From Laboratory to Application: A Survey of Data-Driven Weather Forecasting

PhD Qualifying Examination


Title: "From Laboratory to Application: A Survey of Data-Driven Weather 
Forecasting"

by

Mr. Hao CHEN


Abstract:

The explosive maturation of deep learning has catalyzed a profound paradigm 
shift in Earth system science, fundamentally disrupting the numerical weather 
prediction (NWP) frameworks that have dominated atmospheric modeling for over 
half a century. This survey provides an exhaustive, end-to-end panoramic 
analysis of the rapid evolution of data-driven weather forecasting. Tracing 
the trajectory from early laboratory proof-of-concept explorations to critical 
operational deployments in premier global meteorological centers, we establish 
a rigorous, mutually exclusive four-stage classification system: early 
data-driven exploration, reanalysis-based forecasting, real-data integration, 
and operational commercial services. By systematically evaluating foundational 
data infrastructures and standardized benchmarking ecosystems, we delineate 
the data cornerstones driving algorithmic breakthroughs. Furthermore, we 
meticulously dissect advanced artificial intelligence architectures across 
diverse atmospheric applications, spanning from tropical cyclone tracking to 
extreme precipitation nowcasting. Despite unprecedented computational 
acceleration and deterministic accuracy, data-driven models continue to 
grapple with severe structural vulnerabilities, notably the systematic 
underestimation of extreme weather intensity, physical inconsistency, and 
rapid error accumulation at subseasonal scales. To transcend these 
limitations, this survey synthesizes cross-disciplinary insights to propose a 
mathematically rigorous blueprint for future research. We advocate for 
physics-guided continuous latent alignment, adaptive sparse-orthogonal 
processing for ultra-high-resolution grids, and memory-anchored continuous 
state-space forecasters to conquer the 'predictability desert'. Ultimately, 
this document serves as a foundational roadmap guiding the development of the 
next generation of highly reliable, physically consistent, and operationally 
autonomous AI weather prediction systems.


Date:                   Thursday, 8 May 2026

Time:                   2:00pm - 4:00pm

Venue:                  Room 2126A
                        Lift 19

Committee Members:      Prof. Song Guo (Supervisor)
                        Dr. Dan Xu (Chairperson)
                        Dr. Zihan Zhang