Non-parametric simulation of geotechnical variability

Abstract

The application of bootstrap resampling, a non-parametric simulation technique, to the analysis of uncertainty associated with geotechnical data is investigated. This technique is presented as an alternative to traditional statistical data analysis. An extensive geotechnical database from the Heber Road site, Imperial Valley, Southern California, USA, is utilised as a case example. The database consists of many cone penetration and flat dilatometer tests. Linear correlation between the two types of test results and their vertical autocorrelation properties are analysed using bootstrap resampling, and the outcomes are compared with the same characteristics obtained from direct statistics of the whole database. The convergence and accuracy of bootstrap estimates, as well as their sensitivity to sample size, are studied and discussed. For the type of application investigated through this example, non-parametric bootstrap resampling appears a reliable and robust technique of geotechnical uncertainty analysis. Its main advantage is that statistical estimators of geotechnical properties and associated confidence levels can be computed directly from the empirical data without parametric assumptions on the population distribution.