Effect of the Wavelet Scale Thresholding on the Geonet P-Phase Picker Performance for the borehole and surface stations URZ and EDRZ using the Matata Swarms of 2007-2010

Wednesday, 17 December 2014
Sepideh J Rastin1, Charles P Unsworth1 and Ken R Gledhill2, (1)University of Auckland, Auckland, New Zealand, (2)GNS Science, Wellington, New Zealand
Geonet is responsible for providing information on the geological hazard events including about 15,000 earthquakes recorded by the New Zealand seismograph network per year. Such large numbers of earthquakes confirm the importance of having an accurate P-phase picker in order to find automatic solutions for hypocenters and reduce analyst’s time needed to produce the reviewed catalogs. Geonet uses an Autoregressive-Akaike Information Criterion P-phase picker to estimate the P-phase onsets from the Butterworth filtered waveforms. We studied efficiency of the P-phase picker using waveforms of 3312 shallow earthquakes of the Matata swarms in 2007-2010. For stations EDRZ and URZ with large contributions to the Geonet locator, 48.80% and 32.88% of waveforms were reviewed by analysts. Such high rates of reviewed onsets indicated the need to improve the automatic phase picking’s performance. To do so, we replaced the Butterworth filtering with the Wavelet Scale Thresholding (WST) and divided the re-estimated onsets into the “unchanged”, “revised”, “retrieved” and “adversely affected” classes. We studied effect of the Mexican hat and Haar WST on the P-phase picker’s performance using quality control parameters including total and net improvements, onset corrections, picking errors and picking quality enhancement. Applying the Mexican hat WST mainly on scales 1 and 2 to the EDRZ recordings of chosen events resulted in large percentages for average total improvement (64.79%) and onset retrieval (31.95%) in 2007-2010. Efficiency of the Mexican hat WST at the URZ station was represented by small average retrieval of 2.76%, total and net improvement rates of 35.17% and 17.58% respectively. The small retrieval rate should be due the deep-borehole installation that filters the ambient noise and hence improves the quality of recordings. In spite of the reduced noise, path and site effects at the URZ station, the contribution of the Mexican hat WST to the quality enhancement of the picked onsets with the average rate of 26.28% during 2007-2010 indicated the efficiency of the alternative filtering scheme.