ANSS/NSMP STRONG-MOTION RECORD PROCESSING AND PROCEDURES

Size: px

Start display at page:

Download "ANSS/NSMP STRONG-MOTION RECORD PROCESSING AND PROCEDURES"

Gladys Richardson
6 years ago
Views:

1 ANSS/NSMP STRONG-MOTION RECORD PROCESSING AND PROCEDURES CHRISTOPHER D. STEPHENS AND DAVID M. BOORE U.S Geological Survey 345 Middlefield Road, MS 977 Menlo Park, CA The USGS National Strong Motion Program (NSMP), in cooperation with 17 partners, currently operates a permanent network of more than 675 digital and 265 analog strong motion instruments. These recorders, each with 3 to 18 channels of data, are located at 673 sites, including ground motion reference sites and structural arrays, distributed throughout 32 states and Puerto Rico. About half of these sites are in California. More than 320 of the digital instruments have continuous or dialup telecommunications. Earthquake records obtained from these instruments are collected in Menlo Park for archiving, processing and dissemination. Data obtained in near-real time are automatically processed for preliminary estimates of parametric values (e.g., peak ground motions) that are fed to other applications, such as ShakeMap, and for posting to websites such as the CISN Engineering Data Center ( but all of the records are reviewed and re-processed before final release. This paper summarizes the routine, record processing procedures currently practiced by NSMP. The basic processing steps (e.g., Converse and Brady [1]) for acceleration time series obtained by digitizing analog records or from digital instruments are: baseline removal, conditioning and padding of the ends of the acceleration time series for subsequent filtering operations, instrument response correction and acausal band pass filtering of the acceleration, integration of the corrected acceleration without decimation to velocity and displacement, and computation of response spectra. This relatively simple scheme ensures compatibility among the time series and the spectra when the pads are removed, as they are when the data are served from the NSMP website ( DIGITIZATION OF ANALOG RECORDS Analog records are typically digitized in grayscale images at a density of 600 dpi (about 236 sample/sec) on a desktop flatbed scanner in segments of about 28 cm (or about 28 sec at a typical film transport rate of 1 cm/s). A photo-enhancement program is used to optimize the brightness and contrast (no smoothing) of the grayscale images before they are converted to black-andwhite images for input into a semi-automatic trace-following program. Longer records are digitized in segments that are later joined automatically by a computer program using manually identified registration marks. For output, the digitized time series are interpolated to 200 sps. For multi-segment acceleration traces produced by this program it is often found that at the segment joins there are small offsets and tilts between the segment baselines; these are identified from spikes that appear in the first-difference of the time series which usually mark the onset of an abrupt change in the trend of the integrated time series. In such cases, the adjustments necessary to improve trace alignment are determined using robust L1 fits to each segment.

2 Fixed traces on the analog records are used as reference baselines to correct for lateral translations of the films that may have occurred while being transported during recording. In the process of digitizing a single segment of an analog record, small systematic offsets or distortions in the reference baseline can be introduced by a variety of factors. One that is commonly observed, for example, is a wandering of the fixed reference trace across one or more rows of pixels in the image, which can be due to a variety of factors that include small angular misalignments of the film with respect to the scan bed, or to distortions in the alignment of the pixel elements related to the scanning process. An estimate of the overall noise associated with the digitizing process is made routinely by comparing the Fourier amplitude spectrum (FAS) of a second fixed trace, digitized as if it were a signal, to the FAS of the data channels. BASELINE CORRECTION An initial reference baseline is determined from the mean of a suitable portion of the acceleration record, either the interval prior to the P-wave or, if there is no pre-event, the mean of the entire record. After making the initial baseline correction by subtracting this mean, the record is integrated, without filtering, to velocity to check for long-period drifts that may indicate the presence of offsets in the reference baseline, sometimes not always obvious in the acceleration time series. The cause of these offsets is often uncertain, but can include contamination from tilt or rotational components of ground motion (e.g. Grazier [2]; Trifunac and Todorovska [3]), or systematic effects, such as mechanical or electrical hysteresis in the transducer (Iwan el al [4]; Shakal and Peterson [5]), cross-axis effects (Wong and Trifunac [6]; Todorovska[7]), and the analog-to-digital process (e.g., Boore [8]). The character of such offsets can range from apparent simple step-like offsets to complex, time-dependent variations, and records may contain multiple offsets (Figure 1).

3 Figure 1. Examples of baseline shifts in earthquake records for both strong and weak motions. In each case, a baseline correction is made to the acceleration using the mean of the pre-earthquake signal, and the record is then integrated to velocity. The baseline shifts are evident as abrupt changes in the long-term linear trend in velocity. Baseline offsets occur on both components of HEC, but the one on the N-S trace is much smaller. Note that at least two offsets are present in the record from Olive Dell Ranch for the ML 4.4 event on 02/21/2000. In most cases, baseline offsets are small, and simply filtering the data can effectively remove this noise. In cases where the baseline offsets are considered large enough to severely contaminate the signal at long periods, corrections are made that involve piecewise fitting of linear or loworder polynomials to the velocity, and then subtracting the derivative of these fits from the acceleration time series (e.g. Iwan et al [4], Boore et al [9]). Details of the correction procedures are documented in the headers of the corrected data files. FILTERING For routine processing, acausal filtering is preferred over causal filtering. One reason for this preference is that at periods much shorter than the corner frequency the waveforms, and hence the response spectra, tend to be less sensitive to the low-cut-frequency corner (Figure 2).

4 Figure 2. Effects of causal (left) and acausal (right) filtering on the 5%-damped elastic (top) and inelastic (bottom) displacement response spectrum for the recordings at station Littlerock, 158 km west of the M 7.2 Hector Mine earthquake in Four corner periods are considered for each case. Note that at periods much shorter than the filter corner the responses are significantly less sensitive to the corner period for acausal filtering, particularly for the inelastic response. (From Boore and Akkar [10]). End Effects Each end of the record is conditioned for filtering by first searching inward from the end to find the first zero-crossing, then identifying the sample on either side of the zero-crossing which has the smallest absolute amplitude below a specified threshold, and finally setting to zero all of the samples outward from this point toward the end of the record. In this way, a small part of the

5 time series is sacrificed in order to avoid distorting any of the data with tapering. Note that inadequate corrections for baseline offsets, particularly at the ends of the time series where the signal amplitudes are usually low, can lead to excessive stripping of the data, or can produce ringing at the ends when the data are bandpass-filtered. Padding Prior to the low-cut-frequency filtering, zero-pads are added symmetrically to both ends of the records in order to accommodate the filter transients. The length of the pad at each end, t pad, is calculated using the following empirically determined formula: t pad = 1.5 * nroll / f c,, where nroll is the rolloff of the acausal Butterworth filter and f c is the frequency in Hz of the low-cut-frequency filter corner. The velocity and displacement time series derived by multiple integrations of the filtered acceleration trace are reviewed to ensure that there is adequate padding such that the filter transients decay to near zero at the ends of padded time series; a longterm trend away from zero at the end of the velocity or displacement time series, for example, indicates that there may be insufficient padding (Figure 3).

6 Figure 3. Illustration of the effects on velocity and displacement computed from acceleration that is filtered acausally without (left) and with (right) zero-pads pads. A high cut frequency filter is applied in the spectral domain in the form of a cosine taper over a specified frequency range. The instrument response correction is also simultaneously made at this step. Low-cut-frequency filtering is then applied to the acceleration traces in the time domain using a bi-directional Butterworth filter. The rolloff of the filter is specified by the parameter nroll (filter order = 4*nroll, or 24db/octave for nroll = 1). Selecting the value of nroll is a compromise between effectively removing long-period noise and avoiding the excessive ringing that can be

7 introduced by using higher order filters due to the sharper filter corners (Figure 4). For routine processing a value of nroll=2 is used. Figure 4. Effects of filter order on acceleration, velocity and displacement time series computed using acausal filters. The two case shown are for nroll = 1 and 4. Selection of filter corners The process of selecting filter corners is somewhat subjective, and factors that are considered in addition to the character of the recorded signal include the event magnitude and source

8 mechanism, recording distance, wave propagation path, and for records from structures, the fundamental period of the structure. For analog records, as previously indicated, an estimate of the digitization noise is usually made by comparing the spectra of individual data channels to the spectrum of a second digitized fixed trace. The filter corners are selected to be at a frequency where the level of the FAS of the data channel is about a factor of 2 above the noise, or at high frequencies where the spectrum tends to flatten or begins to rise. Triggered records from digital instruments are more problematic in that the duration of the recorded pre-event interval is often too short to obtain a useful estimate of the noise, particularly at long periods. Most commonly the filter corners are estimated from the FAS of the entire acceleration record (after correcting for large baseline offsets in order to reduce the noise introduced at low frequencies by these distortions). Starting at low frequencies, the noise is assumed to fall off roughly as 1/f towards higher frequencies, whereas the signal is generally assumed to fall off between f 2 and f towards low frequencies; the low-cut filter corner is usually estimated to lie near the intersection of these two trends, and at a frequency where the inferred level of the FAS is about a factor of 2 above the extrapolated noise spectrum. For large earthquakes, however, the possibility that the signal spectrum may be more complex than a single-corner model (e.g., Atkinson and Silva [11]) should also be considered, and thus a lower corner frequency may be required (e.g., Figure 5). At higher frequencies the signal is assumed to fall off approximately as 1/f 2, and the high frequency corner is selected to be at the point where the FAS spectrum flattens or begins to increase in amplitude with increasing frequency, which usually occurs near the natural frequency of the sensor (most commonly near 50 Hz). In some situations selection of the low-cut filter corner may be influenced by other factors, such the need to capture building responses at lower frequencies, or to remove strong contamination of the signal at high frequencies (such as from 60 Hz line noise).

9 Figure 5a. Variations in the Fourier amplitude spectra computed for acausallyfiltered acceleration from the Loma Prieta earthquake (Mw 6.9) recorded on the 333 o component at Anderson Dam downstream using different corner frequencies. A line showing an f 2 falloff at periods below about 1 Hz is shown for reference. Note the bulge in the amplitude of the FAS between about Hz, which, considering the size of this earthquake, is likely part of the signal.

10 Figure 5b. Acceleration, velocity, and displacement time series for the Anderson Dam downstream record using various low-cut filter corners. If the filter corner had been selected on the basis of the upward deflection of the FAS below 0.3 Hz where it deviates from an f 2 falloff, then the large ~0.1 Hz pulse in displacement beginning at 30 s would be lost. After velocity and displacement are obtained by multiple integrations of the corrected acceleration in the time domain, the selection of the filter corners is evaluated using various qualitative measures, such as the coherence of long-period signals at nearby stations (Figure 6) or on similarly-oriented components within structures, and the absence of unusual waveforms, including drifts in the displacement time series indicative of pads being too short to adequately accommodate the filter transients.

11 Figure 6. Comparison of the processed velocity and displacement traces from the N-S components of two stations located about 1.6 km apart and about 160 km west of the Mw 7.2 Hector Mine earthquake in Note the remarkable similarity between the signals at these two stations, particularly at long periods. Comparisons such as this are used to confirm that proper filter corners were selected to process the records. In routine processing, uniform filtering (the same filter corner and rolloff parameter) is applied to all of the channels from a particular site. One of the disadvantages of this choice is that important signal at longer periods may be lost, such as when the polarity of a seismic wave is aligned preferentially with the orientation of a particular sensor, or when a structure responds more strongly in one direction, because the component with the weakest signal (usually the vertical) controls the choice of the filter corner. However, uniform filtering allows motions that involve multiple channels, such as drift and torsion, which are important for engineering purposes, and rotation of ground motion components, to be computed directly from the corrected time series. REFERENCES 1. Converse, AM, Brady, AG. BAP: Basic Strong-Motion Accelerogram Processing Software; Version 1.0. US Geological Survey Open File Report A, 1992: 178 p. 2. Grazier, VM. On inertial seismometry. Izvestiya, Physics of the Solid Earth 1989: Trifunac MD, Todorovska MI. A note on the useable dynamic range of accelerographs recording translation. Earthquake Engineering and Structural Dynamics 2001:

12 4. Iwan, WD, Moser, MA, Peng, C-Y. Some observations on strong-motion earthquake measurement using a digital accelerograph. Bulletin of the Seismological Society of America 1985; Shakal, AF, Peterson, CD. Acceleration offsets in some FBA s during earthquake shaking. Seismological Research Letters 2001; Wong HL, Trifunac MD. Effects of cross-axis sensitivity and misalignment on the response of mechanical-optical accelerographs. Bulletin of the Seismological Society of America 1977: Todorovska MI. Cross-axis sensitivity of accelerographs with pendulum like transducers mathematical model and the inverse problem. Earthquake Engineering and Structural Dynamics 2003: Boore, DM. Analog-to-digital conversion as a source of drifts in displacements derived from digital recordings of ground motion. Bulletin of the Seismological Society of America 2003: Boore, DM, Stephens CD, Joyner, WB. Comments on baseline correction of digital strong-motion data: Examples from the 1999 Hector Mine, California, earthquake. Bulletin of the Seismological Society of America 2002; Boore, DM, Sakkar, S. Effect of causal and acausal filters on elastic and inelastic response spectra. Earthquake Engineering and Structural Dynamics 2003; Atkinson, GM, Silva, W. Stochastic modeling of California ground motions. Bulletin of the Seismological Society of America 2000;

LONG-PERIOD GROUND MOTIONS FROM DIGITAL ACCELERATION RECORDINGS: A NEW ERA IN ENGINEERING SEISMOLOGY

LONG-PERIOD GROUND MOTIONS FROM DIGITAL ACCELERATION RECORDINGS: A NEW ERA IN ENGINEERING SEISMOLOGY David M. Boore U.S. Geological Survey, MS 977, 345 Middlefield Road, Menlo Park, CA 94025 Abstract: