Effects of Demand, Mixing Fraction, and Rate Coefficient Uncertainty on Water Quality Models

The effects uncertainty on water quality predictions by the free water distribution modeling software created by the US Environment Protection Agency (EPANET2) are assessed using both design of experiments (DOE) and Monte Carlo simulation (MCS). For the test system evaluated, the DOE approach describes the effects of uncertainty with as few as eight model executions and yet shows good agreement with the more computationally demanding MCS approach. Surprisingly, first order disinfectant decay models are less sensitive to uncertainty than water age, even though an additional rate coefficient is required. This is likely because disinfectant residuals are bounded between the input concentration and zero, while water age is not. At nodes affected by water storage, up to 40% of the variability in disinfectant residual is attributed to uncertainty in tank mixing fraction (TMF). Due to seasonal and diurnal changes in water temperature, density, and mixing behavior, the TMF is inherently uncertain. Strategies to address this include MCS or sensitivity analysis using DOE. Inclusion of a wall decay coefficient appears to increase model sensitivity to uncertainty.