Abstract
Monte Carlo simulation (MCS) has been commonly applied for uncertainty analysis of model predictions. However, when modelling a water distribution system under unsteady conditions, the computational demand of MCS is quite high even for a reasonably sized system. The aim of this study is to evaluate alternative approximation schemes and examine their ability to predict model prediction uncertainty with less computational effort. Here, MCS is compared with a point estimation method, the first-order second-moment (FOSM) method, and a quasi-MCS method, Latin hypercube sampling (LHS). Hydraulic and water quality simulations are performed using EPANET and the evaluated model outputs are nodal pressure, water age and chlorine concentration. Six input parameters, pipe diameter and roughness coefficient, nodal spatial and temporal demands and bulk and wall decay coefficients, are considered. To examine the effect of the magnitude of input uncertainty on model output, three uncertainty levels are evaluated. The study is performed for a real system with 116 pipes and 90 nodes. Results demonstrate that LHS provides very good estimates of the predicted output range for steady and unsteady conditions compared with MCS, while FOSM did well for steady conditions but poorly for some periods in the extended-period simulation for chlorine concentration.