USING A BAYESIAN NETWORK APPROACH TO MODEL THE SYSTEM OF EFFECTS OF URBANIZATION ON AQUATIC ECOSYSTEMS

Watershed urbanization initiates a complex series of processes many of which lead to harmful consequences for stream biota. Though ecological conceptual models attempt to describe many of the simultaneous, interacting factors caused by urbanization that affect physical, chemical and biological aspects of stream ecosystems, this system-level understanding cannot be translated into a quantified representation using standard statistical modeling techniques. Traditional analysis of stream ecosystem data is limited to finding empirical relationships between pairs of environmental factors using simple regression techniques and does not incorporate the web of interconnected environmental variables, uncertainty characterization, or known ecological information about the system. In contrast, a network modeling approach can represent and parameterize the entire system of urbanization affecting aquatic invertebrates. We construct a Bayesian network model to characterize this system from prior expert knowledge, update this model with USGS EUSE (Effect of Urbanization on Stream Ecosystems) data, and evaluate the resulting model incorporating both sources of information. A Bayesian network model has the flexibility of being able to add in new data as it becomes available in a manner conducive to use in adaptive management. Managers can use the parameterized Bayesian network model to calculate the probabilities of attaining desired aquatic ecosystem goals assuming different levels of urban stress, environmental conditions and management options. This Bayesian approach enables aquatic ecologists to model a comprehensive set of interacting system components in an understandable, probabilistic manner. Many anthropogenic and natural factors affect invertebrates and, rather than investigating each factor individually, a Bayesian network is used to describe the interconnected effect while acknowledging the complexity of the environmental and ecological processes driving biological response, allowing concurrent assessment of all driving factors.