Science · Earth and Space Science
Chapter 1: Earth as a System
A model that matches the past has not yet predicted anything.
In June 1991 Mount Pinatubo in the Philippines blasted about 15 to 20 million metric tons of sulfur dioxide into the stratosphere. Months later, before most of the effect, a NASA team forecast that the world would cool by about 0.5 °C (0.9 °F), most of it in 1992.
If the cooling came as forecast, what would that show that matching the past could not?
Make your guess to turn the page.
Still to come: 10 pages · Lab · Words to Know · What We Learned · 4 questions · Explain It in Writing · Open Questions
The idea this lesson is working on — Students take agreement with past data as proof that a model is right, and so cannot tell a tuned fit from a test. The book sets the Pinatubo forecast, made before the data existed, against two models that fit the same past and diverge. Ask: what data was this model fitted to, and what has it predicted that it was not shown? The idea to reach is that fitting is a minimum and out-of-sample prediction is the test.
Matching the past is not yet a prediction — Hansen, Lacis, Ruedy and Sato published the Pinatubo forecast in Geophysical Research Letters in 1992; the observed cooling matched it closely, which is why it is the standard example of an out-of-sample test. Water-cycle numbers are rounded from Trenberth et al. (2007). The two diverging model curves on the last page are illustrative, not real model output.
Uncertainty is quantified, not hidden — Good models report a range and say where it comes from — measurement error, unknown parameters, or scenario choices. IPCC AR6 (2021) gives equilibrium climate sensitivity as likely 2.5–4 °C, best estimate 3 °C. A single number with no range is the less trustworthy output.