Direct measurement of fault rupture using seismic dense arrays: method and application to the Alpine Fault, New Zealand and SMART-1, Taiwan data

Abstract

The Alpine Fault is the major geological feature in New Zealand. It is a dextral transform fault separating the Pacific plate on the east from the Australian plate on the west, crossing the South Island from Northeast to Southwest (Figure 1- 1). It has an average slip rate of 40 mm per year and is the longest fault in New Zealand with a length, on land, of 650 kIn (Yetton, 2000).

The Alpine Fault has been a source of strong earthquakes in the past A recent field study led by Mark Yetton (2000) shows that the last rupture event occurred in 1717 AD; a minimum rupture length of 375 km extending as far North as the Haupiri River was estimated a moment magnitude of 8.05(+1-0.15} An earlier event, dated between 1480 and 1645, is noticeable throughout a region from the Hokitika River to the Ahaura River. This seems to be the most recent event North of the Haupiri River, with a rupture length of at least 200 km and a moment magnitude estimated to have been as great as 7.8(+-0.1).

The Alpine Fault is a potential source of major earthquakes in the near future. The return period of the fault is approximately 270 years, with no major event occurring over the last 285 years. Yetton applied various probabilistic methods to assess the time of the next major event on the Alpine Fault. Each method converged to give a likelihood of at least 50% of having a major earthquake in the next 50 years. Using the slip rate and the elapsed time since the last event, Yetton estimates a moment magnitude between 7.5 and 8.4. Using the method of Coppersmith et al. (Wells et al., 1994), relating moment magnitudes to the rupture length of the fault, Yetton estimates the two previous events to be of magnitude 8.05 and 8.0. This allows us to assign the expected next event a moment magnitude of at least 8.0. The high level of probability of a major fault rupture in the region provides an ideal opportunity to study directly the process of fault rupture.

Common methods of studying the fault rupture processes involve the inversion of seismic waveforms to fit a proposed fault rupture model. These methods use various computations such as the genetic algorithm applied to strong motion and GPS observations (Zeng et al, 2001), waveform inversion of broadband teleseismic data (Olivieri et al, 1999), waveform inversion of multiple time window strong motion data (Sekiguchi et al., 2002), or broadband P wave inversion (Zobin et al., 2001). All these inversion processes make assumptions about the fault rupture model, and therefore depend on our current knowledge and assumptions of the fault rupture mechanism.

Using a dense array allows one to study the rupture directly without assuming any model. With a dense array, a frequency-wavenumber spectrum analysis can be computed, the spectra projected back onto their source on the fault plane thus giving an image of the source. With a direct image of the rupture process, one can estimate basic source parameters such as rupture velocity, direction of rupture, and possibly the position and extent of asperities. Because dense array analysis does not make any assumptions about the rupture mechanism, it is a powerful tool in understanding the physics of the source.

The method used in this study to carry out dense array analysis applies the mathematical algorithm MUSIC (Multiple Signal Characterization Method, Schmidt 1981, 1986). MUSIC was chosen for its ability to resolve seismic signals with low signal-to-noise ratios. In seismology, earthquake sources are not repetitive and signals are recorded at a limited number of stations. Previous studies have shown the advantages of using MUSIC for seismic signals over other methods: Goldstein (1988) strong motion array study, and Schissele (2002) who applied MUSIC to a broadband seismic antenna. MUSIC performs a frequency slowness analysis over a collection of seismograms. It uses the covariance matrix of the signals to extract information about the rays that propagate through the array. MUSIC looks for the true signal by searching for signals that have a minimum projection in the noise subspace. Results are presented in the form of a 2D slowness spectrum. Careful programming of the algorithm, thorough preparation of the data and accurate determination of time windows are the keys in obtaining reliable results. An innovative way to determine optimal time windows containing only one source is introduced in this study. The MUSIC algorithm as well as the other complementary processes was reprogrammed from scratch using Matlab.

In order to assess the efficiency of dense array analysis, synthetic data were generated for two rupture scenarios with and without an asperity. The synthetic strong-motion records were computed using an empirical Green's function synthetic seismogram program EMPSYN (Hutchings, 1987). EMPSYN allows one to create synthetic strong motion earthquakes with complex geology, as well as, allows the user to choose from a variety of rupture scenarios. Dense array analyses of the vertical components for the two proposed rupture models are presented. They are compared to the original rupture models in order to discuss and validate the method.

Finally, dense array analysis is applied to a real dataset recorded at the dense SMART-1 array in Taiwan. The studied event is Event 5, a magnitude 5.9 earthquake that occurred in 1981. Event 5 is reprocessed applying dense array analysis to the three components of the recordings. Important rupture parameters such as the direction of propagation, velocity and rupture area are directly measured. This analysis of a real dataset provides interesting observations in regards to the fault rupture mechanisms.