A Case Study on Quantile and Percentile Regression to Select Design Lines for Complex, Real Work Profiles

Abstract

Site characterisation necessarily relies on engineering judgement, usually combined with some level of statistics to define characteristic values for design purposes. A suitable method for this task is quantile regression, which allows for the definition of lower, upper, and best-estimate characteristic values. The application of quantile regression to homogeneous profiles is relatively straightforward. Although such sites are common in some areas, there is need for a more comprehensive approach to quantile regression that covers the more general scenario of heterogeneous stratified profiles. This paper takes piezocone penetrometer data from a relatively complex seabed site and demonstrates the streamlined application of quantile regression, highlighting and analysing some of the assumptions and choices behind the approach. The work shows the nuances of the method and suggests workarounds for potential scenarios where its application may be challenging.