Uncertainty in Coupled Nonpoint Source and Stream Water-Quality Models

Abstract

Latin hypercube sampling (LHS) and the mean first-order reliability method are applied to determine the parameters significantly affecting uncertainty in the simulated dissolved oxygen (DO) concentrations at the point in the Zenne River in Brussels, most prone to low concentrations. The DO simulation involves the coupling of a nonpoint pollution load model, a constant treatment efficiency model, and a river water-quality model. LHS found that six of the 53 model parameters significantly contribute to the variance of the annual mean DO concentration, and five model parameters significantly contribute to the variance in the number of hours that DO concentrations are <2 mg/L. The mean first-order reliability method found that these parameters accounted for 73.6 and 84.7% of the variance in the respective output features. Reanalysis with LHS, letting only these six parameters be uncertain, confirmed the identification of these parameters as accounting for 78.1 and 83.0% of the respective variances. The identification of key sources of uncertainty provided insight with respect to treatment plant operation, model improvements, and data collection programs.