Analysis of Model Uncertainty Using Bayesian Monte Carlo

Bayesian Monte Carlo analysis combines Bayesian inference with standard Monte Carlo techniques to provide improved estimates of the uncertainty in water quality model parameters and results. Bayesian Monte Carlo analysis was applied to two models: a chloride/total phosphorus model of Green Bay, Lake Michigan and a dissolved oxygen model of the Grand River near Grand Rapids, Michigan. The ability of Bayesian Monte Carlo to directly estimate cross-correlation between input parameters reduced model prediction uncertainty by a factor of three compared to a standard Monte Carlo application not considering cross-correlation. A secondary benefit of the technique is the ability to determine beforehand the effect on model uncertainty of better field data versus laboratory studies of model process coefficients.