Abed Hammoud

I develop methods that use physical structure and learned correction policies to recover unresolved states from sparse or coarse observations. This includes continuous and discrete data assimilation, physics-informed neural surrogates for downscaling, and reinforcement-learning strategies for chaotic systems. Lorenz models and Rayleigh–Bénard convection provide controlled settings for testing learned estimates under limited information.

Uncertainty is part of the model, not an afterthought. I use Bayesian neural networks, posterior predictive evaluation, model selection, and inverse methods to represent uncertainty in data, parameters, and predictions. This work asks how probabilistic learning can support calibrated retrievals and more defensible decisions when observations are noisy and several models can plausibly explain them.

The methodological work is motivated by atmosphere–ocean and climate questions: turbulent fluxes in the atmospheric boundary layer, satellite ocean-color retrieval, mesoscale eddies, and marine-pollution source identification. Across these applications, the goal is to connect mathematical structure, computation, and observations without hiding the uncertainty that shapes the scientific answer.