“Benchmark forecasts for climate change" is a new working paper by Kesten Green, Scott Armstrong, and Willie Soon. Using forecasting principles the authors selected a naïve no-change method as the appropriate benchmark for forecasting global average temperatures over the 21st Century.
When they compared forecasts from 1992 through 2008, the Intergovernmental Panel on Climate Change’s 1992 forecasts were no more accurate than a naïve benchmark model of “no change from the most-recent-year prior to the forecast.”
The test was limited, as it covered only short-term horizons and was based on few observations, so they conducted a backcasting validation study from 1974 to 1850, used horizons up to 100-years, and obtained 7,550 forecasts. The IPCC backcast errors were seven times larger than the benchmark errors. The IPCC errors increased in relative as well as absolute terms with longer horizons.
The benchmark forecast for the rest of the century is that global average temperatures will be within 0.5°C of the 2008 figure. In effect, global average temperatures cannot be forecast over policy-relevant horizons. There is no scientific justification for climate change policy.
Kesten Green will present the paper at the Heartland Institute’s 2009 International Conference on Climate Change in New York City, March 8-10. It has been accepted subject to further revisions for a Special Issue on decision making and planning under low levels of predictability in the International Journal of Forecasting.