Testing Hypotheses and Predicting Fish Rearing Capacity under Restoration and Land Use Scenarios Using Structural Equation Modeling

Structural equation modeling (SEM) is a statistical tool based on path analysis that accounts for both direct and indirect effects of interrelated predictor variables. Moreover, SEM is a powerful instrument for developing and testing hypotheses about the causal pathways leading to increased benefit or degradation of conditions for a species of concern. This latter feature of SEM allows the researcher to use knowledge gleaned from discussions with local agency biologists or direct observations of the system under study. We demonstrate the use of SEM for hypothesis generation and prediction of land use and restoration activities leading to change in juvenile spring Chinook salmon densities in the upper Grande Ronde River, Catherine Creek, and the Minam River (a reference wilderness stream) in the Blue Mountains of NE Oregon. We build on earlier models linking direct and indirect effects of large wood in the stream channel, pool frequencies, and fish rearing densities to include new data on restoration intensity and land use conditions within various riparian buffer distances laterally and longitudinally along the stream network. Finally, we discuss how SEM can complement data mining or machine learning approaches to exploring relationships in large datasets by providing a priori ecological understanding.