Evaluating Fish-Habitat Relationships in the Multiscale Stream Network

Species distribution models (SDMs) with adequate predictor (covariate) selection and scale specification improve our understanding of species-habitat relationships and informs ecosystem-based conservation and management in stream systems.  In this study, we developed multiscale models for 10 New River, Virginia non-game fish species by relating 1242 historical (1960-1990) metacommunity-based fish samples to environmental covariates retrieved from NHDplusV2 dataset.  Moran’s spatial eigenvectors were incorporated in these models to quantify the underlying spatial structure in fish distributions and to identify adequate scales for the predictors.  Thirty-four environmental covariates and 120 significant spatial eigenvectors respectively explained 25% and 37% of overall variance in fish distributions in a redundancy analysis.  At the scale of inter-confluence stream segment, local conditions (e.g., riparian forest, road density) in addition to fine-scale spatial eigenvectors dominated in the fish-habitat associations; whereas physiography, temperature and hydrology determined fish distribution at the broad (watershed and regional) scale. Independent validation with samples collected in 2012 and within sample cross-validation indicated that incorporating eigenfunction spatial analyses into our metacommunity-based SDMs improved accuracy and precision in prediction of fish occurrence.  This multiscale study provides a comprehensive framework to predict species distributions in the lotic systems, and particularly to guide predictor selection and scale specification.