Multiobjective Design of Water-Quality Monitoring Networks in River-Reservoir Systems

This study develops and tests a method for multiobjective optimization of water-quality monitoring networks in river-reservoir systems. The optimization method identifies optimal sampling locations to detect the sudden release of contaminants to a reservoir, and meets two objectives: (1) minimizing the prediction error of methyl tert-butyl ether (MTBE) at the reservoir's outlet valve; and (2) minimizing the average time during which MTBE is detected at sampling locations. The optimization method considers 36 contaminant scenarios, corresponding to three volumes of contaminant release, three release locations, and four different seasonal release times. The MTBE pollutant chemograph was simulated at the outlet valve of the reservoir and at 16 possible sampling locations with the CE-QUAL-W2 model for each of the 36 scenarios of contaminant release. A support vector regression (SVR) tool is coupled to the nondominated sorting genetic algorithm II (NSGAII) to optimize the water-quality sampling locations. Implementation of the NSGAII-SVR method demonstrates its capacity to design water-quality monitoring networks that meet multiple objectives in a river-reservoir system. (C) 2016 American Society of Civil Engineers.