A graphical probability network model to support water quality decision making for the Neuse River estuary, North Carolina

I present a framework for policy-oriented environmental modeling based on a Bayesian probability network. The graphical model explicitly represents cause-and-effect assumptions between system variables that may be obscured under other modeling approaches. The probabilistic nature of the predictions promotes risk-based decision making and facilitates prioritization of future research and monitoring efforts. I demonstrate the development of a probability network using the problem of eutrophication in the Neuse River estuary, North Carolina. The approach allowed me to factor the complex causal chain linking management actions to ecological consequences into an articulated sequence of conditional relationships. Each of these relationships was then quantified independently using an approach suitable for the type of information available. For example, in modeling the oxygen dynamics of the estuary, a simple process-based mathematical model was formulated that was parameterized using historical data. Results indicated that the duration of bottom water hypoxia in the Neuse is controlled by the balance between the sediment oxygen demand and the frequency of wind-induced vertical mixing events. Because sufficient site-specific data did not exist to characterize how sediment oxygen demand would change in response to changes in algal productivity, cross-system data from a number of estuaries was gathered and used in a hierarchical framework to estimate model parameters. Results suggested that a similar set of aggregate scale mechanisms control the rate of sediment oxygen demand in most of the world's estuaries and coastal zones. When relevant theoretical models and quantitative data of any type did not exist to characterize a model relationship, I relied on the formally elicited judgment of an expert. This was done, for example, to link oxygen concentrations to fish and shellfish health and abundance. Such an approach provides a practical alternative for situations in which the linkage among variables is too complex to describe using fine-scale process description. After linking the multiple relationships, I describe how probability networks provide a useful framework for environmental decision making. I explain how to address frequency-based water quality standards using probabilistic model predictions, and I demonstrate the explicit selection of a margin of safety in setting policy, based on estimates of predictive uncertainty. By separately propagating the effects of natural variability and knowledge uncertainty, the model can be used to assess the degree to which proposed scientific research will improve predictions, thereby reducing the required margin of safety. Results indicate that, for the Neuse, investigating the effects of river flow and nitrogen inputs on algal density is likely to have the greatest benefit. Finally, I describe how probability networks can be updated based on the results of such studies, thus facilitating the ongoing process of adaptive environmental management.