Ground-motion model is not reported. Use six functional forms.
Use four site categories:
Very soft soil
Vs,30< 180m∕s. 0 records.
180 ≤ Vs,30< 360m∕s. 1 record.
360 ≤ Vs,30< 750m∕s. 34 records.
Vs,30≥ 750m∕s. 93 records.
Site conditions are unknown for 35 records. Classify mainly using description of local site conditions owing
to unavailability of Vs measurements.
Exclude data from Mw< 3.0 to exclude data from earthquakes that are likely to be associated with large
uncertainties in their size and location and because ground motions from smaller earthquakes are likely to
be of no engineering significance.
Exclude data from multi-storey buildings, on or in dams or on bridges.
Most data from Mw< 5.5 so believe use of repi is justified.
Records from: eastern N America (78 records), NW Europe (61 including 6 from UK) and Australia (24).
Locations from special studies, ISC/NEIC or local network determinations.
Note distinct lack of data from < 10km for Mw> 5.
Only retain good quality strong-motion data. No instrument correction applied because of the lack of
instrument characteristics for some records. Individually bandpass filter each record with a Butterworth
filter with cut-offs at 25Hz and cut-off frequencies chosen by examination of signal-to-noise ratio and
integrated velocity and displacement traces.
Find use of different functional forms has significant influence on predicted PGA.
Regression on only rock data generally reduced PGA.
Predictions using the functional forms with quadratic M-dependence were unreliable for Mw> 5.5 because
they predict decrease PGA with increasing M since there was insufficient data from large magnitude
earthquakes to constrain the predictions.
Find different regression methods predict similar PGAs with differences of < 5% for a Mw5 event at 5km
when all records were used but differences up to 63% when using only rock data. Prefer the one-stage
maximum-likelihood method since allows for correlation between M and d in dataset and does not ignore
earthquakes recorded by only a single station (25% of data).
Find, from analysis of residuals, that equation generally underpredicts PGA of data from eastern N America
and Australia but overpredicts motions from Europe and UK.
Find no trends in residuals w.r.t. amplitude, distance, magnitude or fault mechanism.
Believe that large σs found are due to: lack of data from close to large magnitude earthquakes, use of
data from different regions with varying source and path characteristics and use of much data from small
earthquakes that are probably associated with higher uncertainty w.r.t. magnitude and location since such
earthquakes have not been as well studied as large earthquakes and there is a lack of data with high
signal-to-noise ratio from which relocations can be made.
Do not recommend equations for practical use due to large uncertainties.