Microhabitat Models of Large Drift-Feeding Brown Trout in Three New Zealand Rivers

Abstract

We examined summer habitat use by 189 drift-feeding brown trout Salmo trutta, 45–65 cm long (fork length), by measuring substrate, depth, mean velocity, focal point velocity use, and adjacent velocity into which fish were feeding in three New Zealand rivers. We compared habitat used with habitat available (simulated by hydraulic modeling), and we derived habitat use, habitat preference, and logistic regression models of habitat selection. Focal points usually were associated with large substrate components including bedrock projections, boulders, and large cobbles. Depths between 0.67 and 0.86 m were most commonly used by brown trout but the preferred depth was 1.0 m. Mean velocities between 0.38 and 0.48 m/s were the most commonly used, but preferred mean velocities were 0.05 m/s higher. Optimal focal point velocities, 0.19–0.28 m/s, were lower than mean velocities. Vertical and lateral velocity shears were calculated as measures of the velocity differentials over which fish were feeding. The velocity shears most commonly used by brown trout were between 0.50 and 0.65 m/s per meter for vertical shear and 0.02 and 0.06 m/s per meter for lateral shear. Those most preferred by brown trout were between 0.50 and 1.20 m/s per meter for vertical shear and between 0.06 and 0.26 m/s per meter for lateral shear. Depth, mean velocity, and substrate were selected independently of each other. Significant differences between rivers were found for the use of substrate, depth, focal point velocity, mean velocity, and lateral velocity shear. Depth and mean velocity were consistently significant variables in logistic regression models of habitat selection, accounting for 33–85% of the explained deviances. Large substrate components accounted for 47% of the explained deviance in one river. Lateral and vertical velocity shear together accounted for 15–27% of the explained deviances. Logistic regression and joint habitat preference models were better predictors of suitable habitat in the rivers for which they were developed than were joint habitat use models. However, joint habitat use models may be more general, because, unlike logistic regression and joint habitat preference models, the predictive success of transferred joint habitat use models was similar to that of models tested on the river for which they were developed. Combined-river habitat use, habitat preference, and logistic regression models are presented for general habitat management applications when habitat criteria are not available for specific rivers.