Bûn-Kim San

PhD Thesis

Prediction of adverse weather events and climate impact using neural networks

Supervisors

Manuel Soler (UC3M) & Javier García-Heras (UC3M)

Abstract

Meteorology and aviation share the same object, the atmosphere, and are linked by a common history. Each pushes the other in the quest for safety (25% of air accidents are associated accidents are associated with a meteorological cause) and improved performance (air delays in Europe cost more than 1 billion a year to airlines, at least as much as air navigation charges. Air Navigation Charges (ANC). On average, 5% of delays in Europe
are weather-related). It is therefore vital to improve our methods of forecasting this phenomenon.
The idea of this work is to use AI methods to perform nowcasting (0-2h) and downscaling. We will study the performance of physically constrained architectures (PINNs) on weather observations (stations, satellite, aircraft observations, etc …)