We used a generalized additive model (GAM) with a Poisson link function to model the proportion of individuals in the telemetered population migrating upstream each day during the spawning season, and included the total number of fish with tracking data available each day as an offset term. The model was fit using the ‘gam’ function in the ‘mgcv’ package [47 ] in R [42 ]. We included daily covariates for temperature, stage, discharge, and an interaction term between stage and discharge because there was not always a linear relationship between the two in 2019 as the reservoir filled and backflow was observed further upstream. Daily covariate values represent the mean stream gauge reading for each day, and all covariates were scaled and centered. Given the long monitoring period encompassed a wide range of water temperatures through spring and summer (April–August) and considering grass carp spawning likely occurs within a relatively narrow temperature range (18–24°C; Jones et al. 2017), water temperature was included in the GAM as a smooth term to account for the nonlinear effect we anticipated temperature to have on fish movement. We modeled linear relationships between migration and all other covariates to avoid overfitting and because we had no a priori assumptions of non-linearity. Parametric coefficients and smooth terms with P values < 0.05 were considered significant. Additionally, we used Lin’s concordance correlation coefficient (CCC) to assess the fit of the predicted proportion from the GAM to the observed portion of fish migrating [48 ].
We used a logistic regression model to determine if ploidy (categorical), total length at stocking, and wet weight at stocking affected the probability of a fish engaging in migratory behavior. For this analysis we incorporated fish with tracking data during the spawning season (April–August). Fish were considered migratory if they were observed making ≥ 1 upstream migratory movement during this time. We scaled and centered total length and wet weight. We considered covariates significant if the 95% confidence interval of the parameter estimate did not encompass 0.
We used a logistic regression model to determine if ploidy (categorical), total length at stocking, and wet weight at stocking affected the probability of a fish engaging in migratory behavior. For this analysis we incorporated fish with tracking data during the spawning season (April–August). Fish were considered migratory if they were observed making ≥ 1 upstream migratory movement during this time. We scaled and centered total length and wet weight. We considered covariates significant if the 95% confidence interval of the parameter estimate did not encompass 0.
Free full text: Click here
