Deep learning for aviation turbulence forecasting with numerical weather prediction models

Abstract

In this thesis, we explore the use of deep learning models to enhance numerical weather prediction (NWP) forecasts for aviation turbulence. NWP models forecast atmospheric state by numerically solving mathematical equations of known physical laws. Currently, it is the primary way to forecast weather including aviation turbulence. Using the NWP forecasts as input, we utilize deep neural network to generate refined prediction with forecast time range from 0 to 60 hours. The model output is the predicted turbulence intensity level: (i) nil or light, (ii) moderate, and (iii) severe. Experiments are performed on a dataset of 53,380 turbulence reports collected from year 2018 to 2022. Results show that the proposed model outperforms commonly-used turbulence forecasting baselines, including the significant weather (SIGWX) charts issued by World Area Forecast Center (WAFC) and diagnostic indices generated by NWP models.

Date of Award	2024
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology
Supervisor	James Tin Yau KWOK (Supervisor)