Do Polar Lows Leave an Infrasonic Signature? A Machine Learning Investigation

Infrasound monitoring within the Comprehensive Nuclear-Test-Ban Treaty (CTBT) framework requires distinguishing between diverse natural and anthropogenic sources. Polar lows–intense mesoscale cyclones developing over high-latitude oceans–have been suggested as potential infrasound sources, although their acoustic signature remains poorly constrained. Here, we investigate whether polar lows leave a detectable infrasonic signature using data from the IS37 station in northern Norway (2014–2019). A supervised machine learning approach is applied to classify detections associated with polar lows based on spatio-temporal and directional matching with a cyclone database. Logistic Regression and Random Forest models are used to assess classification performance and identify controlling factors. Results show that polar-low-related detections cannot be robustly separated from background infrasound. High-confidence detections (90% precision) are obtained only at the expense of a recall below 1%, indicating that most signals remain undetected. Feature importance highlights the dominant role of environmental variables, with the 2-m air temperature accounting for approximately 40% of the total importance. Physical analyses indicate that convective activity is not aligned with inferred source directions and shows weak relationships with signal properties. In contrast, swell-related diagnostics exhibit strong correlations, with the strongest relationship observed between the mean significant wave height and the RMS amplitude (r = 0.64), supporting an ocean-wave origin consistent with microbaroms. Atmospheric propagation further acts as a dominant filter controlling detectability. Polar lows therefore do not act as direct acoustic sources but indirectly modulate infrasound generation and detection through their influence on ocean waves and atmospheric structure.