Using Network Models to Predict Steelhead Abundance, Middle Fork John Day, OR

• Main Author: Blanchard, Monica R.
• Format: Others
• Published: DigitalCommons@USU 2015
• Subjects: Endangered species management; scale up; Pacific salmonids; river; stream; Middle Fork John Day River; Environmental Studies; Other Life Sciences
• Online Access: https://digitalcommons.usu.edu/etd/4477; https://digitalcommons.usu.edu/cgi/viewcontent.cgi?article=5508&context=etd

In the management of threatened and endangered species, informed population estimates are essential to gage whether or not recovery goals are being met. In the case of Pacific salmonids, this evaluation often involves sampling a small subset of the population and scaling up to estimate larger distinct populations segments. This is made complicated by the fact that fish populations are not evenly distributed along riverscapes but respond to physical and biological stream properties at varying spatial extents. We used rapid assessment survey methods and the River Styles classification to explore fish-habitat relationships at a continuous network scale. Semi-continuous surveys were conducted across nine streams in the upper Middle Fork John Day River watershed and increased the number of sites surveyed eight-fold over other monitoring methods within the watershed. Using this increased sample size and continuous habitat metrics we improved watershed-wide steelhead (Oncorhynchus mykiss) abundance models. We first validated the distinctions among River Styles through a classification analysis using physical metrics measured at the rapid assessment sites. Overall classification accuracy, using a combination of reach and landscape scale metrics, was 88.3% and suggested that River Style classification was identifying variations in physical morphology within the watershed that was quantifiable at the reach scale. Leveraging the continuous River Styles classification of physical habitat and a continuous model of primary production improved the prediction of steelhead abundance across the network. Using random forest regressions, a model that included only habitat metrics resulted in R2 = 0.34, while using the continuous variables improved the model accuracy greatly to R2 = 0.65. Random forest allowed for further investigation into the predictor variables through the analysis of the partial dependence plots and identified a gross primary production threshold, below which production might be limiting steelhead populations. This method also identified the rarest River Style surveyed within the watershed, Confined-Valley Step Cascade, as the morphology that had the largest marginal effect on steelhead. The inherent physical properties and boundary conditions unique to each River Style has the potential to inform fish-habitat relationships across riverscapes and improve abundance estimates on a continuous spatial scale.