350 likes | 511 Views
ITACA the Italian strong-motion database Task5 – site classification. L. Luzi, F. Pacor, R. Puglia, M. Massa, D. Bindi Istituto Nazionale di Geofisica e Vulcanologia (INGV). M. R. Gallipoli, M. Mucciarelli Università della Basilicata. R. Paolucci, S. Giorgetti Politecnico di Milano (POLIMI).
E N D
ITACAthe Italian strong-motion databaseTask5 – site classification L. Luzi,F. Pacor, R. Puglia, M. Massa, D. Bindi Istituto Nazionale di Geofisica e Vulcanologia (INGV) M. R. Gallipoli, M. Mucciarelli Università della Basilicata R. Paolucci, S. Giorgetti Politecnico di Milano (POLIMI)
Analysis description • The performance of different classification schemes has been tested through the evaluation of GMPEs • The GM is represented by the acceleration response spectra ordinates (5% damping) • The response variables are: magnitude, distance, style of faulting and soil classes • A GMPE is derived for each classification • The GMPE performance has been evaluated in terms of standard deviation of the GMPEs and of the errors associated to the classes of each scheme
Functional form for regression Functional form(e.g. Akkar & Bommer, 2007): Mref = 5.6, Rref = 1km Response variable: SA (5%, 0.04≤T≤ 4sec) Geomean of H components
Regression approach Random effect model(e.g. Brillinger & Preisler, 1985): Inter-event (hi) Inter-station (qk) Inter-event error = error due to an earthquake recorded by many stations Inter-station error = error due to a station which recorded several events
RANDOM EFFECT MODELinter/intra - event Earthquake i recorded at station k Observation Mean prediction Error distributions Inter-event distribution of error h: it assumes a value for each earthquake and describes the correlation among the errors for different recordings of the same earthquake. It is a normal distribution with standard deviation equal to t Intra-event distribution of error x: it assumes a value for each recording. It is a normal distribution with standard deviation equal to s. The error distributions h and x are assumed to be independent.
RANDOM EFFECT MODELinter/intra - event Earthquake i recorded at station k Observation Mean prediction Error distributions The residuals are decomposed as the sum of the inter- and intra-event error distributions Since the distributions are independent, the total variance is the sum of the two variances:
RANDOM EFFECT MODELinter/intra - station Earthquake i recorded at station k Observation Mean prediction Error distributions Inter-station distribution of error q: it assumes a value for each station and describes the correlation among the errors for different recordings at the same station. It is a normal distribution with standard deviation equal to d Intra-station distribution of error x’: it assumes a value for each recording. It is a normal distribution with standard deviation equal to s’. The error distributions d and x’ are assumed to be independent.
Residualik stot2=0.16763 = = qk d2=0.05867 Bindi et al, 2010 + + x’ik x’2=0.10896 Error distributions variances Example of ITACA , SA at T=1.75 s Recordings ik % qk Stations k % x’ik Recordings ik
Different earthquakes with magnitude 5.5±0.2 recorded at GBP GBP CLC AVZ Bindi et al, 2010 Example of ITACA , SA at T=1.75 s Model for ITACA (black): mean prediction for a M=5.5, class C - EC8
Different earthquakes with magnitude 5.5±0.2 recorded at GBP Inter-station error for GBP GBP qGBP qCLC Red curve= Mean GMPE + inter-station error for GBP CLC qAVZ nearly zero Bindi et al, 2010 AVZ Example of ITACA , SA at T=1.75 s Model for ITACA (black): mean prediction for a M=5.5, class C - EC8
Different earthquakes with magnitude 5.5±0.2 recorded at GBP Inter-station error for GBP GBP qGBP x’GBP,i Intra-station error for event i recorded at GBP CLC Bindi et al, 2010 AVZ Example of ITACA , SA at T=1.75 s Model for ITACA (black): mean prediction for a M=5.5, class C - EC8
Example of ITACA , SA at T=1.75 s Different earthquakes with magnitude 5.5 ± 0.4 recorded by different class Cstations Red curve= Mean + inter-station standard deviation Blue curve= Mean + intra-station standard deviation Dashed curve= Mean + total standard deviation Model for ITACA (black): mean prediction for a M=5.5, class C - EC8
ITACA - EC8 based on Vs,30 when available (~ 80 stations at the present time) OR • based on an expert evaluation when Vs,30 is not available, account for: • detailed geology and stratigraphic profiles when available • H/V from noise and/or earthquake data • 1:100,000 lithologic map
Classification schemes: Sabetta & Pugliese (1987) • Based on geological and geotechnical information and the thickness H of the soil layer, three categories: • Rock sites • Stiff, shallow alluvium (H =< 20 m) • Deep alluvium (H > 20 m) • Stiff sites have average shear-wave velocity greater than 800 m/s • alluvium sites have a shear-wave velocity between 400 and 800 m/s
Classification schemes: Rovelli et al. (2008) FUKUSHIMA et al. (2007) ZHAO et al. (2006) JAPAN ROAD ASSOCIATION PERIOD T (sec) CAT. SC1 T < 0.2 PERIOD T (sec) CAT. SC2 0.2 <= T < 0.6 SCI T < 0.2 T >= 0.6 SC3 SCII 0.2 <= T < 0.4 Generic Rock SC4 SCIII 0.4 <= T < 0.6 SC5 Generic Soil SCIV T >= 0.6 based on predominant period of H/V SA ratios Rovelli et al. PERIOD T (sec) CAT. SCI T < 0.2 SCII 0.2 <= T < 0.4 SCIII 0.4 <= T < 0.6 SCIV T >= 0.6 SCV T unknown & orig. AB site SCVI T unknown & orig. CD site SCVII Unknown
Classification based on f0-Vs,30 • Based on Vs,30 and fundamental frequency of the site, evaluated through H/V of acceleration response spectra • 3 classes are individuated on the base of cluster analysis • Sites are assigned to a class on the base of the membership degree
Cluster analysis C3 C2 C1 Cluster analysis: the error of each cluster is calculated as the mean point – to – centroid distance (normalized to the standard deviation of the cluster)
Degree of membership to a class C3 C2 C1
Cluster analysis (one variable) C3 C2 C1 Degree of membership to a class
Degree of membership to a class • Assuming that the variables of the points in a cluster are normally distributed, the membership to a soil class can be evaluated as probability density • For a normal distribution of one variable, the probability density function is: m is the variable mean s is the standard deviation The assigned class is the one with the highest probability
Data set for regression A common data set of 1000 records Magnitude range 3.5 – 6.3 Distance range 0 – 300 km
Number of stations for each class = rock sites
Preliminary considerations • SP96 has 2 soil classes, therefore the soil coefficients tend to smooth the behaviour of peculiar sites. The classification is efficient, as the curves are clearly separated. • EC8 has 4 soil classes, 2 classes represent sites with well defined response (classes D and E), while classes B and C tend to be very similar at low periods • ROV has 6 soil classes, 2 have well defined response (classes 1 and 4), classes 2 and 3 have intermediate response, but they are too similar (0.2 - 0.4s and 0.4–0.6 s), coefficients of classes 6 and 7 also tend to be very similar (problem in class attribution?) • S4-MI has 3 soil classes, each one with a well defined response.
Preliminary considerations • T (0.04-1s): SP96 EC8 and ROV have similar total standard deviations, S4-MI has lowest • T>1s: EC8 is the classification with the lowest standard deviation, and it depends on the fact that 2 classes amplify the GM, class D (but represented only by 3 stations…..) and C
Error distribution for each class (SP96) 0 1 2 Sigma >= 0.3
A new soil class Sites with broad band amplification: multiple peaks and average amplitude greater than 2.7 for a wide frequency range MNF Rock sites=38 AQG GRR PSC ? LNS FHC 2.5 2.5
Performance Before