Development of a Saginaw Bay Probability Network Model

Primary Investigator:

Craig Stow - NOAA/GLERL

NOAA Research Area:

Advancing understanding of ecosystems to improve resource management

Performance Objective:

Increase number of regional coastal and marine ecosystems delineated with approved indicators of ecological health and socioeconomic benefits that are monitored and understood.

Research Milestones:

Meet annual targets for the number of coastal, marine, and Great Lakes ecological characterizations that meet management needs.

Executive Summary of Rationale

Proposed Work

Current/Ongoing

Refinement of the current conceptual Bayes net model and data analysis to quantify the model nodes.

Scientific Rationale

Water quality models are widely used to guide management decisions regarding allowable pollutant levels in aquatic ecosystems. Because they incorporate uncertainty in a rigorous probabilistic framework, Bayesian probability networks, or Bayes nets (Reckhow 1999), provide an alternative to the more traditional deterministic approach. Bayes nets are causal flow networks that can be concisely expressed as directed acyclic graphs (Pearl 2000, Shipley 2000) with nodes depicting ecosystem processes and arrows depicting the conditional dependencies of these processes. Each dependency indicated by an arrow represents a conditional probability distribution that describes the relative likelihood of each value of the down-arrow node, conditional on every possible combination of values of the parent nodes. A node that has no incoming arrows is said to have no parents, and such a variable can be described probabilistically by a marginal (or unconditional) probability distribution. The graphical network therefore constitutes a description of the probabilistic relationships among the system’s variables that amounts to a factorization of the joint distribution of all variables into a series of marginal and conditional distributions. The Bayes net framework is extremely flexible – probabilistic information for the processes at each node can be derived in several ways:

From detailed, process-based (mechanistic) models
From data-based (empirical) models
From expert elicitation (Morgan and Henrion 1990)
Any combination of the three previous sources

Additionally, several modern Bayesian software programs, such as HUGIN, have “learning” capabilities; model structure and relationships can be estimated from structure and patterns discernible in the available data (Jensen and Nielsen 2007).

Governmental/Societal relevance

Problems associated with declining water levels, continued phosphorus inputs and dreissenid interactions are a major concern in Saginaw Bay. This model will help identify logical management alternatives to alleviate the current problems.

Relevance to Ecosystem Forecasting

Relevance to Ecosystem Forecasting: This activity will lead to the development of a state of the art probabilistic model for ecosystem forecasting.