Developing a Brown Trout Stocking Screening Tool from a Statewide Random Sampling Program and Landscape-Scale Predictor Variables

Evaluation of fisheries management actions are critical but can be challenging when management actions are applied at broad spatial scales. Likewise, evaluations can be limited if monitoring data are not available for comparison. One widely applied management strategy that can be difficult to evaluate at broad spatial scales is the stocking of fish. We used presence/absence data collected by 209 surveys of the Michigan Stream Status and Trends Program, a standardized statewide stream monitoring project, for Brown Trout Salmo trutta over 20.3 cm total length and landscape-scale site predictor variables to generate a random forest model of Brown Trout presence or absence. The model had an overall error rate of 19% and was used to predict the presence and absence of Brown Trout in all Michigan stream segments (n = 68,123 stream segments). We evaluated model predictions using an independent data set containing 773 validation surveys. Validation surveys where Brown Trout were predicted present had catch rates 30% higher than validation surveys occurring at sites where Brown Trout were predicted absent, indicating biological relevance of model predictions. Comparisons with model predictions and recent Brown Trout stocking sites revealed that since 2002 Michigan has stocked nearly 9 million Brown Trout in streams our classification model predicted to be absent of Brown Trout over 20.3 cm, suggesting that field evaluation of those sites may be warranted to avoid ineffective use of resources. The model we developed can be used as a screening tool to identify the potential suitability of future stocking sites and prioritize existing stocking sites for field validation surveys to ensure efficient deployment of agency resources.