2016 年 72 巻 4 号 p. I_715-I_720
The use of a turbidimeter with a particle size-dependent defect is proposed for the estimation of particle size as well as continuous estimation of fine sediment flux by particle size. In this study, an artificial neural network (ANN) and a multiple regression model were built to predict the fine sediment concentration of sand (FSCS) and of silt and clay (FSCSiC) using observation data from the Mimikawa River, where three dams are being planned to implement an integrated sluicing operation.
Both models, which used turbidity, discharge, and hysteresis as predictor variables, were able to generate excellent FSCS and FSCSiC estimates. In particular, estimation of the FSC by the ANN model, which suit for the non-linear system, may reduce the error of estimation by more than 40% of traditional sediment rating curve model. Furthermore, resulting predictions of FSCS and FSCSiC during storm events indicated that both fractions changed every hour. The developed methodology was used to simultaneously obtain reliable and continuous estimates of FSC with respect to particle size.