Comparative Analysis of Three Methods for HYSPLIT Atmospheric Trajectories Clustering

Using the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model to obtain backward trajectories and then conduct clustering analysis is a common method analyze potential sources transmission paths of atmospheric particulate pollutants. Taking Qingdao (N36 E120) as an example, global data assimilation system (GDAS 1°) days from 2015 2018 provided by National Centers for Environmental Prediction (NCEP) used process 72 h trajectory 3 arrival heights (10 m, 100 500 m) through HYSPLIT with interval 6 (UTC 0:00, 6:00, 12:00, 18:00 per day). Three methods data, i.e., K-means, Hierarchical (Hier), Self-organizing maps (SOM), are results compared those released Oceanic Atmospheric Administration (NOAA). Principal Component Analysis (PCA) original data. The internal evaluation indexes Davies–Bouldin Index (DBI), Silhouette Coefficient (SC), Calinski Harabasz (CH), I index quantitatively evaluate three algorithms. show that there little information in height thus only two-dimensional plane clustering. From indexes, SOM K-means better than Hier model. In addition, it found DBI can help select number clusters, which preferred cluster analysis.