Atmosphere Ocean Science Friday Seminar

---

If a (regression) tree falls in a forest and no one is around to hear it, does it still make a sound?

Speaker: Dave Connelly, CAOS

Location: Warren Weaver Hall 1314

Date: Friday, March 25, 2022, 4 p.m.

Notes:

Gravity wave parameterizations in GCMs must simulate the local physics of wave propagation and breaking while also driving realistic climate-scale phenomena, such as the Quasi-Biennial Oscillation. In preparation for forthcoming experiments wherein data-driven parameterizations will attempt to learn gravity wave physics from observations and high-resolution simulations, we are training machine learning emulators of existing parameterizations in order to better understand these methods and their interplay with GCMs.
I will begin with a short-form presentation that I will be giving at a conference the following week, showing experiments with regression tree emulators of a classic gravity wave parameterization coupled to a GCM. During the remainder of the time, I will discuss a few other related topics to varying degrees of detail. These will include mathematical and practical points left out of the presentation; in-progress work studying machine learning parameterizations in a simple one-dimensional model; and, time permitting, application of the one-dimensional studies to analysis of the coupled GCM runs.