Beyond Physics: The AI-Driven Revolution in Numerical Weather Prediction

PhD Qualifying Examination


Title: "Beyond Physics: The AI-Driven Revolution in Numerical Weather 
Prediction"

by

Mr. Tao HAN


Abstract:

Artificial intelligence is reshaping numerical weather prediction by replacing
parts of a costly, hand‑crafted dynamical pipeline with data‑driven models
that emulate, complement, and sometimes surpass traditional cores. This survey
provides a structured overview of this shift, focusing on global data‑driven
forecasters, generative models that represent forecast uncertainty, and
emerging foundation models that treat weather prediction as a representation
learning problem. We outline the data ecosystem that supports AI‑driven NWP,
including reanalyses, operational archives, multimodal remote sensing, and
benchmarks. We then summarize recent modeling advances, covering deterministic
surrogates of global circulation, probabilistic and diffusion‑based models for
extremes, and hybrid physics–AI architectures that integrate learning with data
assimilation and physical constraints. Finally, we discuss evaluation practice,
uncertainty quantification, key application areas, and open challenges in
robustness, non‑stationarity, efficiency, interpretability, and governance that
must be addressed to make AI‑enhanced NWP reliable and operationally useful.


Date:                   Wednesday, 26 November 2025

Time:                   9:00am - 11:00am

Venue:                  Room 5510
                        Lift 25/26

Committee Members:      Prof. Song Guo (Supervisor)
                        Prof. Raymond Wong (Chairperson)
                        Dr. Xiaomin Ouyang