Using Bayesian model averaging to improve ground motion predictions

Bertin, Michaël; Marin, Sylvie; Millet, Christophe; Berge‐Thierry, Catherine

doi:10.1093/gji/ggz486

Cited by 8 publications

(12 citation statements)

References 56 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…we have selected all the GMPEs in Bertin et al ( 2020)'s study except Derras et al (2016) from OpenQuake Engine library (Pagani et al, 2014) and tested their performance against ESM database. Similar to Bertin et al (2020)'s study, the three GMPEs (Bindi2011, Akkar2014, and Bindi2014) are dominant compared to other GMPEs over a set of periods against ESM database. Therefore, in this research, we present the results only for the dominant three GMPEs.…”

Section: Application: Gmpe Modelssupporting

confidence: 59%

“…-Fixed parameter setting, f m (x) is fixed, with added bias term µ m and then estimate the weights. In Bertin et al (2020), fixed parameter setting with added bias term is considered.…”

Section: Stage 3: Model Updatingmentioning

confidence: 99%

“…In a recent study, Bertin et al (2020) presented the BMA approach to improve ground motion predictions for nine GMPE models using RESORCE-2013 database. The BMA approach was implemented using two numerical techniques, Markov Chain Monte Carlo (MCMC) method (Metropolis Hastings algorithm) and the maximum likelihood estimation (MLE) approach with around 1000 records issued from the RESORCE-2013 database (Akkar et al, 2014).…”

Section: Seismotectonic Modelsmentioning

confidence: 99%

“…In current practice, the seismic hazard curves are generated by combining all possible choices for SM and GMPE models using logic trees as shown in Figure 1. Various approaches such as Backbone models, Mixture models, Fuzzy logic, Bayesian inference have been used in the past to obtain the weights of SM and GMPE models in the logic tree (Bertin et al, 2020;Haendel et al, 2015;Pirchiner et al, 2013;Shayanfar et al, 2019;Viallet et al, 2017). In addition to estimation of weights, a unified approach is needed to integrate uncertainties arising from different sources and propagate them to the hazard curve.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

A Bayesian Approach to Estimate Weights for GMPE Models in a Logic Tree

Bodda¹,

Keller²,

Gupta³

et al. 2021

Preprint

View full text Add to dashboard Cite

In current practice, the seismic hazard curves are generated by combining all possible choices for seismotectonic model (SM), earthquake magnitude recurrence models, and ground motion prediction equation (GMPE) models using logic trees. However, studies have shown that significant uncertainties exist in the evaluation of the seismic hazard, especially in low-to-moderate seismicity regions, such as Central \& Eastern United States and Europe, due to the scarcity of the recorded earthquake data and lack of precise knowledge concerning the earthquake generation and propagation mechanisms. Specifically, these uncertainties lead to conservatisms in the seismic hazard which poses a considerable risk for nuclear facilities. Designing or upgrading facilities with such conservative estimates of seismic hazard results in large capital costs. In standard probabilistic seismic hazard assessment (PSHA) practices, the estimate of branch weights for the GMPE models is based on expert judgment and there is no holistic approach for estimating the weights under uncertainty. In this research, we focus on the GMPE part of the PSHA and we propose a statistical methodology that is based on Bayesian inference, to evaluate the GMPE model weights from a list of GMPE models in the logic tree. In this study, Bayesian model averaging (BMA) approach with Bayesian linear models (BLM) is employed to combine GMPE models and the performance of the BMA model is tested against European Strong-Motion (ESM) database. It is shown that the proposed methodology can calculate the weights at different time periods without any approximations using analytical formulas and it is computationally inexpensive.

show abstract

Section: Application: Gmpe Modelssupporting

confidence: 59%

“…-Fixed parameter setting, f m (x) is fixed, with added bias term µ m and then estimate the weights. In Bertin et al (2020), fixed parameter setting with added bias term is considered.…”

Section: Stage 3: Model Updatingmentioning

confidence: 99%

Section: Seismotectonic Modelsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

A Bayesian Approach to Estimate Weights for GMPE Models in a Logic Tree

Bodda¹,

Keller²,

Gupta³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…A strong motion dataset that was considered to be representative of the expected ground motion on site was extracted from the RESORCE database, to challenge the predictions of the different GMPEs against these observed data. The SINAPS@ study clearly showed that without any a priori information on the GMPEs, the Bayesian model averaging provided hierarchy and weighting of the GMPEs that was only based on their relevance with respect to real data (Bertin et al 2019). This finally provided an objective ranking that avoided any expert advice that might be questionable (e.g., on the choice of the candidate GMPEs, on their weight in a logic-tree), as shown in Fig.…”

Section: Bayesian Tool As An Objective Alternativementioning

confidence: 94%

Main Achievements of the Multidisciplinary SINAPS@ Research Project: Towards an Integrated Approach to Perform Seismic Safety Analysis of Nuclear Facilities

Berge‐Thierry¹,

Voldoire²,

López-Caballero³

et al. 2019

Pure Appl. Geophys.

View full text Add to dashboard Cite

This contribution provides an overview of the SINAPS@ French research project and its main achievements. SINAPS@ stands for ''Earthquake and Nuclear Facility: Improving and Sustaining Safety'', and it has gathered the broad research community together to propose an innovative seismic safety analysis for nuclear facilities. This five-year project was funded by the French government after the 2011 Japanese Tohoku large earthquake and associated tsunami that caused a major accident at the Fukushima Daïchi nuclear power plant. Soon after this disaster, the international community involved in nuclear safety questioned the current methodologies used to define and to account for seismic loadings for nuclear facilities during the design and periodic assessment review phases. Within this framework, worldwide nuclear authorities asked nuclear licensees to perform 'stress tests' to estimate the capacity of their existing facilities for sustaining extreme seismic motions. As an introduction, the French nuclear regulatory framework is presented here, to emphasize the key issues and the scientific challenges. An analysis of the current French practices and the need to better assess the seismic margin of nuclear facilities contributed to the shaping of the scientific roadmap of SINAPS@. SINAPS@ was aimed at conducting a continuous analysis of completeness and gaps in databases (for all data types, including geology, seismology, site characterization, materials), of reliability or deficiency of models available to describe physical phenomena (i.e., prediction of seismic motion, site effects, soil and structure interactions, linear and nonlinear wave propagation, material constitutive laws in the nonlinear domain for structural analysis), and of the relevance or weakness of methodologies used for seismic safety assessment. This critical analysis that confronts the methods (either deterministic or probabilistic) and the available data in terms of the international state of the art systematically addresses the uncertainty issues. We present the key results achieved in SINAPS@ at each step of the full seismic analysis, with a focus on uncertainty identification, quantification, and propagation. The main lessons learned from SINAPS@ are highlighted. SINAPS@ promotes an innovative integrated approach that is consistent with Guidelines #22, as recently published by the French Nuclear Safety Authority (Guidelines ASN #22 2017), and opens the perspectives to improve current French practice.

show abstract

A Methodological Approach to Update Ground Motion Prediction Models Using Bayesian Inference

Bodda

Keller

Gupta

et al. 2021

Pure Appl. Geophys.

View full text Add to dashboard Cite

In recent decades, prediction of ground motion at a specific site or a region is of primary interest in probabilistic seismic hazard assessment (PSHA). Historically, several ground motion prediction equation (GMPE) models with different functional forms have been published using strong ground motion records available from NGA-West and European databases. However, low-to-moderate seismicity regions, such as Central & Eastern United States and western Europe, is characterized by limited strong-motion records in the magnitude-distance range of interest for PSHA. In these regions, the available data for the development of empirical GMPEs is very scarce and limited to small magnitude events. For these regions, the general practice in PSHA is to consider a set of GMPEs developed from data sets collected in other regions with high seismicity. This practice generates an overestimation of the seismic hazard for the low seismicity regions. There are two potential solutions to overcome this problem: (i) a new GMPE model can be developed; however, development of such a model can require significant amount of data which is not usually available, and (ii) the existing GMPE models can be recalibrated based on the data sets collected in the new region rather than developing a new GMPE model. In this paper, we propose a methodological approach to recalibrate the coefficients in a GMPE model using different algorithms to perform Bayesian inference. The coefficients are recalibrated for a subset of European Strong-Motion (ESM) database that corresponds to low-to-moderate seismicity records. In this study, different statistical models are compared based on the functional form given by the chosen GMPE, and the best model and algorithm are recommended using the concept of information criteria.

show abstract

Using Bayesian model averaging to improve ground motion predictions

Cited by 8 publications

References 56 publications

A Bayesian Approach to Estimate Weights for GMPE Models in a Logic Tree

A Bayesian Approach to Estimate Weights for GMPE Models in a Logic Tree

Main Achievements of the Multidisciplinary SINAPS@ Research Project: Towards an Integrated Approach to Perform Seismic Safety Analysis of Nuclear Facilities

A Methodological Approach to Update Ground Motion Prediction Models Using Bayesian Inference

Contact Info

Product

Resources

About