An ocean front detection and tracking algorithm

Yishuo Wang, Feng Zhou, Qicheng Meng, Muping Zhou, Zhijun Hu, Chengqing Zhang, Tianhao Zhao

Abstract

Existing ocean front detection methods--including histogram-based variance analysis, Lyapunov exponent, gradient thresholding, and machine learning--suffer from critical limitations: discontinuous outputs, over-detection, reliance on single-threshold decisions, and lack of open-source implementations. To address these challenges, this paper proposes the Bayesian Front Detection and Tracking framework with Metric Space Analysis (BFDT-MSA). The framework introduces three innovations: (1) a Bayesian decision mechanism that integrates gradient priors and field operators to eliminate manual threshold sensitivity; (2) morphological refinement algorithms for merging fragmented fronts, deleting spurious rings, and thinning frontal zones to pixel-level accuracy; and (3) a novel metric space definition for temporal front tracking, enabling systematic analysis of front evolution. Validated on global SST data (2022--2024), BFDT-MSA reduces over-detection by $73\%$ compared to histogram-based methods while achieving superior intensity ($0.16^\circ$C/km), continuity, and spatiotemporal coherence. The open-source release bridges a critical gap in reproducible oceanographic research.

Introduction

Ocean fronts—the boundaries where different water masses meet—are the "weather fronts" of the sea, regulating nutrient cycling, fisheries, and heat transport. However, accurately detecting them from satellite Sea Surface Temperature (SST) data is difficult due to noise and the limitations of traditional methods. Histogram-based methods often suffer from over-detection, while gradient thresholding requires manual "magic numbers" that lead to discontinuous lines. Lagrangian approaches and machine learning models also face issues with missing wind-driven fronts or lacking physical interpretability. This paper introduces the Bayesian Front Detection and Tracking (BFDT-MSA) framework, an open-source tool designed to provide clean, physically interpretable, and temporally coherent ocean front data.

Methodology

The BFDT-MSA framework operates in three main stages: unified Bayesian detection, morphological coherence enforcement, and metric space tracking.

Unified Bayesian Detection

Instead of treating a front as a binary edge based on a hard temperature cutoff, the algorithm treats it as a probabilistic zone. It utilizes a Bayesian fusion of gradient priors and field operators, such as Local Degree of Edge (LDE) and Block Deviation (BD). This approach makes detection robust against noise and multi-scale variability without requiring manual threshold tuning.

Gradient magnitude distribution used for Bayesian prior estimation.

Morphological Coherence Enforcement

To transform probabilistic zones into clean maps, the framework applies several image-processing techniques:

  • Skeletonization: Thinning frontal zones to a precise 1-pixel width.
  • Discrete Skeleton Evolution (DSE): A greedy algorithm that prunes irrelevant branches from the main front.
  • Merging & Ring Deletion: Filling gaps between segments and removing artificial circular artifacts often caused by eddies.

Metric Space Tracking

To follow fronts over time, the paper defines a formal "distance" in a metric space. By calculating the temporal distance between features on consecutive days, the algorithm can rigorously track a front’s identity, growth, and decay.

Results and Discussion

The framework was validated using NOAA Coral Reef Watch SST data (2022–2024) with a focus on the South China Sea. BFDT-MSA reduced over-detection by 73% compared to histogram-based baselines while maintaining superior continuity. It achieved a frontal intensity of 0.16°C/km, providing a much clearer picture of temperature gradients.

Temporal evolution of detected ocean fronts in the South China Sea from t=0 to t=30.

Tracking analysis revealed that most SST fronts are transient, with very few persisting beyond 15–22 days. The model also successfully captured seasonal dynamics, such as the widening of fronts in winter and their intensification in spring.

Conclusion

BFDT-MSA offers a robust, physics-guided approach to ocean front detection and tracking. By moving away from binary classification and employing geometric refinement, it provides high-quality data for oceanographic research. Future work will focus on extending the framework to submesoscale processes and integrating multi-source data like salinity and chlorophyll for a more comprehensive three-dimensional view of ocean boundaries.

Summary updated

Abstract

Existing ocean front detection methods--including histogram-based variance analysis, Lyapunov exponent, gradient thresholding, and machine learning--suffer from critical limitations: discontinuous outputs, over-detection, reliance on single-threshold decisions, and lack of open-source implementations. To address these challenges, this paper proposes the Bayesian Front Detection and Tracking framework with Metric Space Analysis (BFDT-MSA). The framework introduces three innovations: (1) a Bayesian decision mechanism that integrates gradient priors and field operators to eliminate manual threshold sensitivity; (2) morphological refinement algorithms for merging fragmented fronts, deleting spurious rings, and thinning frontal zones to pixel-level accuracy; and (3) a novel metric space definition for temporal front tracking, enabling systematic analysis of front evolution. Validated on global SST data (2022--2024), BFDT-MSA reduces over-detection by $73\%$ compared to histogram-based methods while achieving superior intensity ($0.16^\circ$C/km), continuity, and spatiotemporal coherence. The open-source release bridges a critical gap in reproducible oceanographic research.
BibTeX
@misc{wang2025ocean,
  author = {Yishuo Wang and Feng Zhou and Qicheng Meng and Muping Zhou and Zhijun Hu and Chengqing Zhang and Tianhao Zhao},
  title = {{An ocean front detection and tracking algorithm}},
  year = {2025},
  eprint = {2502.15250},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2502.15250},
}
Figure