A systematic polar-induced signature in infrasound database highlithed by machine learning models

Polar lows are intense but short duration maritime cyclones occurring in both hemispheres. In the northern pole, they are mainly located in the Barents and Norwegian seas, with significant damages for coastal populations. So far, a fully understanding of the physical processes at play is still lacking. This is due to the suddenness of such events, as well as a scarcity of meteorological observations in these areas. Infrasounds are sound waves with frequency ranges below the audible domain. It was shown that polar lows can be a source of infrasound. Only one study looked at the infrasound signature for two particular polar lows using data obtained from two stations, in Northern Norway and on Svalbard. Here we show the potentiality of a systematic polar low-induced signature in infrasound data.

Within the frame of the Comprehensive nuclear-test-ban treaty organization, infrasound stations were set up worldwide. One was settled in northern Norway (IS37NO) in 2003 and made fully operational since 2004. Its records consist in a timeseries of sub-daily pressure data, that are processed through a Progressive Multi Channel Cross Correlation method, resulting in variables such as the mean frequencies, azimuths and amplitudes of the detections, and covering 17 complete years (2004-2021). These variables were used to train statistical models to learn the occurrence of polar lows refered in a polar low database. Our models yield very good results, specially in term of precision and recall. They provide a basis for different research opportunities, such as the prediction of polar lows and a deeper comprehension of its climate controls.