- Ground-motion model is [following Bindi et al. (2014a)]: where Y is in m∕s2. Mh = 6.75, Mref = 5.5 and Rref = 1km. e1 = 2.982, b1 = -0.363, b2 = -0.195,
b3 = -0.406, c1 = -1.231, c2 = 0.272, c3 = -0.00395, h = 6.390, Δc3,Italy = -0.00326 ± 0.00079,
Δc3,Others = 0.00326 ± 0.00076, Δ3,Turkey = 0.00000 ± 0.00034, g1 = 1.407, g2 = -0.234, Δg1,Italy =
-0.360 ± 0.258, Δg1,Others = -0.678 ± 0.212, Δg1,Turkey = 1.038 ± 0.314, Δg2,Italy = 0.063 ± 0.045,
Δg2,Others = 0.119±0.037, Δg2,Turkey = -0.182±0.055, τ = 0.350 (inter-event), ϕ0 = 0.451 (intra-event),
ϕS2S = 0.330 (site-to-site) and σ = 0.657 (total).
- Use V s,30 to characterise sites. Range is from 90 to 2000m∕s but bulk is from 200 to 600m∕s. Recommend
model for 180 ≤ V s,30leq1000m∕s.
- Use data from RESORCE (Akkar et al., 2014c) (2013 version).
- Exclude poor quality, unprocessed records and those without 3 components. Also exclude singly-recorded
earthquakes and those with Mw from empirical conversion formulae. Select data with: Mw ≥ 4, focal
depth < 35km, R < 300km (exclude records with Mw > 5 and/or Repi < 10km without rjb), and known
or inferred V s,30.
- Consider regional effects in Italy (378 records), Turkey (659 records) and rest of Europe and the Middle
East [all those countries with fewer than 200 records, mainly: Greece (137), Montenegro (35), Iran (20) and
France] (214 records) to partially mitigate the ergodic assumption. Note that could prefer a regionalisation
based on tectonics but not possible given unbalanced composition of RESORCE.
- Display Mw-R distribution w.r.t. region and Eurocode 8 site class. Very few records from Turkey in class
A (V s,30 > 800m∕s), which means model for this range of V s,30 controlled by data from elsewhere. Note
model for R > 100km and site classes B and C (360 ≤ V s,30 ≤ 800m∕s) could be mainly controlled by
- Median magnitude is Mw5.5.
- Bulk of data within 150km.
- Conduct preliminary non-parametric analysis (not shown) and find evidence for regional dependency of
- Use mixed-effect regression.
- Try to separate Greece from Others category but find that adjustments are not significant at 5% level so
keep it inside this category.
- Constrain c3 to be ≤ 0 for all periods to avoid unphysical behaviour (this constraint is required for T > 1s).
- Try including style-of-faulting terms but find them poorly-constrained and associated with large standard
errors. Because of this, the unbalanced nature of the database (e.g. most events in Italy are normal), the lack
of reverse-faulting events in database and preliminary analysis showing limited impact of style-of-faulting
these terms were dropped.
- Examine residuals w.r.t. V s,30 and find evidence for regional effects but large scatter, which suggests need
for other parameters to characterise sites.
- Note that attenuation of high-frequency motions could be result of both anelastic attenuation and site
effects. Check for correlation and trade-off between c3 and g2 as well as between regional variation of these
coefficients. Do not find significant correlation (not shown).
- Find regional adjustments are statistically significant at short periods.
- Test model including correlated regional variation on coefficients controlling distance scaling (c1, c2 and
c3) or combinations of them. Do not find appreciable improvements based on Akaike information criterion,
significance tests or residuals.
- Note that regionally dependency in site terms could be related to differences in the average velocity profiles
in different regions.
- Find σ reduces by about 10% and ϕS2S by about 20% when using regional terms.
- Find τ reduces by up to 30% when events with converted Mw estimates are removed.
- Do not recommend model for use outside Europe and the Middle without a compatibility check.
- Provide estimates of epistemic uncertainty in regional adjustments based on bootstrap method.