Guest post by Mark Richardson who is a Research Scientist in the Aerosol and Clouds Group at NASA’s Jet Propulsion Laboratory, California Institute of Technology. All opinions expressed are his own and do not in any way represent those of NASA, JPL or Caltech.
Should scientists choose to believe provably false things? Even though that would mean more inclusive debates with a wider range of opinions, our recent paper Richardson & Benestad (2022) argues no: “instead of repeating errors, they should be acknowledged and corrected so that the debate can focus on areas of legitimate scientific uncertainty”. We were responding to Connolly et al., who suggested that maybe the Sun caused “most” of the warming in “recent decades” based on a simple maths mistake.
Connolly et al. point out that there are many solar activity datasets, then attempted a statistical calculation and said that some datasets support that “most” of the warming in “recent decades” could be due to the Sun. The problem is obvious when you look at the statistical results below, with temperature data (T) in black (NH = northern hemisphere), estimated solar effect in blue and human effect in orange*.
Obviously, no blue line explains “most” of the warming. Connolly et al.’s most important mistake was that instead of calculating solar and human effects at the same time, they decided to first assume that all possible correlation is explained by the Sun, and then any leftovers are from human activity. This is a baseless assumption, and you could just as easily do human activity then the Sun. Both are wrong, and there’s no physical reason to pick either.
Doing humans first gives a solar effect near 0 %. If your conclusions depend on the order in which you enter numbers into a computer, maybe you should check your methods. Some Connolly et al. authors noticed this in 2015 (Soon et al., 2015) but rather than fix it they now chose to report just the (wrong) calculations that supported their conclusions.
There’s a simple test for statistical methods, where you create a toy world in which you know the real answer. If your calculations give that known answer then they pass, while failure means the method should be thrown out. Let’s try that and assume that a Watt of solar heating and a Watt of human-caused heating cause equal warming. One toy world is shown below: blue is its solar effect, orange the human effect and black the combined changes. The thin black line is what’s measured and includes weather.
The thick dashed lines below show the statistical results. Standard regression on the left passes the test, while Connolly et al.’s method on the right fails. As part of its failure, it invents a massive and non-existent solar effect, and this mistake is the only reason they could make a claim about “most” warming “in recent decades” potentially being from the Sun.
Below are real temperature data and a modern solar dataset. The mistaken Connolly et al. calculation (dashed lines) gives a huge solar effect, while the actual result (solid lines) is nearer 0 %.
Some older datasets show huge solar variability before satellite measurements, such as this one:
In this case even Connolly et al.’s wrong dashed-blue line is flat since about 1940 but they reported that this solar dataset supports 58 % of warming being due to the Sun. How does it explain “most” warming in “recent decades”? Also, even the incorrectly calculated solar activity’s 0.3 °C above 1850 is clearly not 58 % of the 1.6 °C total warming.
The way to get those big fractions is to fit a straight line, even though the datasets are obviously not straight lines. This makes the recent solar contribution look bigger, e.g. if we zoom in to “recent decades” since 1950 and zero everything then:
The wrong Connolly et al. solar fit (thin wavy blue line) shows a cooling Sun recently, but the straight-line fit is tilted upwards by the effect of the historical wiggles. This calculation literally turns a recent cooling effect into a warming one. Another result is that human activity and real-world warming accelerated after the 1950s, but the historical changes lower the linear temperature fit (dashed black line) to falsely make the solar fraction look bigger.
Should scientists rely on calculations we know are inaccurate? We strongly believe no: errors should be corrected. In our opinion, this is crucial not just for success in science, but for the credibility of science. Our position is that clearly the Connolly et al. approach is nonsense, there is no evidence for the paper’s main claim and it should be corrected or retracted.
*Technical note: each solar activity dataset gives one blue and one orange line. For “recent decades” I removed solar datasets that end before 2005. I also plotted Northern Hemisphere land temperatures, since that’s what Connolly et al. use
- M.T. Richardson, and R.E. Benestad, "Erroneous use of Statistics behind Claims of a Major Solar Role in Recent Warming", Research in Astronomy and Astrophysics, vol. 22, pp. 125008, 2022. http://dx.doi.org/10.1088/1674-4527/ac981c
- R. Connolly, W. Soon, M. Connolly, S. Baliunas, J. Berglund, C.J. Butler, R.G. Cionco, A.G. Elias, V.M. Fedorov, H. Harde, G.W. Henry, D.V. Hoyt, O. Humlum, D.R. Legates, S. Lüning, N. Scafetta, J. Solheim, L. Szarka, H.V. Loon, V.M. Velasco Herrera, R.C. Willson, H. Yan, and W. Zhang, "How much has the Sun influenced Northern Hemisphere temperature trends? An ongoing debate", Research in Astronomy and Astrophysics, vol. 21, pp. 131, 2021. http://dx.doi.org/10.1088/1674-4527/21/6/131
- W. Soon, R. Connolly, and M. Connolly, "Re-evaluating the role of solar variability on Northern Hemisphere temperature trends since the 19th century", Earth-Science Reviews, vol. 150, pp. 409-452, 2015. http://dx.doi.org/10.1016/j.earscirev.2015.08.010