Infrasonic backprojection with the EarthScope Transportable Array in Alaska: Improving detection and localization of explosive volcanism via noise reduction.

Abstract:

The current deployment of the EarthScope Transportable Array (TA) in Alaska affords an unprecedented opportunity to study explosive volcanic eruptions using a relatively dense regional seismo-acoustic network. Since 2016, seven volcanoes have erupted in Alaska with a range of magnitudes and styles. The 2016–2017 eruption of Bogoslof in particular produced seventy explosive eruptions. Such events provide a unique validation dataset to examine the ability of different network configurations and processing strategies to detect, locate, and characterize remote volcanic eruptions in Alaska. When incorporating the TA and regional infrasound stations, a simple envelope-based backprojection technique, and automated event identification process, is able to capture up to 85% of the infrasound generating events from Bogoslof that were cataloged by the Alaska Volcano Observatory (Sanderson et al., 2020). Notable limitations to the scheme come from anisotropic atmospheric propagation, and low signal-to-noise (SNR) conditions, with wind being the dominant noise source. Here we focus on improving waveform data processing to reduce noise and so enhance signal detection and source location accuracy during backprojection. Additional evaluation metrics include degree of SNR improvement, any signal distortion, need for manual data labeling, and computational cost. The latter point is particularly relevant to real-time data processing applications. We compare a range of recently developed techniques that separate signals and noise in the spectral domain, including within the same frequency band. Examples include block thresholding, non-negative matrix factorization, and singular spectrum analysis. Such noise reduction on individual waveforms (i.e., pre-stack processing) can improve many basic analyses, given that SNR enhancements do not rely on combining traces (or stacking of other functions; i.e., co-stack strategies) in order to improve coherence and isolate events. We further compare results of the pre- and co-stack methods (e.g., phase-weighting, adaptive F-statistic time-series), as well as combinations of both techniques. More broadly, trace-based noise reduction can increase the value of isolated infrasound sensors in places where an array, or spatial averaging equipment, is impractical.