How well have climate models forecast global temperature?
I took the data for global average temperature from climate model simulations in the CMIP5 archive; those are computer models used in the latest IPCC report. I used only those models with the “RCP4.5” emissions scenario (a middle-of-the-road choice). I then aligned them all so their average value was zero during the 1961-1990 “baseline” period. Finally, I calculated yearly averages for each of the 108 models included.
That enables me to compute the “multi-model mean,” the average of all the models at each moment of time. Also at each moment of time, I computed the standard deviation of the model values and recorded the highest and lowest model values (which can be different models at different times).
Now I can graph the multi-model mean over time as a thick red line, together with a yellow outermost envelope showing the range from highest to lowest, a tan-colored middle range the limits of the 2-sigma range (about 95% of the models) and a pink band the 1-sigma range (about 2/3 of the models).
And I can also plot actual observed global temperature from NASA (yearly averages using the same 1961-1990 baseline) as a black line:
Recent values are genuine predictions, in the sense that the observed values of global temperature weren’t known when the models were run (it can take a long time to run these models, even on a supercomputer).
Clearly actual temperature has followed the model results closely, staying within the 1-sigma range most of the time. The latest value (2019 year-to-date) is right in the bull’s eye.
Thanks to all those who generously contributed to the blog. If you’d like to help, please visit the donation link below.
This blog is made possible by readers like you; join others by donating at My Wee Dragon.