Bayesian multi‐state mark‐resight modeling of survival and cause‐specific mortality from telemetered individuals: European fallow deer in Australia

Abstract

Survival and cause‐specific mortality probabilities are commonly estimated from individuals fitted with telemetry devices such as global positioning system (GPS) collars. However, the states and fates of some telemetered individuals are often unknown or unobservable because of imperfect detection (e.g., due to collar loss or transmitter failure). Here, we illustrate how multi‐state mark‐resight (MSMR) models can be used to more robustly estimate the survival and cause‐specific mortality probabilities of GPS‐collared adult female ( n  = 44) and adult male ( n  = 26) European fallow deer ( Dama dama ) in New South Wales, Australia. This MSMR modeling approach separates the observation processes (detected or undetected) from state processes (alive or dead). Some deer were detected long after their collars had failed because they were photographed alive by a trail camera or shot and the collar (with date of death) was reported to us; we included these fates observed after collar failure in the model. Of the deer that died and could be assigned a cause of mortality, humans were the dominant cause of mortality (22 deer were shot), with only one mortality attributed to natural (non‐human) causes. There was no evidence of predation. After accounting for differences in the detection and failure probabilities of the 2 GPS collar types deployed, annual survival was 0.781 (95% credible interval = 0.678–0.869) for adult females and 0.539 (0.379–0.696) for adult males. Annual natural survival was higher for adult females (0.984; 0.943–1.000) than for adult males (0.931; 0.818–0.991). These parameter estimates can be used in models evaluating management options for overabundant fallow deer populations. Our case study illustrates the flexibility of Bayesian MSMR models for incorporating imperfect detection probability (arising from collar loss and failure) and unknown fates to estimate cause‐specific mortality and survival for individuals fitted with telemetry devices. Our modeling approach is applicable to any telemetered wildlife population.