GNGTS 2019 - Atti del 38° Convegno Nazionale

GNGTS 2019 S essione 2.2 477 The GMM is calibrated for the geometrical mean of the horizontal components of the Peak Ground Acceleration (PGA) and 70 acceleration response spectral ordinates (SA) in the period range 0.04-2s. With respect to the predictions, three correction contributions of the systematic and repeatable effect of ground motion are estimated through the residual analysis: • the site-to-site residual term (dS2S) defines the systematic bias of ground motions recorded at a specific stations with respect to the GMM predictions (fixed-effects) at reference sites; in this way the dS2S can be considered as a proxy of the amplification function of the station; • the location-to-location term (dL2L) defines the systematic bias of the source regions. The location terms are here computed by means of two different procedures: (i) GRID approach: the study area is divided into a regular grid with any cell representing the source region in which the events are averaged; (ii) CLUSTER approach: the events are aggregated and averaged within polygonal areas identified on the basis of spatial-temporal criteria of clustering. dL2L may be related to the average stress-drop within the source region (Baltay et al. , 2017; Bindi et al. , 2018); • the path-to-path term (dP2P) defines the systematic deviations along one source-to-site path and is related to anomalies in crustal velocity, density or in the attenuation function (Baltay et al. , 2017). dP2P are here computed on the basis of the two different procedures used to identify the source areas (models GRID and CLUSTER). Geostatistical analysis of the residual terms. To investigate and model the spatial dependence of the corrective terms, a traditional geostatistical analysis is applied. Under the hypothesis of univariate normal distribution and intrinsic stationarity (for the site terms) while non-stationarity (for the path terms), a spatial correlation model is built by fitting the sample semivariograms. A Kriging interpolation technique is applied as a predictor to estimate the value of the corrective terms on the unsampled locations of the computation grid (1.6x1.6 km resolution) and then to reconstruct the fields of the corrections. Maps of such residuals show remarkable path effects: as an example in Fig. 1 for L’Aquila source area, the spatial distribution of dP2P terms for PGA indicates positive corrections (i.e. ground motion amplification) - over about 30 km of correlation distance - for seismic ray-paths travelling from the Apennine chain to the Adriatic coast (red contours), whereas negative values (i.e. deamplifications) affect the area towards the Tyrrhenian coast (blue contours). Maps of interpolated dS2S terms are characterized instead by smaller correlation distances at almost all investigated periods. In any case, the variability of these terms was found to be relevant (in the range 0.1-0.15 log10 units), when removed from the aleatory uncertainty. Example of empirical simulation and discussion Once built the non-ergodic model and obtained correlated fields of the corrective terms through geostatistical analysis, Fig. 1 - Example of spatially correlated fields of the path and location terms computed for L’Aquila area (CLUSTER model) at PGA.