- Ground-motion model for shallow events is:
where a = -1.186±0.169, b1 = 0.726±0.033, c1 = -1.719±0.060, h = 1.551±1.003, eB = 0.357±0.055,
eD = 0.376 ± 0.062, σeve = 0.223 (inter-event), σsta = 0.229 (inter-station) and σT = 0.393 (total)
for horizontal; and a = -1.110 ± 0.168, b1 = 0.691 ± 0.034, c1 = -1.749 ± 0.061, h = 1.245 ± 0.151,
eB = 0.338 ± 0.055, eD = 0.430 ± 0.064, σeve = 0.230 (inter-event), σsta = 0.218 (inter-station) and
σT = 0.393 (total) for vertical.
- Ground-motion model for deep events is:
where Mref = 3.0, Rref = 1; a = 2.527±2.437, b1 = -0.397±0.308, b2 = 0.094±0.039, c1 = -1.998±1.450,
c2 = 0.315±0.070, h = 9.608±5.229, c3 = 0±0.008, eB = -0.200±0.075, eD = 0.002±0.086, σeve = 0.161
(inter-event), σsta = 0.276 (inter-station) and σT = 0.399 (total) for horizontal; and a = 2.459 ± 0.888,
b1 = -0.407 ± 0.293, b2 = 0.100 ± 0.038, c1 = -1.956 ± 0.428, c2 = 0.273 ± 0.061, h = 9.315 ± 2.632,
c3 = -0.001 ± 0.003, eB = 0.229 ± 0.058, eD = -0.013 ± 0.066, σeve = 0.157 (inter-event), σsta = 0.249
(inter-station) and σT = 0.387 (total) for vertical.
- Use 3 Eurocode 8 site classes:
- V s,30 > 800m∕s. About 40% of data. SB = SD = 0.
- 360 < V s,30 ≤ 800m∕s. About 50% of data. SB = 1, SD = 0.
- V s,30 < 180m∕s. 5.4% of data. SD = 1. SB = 0.
Originally included data from site class C (180 ≤ V s,30 ≤ 360m∕s) sites but only 42 (SE) and 77
(DE) records so coefficients poorly constrained. Removed data from this category. Classification based
on GIS maps based on seismic logs in Catania province and geo-lithographical maps of Sicily and other
classifications outside this province.
- Divide data into two classes based on focal depth h:
- Shallow. h < 5km. Events related to dynamics of volcanic edifice rather than regional stress field.
Events within sedimentary substratum. 95% of events have h < 2km. Deepest event has h = 4.5km.
- Deep. 5 < h ≤ 30km.
Find clear difference in records from these two classes with data from shallow events having more low
- Data from 72 24-bit Nanometrics Trillium broadband velocity instruments (sample rate 100Hz) in the Rete
Sismica Permanente della Sicilia Orientale of INGV, from 04/2006 to 11/2012. Stations located between
Aeolian Islands and Hyblean Plateau. Magnitudes and locations from INGV.
- Exclude data from ML < 3 based on quality and homogeneity.
- Visually inspect all data to exclude traces with electronic glitches or that are amplitude saturated. Baseline
correct (offset and linear trend removal) records. Correct for instrument response. Bandpass filter data
with cut-offs of 0.1 and 25Hz. Differentiate records to obtain acceleration.
- Data reasonably well distributed up to 100km and because motions from further distances of limited
engineering interest exclude data beyond this distance. No correlation between magnitude and distance.
Vast majority ( 90%) of data from ML < 4.
- Try various functional forms and report coefficients. Compare results based on σ, F-tests to check reduction
in variance and Akaike and Bayesian Information Criteria. Prefer the models reported above, despite the
coefficients of the deep model not conforming to expectations. Use bootstrap (sampling with replacement)
to assess standard deviations of coefficients.
- Study residuals w.r.t. M and R. Find no significant trends. Histograms suggest that residuals follow a
normal distribution, although the Lilliefors test does not confirm this for low significance levels.
- Compute inter-event and inter-station σ using analysis of variance.
- Examine inter-station errors w.r.t. each station. Examine inter-event errors w.r.t. earthquake ID.
- Compute confidence intervals of predictions. Find that confidence intervals of complex models broaden
greatly at edges of data. Warn against extrapolation of models.
- Use data not used within the boot-strapping to find the confidence limits of the coefficients within
cross-validation exercise to check the root-mean-squared errors. Conclusions on best models match those
obtained through other techniques.
- Compare observations and predictions w.r.t. R for various magnitude ranges. Find good fits.