Preliminary analysis of the acceler-ometric recordings of the August 24 th , 2016 MW 6 . 0 Amatrice earthquake

On 24 August 2016, at 1.36:32 GMT, a MW 6.0 earthquake with epicenter located below the village of Accumoli, struck a wide area among the boundaries of Lazio, Abruzzo, Umbria and Marche regions (Central Italy): the main event caused the collapse of several buildings and about 300 fatalities, mainly in the towns of Amatrice, Arquata del Tronto and Accumoli. The main event was recorded by about 350 sensors, belonging to the Italian Accelerometric Network (Rete Accelerometrica Nazionale, RAN) operated by the Department of Civil Protection (DPC), the Italian Seismic Network (Rete Sismica Nazionale, RSN) managed by the Istituto Nazionale di Geofisica e Vulcanologia (INGV), and to other local networks. All the corrected data are available at the Engineering Strong Motion Database (esm.mi.ingv.it). This paper reports the preliminary results of the analysis of the strong-motion recordings.


I. INTRODUCTION
On 24 August 2016, at 1.36:32 GMT, a M W 6.0 (INGV-CNT Seismic Bulletin) earthquake with epicenter located below the village of Accumoli, struck a wide area among the boundaries of Lazio, Abruzzo, Umbria and Marche regions (Central Italy).The epicenter is located below the village of Accumoli (Latitude 42.70°, Longitude 13.23°, Depth 8.1km, INGV-CNT Seismic Bulletin) and the fault plane solution (TDMT, CNT) indicates normal faulting, which is in agreement with the prevailing tectonic style of the area (according to the seismogenic zonation ZS9, Meletti et al. 2008).The fault geometry was calculated by Tinti et al. (2016) and has the following characteristics: strike 156°, dip 50°, rake -85°, length 26km, width 16km.The Italian Accelerometric Network (RAN), managed by the Department of Civil Protection (DPC), the Italian seismic network (RSN), managed by the Istituto Nazionale di Geofisica e Vulcanologia (INGV) and other regional networks (e.g, managed by OGS and AMRA) provided the records of the mainshock.itaca.mi.ingv.it, Luzi et al. 2008;Pacor et al. 2011).About 350 waveforms were manually processed using the procedure detailed in Pacor et al. (2011).The earthquake was over-all recorded by 14 stations within 30km and 42 within 50km from the epicenter.The location of the epicenter, the surface projection of the causative fault and the spatial distribution of the recording stations is shown in Figure 1.Although the selected ground motion models present similar functional forms, they assume different site effect terms: ITA10 accounts for linear soil amplifications through a site term based on Eurocode 8 (EC8; CEN, 2004) soil categories; ASB14 considers V S,30 as an explanatory variable for site effects and accounts for non-linear soil amplification through PGA.
Since measured V S,30 is available only for 17% of the recording stations, the EC8 soil category for the majority of recording sites was inferred from surface geology (Di Capua et al. 2011).The corresponding V S,30 at these sites , was assigned following the Wald and Allen (2007) approach, that adopts the to-pographic slope as a proxy.In case the V S,30 inferred from the slope and the EC8 category from surface geology do not match, the latter was retained and a preferred V S,30 is assigned (800 m/s for EC8-A, 580 m/s for EC8-B, 270 m/s for EC8-C, 170 m/s for EC8-D).The observed vertical PGAs are well predicted up to 100km (Figure 7).For larger distances, the observations are lower than predictions, similarly to horizontal components.Observed vertical PGVs are, in general, higher than predictions (Figure 7).The residuals (R es ) were calculated as logarithmic difference between observations and predictions and were decomposed in between-event δB e (mean of residual per event) and within-event δW es residuals (δW es = R es -δB e ), as proposed by Al- Atik et al. (2010).The between-event value is reported in the title.
Figure 8 shows the within-event residuals against event-to-station azimuth, for the stations up to R JB =80km.Positive residuals in-dicate under-predictions and are observed in the Northern sector at SA 0.3s (0-45° and 315-360°).Negative residuals (over-predictions) are instead obtained in the range 135-270°, corresponding to SW direction.This represents an additional evidence of directivity effects at high frequencies, as highlighted in Figure 4.No clear trend with azimuth is instead observed at SA 3s; large positive residuals are only observed for some stations located in alluvial basins (see also Figure 5).The between-event residual is found to be small at SA 0.3s, while it is larger and positive at SA 3s, which means a general underprediction of the GMPEs at long periods.
IV. WAVEFORM FEATURES OF NEAR-FAULT STATIONS: AMT AND NRC The near-fault recordings of AMT and NRC stations, both located at about 1.5 km from the fault projection but in opposite positions from the epicenter, present different characteristics in terms of peak values and frequency content.

V. CONCLUSIONS
The recent earthquake of Amatrice (24/08/2016, M W =6.0) was recorded by about 300 strong-motion stations.The accelerometric data were manually processed and are available at the ESM website (esm.mi.ingv.it).The spatial distribution of the PGA and low period spectral ordinates show a pattern that can be related to the source-to-site azimuth.The residual analysis and the interpolation of observed data highlight the nature of the rupture, which is bilateral, dominated by the NW propagation.Different attenuation patterns are also evident towards West and East.
The vertical components are comparable to the horizontal ones only in the near-fault or when site/propagation effects contribute to increase the vertical components in the far field.GMPEs generally fit the observations in the near-fault for low periods spectral ordinates; at distances larger than 80km, the residuals show a positive bias in the sense that the GMPEs tend to under-predict the observed ground motions.GMPEs for PGV and SA at long periods generally underestimate the observations, highlighting the necessity of reviewing the most recent GMPEs developed for Italy, widely based on analog records.

Figure 2 .
Figure 2. Spatial distribution of the maximum horizontal Peak Ground Acceleration (cm/s 2 ).

Figure 3 .
Figure 3. Spatial distribution of the maximum horizontal Peak Ground Velocity (cm/s).

Figure 5 .
Figure 5. Map of interpolated horizontal spectral acceleration (T = 3s).III.COMPARISON OF OBSERVED AND PREDICTED GROUND MOTIONS Some ground motion parameters (PGA, PGV and acceleration spectral ordinates at 0.3, 1 and 3 seconds) were compared to the predictions of the most recent GMPEs by Bindi et al (2011a), for Italy (ITA10), and Akkar et al. (2014), for Europe (ASB14).Both GMPEs use the Joyner-Boore distance up to 200km.Although the selected ground motion models present similar functional forms, they assume different site effect terms: ITA10 accounts for linear soil amplifications through a site term based on Eurocode 8 (EC8;CEN,  2004)  soil categories; ASB14 considers V S,30 as an explanatory variable for site effects and accounts for non-linear soil amplification through PGA.Since measured V S,30 is available only for 17% of the recording stations, the EC8 soil category for the majority of recording sites was inferred from surface geology(Di Capua et al. 2011).The corresponding V S,30 at these sites , was assigned following the Wald and Allen (2007) approach, that adopts the to-

Figure 6 and
Figure 6 and Figure 7 show the comparison between prediction and observations for the geometrical mean of the horizontal and ver-

Figure 8 .
Figure 8. Within-event residuals δWes vs. azimuth for SA at T=0.3s (top) and 3s (bottom) for the stations within R JB =80km.The event-to-site azimuth is computed clockwise from the north.The between-event value is reported in the title.

Figure 9 .
Figure 9. Velocity time history (a) and power spectrum (b) of the E component at station AMT

Figure 10 .
Figure 10.Velocity time history (a) and power spectrum (b) of the E component at station NRC