LAUSR

Abstract A recent trend of geotechnical correlation analysis is to construct a multivariate distribution model to describe the multiple dependency characteristics of different soil parameters. In this study, the feasibility of this multivariate method to model the multivariate dependency of seven soil parameters including the undrained shear strength ratio, overconsolidation ratio, sensitivity, normalized cone tip resistance, normalized sleeve frictional resistance, pore pressure parameter, and normalized shear wave velocity.is studied. A database containing 372 sets of observations of these seven parameters is collected from 25 sites in the Jiangsu province, China. Two data transformations including the Johnson system of distribution and Box-Cox method were used in the construction of the multivariate model for the Jiangsu cohesive soils. The correlations derived from these two local models were compared with the compiled Jiangsu database, data points at other testing sites in China, and database and correlations published in the literature for validation. It was shown that the local models developed in this study were capable of the multivariate dependency of the soil parameters for the Jiangsu cohesive soils, whereas the global model developed in the literature tends to overestimate the uncertainties associated with the design parameters for the Jiangsu soils. A potential explanation is that the soil parameters in a local region were less variable with stronger correlations than those over the global sites, due to the site-specific effect. However, caution shall be taken if the local models for the Jiangsu soils are used at other sites. Besides, the Johnson system of distribution is superior to the Box-Cox transformation in the construction of the multivariate distribution model in the sense that the former is more effective to achieve the standard normal distribution, especially for the soil parameters with a bounded distribution.