Sensitivity Analysis of a Water-Quality Model Using Latin Hypercube Sampling

Abstract

Decisions about water pollution control and management in river systems are increasingly based on predictions made with water quality models. Many uncertainties are associated with the development and calibration of such models and thus affect the model predictions. Therefore it often is necessary to perform a model reliability analysis prior to or as a supplement to model application to water pollution control and planning problems. In this paper, only the effect of parameter uncertainties on model prediction is investigated by applying sensitivity and uncertainty analyses. The Latin hypercube sampling technique in combination with regression and correlation analyses has been applied to the DUFLOW model developed for the Dender River in Belgium. These analyses aim to (1) identify the model parameters that significantly affect the model output uncertainty; (2) define the variables that should be measured for a successful model calibration; and (3) examine how model sensitivity changes with changing level of waste treatment. The results obtained indicate that three parameters related to the algal growth process have the greatest effect on the uncertainty of the simulated dissolved oxygen concentrations in the Dender River.