Research

I am a seismologist at the University of Southern California. My research focuses on the development and application of automated techniques to large seismic data sets for efficient extraction of seismological information. The amount of seismic data recorded worldwide is rapidly increasing due to the availability of inexpensive, high-quality sensors and unprecedented computating power and data storage. Historically, trained experts would process these data sets by hand, but at the present day this is generally no longer feasible. Thus, seismologists must rely on the aid of automatic techniques in performing these tasks. For my work, these techniques fall into three general categories: phases, source, and structure.

Automatic detection and picking of seismic phases

We developed robust automated techniques for picking of P and S body waves. The techniques require no prior knowledge about whether an earthquake has occurred and make phase detections directly on the continuous waveform data. Our S-wave picking algorithm, DBSHEAR (Ross & Ben-Zion, 2014; Ross et al. 2016), is capable of making S-wave picks that are within 0.25 sec 90% of the time for local-regional seismic networks. The method is currently employed by several permanent real-time seismic networks. Two versions of this software are available upon request: an Antelope-based program and a stand-alone python phase picking code for working with event data (e.g. SAC/MSEED files).

Recently, we developed an earthquake detection algorithm that is able to pick out microearthquakes from continuous data that were often missed. The algorithm allows for combining numerous detectors together using "pseudo-probabilities", resulting in significantly enhanced detection rates. The method has great flexibility in choosing the indicator variables to be used for detection, and provides a single, robust time series indicator variable as an output to replace classical detectors like STA/LTA.

Automatic estimation of earthquake source properties

We have developed different algorithms for estimating a variety of earthquake source properties from body-wave spectra. In particular, we developed a simple technique for reliable estimation of the seismic moment in small local earthquakes. This technique was used to calculate seismic moment and moment magnitudes for > 11,000 earthquakes with 0 < ML < 3.5 that occurred in the San Jacinto fault zone in 2013. We used the derived values to show that the b-value in the moment magnitude scale is 25% larger than in the ML scale, due to a deviation of the scales for ML < 3.5. We found that the average event with ML 0 has an Mw of 0.9.

We also developed a complete procedure for estimating stress drops, seismic energy, corner frequencies and ratios, apparent stress, and extent of directivity. The procedure uses spectral ratios and is fully automated. It is capable of being applied to earthquakes with M > 3.0. We applied the technique to many earthquakes in Southern California and found that a significant number of them had strong evidence for unilateral directivity.

My research has also focused on resolving signatures of volumetric changes during the source process. For example, both theoretical and observational studies have shown that tectonic earthquakes are capable of fracturing the surrounding rocks during the source process. The fracture of brittle fault zone rocks leads to sudden changes of the elastic moduli, and a rapid increase of the source volume in addition to the shear deformation on the fault itself. The brittle fracture process is expected to produce small amounts of "damage-emitted radiation", resulting from the sudden volume change. This damage-emitted radiation is expected to contain a significant isotropic component and be much higher in frequency content than the radiation from shear dislocations. In practice, routine derivations of earthquake source properties constrain the solutions to allow no volume change, which is appropriate under many circumstances in seismology. The presence of even small volumetric source components, however, has fundamental implications for the local physics (such as assumptions of constant normal stress during rupture). I am investigating the possibility of isotropic source components in standard, tectonic earthquakes from a number of different perspectives including:

Automatic identification of fault zone generated phases

Large regional strike-slip faults often are the interface between two different lithologies with a significant velocity contrast. Earthquakes produced on such faults can have fault zone head waves (FZHW) as the first arriving phase on a seismogram, instead of the direct P-wave, for stations close to the fault. These phases are similar to Pn phases refracting along the Moho, but have opposite first motion-polarity from the direct P-wave. Since they propagate along the fault interface, they provide the highest resolution information of fault zone structures. Neglecting FZHW can introduce significant bias in earthquake locations, focal mechanisms, and other derived quantities since they are routinely mistaken for P-waves at near-fault stations. We have developed an automatic technique capable of distinguishing between P and FZHW, and accurately picking the arrival times for both phases. This allows for the automatic processing of large seismic data sets.

Large fault systems typically accumulate rock damage with each seismic cycle, leading to low velocity fault zone layers that can act as a waveguide. These damage zones can produce fault zone trapped waves (FZTW), associated with Love-type resonance modes. They are longer period and have usually higher amplitude than the direct S-wave. FZTW provide critical high-resolution information on the internal structure of fault zones. We have developed a method for robust, automated identification of FZTW recorded by dense arrays deployed around fault zones. The method uses the automatic S-picking algorithm of Ross and Ben-Zion (2014) to identify time windows to search for features of FZTW, and identifies recordings that have features which are statistical outliers with respect to the rest of the array. No prior information is required about which stations (if any) are located inside of a low velocity damage zone.