Key points are not available for this paper at this time.
Traditional empirical models for ship–bridge collision risk are constrained by sparse accident records and assumptions of normally distributed tracks. Consequently, detection of rare yet consequential yaw extremes is hindered. To address these limits, a ship–bridge collision risk assessment framework is proposed using surrogate safety measures (SSM) and bivariate extreme value theory (B-EVT). From Automatic Identification System trajectories, three SSMs—negative spatial clearance (NSC), negative trajectory deviation (NTD), and time of risk exposure (TRT)—are constructed to capture complementary aspects of collision propensity. Joint extremes are modeled for the pairs NSC–NTD, NSC–TRT, and NTD–TRT using a bivariate threshold-excess approach under B-EVT, by which high-risk trajectories are identified. In 2543 passages under the Jintang Bridge, 152 true high-risk trajectories are detected. Performance is validated against a high-risk sample compiled from Vessel Traffic Service warnings and penalty records, with standard confusion-matrix metrics (accuracy, recall, specificity, precision, F1-score) computed. Inclusion of TRT yields the best results: for NTD–TRT, 99.61 % accuracy, 92.76 % recall, 99.96 % specificity, 99.29 % precision, and a 95.91 % F1-score are achieved. The proposed framework effectively identifies ship–bridge high-risk trajectories without relying on accident samples, demonstrating strong robustness and interpretability for refined navigational safety management.
Liu et al. (Thu,) studied this question.