Combined Estimation of Hydrogeologic Scenario, Model, and Parameter Uncertainty with Application to Groundwater Reactive Transport Modeling
Groundwater contamination has been a serious health and environmental problem in many areas over the world nowadays. Groundwater reactive transport modeling is vital to make predictions of future contaminant reactive transport. However, these predictions are inherently uncertain, and uncertainty is one of the greatest obstacles in groundwater reactive transport.
We propose a Bayesian network approach for quantifying the uncertainty and implement the network for a groundwater reactive transport model for illustration. In the Bayesian network, different uncertainty sources are described as uncertain nodes. All the nodes are characterized by multiple states, representing their uncertainty, in the form of continuous or discrete probability distributions that are propagated to the model endpoint, which is the spatial distribution of contaminant concentrations.
After building the Bayesian network, uncertainty quantification is conducted through Monte Carlo simulations to obtain probability distributions of the variables of interest. In this study, uncertainty sources include scenario uncertainty, model uncertainty, parameter uncertainty. Variance decomposition is used to quantify relative contribution from the various sources to predictive uncertainty. While these new developments are illustrated using a relatively simple groundwater reactive transport model, our methods is applicable to a wide range of models. The results of uncertainty quantification are useful for environmental management and decision-makers to formulate policies and strategies.