So, after learning about evolutionary uncertainty from Inga Hense last week, model uncertainty was on our agenda today. Johanna Baehr showed very nice graphs, distinguishing between different sources of uncertainty.
Uncertainty due to sensitivity to initial conditions: It is important to note that most of the earth system models we look at are basically solving transient (i.e. time-dependent) non-linear partial differential equations. In a limited time frame, the solution to such mathematical equation depends on initial conditions. Johanna showed in an example that the variability (i.e. uncertainty with respect to one fixed solution) is somewhat bounded in a narrow corridor.
Uncertainty due to boundary conditions/ external forcing: Again, certain types of partial differential equations are almost uniquely determined by their boundary conditions, so it was (to me) not surprising to see that the temperature change as a modeling result is almost purely determined by assumptions on the given carbon emission scenario, in other words the boundary conditions.
What Johanna did not cover is uncertainty due to the choice of the model. While this is somewhat included in the “fitness for purpose” quest, different modeling approaches are in fact determining some uncertainty.
And another uncertainty that is usually considered to be non-existent in the model-user community is uncertainty due to insufficient numerical implementation. This is a topic little explored. An (old) article published in American Scientific took up this issue (here is a link).