- Ground-motion model is: where Z is in cm∕s2, a = 0.7401, b = -0.0098, c = 1.0273, d = -0.1099, e = 0.0071, si = 0.2841,
ss = 4.1648, ssl = -0.6199, x = 0.4998, qi = -0.0662, qs = -0.1807, ϕ = 0.3745, τ = 0.1568 and σtot =
- Use NEHRP site classes. Most are NEHRP B (rock).
- Classify events into two types:
- 374 records from 48 earthquakes
- 720 records from 90 earthquakes
based on geometry and stress regime in subduction zone. Finally classify events manually based on depth,
distance to trench, style-of-faulting and dip angle. Plot vertical cross sections to visualize downgoing Nazca
plate and to display positions of studied events w.r.t. slab. Hence remove crustal and spurious events.
Remove those that cannot be clearly classified.
- Functional form based on Zhao et al. (2006) but simplified because of characteristics of available database
and difficulty in obtaining stable coefficients.
- Data mainly from 20 stations in a dense network deployed as part of Integrated Plate Boundary
Observatory Chile (IPOC) supplemented with data from 8 triggered stations in area and 47 records (11
earthquakes) from 40 stations in other networks to extend model to higher magnitudes.
- For data from IPOC use a semi-automatic tool to extract records from continuous data streams for
earthquakes with Mw ≥ 5.
- IPOC data from 2006 to May 2012 and other data from 1966 to 2007.
- Generally prefer locations from local agency (Chilean Seismological Service) over those from teleseismic
- Only use those earthquakes with focal mechanisms from Global CMT so that all required information
available. Only use those earthquakes with Mw > 5 since because majority of models calibrated for large
- Use empirical relations between rupture length, width and Mw to estimate location of rupture plane and
hence compute rrup. For those events where fault plane is ambiguous use distance-conversion equations.
- Do not correct for instrument response because not required up to 100Hz. Apply zeroth-order baseline
correction. Subtract mean of pre-event portion of record. Integrate some of the records to check for
long-period drifts that would indicate changes in reference baseline. Generally baseline offsets small so
band/high-pass filter using acausal 4th-order Butterworth after zero padding. Choose cut-off frequencies
based on smoothed Fourier amplitude spectra and signal-to-noise ratios of 2, with a default of 0.05Hz if
signal-to-noise is always higher than 2. Reject those records with signal-to-noise ratio barely larger than 3
over whole frequency range. Randomly check velocity and displacement traces. Use same filter cut-offs for
all components with cut-offs chosen based on horizontal components.
- Most data from Mw < 6.5. Data well-distributed w.r.t. rrup.
- Plot residuals w.r.t. Mw and rrup and find no trends.