Reliable ground-motion measurements are essential for seismic hazard assessment and require seismological stations to be installed in free-field conditions, away from structural interferences. However, global network analyses show diverse installation configurations that can affect measurements. Although topographic effects are natural, their influence can extend over several meters, making them closely related to operators' installation choices. Although summit effects are well studied and known to cause strong amplifications, the impact of cliffs or escarpments remains less explored. In this study, 16 SmartSolo IGU-16HR 3C nodes were placed on either side of a 30-m cliff located in Cephalonia (western Greece), for 4-10 months, recording 307 low-to-moderate-magnitude seismic events. The main results reveal that de-amplifications recorded at the cliff base are greater than amplifications determined at the cliff top, particularly between 5 and 15 Hz. This study shows that amplification and de-amplification are primarily driven by topography, whereas their directional dependence is likely shaped by both topographic and lithologic factors. Furthermore, the anisotropy related to cliff fracturing interacts with the effects of source back azimuth. Amplification and de-amplification can be correctly predicted using the frequency-scaled curvature (FSC) and illuminated FSC proxies, provided that an exponential functional form is used for determining de-amplifications at the base of near-vertical cliffs. Furthermore, the azimuthal dependence observed, inducing strong de-amplifications depending on the direction observed, seems related more to the anisotropy of geologic formations than to the cliff topography itself. These effects should be considered when using data from stations in similar topographic settings. Neglecting them can bias ground-motion models and underestimate earthquake magnitudes, particularly for stations at the base of cliffs. From these considerations, we strongly recommend installing future seismological stations a few tens of meters away from the bases and ridges of even small cliffs or escarpments.

ABSTRACT Fourier transfer function FTF(f), defined as a Fourier transform of the impulse response of the medium between a source and receiver, fully characterizes transfer properties in the frequency domain. Modulus of FTF(f) is one of the most used characteristics of seismic site response in both theoretical and numerical analyses of effects of local surface structures on seismic motion. It is especially useful for identifying resonance peaks and/or the amplified frequency bands. By definition, however, |FTF(f)| cannot provide information on temporal development of the site response. In this article, we introduce a novel characterization of seismic site response—time–frequency transfer function TFTF(t, f) that properly characterizes temporal development of frequency content of transfer function. This feature significantly helps in interpretating wavefield composition. We present theory that is readily applicable to any time-dependent output signal. We demonstrate superior properties of TFTF(t, f) in comparison with standard |FTF(f)| for selected quasi-1D canonical models of local surface sedimentary structures. In the model of a semi-infinite layer, TFTF(t, f) made it possible to distinguish modes of vertical resonance and modes of surface waves, which is impossible based on standard |FTF(f)|.

SUMMARY Seismic ambient-noise horizontal-to-vertical spectral ratios (H/V, HVSR) are widely used to characterize near-surface structure and to identify site resonant frequencies. In this article, we propose a non-parametric statistical descriptor of ambient-noise HVSRs. In single-station practice, HVSR curves are typically represented by the geometric mean of window-wise ratios at each frequency, that is, by the median of a lognormal model. However, distributions of window ensembles are frequently skewed or multimodal. The lognormal median and symmetric uncertainty bands can then depart from the most probable amplitudes and distort the spread. The lognormal mode can mitigate this bias in near-lognormal cases, but it remains tied to a unimodal parametric form. As a consequence, many workflows rely on strict window selection or rejection to justify lognormal assumption. To address this problem, we present a data-driven, non-parametric alternative that provides a marked improvement when ensembles deviate from lognormality. At each frequency, we estimate the probability density of the log-amplitude distribution using kernel density estimation (KDE), transform it to linear space and take the mode of the linear-domain density as the representative curve. Uncertainty is quantified by highest density intervals, which naturally accommodate asymmetry and multibranch distributions. Using an example of a near-lognormal microtremor, we demonstrate that the KDE mode closely tracks the lognormal mode while revealing a systematic upward shift of the commonly used lognormal median. Using a microtremor at a structurally complex site, we observe strongly multimodal amplitude and directional statistics. Both the lognormal-median and lognormal-mode curves fall between competing modes. However, the KDE-based modes and intervals follow the dominant branches and capture multimodal spread. Because the density is inferred directly from window-wise data, the method reduces the need for strong window rejection. We also compare the KDE-mode descriptor with the energy-ratio estimator which averages horizontal and vertical component energies before taking their ratio. Agreement between the energy-ratio estimator and the KDE mode is consistent with a stable single HVSR population, whereas discrepancies help identify frequency bands affected by multimodality, non-stationarity or directional effects. Finally, we extend the same framework to directional HVSR using the horizontal spectral matrix and quantify directional variability through a $\pi $-periodic circular KDE of the axial principal horizontal directions. We show that KDE-based modes, highest density intervals and directional spread provide robust distribution-aware observational diagnostics that can guide data selection, uncertainty assignment and frequency weighting in site-characterization and inversion workflows.

Ground Motion Models (GMMs) are fundamental to seismic hazard assessment, yet often rely on data assumed to be recorded in free-field conditions. In reality, stations installation conditions –often undocumented or poorly documented in networks metadata – can vary and significantly affect ground motion recordings in certain frequency ranges, potentially impacting GMMs reliability.This study focuses on the influence of buildings on ground motion and ambient vibrations wave field. Nine buildings housing permanent accelerometric stations in Greece have been temporarily instrumented using SmartSolo IGU-16HR 3C nodes, recording between 98 to 425 earthquakes. Sensors were deployed at four “key locations”: co-located with the permanent station at the foundation level (STA), on the top floor (TBD), near the building (CBD), and in free-field (FF1, FF2) as reference. Data were primarily analyzed using the Standard Spectral Ratios (SSR) method.Results reveal significant foundation-level de-amplification, reaching factors of 2 to 5 at sites with VS30 < 500 m/s in the 3-10 Hz frequency ranges. These effects typically start near the building resonance frequency and may extend to lower frequencies on softer sites, whereas they are negligible on hard rock – using earthquake recordings. Furthermore, ambient noise SSR provide results consistent with earthquake recordings up to 10 Hz. However, strong anthropogenic activity (e.g., in buildings like hospitals), noisy free-field reference stations, and unsuitable sensor configurations can introduce bias. Nighttime measurements and multiple reference stations are therefore recommended.These findings demonstrate that building installation effects can bias GMMs in engineering-relevant frequency bands and should be considered when developing models.

During earthquakes, a high degree of spatial variation in damage distribution, encompassing both structural damage to buildings and co-seismic landslides, is commonly observed in mountainous regions near the seismic source. Among other factors, this spatial variability can be partly attributed to the amplification of seismic waves caused by surface topography. Our study focuses on predicting ground-motion amplification due to topography in close proximity to earthquakes and examining its potential influence on co-seismic landslide distribution patterns. To achieve this goal, we employ neural network analysis on previously available synthetic data from 3D finite-differences simulations of seismic wave propagation. The analysis aims at developing a physics-based estimator of topographic site effects in close distances to the source, referred to as the i-FSC proxy (Illuminated Frequency Scaled Curvature). This proxy depends on the S-wavelength, the curvature of the topographic surface, and a new parameter called the "normalized seismic illumination angle", which quantifies the slope's exposure to the incoming wavefield. The inclusion of the illumination parameter substantially decreases the uncertainties of the proxy by a factor of 2 compared to estimators that rely solely on curvature as a key parameter. The i-FSC proxy is a user-friendly tool that does not require high computational resources; it utilizes only a digital elevation map and the position of the seismic source to predict amplification factors at any point on the surface topography. This estimator allows exploring the spatial variations in topographic amplification caused by nearby seismic sources, representing a significant breakthrough as areas closest to the fault typically sustain the most damage during earthquakes. Subsequently, the i-FSC proxy is used to investigate the correlation between ground-motion amplification and the spatial distribution of earthquake-induced landslides triggered by large events such as the 2015 Gorkha earthquake (MW 7.8). The results indicate that more than 71% of co-seismic landslides tend to be localized in amplified areas. Different physical controls on the landslide triggering at different frequencies have been identified. The results also highlight the crucial importance of considering the effect of topographic amplification, together with other classical factors such as slope steepness, for a better understanding of the complex mechanisms governing the spatial distribution of earthquake-induced landslides at local and regional scales. The obtained results could provide valuable insights for future researches, guiding efforts towards more effective risk assessment and mitigation strategies in mountainous regions.

This study presents a new, data-driven, region-specific ground motion model for Greece. This model utilizes a neural network approach that eliminates the need for any a priori functional form. Due to limitations in the recent Greek dataset, selected records from the RESORCE database have been incorporated. A fully connected multi- layer perceptron is employed to predict several ground motion intensity measures (GMIMs), including peak ground velocity (PGV), peak ground acceleration (PGA), and the 5 % damped pseudo-spectral acceleration (PSA) at 18 periods ranging from 0.01 to 4.00 s, for active shallow crustal earthquakes (h <= 30 km). Given the available dataset information, this GMM is driven by three input parameters; moment magnitude (Mw), Joyner-Boore distance, RJB (km), and the average seismic shear-wave velocity of the uppermost 30 m at the station site, VS30 (m/s). Additional source parameters, such as focal mechanism and depth, were also tested. The linear mixed-effects algorithm of the lme4 package [1] is used to decompose the total ground-motion aleatory variability (GMAV) into inter-event residuals (SBe) and Site-to-Site residuals (SS2S) while analyzing their dependence on the magnitude and distance (heteroscedasticity). The sensitivity of GMIMs predictions to various input parameters is also analyzed. Results indicate that combining the partially non-ergodic assumption (SS2S) with the heteroscedastic model significantly reduces GMAV, while these data-driven predictions exhibit physical trends consistent with classical GMMs. This new GMM enables site-specific predictions throughout Greece, provided sufficient on-site recordings exist to derive the site-specific term SS2Ss.

Our study focuses on predicting topographic amplification of ground motion in the near-source region, where seismic rays reach the free-surface at varying incidence angles. We rely on data from previous 3D numerical simulations conducted on a topographic relief with a homogeneous medium. First, using neural networks, we identify which key parameters, describing the geometric characteristics of the relief relative to the seismic source position, control ground motion amplification. Then, we determine the functional form that relates these parameters to the simulated amplifications. Subsequently, we conduct a regression study to develop a model of topographic amplification, referred to as the i-FSC proxy (Illuminated Frequency-Scaled Curvature). Our estimator depends on the frequency-scaled (1) curvature, a parameter that accounts for the occurrence of amplifications over convex topographies and de-amplification over concave ones; (2) normalized illumination angle, a newly defined parameter that quantifies the slope exposure to the incoming wavefield, accounting for high amplification on slopes oriented opposite to the seismic source. The illumination parameter reduces the uncertainties of the proxy by a factor of 2 compared to estimators that rely solely on curvature. The proxy does not require high computational resources. It uses a digital elevation map and a seismic source position to predict amplification factors (without lithological effects) for an S-wave at any site on the surface topography. It allows exploration of variations in topographic amplification near seismic sources, representing a significant breakthrough as areas closest to the fault typically sustain the highest damages. A MATLAB script performing the i-FSC calculations is provided.

Local soil conditions significantly influence the characteristics of observed seismic waves, often obscuring or distorting the original source signature and modifying recorded waveforms. Local backprojection (BP) imaging, a common technique for identifying the spatiotemporal high-frequency energy release during seismic ruptures, typically relies on strong motion data that is frequently not installed on bedrock. We evaluate the impact of soil effects on local BP images by analyzing data obtained from the 2008 Iwate-Miyagi Nairiku earthquake and recordings from both surface and downhole stations of the Kiban-Kyoshin network. Initially, we conducted a detailed study of sites near the seismic source. Surface station data were corrected for site effects using horizontal-to-vertical spectral ratios and established correction methods that account for nonlinearity effects, providing accurate estimations of horizontal amplification, which we then eliminate from the seismic signals. Extensive comparisons between corrected surface records, uncorrected surface records, and borehole records reveal that uncorrected surface records can distort BP coherence, leading to inaccurate results. However, applying site effect corrections significantly enhances BP image quality, aligning outcomes more closely with those from borehole records. In addition, we examine the site independence of borehole data concerning site effects. This study highlights the importance of incorporating site effect corrections in BP analyses to improve seismic data accuracy. Finally, we propose a comprehensive workflow for integrating these corrections into future BP studies.

Considering the purpose of the session relating early engineering developments in site response and soil-structure interaction, this paper focuses on the development of studies regarding site-city interaction following the striking site response observations obtained in Mexico City during the 1985 Guerrero-Michoacan event, The first part presents an overview of the investigations on multiple structure-soil-structure interaction, starting with Mexico-city like environments with dense urbanization on soft soils, which later evolved with the concept of metamaterials. Up to now, such investigations have been largely relying on numerical simulations in 2D and 3D media, coupling soft surface soil layers and simplified building models, including also some theoretical developments using various mechanical concepts. They also relied on a number of laboratory experiments on reduced-scale mock-ups with diverse vibratory sources (shaking table, acoustic devices). The latest studies coupled full-scale experiments on mechanical analogs such as forests or wind turbine farms involving sets of resonators with similar frequencies, and numerical simulation to investigate their impact on the propagation of surface (Rayleigh) waves. Almost all such studies converge in predicting lower ground motion amplitude for sites located within the ”urbanized” area, but none of them can be considered a ”groundtruth” proof for a real earthquake in a real city. The second part thus takes advantage of the long duration of strong motion observations in the Kanto area thanks to the KiK-net, K-NET and JMA (Shin-dokei) networks, to investigate the possible changes in site response with time. The first results obtained with the event-specific site terms derived from Generalized Inversion Techniques (Nakano et al., 2015) indicate a systematic reduction of the low frequency (0.2 -1 Hz) site amplification, in the central-south Tokyo area. As this frequency band corresponds both to the site frequency (very thick deposits) and to the high-rise buildings, the discussion focuses on the possible relation with the extensive construction in some areas of downtown Tokyo over the last 2 decades.

As a contribution to step 3 of the ESG6 blind prediction exercise, we present an application of two different, purely empirical approaches to estimate the strong ground motion at a soft site ("KUMA") from the observed ground motion at a reference rock site ("SEVO") for the two largest shocks of the Kumamoto 2016 sequence. The two methods estimate the non-linear transfer function between a reference rock and a sedimentary site by modifying the linear transfer function derived from weak motion recordings. The modification is based either on a machine learning tool based on a wide collection of Japanese weak and strong motion recordings and the associated site metadata (method 1), or on an estimate of a site-specific parameter related to an average non-linear site response (method 2). The acceleration time series are then derived at the sedimentary site of interest using an estimation of the time delay between wave arrivals at the rock and site stations, and a minimum phase assumption for the site transfer function. These predictions were made blindly, but after the ESG6 conference they could be compared both with the actual ground motion recorded at KUMA during the two shocks, and the average and range of all other predictions preformed for this benchmark. Both of these purely empirical methods provide an honorable prediction of usual engineering ground motion parameters of the two target events. The performance of these two purely empirical approaches is at least comparable to those of the numerical simulation methods for the foreshock—if not better—and slightly worse for the (largest) mainshock. As the methods required only recordings of weak motions at the target and a referent sites and very simple description of the soil profile. The use of moderate motions to constrain the frequency shift prediction for the second method and the consideration of an alternative phase modification are possible ways to improvement.

The 2020 update of the European Seismic Hazard Model (ESHM20) is the most recent and up-to-date assessment of seismic hazard for the Euro-Mediterranean region. The new model, publicly released in May 2022, incorporates refined and cross-border harmonized earthquake catalogues, homogeneous tectonic zonation, updated active fault datasets and geological information, complex subduction sources, updated area source models, a smoothed seismicity model with an adaptive kernel optimized within each tectonic region, and a novel ground motion characteristic model. ESHM20 supersedes the 2013 European Seismic Hazard Model (ESHM13; Woessner et al., 2015) and provides full sets of hazard outputs such as hazard curves, maps, and uniform hazard spectra for the Euro-Mediterranean region. The model provides two informative hazard maps that will serve as a reference for the forthcoming revision of the European Seismic Design Code (CEN EC8) and provides input to the first earthquake risk model for Europe (Crowley et al., 2021). ESHM20 will continue to evolve and act as a key resource for supporting earthquake preparedness and resilience throughout the Euro-Mediterranean region under the umbrella of the European Facilities for Seismic Hazard and Risk consortium (EFEHR Consortium).

In this supplementary materials section, we provide a comprehensive exploration of the differences of the earthquake rate forecasts of ESHM20 and ESHM13.Additionally, we present a series of trellis plots to facilitate a comparative analysis of the ground motion characteristics models.Changes in the seismogenic sources cause many of the local differences across the entire region.Regional

Historical and archeological data report that an earthquake was felt over the whole eastern Mediterranean on 21 July A.D. 365. The impact of the tsunami it generated, which may have caused several thousands of fatalities, has been widely studied, whereas the impact of the seismic waves has hardly been explored. Here, we present simulations of the ground motion caused by the A.D. 365 event at 316 sites now instru-mented by seismological stations throughout Europe. The simulation approach is based on the modeling of a series of rupture scenarios coupled with empirical Green's func-tions (EGFs) obtained at the stations from a recent M-w 6.4 earthquake. The broadband and remarkably also the accelerometric records in urbanized areas can be exploited at distances as far as similar to 2000 km. Then, we use three empirical models to estimate the macroseismic intensity across Europe from the simulated peak ground accelerations and peak ground velocities. The presence of stations in thick sedimentary basins (lower Danube valley, Po plain, urban accelerometric network in the alpine valley of Grenoble) shows that local basin amplification effects can dominate acceleration values at frequencies favorable to human earthquake perception (similar to 0.5-1 Hz), even for basins located at more than 1500 km from the earthquake. Thus, our simulations indicate that the A.D. 365 earthquake was likely felt by the populations as far as the Po plain and as the city of Grenoble, about 1800 km away, and presumably in other large European basins such as the Pannonian basin. It is possible that the perception of the earthquake up to such distances contributed to its "universal" character perpetuated in archival sources. At closer distances (300-500 km), the simulated intensity levels indicate that the earthquake probably caused no damage.

The present work develops a comprehensive probabilistic seismic hazard study for Lebanon, a country prone to a high seismic hazard since it is located along the Levant fault system. The historical seismicity has documented devastating earthquakes which have struck this area. Contrarily, the instrumental period is typical of a low-to-moderate seismicity region. The source model built is made of a smoothed seismicity earthquake forecast based on the Lebanese instrumental catalog, combined with a fault model including major and best-characterized faults in the area. Earthquake frequencies on faults are inferred from geological as well as geodetic slip rates. Uncertainties at every step are tracked and a sensitivity study is led to identify which parameters and decisions most influence hazard estimates. The results demonstrate that the choice of the recurrence model, exponential or characteristic, impacts the most the hazard, followed by the uncertainty on the slip rate, on the maximum magnitude that may break faults, and on the minimum magnitude applied to faults. At return periods larger than or equal to 475 years, the hazard in Lebanon is fully controlled by the sources on faults, and the off-fault model has a negligible contribution. We establish a source model logic tree populated with the key parameters, and combine this logic tree with three ground-motion models (GMMs) potentially adapted to the Levant region. A specific study is led in Beirut, located on the hanging-wall of the Mount Lebanon fault to understand where the contributions come from in terms of magnitudes, distances and sources. Running hazard calculations based on the logic tree, distributions of hazard estimates are obtained for selected sites, as well as seismic hazard maps at the scale of the country. Considering the PGA at 475 years of return period, mean hazard values found are larger than 0.3 g for sites within a distance of 20–30 km from the main strand of the Levant Fault, as well as in the coastal region in-between Saida and Tripoli (≥ 0.4 g considering the 84th percentile). The study provides detailed information on the hazard levels to expect in Lebanon, with the associated uncertainties, constituting a solid basis that may help taking decisions in the perspective of future updates of the Lebanese building code.

ABSTRACT Site-specific seismic hazard assessment involves the prior knowledge of (1) the input ground motion at the local bedrock and (2) the site response. In this article and its companion, we address the deconvolution approach to obtain a reference ground-motion model, which consists of removing site effects from surface ground-motion recordings. Laurendeau et al. (2018) applied this approach on the KiK-net network with site response from 1D SH-wave (1DSH) simulations, calculated using the VS profiles available for most sites. Indeed, this approach presents several limitations with 1DSH site response if it is considered to be applied to other networks, especially in the European context. First, the approach requires identification of sites with dominant 1D effects. Second, it needs the presence of accurately measured VS profiles. In this context, we propose to derive deconvolved ground-motion models using site response from generalized inversion techniques () for two main reasons. The first reason is that the GIT delivers empirical site response for all types of sites, conditioned by the presence of sufficient amount of data, and the second is that it reduces the need for VS profiles. We focus on the estimation of site effects from different approaches and present a methodology to obtain reliable site terms from GIT based on the experience from previous studies. We also introduce and detail the difference between absolute and relative site response, which mainly depends on the chosen reference. We estimate and compare site response for the KiK-net stations with different methods, that is, empirical and theoretical 1DSH. We also conclude a list of 1D sites based on comparisons between theoretical and empirical estimates of site response. The results obtained in this article will be the primary input for the deconvolution approach applied in the companion article.

ABSTRACT In the framework of site-specific seismic hazard assessment, the definition of reference motion is a crucial step. Reference motion is generally associated with hard-rock conditions, characterized by S-wave velocity exceeding 1500 m/s. However, ground motion recorded at sites with such conditions is poorly represented in existing strong-motion databases. Thus, the validity domains of most empirical ground-motion prediction equations (GMPEs) are not representative of reference rock conditions. To overcome this limitation and assess ground motion at reference conditions, the so-called “deconvolution approach” was proposed by Laurendeau et al. (2018) to correct surface recordings from theoretical 1DSH site response before GMPE developments. With the same purpose, in this article, we propose to apply the deconvolution approach using empirical site-response estimates as an alternative to theoretical ones. Using the Kiban–Kyoshin network (KiK-net) data, we estimate empirical site responses at KiK-net stations using generalized inversion techniques in addition to those from 1DSH numerical simulations, as presented in the companion article. Finally, a reference ground-motion model (RGMM) is determined based on empirically deconvolved ground motions. The advantage of using empirical rather than 1DSH site responses in the deconvolution approach is that in the former case the RGMM can be built based on records from an extensive set of sites, whereas the latter case is restricted to well-characterized sites with dominant 1D behavior. This makes the proposed approach easily exportable to different regions of the world, where precise site characterizations are not systematically available, and the knowledge of site behavior is limited.