Probabilistic inversion for submerged source depth and strength from infrasound observations

In seismology, the depth of a near-surface source is hard to estimate in the absence of local stations. The depth-yield trade-off leads to significant uncertainties in the source's depth and strength estimations. Long-range infrasound propagation from an underwater or underground source is very sensitive to variations in the source's depth and strength. This characteristic is employed in an infrasound based inversion for the submerged source parameters. First, a Bayesian inversion scheme is tested under the variations of the number of stations, the signal's frequency band, and the signal-to-noise ratio (SNR). Second, an ensemble of realistic perturbed atmospheric profiles is used to investigate the effect of atmospheric uncertainties on the inversion results. Results show that long-range infrasound signals can be used to estimate the depth and strength of an underwater source. Using a broadband signal proved to be a fundamental element to obtain the real source parameters, whereas the SNR was secondary. Multiple station inversions perform better than one-station inversions; however, variations in their position can lead to source strength estimations with uncertainties up to 50%. Regardless of the number of stations, their positions, and SNRs, all of the estimated depths were within 10% from the real source depth.