Kuskokwim River Harvest Prediction Tools

Since 2016, harvest management of lower Kuskokwim River subsistence salmon fisheries has relied on in-season estimates of harvest and effort outcomes from periodic, short-duration fishing opportunities. There is little doubt that these data have been useful for their originally intended purpose: to track the cumulative harvest for determining how much harvest remains to be taken for the season. In this role, they reduce uncertainty in the outcomes of management actions already taken. However, after five years of rigorous monitoring, the time has come to evaluate their utility in reducing uncertainty with respect to outcomes that may occur from proposed future management actions. That is, managers may be able to look to the outcomes in previous years to reduce uncertainty in what will occur in the current year under alternative fishing schedules. Total harvest by species per day can be predicted as the product of total effort, total catch per effort, and species composition. There is reason to believe that these quantities vary in reasonably predictable ways, but rigorous analyses targeted at determining the best predictors and quantifying their reliability have yet to be conducted. This proposal presents a project designed to bring the historical data into greater focus for the in-season management process. The project will facilitate reproducible analyses of historical data by compiling them into an accessible and updatable data storage framework. Then, a complete analysis of variables that may predict the three critical quantities will be performed to determine which variables are most useful for predicting outcomes and their reliability (accuracy and precision) will be assessed. The final relationships that come out of this analysis will be non-trivial to implement without statistical training, so a user-friendly interactive predictive tool will be created in order to facilitate uptake of predictions into the management framework. The tool would accept inputs from the user in the form of predictor variables for the current season, and return predictions of the three critical quantities, their uncertainty, and the implied harvest. It will also include features for exploring the historical data and relationships and for performing sensitivity analyses. Biologists and managers in the region will be invited to provide input on the analyses and form and function of the tool to ensure the final products are as inclusive, informative, and user-friendly as possible. It is anticipated that these three key project components (data compilation, predictive analyses, and an interactive tool) will add substantial value to the in-season harvest monitoring program by making more complete use of the data it collects.

The overarching intent of this proposed project is to bring the historical in-season harvest and effort data into greater focus for in-season decision-making, to the extent justified by the data. Rather than viewing the in-season harvest monitoring data as having primary utility for the season in which they are collected, there is a strong possibility that they could additionally be used to develop predictive models to reduce uncertainty when considering alternative fishing strategies for the rest of the season. The primary research question of this proposed project is whether the three key quantities for predicting harvest (effort, catch rate, and species composition) can themselves be reliably predicted from variables readily available in-season. This overarching question is made up of many “sub-questions” and hypotheses. For example, we have reason to believe that the catch rate of the fishery should be related somehow to the in-river abundance of salmon, and the specific hypothesis would be that with more salmon in the river, more salmon will be captured by each individual fisher. The analysis will test this hypothesis (and many like it) for whether it is likely to be true, quantify the relationship, and conduct further calculations to determine whether the strength of the relationship is informative enough to be useful for prediction. If useful relationships can be identified, then it is critical that the data and estimated relationships can be readily accessed, updated, and used by non-technical users; these will be other intents of the project.