An integrated data model to estimate abundance from counts with temporal dependence and imperfect detection

Area of Science:

• Ecology
• Wildlife Population Dynamics
• Statistical Modeling

Background:

• Estimating animal population abundance is crucial for conservation.
• Traditional survey counts often miss individuals, leading to underestimation.
• Sightability models aim to correct for missed individuals but can be limited.

Purpose of the Study:

• To develop an improved Bayesian hierarchical model for estimating population abundance.
• To integrate animal survey counts with separate detection data to account for missed individuals.
• To apply the model to harbor seal (Phoca vitulina richardii) population dynamics in Prince William Sound.

Main Methods:

• Developed a logistic-binomial-Poisson hierarchical model combining survey counts and detection data.
• Modeled detection probability using logistic regression with autocorrelated random effects.
• Incorporated temporally autocorrelated Poisson models for true abundances.
• Utilized two-stage sampling with AR1 and random walk models for computational efficiency.

Main Results:

• Identified time-of-year and time-from-low-tide as key predictors of detection probability.
• Harbor seal abundance in Prince William Sound showed a decline (1996-2001), an increase (2001-2015), and a subsequent decline (2015-2023).
• The model successfully accounted for missed individuals in aerial surveys.

Conclusions:

• The developed Bayesian model provides a robust framework for estimating population abundance using combined count and detection data.
• This approach enhances the accuracy of long-term population monitoring programs.
• The methodology is adaptable for various species and survey types, including those used in traditional sightability models.