Let me say, at the outset, that this will (hopefully) not be terribly technical in nature. I’m not looking to convince anyone regarding any particular “climate-related” policy or position. I am, however, arguing that scientism is rampant within the climatology discipline, and even more so in the world of climate-related public policy and media. As a result, I will not be making much effort to be balanced: one side of the argument (namely, the “received wisdom” position that fossil fuel use at current levels poses an existential threat to humanity) is very well-represented in academia, public policy and the media. The other side (or, indeed, the fact that there is another side) of the debate on the relationships among carbon emission, climate and human flourishing, is not well-represented at all, being actively suppressed by many supposed neutral actors.
Also, the title of this post notwithstanding, I will be principally dealing with the phrase CAGW (Catastrophic Anthropogenic Global Warming) rather than “climate change.” This is because “climate change” is nearly meaningless as a phrase, and so broad as to be unfalsifiable – here is a discussion of climate change being adduced as the cause of low water levels in the Great Lakes, high water levels in the Great Lakes, and rapid changes in water levels in the Great Lakes. This is the actual, relevant time series over a 30-year period, with a 100-year mean superimposed (H/T WSJ via PowerLine):
Given that CO2 emissions have gone up steadily during this time, I’m not sure what signal I’m supposed to be seeing here. Climate change: is there anything it can’t do?
A word about sources
One of the unfortunate features in the climate change “debate” (or rather shout-down) is that anyone who exhibits less than full-throated support for the “received wisdom” ruling-class position described above is labelled a “climate denier.” I will be using data that has been publicized (and, in some cases, generated) by people who have been labelled in this fashion. This group, however, includes eminent academic (and former academic) climatologists John Christy, Judith Curry, Richard Lindzen, Roger Pielke (Sr.), Fred Singer, and Roy Spencer, as well as public policy specialists like Steven Hayward, Bjørn Lomborg, Ross McKitrick, and Roger Pielke (Jr.). Also, physicists like Robert Brown, William Happer, Steven Koonin and Andy May. It also includes activists like Tony Heller (an engineer/geologist by training) and Anthony Watts (a meteorologist by training) who can be, well, a bit intemperate in their phrasing. They’re a diverse lot – probably each of these has a different major point of conflict with the “received wisdom” position. What they have in common is that they don’t believe that the “received wisdom” position should be believed simply because other very smart (or very powerful) people seem to believe it.
Similarly, one shouldn’t believe what they (the “deniers”) say just because they say it. If you don’t like one or more of my sources, that’s fine, but even if what they say isn’t “right because they say it” it’s not “wrong because they say it,” either. Examine the data for yourself – the sources I am using are usually pretty good at telling you exactly where they got their data.
TL;DR
This essay focuses on three aspects of the current CAGW debate shouting match that are particularly related to “scientism”
- Overconfidence in models – specifically, a discussion of John Christy’s “tropical hot spot” metric for model assessment, and the current state of play regarding cloud modelling.
- Misuse of models – specifically, the inappropriate use of ensemble averages of models, and a discussion of Judith Curry’s notions of “spinning the models” (both of these are examples of politicization of science).
- Lying with statistics – specifically, a discussion of the main summary infographic of the 2018 National Climate Assessment, in which many of the graphs are apparently designed to mislead, not inform (another example of politicization of science).
Closing summary: climate science is a science in its infancy; I have every confidence that it will get better, but in the meantime it will necessarily get a LOT wrong. It is irrational to destroy society and attempt to build it anew on the say-so of an infant scientific discipline!
Overconfidence in models
There are many good possibilities for examples here. I will choose two: the missing “tropical hot spot” (good video of John Christy lecture on the subject here) and the problem of clouds (good overall introduction to current state of play here).
Climate prediction is necessarily predicated on the use of mathematical models. Since the stakes are so high (the costs of actually doing even a fraction of what is recommended by climate alarmists are truly stunning), it is incumbent on all of us to make sure we have our models as accurate as possible. Some hindcasting is now done to validate the IPCC models, but the validation suite includes no “null model” comparison or “tuning stability” testing (as I recommended here). An alternative assessment method, proposed by McKitrick and Christy in 2018, is to use a metric that is:
- seen clearly in all models
- present only when extra greenhouse gas is present in model
- not used in tuning any of the models
- derived using measured data from multiple, independent sources
Their proposed metric is the “tropical hot spot” – an area approximately 30-40k feet above the tropics, that in all the (CMIP5 and CMIP6) models shows a striking warming (graphs all taken from McKitrick and Christy’s 2018 paper or Christy’s 2021 lecture).
Furthermore, the warming of this region (in the models) should have already happened (and is measurable by both balloons and satellites). For their 2018 paper, they ran all the models from the CMIP5 set, to produce these results:
Note that only one model in this suite (the Russian INM-CM4) has a metric value remotely in the ballpark of the actual values.
This experiment was recently run again with the newer CMIP6 model set, which produced these results:
The newer models were slightly closer to reality over the region covered by actual observation (except that the revised INM-CM4-8 and INM-CM5 models joined the main group), but when they were continued beyond that point, the “tropical hot spot” warming accelerated. Furthermore, the detrended variance of the newer models on this metric is huge, compared to reality – clearly the feedback is just wrong on these models (leading to a lack of temporal stability). Here’s another view of the individual model trends:
To be fair, Christy’s approach has been criticized by saying, “yes, but we live at sea level and the models reproduce temperature there SO much better.” This is of course true, but irrelevant – the surface temperature time series is used in tuning the models, so of course they’re going to be better on that metric; that’s the whole point of the tuning! The key advantage of this metric is that it is NOT used in tuning: since reality doesn’t get reproduced by the models, clearly they have the physics wrong at some point! Somehow, at least over the tropics, the real climate system has a much more effective cooling component than the models predict. Will this affect ground temperatures further into the future? Who knows? Until we get the models right on the physics, they can’t tell us!
And then there’s the effect of clouds: described here (and summarized here without paywall). Clouds are always a particular problem in climate modeling – depending on exact conditions, they both retain and reflect heat, so they contribute to both warming and cooling. Modelers concluded some years ago that current ways of incorporating clouds into their models are just incorrect, so they started developing much more sophisticated models, incorporated into NCAR‘s CESM2 modeling toolkit. The new approach led to much higher climate sensitivity than even the previous alarming ones. Andrew Gettelman at NCAR: “The old way is just wrong, we know that. I think our higher sensitivity is wrong too. It’s probably a consequence of other things we did by making clouds better and more realistic. You solve one problem and create another.”
My point here is not to slam the folks at NCAR or anyone else in climate modeling, but just to point out how immature a scientific discipline climatology is. Clouds are fundamental to the climate regulatory system! And the people doing cloud modeling just don’t believe that we’ve gotten them right (and admit this).
The massive uncertainties present in all these models are spelled out in the literature clearly, but somehow those uncertainties never make it into the Summary for Policy Makers, which is completely confident that the worst case will happen (and that we won’t be able to adapt to the projected changes, so we must prevent them at any cost – not just “climate change” but Catastrophic Anthropogenic Global Warming).
Misuse of models
The misuse of models extends beyond asserting an unmerited certainty about their projections – we also see inappropriate use of the model outputs in other ways. I will mention a couple of these: use of “ensemble average” of models, and what Judith Curry refers to as “spinning the model” (model outputs are presented in ways that make them appear more closely aligned with actual temperatures than they are).
First, ensemble averages: Andy May refers to this as “model democracy.” The notion is that all models in a CMIP set are seen as equally “good,” so that we can average over the whole set and obtain something useful. You’ll often see it said that this allows the IPCC to separate “natural variability” (or “noise”) from “model uncertainty.” (see, IPCC AR6, pp. 4.21-24). The models (after some initial filtering) are assumed to be independent enough that any trends that survive in an ensemble mean are likely to be “correct” trends.
In fact, though, one would only expect this to be true if the ensemble in question comprised a random sample of models from some overall “model population” – even then, the mean would only give you some idea of the population mean, which has nothing a priori to do with actual data. The only reason to believe the ensemble average approximates “ground truth” is if you believe that the models in the population are actually independent, unbiased estimates of the true climate; that is, if you believe that the modelers as a group have included all appropriate variables, in appropriate ways so as to sufficiently approximate reality. More points in the sample (even if they are independent from each other) only get you closer to the model population mean, not reality.
Furthermore, the models aren’t independent and unbiased – many of them use the same assumptions about clouds, ocean, atmosphere, land surface and sea ice. Some models were apparently removed from the ensemble mean in AR6 as being too close to another model, but this really underscores the problem: there is a lot of model “groupthink” going on here. Some of that is inevitable (after all, they are all modeling the same climate!) but over time the model outputs become more and more similar. Their projections get closer to each other (though year-on-year variability within each model is often increasing), but are they getting closer to ground truth (see above with special attention to the fate of the lone dissenter from CMIP5)?
Even more striking is the decision of the IPCC to describe the 5-95% model range as the likely climate range. They say this very clearly in the paper, but not so much in the Summary for Policy Makers. This range is certainly a likely range for other models produced at the same time by the same researchers, but it’s a real leap of faith to say that this is the likely range for the actual climate!
Proposals have been made to score models on performance, and weight the mean based on the scores, but so far there is no agreement on precisely how to do this.
Suppose that, in AR7, say, researchers were to take one particular model and submit 30 minor variants of it to the suite. The ensemble spread will shrink and the averages will look a lot better, but they’ll only be a lot better if that particular model is a particularly good one!
Whenever I think about ensemble averages of a non-random sample like this, I’m reminded of a story that an old-school rocket scientist told me about an experience he had decades ago: he and his group were designing a missile guidance system, with certain specs, and a status meeting was held with lots of corporate brass, lawyers, accountants, military types, etc. In this meeting, the question came up as to what the time constant of the homing and guidance system should be, given the specs. The decision was somehow made to go around the room and ask everyone present what they thought it should be, then average the results! Fortunately, my interlocutor was the only one writing down the guesses – he thanked them all, then went back to his group and they did the hard work of calculating the correct time constant from requirements, ignoring the “democratic” result.
As to Curry’s “spinning the model” – her discussion (again, largely here and here) focuses on the fact that hindcasting with model climatology produces models that are significantly different from observed numbers (though often with correct trends). These are reported in aggregate, then, by subtracting out some reference climatology and reporting the “anomalies.” If you don’t do this, you get graphs like this (taken from the MPI modeling group’s paper here):
Note that the model lines (gray) to the left of the present differ substantially from historical data (by 1-2° C). As a side note (which was the point of the paper), this is a real problem, since a number of climate system features (ice, cloud formation) depend very tightly on the absolute temperature.
At any rate, by choosing the “reference climatology” carefully, you can make all sorts of things appear in your graphs. Consider this graph:
This shows the HadCRUT4 temperature series superimposed on the CMIP5 model data, normalized to some reference climatology – note that the width of the ensemble is nowhere near as broad as the previous graph (that is the point of the normalization) AND that the observations are coming perilously close to being outside the 5-95% model curves. An early draft of the AR5 IPCC Summary for Policy Makers report contained the following figure (graphs taken from Curry’s post here):
showing much the same situation. However, by the time the report was published, this figure had been replaced by this one:
Note that here the model ensemble has moved down considerably (by changing the reference point – note that the MPI figure was labelled as “temperature” and the IPCC figures are labelled “temperature anomaly”). Now, magically, the current temperatures are solidly inside the ensemble envelope!
In addition, the (correct) comment that was included in the final draft (yes, you read that right: the final draft) that “Models do not generally reproduce the observed reduction in surface warming trend over the last 10 –15 years,” was removed before publication.
Comments about this:
- Ross McKitrick – “Playing with the starting value only determines whether the models and observations will appear to agree best in the early, middle or late portion of the graph. It doesn’t affect the discrepancy of trends, which is the main issue here. The trend discrepancy was quite visible in the 2nd draft Figure 1.4. All they have succeeded in doing with the revised figure is obscuring it.“
- Curry – “There may be nothing technically wrong with Figure 1.4, although it will mislead the public … to infer that climate models are better than we thought, especially with misleading accompanying text in the Report.“
Deliberate obfuscation to go along with the politically supported narrative is NOT a good look, folks.
Lying with statistics
The classic How to Lie with Statistics is a wonderful little book that is essentially a cautionary tale in how to avoid being lied to with statistics. One of the standard tricks is to carefully select the range of a time series to show one thing, when a broader view shows something else.
There is a graphic in the 2018 National Climate Assessment (the summary infographic on page 38) that is absolutely laced with this sort of malfeasance. I’m normally reluctant to ascribe ulterior motive to things like this, but truly it’s hard to see any other reason for the choices made in producing the graphic. I have no idea who actually made the choices, but, at the very least, someone at USGCRP should have noticed the bias in this and pulled/reworked it. Most of these problems were noticed by Tony Heller, but even if you don’t like him (or his tone, or…), it’s hard to refute the evidence he presents. Here is the offending graphic:
First, let me say that the graphic includes a few facts that are presented as “no clear trend” (like drought conditions) which gives an appearance of even-handedness, while being biased in many other ways. I’ll take the subgraphs in alphabetical order:
a: this map represents the difference between annual average temperature 1986-2016 and 1901-1960, showing a geographically widespread increase in average temperature between those two timeframes. This one is likely an honest (though misleading) graph, as information like this is often promulgated. Here are the more relevant time series/maps (from the previous year 2017 CSSR produced by the same group):
These make it clear that the “increasing average temperature” shown in NCA subgraph “a” is largely a matter of warming the low temperatures, rather than increasing the high temperatures. This has often been observed, but gets no traction in media reporting: “moderating cold weather” isn’t nearly as alarming as “setting the planet on fire.”
b: this is one of the worst offenders – here it is, expanded in size:
Wow! How are we not already cooked? The answer lies in these graphs:
Here are a couple of useful superimpositions:
One observation that one might make is that these graphs represent different metrics – the EPA and CSSR graphs represent a “Heat Wave Magnitude” or “Heat Wave Index” while the NCA graph represents some “hot weather” metric measured in days. Here, though, is the NOAA time series for number of days above 90°:
So, yes, the trend is slightly upward since 1960, but 1960 was a historical low point. To choose that as the origin of the graph is to deliberately obscure the long-term trend of the time series.
c,d,e: these are unexceptionable graphs, I suppose – a slight upward trend in the percentage of land area experiencing occasional intense, single-day rainfall events, a sizeable decrease in western snowpack from 1955 to the present (I don’t have access to more of the data, so I have no idea whether these represent anything significant or not), and a graph showing no observable trend in the average Palmer Drought Index over the last century or so (I believe that the Ketch-Byrum Drought Index shows a similar lack of trend).
f: this is one of the now-standard graphs showing decline in Arctic sea ice – here is a larger version:
For this one, I will point out that, while it is now routinely asserted that this is the entire dataset for satellite measurement of Arctic sea ice, the 1990 IPCC report said otherwise:
Here is the longer time series, again extracted and annotated by Tony Heller
Reads a little differently when the earlier data is added back in, doesn’t it? There were even earlier (non-satellite) assessments in DOE publications with a time series dating back to 1925. Here is that data:
Here is an overlay of that data with the satellite data (overlay and annotation by Tony Heller):
Again, it reads much differently when the origin isn’t set near the high point in 1979.
g: The sea level graph is, at least, more honest than most, showing a pretty much consistent rate of sea level rise from 1920 to the present (note that this has essentially no correlation with CO2 since those levels didn’t begin to rise until around 1975). But, the sea level record goes back a good deal farther than that. This is a typical graph:
Here’s another:
You’ll see the assertion made that this is accelerating since 1990, but I don’t see it in any of these graphs. I believe that to get that result, you have to involve the change from tide gauges to satellites (which have particular problems with measuring sea level – I’m sure we’ll get those sorted out at some point, but I see no reason to abandon the tide gauge data series now, while the satellite data is so uncertain). Our (long-held) best guess about the reason for the slow, steady sea level rise (for as long as we’ve been measuring it) is a continuing effect of the end of the Ice Age.
h,i,j: h and i are short time series (marine species distribution and Hawaiian ocean acidity). Frankly, I have no idea whether these have any statistical significance in an appropriate context or not. I also mention that the y-axis origin on i is set so as to exaggerate the trend (which runs from roughly 8.12 to 8.08 – is that amount of change significant in any way? They don’t say). j is an unsurprising (and positive) result of the fact that the coldest weather is warming (see “a“). Beginning and end of growing season are largely determined by the evening low temperatures dropping below some point (last frost/first frost), so warming the lows will necessarily increase the length of the growing season.
k: Another in the crowd of cherry-picked x-axis origins. Here’s a larger version:
and here is a longer look at the same time series from the USDA:
And here’s an overlay of the two, again produced and annotated by Tony Heller:
l: I have no reason to believe this graph is inaccurate or misleading (it purports to show that the number of days with averages above 65° is increasing, and the number of days with averages below 65° is decreasing), though the “cooling” result is a bit difficult to square with other results I am aware of, which indicate that particularly hot days (with highs above the thresholds of either 90° or 95° – see here for example) show no significant recent trends. It is, I suppose, possible that days with averages above 65° but highs below 89° could be increasing but NOT the hotter days. If that is the case, I don’t find it particularly alarming, given that insulation often eliminates the need for cooling on a day as described – certainly in Texas we wouldn’t run our AC on such a day (as I write this, my personal weather station indicates that yesterday had an average of 70.2° and a high of 86.9° – our AC is off and we were quite comfortable).
So, to sum up: 12 subgraphs, one (temperature) is misleading (but accurate), one (drought) shows no trend, four (precipitation, snowpack, marine species, acidity) are presented without sufficient statistical context to ascertain significance, one (growing season) is both positive and unsurprising given earlier information, one (cooling days) is accurate but not alarming when harmonized with other readily available data, one (sea level) is consistent with long-term trends going back to the 19th century, and three (heat waves, sea ice, wildfires) have their origin cherry-picked so as to inflate the significance of recent data markedly.
When appropriate context is supplied for all 12 subgraphs, none of them are particularly “catastrophic” or alarming, and yet 10 of the 12 arrows point in the direction that supports the “received wisdom” narrative. I think it’s pretty safe to say that communication of truth was NOT the point of this infographic.
Summary
As I said at the outset of this essay, my intent is not to claim that I am certain about any particular proposition about CAGW. Rather, I want to point out that “scientism” – specifically, “certainty vs. confidence,” over-reliance on models, and suppression of evidence opposed to the “ruling class” narrative – is rampant in the promulgation of the “received wisdom” position (for more evidence and discussion of this, see here, here and here; also see Judith Curry’s story about why she left academia).
As far as actual climate science policy goes, absent a huge increase in the confidence associated with predictions of doom (and the mechanisms for such doom), any reasonable cost/benefit analysis would suggest that adaptation, not mitigation is the appropriate response to whatever is happening.
Climate science is a science in its infancy; I have every confidence that it will get better, but in the meantime it will necessarily get a LOT wrong. It is irrational to destroy society and attempt to build it anew on the say-so of an infant scientific discipline!
Progress toward eradicating poverty based on accessible and affordable energy (which is carbon today) is continuing … an undeniable force: no one wants to be poor!
–John Christy (video: my transcript)
Postscript: after this essay was originally written, William Happer and Richard Lindzen testified before the SEC (of all places!) against a proposed rule forcing businesses to disclose their “climate-related business risks,” assuming in the process what I described in the first paragraph as the “received wisdom” position on the climate (that fossil fuel use at current levels poses an existential threat to humanity). Their testimony, available here, touches on many of the themes mentioned in this essay. I wholeheartedly recommend reading it carefully.