Seismic detection using submarine cable polarization signals with machine learning

Submarine fiber-optic communication cables are emerging as potential platforms for large-scale geophysical sensing. This study evaluates the feasibility of exploiting the state of polarization of optical signals in the live Med-Nautilus submarine cable for earthquake detection. We analyze state of polarization time series collected between 2022 and 2024 to assess whether seismic events induce detectable perturbations. Supervised and unsupervised Machine Learning models, including Logistic Regression, Extreme Gradient Boosting, and Deep Autoencoders, are applied for classification and anomaly detection. Results indicate that moderate-to-large earthquakes (magnitude ≥ 5) generate measurable state of polarization anomalies, although the cable was not designed for sensing. The Extreme Gradient Boosting model achieves up to 60% accuracy in distinguishing seismic signals from background noise. These findings provide the first empirical validation of state of polarization-based seismic detection under real operating conditions and support the potential of existing submarine cable networks as scalable, cost-effective geophysical monitoring systems.