Sampling dependent parameters in traffic simulation models with Gaussian copula

Making samples with certain marginal distributions and dependence structures is an essential but difficult step to perform sampling-based SA for traffic simulation models with dependent parameters. In this paper, we present a general approach for generating samples for dependent parameters. It utilizes the Gaussian copula in the sampling process, which makes it attractive for sampling parameters from any arbitrary marginal distribution. Furthermore, the Spearman’s rank correlation coefficient is employed instead of the traditional linear correlation coefficient, so that the dependence structure of the empirical data can be retained throughout the non-linear transform of the Gaussian copula. A case study that generates samples for the kinematic parameters of Wiedemann-74 car-following model is included to demonstrate the application of this approach. It has shown that the marginal distributions and correlation coefficients of the generated samples are comparable with that of the empirical data. Specifically, the 1,024 samples, which are generated by employing the Sobol sequence in the sampling process, also present consistent marginal distributions and correlation coefficients as the empirical data. This has demonstrated that the proposed sampling approach is also useful for making proper samples of computationally expensive models, for which a big number of model runs are not always affordable.