Probabilistic acoustic localization method (PALM): A statistical framework for infrasound localization using ray-tracing

As part of the Comprehensive Nuclear-Test Ban Treaty, there is an interest in the capacity to localize infrasound sources on a global scale. As such, it is necessary to consider the effects of the atmospheric conditions on infrasound propagation to accurately predict the source location using direction of arrival observations from several infrasound stations. Geometric ray-tracing is used to predict infrasound trajectories using a realistic atmosphere. However, the presence of shadow zones, where the geometric rays do not propagate, results in blind zones using this method. To help fill in these shadow zones, we propose to use kernel density estimation as part of a Bayesian framework to perform a robust, global infrasound localization scheme referred to as the probabilistic acoustic localization method (PALM). We studied several ground-truth events on both regional (hundreds of kilometers) and global (thousands of kilometers) scales to demonstrate the robustness of PALM. We show that PALM accurately predicts these event locations and times. Given an atmospheric model, we demonstrate improved uncertainties relative to other state-of-the-art infrasound localization methods as well as robustness against atmospheric uncertainties provided by an ensemble spread. These conclusions remain conditional on the chosen atmospheric model.