where Y is in g, c1= 0.14, c2= -6.25, c3= 0.37, c4= 2.237, c5= -7.542, c6= -0.125, c7= 1.19,
c8= -6.15, c9= 0.525, bv= -0.25, VA= 484.5, R1= 100km and σ = 0.552.
Characterise sites by Vs,30 (average shear-wave velocity in upper 30m). Note that approximately half the
stations have measured shear-wave velocity profiles.
Include basin effects through modification of D1. For sediment depth (Z ≥ 1kmD1= 0.35; otherwise
Use three faulting mechanism classes:
1120 records. F = 1.00.
1450 records. F = 1.28 (taken from previous studies).
but only retain two (strike-slip and reverse) by combining normal and strike-slip categories.
Only use earthquakes with focal depths < 20km. Focal depths between 4.6 and 19km.
Exclude data from aftershocks.
Use data from: Alaska (24 records), Armenia (1 record), California (2034 records), Georgia (8), Iran (7
records) Italy (10 records), Nevada (8 records), Taiwan (427 records), Turkey (63 records) and Uzbekistan
Most data from 5.5 ≤ Mw≤ 7.5.
Adopt functional form to model: a constant level of ground motion close to fault, a slope of about R-1 for
> 10km and R-1.5 at greater distances (> 100km) and observation (and theoretical results) that highest
amplitude ground motions do not always occur nearest the fault but at distances of 3–10km.
Choose functional form based on transfer function of a SDOF oscillator since this has similar characteristics
to those desired.
Note that magnitude scaling may need adjusting for small magnitudes.
Firstly regress for magnitude and distance dependency and then regress for site and basin effects.
Examine residual w.r.t. magnitude and distance and observe no significant trends.
Compare predictions to observations for 12 well-recorded events in the dataset and find that the
observations are well predicted for near and far distances.
Demonstrate (for the 2004 Parkfield earthquake) that it is possible to add an additional ‘filter’ term in
order to predict ground motions at large distances without modifying the other terms.