{"title":"The Effect of Bayesian Updating in the Hazard Assessment of Submarine Landslides","authors":"Roneet Das, P. Varela, Z. Medina-Cetina","doi":"10.4043/29669-MS","DOIUrl":null,"url":null,"abstract":"\n This paper introduces a Bayesian methodology to conduct landslide hazard assessment. The proposed approach demonstrates how a probabilistic method can incorporate evolving information about a site for progressively more certain geotechnical characterization. The probabilistic method presented herein is called the Bayesian framework, which integrates a physics-based model defining certain characteristic or phenomenon related to the site, state of evidence on the model parameters, and experimental observations to produce an updated state of evidence on the model parameters and more confident model predictions. This study focuses on landslide geohazard of a site using the physics-based infinite block slope model to estimate the probability of submarine slope failure. The probability of failure against sliding is estimated using the predictions of the infinite slope model under static loading condition for different states of evidence on the model parameters. A state of evidence reflects the level of knowledge about a parameter which describes an attribute of the site such as bathymetry or geotechnical properties of the in-situ soil. This research studies the influence of increasing states of evidence on the confidence gain in model predictions and subsequent updates in the estimates of probability of failure. Predictions based on the infinite slope model are made using the Monte-Carlo algorithm through random sampling of the model parameters. The state of evidence on the model parameters is incorporated in the algorithm by considering the model parameters as random variables following a probability distribution function. These probability distributions, also known as the prior probability distributions, represent the initial state of evidence on the model parameters. The Bayesian framework is used to conduct sequential calibration of the infinite slope model using synthetically generated data on the shear strength of the in-situ soil. These experimental observations represent the state of evidence on the soil conditions. In this paper two sets of data containing 5 and 20 data ‘sample’ points, respectively are used to calibrate the infinite slope model. Calibration of the model results in an updated state of evidence on the model parameters and generates a new set of probability distributions known as the posterior probability distributions. The posterior distributions more accurately describe the potential range of value that the parameters can attain. Comparison between the model predictions based on the initial state of evidence and the updated states of evidence shows a gain in the certainty of the model predictions.","PeriodicalId":10968,"journal":{"name":"Day 3 Wed, May 08, 2019","volume":"197 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2019-04-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"2","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Day 3 Wed, May 08, 2019","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.4043/29669-MS","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 2
Abstract
This paper introduces a Bayesian methodology to conduct landslide hazard assessment. The proposed approach demonstrates how a probabilistic method can incorporate evolving information about a site for progressively more certain geotechnical characterization. The probabilistic method presented herein is called the Bayesian framework, which integrates a physics-based model defining certain characteristic or phenomenon related to the site, state of evidence on the model parameters, and experimental observations to produce an updated state of evidence on the model parameters and more confident model predictions. This study focuses on landslide geohazard of a site using the physics-based infinite block slope model to estimate the probability of submarine slope failure. The probability of failure against sliding is estimated using the predictions of the infinite slope model under static loading condition for different states of evidence on the model parameters. A state of evidence reflects the level of knowledge about a parameter which describes an attribute of the site such as bathymetry or geotechnical properties of the in-situ soil. This research studies the influence of increasing states of evidence on the confidence gain in model predictions and subsequent updates in the estimates of probability of failure. Predictions based on the infinite slope model are made using the Monte-Carlo algorithm through random sampling of the model parameters. The state of evidence on the model parameters is incorporated in the algorithm by considering the model parameters as random variables following a probability distribution function. These probability distributions, also known as the prior probability distributions, represent the initial state of evidence on the model parameters. The Bayesian framework is used to conduct sequential calibration of the infinite slope model using synthetically generated data on the shear strength of the in-situ soil. These experimental observations represent the state of evidence on the soil conditions. In this paper two sets of data containing 5 and 20 data ‘sample’ points, respectively are used to calibrate the infinite slope model. Calibration of the model results in an updated state of evidence on the model parameters and generates a new set of probability distributions known as the posterior probability distributions. The posterior distributions more accurately describe the potential range of value that the parameters can attain. Comparison between the model predictions based on the initial state of evidence and the updated states of evidence shows a gain in the certainty of the model predictions.