Application of MCMC-GSA model calibration method to urban runoff quality modeling

In stormwater quality modeling, estimating the confidence level in conceptual model parameters is necessary but difficult. The applicability and the effectiveness of a method for model calibration and model uncertainty analysis in the case of a four parameters lumped urban runoff quality model are illustrated in this paper. This method consists of a combination of the Metropolis algorithm for parameters' uncertainties and correlation assessment and a variance-based method for global sensitivity analysis. The use of the Metropolis algorithm to estimate the posterior distribution of parameters through a likelihood measure allows the replicated Latin hypercube sampling method to compute the parameters' importance measures. Calibration results illustrate the usefulness of the Metropolis algorithm in the assessment of parameters' uncertainties and their interaction structure. The sensitivity analysis demonstrates the insignificance of some parameters in terms of driving the model to have a good conformity with the data. This method provides a realistic evaluation of the conceptual description of the processes used in models and a progress in our capability to assess parameters' uncertainties.