Generating bunkering statistics from AIS data: A machine learning approach

In shipping, the optimization of the bunkering location is dependent on price, deviation from the planned route and the cost of delays incurred by the bunkering operation itself. Despite their potential importance, detailed statistics for bunkering operations at the individual port call level (e.g. waiting times, barge capacity, location - anchorage or terminal) are not available. I develop a new method, based on the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm, that can a) identify tanker vessels used as bunkering tankers, b) detect stationary ocean-going vessels at anchorage or alongside terminals and c) automatically recognize bunkering operations as a rendezvous between an ocean-going vessel and a bunkering barge. I find that the high time complexity of the DBSCAN algorithm in this setting can be compensated by adjusting the algorithm to distributed computer settings. In the empirical study, I use the output to describe the relative importance of Mediterranean ports for bunkering and provide statistics on waiting and servicing times. The empirical findings are important for the optimization of the bunkering location decision in shipping and studies on regional port competitiveness.