Search

The latest generation of national and regional probabilistic seismic hazard assessments (PSHAs) in Europe presents stakeholders with multiple representations of the hazard in many regions. This raises the question of why and by how much seismic hazard estimates between two or more models differ, not only where models overlap geographically but also where new models update existing ones. As modern PSHA incorporates increasingly complex analysis of epistemic uncertainty, the resulting hazard is represented not as a single value or spectrum but rather as probability distribution. Focusing on recent PSHA models for France and Germany, alongside the 2020 European Seismic Hazard Model, we explore the differences in model components and highlight the challenges and strategy for harmonising the different models into a common PSHA calculation software. We then quantify the differences in the source model and seismic hazard probability distributions using metrics based on information theory, illustrating their application to the Upper Rhine Graben region. Our analyses reveal the spatial variation in and complexity of model differences when viewed as probability distributions and highlight the need for more detailed transparency and replicability of the models when used as a basis for decision-making and engineering design. [ABSTRACT FROM AUTHOR]

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). [ABSTRACT FROM AUTHOR]

Current practice in strong ground motion modelling for probabilistic seismic hazard analysis (PSHA) requires the identification and calibration of empirical models appropriate to the tectonic regimes within the region of application, along with quantification of both their aleatory and epistemic uncertainties. For the development of the 2020 European Seismic Hazard Model (ESHM20) a novel approach for ground motion characterisation was adopted based on the concept of a regionalised scaled-backbone model, wherein a single appropriate ground motion model (GMM) is identified for use in PSHA, to which adjustments or scaling factors are then applied to account for epistemic uncertainty in the underlying seismological properties of the region of interest. While the theory and development of the regionalised scaled-backbone GMM concept have been discussed in earlier publications, implementation in the final ESHM20 required further refinements to the shallow-seismicity GMM in three regions, which were undertaken considering new data and insights gained from the feedback provided by experts in several regions of Europe: France, Portugal and Iceland. Exploration of the geophysical characteristics of these regions and analysis of additional ground motion records prompted recalibrations of the GMM logic tree and/or modifications to the proposed regionalisation. These modifications illustrate how the ESHM20 GMM logic tree can still be refined and adapted to different regions based on new ground motion data and/or expert judgement, without diverging from the proposed regionalised scaled-backbone GMM framework. In addition to the regions of crustal seismicity, the scaled-backbone approach needed to be adapted to earthquakes occurring in Europe's subduction zones and to the Vrancea deep seismogenic source region. Using a novel fuzzy methodology to classify earthquakes according to different seismic regimes within the subduction system, we compare ground motion records from non-crustal earthquakes to existing subduction GMMs and identify a suitable-backbone GMM for application to subduction and deep seismic sources in Europe. The observed ground motion records from moderate- and small-magnitude earthquakes allow us to calibrate the anelastic attenuation of the backbone GMM specifically for the eastern Mediterranean region. Epistemic uncertainty is then calibrated based on the global variability in source and attenuation characteristics of subduction GMMs. With the ESHM20 now completed, we reflect on the lessons learned from implementing this new approach in regional-scale PSHA and highlight where we hope to see new developments and improvements to the characterisation of ground motion in future generations of the European Seismic Hazard Model. [ABSTRACT FROM AUTHOR]

We present a seismic catalog (, https://doi.org/10.5880/GFZ.2.6.2023.010) including energy magnitude Me estimated from P waves recorded at teleseismic distances in the range 20°≤Δ≤98° and for depths shorter than 80 km. The catalog is built starting from the event catalog disseminated by GEOFON (GEOFOrschungsNetz), considering 6349 earthquakes with moment magnitude Mw≥5 occurring between 2011 and 2023. Magnitudes are computed using 1 031 396 freely available waveforms archived in EIDA (European Integrated Data Archive) and IRIS (Incorporated Research Institutions for Seismology) repositories, retrieved through the standard International Federation of Digital Seismograph Networks (FDSN) web services (https://www.fdsn.org/webservices/ , last access: March 2024). A reduced, high-quality catalog for events with Mw≥5.8 and from which stations and events with only few recordings were removed forms the basis of a detailed analysis of the residuals of individual station measurements, which are decomposed into station- and event-specific terms and a term accounting for remaining variability. The derived Me values are compared to Mw computed by GEOFON and with the Me values calculated by IRIS. Software and tools developed for downloading and processing waveforms for bulk analysis and an add-on for SeisComP for real-time assessment of Me in a monitoring context are also provided alongside the catalog. The SeisComP add-on has been part of the GEOFON routine processing since December 2021 to compute and disseminate Me for major events via the existing services. [ABSTRACT FROM AUTHOR]

To test whether a globally inferred sediment thickness value from geomorphological studies can be used as a proxy to predict earthquake site amplification, we derive site-amplification models from the relation between empirical amplification for sites in Europe and Türkiye and the geomorphological sediment thickness. The new site-amplification predictions are then compared to predictions from site-amplification models derived using the traditional site proxies, VS30 inferred from slope, slope itself, and geological era and slope combined. The ability of each proxy to capture the site amplification is evaluated based on the reduction in site-to-site variability caused by each proxy. The results show that the highest reduction is caused by geological era and slope combined, while the geomorphological sediment thickness shows a slightly larger or equal reduction in site-to-site variability as inferred VS30 and slope. We therefore argue that including geology and geomorphology in site-amplification modelling on regional scale can give an important added value and that globally or regionally inferred models for soil and sediment thickness from fields beyond engineering seismology can have a great potential in regional seismic hazard and risk assessments. Furthermore, the differences between the site-amplification maps derived from different proxies capture the epistemic uncertainty of site-amplification modelling. While the different proxies predict similar features on a large scale, local differences can be large. This shows that using only one proxy when predicting site amplification does not capture the full epistemic uncertainty, which is demonstrated by looking into detail on the site-amplification maps predicted for eastern Türkiye and Syria, where the devastating Kahramanmaraş earthquake sequence occurred in February 2023. [ABSTRACT FROM AUTHOR]

To design user-centred and scientifically high-quality outreach products to inform about earthquake-related hazards and the associated risk, a close collaboration between the model developers and communication experts is needed. In this contribution, we present the communication strategy developed to support the public release of the first openly available European Seismic Risk Model and the updated European Seismic Hazard Model. The backbone of the strategy was the communication concept in which the overall vision, communication principles, target audiences (including personas), key messages, and products were defined. To fulfil the end-users' needs, we conducted two user testing surveys: one for the interactive risk map viewer and one for the risk poster with a special emphasis on the European earthquake risk map. To further ensure that the outreach products are not only understandable and attractive for different target groups but also adequate from a scientific point of view, a two-fold feedback mechanism involving experts in the field was implemented. Through a close collaboration with a network of communication specialists from other institutions supporting the release, additional feedback and exchange of knowledge was enabled. Our insights, gained as part of the release process, can support others in developing user-centred products reviewed by experts in the field to inform about hazard and risk models. [ABSTRACT FROM AUTHOR]

We present a seismic catalogue including energy magnitude Me estimated from P-waves recorded at teleseismic distances in the range 20o ≤ Δ ≤ 98o and for depths less than 80 km. The catalogue is built starting from the event catalogue disseminated by GEOFON, considering 6349 earthquakes with moment magnitude Mw ≥ 5 occurring between 2011 and 2023. Magnitudes are computed using 1031396 freely available waveforms archived in EIDA and IRIS repositories, retrieved through standard FDSN webservices (https://www.fdsn.org/webservices/). A reduced, high quality catalogue for events with Mw ≥ 5.8 and from which stations and events with only few recordings were removed forms the basis of a detailed analysis of the residuals of individual station measurements, which are decomposed into station and event specific terms, and a term accounting for remaining variability. The derived Me values are compared to Mw computed by GEOFON and with the Me values calculated by IRIS. Software and tools developed for downloading and processing waveforms for bulk analysis and an add-on for SeisComP for real-time assessment of Me in a monitoring context are also provided alongside the catalogue. The SeisComP add-on is part of the GEOFON routine processing since December 2021 to compute and disseminate Me for major events via the existing services. [ABSTRACT FROM AUTHOR]

The latest generation of national and regional probabilistic seismic hazard assessments (PSHA) in Europe presents stakeholders with multiple representations of the hazard in many regions. This raises the question of why and by how much seismic hazard estimates between two or more models differ, not where models overlap geographically but also where new models update existing ones. As modern PSHA incorporates increasingly complex analysis of epistemic uncertainty, the resulting hazard is represented not as a single value or spectrum but rather as probability distribution. Focusing on recent PSHA models for France and Germany, alongside the 2020 European Seismic Hazard Model, we explore the differences in model components and highlight the challenges and strategy for harmonising the different models into a common PSHA calculation software. We then quantify the differences in the source model and seismic hazard probability distributions using metrics based on information theory, illustrating their application to Upper Rhine Graben region. Our analyses reveal the spatial variation and complexity of model differences when viewed as probability distributions and highlight the need for more detailed transparency and replicability of the models when used as a basis for decision making and engineering design. Short Summary. New generations of seismic hazard models are developed with sophisticated approaches to quantify uncertainties in our knowledge of earthquake process. To understand why and how recent state-of-the-art seismic hazard models for France, Germany and Europe differ despite similar underlying assumptions, we present a systematic approach to investigate model-to-model differences and to quantify and visualise them while accounting for their respective uncertainties. [ABSTRACT FROM AUTHOR]

Current practice in strong ground motion modelling for probabilistic seismic hazard analysis (PSHA) requires the identification and calibration of empirical models appropriate to the tectonic regimes within the region of application, along with quantification of both their aleatory and epistemic uncertainties. For the development of the 2020 European Seismic Hazard Model (ESHM20) a novel approach for ground motion characterization was adopted based on the concept of a regionalized scaled backbone model, wherein a single appropriate ground motion model (GMM) is identified for use in PSHA, to which adjustments or scaling factors are then applied to account for epistemic uncertainty in the underlying seismological properties of the region of interest. While the theory and development of the regionalized scaled backbone GMM concept has been discussed in earlier publications, implementation in the final ESHM20 required further refinements to the shallow seismicity GMM in three regions, which were undertaken considering new data and insights gained from the feedback provided by experts in several regions of Europe: France, Portugal and Iceland. Exploration of the geophysical characteristics of these regions and analysis of additional ground motion records prompted re-calibrations of the GMM logic tree and/or modifications to the proposed regionalization. These modifications illustrate how the ESHM20 GMM logic tree can still be refined and adapted to different regions based on new ground motion data and/or expert judgement, without diverging from the proposed regionalized scaled backbone GMM framework. In addition to the regions of crustal seismicity, the scaled backbone approach needed to be adapted to earthquakes occurring in Europe's subduction zones and to the Vrancea deep seismogenic source region. Using a novel fuzzy methodology to classify earthquakes to according to different seismic regimes within the subduction system, we compare ground motion records from non-crustal earthquakes to existing subduction GMMs and identify a suitable backbone GMM for application to subduction and deep seismic sources in Europe. The observed ground motion records from moderate and small magnitude earthquakes allow us to calibrate the anelastic attenuation of the backbone GMM specifically for the Eastern Mediterranean region. Epistemic uncertainty is then calibrated based on the global variability in source and attenuation characteristics of subduction GMMs. With the ESHM20 now completed, we reflect on the lessons learned from implementing this new approach in regional scale PSHA and highlight where we hope to see new developments and improvements to the characterization of ground motion in future generations of the European Seismic Hazard Model. [ABSTRACT FROM AUTHOR]

To test whether a globally inferred sediment thickness value from geomorphological studies can be used as a proxy to predict earthquake site amplification, we derive site amplification models from the relation between empirical amplification for sites in Europe and Turkey, and the geomorphological sediment thickness. The new site amplification predictions are then compared to predictions from site amplification models derived using the traditional site proxies, V S30 inferred from slope, slope itself, as well as geological era and slope combined. The ability of each proxy to capture the site amplification is evaluated based on the reduction in site-to-site variability caused by each proxy. The results show that the highest reduction is caused by geological era and slope combined, while the geomorphological sediment thickness show a slightly larger or equal reduction in site-to-site variability as inferred V S30 and slope. We therefore argue that including geology and geomorphology in site amplification modelling on regional scale can give an important added value and that globally or regionally inferred models for soil and sediment thickness from fields beyond engineering seismology can have a great potential in regional seismic hazard and risk assessments. Furthermore, the differences between the site amplification maps derived from different proxies capture the epistemic uncertainty of site amplification modelling. While, albeit the different proxies predict similar features on a large scale, local differences can be large. This shows that using only one proxy when predicting site amplification does not capture the full epistemic uncertainty, which is demonstrated by looking into detail on the site amplification maps predicted for Eastern Turkey and Syria, where the devastating Kahramanmaras Earthquake Sequence occurred in February 2023. [ABSTRACT FROM AUTHOR]

To design user-centred and scientifically high-quality outreach products to inform about earthquake-related hazards and the associated risk, a close collaboration between the model developers and communication experts is needed. In this contribution, we present the communication strategy developed to support the public release of the first openly available European Seismic Risk Model and the updated European Seismic Hazard Model. The backbone of the strategy was the communication concept in which the overall vision, communication principles, target audiences (including personas), key messages, and products were defined. To fulfil the end-users' needs, we conducted two user testing surveys; one for the interactive risk map viewer and one for the risk poster with a special emphasis on the European earthquake risk map. To further ensure that the outreach products are not only understandable and attractive for different target groups but also adequate from a scientific point of view, a two-fold feedback mechanism involving experts in the field was implemented. Through a close collaboration with a network of communication specialists from other institutions supporting the release, additional feedback and exchange of knowledge was enabled. Our insights, gained as part of the release process, can support others in developing user-centred products reviewed by experts in the field to inform about hazard and risk models. Short Summary. For the release of (cross-country harmonised) hazard and risk models, a communication strategy co-defined by the model developers and communication experts is needed. The strategy should consist of a communication concept, a user testing, expert feedback mechanisms, and the establishment of a network with outreach specialists. Here we present our approach for the release of the European Seismic Hazard and Risk Models and provide practical recommendations for similar efforts. [ABSTRACT FROM AUTHOR]

Probabilistic seismic hazard estimates are a key ingredient of earthquake risk mitigation strategies and are often communicated through seismic hazard maps. Though the literature suggests that visual design properties are key for effective communication using such maps, guidelines on how to optimally design hazard maps are missing from the literature. Current maps use color palettes and data classification schemes which have well-documented limitations that may inadvertently miscommunicate seismic hazard. We surveyed the literature on color and classification schemes to identify design criteria that have empirical support for communicating hazard information. These criteria were then applied to redesign the seismic hazard map for Germany. We established several communication goals for this map, including essential properties about moderate-hazard seismic regions and a critical hazard threshold related to the German seismic building codes. We elucidate our redesign process and the selection of new colors and classification schemes that satisfy the evidence-based criteria. In a mixed-methods survey, we evaluate the original and redesigned seismic hazard maps, finding that the redesign satisfies all the communication goals and improves users' awareness about the spatial spread of seismic hazard relative to the original. We consider practical implications for the design of hazard maps across the natural hazards. [ABSTRACT FROM AUTHOR]

Multi-hazard risk assessments for building portfolios exposed to earthquake shaking followed by a tsunami are usually based on empirical vulnerability models calibrated on post-event surveys of damaged buildings. The applicability of these models cannot easily be extrapolated to other regions of larger/smaller events. Moreover, the quantitative evaluation of the damages related to each of the hazard types (disaggregation) is impossible. To investigate cumulative damage on extended building portfolios, this study proposes an alternative and modular method to probabilistically integrate sets of single-hazard vulnerability models that are constantly being developed and calibrated by experts from various research fields to be used within a multi-risk context. This method is based on the proposal of state-dependent fragility functions for the triggered hazard to account for the pre-existing damage and the harmonisation of building classes and damage states through their taxonomic characterisation, which is transversal to any hazard-dependent vulnerability. This modular assemblage also allows us to separate the economic losses expected for each scenario on building portfolios subjected to cascading hazards. We demonstrate its application by assessing the economic losses expected for the residential building stock of Lima, Peru, a megacity commonly exposed to consecutive earthquake and tsunami scenarios. We show the importance of accounting for damage accumulation on extended building portfolios while observing a dependency between the earthquake magnitude and the direct economic losses derived for each hazard scenario. For the commonly exposed residential building stock of Lima exposed to both perils, we find that classical tsunami empirical fragility functions lead to underestimations of predicted losses for lower magnitudes (Mw) and large overestimations for larger Mw events in comparison to our state-dependent models and cumulative-damage method. [ABSTRACT FROM AUTHOR]

Probabilistic seismic hazard estimates are a key ingredient of earthquake risk mitigation strategies and are often communicated through seismic hazard maps. Though the literature suggests that visual design properties are key for effective communication using such maps, guidelines on how to optimally design hazard map are missing from the literature. Current maps use color palettes and data classification schemes which have well-reported limitations that may inadvertently miscommunicate seismic hazard. We surveyed the literature on color and classification schemes to identify design criteria that have empirical support for communicating hazard information. These criteria were then applied to redesign the seismic hazard map for Germany. We isolated several communication goals for this map, including essential properties about moderate-hazard seismic regions and a critical hazard threshold related to the German seismic building codes. We elucidate our redesign process and the selection of new colors and classification schemes that satisfy the evidence-based criteria. In a mixed-methods survey, we evaluate the original and redesigned seismic hazard maps, finding that the redesign satisfies all the communication goals, and improves users' awareness about the spatial spread of seismic hazard, relative to the original. We consider practical implications for the design of hazard maps across the natural hazards. [ABSTRACT FROM AUTHOR]

Multi-hazard risk assessments for building portfolios exposed to earthquake shaking followed by a tsunami are usually based on empirical vulnerability models calibrated on post-event surveys of damaged buildings. The applicability of these models cannot easily be extrapolated to other regions of larger/smaller events. Moreover, the quantitative evaluation of the damages related to each of the hazards type (disaggregation) is impossible. To investigate cumulative damage on extended building portfolios, this study proposes an alternative and modular method to probabilistically integrate sets of single-hazard vulnerability models that are being constantly developed and calibrated by experts from various research fields to be used within a multi-risk context. This method is based on the proposal of state-dependent fragility functions for the triggered hazard to account for the pre-existing damage, and the harmonisation of building classes and damage states through their taxonomic characterization, which is transversal to any hazard-dependent vulnerability. This modular assemblage also allows us to separate the economic losses expected for each scenario on building portfolios subjected to cascading hazards. We demonstrate its application by assessing the economic losses expected for the residential building stock of Lima, Peru, a megacity commonly exposed to consecutive earthquake and tsunami scenarios. We show the importance of accounting for damage accumulation on extended building portfolios while observing a dependency between the earthquake magnitude and the losses derived for each hazard scenario. For the commonly exposed residential building stock of Lima exposed to both perils, we find that classical tsunami empirical fragility functions lead to an underestimation of predicted losses for lower magnitudes (Mw) and large overestimations for larger Mw events in comparison to our state-dependent models and cumulative damage method. [ABSTRACT FROM AUTHOR]

We propose the use of variable resolution boundaries based on central Voronoi tessellations (CVTs) to spatially aggregate building exposure models for risk assessment to various natural hazards. Such a framework is especially beneficial when the spatial distribution of the considered hazards presents intensity measures with contrasting footprints and spatial correlations, such as in coastal environments. This work avoids the incorrect assumption that a single intensity value from hazards with low spatial correlation (e.g. tsunami) can be considered to be representative within large-sized geo-cells for physical vulnerability assessment, without, at the same time, increasing the complexity of the overall model. We present decoupled earthquake and tsunami scenario-based risk estimates for the residential building stock of Lima (Peru). We observe that earthquake loss models for far-field subduction sources are practically insensitive to the exposure resolution. Conversely, tsunami loss models and associated uncertainties depend on the spatial correlations of the hazard intensities as well as on the resolution of the exposure models. We note that for the portfolio located in the coastal area exposed to both perils in Lima, the ground shaking dominates the losses for lower-magnitude earthquakes, whilst tsunamis cause the most damage for larger-magnitude events. For the latter, two sets of existing empirical flow depth fragility models are used, resulting in large differences in the calculated losses. This study, therefore, raises awareness about the uncertainties associated with the selection of fragility models and spatial aggregation entities for exposure modelling and loss mapping. [ABSTRACT FROM AUTHOR]

The propagation of a seismic rupture on a fault introduces spatial variations in the seismic wave field surrounding the fault. This directivity effect results in larger shaking amplitudes in the rupture propagation direction. Its seismic radiation pattern also causes amplitude variations between the strike-normal and strike-parallel components of horizontal ground motion. We investigated the landslide response to these effects during the 2016 Kumamoto earthquake (Mw 7.1) in central Kyushu (Japan). Although the distribution of some 1500 earthquake-triggered landslides as a function of rupture distance is consistent with the observed Arias intensity, the landslides were more concentrated to the northeast of the southwest–northeast striking rupture. We examined several landslide susceptibility factors: hillslope inclination, the median amplification factor (MAF) of ground shaking, lithology, land cover, and topographic wetness. None of these factors sufficiently explains the landslide distribution or orientation (aspect), although the landslide head scarps have an elevated hillslope inclination and MAF. We propose a new physics-based ground-motion model (GMM) that accounts for the seismic rupture effects, and we demonstrate that the low-frequency seismic radiation pattern is consistent with the overall landslide distribution. Its spatial pattern is influenced by the rupture directivity effect, whereas landslide aspect is influenced by amplitude variations between the fault-normal and fault-parallel motion at frequencies <2 Hz. This azimuth dependence implies that comparable landslide concentrations can occur at different distances from the rupture. This quantitative link between the prevalent landslide aspect and the low-frequency seismic radiation pattern can improve coseismic landslide hazard assessment. [ABSTRACT FROM AUTHOR]

The propagation of a seismic rupture on a fault introduces spatial variations in the seismic wavefield surrounding the fault during an earthquake. This directivity effect results in larger shaking amplitudes in the rupture propagation direction. Its seismic radiation pattern also causes amplitude variations between the strike-normal and strike-parallel components of horizontal ground motion. We investigated the landslide response to these effects during the 2016 Kumamoto earthquake (MW 7.1) in central Kyūshū (Japan). Although the distribution of some 1,500 earthquake-triggered landslides as function of rupture distance is consistent with the observed Arias intensity, the landslides are more concentrated to the northeast of the southwest-northeast striking rupture. We examined several landslide susceptibility factors: hillslope inclination, median amplification factor (MAF) of ground shaking, lithology, land cover, and topographic wetness. None of these factors can sufficiently explain the landslide distribution or orientation (aspect), although the landslide headscarps coincide with elevated hillslope inclination and MAF. We propose a new physics-based ground motion model that accounts for the seismic rupture effects, and demonstrate that the low-frequency seismic radiation pattern consistent with the overall landslide distribution. The spatial landslide distribution is primarily influenced by the rupture directivity effect, whereas landslide aspect is influenced by amplitude variations between the fault-normal and fault-parallel motion at frequencies < 2Hz. This azimuth-dependence implies that comparable landslide concentrations can occur at different distances from the rupture. This quantitative link between the prevalent landslide aspect and the low-frequency seismic radiation pattern can improve coseismic landslide hazard assessment. [ABSTRACT FROM AUTHOR]

The steady increase of ground-motion data allows new possibilities but also comes with new challenges in ground-motion modelling and the development of ground-motion prediction equations.One challenge is data selection - not all data can be processed equally - and we introduce an extension of the widely used mixed effect model, i.e. a model of fixed and random effects.The extension incorporates data selection by assigning weights per event and model parameters are inferred by weighting the record data accordingly.This advancement allows for the possibility of data classification beyond the more classical, expert-driven, binary classification based e.g. on the depth of the event, distance to the trench, style-of-faulting and dip angle of the fault plane.We apply ACE - (A)ngular (C)lassification with (E)xpectation-maximization - a method to efficiently identify clusters of nodal planes from focal mechanisms to differentiate between interface and intraslab type events.The classification is continuous, i.e. no event belongs completely to one class and classification uncertainty is then evaluated and taken into account in the ground-motion modelling.As an example, we developed a ground-motion prediction equation from a database of approximately 2400 records from 319 events in the Chilean subduction zone.Our ground-motion model with the data-driven and reproducible classification is comparable to the expert classification based model.Furthermore, the models show temporal variations of the between-event residuals before and after large earthquakes in the region. [ABSTRACT FROM AUTHOR]