Prediction of Longitudinal Dispersion Coefficient in Natural Channels Using Soft Computing Techniques


Department of Civil Engineering,Santa Cruz


Accurate estimate of longitudinal dispersion coefficient is essential in many hydraulic
and environmental problems such as intake designs, modeling
ow in esturies and risk assessment of
injection of hazardous pollutants into river
ows. Recent research works show that in the absence of
knowledge about explicit relationships concerning longitudinal dispersion coefficient and its in
parameters, data driven techniques can be used to predict it with reasonable degree of accuracy. In this
paper, the usefulness of Support Vector Machines (SVM) and Genetic Programming (GP) are examined
for predicting longitudinal dispersion coefficient in natural channels. The hydraulic variables such as

ow depth (H),
ow velocity (U) and shear velocity (u) along with the width of channel (B) are used
as input variables to predict longitudinal dispersion coefficient (Kx). The performance evaluation based
on multiple error criteria confirm that GP shows remarkably good performance in capturing non-linear
relationship between the predictors and predictant in the estimation of longitudinal dispersion coefficient
when compared with empirical approaches, the traditional Artificial Neural Networks (ANN) and SVM.
Hence GP can be used as an eficient computational paradigm in the prediction of longitudinal dispersion
coeficient in natural channels.