I am a third-year Ph.D. student working with Eric Kiser in the Global Seismology and Tectonics (GSAT) group at the University of Arizona. I investigate the relationships between earthquake source properties, physical mechanisms that control slip on faults, and social hazards. Our primary tool for investigating the source properties of large earthquakes is the back-projection method, for which we have developed a new genetic algorithm-based approach that selects an optimal subset of seismic stations that reduces artifacts and increases the resolution of back-projection results (see Projects).
2018 MW 6.9 Hawaiʻi Earthquake
Large volumes of volcanic output and high rates of seismicity occurred near the south flank of Kīlauea beginning in April 2018. Associated with this activity was a MW 6.9 earthquake on 4 May 2018. Using a novel genetic algorithm-based back-projection technique, the complex rupture properties of this event are imaged. The dominant feature of this earthquake is a slow western-propagating rupture that overlaps with the location of previous slow slip events. The rupture properties imaged in this study can be explained by slip on a décollement composed of soft sediments with small velocity-weakening asperities.
In the back-projection result shown above, the normalized waveform coherence of the 2018 MW 6.9 Hawaiʻi mainshock is plotted as a function of time (hypocentral time is 0 s). The white line is the coastline of Hawaiʻi and the black lines are faults. The white dots indicate the maximum coherence value location for each 1 s time step.
Read more: https://doi.org/10.1029/2018GL080397
Genetic Algorithm-based Back-Projection Method
Back-projection studies typically use a single dense, teleseismic array to image the rupture properties of a large earthquake. Waveforms recorded across these types of arrays are generally coherent, assisting the back-projection process and reducing artifacts in the source image. However, the small aperture of a single teleseismic array limits the overall resolution of the back-projection result. Some studies have introduced larger (e.g., global) distributions of seismic stations for the purpose of increasing array aperture and enhancing back-projection resolution. However, this approach samples additional earth structure and reduces waveform coherence across the array, resulting in artifacts in the back-projection result. One way to mitigate this problem is to remove the stations responsible for producing these artifacts. This can be accomplished by visual inspection or by introducing a measure of waveform similarity across the array (i.e., coherence values), but these techniques are not always successful.
Instead, a genetic algorithm can be used to determine an optimal distribution of stations for back-projection. The key idea behind this approach is that small earthquakes should be imaged by the back-projection method as point sources and any additional imaged energy is likely an artifact. For a seismic network of M stations, an optimal back-projection network that maximizes the point source nature of a small earthquake can be found in 2M attempts. A genetic algorithm is used to reduce this number of attempts and find a near-optimal solution. This solution is then applied to the mainshock back-projection analysis, providing a comparable reduction in imaged artifacts.
The genetic algorithm is an iterative process (see below). N potential solutions of length M binary strings that represent a subset of stations are assigned a value according to a fitness function. Subsets of stations are identified proportionally to their fitness (roulette wheel selection) and are used to form a new set of N subsets of stations (crossover). These subsets of stations are, on average, better potential solutions to the problem. Random individuals undergo random changes to increase diversity and search more of the solution space (mutation). This cycle is repeated until a termination condition is reached.
Department of Geosciences
University of Arizona
1040 E. 4th Street
Tucson, AZ 85721