 1 above shows that one of the consequences is error in simulated cloud fraction.
In a projection of future climate states, the simulation physical errors are unknown because future observables are unavailable for comparison.
However, rss propagation of known model calibration error through the iterated steps produces a reliability statistic, by which the simulation can be evaluated.
The above summarizes the method used to assess projection reliability in the propagation paper and here: first calibrate the model against known targets, then propagate the calibration error through the iterative steps of a projection as the root-sum-square uncertainty. Repeat this process through to the final step that describes the predicted final future state.
The final root-sum-square (rss) uncertainty indicates the physical reliability of the final result, given that the physically true error in a futures prediction is unknowable.
This method is standard in the physical sciences, when ascertaining the reliability of a calculated or predictive result.
Emulation and Uncertainty: One of the major demonstrations in the error propagation paper was that advanced climate models project air temperature merely as a linear extrapolation of GHG forcing.
Figure 3, panel a: points are the IMAGE 3.0 air temperature projection of, blue, scenario SSP1; and red, scenario SSP3. Full lines are the emulations of the IMAGE 3.0 projections: blue, SSP1 projection, and red, SSP3 projection, made using the linear emulation equation described in the published analysis of CMIP5 models. Panel b is as in panel a, but also showing the expanding 1 s root-sum-square uncertainty envelopes produced when ±2.7 Wm⁻² of annual average LWCF calibration error is propagated through the SSP projections.
In Figure 3a above, the points show the air temperature projections of the SSP1 and SSP3 storylines, produced using the IMAGE 3.0 climate model. The lines in Figure 3a show the emulations of the IMAGE 3.0 projections, made using the linear emulation equation fully described in the error propagation paper (also in a 2008 article in Skeptic Magazine). The emulations are 0.997 (SSP1) or 0.999 (SSP3) correlated with the IMAGE 3.0 projections.
Figure 3b shows what happens when ±2.7 Wm⁻² of annual average LWCF calibration error is propagated through the IMAGE 3.0 SSP1 and SSP3 global air temperature projections.
The uncertainty envelopes are so large that the two SSP scenarios are statistically indistinguishable. It would be impossible to choose either projection or, by extension, any SSP air temperature projection, as more representative of evolving air temperature because any possible change in physically real air temperature is submerged within all the projection uncertainty envelopes.
An Interlude –There be Dragons: I’m going to entertain an aside here to forestall a previous hotly, insistently, and repeatedly asserted misunderstanding. Those uncertainty envelopes in Figure 3b are not physically real air temperatures. Do not entertain that mistaken idea for a second. Drive it from your mind. Squash its stirrings without mercy.
Those uncertainty bars do not imply future climate states 15 C warmer or 10 C cooler. Uncertainty bars describe a width where ignorance reigns. Their message is that projected future air temperatures are somewhere inside the uncertainty width. But no one knows the location. CMIP6 models cannot say anything more definite than that.
Inside those uncertainty bars is Terra Incognita. There be dragons.
For those who insist the uncertainty bars imply actual real physical air temperatures, consider how that thought succeeds against the necessity that a physically real ±C uncertainty requires a simultaneity of hot-and-cold states.
Uncertainty bars are strictly axial. They stand plus and minus on each side of a single (one) data point. To suppose two simultaneous, equal in magnitude but oppositely polarized, physical temperatures standing on a single point of simulated climate is to embrace a physical impossibility.
The idea impossibly requires Earth to occupy hot-house and ice-house global climate states simultaneously. Please, for those few who entertained the idea, put it firmly behind you. Close your eyes to it. Never raise it again.
And Now Back to Our Feature Presentation: The following Table provides selected IMAGE 3.0 SSP1 and SSP3 scenario projection anomalies and their corresponding uncertainties.
Table: IMAGE 3.0 Projected Air Temperatures and Uncertainties for Selected Simulation Years
|Storyline||1 Year (C)||10 Years (C)||50 Years (C)||90 years (C)|
Not one of those projected temperatures is different from physically meaningless. Not one of them tells us anything physically real about possible future air temperatures.
Several conclusions follow.
First, CMIP6 models, like their antecedents, project air temperatures as a linear extrapolation of forcing.
Second, CMIP6 climate models, like their antecedents, make large scale simulation errors in cloud fraction.
Third, CMIP6 climate models, like their antecedents, produce LWCF errors enormously larger than the tiny annual increase in tropospheric forcing produced by GHG emissions.
Fourth, CMIP6 climate models, like their antecedents, produce uncertainties so large and so immediate that air temperatures cannot be reliably projected even one year out.
Fifth, CMIP6 climate models, like their antecedents, will have to show about 1000-fold improved resolution to reliably detect a CO2 signal.
Sixth, CMIP6 climate models, like their antecedents, produce physically meaningless air temperature projections.
Seventh, CMIP6 climate models, like their antecedents, have no predictive value.
As before, the unavoidable conclusion is that an anthropogenic air temperature signal cannot have been, nor presently can be, evidenced in climate observables.
I’ll finish with an observation made once previously: we now know for certain that all the frenzy about CO₂ and climate was for nothing.
All the anguished adults; all the despairing young people; all the grammar school children frightened to tears and recriminations by lessons about coming doom, and death, and destruction; all the social strife and dislocation. All of it was for nothing.
All the blaming, all the character assassinations, all the damaged careers, all the excess winter fuel-poverty deaths, all the men, women, and children continuing to live with indoor smoke, all the enormous sums diverted, all the blighted landscapes, all the chopped and burned birds and the disrupted bats, all the huge monies transferred from the middle class to rich subsidy-farmers:
All for nothing.
Finally, a page out of Willis Eschenbach’s book (Willis always gets to the core of the issue), — if you take issue with this work in the comments, please quote my actual words.