Done by the Virginia Institute of Marine Science, new sea-level report cards offer a look at current sea-level rise rates, and projections. What’s interesting to me is making the projections conditional upon the character of the model used to project. In particular, this “character” is there simple — they show differences between linear and quadratic projections — but the 2050 projection is in most cases markedly different depending upon which model is chosen.

This is very good, because it shows how modeling matters, and how, as Tamino and others have noted elsewhere, proper model criticism and treatment of uncertainties are key.

I think the VIMS presentation is exactly right for public consumption.

For a more technical audience, one familiar with, say, the “advanced” level of presentation at SkepticalScience, I am increasingly fond of ensemble methods^(*), like spaghetti plots. These are very flexible, and can even support a model averaged version of, say, linear and quadratic projections, even if, I think, neither is necessarily defensible on its own.

^(*) Even though ensemble methods are tied tightly in the popular technical literature with machine learning, in particular, within techniques like boosting, here I do not mean to make that connection. Here I mean techniques like those discussed by Vrugt and Robinson (2007) or by Moradkhani, DeChant, and Sorooshian, or by Thuiller, Lafourcade, Engler, and Araújo.