- Ground-motion model is: where Y is in g; a8 = -0.1091, a9 = 0.0937, a2 = 0.0029, a5 = 0.2529, a6 = 7.5 and a7 = -0.5096 for
all distance metrics; a1 = 1.85329, a3 = -0.02807, a4 = -1.23452, ϕ = 0.6201 (intra-event), τ = 0.3501
(inter-event) and σ = 0.7121 (total) for rjb; a1 = 2.52977, a3 = -0.05496, a4 = -1.31001, ϕ = 0.6375
(intra-event), τ = 0.3581 (inter-event) and σ = 0.7312 (total) for repi; a1 = 3.26685, a3 = -0.04846,
a4 = -1.47905, ϕ = 0.6475 (intra-event), τ = 0.3472 (inter-event) and σ = 0.7347 (total) for rhypo;
V REF = 750m∕s, V CON = 1000m∕s, b1 = -0.41997, b2 = -0.28846, c = 2.5 and n = 3.2 are from
Sandikkaya et al. (2013).
- Use V s,30 to characterise sites. Most sites classified in Eurocode 8 classes B and C, i.e. 180 ≤ V s,30 ≤
800m∕s. Note limited data from V s > 800m∕s. Use nonlinear site amplification model of Sandikkaya
et al. (2013) for model. Recommend model for 150 ≤ V s,30 ≤ 1200m∕s.
- Focal depths between roughly 0 and 29km. No dependency on mechanism. Vast majority of earthquakes
with Mw > 6 have depths < 20km and distribution for smaller events roughly uniform.
- Data from 322 stations.
- Use 3 mechanisms:
- FN = FR = 0.
- FN = 1, FR = 0. Most data from this mechanism.
- FR = 1, FN = 0. Relatively few records.
- Derive using RESORCE (Akkar et al., 2014c) as part of special issue (Douglas, 2014) including 4 other
ground-motion models (Douglas et al., 2014).
- Most data from Italy, Turkey and Greece but believe models can be used for seismically-active areas in S.
Europe and Middle East.
- Derive models using rjb, repi and rhypo so as to avoid requirement for distance conversion or virtual faults
when using the models in probabilistic seismic hazard assessments.
- Include records from aftershocks because: difficult to classify European events into mainshocks and
aftershocks, about half records come from aftershocks, and there is limited evidence for differences in
motions for European data (Douglas and Halldórsson, 2010). Note that this inclusion could increase σ.
- Note that possible bias in data at great distances because of trigger thresholds but conclude, based on
predictions from previous model and various instrument resolutions, that data roughly unbiased for Mw > 4
and rjb < 200km.
- Exclude data from 163 singly-recorded events so as not to inflate τ (inter-event variability).
- Only include data from 3-component accelerograms so that a consistent model for vertical-to-horizontal
spectral ratio can be derived.
- Remove events with Mw < 5 with < 3 records to make the distribution w.r.t. mechanism more uniform
and to prevent small events dominating derivation of mechanism terms.
- Note that data covers Mw ≤ 7 well, particularly for normal and strike-slip mechanisms. For Mw > 7
almost no records from normal and reverse events and most data from 3 strike-slip earthquakes: 1990
Manjil, 1999 Kocaeli and 1999 Düzce.
- Undertake trial regressions adjusting motions to V s,30 = 750m∕s using nonlinear site amplification model
of Sandikkaya et al. (2013) to choose functional form. Also regress using simple site classes to check
Sandikkaya et al. (2013) and find similar results. Consider quadratic, cubic and hinged magnitude scaling
and visually compare observations and predictions and study reduction in τ (inter-event). Impact on τ was
limited so mainly use visual comparisons. Plot predicted and observed ground motions scaled to rjb = 10km
and V s,30 = 750m∕s against Mw. Find similar results for Mw < 6 and significant differences for Mw > 7.
Find oversaturation predicted by cubic model that is not seen in data. Find hinged magnitude scaling best
matches observations but note that this is somewhat unconservative and higher epistemic uncertainty at
these magnitudes because of lack of data.
- Try including anelastic attenuation term but find non-physical positive coefficients so remove it.
- Try including magnitude-dependent distance saturation but find similar predictions and no significant
impact on σ. Hence remove it to reduce number of coefficients.
- Do not consider effect of depth to top of rupture because of limited information.
- Find that a2, a5, a6 and a7 show little variation with T and hence make them period-independent
coefficients, which leads to smooth spectra.
- Find mechanism coefficients a8 and a9 are very similar for three distance metrics so use same coefficients
for rjb, repi and rhypo.
- Plot residuals, grouped into bins, w.r.t. Mw, R and V s,30. Intra-event residuals do not show trends.
Find model overestimates observations for V s,30 < 180m∕s and underestimates short-period motions
for V s,30 > 800m∕s. But note that data in these bins are sparse and poorly distributed. Inter-event
residuals suggest some bias for large magnitudes. Narrowing in residuals for large magnitudes could suggest
magnitude-dependent σ but note sparse data for Mw > 7. Decide not to model magnitude-dependent σ
since apparent dependency in residuals could be due to uncertain metadata (particularly Mw) for smaller
events and data from only handful of events for Mw > 6.5 leading to underestimation of true σ here.
- Note that σs for repi may underestimate true variability because of lack of data from Mw > 6 and
rjb < 15km for which the impact of using repi rather than rjb is largest.
- Note possible overprediction of motions for Mw < 5 based on comparisons to previous models.
- Note uncertainty in model beyond the data (Mw > 7.6) but believe that can be used up to Mw8 based
on comparisons to other models.
- Note that extrapolation to rjb > 200km can be done with some caution.