4. Attenuation parameters for regional earthquakes in the northern Indian Ocean –
4.2 Database sampling and processing procedure
Geological Survey and Mines Bureau in Sri Lanka, in collaboration with USGS and German Research Centre for Geosciences (also known as GFZ-GEOFON), is responsible for maintaining 3 broadband seismic stations located at Pallekele (PALK), Hakmana (HALK) and Mahakanadarawa (MALK) rock sites. While PALK station (with USGS) has been operating since 2000, the other two had been established in 2011 under GEOFON permanent network.
Digital waveforms of recorded events at the above stations are available to download from their respective data management system, i.e., IRIS (Incorporated Research Institutions for Seismology) data management centre for PALK and GFZ seismological data archive for HALK and MALK. 181 such records of 71 shallow crustal earthquakes magnitude ranging from 4.0 to 5.9 in short period body wave magnitude scale (mb), have been downloaded and processed for further analysis (Refer Figure 4.1 and for a detailed list of events refer Appendix B).
Figure 4.1 Recorded events (circles) used in the study for deriving the regional Q and assessing H/V ratio calculations (Sources; ISC-UK, GEOFON-Germany and ANSS-USA). Triangles denote broadband stations at Pallekele (PALK), Hakmana (HALK) and Mahakanadarawa (MALK).
61
Distribution of data in magnitude-hypocentral distance is presented as a dot plot in Figure 4.2.
Location and magnitude data of older events (events before 2010) which represent about 17% of the total are directly obtained from “Reviewed ISC bulletin”, whereas those of newer events (about 83% of the total) are taken from GEOFON and ANSS data archives depending on the availability of the event in the respective data archive. The data archives assign short period body wave magnitude as the common scale. Therefore this magnitude parameter is chosen for regression analysis to ensure consistency. Epicentral distance varies from 2.30 (260 km) to 14.10 (1578 km) for the whole dataset. The database consists of a large number of records from aftershocks of the aforementioned Mw 8.6 earthquake, and this in turn strengthen the dataset by increasing the original sample size. Magnitude and location uncertainties, if any, are omitted in all newer records even though it is presumed that an uncertainty associated with the distance alone may not produce any notable deviations in the final results in the case of long-distant earthquakes. After removing random outliers (discussed under “4.5 Results and discussion”) identified in the database, original records and events dropped down to 162 and 62, respectively.
Thus the quality of data sampling has been ensured.
Figure 4.2 Distribution of the dataset in magnitude (mb) and hypocentral distance space.
Magnitude ranges from mb 4.0 to 5.9, whereas hypocentral distance varies between 260 and 1578 km.
4.2.2 Processing procedure
Broadband High-Gain vertical component data (BHZ) are selected in each of the waveform record. At PALK rock site, since there are two seismometers installed at two different depths (one at 1 m and the other at 90 m depth), vertical components corresponding to upper one (at 1
62
m depth) are selected to be compatible with records at other two sites. Horizontal components are not considered in the regression analysis as they would be susceptible to inherent local site amplifications that can be quite significant (sometimes) at high frequencies and may result in extensive spatial variations. However, the horizontal motion has been taken into account in the H/V calculations as described in section 4.4 of the Chapter. Vertical motion is said to be less affected by near-source upper crustal modifications (in terms of attenuation and amplification) offsetting any decay in wave propagation velocity with refracted waves travelling in the vertical direction (Atkinson, 2004b). Selected vertical components are bandpass filtered between 0.1 – 9.0 Hz frequencies after applying tapering of 10% at both ends of the each trace. This also covers the range of natural periods of engineering significance (between 0.1 sec and 10 sec).
The maximum frequency 9.0 Hz is selected in such a way that the maximum frequency limit should be less than half of the sample rate of the instrument (Havskov and Ottemoller, 2012). In the current case they are 40, 20 and 20 sps (samples per second) for seismometers at PALK, HALK and MALK, respectively. Next, filtered traces are corrected for instrument effects to produce acceleration time histories. Fourier Acceleration Spectra (FAS) of each processed record are obtained by selecting the “shear/strong window” containing the multiple reflected and refracted S phases. FAS are visually inspected for signal strength relative to the background noise spectra and amplitudes of frequencies at 0.1, 0.2 and 0.3 Hz are left out owing to low frequency noise amplifications. FAS are smoothed for the selected whole frequency range at 0.1 log frequency bins by introducing a box/rectangular weighting function and thus taking the geometric average of amplitudes in each frequency bin. This frequency dependent smoothing (due to constant width in log frequency) allows uniform smoothing of amplitudes over the entire frequency range in the log frequency scale and is considered to be more desirable for regression analysis due to reductions in noise. Box weighting smoothing has been shown to produce little scatter at low frequencies with respect to other smoothing options (such as triangular weighting) although Boore (2012) notes the effect can be negligible in the final results. Acceleration amplitudes of smoothed spectra in each vertical record are then tabulated in 0.1 frequency increments for the selected frequency range. A processed sample accelerogram with respective FAS are shown in Figure 4.3.
63
Figure 4.3 Data processing: (a) A sample acceleration time history produced after filtering (between 0.1-9.0 Hz), tapering (10% at both ends) and correcting for instrument. Main phases along with the shear/strong window to be used for Fourier Acceleration Spectra (FAS) are also shown. (b) FAS of the record in (a). Dark line represents the smoothed spectrum (smoothing method is described in the text).