2.188 Herak et al. (2001)

• Ground-motion model is:

where amax is in g, for horizontal PGA c1 = -1.300 ± 0.192, c2 = 0.331 ± 0.040, c3 = -1.152 ± 0.099, c4 = 11.8 ± 4.8km and σ = 0.311 and for vertical PGA c1 = -1.518 ± 0.293, c2 = 0.302 ± 0.035, c3 = -1.061 ± 0.096, c4 = 11.0 ± 5.5 and σ = 0.313.

• Records from 39 sites. Records from instruments on ground floor or in basements of relatively small structures.
• Site information only available for a small portion of the recording sites and therefore is not considered. Believe that most sites are ‘rock’ or ‘stiff soil’.
• All records from Kinemetrics SMA-1s.
• Select records with ML 4.5 and D 200km because of poor reliability of SMA-1 records for small earthquakes and to avoid problems related to a possible change of geometrical spreading when surface waves start to dominate over body waves at large distances.
• Bandpass filter with passbands selected for which signal-to-noise ratio is > 1. Widest passband is 0.0725Hz.
• Do not use rjb because do not accurately know causative fault geometry for majority of events.
• Do not include an anelastic decay term because data is inadequate to independently determine geometric and anelastic coefficients.
• Note correlation between magnitude and distance in data distribution therefore use two-stage regression. Because many earthquakes have only a few records data is divided into classes based on magnitude (details not given).
• Most data from ML < 5.5, particularly data from D < 20km.
• Find all coefficients significantly different than 0 at levels exceeding 0.999.
• Also regress using one-stage method and find practically equal coefficients and larger standard errors.
• Find residuals are approximately lognormally distributed with slight asymmetry showing longer tail on positive side. Relate this to site amplification at some stations balanced by larger than expected number of slightly negative residuals.
• Find no distance or magnitude dependence of residuals.
• Compute ratio between larger and average horizontal component as 1.15.
• Believe that higher than normal σ is due to lack of consideration of site effects and due to the use of repi rather than rjb.