Bayesian Update Method for Contaminant Source Characterization in Water Distribution Systems

Abstract

Bayesian analysis has application to probabilistic source characterization in water distribution systems. A new implementation of Markov-chain Monte Carlo (MCMC) for this problem is described. The solution addresses the discrete nature of water distribution networks that precludes the application of MCMC methods of general applicability that have been reported elsewhere in the water resources literature. The method is applied to a hypothetical network that has been used by others to test source identification methods. The likelihood function, a key component of Bayes’ rule, is evaluated using a Monte Carlo–based stochastic water-demand model. The results reinforce the need to address the multiple sources of uncertainty in the source characterization, including the stochastic variation of water demand. Further research is needed to make the approach feasible in operational environments. Limitations of the approach and future research directions are discussed.