The weather is something that impacts everyone on a daily basis and much research has been done to improve weather forecasting in the past. This is also the case in the aviation industry, using traditional data sources as well as aircraft measurements. Previous research in this domain has developed the Meteo-Particle (MP) model which constructs a wind field based on data collected by aircraft and UAVs. Research has also been done with new physically inspired machine-learning approaches to create wind fields. This research, with the ultimate goal being to reduce uncertainty in aircraft estimated time of arrival (ETA), intends to approach the problem of creating accurate 3D wind fields with a diffusion neural network, filling in the gaps where there is no aircraft data available. Previous models struggle with non-uniform wind fields, this new approach presents an opportunity for better reconstruction of the wind fields under these conditions.
Graduated: July 2025