Atmosphere Ocean Science Colloquium

---

Ill-Posed Inverse Problems in Multiscale Geophysical Fluid Flows

Speaker: Sai Prasanth, NASA JPL

Location: Warren Weaver Hall 1302

Date: Wednesday, April 1, 2026, 3:30 p.m.

Synopsis:

Space-borne observations of precipitating tropical convective storms create ill-posed inverse problems where each outcome is consistent with many possible inputs. I will present three connected variants of this class of problems where in each case, classical deterministic approaches fail. First, is the problem of estimating geophysical variables from satellite observations where each observed state maps to many possible physical states. Second, we will talk about relating convective outcomes to environments where each outcome may be consistent with many environmental states. Third, we will talk about forecasting storm evolution from cloud-top imagery, where each image is consistent with many possible 3D vertical structures and developments beneath. I will present a probabilistic framework that we developed at JPL to address these problems combining machine learning and high-resolution physical models to represent uncertainty explicitly and constrain families of plausible solutions. The result is a common mathematical foundation across observation, environment, and prediction tasks, with direct implications for current NASA satellite science efforts including the upcoming INCUS mission.