I’ve been struggling, more than usual, to find things to write about. Everything seems to just be a bit of a mess. The pandemic itself, how it’s been handled in some cases, and the protests in the USA, especially how the protestors are being treated by the police. I just don’t feel that I really have the words to describe what’s currently happening in a way that would do it justice.
However, given that it’s been rather quiet here, I thought I would just highlight one paper that I found interesting, and useful. The lead author is Flavio Lehner, and the paper is called Partitioning climate projection uncertainty with multiple large ensembles and CMIP5/6. The paper seems to be open access, so I don’t need to say too much. Essentially, it ues ensembles of models to estimate the sources of uncertainty and their magnitudes.
One of the key figures is below. It shows 3 different model ensembles, their global surface temperature projections for different scenarios, and – finally – how the fractional contribution to total uncertainty. The key results are that for long-timescales (many decades) internal variability contributes little to the total uncertainty (essentially, it averages out). The largest source of uncertainty is scenarion uncertainty (i.e., how much are we going to emit). A similar result is obtained if you consider changes in global mean precipitation.
Credit: Lehner et al. (2020)
Although the model uncertainty (defined as uncertainty in the forced response and structural differences between models) is not negligible, it’s clear that a dominant source of uncertainty essentially relates to what we do (i.e., how much do we emit). I realise that using “we” is a bit simplistic, since a small fraction of the world’s population dominate the emissions budget, but it’s still clear that future climate change, and what we will have to deal with, depends mostly on future emissions. This is something that we can influence, even if determining how we do so is not a trivial. I also realise that some might argue that the scenario uncertainty is somewhat smaller than indicated in this paper, since some of the scenarios are much less lilely than others. Although true, I don’t think it really changes the basic message.
I have noticed some discussion about how we tend to ignore adaptation over emission reductions. There’s some truth to this, and we will certainly have to develop some adaptation strategies to deal with the changes that are now unavoidable. However, until we get net emissions to ~zero, the climate will continue to change, as will our adaptation strategies. I still think that John Holdren’s comment that [w]e basically have three choices: mitigation, adaptation and suffering. We’re going to do some of each. The question is what the mix is going to be. The more mitigation we do, the less adaptation will be required and the less suffering there will be, is worth bearing in mind.
Links:
Partitioning climate projection uncertainty with multiple large ensembles and CMIP5/6, paper by Flavio Lehner et al.
Mitigation, adaptation and suffering – short post by the late Andy Skuce, where I found the John Holdren quote.
...and Then There's Physics 
You will not understand what is going on in America if you refer to protesters. They are looters who have no respect for the law and decent people. They have no respect for the hard working people who have created the business and homes they are destroying.
You will never understand the climate if you continue to believe that reducing human carbon emissions to zero will stop climate change.
morpheusonacid — Most are just protesters, in peaceable assembly. There have been a few looters as well, unfortunately.
morpheus,
I don’t think your comment deserves a response.
About uncertainty:
https://www.insidescience.org/news/when-scientists-find-nothing-value-null-results
Bayes to the rescue!
morpheus will never understand much of anything, period.
ATTP are we reading this the same way?
–
“The key results are that for long-timescales (many decades) internal variability contributes little to the total uncertainty (essentially, it averages out).”
–
I do not see this as the key finding, rather a statement of the parameters being put in.
By definition internal variability is defined as fluctuations around some predetermined real value.
As time goes by the fluctuations balance out and the true value is revealed shed of dross. In other words it must always reduce to zero.
angech,
Internal variability emerges from the models. It’s not really an input. That it averages out is not a surprise, but that’s mostly because of conservation of energy, not because it is prescribed to average out.
“I do not see this as the key finding,”
And yet so many skeptics don’t understand that the fact that the observations fall within the spread of the model is all that you can reasonably expect from the models, and that the observations being in the lower half of the distribution (at the time) does not imply the models are running hot.
“rather a statement of the parameters being put in.”
Not parameters, the initialization of the model state at the start of the run. This is the basis of Monte Carlo simulation that we discussed on the previous thread. We can’t predict the exact course of internal climate variability (which is chaotic and highly sensitive to initial conditions) – the best we can do is to simulate different realisations of internal variability (which is why each model run is diffierent) and produce a distribution over plausible future climate states.
“By definition internal variability is defined as fluctuations around some predetermined real value.”
I don’t think that actually is the definition. Citation required.
“By definition internal variability is defined as fluctuations around some predetermined real value.”
I’m pretty sure that is wrong. It assumes that there is only one correct “real” value, and that any time a model differs from the One Value That Rules Them All then the model has an error.
Even in as simple a system as a normal distribution, a different sample will have a different sample mean, and that mean is the “real value” of the mean of that sample. It is not an error – it’s just not a perfect representation of the mean of the entire population.
Anders: morpheus’ comment deserves an insult. Part of the reason why epidemiological data is broken down by race in the US is because there is much history behind it;
DM said:
Who says that internal climate variability is ” chaotic and highly sensitive to initial conditions”? Citation required.
“Any climate variable projection derived from a single simulation of an individual climate model will be affected by internal variability (stemming from the chaotic nature of the system), ”
Click to access WG1AR5_Chapter12_FINAL.pdf
“highly sensitive to initial conditions” is part of the definition of chaotic.
Now you may disagree, that is fine, but if you are trying to overturn the mainstream scientific postion, then it is incumbent on you to know what the mainstream position is and not require me to give references on the basics. You should already be aware of them.
“By definition internal variability is defined as fluctuations around some predetermined real value.”
The IPCC define “climate variability” (glossary, page1451) as:
i.e. it is defined in essentially statistical terns, however that is not fluctuations around predetermined real values. They go on to differentiate this from internal variability:
So internal variabilty is defined in terms of processes (i.e. physics) other than forcings, not in statistical terms. In other words, it is the variability in climate that would remain if forcings were held constant.
BTW internal variability being chaotic doesn’t mean it won’t average out. If you have an iron double pendulum in a vacuum with frictionless joints, there will be an average value for the angles of the joints if you leave it to run long enough that will not depend on the initial configuration. If you place a magnet to one side, those mean values will be different in an entirely predictable manner (e.g. by Monte Carlo simulation).
” that will not depend on the initial configuration.”
… except in the potential/kinetic energy available at the start.
ATTP,
“Internal variability emerges from the models. It’s not really an input. That it averages out is not a surprise, but that’s mostly because of conservation of energy, not because it is prescribed to average out.“
_
Agreed, the fact though is that it diminishes to zero over time. In all the scenarios.
And usually by halfway through the graphs as shown.
I guess it indicates that the system is self balancing without forcing upsets which will be a blessing in the long term even if not for us.
“Agreed, the fact though is that it diminishes to zero over time.”
And yet so many skeptics are sure that the “pause” meant that the models were running hot…
It doesn’t diminish to zero over time angech. It averages out to zero. Like a coin toss would if you labelled the sides -1 and +1 rather than heads and tails. The coin doesn’t go over time to landing on its edge.
The interesting feature is the model uncertainty of CMIP6 being higher than CMIP5 (which I suspect may have been higher than CMIP3?). Those pesky “unknown unknowns” becoming “known unknowns” is good scientific progress.
I don’t think it will be too much of a surprise to anybody that internal variability isn’t very important on centennial scales or that scenario uncertainty is the dominant factor, but partitioning the uncertainty ought to be useful information for those needing to make the decisions.
DM said:
Yes, I am aware of what internal climate variability is, which is best exemplified by a regular change in temperature on an annual basis. At very long time-scales it appears to track orbital changes.
Chaos as an explanation is used as a placeholder to aspects that are not well understood, IMO.
Daily cycles of climate : well understood, not chaotic.
Annual cycles of climate: well understood, not chaotic.
Volcanic episodes in climate: well understood, response not chaotic.
El Nino cycles in climate: not well understood, but more than likely not chaotic.
Solar spot cycles impact on climate: weak, not chaotic
GHG forcing of climate: well understood, response not chaotic.
Glacial cycles of climate: not completely understood, but match to Milankovitch cyclic forcing does not suggest chaotic limit cycles as a mechanism.
So what exactly is the chaotic behavior that is assumed to be mainstream? I do understand that weather is chaotic, but weather is not climate.
Paul Pukite — Certain ocean currents, for example in the North Pacific, either north or south of the Aleutian chain.
“So what exactly is the chaotic behavior that is assumed to be mainstream? I do understand that weather is chaotic, but weather is not climate.”
I answered your question, you did nothing with the answer and have gone on to ask another question. Sorry, I am not playing along with your rhetorical game. If you want to know why I lost interest in your work on this topic, this sort of behaviour is the a good example.
Especially as you provided the answer yourself “El Nino cycles in climate: not well understood, but more than likely not chaotic.” Not everybody agrees with you.
FWIW, the obvious Google Scholar query provides a fair bit of backing for the idea that ENSO is chaotic.
Here are the latest Google scholar searches (since 2020) on ENSO & chaotic, which suggests how uncertain the belief is:
“Chaotic signature of climate extremes” in Theoretical and Applied Climatology (2020) DOI:10.1007/s00704-019-02987-6
“The observed trends in climate time series can either be deterministic (external forcing) or stochastic (intrinsic variability) (Franzke 2011). …. In the case of El Nino Southern Oscillation (ENSO), the discussion is about whether a deterministic or stochastic model best describes it. Rejection of chaos in the time series of ENSO has been disputed based on analysis carried
out using Largest Lyapunov Exponent (Kawamura et al 1998), Correlation Dimension (Kawamura et al 1998), close return plots (Ahn and Kim 2005) and determinism (Binder and Wilches 2012). The theory that ENSO time series is a stochastic system rather than a chaotic one has led to the development of stochastic approach for ENSO (Ubilava and Helmers 2013; Hall et al 2001).”
“Dynamics with Chaos and Fractals” (2020) – Springer – DOI:10.1007/978-3-030-35854-9_10
“The impact of SST variability on global climate is clear during global climate patterns, which
involve large-scale ocean-atmosphere fluctuations similar to the El Niño-Southern Oscillation (ENSO). Sensitivity (unpredictability) is the core ingredient of chaos. Several researches suggested that the ENSO might be chaotic. It was Vallis [65, 66] who revealed unpredictability of ENSO by reducing his model to the Lorenz equations.”
The belief is based on the fact that since Lorenz equations are chaotic then the actual climate system may be chaotic. Yet, there is no indication that the climate follows the Lorenz equations. A Lyapunov exponent can only be extracted from a mathematical model such as Lorenz, as it is impossible to extract from data alone since a Lyapunov instability can only be detected from tweaking model parameters.
There are also a couple of ML investigations from 2020 on the topic, trying to determine the predictability of ENSO
“Extended-range statistical ENSO prediction through operator-theoretic techniques for nonlinear dynamics” — DOI:10.1038/s41598-020-59128-7
“Temporal Convolutional Networks for the Advance Prediction of ENSO” — DOI:10.1038/s41598-020-65070-5
These last two are from Nature Scientific Reports, who have a reputation for publishing bleeding-edge research.
Paul if you ask questions and then demonstrate that you were not interested in any answer I might have given, don’t expect me to show any interest in following your games any further. Especially when you move the goal posts from “more than likely not chaotic.” to “which suggests how uncertain the belief is:”.
The ML stuff you quote is a bit disingenuous as they appear to be only relatively short term predictors of ENSO (~ 1 year compared to its ~decadal characteristic time-scale) and so are not evidence that it is isn’t chaotic. In fact one of the mentions “Yet, due to highly nonlinear and chaotic dynamics (particularly during ENSO initiation),”.
Hint: a double pendulum is a classic chaotic system, but if you sample it at a sufficiently high rate, you can use Machine Learning to make short term predictions of its motion.
BTW the distinction between stochastic and chaotic is a distinction in models, not necessarily reality. It is not at all clear that true randomness exists in the real world on macroscopic scales (i.e. except perhaps quantum effects). Random is most often used to mean something we can’t predict, but that doesn’t mean we can’t predict it because it is chaotic (i.e. deterministic (non-random) but dependent on imperfectly known initial conditions). It is important not to confuse models with reality. Modelling something with random variables does not mean it actually is random.
“But Enso” drive-by done.
Thanks.
Matt gives the lukewarmer uncertainty treatment to COVID-19:
https://twitter.com/CT_Bergstrom/status/1270226183485976584?s=20
COVID19 is fatal in hospitals because that is where you take the people who have been so severely affected by it that they have a high risk of death. It quarantines those people in a space where they are in contact with people taking a LOT of precautions to avoid cross infection.
If you left them out in the community, or in an old peoples home, they would be much more likely to infect other that shared that environment before they died, so the number of cases would spread further and faster.
I suppose the ‘evolutionary process’ that Ridley alludes to is the one where a virus that is endemic becomes less virulent because the versions that fail to kill quickly can infect more people. But that process does not reduce the amount of deaths from the virus while that selection process occurs. The extra virulent forms still kill.
Several virus infections adapted in a different way to the challenge of infecting as many people as possible before the person died. They had long asymptomatic infectious states, or just kill very slowly. HIV being the obvious example.
Hoping that a virus will become less virulent if the infection spreads widely is basically betting that the virus will become its own ‘vaccine’ and create herd immunity. While miraculously failing to kill the most susceptible in the process.
You could almost forgive Ridley his stupidity over AGW because he’s not a physical scientist. Still, he does have a science education (zoology PhD and presumably biology as an undergrad) and was a science journalist before he went off the deep end. So he ought to know (a) how to assess evidence and (b) the risk of going public without evidence. Here, there is no excuse. Indeed, I bet if I read his PhD he’d have used game theory at some point. Shame he didn’t engage brain and apply it before engaging fingers and tweeting. Proof if ever it was needed that politics, like religion, makes even smart people stupid.
From Bergstrom’s thread – a neater version of what I said in the previous ATTP thread:
On the suffering caused by not mitigating: The economic costs of Hurricane Harvey attributable to climate change and Climate crisis to blame for $67bn of Hurricane Harvey damage – study.
DG
“Another way to think about it is that the bad things that COVID does occur after transmission and only to a small fraction of people, so natural selection can’t really “see” those things or operate effectively on them.“
Good point?
Natural selection does not work by seeing things though, that is a human interpretation.
The book is called the Blind watchmaker for a reason.
Bersgtrom knows that angech, just as I do.
The scare quotes around “see” were there for a reason.
And in this context “see” doesn’t mean seeing things through in a teleological sense. It means not being a part of the extended phenotype (in Dawkins-speak) which impacts transmission, and so not something that can be selected for (blindly to intent but not to the phenotype) among mutations which change the transmission rate.
angech:
True, but irrelevant. In context, Bergstrom’s figurative quote marks around “see” mean only that selection is weak for reduced virulence in the pathogen, because it propagates before killing its host; and because lethality to humans is already low. IMO, while the two selective forces may be weak, ceteris paribus they would tend to reduce COVID-19 mortality over centuries to millennia. OTOH, an effective vaccine would minimize both factors, although perhaps the virus will abide in non-human hosts, with new novel strains repeatedly jumping to us. It’s complicated, IOW, so make predictions at your own risk!
Heh. I’m proud to claim my mind and D_G’s think alike ;^).
You’re right Mal, selection is weak but not absent. As a geologist I have no problem with a 1% advantage leading to dominance over time, as long as the advantage remains during that time. But as a geologist I’m used to loooong timescales. It’s one reason I would argue that form following function is a stronger statement in the fossil record than at present day. Fossil preservation is so rare that everything we find has to represent species which were around for a long time over a large area, and can be presumed to have been successful because they were well adapted to that environment. We don’t see the unsuccessful variants or Hopeful Monsters which might crop up in real time and confuse biologists. We don’t have to worry about things like whether the salmon-eating Orca are well adapted to eating salmon because they always have, or maladapted 😉 because their large-mammal prey has been over-hunted by humans and are hanging in there but on their way out.
If Covid-19 becomes just another common cold coronavirus in a hundred or a thousand years’ time (or even in decades), that has no relevance to policy or to human misery in the near term.
@-Dave_G
“…that has no relevance to policy or to human misery in the near term.”
I suspect Ridley is conflating the mote of truth about the possible evolution of a less virulent virus over large timescales with the immediate response to the epidemic to try and make a policy relevant argument.
That the policy response has been very damaging to the economic well being of BAU and that the policy balance between reducing human misery and causing economic damage has shifted too far in favour of preserving life at the expense, (literally) of business.
But then for historical reasons I tend to attribute malicious motivation to Ridley rather than give him a pass for wilful ignorance. The economic hardship from the failure of Northern Rock was disproportionately skewed towards the smaller stakeholders.
Some more from Bergstrom, which is the flip side of my distinction between things we can be (reasonably) complacent about in the fossil record because we only see systems that have reached equilibrium (or we see a flip to a new equilibrium, but usually with a gap rather than seeing the transition in full), but not in real-time biology where systems can be out of equilibrium. An evolutionary biologist considers the virulence of emerging infectious diseases.
Trade-off models would be equivalent to letting a game-theory model run to a Nash Equilibrium. You don’t care about how it gets to equilibrium, so you can make simplifications and take short-cuts, and there might even be a simple, one-step closed-form solution. If you care about now and the near future, not the far future, you need to model the path to the far future, probably only a little bit of the way there, with all the dead ends, U-turns, stochastic blips and local fitness highs that entails.
The paper Bergstrom was referring to is here: Invasion thresholds and the evolution of nonequilibrium virulence (the link in Bergstrom’s commentary is broken).
There are exceptions to every rule of course, and in geology one is the Al2SiO5 triple point between andalusite, kyanite and sillimanite. It anchors a mesh of mineral and melting reactions in much the same way the H2O triple point does in other fields. Unfortunately the kinetics of the phase change are very slow and it’s not unusual to find two or even three phases present 😦 . However we know from other reactions that we don’t fortuitously have lots of rocks which formed at the triple point, and often you can tell which is the new phase and which is the disappearing phase.
To compound the problem, the entropy contribution to the free energy change is unusually large relative to the enthalpy, so things we normally ignore like grain size, degree of crystallinity and presence of impurities start to matter. Lab experiments become difficult because you want finely crushed starting material, especially for slow reactions, but then the surface area increase and defects introduced to the crystal lattice change the entropy and move the equilibrium in P,T space. And to make matters worse, sillimanite in nature often occurs as fibrolite, extremely fine fibres like asbestos which are fine enough to change the entropy of the natural mineral, and of course each case is different. So even if you had a perfect lab experiment, it would give the wrong answer when applied to a real-world rock. Nevertheless, we do our best.
It sticks in my mind because an international collaboration had come up with a consensus mesh of mineral and melting reactions which was upended just as I was doing my PhD, with the onset of results on Al2SiO5 and other systems using tiny samples benefiting from miniaturisation, semiconductors, wide availability of SEMs with EDAX ,etc. They broke the mesh, but a new one was not built until after I had written up. Hence my thesis contains the paragraph:
Fortunately my examiners bought that line. Sometimes science is messy.
Pingback: Extreme precipitation events | …and Then There's Physics