Sensitivity Analysis of Empirical and Data-Driven Models on Longitudinal Dispersion Coefficient in Streams

Abstract

Longitudinal dispersion coefficient (LDC) is a key element in pollutant transport modeling in streams. Several empirical and data-driven models have been proposed to evaluate this parameter. In this study, sensitivity analysis was performed on four key parameters affecting the LDC including: channel width, flow depth, mean flow velocity and shear velocity. In addition, Monte Carlo simulation was used to generate new datasets and evaluate performance of LDC estimation methods based on uncertainty of input parameters. Sensitivity indices of the input parameters in selected empirical equations and differential evolution model follow almost the same trend, where mean flow velocity is the most sensitive parameter among input parameters and the prediction accuracy depends heavily on the value of this parameter. In above mentioned models, shear velocity had a negative value and a reverse effect on LDC estimation. Channel width and mean flow velocity have the highest sensitivity in M5 model for narrow and wide streams, respectively. Based on sensitivity indices, the efficiency of empirical and data-driven models in different conditions, according to uncertainties in the input parameters, has been investigated. Result of LDC estimation based on the data of Monte Carlo simulation, showed that most LDC estimation models have a high uncertainty for upper LDC values.