Prediction of Water Quality Parameters at the Confluence of Nakdong and Kumho Rivers using Artificial Neural Network Model

Abstract

The pollutants in the tributaries with sufficient flow have a great influence on the water quality of the river after confluence. Stream confluences are elements of river networks that play a major role in the dynamics of fluvial systems. Substantial changes in flow hydrodynamics generally occur immediately downstream of confluence, and mixing of tributary flows may extend many kilometers downstream of confluences. Up-to-data investigations of the flow physics at river confluences rely primarily on physically-based numerical modeling. However, verification through field experiments is essential when a physics-based numerical model is applied to analyze the behavior of contaminants flowing through the tributaries downstream of the confluence. In addition, time and money should be invested in model construction and operation. However, in the case of data-based model, it is possible to make predictions with only accumulated data. Among the data-driven model, the ANN model is often used for application of water quality prediction. Many researchers used ANN technique to predict water quality parameters in river systems. In this study, the ANN ensemble model with resilient propagation method was developed to predict the water quality parameters at the confluence of Nakdong and Kumho Rivers. The data of EC tracing data conducted in 2015 were used to accurately understand the behavior of contaminants after confluence. The best fitted prediction results is shown when using water quality values of mainstream and tributary both as the input data (0.56 r squared value of pH, 0.75 of DO, 0.80 of EC, 0.66 of Chl-a). The point where the best fitted prediction results shown is ARCWQ-2 and ARCWQ-3 (where the transverse mixing was completed.). And the improvement rate was also the largest at the same case and point for pH, EC, and EC(22% of pH, 77% of DO, 26% of EC and 19% of Chl-a).