Abstract
The water quality of the Dongjin River deteriorates during the irrigation period because the supply of river maintenance water to the main river is cut off by the mass intake of agricultural weirs located in the midstream regions. A physics-based model and a data-driven model were used to predict the water quality in the Dongjin River under various hydrological conditions. The Hydrological Simulation Program–Fortran (HSPF), which is a physics-based model, was constructed to simulate the biological oxygen demand (BOD) in the Dongjin River Basin. A Gamma Test was used to derive the optimal combinations of the observed variables, including external water inflow, water intake, rainfall, and flow rate, for irrigation and non-irrigation periods. A data-driven adaptive neuro-fuzzy inference system (ANFIS) model was then built using these results. The ANFIS model built in this study was capable of predicting the BOD from the observed hydrological data in the irrigation and non-irrigation periods, without running the physics-based model. The predicted results have high confidence levels when compared with the observed data. Thus, the proposed method can be used for the reliable and rapid prediction of water quality using only monitoring data as input.
Highlights
The Saemangeum project in Korea, pursued for over three decades, involves building the longest sea dike in the world for the reclamation of land and lakes with the goals of expanding land area, developing water resources, and providing extra land for farming
The water quality of the Dongjin River deteriorates during the irrigation period due to the mass intake of river maintenance water at the Nakyang Weir, a diversion weir located in the midstream region of the Dongjin River
With regard to the input variables, the discharges of the Seomjin River Dam (Unam and Chilbo), Sanseong water intake, Nakyang Weir water intake (Gimje and Jeongeup irrigation canals), rainfall, and flow rate (t), and the input variables of one day before (t − 1) and one day after (t + 1) were used. These input variables were selected considering the usability of the data-driven model because the corresponding observation data were generated every day, and when they were input into the data-driven model, the water quality (BOD) for Dongjin River 3 point could be predicted without running the Hydrological Simulation Program–Fortran (HSPF) model
Summary
The Saemangeum project in Korea, pursued for over three decades, involves building the longest sea dike in the world for the reclamation of land and lakes with the goals of expanding land area, developing water resources, and providing extra land for farming. Najah et al [7] used an ANN to predict dissolved solids, electrical conductivity, and turbidity These studies constructed data-driven models based on actual observed data obtained on a monthly timescale or longer. We propose an efficient technique for predicting the water quality of the Dongjin River Basin, which combines the advantages of physics-based and data-driven models (Figure 1). The Adaptive Neuro-Fuzzy Inference System (ANFIS) model, which is a data-driven model, was constructed to maximally reflect the variability of the water quality according to various hydrological conditions, enabling rapid and accurate water quality predictions. Dongjin River 3 is a river management monitoring station situated upstream of the Saemangeum Lake, wherein the target water quality, in terms of biological oxygen demand (BOD), must be achieved within a specific planned year. 10 years of data from 2008 to 2017 were used, and the applicability of the data-driven model developed in this study was assessed using the HSPF simulation results for 2018
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
