Characterization of parameters in mechanistic models: A case study of a PCB fate and transport model

As a first step in a Bayesian analysis of PCB fate and transport in the upper Hudson River, a joint probability density function for parameters in a simulation model is created. The density function describes the joint probabilities of the following parameters: the anaerobic dechlorination rate constant, the volatilization rate constant, the aerobic biodegradation rate constant, the sedimentation rate, and the contaminated sediment depth. Difficulties in forming this probability density function are shown to result from problems with extrapolating data from the laboratory to the field, non-stationarity and aggregation, extrapolating information and analyses from other sites, and bias due to study design. These difficulties result in a density function characterized by high variances, and imply that predictions from this simulation model, and similarly large fate-and-transport models, are apt to be highly uncertain. Bayesian analysis is proposed as a rigorous mathematical technique for including observational data in density function generation in order to reduce prediction uncertainty.