Bayesian yield-per-recruit modeling

Yield-per-recruit modeling of two piscivores in a Midwestern reservoir: A Bayesian approach

Jason C. Doll, Thomas E. Lauer, and Sandra Clark Kolaks

Accompanying code can be found here.


Walleye Sander vitreus and hybrid striped bass Morone chrysops x M. saxatilis fisheries are supported byannual stockings in many US midwestern reservoirs. To maximize return to the angler, yield-per-recruitmodels are often used to evaluate expected yield and assist managers to determine which regulation toimplement, generally length or bag limits. However, yield-per-recruit models are typically formulatedwith point estimates of life history parameters, which ignore uncertainty. Our objective was to estimateyield from yield-per-recruit models of walleye and hybrid striped bass under various harvest strate-gies (e.g., alternative minimum length limits and conditional fishing mortality rates) while incorporatinguncertainty about the input model parameters. We estimated parameters of age and growth and weight-length models simultaneously using Bayesian inference. The full posterior distribution of these modelparameter estimates were then used to estimate yield. We found that yield differed among length limitsfor both species at high conditional fishing mortality. We also found yield decreased for both species asminimum length limits increased for low conditional fishing mortality. Finally, we presented a proba-bilistic framework to determine how changing minimum length limits and conditional fishing mortalityaffects the probability of achieving 70–90% of the maximum yield. Our results provide insight on theexpected yield under different minimum length limits and bag limits, while incorporating uncertainty inthe model inputs, and add to the sparse literature on hybrid striped bass population dynamics.

Figure. 1. Yield (kg) estimates of walleye under exploitation rates ranging from 0.05 to 0.65 and minimum harvest length limits ranging from 203 to 508 mm. Symbols represent medians of the posterior distribution and error bars represent 95% credible intervals.