Atmosphere Ocean Science Colloquium

---

Reconstructing SWOT sea surface height time series using diffusion posterior sampling

Speaker: Tatsu Monkman, Post Doctoral Associate, Courant

Location: Warren Weaver Hall 1302

Date: Wednesday, May 6, 2026, 3:30 p.m.

Synopsis:

Since the start of data collection in 2023, the Surface Water and Ocean Topography (SWOT) mission has exceeded its original requirements and revealed exciting potential for studying previously unresolved submesoscale processes and the upper-ocean kinetic energy cycle. However, significant challenges remain in fully exploiting SWOT's sea surface height (SSH) data, particularly with respect to estimating and validating submesoscale surface and subsurface currents and vertical velocities. One major limitation is the temporal and spatial sparsity of SWOT observations, which complicates the use of traditional dynamical reconstruction methods such as those based on quasi-geostrophic (QG) theory that require SSH data on a uniform grid.

We aim to generate and validate realistic, high-resolution 2D SSH fields by extrapolating altimetry and infrared satellite observations onto a regularly spaced grid using diffusion posterior sampling. Specifically, we employ diffusion models trained on a high resolution numerical ocean model, MITgcm llc4320, to generate realistic, time evolving surface fields, and use a stabilized version of diffusion posterior sampling (Chung, 2023, Rozet and Louppe, 2023) to guide the generation process towards fields that are consistent with time series of real SSH and SST observations. We further employ the sliding window technique from the “score-based data assimilation” algorithm (Rozet and Louppe, 2023) to demonstrate how the generation and steering method can be extended to arbitrarily long time series.