GBM: An Ocean Front Detection Framework

Ocean fronts are fundamental dynamical interfaces in the ocean, regulating heat exchange, material transport, and ecosystem functioning. Despite extensive studies, existing front detection approaches—including histogram-based, entropy-based, Lyapunov exponent-based, gradient-based edge detection, and machine learning methods—often face challenges such as fragmented detections, excessive false positives, sensitivity to fixed thresholds, and limited interpretability or reproducibility. To address these issues, this study proposes a physically informed and fully interpretable framework for ocean front detection, termed Gradient-Bayesian-Morphology (GBM). The framework integrates three complementary components: (1) a gradient-based thresholding strategy that introduces a physically meaningful prior emphasizing sharp field transitions; (2) a Bayesian inference scheme that combines gradient priors with local field descriptors to enable adaptive, data-driven classification of frontal zones without ad hoc threshold tuning; and (3) a suite of morphological and graph-based refinement procedures that enforce topological consistency by thinning frontal zones, merging fragmented segments, and systematically eliminating spurious ring structures. GBM is evaluated using daily sea surface temperature and salinity fields from 2022 to 2024 over the South China Sea and the Kuroshio region, representing weak and strong frontal regimes, respectively. Results demonstrate that GBM effectively suppresses over-detection while producing fronts with enhanced spatial continuity. Statistical analyses further indicate stable frontal characteristics with minimal parameter sensitivity, highlighting the robustness and reproducibility of the proposed framework. Overall, GBM provides a transparent alternative to black-box approaches and offers a unified framework for multi-variable ocean front analysis.