Survival, harvest, and abundance of waterfowl populations using tag-recoveries, harvest data, and Bayesian estimation

Abstract

Annual harvest, survival, and abundance are three of the most monitored vital rates of waterfowl populations in North America. Annual harvest and survival probabilities are estimated from harvest tag-recovery data, which is a form of capture-mark-recapture data, while abundance estimation varies with species. While tag-recovery models were parameterized over 30 years ago, modern applications of these models often estimate harvest and survival probabilities as random effects drawn from Normal distributions. We evaluated tag-recovery models for reliable parameter estimation relative to different 1) monitoring scenarios varying by total individuals tagged annually, 2) prior distributions for the standard deviations of random effects, and 3) life history traits. When sample sizes of tag-recovery datasets were modest, hierarchical mean estimates of harvest and survival probability were reliably estimated but annual estimates of these parameters tended to underestimate true variation due to parameter shrinkage. The sample sizes required to capture true parameter variation with tag-recovery models likely only exist for a few species like mallards (Anas platyrhynchos) and lesser snow geese (Anser caerulescens caerulescens). Abundance of waterfowl populations is increasingly monitored using Lincoln’s index, which estimates annual abundance by dividing annual harvest count data by annual harvest probability. We demonstrated that harvest probabilities estimated from tag-recovery models can be used with Lincoln’s method, which improves the certainty of parameter estimates and increases the power to evaluate different hypotheses about harvest vulnerability within tagged samples. Our work indicates tag-recovery models should be used to assess for different harvest vulnerabilities between recently tagged individuals and individuals alive for at least one year after being tagged. If unmodeled heterogeneity in the form of different harvest vulnerabilities exist within a tagged sample and are not accounted for, inference about annual harvest, survival, and Lincoln’s abundance index will be biased.