Atmosphere Ocean Science Friday Seminar

---

A good baseline is hard to find

Speaker: Dave Connelly

Location: Warren Weaver Hall 1314

Date: Friday, March 7, 2025, 4 p.m.

Synopsis:

Ray tracing offers a framework for gravity wave parameterization that is arguably closer to the underlying physics than traditional steady-state schemes. I am working on data-driven approaches to making it more computationally practical. But whenever you do machine learning, it’s important to establish a fair baseline against which to compare. In this talk, I will briefly review the fundamentals of ray tracing before discussing the various rabbit holes I’ve gone down over the last few months while attempting to find such a baseline. In the process, I will examine the sensitivity of the ray tracer to various parameters and try to comment on the accuracy of the GCM integration with ray tracing shown in Bölöni et al. (2021). Finally, I will review some of the machine learning work I’ve done thus far and talk about some ML ideas I’m in the process of implementing.