Climate science needs a critical review by skeptical experts

Anthony …. I went and read your link, and I find a lot of Cherry picked, subjective clap trap. In fact, your link is a perfectly good reason as to why a panel of interested and non interested parties should convene and iron out the truth. ……. Funny, the folks in your link oppose such. Wonder why?

Strange: As I understood it, the fate of Schrodinger’s cat was supposed to be determined by the decay of a single nucleus. However, you are right, I should have limited my discussion to chaos.

I’m dissatisfied with Essex’s presentation. I understand vaguely that their are significant problem with computational fluid dynamics. Nevertheless, as I understand it, engineers today design airplanes using computational fluid dynamics. They no longer go to wind tunnels with a scale model. Some CFD calculation apparently are reliable, but I’d never know it from listening to Essex.

Climate models may have been Alice-in-Wonderland when Essex started, but they stopped needing flux adjustments to deal with computational instabilities almost two decades ago. (It’s disgraceful that two IPCC reports were written without clearly raising the issue of flux adjustments.) What are the other signs that models can’t do what we are asking then to do? Initialization uncertainty can be addressed by multiple runs. We have 100 runs from one model today, Parameter uncertainty is currently unsolvable, but ensembles with perturbed parameters suggests that it is difficult to find a model with ECS less than 2. Reducing grid cell size allowed the QBO to be properly modeled in the stratosphere and may someday make an MJO appear. We have cloud resolving models that can be compared with AOGCMs.

Can you suggest anything more informative/compelling that Essex’s propaganda?

Dr. Strangelove replied: “Where is the wind tunnel where climate modelers test Earth’s climate? They don’t have empirical parameters derived from experiments. They just plug in the parameters and call the results, scenarios.”

Thoughtful answer. The ability to reproduce today’s climate might be considered to be the “wind tunnel” of AOGCMs, but experiments with ensembles of perturbed parameters and a panel of observed properties of today’s climate have failed to narrow the plausible range for any particular parameter. Model with N parameters requires finding a global optimum an N-dimensional space with many local optima. With ensembles of simpler models, a global optimum can’t be found by trial and error. Such ensembles show a wide range of ECS, but rarely below 2. (In one case, I know of this may be because low climate sensitivity models are unstable towards cooling.) With sophisticated computationally-intensive models, tuning parameters one-by-one is likely to end up at a local optimum.

Hindcasting certainly doesn’t prove anything when you can’t be sure what fraction of recent warming is unforced. And models validated by hindcasting can be used to make attribution statements about past warming, that is circular reasoning.

Pat: You are correct that I shouldn’t have mentioned the non-deterministic nature of quantum mechanics in a discussion of the macroscopic world. However, you totally ignored my comments on how deterministic chaos changes the nature of the predictions that can be made about phenomena such as weather, climate, coupled pendulums, the solar system on long time scales, etc.

Pat says: “Parameter uncertainty means the underlying physics is not known. The climate energy state is misrepresented. The projections are an expression of ignorance, no more.”

This is incorrect. We know the underlying physics of turbulence, cloud microphysics, and other phenomena. We can’t apply the physics we know to a large grid cell. Clouds start appearing in a grid cell before the average relative humidity for the grid cell reaches 100%. IIRC, accurate calculations of the transport of heat and water vapor perpendicular to the surface of stationary ocean requires working with grid cells of 1 mm^3. Nevertheless, an equation for bulk transfer by turbulent mixing with an empirical constant provides reasonably accurate fluxes: the rate of evaporation is proportional to wind speed and undersaturation (both assessed a specified distance, usually 2 meters, above the surface). The validity of such empirical constants has been tested under carefully controlled conditions in the laboratory (where the surface is no longer flat) and with fine scale CFD computations, and in the field (where definitive experiments are challenging). Similar approximations are used in aeronautical engineering (where CFD is practical): Lift and drag are proportional to the wing area, the density of air and velocity squared. An empirical coefficient of lift and drag for a given wing shape is used. Careful experiments show these coefficients are not constant and applicable to all situations, but they provide useful approximations. They are NOT, as you claim, expressions of IGNORANCE. They are practical approximations used in the real world.

We may be ignorant of the limits beyond which these approximations work. What happens to the above equations for lift and drag when the flow over wing is no longer laminar? We do experiments to find out. These equations fail when the angle of attack of a wing is too steep and the flow detaches from the wing. What happens when the wind over the ocean is strong enough to produce a fine mist of water droplets. We’ve done those experiments too. However, it is my understanding that the output from CFD and other simulations varies in unpredictable ways. In the laboratory, the number of cloud condensation nuclei determines the size of the water droplets that initially makes up a cloud and determines its reflectivity (Toomey effect?) However, small water droplets evaporate and form more stable large droplets? How long does that take? Apparently this happens faster in the real atmosphere than in the laboratory, because AR5 now says the Toomey effect is usually insignificant in the field. Modeling groups that accept this conclusion are re-parameterizing cloud reflectivity.

AOGCMs provide reasonable representations of current climate, in part because key parameters have been tuned to do so. The question is whether they provide a useful estimate of what our world with 2X or 3X more CO2 will be like. There is plenty of room for skepticism here! IMO, skeptics would be far better discussing this problem rather than making dubious and inaccurate assertions that are easy for the consensus to reject. Nevertheless, many prefer the Stephen Schneider approach: Telling scary stories, making simplified dramatic statements, and hiding doubts. I personally prefer his definition of ethical science.

Pat Frank mistakenly claims: “Argo floats, by the way, exhibit about ±0.6 C of systematic uncertainty, when calibrated in the field (not back in the lab). They cannot give anywhere near the accuracy necessary to measure the tiny change in ocean heat content required to resolve any thermal effect of CO2 emissions.”

This is random error, not systematic uncertainty. Thus, the authors of the paper you are referring to show their error going down over longer periods. When averaged over a decade or two – rather than a few months – and over the entire planet – not just a problematic piece of the Atlantic where the spacial resolution of the Argo buoys is poorly sampled – random errors average out. It isn’t clear that the authors of the paper believe they have proven ARGO isn’t meeting its design objectives and they certainly don’t make this claim. And the developers of ARGO, weren’t inexperienced idiots either. They understood the challenges they faced from previous experience inadequate equipment and sampling and designed a system that would achieve their goals.

https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2006JC003825

“Results for the main North Atlantic region show a reduction in RMS difference from 29 ± 3 Wm−2 for monthly heat storage to 12 ± 2 Wm−2 for seasonal (i.e., 3 monthly) heat storage and 4 ± 1 Wm−2 at biannual scales.”

Any error that is going down with time is clearly not a systematic error in the eyes of these authors. I see no reason to assume that the current value for ocean heat uptake (0.7? W/m2) averaged over the planet over more than decade is unreliable.

Furthermore, a CONSTANT systematic error in a single point calibration (for example, the error in b for y = mx + b), doesn’t interfere with out ability to measure a CHANGE in y accurately. dy depends only on the accuracy of m and dx. A VARIABLE systematic error is a whole different matter. However, we debated this point several years ago.

To the extent that poorly-sited temperature stations have a CONSTANT systematic error, one might expect them to show the same trend as well-sited stations. This is what Andy found. And if urban stations have a CONSTANT bias compared to non-urban stations (a more dubious hypothesis), then you might expect them to have similar trends in temperature. This is what BEST found. If you created model data sets with constant and variable systematic errors, this is what you would find too.

Pat Frank wrote: “The same limit is true for parameterized climate models. Their parameters are adjusted to produce a fit to bounded observables. There is no reason whatever to suppose they are applicable to conditions outside those bounds. Climate model projections fail within the limits of your own argument.

Frank wrote: “AOGCMs provide reasonable representations of current climate, in part because key parameters have been tuned to do so.”

Pat Frank replied: No, they do not. AOGCMs provide a reasonable reproduction of target observables; the observables that the AOGCM was tuned to reproduce. That is not the same at all of achieving the correct climate energy state, and deriving the observables from theory.

Frank now adds: This blog post from Isaac Held shows that an AOGCM is able to reproduce features of our climate that were not used to tune the model. In this case, the AOGCM is predicting the location and strength of the jet stream and how it changes between the seasons. The detail is awesome. The jet stream is a result of the large scale motion of the Hadley, Ferrel, and Polar cells being acted on by the Coriolis effect. The model is clearly doing something right!

https://www.gfdl.noaa.gov/blog_held/60-the-quality-of-the-large-scale-flow-simulated-in-gcms/

It is worth noting what the model does wrong. The model is run at higher resolution than most CMIP5 models. And this happens to be the output of a AMIP run where SSTs are specified and not controlled by the model. (When the CMIP5 models were forced with historic SSTs, their climate feedback parameters were appropriate for an ECS from 1.5-2.0 K. One might speculate that normal model runs with historic forcing were directing heat to locations on the planet where fewer W/m2 escape to space per degK of average surface warming.) I’d also like to ask how the model’s ability to properly reproduce the jet stream varies with how the model is parameterized.

My favorite climate paper is Tsushima and Manabe (2013), because it compares observed and modeled changes in OLR and reflected SWR during the seasonal cycle. GMST (not an anomaly) rises and falls 3.5 K every year, so seasonal changes observed by CERES are huge. And reproduced every year. The models all do a great job with the changes in OLR emitted through clear skies, where only water vapor and lapse rate feedbacks operate. The models do poor and mutually inconsistent job at reproducing the changes in OLR from cloud skies and SWR from both clear and cloud skies. The changes in LWR are highly linear and look like responses that are complete within one month, but the changes in SWR are not and appear to have some lagged components (a phenomena that has been reported by others). IMO, this is proof that we can’t trust the feedbacks (and therefore the climate sensitivity) of AOGCMs.

IMO, an honest and candid assessment of AOGCMs is far better than mischaracterizing the importance of getting the cloud fraction wrong (at least according to 1980’s cloud data). It’s too easy to dismiss all skeptics if we take such arguments seriously. As I mentioned before, models are tuned to have no radiative imbalance with a Pre-industrial atmosphere. All models are spun up and run for a couple of centuries to prove that the model can produce a stable temperature (with about 0.1 K of unforced internal variability) before a forcing is applied. The existence of disagreements between various models and disagreements with observations of the type you have noted can be explained by offsetting errors in cloud albedo or emissivity, or surface albedo, among other factors. This must be the case, because the model wouldn’t produce a stable temperature with an large imbalance at the TOA. Or it would equilibrate to a radically different temperature. Before 2000, models required “flux adjustments” because they couldn’t be tuned to afford a stable temperature near 287 K. But that isn’t the case today.

Pat wrote: “Your Isaac Held evidence of model/observational conformity, is model/model conformity. Held is using reanalysis as his observations. Reanalysis uses modeled extrapolation of observational data and calls that extrapolation “observations.” Data that have gone through models extrapolations suffer from the same sort of errors that the model projections themselves exhibit. It’s no surprise that models can emulate models.”

Of course, one always needs to ask if a feature found in a reanalysis is present in the re-analyzed data or introduced by the biases of the model used to assemble climate data from a variety of sources. In fact, I tried to ask this question when the post first appeared and was mis-understood. The original post appears to have revised to address this question.

However, you need to remember that most AOGCMs don’t reproduce the jet stream nearly as well Held’s AOGCM, which possessed higher resolution than normal and was run in AMIP mode (ie with historic SSTs). The seasonal change in the jet stream shown by Held is not a common pattern produced by all climate models. The degree of agreement between re-analysized observations of winds and the prediction of winds in AOGCMs is something that has evolved over time as models have become more sophisticated. So this agreement shouldn’t be dismissed as a models vs models comparison.

Pat Frank mischaracterizes some of Held’s caveats: “Secondly, Held himself admits that time-wise projections are unreliable. He says, “Importantly, the model can propagate information from data rich regions into data poor regions if this propagation of information is fast enough compared to the time scale at which errors grow. (my bold)”

Frank notes that Held continued: “For climatological circulation fields such as the ones that I have shown here reanalyses provide our best estimates of the state of the atmosphere. For the northern hemisphere outside of the tropics these estimates are very good — I suspect that they provide THE MOST ACCURATE DESCRIPTION OF ANY TURBULENT FLOW IN ALL OF SCIENCE. [My emphasis.] For the tropics and for the southern hemisphere the differences between reanalyses can be large enough that estimating model biases requires more care”

Pat continues: “The upper limit time scale over which errors grow 100x larger than the perturbation of CO2 emissions is a year. A better estimate of the time over which a model projection becomes unreliable is a week.”

Weather projections become less skillful within a week, because chaos amplifies miniscule errors in initialization. Climate models are run for long enough periods that chaotic fluctuations in temperature average out in the long run. Negative feedback prevents climate from drifting too far from a stable equilibrium temperature. For most climate models, the standard deviation of the temperature variation in a century-long pre-industrial run is about 0.1 degK and this stable temperature is within 2 degK of our best estimate of pre-industrial and often much closer. In systems with negative feedback, errors do NOT grow exponentially.

For example, we know that a random walk process on the average moves an increasing distance from the starting position (that grows with the square root of the number of steps). Now modify your random walk with a negative feedback: Heads takes you one step to the right minus 0.1 steps times the number of steps you already are to the right of the origin. Tails does the opposite. Now your random walk is constrained by negative feedback to remain near the origin. Negative feedback is inherent in the Stefan-Boltzmann law and the fact that our planet’s temperature hasn’t drifted far for the last 70 centuries of the Holocene.

Frank earlier wrote, “The existence of disagreements between various models and disagreements with observations of the type you have noted can be explained by offsetting errors in cloud albedo or emissivity, or surface albedo, among other factors.”

Pat notes: “You offer an exercise in handwaving: “…can be explained by…” A complete physical theory would allow one to say, ‘Is known to be caused by…’ If you don’t know that offsetting model errors do not remove uncertainty from an expectation value, then you don’t know anything about physical error analysis. Every single climate modeler I’ve encountered is in that group. The ±4 W/m^2 cloud forcing error is not a mischaracterization. It’s a systematic model error that appears in every single step of a projection.

Frank revises: Model developers are no longer developing models with a significant radiative imbalance at the TOA that must be corrected by flux adjustments. When a model has a significant bias in cloud cover, then there MUST BE a compensating bias somewhere else. OTHERWISE these models wouldn’t exhibit a reasonable and stable pre-industrial temperature for a century or two before forcing begins. And despite these offsetting errors, such models all reproduce current climate reasonable well. They get temperature about right, deserts and rain forests in the proper locations, ocean current and wind blowing in the right direction, etc. These models clearly have some utility in reproducing what we observe. That is not the important question. The important question is: Do the models properly represent the increased amount of OLR emitted through the atmosphere and SWR reflected per degK of surface warming. If not, the models aren’t useful for predicting climate sensitivity.

Was error propagation an important part of your work on the determination of molecular structures?