A Bayesian network for prioritizing restoration of aquatic connectivity

Graduation date: 2011Re-establishing connectivity is a primary restoration activity for enhancing the\ud recovery of migratory fishes, but actions are often limited by lack of funds and\ud understanding of the benefits of individual projects. The objective of this study was to\ud develop a Bayesian Network (BN) to assess priorities for restoration of aquatic\ud connectivity as accomplished by replacement of culverts at road stream crossings that\ud may act as passage barriers to winter run Steelhead (Oncorhynchus mykiss) in the\ud North and South Santiam Rivers (state of Oregon). The model predicted the\ud probability of biological benefit obtained by removal or replacement of a culvert. The\ud degree of passage impairment, habitat suitability and probability of habitat use\ud influenced the predicted biological benefit. This model structure was populated with\ud conditional probability table values derived from expert opinion and a Bayesian\ud learning algorithm to produce outcomes based on different model inputs. Both models\ud were then used to assess 141 data scenarios land and fishery managers would likely\ud encounter. Results of the BN indicate that culverts that 1) are barriers to adult and\ud juvenile steelhead, 2) are located in Oregon Department of Environmental Quality\ud (DEQ) designated cold core water habitat, 3) have a high capacity for rearing juvenile\ud fish, and 4) have a high probability of habitat use will provide the highest overall\ud benefit. As anticipated, culverts that are not barriers to upstream migrating fish\ud provided the lowest benefit, regardless of habitat suitability or habitat use. In addition\ud to specific results for the Santiam basin, comparison between the two models and\ud across information scenarios illustrated the sensitivity of such models to various\ud conditions likely to be encountered by decision makers; in general, the two models\ud agreed when all input nodes were engaged by having a state value entered, yet\ud disagreed as fewer input nodes were engaged. The passage impairment of a culvert\ud and the probability of habitat use exerted a strong influence on model output. Finally,\ud this model may serve as a template for providing a coarse evaluation of culverts in\ud other basins or may be a foundation upon which additional nodes may be added