Ground motion prediction maps using seismic microzonation data and machine learning

Past seismic events worldwide demonstrated that damage and death toll depend on both the strong ground motion (i.e., source effects) and the local site effects. The variability of earthquake ground motion distribution is caused by local stratigraphic and/or topographic setting and buried morphologies, that can give rise to amplification and resonances with respect to the ground motion expected at the reference site. Therefore, local site conditions can affect an area with damage related to the full collapse or loss in functionality of facilities, roads, pipelines, and other lifelines. To this concern, the near real time prediction of damage pattern over large areas is a crucial issue to support the rescue and operational interventions. A machine learning approach was adopted to produce ground motion prediction maps considering both stratigraphic and morphological conditions. A set of about 16'000 accelometric data and about 46'000 geological and geophysical data were retrieved from Italian and European databases. The intensity measures of interest were estimated based on 9 input proxies. The adopted machine learning regression model (i.e., Gaussian Process Regression) allows to improve both the precision and the accuracy in the estimation of the intensity measures with respect to the available near real time predictions methods (i.e., Ground Motion Prediction Equation and shaking maps). In addition, maps with a 50 × 50 m resolution were generated providing a ground motion variability in agreement with the results of advanced numerical simulations based on detailed sub-soil models. The variability at short distances (hundreds of meters) was demonstrated to be responsible for 30–40 % of the total variability of the predicted IM maps, making it desirable that seismic hazard maps also consider short-scale effects.

Seismic Wave Propagation and Basin Amplification in the Wasatch Front, Utah

Ground-motion analysis of more than 3000 records from 59 earthquakes, including records from the March 2020 Mw 5.7 Magna earthquake sequence, was carried out to investigate site response and basin amplification in the Wasatch Front, Utah. We compare ground motions with the Bayless and Abrahamson (2019; hereafter, BA18) ground-motion model (GMM) for Fourier amplitude spectra, which was developed on crustal earthquake records from California and other tectonically active regions. The Wasatch Front records show a significantly different near-source rate of distance attenuation than the BA18 model, which we attribute to differences in (apparent) geometric attenuation. Near-source residuals show a period dependence of this effect, with greater attenuation at shorter periods (T<0.5 s) and a correlation between period and the distance over which the discrepancy manifests (∼20–50 km). We adjusted the recorded ground motions for these regional path effects and solved for station site terms using linear mixed-effects regressions, with groupings for events and stations. We analyzed basin amplification by comparing the site terms with the basin geometry and basin depths from two seismic-velocity models for the region. Sites over the deeper parts of the sedimentary basins are amplified by factors of 3–10, relative to sites with thin sedimentary cover, with greater amplification at longer periods (T≳1 s). Average ground-motion variability increases with period, and long-period variability exhibits a slight increase at the basin edges. These results indicate regional seismic wave propagation effects requiring further study, and potentially a regionalized GMM, as well as highlight basin amplification complexities that may be incorporated into seismic hazard assessments.

Hard as a rock? Looking for typical and atypical reference sites in the Greek network

<p>Although it is nowadays desirable and even typical to characterise site conditions in detail at modern recording stations, this is not yet a general rule in Greece, due to the large number and geographical dispersion of stations. Indeed, most of them are still characterised merely through geological descriptions or proxy-based parameters, rather than through in-situ measurements. Considering: 1. the progress made in recent years with sophisticated ground motion models and the need to define region-specific rock conditions based on data, 2. the move towards large open-access strong-motion databases that require detailed site metadata, and 3. that Greek-provenance recordings represent a significant portion of European seismic data, there are many reasons to improve our understanding of site response at these stations. Moreover, it has been shown recently in several regions that even sites considered as rock can exhibit amplification and ground motion variability, which has given rise to more scientific research into the definition of reference sites. For Greece, in-situ-characterisation campaigns for the entire network would impose unattainable time/budget constraints; so, instead, we implement alternative empirical approaches using the recordings themselves, such as the horizontal-to-vertical spectral ratio technique and its variability. We present examples of 'well-behaved', typical rock sites, and others whose response diverges from what is assumed for their class.</p><p> </p>

High-frequency ground-motion variability for rough-fault ruptures

<p>Geological observations show variations in fault-surface topography not only at large scale (segmentation) but also at small scale (roughness). These geometrical complexities strongly affect the stress distribution and frictional strength of the fault, and therefore control the earthquake rupture process and resulting ground-shaking. Previous studies examined fault-segmentation effects on ground-shaking, but our understanding of fault-roughness effects on seismic wavefield radiation and earthquake ground-motion is still limited.  </p><p>In this study we examine the effects of fault roughness on ground-shaking variability as a function of distance based on 3D dynamic rupture simulations. We consider linear slip-weakening friction, variations of fault-roughness parametrizations, and alternative nucleation positions (unilateral and bilateral ruptures). We use generalized finite difference method to compute synthetic waveforms (max. resolved frequency 5.75 Hz) at numerous surface sites  to carry out statistical analysis.  </p><p>Our simulations reveal that ground-motion variability from unilateral ruptures is almost independent of  distance from the fault, with comparable or higher values than estimates from ground-motion prediction equations (e.g., Boore and Atkinson, 2008; Campbell and Bozornia, 2008). However, ground-motion variability from bilateral ruptures decreases with increasing distance, in contrast to previous studies (e.g., Imtiaz et. al., 2015) who observe an increasing trend with distance. Ground-shaking variability from unilateral ruptures is higher than for bilateral ruptures, a feature due to intricate seismic radiation patterns related to fault roughness and hypocenter location. Moreover, ground-shaking variability for rougher faults is lower than for smoother faults. As fault roughness increases the difference in ground-shaking variabilities between unilateral and bilateral ruptures increases. In summary, our simulations help develop a fundamental understanding of ground-motion variability at high frequencies (~ 6 Hz) due small-scale geometrical fault-surface variations.</p>

Spatial Variability of Near-field Ground Motions from Pseudo-Dynamic Rupture Simulations

<p>We analyze the effect of earthquake source parameters on ground-motion variability based on near-field wavefield simulations for large earthquakes. We quantify residuals in simulated ground motion intensities with respect to observed records, the associated variabilities are then quantified with respect to source-to-site distance and azimuth. Additionally, we compute the variabilities due to complexities in rupture models by considering variations in hypocenter location and slip distribution that are implemented a new Pseudo-Dynamic (PD) source parameterization.</p><p>In this study, we consider two past events – the Mw 6.9 Iwate Miyagi Earthquake (2008), Japan, and the Mw 6.5 Imperial Valley Earthquake, California (1979). Assuming for each case a 1D velocity structure, we first generate ensembles of rupture models using the pseudo-dynamic approach of Guatteri et.al (2004), by assuming different hypocenter and asperities locations (Mai and Beroza, 2002, Mai et al., 2005; Thingbaijam and Mai, 2016). In order to efficiently include variations in high-frequency radiation, we adopt a PD parameterization for rupture velocity and rise time distribution in our rupture model generator. Overall, we generate a database of rupture models with 50 scenarios for each source parameterization. Synthetic near-field waveforms (0.1-2.5Hz) are computed out to Joyner-Boore distances Rjb ~ 150km using a discrete-wavenumber finite-element method (Olson et al., 1984). Our results show that ground-motion variability is most sensitive to hypocenter locations on the fault plane. We also find that locations of asperities do not alter waveforms significantly for a given hypocenter, rupture velocity and rise time distribution. We compare the scenario-event simulated ground motions with simulations that use the rupture models from the SRCMOD database (Mai and Thingbaijam, 2014), and find that the PD method is capable of reducing the ground motion variability at high frequencies. The PD models are calibrated by comparing the mean residuals with the residuals from SRCMOD models. We present the variability due to each source parameterization as a function of Joyner-Boore distance and azimuth at different natural period.</p>

Ranking and calibration of ground-motion models using the stochastic area metric.

<p>The selection and ranking of  Ground Motion Models (GMMs) for scenario earthquakes is a crucial element in seismic hazard analysis. Typically model testing and ranking do not appropriately account for uncertainties, thus leading to improper ranking. We introduce the stochastic area metric (AM) as a scoring metric for GMMs, which not only informs the analyst of the degree to which observed or test data fit the model but also considers the uncertainties without the assumption of how data are distributed. The AM can be used as a scoring metric or cost function, whose minimum value identifies the model that best fits a given dataset. We apply this metric along with existing testing methods to recent and commonly used European ground motion prediction equations: Bindi et al. (2014, B014), Akkar et al. (2014, A014) and Cauzzi et al. (2015, C015). The GMMs are ranked and their performance analysed against the European Engineering Strong Motion (ESM) dataset. We focus on the ranking of models for ranges of magnitude and distance with sparse data, which pose a specific problem with other statistical testing methods. The performance of models over different ranges of magnitude and distance were analysed using AM, revealing the importance of considering different models for specific applications (e.g., tectonic, induced). We find the A014 model displays good performance with complete dataset while B014 appears to be best for small magnitudes and distances. In addition, we calibrated GMMs derived from a compendium of data and generated a suite of models for the given region through an optimisation technique utilising the concept of AM and ground motion variability. This novel framework for ranking and calibration guides the informed selection of models and helps develop regionally adjusted and application-specific GMMs for better prediction. </p><p> </p>

Local and Moment Magnitude Analysis in the Ridgecrest Region, California: Impact on Interevent Ground-Motion Variability

ABSTRACT We investigate the dependence of event-specific ground-motion residuals in the Ridgecrest region, California. We focus on the impact of using either local (ML) or moment (Mw) magnitude, for describing the source scaling of a regional ground-motion model. To analyze homogeneous Mw, we compute the source spectra of about 2000 earthquakes in the magnitude range 2.5–7.1, by performing a nonparametric spectral decomposition. Seismic moments and corner frequencies are derived from the best-fit ω−2 source models, and stress drop is computed assuming standard circular rupture model. The Brune stress drop varies between 0.62 and 24.63 MPa (with median equal to 3.0 MPa), and values for Mw>5 are mostly distributed above the 90th percentile. The median scaled energy for Mw<5 is −4.57, and the low values obtained for the Mw 6.4 and 7.1 mainshocks (−5 and −5.2, respectively) agree with previous studies. We calibrate an ad hoc nonparametric ML scale for the Ridgecrest region. The main differences with the standard ML scale for California are observed at distances between 30 and 100 km, in which differences up to 0.4 magnitude units are obtained. Finally, we calibrate ground-motion models for the Fourier amplitude spectra, considering the ML and Mw scales derived in this study and the magnitudes extracted from Comprehensive Earthquake Catalog. The analysis of the residuals shows that ML better describes the interevent variability above 2 Hz. At intermediate frequencies (between about 3 and 8 Hz), the interevent residuals for the model based on Mw show a correlation with stress drop: this correlation disappears, when ML is used. The choice of the magnitude scale has an impact also on the statistical uncertainty of the median model: for any fixed magnitude value, the epistemic uncertainty is larger for ML below 1.5 Hz and larger for Mw above 1.5 Hz.

Diffracted wavefield decomposition and multidimensional site effects in the Argostoli valley, Greece

SUMMARY Effects of seismic ground motion induced by surface geology and geometry are known to be associated with the generation of a substantial proportion of surface waves. As a consequence, surface waves significantly contribute to ground-motion variability and site amplification. There is a growing body of literature recognizing that an understanding of physical patterns of the wavefield crossing a site is the key aspect to characterize and quantify them. However, this task remains technically challenging due to the complexity of such effects as well as the limitations of geophysical investigations, especially in case of small sedimentary valleys. The present study attempts to investigate the waves propagating across two 2-D dense seismic arrays from a number of earthquakes and explore the extent to which they are contributing to the multidimensional site effects. The arrays were deployed in the small-size, shallow alluvial valley of Koutavos-Argostoli, located in Cephalonia Island, Greece, and consisted of three-component velocimeters with interstation distances ranging from 5 to 160 m. A set of 46 earthquakes, with magnitudes between 2 and 5 and epicentral distances up to 200 km, was analysed by using an advanced seismic array processing technique, MUSIQUE. The phase velocity, backazimuth and energy of the dominant waves crossing the array were extracted, and their identification as Love or prograde/retrograde Rayleigh waves was obtained. The results clearly indicate a predominance of scattered surface waves (up to 60 per cent of total energy), mainly from the closest valley edges, above the fundamental frequency (∼1.5 Hz) of the valley. Love waves dominate the low-frequency wavefield (<3 Hz) while Rayleigh waves dominate some high-frequency bands. An excellent consistency is observed, in a given frequency range, among the dominance of the type of diffracted surface waves, group velocities estimated from the ground velocity structure and site amplification. The outcomes of this research provide a better understanding of the contribution of edge-diffracted surface waves and the 2-D/3-D site amplification at small and shallow alluvium valleys like Argostoli. The method applied here can be used to calibrate and validate 3-D models for simulating seismic ground motion.

Ground-Motion Attenuation, Stress Drop, and Directivity of Induced Events in the Groningen Gas Field by Spectral Inversion of Borehole Records

ABSTRACT Production-induced earthquakes in the Groningen gas field caused damage to buildings and concerns for the population, the gas-field owner, and the local and national authorities and institutions. The largest event (ML=3.6) occurred in 2012 near Huizinge, and, despite the subsequent decision of the Dutch government to reduce the gas production in the following years, similar magnitude events occurred in 2018 and 2019 (ML=3.4). Thanks to the improvement of the local seismic networks in the last years, recent events provide a large number of recordings and an unprecedented opportunity to study the characteristics of induced earthquakes in the Groningen gas field and related ground motions. In this study, we exploit the S-wave Fourier amplitude spectra recorded by the 200 m depth borehole sensors of the G network from 2015 to 2019 to derive source and attenuation parameters for ML≥2 induced earthquakes. The borehole spectra are decomposed into source, attenuation, and site nonparametric functions, and parametric models are then adopted to determine moment magnitudes, corner frequencies, and stress drops of 21 events. Attenuation and source parameters are discussed and compared with previous estimates for the region. The impact of destructive interference of upgoing and downgoing waves at borehole depth on the derived parameters is also discussed and assessed to be minor. The analysis of the apparent source spectra reveals that several events show rupture directivity and provides clear observations of frequency-dependent directivity effects in induced earthquakes. The estimated rupture direction shows a good agreement with orientation of pre-existing faults within the reservoir. Our results confirm that rupture directivity is still an important factor for small-magnitude induced events, affecting the amplitude of recorded short-period response spectra and causing relevant spatial ground-motion variability.