Implications of a Bayesian Approach for Simulating Salmon Population Dynamics
R.A. Hill and B.J. Pyper
- Price: $0.75 Sale: $0.00
This is part of Fishery Stock Assessment Models| Format | Price | |
|---|---|---|
| PDF download [587.6 KB] Bypass cart and download |
Free | Add to Cart |
Description
Management options for Pacific salmon (Oncorhynchus spp.) populations are often analyzed using stochastic simulations that project spawner and recruit abundances over time. These forward simulations are commonly based on "best-fit" parameter estimates of the Ricker stock-recruitment relationship, despite the fact that such estimates are often highly uncertain. To examine if explicitly accounting for parameter uncertainty using Bayesian methods would affect the results of forward simulations, we compared simulations based on best-fit Ricker parameters to simulations where parameter uncertainty was incorporated through a Bayesian formulation with noninformative priors. Forward simulations using fixed harvest rates were conducted using data for two stocks of Fraser River sockeye (O. nerka) salmon. We found that expected values of spawner and recruit projections for the Bayesian simulations were consistently larger than projections for the best-fit case. Increases in the Bayes projections over those of the bestfit case typically ranged from 5 to 15% for the Stellako stock, and 20 to 60% for the Late Stuart stock. However, these differences were largely attributable to low values of the Ricker β parameter that were included in the Bayesian prior distribution. Results of Bayesian simulations were found to be highly sensitive to changes in the lower bound of the prior on β. We conclude that the common approach of specifying noninformative priors by extending uniform distributions across a broad range of parameter values can be inappropriate for the Ricker model, and recommend the use of informative priors whenever possible.
Item details
- Item number: AK-SG-98-01av
- Year: 1998
- DOI: https://doi.org/10.4027/fsam.1998.48
