Automatic identification system (AIS) data are used in a myriad of applications related to marine vessel activity, including emissions modeling efforts. The classification of a vessel’s operating mode (i.e., hotelling, anchorage, maneuvering, and transit) is of relevance to such analyses to accurately estimate emissions, but is not reliably provided in the raw data owing to human error and thus must be inferred. Previous works have identified vessel operating mode by using supervised machine learning, applying speed and engine load factor corrections, or incorporating external datasets (i.e., shapefiles) to geospatially designate anchorage and hotelling sites. Rather than relying on external datasets, we propose a method to classify vessel activity into operating modes using only the speed, heading, and position features in AIS data. To accomplish this, we leverage a combination of comparative segment-wise tiered filtering and unsupervised machine learning techniques (i.e., K-means and DBSCAN density-based spatial clustering of applications with noise clustering algorithms). The results at the port level for ocean-going vessels displayed clear separation between the behavior characteristics of each operating mode. Furthermore, the method demonstrated improved accuracy for identifying vessel activities over simply using the AIS-supplied fields or shapefile geofences.
Sobczak et al. (Tue,) studied this question.