John Christy, Climate Models, and Long-term Tropospheric Warming

The outline for this post is as follows:

1. Introduction
2. Summary of the Objections
3. Elaboration on the Objections
4. References

If you want the "tl;dr" for this post, then I suggest reading sections 1 and 2. Alternatively, if you are familiar with John Christy's claims on tropospheric temperature trends, then simply skip ahead to section 2.

Each numbered point in section 2 corresponds with a numbered section of section 3. So there is no need to read this entire post; instead, you can see which numbered point you find interesting in section 2, and then for further details you can skip to the corresponding numbered portion in section 3.

This is the "main version" of this post, which means that this post lacks of most my references and citations. If you would like a more comprehensive version with all the references and citations, then please go to the "+References" version of this post.

References are cited as follows: "[#]", with "#" corresponding to the reference number given in the References section at the end of this post.

1.  Introduction

Earth's atmosphere contains multiple layers. The layer closest to the Earth's surface air is known as the troposphere. Above the troposphere is the stratosphere. Climate models and basic physical theory predict how carbon dioxide (CO2) will affect temperature trends in the troposphere and stratosphere. I addressed some of these predictions in my two part series on "John Christy and the Tropical Tropospheric Hot Spot" (part 1 and part 2). I also critiqued some of the claims made by the climate scientist John Christy. In this post I will critique some of Christy's other claims on tropospheric temperature changes and model-based atmospheric predictions.

Climate models predict that CO2 will cause more warming in the mid-troposphere than in the lower troposphere and surface, especially in the tropics (for more on this, see part 1 and part 2 of my series on "John Christy and the Tropical Tropspheric Hot Spot"). Climate models also predict how much tropospheric warming should result from a given increase in atmospheric CO2 levels in combination with other factors. These model-based predictions over-estimated the observed rate of tropospheric warming, resulting in a discrepancy between observations vs. climate models. I will refer to this as the model-observations discrepancy.

John Christy often presents this discrepancy in the form of graphs. Many denialists rely on these graphs in their discussions of anthropogenic climate change (ACC).
(Note: "Denialist" is a fairly well-defined term in science and philosophy, with AIDS denialists being one of the standard examples of a denialist group)
You have likely seen one of these graphs if you recently read an ACC denialist book, visited an ACC denialist blog, or viewed politically conservative media outlet discussing ACC. Here is one of Christy's graphs depicting the model-observations discrepancy [1]:

Figure 1: "TMT" is mid-tropospheric temperature [1].

Christy uses these graphs and the discrepancy to argue that:

1. Climate models over-estimate the real global warming trend, and thus climate models are deeply flawed.
2. Climate models likely exaggerate the warming impact of CO2 and under-estimate the ability of clouds to limit global warming. This means that the models are too sensitive to greenhouse gases and too insensitive to negative feedback from clouds.
3. There has been no tropospheric warming for 18 years.
4. Non-anthropogenic, as opposed to anthropogenic, factors best explain post-1970s atmospheric temperature trends.
5. Politicians should reconsider environmental regulations with regards to climate change.

Christy's arguments hold up poorly when compared to evidence published in peer-reviewed journals. So let's go over some of the issues with Christy's arguments.

2. Summary of the Objections

Christy's data analysis is not peer-reviewed; competent peer review may have remedied some of the issues with Christy's analysis, as was the case with Christy's non-peer-reviewed claims regarding the hot spot (for more on this, see part 1 and part 2 of my series on "John Christy and the Tropical Tropospheric Hot Spot"). So it is quite telling that Christy did not first get his analysis peer reviewed, before citing his analysis to Congress. Christy's analysis seems less about providing credible research to informed, skeptical scientists, and more about politically influencing uninformed, credulous politicians, while providing debate material for denialist sources. Christy's avoidance of peer review has even led to exasperation among some scientists.

Below is a brief summary of some of the shortcomings of Christy's argument:


1) Christy does not use data that is more appropriately homogenized, nor is his congressional testimony adequately forthcoming when it comes to issues regarding homogenization. Appropriately homogenized data likely reduces the model-observations discrepancy.

2) Christy leaves out one satellite data-set showing tropical tropospheric warming. Including this data would likely reduce the model-observations discrepancy. Christy also under-estimates the tropospheric warming trend presented in other data sources.

3) Christy suggests that the model-observations discrepancy is most likely due to climate models being too sensitive to CO2. However, there are multiple other plausible explanations for the model-observations discrepancy, such as errors in data homogenization.

4) Christy cherry-picks short time periods, in order to argue that there has been no tropospheric warming for 18 years. This cherry-picking is unjustified because such short time periods are particular prone to statistical noise, shifts to temperature monitoring equipment, and the effects of short-term control variables such as ENSO. These short-term trends do not undermine model-based predictions of underlying, long-term trends. Even given Christy's cherry-picking, there still has been tropospheric warming over the past 18 years.

5) Christy falsely asserts that non-anthropogenic factors better explain recent atmospheric temperature trends. Instead, anthropogenic factors better explain the radiosonde and satellite atmospheric temperature record.

6) Christy's graphs are very misleading and seemingly designed to exaggerate the difference between the models and the observations.

7) Christy claims that the models do not accurately represent cloud responses, even though Christy does not adequately address observations that support model-based predictions regarding cloud behavior.


Let's examine each of these objections in turn.

3. Elaboration on the Objections

3.1 Paying inadequate attention to homogenized data

Homogenization is a process for correcting heterogeneities, which are factors that artificially skew temperature records. For example, suppose a temperature recording station changes to new equipment for recording temperature. This new equipment may have a cold bias relative to the older equipment, such that the new equipment will record a temperature 0.05K lower than the older equipment, even if the older equipment and new equipment are placed in identical temperatures conditions. So if the temperature station switched from the older equipment to the new equipment, this will cause an heterogeneity of about -0.05K in the temperature trend. Homogenization would correct this heterogeneity, removing the artificial cold bias from the data.

For atmospheric temperature records, scientists homogenize weather balloon (radiosonde) and satellite temperature records, correcting for factors such as the decay/drift in the orbits of satellites and stratospheric cooling contaminating measurements of tropospheric warming. At this point, many ACC denialists make claims such as:

Objection 1: Homogenization is part of the climate science conspiracy (or at least a product of scientists' implicit bias), since homogenization lets climate scientists collude together to make the data show whatever the scientists want. This allows the scientists to maintain their alarmism, the status quo, and the credibility of their climate models. Furthermore, politicians use climate science research to justify regulation and other policies. This creates a further political motive for scientists to be biased in their homogenization. So we should not trust the data homogenization and other climate science research, given the political environment surrounding of climate science research. Homogenization is sometimes just scientists trying to "massage" the data.

I will address this objection at length, because it represents one of the main layperson objections to homogenization and climate science research in general. You may want to skip the following bullet points if you do not find objection 1 persuasive.

Objection 1 fails for many reasons. For instance (the counter-objections below apply to the "political conspiracy theory" version of objection 1; however the counter-objections can be modified to apply to the "implicit bias / massaging the data" version of objection 1 as well):

• Numerous scientists have pointed out model-observation discrepancies and imperfect homogenization, without those scientists losing their jobs or funding. Christy is among these scientists. Therefore, climate scientists can maintain their jobs/funding even if they do not support the status quo on homogenization and mainstream climate science. 

• Contrarians such as John Christy, Roy Spencer, Richard Lindzen, Roger Pielke Sr., Judith Curry, Willie Soon, Craig Idso, David Legates, and Anthony Watts benefited from government funding, even though these contrarians challenged mainstream climate science. For example, Christy received funding for his poorly homogenized satellite-based temperature analysis. So objection 1 fails when it implies that government funding of research would necessarily push climate scientists into misrepresenting evidence of the sake of the government's political ideology.

• The scientific research that gets the most attention, prestigious prizes such as the Nobel Prize, etc. is research that overturns expectations or breaks new ground, as opposed to research that simply supports the consensus position. So even if scientists were just self-interested attention-seekers, most scientists likely would not use deceptive homogenization to support the consensus position.

• Homogenization techniques are often validated. And even non-scientists / non-experts can check the accuracy of these homogenization techniques, including by accessing the raw data.

• Different, independent research groups apply different homogenization techniques to the same data. These research groups and methods serve as a check on one another, helping remedy the mistakes involved in any one research group or method. This is one reason why scientists use different approaches/methods to test a conclusion: the strengths of one method can compensate for the weaknesses in another method, so that one knows that the results are not just due to the flaws of one particular method. This leads to consilient/convergent lines of evidence supporting a conclusion. 

• A scientist's reputation could be tarnished if another research group caught the scientist using a deeply flawed homogenization technique. And if a scientist's homogenization technique was found to be clearly fraudulent or deceptive, then that scientist's research career would be over. This provides a strong incentive for scientists not to engage in deceptive homogenization.

• The presence of multiple homogenization research groups makes a conspiracy unlikely, since conspiracies become more difficult to conceal as more people get involved in the conspiracy. This is especially the case when dealing with different research groups in different countries with different political climates. If homogenization results agree between such diverse groups, then this should increase one's confidence that the results are not merely a product of political pressure. Similar reasoning applies to another a variant of objection 1; this variant states that anthropogenic climate change is a politically-motivated myth, invented by conspirators for the purpose of justifying government regulation and globalization. Such a conspiracy would need to involve a ludicrous amount of people because there is a large scientific consensus on anthropogenic climate change. This consensus extends outside of the climate science community, as expected, since non-climatologists can document how anthropogenic climate change affects marine environments, the economy, the health of various organisms, and so on. Furthermore, evidence of CO2-induced climate change comes from diverse fields, including paleoclimatology, atmospheric physics, oceanography, and the study CO2-rich atmospheres on other planets. So the conspiracy would need to involve the scientists working in these fields as well. The conspiracy would also need to date back to at least the late 1800s, since the idea of CO2-induced global warming dates back to at least that point. Such a conspiracy is unlikely to exist, given the enormous amount of proposed conspirators spread across more than a century.

• There have been previous instances in paleoclimatology, atmospheric physics, etc. in which a model-observations discrepancy was due to incorrect data analysis, as opposed to an error in the models (as in the case of Christy's flawed satellite data analysis, a topic I discuss later in this section). Thus it would not be shocking if the tropospheric model-observations discrepancy was largely due to flawed data homogenization. So if objection 1 states that homogenization is part of a scientific conspiracy, then this conspiracy includes paleoclimatologists, atmospheric scientists, and other scientists who used corrected data analysis methods in order to resolve a model-observations discrepancy. And it is unlikely that there is such a large conspiracy of diverse scientists across multiple decades, for the reasons I previously explained.

• Homogenization does not always increase temperature trends nor does it always reduce model-observations discrepancies. In fact, models often under-estimate the trends observed in homogenized data, as I discussed in section 3.7 of "John Christy and the Tropical Tropospheric Hot Spot, Part 1". But suppose that, for the sake of argument, homogenization tended to reduce model-observations discrepancies. This is what one would predict if the models' average prediction was accurate. Yet objection 1 draws a different conclusion: the reduced discrepancy is evidence that the homogenization is flawed, and likely a result of a conspiracy (or bias) on the part of the homogenization scientists. The objector can only draw this conclusion if they assume the models' accuracy is not a good explanation for the resolved discrepancy. Thus objection 1 begs the question by assuming from the outset that the models' accuracy is not a good explanation. This is akin to a committed flat-earther claiming that homogenized data showing a round-Earth, is actually evidence of a conspiracy of scientists (or scientific bias) massaging data in order to support a round-Earth model over a flat-Earth model. In doing so, the flat-earther would beg the question by assuming that the accuracy of round-Earth models is not a good explanation for the convergence of homogenized data, and for why homogenization of sometimes discordant data brings that data into line with round-Earth models. Objection 1 similarly begs the question against the accuracy of climate models. The objector offers an unfalsifiable, "heads, I win; tails, you lose" situation: According to the objector's reasoning, if there is a significant model-observations discrepancy, then this shows that the models' predictions are wrong and that the objector's position is correct. If there is no significant discrepancy (due to homogenization, for example), then this shows scientists' conspiracy/bias and that the models' predictions are likely still wrong.

• Many denialists who use objection 1 render the objection unfalsifiable, since the denialists will resort to almost any claim in order to defend objection 1. For instance, denialists will often move the goalposts in order to avoid accepting evidence against their position. In the denialist's mind: Contrary evidence against their conspiracy theory is actually evidence for their conspiracy theory, since the conspirators must have fabricated the contrary evidence. Absence of evidence for the conspiracy theory is also evidence for the conspiracy theory, since the conspirators must have suppressed the evidence. Baseless claims count as support for the conspiracy theory, even if the baseless claims come from uninformed people who contradict scientific evidence. And objecting to the conspiracy theory means you are either part of the conspiracy or you are among the ignorant sheeple duped by the conspiracy. In this way, the denialist offers an impossible burden of proof that no scientific research could ever meet, and the denialist immunizes their position against falsification.

• Objection 1 is self-serving and self-perpetuating. The self-perpetuation cycle is as follows: ideologues help manufacture controversy about ideologically-inconvenient evidence. Then the ideologues appeal to the controversy as a rationale distrusting the ideologically-inconvenient evidence. This appeal maintains the lack of public agreement and thus the controversy, then ideologues again appeal to the controversy, and so on in a self-perpetuating cycle. To elaborate on this point further: laypeople's ideological values can shape their beliefs regarding scientific matters. So, for instance, many (but not all) religious people may object to climate science on the grounds that humans cannot detrimentally affect God's global climate or climate science may be associated with nature worship. This leads to a negative correlation between certain religious positions vs. environmental concern and accepting various scientific claims regarding climatology. Similarly, many (but not all) political conservatives might distrust climate science research because they think this research could be used to support regulation and environmentalism. This produces a negative correlation in the US between political conservatism vs. trusting climate scientists and accepting various scientific claims regarding climatology. These two correlations result in an ideological controversy with both economic and religious/social dimensions. Industry-funded think tanks, the media, and other denialists further maintain the controversy by manufacturing more false doubt about climate science. And this is where many conservatives apply objection 1: these conservatives appeal to the ideological controversy as a rationale for distrusting climate science research, even though these same conservatives do not accept much mainstream climate science research and thus help perpetuate the very controversy they appeal to. This cycle represents another way in which denialists can unjustifiably immunize their position against contrary evidence, since the cycle can be used to discredit any research that ideologues do not already accept. A similar cycle can be used to shield virtually any ideologically-motivated position, no matter how implausible that position may be.

• Objection 1 amounts to special pleading, since one could raise a similar objection to scientific research the objector accepts and relies upon. For example, one could claim that politicians use HIV/AIDS research to justify foreign aid to poorer nations, compulsory HIV testing of sexual offenders, and punishing HIV-positive people who do not inform their sexual partners of their HIV-positive status. Given this political environment, one could then reason that one should distrust any research showing that HIV causes AIDS, in keeping with the logic of objection 1. Similarly, a fiscally conservative tobacco executive could argue that since politicians use biomedical research to support cigarette taxes, regulating the tobacco industry, and banning smoking in certain public places, one should distrust any biomedical research showing a smoking-cancer link. This conservative could then claim that physician-scientists fabricate the alarmist smoking-cancer link so that more smokers make appointments to meet their physician, thereby providing more funding for these physician-scientists. And suppose homogenized estimates of Earth's circumference kept converging on a circumference estimate predicted by round-Earth models. A committed flat-earther could claim this as evidence of a conspiracy of scientists (or scientific bias) massaging data in order to support a round-Earth model over a flat-Earth model. A committed denialist could raise similar objections against almost any scientific research. This includes research into inventions many denialists rely on, such as computers, the Internet, and medicines. To put the issue another way: one can accept data homogenization on the same grounds as one accepts other scientific research. These grounds include validation, peer review, and checks by other independent research groups. So unless the objector wishes to engage in special pleading, the objector needs to explain why objection 1 applies to climate science and homogenization, but not to other scientific research. The objector therefore bears the burden of proof. They cannot meet this burden by inventing baseless political narratives, since those narratives can be invented for virtually any scientific research, as previously shown.

• Objection 1 is self-undermining. To see this, first note that one could propose a political narrative about the proponent of objection 1. For example, one might claim that the objector offered objection 1 in order to evade scientific research that might be used to support regulation. So based on the logic of objection 1, this implies that one should be skeptical of both objection 1 and proponents of objection 1. Thus objection 1's reasoning is self-undermining and internally inconsistent.

To put the point more bluntly: the denialists' objection is a paranoid, baseless conspiracy theory. ACC denialists, AIDS denialists, and other denialists often offer similar conspiracy theories in order to discredit evidence presented by the mainstream scientific community. These self-serving conspiracy theories allow denialists to engage in unjustifiable cherry-picking of what evidence the denialists will accept.

The bad faith and special pleading / double-standard of many denialist objectors becomes evident when they object to homogenization in principle, even though said objectors happily accept homogenized data analyses that fit the objectors' biases and pre-conceived notions. Many denialists, for instance, accepted Remote Sensing System's (RSS) satellite data analysis and Christy's University of Alabama in Huntsville (UAH) satellite data analysis, even though both analyses use homogenization and the non-expert public is unable to easily access the non-homogenized raw data for these satellite-based analyses (unlike the raw data that non-scientists and non-experts can access for surface temperature trend analyses, as discussed above). Denialists latched onto RSS' analysis because faulty homogenization caused the RSS analysis to show very little tropospheric warming.

Christy's UAH colleague Roy Spencer warned denialists that RSS' tropospheric warming trend was artificially low. Spencer, however, did try to facilitate denialists' self-serving double-standard by suggesting that denialists use the RSS analysis, even though Spencer thought the RSS analysis was worse than his UAH analysis:

"But, until the discrepancy [between the RSS analysis vs. the UAH analysis] is resolved to everyone’s satisfaction, those of you who REALLY REALLY need the global temperature record to show as little warming as possible might want to consider jumping ship, and switch from the UAH to RSS dataset [10]."

Amazing. RSS then corrected their homogenization, increasing their reported tropospheric warming trend. This caused many denialists to launch into baseless, self-serving conspiracy theories about RSS, despite Spencer's aforementioned warning.

RSS was not the only group to generate a faulty conclusion due to erroneous homogenization; over two decades earlier, Christy and Spencer claimed that their UAH satellite data analysis showed virtually no tropospheric warming. Subsequent research pointed out errors in the UAH data analysis. Correction of these errors with improved homogenization, along with over two decades of subsequent global warming, resulted in a tropospheric warming trend in the UAH data. Yet this did not cause many denialists to launch into conspiracy theories about Spencer and Christy, in stark contrast to the denialists' response to RSS' tropospheric warming trend. This illustrates the self-serving double-standard implicit in many denialists' objections to homogenization.

As Christy and Spencer note:

"Of course, all data require adjustments [...] [9, page 20]."

So Christy should recognize the importance of appropriate homogenization, since Christy employs homogenization in his own work with UAH and other scientists have noted deficiencies in Christy's UAH homogenization process (deficiencies that cause UAH's atmospheric temperature trends to greatly differ from the trends revealed by other research groups). Yet Christy uses inadequately homogenized data in his discussion of the model-observations discrepancy. Christy's models-observation graphs, for instance, present tropical radiosonde temperature records, even though these records may contain a cold bias. Christy and Spencer previously discussed this cold bias and the cold bias was even pointed out in a report that Christy co-authored. If scientists correct this spurious cooling via homogenization, then this would likely reduce Christy's model-observation discrepancy.

Other homogenization issues include:

• Though RSS updated the homogenization for their satellite data analysis, some of Christy's graphs were made before this update. These graphs therefore use a pre-update RSS analysis that is spuriously cool. RSS used homogenization to correct this spurious cooling in both the mid- to upper troposphere and the lower troposphere. It remains unclear whether Christy's newer graphs incorporate the updated RSS analysis. If Christy's new graphs do not incorporate these updates, then Christy's graphs show spuriously cool satellite tropospheric data and thus the graphs exaggerate the model-observations discrepancy.

• Other scientists have critiqued Christy's UAH satellite data homogenization and Christy's UAH analysis conflicts with his own position on tropospheric warming caused by the Sun (for more on this latter point, see sections 3.1 and 3.2 of "John Christy and the Tropical Tropospheric Hot Spot, Part 1"). Christy's UAH analysis also shows much less tropical tropospheric warming than do three other independent, validated analyses. This suggests that UAH's tropical temperature trend contains a spurious tropical tropospheric cold bias due to faulty homogenization. Correcting the UAH cold bias would likely lessen Christy's model-observations discrepancy. This also illustrates how other independent research groups can serve as a check on the faulty homogenization done by another research group, as I previously discussed.

• Christy's graphs incorporate satellite tropospheric data that is spuriously cool due to contributions from stratospheric cooling (see section 3.4 of "John Christy and the Tropical Tropospheric Hot Spot, Part 1" for more evidence of long-term stratospheric cooling). When scientists used homogenization to correct this spurious cooling, this correction greatly reduced the model-observations discrepancy.

Taken together, these results imply that proper homogenization of radiosonde and satellite records would likely reduce the Christy's model-observations discrepancy. Yet Christy's congressional testimony does not pay adequate attention to how homogenization might reduce the model-observations discrepancy. So Christy (unintentionally or intentionally) misleads his audience about the magnitude of the model-observation discrepancy and about the factors causing the discrepancy.

3.2 Leaving out data showing tropospheric warming

Christy excludes the University of Washington (UW) satellite data analysis, even though this data analysis has been validated, shows tropical tropospheric warming, and is fairly consistent with two other independent satellite data analyses. This exclusion seems rather telling, since the UW team has debated Christy's UAH team in the scientific literature and UW's analysis conflicts with the reduced rate of tropical tropospheric warming trend shown in Christy's UAH analysis. In fact, UAH's tropical tropospheric data analysis is an outlier among the four research groups investigating the satellite data (UAH, RSS, the National Oceanic and Atmospheric Administration Center for Satellite Applications and Research (NOAA/STAR), and UW) [2; 4]. This is illustrated in the following two figures[2; 4]:

Figure 2: HadCRUT4 tropical surface warming trends and tropical mid- to upper tropospheric warming trends (in K per decade) above the land, oceans, and both land and oceans from 1979 - 2012. Tropospheric warming trends are from UW, NOAA, RSS, and UAH satellite data analyses. UW(GCM) and UW use different methods for processing the satellite data. The value in parentheses is the ratio of the tropospheric warming to the surface warming for a given tropospheric temperature trend [2]. The RSS tropospheric warming trend is spuriously low due to an error in homogenization. The RSS team later corrected this error. This resulted in a RSS tropical mid- to upper tropospheric warming trend that is between the NOAA trend and the UW trend, as shown in figure 3B.

Figure 3: (A),(B) 1979 - 2016 mid- to upper tropical tropospheric warming trends predicted by climate models and observed in satellite data analyses. Trends are presented as an average of all the trend values for a given trend length. Trends are not corrected for stratospheric cooling (A) or corrected for stratospheric cooling (B). (C),(D) Ratio between the tropospheric warming trend predicted by the climate models vs. the tropospheric warming trend observed in the satellite data analyses. The dotted lines in (C) show the ratios Christy reported to Congress [3, page 3]. Trend ratios are not corrected for stratospheric cooling (C) or corrected for stratospheric cooling (D) [4].

Christy and Christy's UAH colleague Roy Spencer claim that the greater warming trends from RSS, NOAA, and UW do not match the trends from temperature re-analyses and weather balloon (radiosonde) data, while the re-analyses and radiosonde data do match the lower UAH warming trend. So the re-analyses and radiosonde data support Christy et al. UAH analysis, and argue against the RSS, NOAA, and UW analysis.

(The denialist Anthony Watts credulously cites Spencer's discussion of how the tropical radiosonde record argues against the RSS and NOAA satellite data analyses. This despite the fact that Watts previously objected to the tropical radiosonde record, since, according to Watts' article, radiosonde coverage is sparse in the tropics and satellite analyses offer better coverage. So Watts uses satellites to object to radiosondes, when the radiosondes inconvenience Watts' position. Then Watts uses radiosondes to object to satellites, when the satellites inconvenience Watts' position. And Watts does this without resolving the apparent contradiction between his two cited lines of reasoning. This falls in line with Watts' habit of evading, and not changing his mind in response to, evidence that shows he is wrong (for further examples of Watts' tendency, see section 3.4 of part 2 of my series on "John Christy and the Tropical Tropospheric Hot Spot", and section 3.5 of part 1 of "Christopher Monckton and Projecting Future Global Warming"). Such a tendency is a classic sign of a denialism, in contrast to scientific skeptics who change their mind in response to evidence.)

Spencer and Christy's reply fails for a number of reasons.

First, radiosonde analyses have a well-known cold bias, as pointed out in a report that Christy co-authored. Christy and Spencer have commented on this cold bias before, so they have excuse for not being aware of it. They should also know better for another reason: over a decade ago, Christy and Spencer emphasized how radiosonde analyses fit with UAH's lower warming trend. The RSS team, however, then showed Christy and Spencer that their UAH analysis was too low and needed to be adjusted upwards. Thus Christy and Spencer should aware of the dangers on relying on low, radiosonde-based tropospheric warming trends. Yet Christy and Spencer still make the same error over a decade later. Amazing.

Second, Christy under-estimates the satellite warming trend. For example, Christy claims that the tropical mid-tropospheric warming trend is (in K per decade) 0.160 for the NOAA analysis and 0.137 for the RSS analysis. But based on figure 3B, the NOAA trend is actually ~0.20 and the RSS trend is ~0.18; so both trends are larger than Christy claimed.

Third, the greater warming trends from RSS, NOAA, and UW agree fairly well with the warming trends from temperature re-analyses and radiosonde data, in contrast to the lower UAH trend. To see this, let's first take note of the tropical upper tropospheric warming trends (in K per decade) for the four satellite data analyses in figure 3D:

• RSS  :  ~0.18
• NOAA  :  ~0.20
• UW  :  ~0.16
• UAH  :  ~0.10

So UAH is clearly the outlier, as I previously mentioned. The aforementioned RSS, NOAA, and UW trends are fairly consistent with 5 out of 6 radiosonde trends for 1979-2014 and 2 out of 3 temperature re-analyses for 1979-2012 (in K per decade):

• Five radiosonde analyses each generally have upper tropospheric temperature trends of  :  >0.17 
• Met Office Hadley Centre (HadAT2) radiosonde analysis  :  >0.11
• Modern Era Retrospective-Analysis for Research and Applications (MERRA) re-analysis  :  ~0.2 
• European Centre for Medium-Range Weather Forecasts Interim re-analysis (ERA-I)  :  ~0.2
• National Centers for Environmental Prediction (NCEP-2) re-analysis  :  ~0.1

The low NCEP-2 trend should be taken with a grain of salt, since the NCEP re-analysis has a history of under-estimating tropospheric warming; this may explain why Watts gladly trumpets the NCEP re-analysis. Other re-analyses, such as MERRA and ERA-I, tend to perform better than NCEP when it comes to representing atmospheric phenomena.  So one should consider using another re-analysis instead of NCEP-2, though ERA-I likely under-estimates lower tropospheric warming. But overall, despite the fact that the radiosonde analyses have a well-known cold bias, the radiosonde data and temperature re-analyses fit better with the higher RSS/NOAA/UW trends than the lower UAH trend. So Christy and Spencer are likely wrong when they claim otherwise. Christy also sometimes excludes other data that shows significant tropospheric warming, a tendency shared by many of those who repeat Christy's claims. In any event, including the UW data would likely reduce Christy's model-observations discrepancy.

3.3 Disregarding other plausible explanations of the model-observations discrepancy

Christy initially did not come to a firm conclusion on what caused the model-observations discrepancy. However, Christy now suggests that the model-observations discrepancy is most likely due to models being too sensitive to CO2, such that the models exaggerate the amount of warming caused by CO2. However, there are multiple other plausible explanations for the model-observations discrepancy. These explanations include:

1. Uncertainty in the observations, due to data homogenization and differences between the temperature trends produced by different research groups. I elaborated on this explanation in sections 3.1 and 3.2. This observational uncertainty remain distinct from possible flaws within the climate models.
2. Errors in measurements of human, solar, volcanic, and other factors. Estimates of these factors serve as input for the climate models, and the climate models then use this input (in addition to other information) to project future trends under various scenarios. So errors in this model input would lead to incorrect model projections, even if the climate models themselves were perfect with respect to the physical processes relevant to climate change. 
3. Differences in short-term variability. Certain factors strongly influence shorter-term climate variability, while different factors more strongly influence longer-term climate trends. Shorter-term factors include aerosols and the El Niño phase of El Niño-Southern Oscillation (ENSO). Shorter-term temperature variability is often due to the randomness / stochastic noise that afflicts smaller sample sizes. This randomness tends to have less of an effect on larger sample sizes; I discuss this more in section 3.4. This randomness could skew shorter-term model-based temperature projections, even if the climate models themselves were perfect with respect to the physical processes relevant to climate change.

Scientific evidence shows that all three of these factors contribute to the model-observations discrepancy; this includes evidence from a report that Christy co-authored. Furthermore, scientific evidence rebuts Christy's attempt to attribute the model-observations discrepancy to the models being too sensitive to CO2. So Christy jumped the gun when he claimed that excessive sensitivity to CO2 is the most plausible explanation of the model-observations discrepancy.

3.4 Cherry-picking short time periods

Christy argues that there has been no significant tropospheric warming for 18 years, resulting in a pause/hiatus in warming. US Senator Ted Cruz concurred with Christy's assessment, as have others. In making these claims, both Cruz and Christy illegitimately cherry-pick short time periods and small sample sizes. To see the problems with this, let's examine two scenarios:

• Suppose you flip a fair coin 4 times. I cannot confidently tell you whether or not you will get close to a 2:2 ratio of heads to tails. You might get a ratio of 3:1, 1:3, 4:0, or 0:4 by chance. Chance does not mean magic, since we can (in principle) give a detailed scientific explanation for why each flip came up heads. This explanation would include information about the motion of your hand during the flip, differences in air pressure as the coin moved through the air, how the coin was positioned on your hand before the flip, etc. These are some of the factors influencing our small size of 4 coin flips. Now, suppose you flip that fair coin 4 million times. I can confidently say that you will get close to a 2:2 ratio, since the large sample size makes it very unlikely that you will wander far from a 2:2 ratio by chance. I can predict this 2:2 ratio, since I know that the coin is not strongly loaded to one side of the coin versus the other side of the coin (meaning that the coin is fair), and I know the coin's fairness is the dominant factor explaining heads-to-tails ratios in a large sample size of coin flips. I can say this, even though I do not understand all the factors that affect the heads-to-tail ratio across a smaller sample size of 4 flips. One can apply the same reasoning to a small number of dice rolls vs. a large number of dice rolls.

• Suppose Ted, Chris, Benjamin, Peter, Carlos, Veronica, and Victor live in Canada. Chris is "skeptical" of whether the Earth's axial tilt (relative to the sun) causes Canada's average temperature to increase from mid-winter to mid-summer. To support his "skepticism", Chris cherry-picks a particularly warm week in autumn and a subsequent colder week in autumn. He does not control for the short-term weather pattern that made one autumn week warmer than the subsequent autumn week. Chris then uses this short-term cooling trend to argue that there has been no warming for two weeks and thus there has been a pause/hiatus/stall/stoppage in the effect of Earth's axial tilt on Canada's average temperature. He then claims that this justifies his skepticism about how much Earth's relative axial tilt affects Canadian average temperature. Teddy concurs with Chris' assessment. Benjamin, Peter, and Carlos point out that Chris cherry-picked a small sample size without controlling for short-term weather trends that would affect such a short term trend. They emphasize that multi-month temperature trends are not as biased by these short-term weather patterns. A multi-month temperature increase remains, even after Benjamin and Carlos correct for short-term weather trends. Benjamin and Peter then note that the long-term, multi-month increase in Canada's average temperature is best explained (and predicted) by model's incorporating Earth's axial tilt relative to the Sun. Benjamin and Peter also point out that Chris is engaged in end-point bias, which is a psychological tendency to view a recent short-term fluctuation as debunking a long-term trend. Victor takes Chris' two weeks worth of data, and uses that data to provide insights into how weather causes shorter-term, multi-week temperature variability. Victor does this while still acknowledging how much Earth's relative axial tilt affects multi-month temperature trends. Veronica appreciates what Victor is doing, but she warns Victor that such a small sample size may not lead to statistically significant results; she doubts that the two week trend is statistically significant, or above the threshold for statistical noise.

These scenarios illustrate how certain factors strongly influence smaller sample sizes, while different factors can more strongly influence larger sample sizes. This explains why, for instance, genetic drift has a larger effect on smaller populations as opposed to larger populations. These examples also illustrate how short-term trends can be biased by chance / statistical noise, while larger sample sizes are not as prone to these effects. So let's apply these lessons to Christy's and Cruz's claim.

Christy and Cruz argued for a pause/hiatus in tropospheric warming, based on a short, 18 year time period that began with a very strong El Niño in 1997-1998. During the El Niño phase of ENSO, both the oceans' surface and the troposphere temporarily warm. This means that Christy and Cruz, like Chris and Teddy, cherry-picked a warm start point for their trend and made sure that the start point was biased by a short-term factor.

In contrast to Christy and Cruz, scientists can follow Victor's example; they can justifiably investigate short-term variability and the factors influencing short-term trends, while accepting that there is a long-term global warming trend underlying this short-term variability. So, for instance, the same data that shows long-term global warming and global warming over the past 18 years, can also be used to investigate the short-term climate variability that can temporally dampen the rate of global warming. This point is acknowledged even by researchers skeptical of whether there has been a recent pause/hiatus in global warming (in keeping with Veronica's reasoning), since one would expect variability to temporally augment or mitigate a global warming trend. Such behavior occurs in climate model simulations and in past warming periods. Scientists have commented on past periods of mitigated warming since at least the 1980s.

Despite this respect for variability research, scientists and non-scientists alike follow Benjamin's, Peter's, Carlos', and Veronica's example; they emphasize the dangers of choosing short-term trends that begin with a strong El Niño year like 1998, especially given the 1998 transition in satellite temperature monitoring equipment. They also often examine temperature trends after correcting for short-term cyclical factors, such as ENSO (for more on this, see section 3.5 of "John Christy and the Tropical Tropospheric Hot Spot, Part 1"). So, in keeping with this reasoning, one cherry-picks if one chooses 1997-1998 as start years in an attempt to rebut a long-term warming trend.

Yet Christy and Cruz cherry-pick anyway, while also displaying end-point bias. Ironically, given the very strong El Niño of 2015-2016, tropospheric temperatures increased since 1997-1998. El Niño alone cannot account for this warming trend since the 2015-2016 El Niño was about as strong as (or possibly weaker than) the 1997-1998 El Niño (with the 2015-2016 El Niño having less ocean warming in some regions, and greater ocean warming in other regions, than did the 1997-1998 El Niño), yet 2015-2016 was warmer than 1997-1998. The 1998-2016 warming trend debunks Christy and Cruz's claim of no tropospheric warming for 18 years, despite Christy conveniently choosing a time-frame that minimized the tropospheric warming trend. Take, for example, an average of the RSS, UAH, and NOAA data analyses [5]:

Figure 4: Near global mid- to upper tropospheric temperature averaged for the UAH, NOAA, and RSS analyses [5]. 

Or take the RSS data analysis that many denialists used to show a "pause/hiatus" in global warming (as I discussed in section 3.1) [4; 6]:

Figure 5: RSS near-global lower tropospheric temperature, with a trend-line for 1979 to 2016 [6].

Figure 6: RSS near-global mid- to upper tropospheric temperature, with trend-lines for different 18 year periods [4].

Figures 4, 5, and 6 shows a warming trend from 1998 to 2016. Once again, this warming trend is not wholly attributable to ENSO, since the El Niño of 2015-2016 was about as strong as (or possibly weaker than) the El Niño of 1997-1998 in terms of warming of the ocean surface. This point bears repeating given the number of denialists who follow Roy Spencer's misguided example and claim that El Niño is responsible from the warming between 1998 and 2016. El Niño does not account for the warming trend between 1998 and 2016, since the El Niño of 2015-2016 was about as strong as (or possibly weaker than) the El Niño of 1997-1998. It's as if some non-ENSO factor was also driving temperature up during this time period... (*cough* CO2 *cough*).

The relative magnitudes of the 1997-1998 and 2015-2016 El Niño events are shown in figures 7 and 8:

Figure 7: Multivariate ENSO index (MEI). The red peaks represent El Niño events and the blue troughs represent La Niña events. The relative magnitudes of the peaks and troughs are proportional to the strength of corresponding El Niño and La Niña events [7]. See figure 8 for a representation of some of the uncertainties involved in calculating the magnitude El Niño and La Niña events.

Figure 8: ENSO index (Niño3.4) based on ocean temperature in the east central equatorial Pacific. Niño3.4 is one of several non-cumulative indexes used in generating the non-cumulative MEI shown in figure 7. Red lines indicate the Niño3.4 index values, while the black region represents uncertainty at the 95% confidence level [8].

Spencer should be aware of this point, since he previously relied on figure 7's ENSO index [7] and on a source that used this index. So Spencer really has no excuse for claiming that ENSO caused the 1998 - 2016 warming trend.

And based on figures 4, 5, and 6, if one wanted to investigate the underlying global warming trend, then comparing 1998 to any subsequent year other than an equally strong El Niño year (such as 2016) would be misleading, since ENSO would skew the start point for that short-term trend. But that fact did not stop Christy and Cruz from engaging in that sort of misleading cherry-picking.

Christy's, Cruz's, and Chris' cherry-picking is not particularly novel. For example, one could follow Chris' example by distorting temperature for the middle 1980s to late 1990s in much the same way Christy and Cruz distorted temperature for the late 1990s to middle 2010s. The process would be something like the following:

1. Wait for an anomalously warm year W1 (ex: 1982, 1983, 1987, or 1988; see figures 4 and 5).
2. Do not account for the short-term factors F (ex: ENSO) that contributed to the anomalous warmth of W1.
3. Keep saying that there has been no warmth for X years after W1, and thus there has been a "pause/hiatus". Claim that this pause/hiatus creates a problem for proponents of a CO2-induced, long-term global warming trend. Also make sure that X is a relatively small number, usually shorter than the length of a full cycle of F.
4. Eventually, F will likely cause another anomalously warm year W2 (ex: 1998). If so, then ignore the fact that W2 is warmer than W1.
5. Replace anomalously warm year W1 with anomalously warm year W2.
6. Repeat the process beginning with step 1, replacing W1 with W2.
7. Continue this cycle over and over, with each anomalously warm year that appears.

No doubt future "skeptics" will follow Christy and Cruz's aforementioned process. So prepare to constantly hear about "pauses/hiatus" in global warming, and how these "pauses/hiatuses" supposedly debunk scientific explanations of long-term global warming.

3.5 Misattributing anthropogenic warming to non-anthropogenic causes

Christy incorrectly claims that non-anthropogenic, as opposed to anthropogenic factors, better explain post-1970s atmospheric temperature trends. In reality, anthropogenic factors better explain the radiosonde and satellite atmospheric temperature record, in terms of increased radiative forcing from CO2 within specific infrared wavelengths and Earth's energy balance (the amount of energy the Earth takes up vs. the amount of energy the Earth releases), along with both stratospheric cooling and tropospheric warming. Christy should be aware of this since this was pointed out in a report co-authored by Christy. I discuss this further in "John Christy and the Tropical Tropospheric Hot Spot", sections 3.3 to 3.5 of part 1  and section 3.5 of part 2.

3.6 Misleading graphs that exaggerate the model-observations discrepancy

Christy's graphs of the tropospheric model-observation discrepancy seem designed to exaggerate the difference between the models and the observations. For instance, Christy's graphs:

• Average the temperature data together in a way that obscures observational uncertainty. This uncertainty results from factors such as homogenization and differences the temperature trends produced by different research groups. Displaying observational uncertainty would reveal that this uncertainty accounts for much of the model-observations discrepancy, as I discussed in section 3.3 (see figure 3 for an example of how to display observational uncertainty).

• Display climate model runs, without adequately representing the structural uncertainty in the model average. A scientist could introduce a metric to quantify the uncertainty in model runs, just as scientists use metrics such as standard deviation to quantify uncertainties in averaged data. As with observational uncertainty, displaying model uncertainty would show that this uncertainty explains much of the model-observations discrepancy (for example, see figure 3).

• Place the models and data on a very short baseline that is not commonly used in the scientific literature. This baseline needlessly expands the range of values taken by models and exaggerates the model-observations discrepancy. It can also distort the analysis, if the annual cycle in the short baseline period greatly differs from the annual cycle in the rest of the data.

Other scientists addressed some of these misleading graphical features and incorporated some of the homogenized data I discussed in sections 3.1 and 3.2. This produced better graphs of the tropospheric model-observations discrepancy (see figure 3, for example). These improved graphs show a reduced model-observations discrepancy in comparison to Christy's misleading graphs. For example, compare figure 1 from Christy to figure 9 below:

Figure 9: Mid- to upper tropical tropospheric warming trends predicted by climate models and observed in satellite data analyses. The pink line is the observed tropospheric warming trend, corrected for stratospheric cooling and shown as an average of the UAH, RSS, NOAA, and UW satellite data analyses. The black line shows the average warming trend from an ensemble of climate models, while the gray region shows the range of values taken by different realizations of each model [4].

(In figure 9, 2016 is cooler than 1998. This reflects tropical upper tropospheric temperature, as opposed to global upper tropospheric temperature. When one examines near-global upper tropospheric temperature, 2016 is warmer than 1998, as shown in figure 4. So figure 9 does not show a "pause/hiatus" in global warming and thus does not conflict with my points in section 3.4.)

In contrast to figure 1 from Christy, figure 9 addresses observational uncertainty by including the UW analysis and correcting for satellites erroneously reading stratospheric cooling as tropospheric temperature. Figure 9 also presents a clearer representation of model uncertainty, as seen in the gray region of figure 9. One could produce similar figures using the re-analysis and radiosonde data I discussed in section 3.2. The remaining model-observations discrepancy in figure 9 is likely due to a number of factors, as I discussed in section 3.3. But even with these remaining discrepancies, figure 9 helps remedy Christy's exaggeration of the model-observations discrepancy.

3.7 Neglecting research confirming model-based cloud predictions

Christy indirectly links the model-observations discrepancy to a failure of the models to accurately represent cloud responses. To do this, Christy first claims that the model-observations discrepancy is due to the models being too sensitive to CO2. Then Christy argues that models are too sensitive because the models under-estimate the ability of clouds to limit global warming. Of course, Christy conveniently does not seem to think that clouds will similarly limit global warming caused by other factors, such as the Sun or El Niño (for more on this, see section 3.5 of part 1 of my series on "John Christy and the Tropical Tropospheric Hot Spot"). In any event, Christy also emphasizes how uncertain climate science is with respect to respect to clouds , and other scientists agree that climate models have some issues with respect to clouds.

To see where Christy may have gone wrong, let's examine the relationship between CO2 and another powerful greenhouse gas, water vapor. CO2 absorbs energy at wavelengths missed by water vapor, which helps explain why CO2 can contribute to global warming even in the presence of water vapor. Furthermore, water vapor is a condensing greenhouse gas that condenses into liquid water at colder temperatures. This makes water vapor very responsive to temperature changes, and thus very poor at driving up long-term temperature to very high levels in Earth's current climate. CO2, in contrast, is a non-condensing greenhouse gas that does not condense at the temperatures and pressures normally seen in the troposphere. This allows CO2 to accumulate, with a longer atmospheric residence time than that of water vapor, even in the presence of short-term cooling. CO2 can thus drive temperatures up in the long-term, in contrast to water vapor, resulting in a long-term correlation between CO2 and temperature.

As CO2 warms the atmosphere, we can predict that atmospheric water vapor levels will increase since warmer air can hold more water vapor. And since water vapor acts a greenhouse gas, increased water vapor will serve as a fast positive feedback that amplifies the warming caused by CO2. Water vapor can also result in the formation of clouds. These clouds with then act as a positive feedback on warming or as a negative feedback, depending on the nature of the clouds and how high the clouds are in the atmosphere. Clouds can reflect solar radiation and thus act as a negative feedback, or clouds can reflect/absorb radiation emitted by the Earth and thus act a a positive feedback. Lower level clouds tend to act as a negative feedback, while higher level clouds tend to act as a positive feedback. Climate models predict a net positive feedback from clouds due to increases in higher level clouds and reductions in lower level clouds, though different models disagree on some aspects of cloud feedback.

So let's synthesize all these claims together into two model-based predictions:

1. Water vapor levels increase during warming, since warmer air can hold more water vapor.
2. Higher water vapor levels act as a positive feedback on warming, including a net positive feedback from clouds that form from water vapor.

Observational evidence supports each of these predictions. Water vapor levels increased during recent periods of global warming, with much of the increase occurring in response to anthropogenic warming. There is some additional evidence of water vapor acting as a positive feedback that amplified warming. Clouds also acted as a positive feedback, consistent with model-based predictions. Cloud responses may therefore augment CO2-induced global warming. Scientists also further improved how climate models simulate clouds and scientists reduced uncertainty in estimates of cloud feedback.

Taken together, these results undercut Christy's critique of model-based cloud predictions, since it turns outs that Earth has cloud-based mechanisms that amplify the warming caused by greenhouse gases, as predicted by the climate models.

4. References

1. https://via.hypothes.is/http://climatefeedback.org/wp-content/uploads/2017/04/model-obs-middle-tropo.png
2. "Removing diurnal cycle contamination in satellite-derived tropospheric temperatures: understanding tropical tropospheric trend discrepancies"
3. "Testimony. Data or dogma? Promoting open inquiry in the debate over the magnitude of human impact on Earth’s climate. Hearing in front of the U.S. Senate Committee on Commerce, Science, and Transportation, Subcommittee on Space, Science, and Competitiveness, 8 December 2015"
4. "Comparing tropospheric warming in climate models and satellite data"
5. "Tropospheric warming over the past two decades"
6. http://images.remss.com/msu/msu_time_series.html (accessed June 5, 2017)
7. https://www.esrl.noaa.gov/psd/enso/mei/
8. "Ranking the strongest ENSO events while incorporating SST uncertainty"
9. "UAH version 6 global satellite temperature products: Methodology and results"
10. "On the divergence between the UAH and RSS global temperature records" [http://archive.is/6EIbO#selection-191.0-191.247]