Do not report coefficients as aim of study is to examine the residuals and the various components of
Characterise sites by Vs,30. 2 sites classed as Eurocode A, 6 as Eurocode B and the rest as Eurocode C.
Data from 3D strong-motion array of 16 surface and 6 downhole stations in Mygdonia valley. Number of
records per station is between 9 and 90 with an average of about 33.
Records filtered using acausal 4th-order bandpass Butterworth filter with a threshold signal-to-noise ratio
of 3. Lowest usable frequencies from 0.1 to 3Hz (decreasing with magnitude) and highest usable frequencies
from 10 to 50Hz.
Exclude data from before 2003 as most data from low-resolution instruments and not all array operational.
Exclude data from > 220km. Exclude records without both horizontal components. Exclude data from
events with focal depths > 25km to reject subduction events. Finally exclude records from earthquakes
not recorded by ≥ 3 stations because this could increase τ.
Examine influence of various steps on τ, ϕ, σ and ϕSS.
Find improving seismological data (going from routine locations and magnitudes to reassessed values)
reduces τ by 30–50%. Most of this reduction is due to better magnitude estimates.
Find improving site information (i.e. changing the site term from S = 0 to more complex forms: binary
soil/rock, Vs,30 and finally individual site terms) reduces system site-to-site variability ϕS2S by 20–30%.
Find through sensitivity checks (increasing the minimum number of records to 5, 8, 10, 12 and 14) that 3
records per event is sufficient for robust results.
Examine ϕSS,S and δS2S, the correlations between them and with geology for the various stations.