How accurate were James Hansen’s 1988 testimony and subsequent JGR article forecasts of global warming?
According to a laudatory article by AP’s Seth Borenstein, they “pretty much” came true, with other scientists claiming their accuracy was “astounding” and “incredible.” Pat Michaels and Ryan Maue in the Wall Street Journal, and Calvin Beisner in the Daily Caller, disputed this.
The whole debate has focused on comparisons of the 1988 and 2017 endpoints. Skeptical Science waived away the differences by arguing that if one adjusts for an overestimation in the rise of greenhouse gas (GHG) forcing, Hansen’s 2017 Scenario B prediction was not far off reality.
There are two problems with the debate as it has played out. First using 2017 as the comparison date is misleading because of mismatches between observed and assumed El Nino and volcanic events that artificially pinched the observations and scenarios together at the end of the sample. What really matters is the trend over the forecast interval, and this is where the problems become visible. Second, applying a post-hoc bias correction to the forcing ignores the fact that converting GHG increases into forcing is an essential part of the modeling. If a correction were needed for the CO2 concentration forecast that would be fair, but this aspect of the forecast turned out to be quite close to observations.
Let’s go through it all carefully, beginning with the CO2 forecasts. Hansen didn’t graph his CO2 concentration projections, but he described the algorithm behind them in his Appendix B. He followed observed CO2 levels from 1958 to 1981 and extrapolated from there. That means his forecast interval begins in 1982, not 1988, although he included observed stratospheric aerosols up to 1985.
From his extrapolation formulas, we can compute that his projected 2017 CO2 concentrations were: Scenario A 410 ppm; Scenario B 403 ppm; and Scenario C 368 ppm. (The latter value is confirmed in the text of Appendix B). The Mauna Loa record for 2017 was 407 ppm, halfway between Scenarios A and B. Note that Scenarios A and B also represent upper and lower bounds for non-CO2 forcing as well, since Scenario A contains all trace gas effects and Scenario B contains none. So, we can treat these two scenarios as representing upper and lower bounds on a warming forecast range that contains within it the observed post-1980 increases in greenhouse gases. Consequently, there is no justification for a post-hoc dialling down of the greenhouse gas levels; nor should we dial down the associated forcing, since that is part of the model computation.
Now note that Hansen did not include any effects due to El Nino events. In 2015 and 2016 there was a very strong El Nino that pushed global average temperatures up by about half a degree C, a change that is now receding as the oceans cool. Had Hansen included this El Nino spike in his scenarios, he would have overestimated 2017 temperatures by a wide margin in Scenarios A and B.
Hansen added in an Agung-strength volcanic event in Scenarios B and C in 2015, which caused the temperatures to drop well below trend, with the effect persisting into 2017. This was not a forecast, it was just an arbitrary guess, and no such volcano occurred.
Thus, to make an apples-to-apples comparison, we should remove the 2015 volcanic cooling from Scenarios B and C, and add the 2015/16 El Nino warming to all three Scenarios. If we do that, there would be a large mismatch as of 2017 in both A and B.
The main forecast in Hansen’s paper was a trend, not a particular temperature level. To assess his forecasts properly we need to compare his predicted trends against subsequent observations. To do this we digitized the annual data from his Figure 3. We focus on the period from 1982 to 2017 which covers the entire CO2 forecast interval.
The 1982 to 2017 warming trends in Hansen’s forecasts, in degrees C per decade, were:
- Scenario A: 0.34 +/- 0.08,
- Scenario B: 0.29 +/- 0.06, and
- Scenario C: 0.18 +/- 0.11.