A general modelling framework for open wildlife populations based on the Polya Tree prior

Alex Diana, Eleni Matechou, Jim E. Griffin, Todd W. Arnold, R. E. Griffiths, John W. Pickering, Simone Tenan, Stefano VolponiarXiv 2020 · DOI · arXiv

This paper presents a flexible, unified statistical framework for understanding how wildlife populations change over time. By leveraging Bayesian nonparametrics, the authors provide a way to model animal "demographic schedules"—when they arrive, how long they stay, and when they leave—without being trapped by the rigid assumptions of traditional models.

TL;DR

The authors introduce a general modelling framework for open wildlife populations using Polya Tree (PT) priors. This Bayesian nonparametric approach allows researchers to capture complex entry and exit patterns across various survey types (counts, capture-recapture, and ring-recovery) while avoiding the pitfalls of over-parameterization and rigid parametric curves.

What Problem Does This Paper Address?

In ecology, "open" populations are those where individuals enter (birth/migration) and exit (death/emigration) throughout a study. Estimating population size and survival is difficult because we rarely see every individual.

Existing statistical models usually fall into two camps:

  1. Unconstrained models: These assign a unique parameter to every single sampling occasion. While flexible, they often overfit the data, especially when sightings are sparse.
  2. Parametric models: These assume entry/exit follows a specific shape (like a bell curve). While stable, they often fail to capture the messy, multi-modal realities of nature.

The gap addressed here is the need for a "Goldilocks" solution: a model flexible enough to follow the data’s true shape but structured enough to "borrow strength" across different years, sites, or species to remain statistically robust.

Methods

The core innovation is the application of Polya Tree (PT) priors to model densities of entry and exit times. A Polya Tree works by recursively partitioning the timeline into smaller and smaller intervals, assigning probabilities to each.

Key methodological extensions include:

  • Replicate PT (RPT): Allows different groups to share the same underlying "tree" structure, which is useful for assuming that survival patterns are consistent across different years.
  • Hierarchical Logistic PT (HLPT): A novel extension that allows for "hierarchical borrowing." For example, if you are studying three species of newts, the model assumes their behaviors are related (drawn from a common "base" behavior) but allows each species to deviate based on its own data.
  • Optional PT (OPT): Used for long time series (like daily counts over years). It automatically decides how deep the "tree" should go, essentially "merging" time intervals where no significant change in population density is occurring.

Datasets

The authors demonstrate the framework's versatility using four distinct ecological case studies:

  1. Italian Spoonbills: A joint analysis of capture-resight data and unmarked bird counts in the Po delta.
  2. Female Mallards: A massive 51-year dataset of 100,127 banded birds in Minnesota to estimate age-specific survival.
  3. Newts: Weekly detections of three species (Great Crested, Smooth, and Palmate) at the University of Kent, separated by sex.
  4. Moths: A 5-year daily count series of two indistinguishable species in North America.

Key Insights and Ideas

  • Computational Efficiency: Unlike other popular Bayesian nonparametric methods (like Dirichlet Process mixtures), the PT approach's complexity scales with the number of sampling occasions, not the number of individuals. This makes it much faster for large-scale wildlife datasets.
  • Centering the Prior: The PT can be "centered" on a standard distribution (like a Normal or Exponential). If the data is sparse, the model defaults to this simple shape; if the data is rich, the model "grows" to capture the complexity.
  • Age-Specific Survival without Curves: In the mallard study, the model successfully captured the classic "U-shaped" survival curve (where very young and very old birds have lower survival) entirely from the data, without being told to look for that specific shape.

Key Results

  • Spoonbills: The model estimated that ~40% of the population is already present when the spring study begins, and identified that camera traps have a lower resighting probability than manual nest visits.
  • Newts: The hierarchical model revealed that Great Crested newts arrive and depart later than other species, and that males consistently arrive earlier than females across all species.
  • Moths: Using the Optional PT, the authors identified that 2013 was a biological outlier; a warm start to the year caused the first "brood" of moths to emerge significantly earlier than in the other four years of the study.

Limitations and Future Work

The authors acknowledge that the Optional Polya Tree (OPT), while powerful for long time series, is computationally expensive because it requires recursive calculations across the entire tree in every iteration of the model. They suggest that future versions could use "tree pruning" algorithms to speed up the process.

Additionally, they propose extending the framework to include covariates (like temperature or rainfall) directly into the hierarchical structure to see how environmental changes drive the entry and exit patterns they've observed.

Summary updated

Abstract

Wildlife monitoring for open populations can be performed using a number of different survey methods. Each survey method gives rise to a type of data and, in the last five decades, a large number of associated statistical models have been developed for analysing these data. Although these models have been parameterised and fitted using different approaches, they have all been designed to model the pattern with which individuals enter and exit the population and to estimate the population size. However, existing approaches rely on a predefined model structure and complexity, either by assuming that parameters are specific to sampling occasions, or by employing parametric curves. Instead, we propose a novel Bayesian nonparametric framework for modelling entry and exit patterns based on the Polya Tree (PT) prior for densities. Our Bayesian non-parametric approach avoids overfitting when inferring entry and exit patterns while simultaneously allowing more flexibility than is possible using parametric curves. We apply our new framework to capture-recapture, count and ring-recovery data and we introduce the replicated PT prior for defining classes of models for these data. Additionally, we define the Hierarchical Logistic PT prior for jointly modelling related data and we consider the Optional PT prior for modelling long time series of data. We demonstrate our new approach using five different case studies on birds, amphibians and insects.
BibTeX
@misc{diana2020general,
  author = {Alex Diana and Eleni Matechou and Jim E. Griffin and Todd W. Arnold and R. E. Griffiths and John W. Pickering and Simone Tenan and Stefano Volponi},
  title = {{A general modelling framework for open wildlife populations based on the Polya Tree prior}},
  year = {2020},
  eprint = {2011.00591},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2011.00591},
}
Figure