A flexible Bayesian approach for estimating survival probabilities from age‐at‐harvest data

Abstract

Abstract Understanding survival probabilities is critical for the sustainable harvest of wildlife and fisheries populations. Age‐ and stage class‐specific survival probabilities are needed to inform a suite of population models used to estimate abundance and track population trends. However, current techniques for estimating survival probabilities using age‐at‐harvest methods require restrictive assumptions or incorporate potentially unknown parameters within the model. Using a Bayesian approach, we developed a flexible age‐at‐harvest model that incorporates either age‐ or stage‐structured populations, while accounting for uncertainty in age structure, population growth rates and relative selectivity. Survival probabilities can vary by age or stage class, as well as by environmental covariates, and both population growth rates and selectivity for each age or stage class can be specified as fixed and known or these parameters can be specified as informative priors, allowing for the incorporation of expert opinion. We evaluated our model with simulations and empirical data from harvested bobcats Lynx rufus and American paddlefish Polyodon spathula. Models fit to simulated age‐at‐harvest data yielded unbiased estimates of survival probability when population growth rates and selectivity were centered on the data‐generating parameter. We obtained unbiased estimates of survival probability even with biased prior estimates of selectivity and random departures from the assumed stage distribution, although the latter increased uncertainty in those estimates. We found biased estimates of survival probability when the prior distribution for population growth rate was not centered on the data‐generating value. When fit to empirical harvest data, our proposed age‐at‐harvest model produced estimates of survival probability congruent to those reported in the literature within similar geographic regions. We demonstrate the utility of a novel age‐at‐harvest model that estimates survival probability and realistically account for uncertainty in model parameters, transcending the restrictive assumptions and auxiliary data requirements of other methods. Furthermore, we advise collecting information about population trends and age structure alongside age‐at‐harvest data to help reduce bias. Although our model cannot replace more rigorous methods, we believe our model will be transformative for wildlife and fisheries practitioners who collect age‐at‐harvest data to estimate age‐ or stage‐specific survival probabilities to help inform management decisions.