A machine learning method for the evaluation of ship grounding risk in real operational conditions

Ship groundings may often lead to damages resulting in oil spills or ship flooding and subsequent capsizing. Risks can be estimated qualitatively through experts' judgment or quantitatively through the analysis of maritime traffic data. Yet, studies using big data remain limited. In this paper, we present a big data analytics method for the evaluation of grounding risk in real environmental conditions. The method makes use of big data streams from the Automatic Identification System (AIS), nowcast data, and the seafloor depth data from the General Bathymetric Chart of the Oceans (GEBCO). The evasive action of Ro-Pax passenger ships operating in shallow waters is idealized under various traffic patterns that link to side - or forward - grounding scenarios. Consequently, an Avoidance Behaviour-based Grounding Detection Model (ABGD-M) is introduced to identify potential grounding scenarios, and the grounding probabilistic risk is quantified at observation points along ship routes in various voyages. The method is applied on a Ro-Pax ship operating over 2.5 years ice-free period in the Gulf of Finland. Results indicate that grounding probabilistic risk estimation may be extremely diverse and depends on voyage mutes, observation points, and operational conditions. It is concluded that the proposed method may assist with (1) better identification of critical grounding scenarios that are underestimated in existing accident databases; (2) improved understanding of grounding avoidance behaviours in real operational conditions; (3) the estimation of grounding probabilistic risk profile over the life cycle of fleet operations and (4) better evaluation of waterway complexity indices and ship operational vulnerability.