A subjective-objective composite risk perception field for AIS-based ship collision avoidance behaviour analysis in coastal waters

Coastal waters are characterised by dense traffic, constrained navigable space, heterogeneous vessel types and frequent encounters. Collision avoidance behaviour reflects not only physical collision risk but also navigators’ perceived spatial pressure, which conventional DCPA, TCPA and ship-domain indicators cannot fully explain. This study proposes a subjective-objective composite risk perception field (CRPF) for AIS-based collision avoidance behaviour analysis. The framework separates encounter risk into a subjective risk perception field (SRPF) and an objective collision risk field (OCRF). The SRPF is formulated as an AIS-derived, navigator-centred proxy for perceived navigational pressure associated with target ships, navigational boundaries and limited manoeuvring space, rather than as a direct measure of psychological state. The OCRF quantifies physical collision tendency based on relative motion, closest point of approach, ship-domain intrusion and encounter geometry. The framework was validated using 1670 encounter events extracted from real-world AIS data in the Ningbo–Zhoushan coastal waters. Head-on and multi-ship encounters showed coupled increases in subjective and objective risk, whereas crossing and overtaking encounters more frequently exhibited pressure-dominated patterns. Under identical AIS inputs, encounter-level data splits and warning criteria, the proposed CRPF was compared with an LSTM model and a spatiotemporal graph attention network (ST-GAT). On the independent testing set, ST-GAT achieved the highest behavioural consistency of 83.8% and an F1-score of 0.85, while the CRPF achieved comparable values of 82.9% and 0.84, respectively. The CRPF provided the longest mean warning lead time of 9.5 min, compared with 9.1 min for ST-GAT, while enabling explicit decomposition of risk sources and their spatiotemporal evolution. The framework therefore provides an interpretable basis for maritime risk monitoring, collision avoidance modelling and navigator behaviour analysis in complex coastal waters.