Investigation of Tree Models Performance for Estimation of Longitudinal Dispersion Coefficient in Straight River

Modeling pollution transmission in rivers is an important subject in environmental studies. Longitudinal dispersion coefficient is one of the key factors in the modelling of lateral dispersion of pollutants. Several researchers have attempted to estimate this coefficient using empirical and semi-empirical methods. However, robust models that can accurately estimate longitudinal dispersion coefficient in river streams are still required. In this study, data driven models were developed using the hydraulic and geometric parameters of rivers. The classification and regression tree (CART), M5 and genetic programming (GP) were used for this purpose. The models performances were then compared quantitatively with those of existing ones using accuracy parameters such as root mean square error (RMSE), mean absolute error (MAE) and discrepancy ratio (DR). The results illustrated that data driven models outperform the existing formulae in term of accuracy. CART model outperform other models in training step, but its performance decrease for testing data. M5 and GP models have RMSE of 0.41 and 0.44 and accuracy of 61% and 62%, respectively. According to small difference between M5 and GP performances, and simple structure of M5 algorithm, this model can be used for estimating longitudinal dispersion coefficient in streams.