Application of OpenDrift-based trajectory prediction for maritime search and rescue: a case study in the South Sea of Korea

Accurate trajectory prediction of drifting objects is critical for improving survival rates and optimizing search efficiency in maritime search and rescue (SAR) operations. This study presents a comprehensive, end-to-end evaluation of drift prediction performance in the South Sea of Korea. To achieve this, object-specific leeway coefficients were derived from field experiments using three types of manikin drifters (with lifejacket, without lifejacket, and with wetsuit), and their trajectories were simulated using both a probabilistic Monte Carlo framework (OpenDrift) and a deterministic empirical approach based on the IAMSAR manual. The simulations were driven by high-resolution hydrodynamic (SCHISM) and atmospheric (ECMWF) forcing fields, and prediction performance was evaluated using multiple complementary metrics, including Normalized Cumulative Lagrangian Separation (NCLS), Root Mean Square Error (RMSE), Location Prediction Conformance (LPC), and a newly introduced metric, the Conformance-Effort Ratio (CER), which explicitly quantifies the trade-off between search coverage and required search effort. The results show that NCLS can yield systematically conservative evaluations under short travel-distance conditions, primarily due to its normalization structure, where the denominator (i.e., cumulative trajectory-length sum) remains small. This highlights a structural limitation of single-metric evaluation and underscores the necessity of a multi-metric assessment framework. When evaluated using CER, the IAMSAR approach achieves near-complete containment (LPC > 99%) by conservatively expanding the search area, but at the cost of substantially increased search effort. In contrast, the OpenDrift approach maintains a reasonable containment level within a significantly smaller search area, demonstrating intensive spatial distribution characteristics. These findings demonstrate that probabilistic drift modeling, supported by auxiliary indicators like CER, can provide a robust decision support framework for prioritizing high-probability search zones and optimizing resource allocation in SAR operations. Rather than serving as a direct replacement for conventional methods, such approaches offer strong potential as a complementary tool for improving operational efficiency under resource-constrained conditions.