Abstract

The complexity of pollutant-mixing mechanism in open channels generates large uncertainty in estimation of longitudinal and lateral dispersion coefficients (K_x and K_y). Therefore, K_x and K_y estimation in rivers should be accompanied by an uncertainty analysis, a subject mainly ignored in previous studies. We introduce a method based on thorough analysis of different calibration datasets, resampled from a global database of tracer studies, to determine the uncertainty associated with five applicable intelligent models for estimation of K_x and K_y (model tree, evolutionary polynomial regression (EPR), gene-expression programming, multivariate adaptive regression splines (MARS), and support vector machine (SVM)). Our findings suggest that SVM gives least uncertainty in both K_x and K_y estimation, while EPR and MARS generate most uncertainty in K_x and Ky estimation, respectively. By considering significant uncertainty in the model estimations, we suggest that the methodology we introduce here for uncertainty determination of the models be incorporated in empirical studies on estimation of K_x and K_y in rivers.