- Ground-motion model is: Use SR only for crustal reverse events, SI only for interface events, SS only for subduction slab events
and Sk for each of the site classes (k = 1, …, 4). δh = 0 for h < hc and 1 otherwise. For h > 125km use
h = 125km.
- Use four site categories:
- Rock, natural period T < 0.2s, V s,30 > 600m∕s, approximately NEHRP classes A and B. 1381 records.
- Hard soil, natural period 0.2 ≤ T < 0.4s, 300 < V s,30 ≤ 600m∕s, approximately NEHRP class C.
- Medium soil, natural period 0.4 ≤ T < 0.6s, 200 < V s,30 ≤ 300m∕s, approximately NEHRP class D.
- Soft soil, natural period T ≥ 0.6s, V s,30 ≤ 200m∕s, approximately NEHRP classes E and F. 938
Site classification unknown for 62 records. Prefer using site classes rather than individual coefficients for
each station because avoids possibility of source effects being shifted into site terms and can be used when
there are only a few records per station.
- Response parameter is acceleration for 5% damping.
- Classify earthquakes into three types:
- Focal depths ≤ 25km. 81 earthquakes, 1497 records.
- 88 earthquakes, 1188 records.
- 101 earthquakes. 1715 records.
- Classify earthquakes into four mechanisms:
- 160 earthquakes (28 crustal), 1969 records (373 crustal).
- 82 earthquakes (39 crustal), 1674 records (1100 crustal).
- 26 earthquakes (4 crustal), 749 records (24 crustal).
- 2 earthquakes (0 crustal), 8 records (0 crustal).
Consider differences between reverse and strike-slip motions for crustal earthquakes because enough data
but note there is not enough data to consider normal earthquakes as a separate group.
- Focal depths, h, between about 0 and 162km with most < 60km.
- Exclude data from distances greater than a specified limit for a given magnitude in order to eliminate bias
due to untriggered instruments. For subduction slab events, fix maximum distance as 300km.
- Note that there is little near-source data from Japan from within 30km. All Japanese data from within
10km is from two earthquakes (Kobe 1995 and Tottori 2000). Add data from with 40km from earthquakes
in western USA (h < 20km) and from the Tabas 1978 (Iran) earthquake to help constrain near-source
behaviour of derived equations. Use data from: Japan (61 crustal earthquakes, 1301 records; 87 interface
earthquakes, 1176 records; 101 slab earthquakes, 1715 records) and Iran and western USA (20 crustal
earthquakes; 196 records; 1 interface earthquake, 12 records).
- Note that reasonably good distribution of data for all magnitudes and focal depths.
- Note strong correlation between focal depth and distance.
- Use ISC relocations rather than JMA locations because find that they are more reliable.
- Use Mw values from Harvard CMT unless value from special study is available.
- Prefer the one-stage maximum-likelihood method to the two-stage method because when there many events
with only a small number of records and many individual site terms, the coefficients must be determined
using an iterative method and hence their reliability is questionable.
- Find that, by residual analysis (not shown), that equations predict unbiased ground motions for crustal
and interface events but biased ground motions for slab events with bias that depends on distance. Apply
this magnitude-independent path modification factor SF for slab events: log(SF) = SSL[log() -
log(Rc)] where Ra = 90.0km and Rc = 125.0km.
- Find that, because of lack of near-source data, it is not possible to find reliable estimates of c and d so use
a iterative method to find d by fixing c.
- Estimate site coefficient, SH, for hard rock sites (V s,30 = 1500m∕s) from 10 stations with 1020 ≤ V s,30 ≤
2200m∕s with 1436 records, based on residuals.
- Examine residuals w.r.t. magnitude, distance and focal depth for all three source types and find no
significant bias. Find that PGAs from two events on east coast of Hokkaido are under-estimated and note
that investigation needed to see if it is a regional anomaly. Also find that ground motions from 2003 Miyagi
(Mw7.0) event are under-estimated, which note is due to a known regional anomaly.
- Believe model more robust than other models for subduction events due to lower prediction errors.
- Note that predictions for near-source ground motion for subduction events are largely constrained by data
from shallow crustal events from western USA hence adding subduction records from < 50km could result