Modelling runoff and sediment rate using a neuro-fuzzy technique

This paper demonstrates the estimation and simulation of discharge and sediment concentration for two river basins in the USA and India. The first-order Sugeno fuzzy inference system was utilised to model the stage, discharge and sediment concentration relationship. A subtractive clustering algorithm, along with a least-squares estimation, was used to generate the fuzzy rules that describe the relationship between input and output data of stage, discharge and sediment concentration, which change over time. The fuzzy rules were tuned by a back-propagation algorithm. The results are illustrated using simulation and virtual reality. A comparison was made between the estimates provided by the neuro-fuzzy model and a multi-linear regression model. Different statistical criteria were used to evaluate the performance of both models in estimating discharge and sediment concentration. Comparison of the results reveals that, in general, the neuro-fuzzy model gives better estimates than the multi-linear regression model in terms of root mean square and sum of squares errors. Furthermore, compared with the multi-linear regression model, the neuro-fuzzy model yields statistical properties of estimates that are closer to actual historical data.