Abstract

Reactive transport and Earth system processes involve strongly coupled nonlinear dynamics across heterogeneous spatial, temporal, and spectral scales, posing significant challenges for conventional neural operators that rely on a single representation domain. In this talk, I present UFO, a generalized operator learning framework that enables adaptive interactions among physical, spectral, and latent representations without enforcing domain unification. UFO supports discretization-decoupled learning, allowing input functions to be observed at resolutions or locations different from training while enabling flexible solution querying across output discretizations. I will discuss the application of UFO to Earth system modeling problems, including reactive transport processes, and show how cross-domain operator realization can improve robustness under irregular sampling, nonlinear dynamics, and multi-scale structure.

Speaker Bio

Hanli Qiao is an invited speaker at AIEPS 2026.

Materials and Images

  • Slides: To be added.
  • Related links: To be added.
  • Images: To be added.
Back to top