Climate sensitivity, defined as the increase in the Earth’s surface air temperature caused by a doubling of atmospheric CO2, is a curious metric. It isn’t used as direct input for anything – it’s derived from the output of climate models or from observational analyses – but no other variable in all of climate science is so controversial. This is because it tells us, in one single number, how serious a problem CO2-induced warming might be. If climate sensitivity is down around 1.5C, the low end of the IPCC’s range, the impacts probably won’t be serious, maybe not even noticeable. But if it’s up around 4.5C, the high end of the range, watch out.
So what is the true value of climate sensitivity?
The first estimate – 5.5C – was made by Svante Arrhenius in 1896. Ten years later in 1906 he lowered it to 2.1C. Then came Hulbert in 1931 with 4C, Callendar in 1938 with 2C, Plass in 1956 with 3.8C, Möller in 1963 with 9.6C, Manabe & Wetherald in 1971 with 2.4C and Sellers in 1973 with 0.1C. Figure 1 plots these and a couple more contemporaneous estimates by year. Clearly scientists were having difficulty putting a value on climate sensitivity in the early days:
Figure 1: Early climate sensitivity estimates
But things are different now, right?
Figure 2: 120 observational estimates of equilibrium climate sensitivity from 1896 to 2014. Only median or average values are plotted; the scatter would be much greater if high/low estimates were included. Data sources are listed at the end of the post.
A casual observer looking at Figure 2 might be excused for concluding that scientists are no closer to reaching agreement on the true value of climate sensitivity now than they ever were. As unlikely as it seems, however, there are trends, although the data have to be segregated into time groups to reveal them. I used six with a roughly equal number of estimates in each:
Before 1980: 19 measurements centered on 1960.8
1980 through 1993: 22 measurements centered on 1986.7
1995 through 2005: 20 measurements centered on 1999.9
2006 through 2011: 19 measurements centered on 2008.9
2012: 25 measurements centered on 2012
2013 & 2014: 15 measurements centered on 2013.3
Figure 3 plots the means, medians and standard deviations of these six groups. Means and medians show a downward trend since 1985, indicating that climate sensitivity estimates have been generally decreasing over the last 30 years, and the erratic downward trend in standard deviation since 1960 indicates that the level of estimation uncertainty has also been generally decreasing:
Deleting the three high-value outliers seen in Figure 2 – Möller 1963, Pagani 2010 and Hansen 2012 – has only a small impact on the mean and median values but cleans up the standard deviation plot, which now shows a coherent and accelerating downtrend:
Table 1 summarizes the results numerically. The means and medians of the estimates have fallen from 3 – 3.5C since 1981-93 to around 1.9C now – a substantial drop – and the decrease in the spread of the 90% confidence limits (equivalent to IPCC “very likely”) from 4.45C to 1.8C represents a significant improvement in the level of agreement:
Climate models are also used to obtain estimates of climate sensitivity, and while there aren’t enough estimates in Figure 5 to demonstrate convergence they also show a downward trend heading towards a 2013-14 value of around 2C, comparable to the 2013-14 observational values shown in Table 1:
Figure 5: Climate sensitivity estimates from climate models, excluding IPCC data
Figure 5, however, excludes the IPCC’s climate-model-based estimates, which place the value of climate sensitivity at around 3C with a “likely” range of from 1.5C to 4.5C. The IPCC’s estimates have in fact shown remarkable consistency over the years. Except for the increase in the lower limit from 1.5C to 2.0C in the AR4 they haven’t changed since the Charney report got the global warming ball rolling 35 years ago:
Figure 6: Charney & IPCC climate sensitivity estimates
The IPCC’s data nevertheless exhibit one interesting feature. The model-derived climate sensitivity estimates don’t match the model-derived temperatures. The mean of the 26 individual AR5 model-derived climate sensitivity estimates is 3.3C, but when the same models predict temperatures they operate at an effective climate sensitivity of only 2.2C. Figure 7 shows the near-exact fit at 2.2C for RCP8.5, the AR5’s worst-case scenario, which projects 1,250ppm CO2 by 2100 and 3.7C of warming to go with it. The fits for the RCP6.0 and RCP4.5 scenarios are similar:
Figure 7: Model-predicted surface air temperatures vs. temperatures calculated using climate sensitivity = 2.2C, IPCC AR5 RCP8.5 scenario.
Why do the IPCC’s climate models operate at 2.2C and not 3.3C? Because according to global warming theory temperature equilibrium isn’t reached for hundreds of years after CO2 concentrations stabilize, meaning that model temperatures will keep climbing and eventually flatten out at a level compatible with a +/-3.3C equilibrium climate sensitivity if the models run long enough. But are computer model projections of what might happen hundreds of years into the future meaningful? Are they even relevant? Figure 7 shows an excellent fit to a climate sensitivity of 2.2C over a period of 200 years, which exceeds human predictive capability by a large margin and certainly my attention span. How far out do we have to go?
I contend that 200 years is quite far enough, in which case the IPCC’s best current estimate of climate sensitivity is 2.2C, not 3C. And at 2.2C the IPCC falls into line with everyone else. We have established a consensus.
And having done so the question becomes, what value will climate sensitivity estimates ultimately zero in on? The current value is around 2C, with estimates decreasing with time and with every prospect that they will continue to do so at least while the warming “pause” persists. My guess is that they will eventually settle somewhere below 1.5C – maybe even as low as 1C, which is generally agreed to be the climate sensitivity to a doubling of CO2 in the absence of feedbacks. At these levels the perceived risks of man-made global warming decrease very significantly, but that will have to be the subject of another post.