Continuous measurement of suspended sediment concentrations based on image and environmental variables

The continuous measurement of the suspended sediment concentration (SSC) is essential for understanding sediment transport dynamics in waterway engineering. However, a single laboratory measurement for SSC is time-consuming and continuous measurements are difficult to realize in the field. This study developed predictive models combining laboratory measurement and image processing for the continuous measurement of SSC. The results revealed that factors the correlated with SSC were visibility (V), temperature (T), non-highlight areas of the green variable (G_n), green variable of whole region (G), non-highlight areas of the red variable (R_n), red variable of whole region (R), non-highlight areas of the gray variable (Gray_n), and gray variable of whole region (Gray), and that their factor loadings were 0.980, 0.980, 0.846, 0.938, 0.946, 0.939, 0.922, and 0.943, respectively. The predictive model for SSC used temperature and visibility as key variables with a coefficient of determination (R²) of 0.760. The model was applicable for SSC of 0.3–0.8 g/L. When color variables were used to build SSC prediction models, the R² values were 0.766, 0.539, and 0.389 for RF, SVR, and BPNN, respectively. However, when environmental variables were used as inputs for the artificial intelligence models, the R² were 0.853, 0.792, and 0.820 for RF, SVR, and BPNN, respectively. This study demonstrated that the models based on color variables and environmental variables contributed to the continuous measurement of SSC in the field.