Improved computational tools for infrasound analysis

Abstract

Infrasound describes low frequency (≤ 20 Hz) acoustic waves that can propagate long ranges (≥ 1000 km) through wave guides in the atmosphere. This characteristic makes infrasonic waves a useful monitoring technology for a variety of violent phenomena such as volcanic eruptions. However, source processes may be complex, and infrasound waves are continually modified as they interact with the ground and dynamic atmosphere along their propagation path. Additionally, infrasound still has a data quantity problem: vast amounts of raw data are now continuously generated from permanent arrays distributed around the world, but ground truth information, when possible, may be difficult to obtain. Globally-recorded events are more rare still. Under these conditions, computationally intensive approaches are an increasingly necessary tool to further exploit the information contained in infrasonic waveforms. This dissertation focuses on advanced computational approaches to infrasound analysis. Infrasound arrays may be located in remote environments, and an in-situ indicator of data quality would be useful to ensure a properly working array. Assuming that acoustic signals traverse the array as a plane wave, we document how some elements in both International Monitoring System and Alaska Volcano Observatory arrays produce outliers in inter-element travel time. These outliers, due to timing errors or other issues that cause an apparent deviation from plane wave behavior, produce inaccurate plane wave parameters (back-azimuth and trace velocity) when processed with conventional least squares time-domain array processing. In Chapter 2, we investigate how robust statistical regression methods, particularly least trimmed squares, M-estimation, and L1-norm regression, perform for time-domain infrasound array processing. Least trimmed squares processing returns accurate values across a variety of synthetic tests by using a subset of element pairs to estimate optimal back-azimuth and trace velocity values. By examining the element pairs not included in the subset, we find that the element producing outlying travel times can be identified and removed. We proceed to show how least trimmed squares processing improves infrasound array processing results at arrays I53US, I55US, and ADKI. We investigate the effect of terrain on infrasound propagation in Chapter 3. However, here our emphasis is on finite-frequency effects, specifically diffraction, on propagation ranges longer than 100 km. Simulations in the geometric acoustics approximation have shown that realistic terrain can reflect acoustic waves into shadow zones and scatter acoustic energy from tropospheric ducts. However, finite-frequency effects such as partial reflection and diffraction are not modeled under this approximation. We develop a finite-difference timedomain method to simulate linearized, inviscid, Euler equations for infrasound propagation. We first compare our finite-difference results with ray predictions with both flat terrain and a Gaussian hill at different ranges. We note an extended spatial footprint on the ground for our finite-frequency method compared to ray tracing, and evidence of partial reflections from the tropospheric duct. We build on these findings to investigate infrasound recordings from an explosion at the Utah Testing and Training Range. We examine recordings of this 2012 explosion on two infrasound arrays, NOQ and WMU, which are located at approximately 84 km and 148 km from the source respectively. Evidence from array processing suggests propagation paths through the troposphere, but no eigenrays were identified due to the weak tropospheric ducting conditions at the time of the explosion. We predict infrasonic signals at these arrays with our finite-difference method which show qualitative matches in waveform shape. Moreover, we track changes in waveform shape from source to receiver due to diffraction over terrain along the propagation path. Our results suggest that geometric acoustics underestimates acoustic arrivals through the troposphere, and that terrain along the propagation path affects waveform shape at distances greater than 100 km. As noted above, propagating acoustic waves frequently interact with the ground as they travel over sometimes complex topography. As part of this interaction, infrasound waves are commonly recorded to couple into the ground and are recorded on seismometers. Acoustic to seismic coupling is not commonly considered in simulations of infrasound propagation. In Chapter 4, we quantify the amount of acoustic to seismic coupling that occurs over both flat topography and meshed, complex, topography using a spectral element method. In the course of this research, we also derive expressions relating a seismic moment tensor to an acoustic quadrupole as well as conditions for elastic particle motion from the ground coupled airwave to switch from retrograde to prograde at the surface of an elastic halfspace. Using a suite of Earth models that span a range of specific acoustic impedances, we find a wide variety of energy admittances as a function of incidence angle (≤ 1% to ≈ 78%). However, in simulations over the complex terrain of Sakurajima Volcano, we find that the effect of coupling reduces peak acoustic amplitudes over a 15 km distance from the volcano by ≤ 2%. While this value is relatively small, the cumulative effect over long ranges, and multiple acoustic bounce points, may be nontrivial.