Coupling Optimal Sampling Designs with Models for Imperfect Detection: Modeling Habitat Selection of Juvenile Chinook Salmon of the Trinity River (CA)

Evaluations of the characteristics of habitats utilized by various species and life-stages of fish are often used to improve ecological understanding, inform population dynamics models, and evaluate restoration activities. In lotic ecosystems, the relationship between fish use and habitat is often cast in a suitability framework resulting in the familiar habitat suitability indices. These indices fall between 0 (lowest quality) and 1 (highest quality), and methods to create them range from professional judgment to sophisticated statistical models. Frequently, the data that feed these models are based on point-counts. It is well known that observer efficiencies are less than perfect, and therefore, these methods generally don’t lead to estimated relationships between metrics that relate to the local abundance of individuals, but instead a relative measure. Ignoring the imperfect detection can be especially problematic when detection probabilities vary based on the physical characteristics of the sampling units. N-mixture models allow, given suitably designed data collection, estimation of the relationships between habitat characteristics and local species abundance. We describe our sampling and modeling approach to evaluate the habitat characteristics of juvenile Chinook Salmon of the Trinity River (CA), and how the results can provide ecological inference and seamlessly integrate into population dynamics models.