Numerical weather prediction (NWP) is like excellent coffee in the Bay Area: so common that it is now taken for granted, obscuring the decades of expertise, knowledge, and technique underlying the whole operation. In episode 77 of Forecast, Peter Bauer from the European Centre for Medium-range Weather Forecasts tells Mike about the massive and decades-long efforts that have made NWP so incredibly useful for modern society.

But the field still grapples with historically-intractable issues, such as the need to parameterize critical processes like cloud convection, ocean eddies, and atmospheric gravity waves. Maybe the best way to deal with parameterizations … is not to do them at all. Towards this goal, Peter and his colleagues are now pushing forward on a preposterously ambitious proposal called Extreme Earth, in which they would conduct NWP at ~ 1 km resolution, thereby allowing simulation of the key physical processes. Doing so, however, demands a 10,000x increase in computational power, a mind-boggling challenge that goes way beyond the usual approach of bolting together thousands of processors.

Extreme Earth also proposes to invert the usual scientific information flow in NWP in particular and application-oriented science in general. The normal process goes something like: (1) scientists decide what they think the user community might want (2) scientists spend years developing such products (3) scientists show policy/management/public the new product (4) intended audience yawns (5) repeat process. In Extreme Earth, the goal instead would be to hand the reins over to the users, so that they would be able to design the experiments and information flow that would best suit their needs.

How any of this would work is a research question, but it’s one that Peter and the Extreme Earth community are keen to tackle. If funded, to the tune of about a billion euro, the project would certainly represent the most ambitious current program to take weather and climate modeling to a new phase of scientific rigor and societal relevance.