GNGTS 2019 - Atti del 38° Convegno Nazionale

478 GNGTS 2019 S essione 2.2 we are able to compute the pattern of the ground shaking by summing each contribution (median prediction and residual corrections) at any point of the regular grid. Here we show an example of shaking map (mean field) related to the scenario of L’Aquila M w 6.1 earthquake occurred on 6 th April 2009 for PGA (Fig. 2a). Following the proposed method, equiprobable realizations of the shaking maps are obtained by incorporating the prediction uncertainty (accounting both for the spatial interpolation variance and the residual aleatory variability) at different percentiles (Sgobba et al. , 2019); as shown in Fig. 2b for L’Aquila example. A comparison of the acceleration spectral ordinates with the observed values for different recording stations, performed in terms of different error metrics (Root Mean Squared Error and R-squared error), confirms the reliability of the GMM predictions adjusted for the regional corrections, with small differences between the GRID and the CLUSTER models. The obtained spatial distribution reproduces the main ground motion patterns as documented in the literature for the event of L’Aquila with reference to instrumental data or according to macroseismic observations (see Ameri et al. , 2011 for example). The most relevant similarity with the patterns proposed in the literature can be found on the systematic amplification effect observed at the south edge of the surface projection of the fault, which maybe more likely related to combined effects of site and path. Fig. 2 - Example of shaking maps simulation of the Mw6.1 L’Aquila earthquake (median field – 50° percentile) at PGA (a); zoomed map around the fault area (random field realization at 65° percentile) (b). Star indicates the epicenter of the event whereas black triangles indicate the recording stations. Continuous blue lines inside the area represent the main thrusts. a) b) Other tests have been performed also on independent events (i.e. not included in the dataset calibration), confirming the good agreement between predictions and observations. Some inconsistencies (average residual ~0.3 log10 units) have been detected at longer periods for events with larger magnitudes and conversely at short periods for smaller events at stations above or in the proximity of the faults, thus suggesting that the ground motion is here affected by more complex near-source effects, not fully captured by the adjusted model. This may be due to the fact that near-source effects mainly depend on the specific source and thus cannot be mapped into the repeatable terms of variability. Further improvements of the proposed approach include the implementation of the directivity effects into GMM by modelling the azimuthal distribution of the aleatory residuals (i.e. the final residuals after removing the systematic components). The procedure could also take advantage from the application of advanced geostatistical techniques to obtain more accurate spatial interpolation of the corrective terms.