Improving the performance of earthquake detection in New Zealand with wavelets and ambient noise models

Abstract

More than 14000 local and regional earthquakes are catalogued automatically using the New Zealand GeoNet facilities each year. Thus, improving the quality and accuracy of automatic estimations of event locations and magnitudes are crucial for reducing the amount of manual analyses required to refine the automatic solutions. The motivation of this research is to evaluate and improve the current GeoNet automatic signal processing methods that are applied to the recordings of the New Zealand Seismograph Network (NZSN). The thesis presented addresses issues along this theme making three main scientific contributions.

The first scientific contribution is using the Mode Low Noise Models (MLNM) to represent the ambient noise for the NZSN of the North Island by analysis recordings five years (2005–2009). The MLNMs are used to evaluate sensor functionality, installation quality and to characterize ambient noise at each seismographic station. The NZSN long-term noise baselines can be used to prioritize the maintenance issues and to estimate earthquake detection capability.

The second scientific contribution is the development and validation of an accurate method to evaluate the performance of the GeoNet pickers using real and synthetic seismograms of Matata earthquakes that occurred in 2008. We quantify the effect of radiation pattern and noise on the pickers’ performance which allows us to identify optimized locations for the seismographic stations and to provide efficient seismic signal processing schemes.

The final contribution of the thesis is in the area of Wavelet Scale Thresholding (WST). We demonstrate when a suitably designed WST scheme is substituted for the current filtering scheme, enhancements in detection, accuracy and quality of automatic P-phase onsets can be achieved. We also investigate how the P‐phase picker’s performance behaves both spatially and temporally using 6471 waveforms from 3312 Matata earthquakes for a 4-year period (2007-2010). We demonstrate that WST provides superior time-frequency localization improving the detection capability and noise-signal modeling quality for 45% of the waveforms.

It is hoped that the findings disseminated in this thesis will be of benefit to future researchers in the development of the NZSN and advanced signal processing implementations which improve the efficiency of the current GeoNet facility.