Path2Vec: A Deep Representation Learning Method for Trajectory Feature Extraction and HYSPLIT Uncertainty Quantification

Accurate quantification of the uncertainty in HYSPLIT model simulations is crucial for analyzing atmospheric pollution propagation paths and assessing environmental risks. This study introduces Path2Vec, a method based on deep representation learning for extracting trajectory features and measuring uncertainty. The method is capable of mining spatiotemporal-independent trajectory motion patterns in the HYSPLIT model. We first extract spatiotemporal-invariant features of the trajectories using a sliding window technique. Subsequently, we utilize a deep representation learning model that integrates a variational autoencoder (VAE) with long short-term memory (LSTM) networks to encode high-quality deep representations of the trajectories. By measuring the similarity and performing clustering analysis on the generated trajectory deep representations, we can identify and classify different motion patterns, and quantify the uncertainty of HYSPLIT. Experimental results indicate that the Path2Vec method surpasses traditional similarity measurement techniques, such as Euclidean distance and Edit Distance on Real sequence, in extracting spatiotemporal-independent motion patterns and quantifying uncertainty. This study provides a novel and effective approach for trajectory feature extraction and uncertainty quantification, with wide-ranging applications in fields such as meteorological simulation and air pollution propagation path analysis.