ANALYSES OF TURBIDITY AND ACOUSTIC BACKSCATTER SIGNAL WITH ARTIFICIAL NEURAL NETWORK FOR ESTIMATION OF SUSPENDED SEDIMENT CONCENTRATION

Abstract

The commonly used sampling method is restrictive for the spatial and temporal measurement of suspended sediment and requires intensive labor. These limitations and technological advances have led to methods based on sound or light scattering in water. In this study, the turbidity and acoustic backscattering signal (ABS) values were used with the aim of improving these methods with different artificial neural network (ANN) models; Multilayer Perceptron (MLP), Radial Basis Neural networks (RBNN) and General Regression Neural Network (GRNN). Measurements were taken in a vertical sediment tower for two different sediment sizes (< 50 mu m and 50-100 mu m) and concentrations (0.0-6.0 g L-1). In the results of the regression analyses, turbidity values had strong relationships with sediment concentration for both sediment size groups (R-2 = 0.937 and 0.967). Although the ABS values had a reasonable R-2 value (0.873) for the 50-100 mu m group, the < 50 mu m group did not produce a significant R-2 value with regression analyses. The remarkable differences were not observed among MLP, RBNN and GRNN model for this sediment size group, and the reasonable R-2 and RMSE results were not produced with any ANN model that had a single ABS input for the < 50 mu m sediment group. On the other hand, for the other sediment group (50-100 mu m), ABS values were used as a single input, and the highest R-2 (0.917) value was obtained with MLP model and it was improved with the turbidity input (up to R-2 = 0.999). The results show that the ANN model could be considered as an alternative method because it was applied successfully to estimate suspended sediment concentration using with turbidity and ABS under different particle size conditions.