North American Breeding Bird Survey status and trend estimates to inform a wide range of conservation needs, using a flexible Bayesian hierarchical generalized additive model

Abstract

The status and trend estimates derived from the North American Breeding Bird Survey (BBS) are critical sources of information for bird conservation. However, the estimates are partly dependent on the statistical model used. Therefore, multiple models are useful because not all of the varied uses of these estimates (e.g., inferences about long-term change, annual fluctuations, population cycles, and recovery of once-declining populations) are supported equally well by a single statistical model. Here we describe Bayesian hierarchical generalized additive models (GAMs) for the BBS, which share information on the pattern of population change across a species’ range. We demonstrate the models and their benefits using data from a selection of species, and we run full cross-validation of the GAMs against 2 other models to compare the predictive fit. The GAMs have a better predictive fit than the standard model for all species studied here and comparable predictive fit to an alternative first difference model. In addition, one version of the GAM described here (GAMYE) estimates a population trajectory that can be decomposed into a smooth component and the annual fluctuations around that smooth component. This decomposition allows trend estimates based only on the smooth component, which are more stable between years and are therefore particularly useful for trend-based status assessments, such as those by the International Union for the Conservation of Nature. It also allows for the easy customization of the model to incorporate covariates that influence the smooth component separately from those that influence annual fluctuations (e.g., climate cycles vs. annual precipitation). For these reasons and more, this GAMYE model is a particularly useful model for the BBS-based status and trend estimates.

Current contrasting population trends among North American hummingbirds

“…The authors acknowledge the potential for observer biases, wherein more resources for identification of rufous hummingbirds promotes more frequent reporting as well as an unknown increase in supplemental resource provisioning 31 . This highlights an advantage of BBS analyses for modelling population trends, because the data are systematically collected by skilled observers and are modelled with first-year-observer effects and inter-observer variability contributing to uncertainty estimates 32 . Observed declines for disparate populations in North America might also reflect threats on their non-breeding grounds 33 .…”

“…www.nature.com/scientificreports/ Estimating species population trends and population sizes using BBS data has drawn criticism because surveys along routes could potentially fail to transect productive habitats, or underestimate populations 13,42,43 . Still, the current and rapidly developing Bayesian statistical methods used to analyze these data have matured considerably in the past half-century to extract ecologically important information from species in regions with low survey coverage 32,[44][45][46] . Furthermore, these are the most rigorous methods available because it is possible to quantify uncertainties in trend estimates 47 .…”

Long-Term Modelling Reveals Contrasting Population Trends Among North American Hummingbirds

“…Estimating species population trends and population sizes using BBS data has drawn criticism because surveys along routes could potentially fail to transect productive habitats, or underestimate populations 13,35,36 . Still, the current and rapidly developing Bayesian statistical methods used to analyze these data have matured considerably in the past half-century to extract ecologically important information from species in regions with low survey coverage [37][38][39][40] . Furthermore, these are the most rigorous methods available because it is possible to quantify uncertainties in trend estimates 41 and BBS monitoring still…”

“…Population trends and trajectories were estimated using a Bayesian hierarchical Generalized Additive Model with Year Effects (GAMYE) 40 in R 44 . The survey-wide analyses were run during the annual analysis of the BBS data conducted by the Canadian Wildlife Service 45 , and additional summaries and maps were created with the R-package bbsBayes 46 .…”

Regional variability in trajectories of Barn Swallow populations across Canada are not predicted by breeding performance

“…We estimated annual indices of abundance and trends from the BBS using hierarchal Bayesian models and Markov chain Monte Carlo (MCMC) methods with the bbsBayes R package (Smith and Edwards 2020). The annual index represents the estimated average abundance of a species on BBS routes run in a given year by an average observer in that stratum.…”