Using Spatial Autocorrelation to Improve Network Scale Models of Salmonid Abundance

Salmonids respond to physical habitat and stream temperatures at multiple spatial scales. At a network scale, these habitat characteristics can vary greatly among tributaries, across property ownership boundaries, and among various landscape settings. Physical properties in conjunction with biological factors, such as spawning density, dictate spatial structure of a population, the degree to which fish populations are independent from one another. In an effort to model salmonid abundance we employed spatial stream network models (SSNM) to leverage spatial autocorrelation of fish populations across two watersheds, the Middle Fork John Day River and the Grande Ronde River. Both basins originate in the Blue Mountains of eastern Oregon and offer an opportunity to explore fish-habitat relationships and patterns of spatial autocorrelation among neighboring systems. Our models utilized empirical habitat metrics and temperature data in addition to spatial structure to model abundance of juvenile salmonids in these basins. By incorporating spatial autocorrelation into our models of salmonid populations we improved our ability to model fish abundance. We also found that observed physical differences among the tributaries and mainstem rivers along with the wide range of thermal regimes present in the watersheds led to strong spatial patterns that set the framework for population response.