Integrating multiple data sources improves prediction and inference for upland game bird occupancy models

Abstract

Bird populations have declined across North America over the past several decades. Bird monitoring programs are essential for monitoring populations, but often must strike a balance between efficiency of data collection and spatial biases. Species- or habitat-specialist-specific monitoring programs may be helpful for increasing efficiency of sampling and understanding effects of management actions, but may be subject to preferential sampling bias if they are used to assess large-scale occupancy or abundance and monitoring is largely focused in high-quality habitat. More general monitoring programs, such as the North American Breeding Bird Survey (BBS) and eBird, may not preferentially sample specialists’ habitats but are subject to other forms of bias and often do not efficiently sample specialists’ habitats. We used an integrated occupancy model combining data from eBird, BBS, and Illinois state surveys of upland game bird habitat areas to estimate drivers of Northern Bobwhite (Colinus virginianus) and Ring-Necked Pheasant (Phasianus colchicus) occupancy and compare inference from single-visit, multi-visit, and integrated monitoring programs. We fit sets of candidate models using every combination of the 3 datasets except for eBird by itself, to better understand how differences in spatial biases between programs affect ecological inference. We found that, for both bobwhite and pheasant, state surveys of upland habitat increased the predictive ability of models, and BBS data usually improved inference on occupancy parameters when it was integrated with other data sources. Integrating multiple data sources partially resolved the spatial gaps in each monitoring program, while also increasing precision of parameter estimates. Integrated models may be capable of combining the higher sampling efficiency of targeted monitoring programs with the more even spatial coverage of broad-scale monitoring programs.

Considerations for fitting occupancy models to data from eBird and similar volunteer-collected data

“…eBird's data can be used to guide the planning of formal surveys, such as in the initial stage of the two-stage study design described by Pacifici et al (2012). Volunteer-collected data such as eBird's can also be combined with formally collected data from other sources in occupancy models (von Hirschheydt et al 2023), although idiosyncratic differences between data sources may need to be accounted for during modeling (Schindler et al 2022, Emmet et al 2023).…”

“…Regardless of how eBird's data are used, the variability in the data-collection process will need to be modeled (Johnston et al 2021), for example variation in the duration of observation periods, distance traveled during an observation period, and number of observers. Unless eBird's data are filtered to contain only point counts (Emmet et al 2023), the interpretation of “site” will depend on the range of travel distances encompassed in the data that are used. This is similar to the way that occupancy needs to be interpreted with any non-point counting (Altwegg and Nichols 2019).…”