where A is in cm∕s2, b0= 2.305, b1= 0.178, b2= -0.666 and b3= 30 (σ not reported7).
Use N-value profiles from standard penetration tests (SPTs) to characterise sites. Use data from alluvial
and diluvial sites. Exclude data from rock and very soft soils. Define Sn as a weighting function for SPT
profile to characterise softness of surface layers. Plot residuals from model against Sn and find correlation.
Derive site correction factors for model. Find coefficient of variation decreases after applying correction.
Use 346 uncorrected components (magnitudes from 5 to about 7.8 and repi from about 7 to 500km) to
derive preliminary model without site term: A= b010b1M∕(repi+30)b2. Derive models using different data
selections: all data, M < 6.6, M ≥ 6.6, repi≤ 119km, repi> 119km, M-repi region where expected PGA
(from model using all data) ≥ 39cm∕s2 or expected PGA < 39cm∕s2 (these selected divide data into two
equal halves). Examine scaling of the various models in 3D plots. Based on this analysis, conclude that
model depends on M-repi range used for data selection. Because of engineering interest in PGA> 10gal
believe model should be derived using M-repi region defined by expected PGA.
18 records from 1978 Off Miyagi earthquakes and 6 records from 1978 Izu-oshima-kinkai earthquake.
Strong correlation between M and repi with almost all data from M > 7 being from repi> 100km.
Try different b3 values but find influence on coefficient of variation minimal so fix to 30km.
For final model use corrected accelerograms (PGAs generally 10% to 30% higher than uncorrected values).
Most data from SMAC instruments.
Plot residuals w.r.t. repi and M and find no trends.
7Report coefficient of variation of 0.578 which could be the value of σ in terms of natural logarithms.