INTRODUCTION

North America’s boreal forest supports billions of breeding birds from more than 300 species (Niemi et al. 1998). Increasing industrial development in this region (Hobson et al. 2013, Mahon et al. 2014), changing forest dynamics due to rapid rates of climate change (Stralberg et al. 2015), and numerous pressures during non-breeding periods (i.e., migration, wintering; Kirby et al. 2008) have led to concerns over the status of boreal avian populations. Recent studies suggest that boreal birds appear to have experienced among the steepest population declines of any avian group, owing to large declines of several previously abundant and widespread species (Rosenberg et al. 2016, 2019). There is therefore an urgent need to improve avian monitoring in the boreal forest (Cumming et al. 2010).

Population trends for most North American landbirds are derived from the North American Breeding Bird Survey (BBS), but this roadside survey has limited coverage in the mostly roadless core of the boreal zone. Consequently, the BBS samples a biased subset of boreal habitat (Van Wilgenburg et al. 2015), potentially leading to unrepresentative or biased trend estimates for populations of boreal bird species (Machtans et al. 2014). Many boreal-breeding species migrate to Neotropical regions that are not adequately monitored by nonbreeding (i.e., “wintering”) surveys such as the Christmas Bird Count (CBC; Meehan et al. 2019). Thus, although there are substantial data with which to estimate population status in select regions of northern forest, range-wide trends of most boreal species are lacking (Dunn et al. 2005).

There has long been interest in using standardized counts of migrating birds to evaluate population change for species that are not well-monitored during the non-migratory part of their life cycle (Francis and Hussell 1998, Dunn et al. 2005). Counting migrants for use in population monitoring is one goal of The Canadian Migration Monitoring Network, a collaborative initiative of bird observatories across Canada, Birds Canada, and Environment and Climate Change Canada (Canadian Migration Monitoring Network 2021). Similar migration count data are also collected by several long-running migration monitoring stations in the United States. However, migration monitoring data have typically been analyzed on a station-by-station basis because of a lack of knowledge of the breeding origins of migrants passing each count site. This gap in knowledge precludes appropriate weighting of site-specific trends in combined analyses to derive regional and/or range-wide trends, while also hampering appropriate targeting of conservation actions.

Advances in probabilistic origin assignments using stable hydrogen ratios in feathers have led to cost-efficient methods for broadly estimating the breeding ground origins of birds captured at migration count sites (Van Wilgenburg and Hobson 2011, Wunder 2012). The ability to assign birds to geographic regions where they molted feathers provides a powerful opportunity to appropriately weight information across a network of migration monitoring stations and to estimate large-scale patterns in population trends (Crewe et al. 2016a). This is particularly relevant to boreal-breeding birds that are not well-monitored by other methods, thereby delivering enhanced information for conservation decision making.

Blackpoll Warbler (Setophaga striata) is an abundant Nearctic-Neotropical migrant landbird species that breeds mostly in the northern boreal forest. Trends derived from roadside BBS surveys suggest populations have declined by over 90% since 1970; among the steepest declines of any landbird over that period (Rosenberg et al. 2016, Sauer et al. 2020). However, these trend estimates are potentially unreliable because the BBS lacks coverage throughout the core of the species’ breeding range (Environment and Climate Change Canada 2019) and is therefore unable to distinguish genuine population declines from climate-driven northward range shifts. Migration monitoring provides an attractive alternative method for tracking population status in boreal species, because populations from all parts of the breeding range pass through southern Canada and the eastern United States en route to and from their Neotropical non-breeding areas.

Here, we develop a Bayesian integrated population model to estimate regional and continental population trends using standardized daily counts of migrants from a series of monitoring stations, combined with estimates of the proportions of individuals coming from two geographic strata based on feather stable isotope assignments. Our method synthesizes trend information at a range-wide scale by weighting information across an international network of migration monitoring stations, thereby moving beyond station-by-station analyses. We apply this analysis to Blackpoll Warbler and compare the resulting trend estimates to those from conventional Breeding Bird Survey analysis. We also describe how this analytical framework can be applied to other migratory boreal species to generate estimates of range-wide population trends. We provide fully documented R code and recommendations for expanding this method to other avian species.

METHODS

Empirical data

Our empirical analysis relies on time series of standardized daily migration counts of Blackpoll Warblers collected at permanent monitoring locations across Canada and the United States (Fig. 1, station-specific data in Appendix 1). We limited our analysis to the time period from 1998 to 2018, for which migration counts were available at the majority of sites, and feather stable isotopes had been collected intermittently (Hobson et al. 2015). Canadian sites are members of the Canadian Migration Monitoring Network, which use station-specific standards to collect avian counts during the pre-breeding (North American spring) and/or post-breeding (North American fall) migration seasons. Count protocols differ among stations, but each site attempts to maintain relatively standardized protocols and effort over time (Canadian Migration Monitoring Network 2021). The most used count approaches include banding captures using mist nets and visual counts that record all birds detected in a specified area during a specified time. Banding at most sites occurs for six hours starting a half-hour pre-dawn at fixed net locations. The length of the daily survey period ranges from one hour (e.g., when counting birds along a fixed route) to more than six hours (e.g., continuous counts from a fixed point). Data from U.S. bird observatories consist of daily captures with mist-nets, usually with fixed locations and operated for similar hours as at Canadian sites for least five days per week during spring and/or fall migrations. Because some U.S. stations had variable daily hours and/or number of nets, the daily number of net-hours was used as a statistical offset in analyses. Daily counts for Blackpoll Warbler were available for 13 monitoring sites during pre-breeding migration and 18 sites in post-breeding migration (Fig. 1, Appendix 1). Data from each site were restricted to dates that were sampled in at least two-thirds of years in which the station operated. Within those date limits, we also restricted the analysis to data where surveys were conducted during at least 75% of days within local migration season for Blackpoll Warbler. Specific days with < 50 total net hours were omitted from analyses. For sites with notable changes in count effort between 1998 and 2018, data were limited to the years within which coverage was standardized.

For analysis, we divided the boreal breeding range of the Blackpoll Warbler into two geographic strata: “West” and “East” (Fig. 1), based on feather stable hydrogen isotope analysis and other information in Dunn et al. (2023). Those authors defined three strata, but we combined the two covering all areas west of the Great Lakes into our single West stratum because of sample size considerations, and to ensure migrants could be assigned to each stratum with higher assignment accuracy. Breeding ground origins of individual Blackpoll Warbler can be confidently assigned to one or the other stratum based on stable hydrogen isotope ratios in their tail feathers (d²H_f; Dunn et al. 2023). Boundaries of these strata are also consistent with a priori knowledge of Blackpoll Warbler migration routes, based on banding, geolocator, and additional stable isotope studies (DeLuca et al. 2015, 2019, Holberton et al. 2015, Morris et al. 2016, Covino et al. 2020).

We used feather isotope results from Dunn et al. (2023) as data in our statistical population model to indicate the proportions of migrants at each capture site that originated from East and/or West strata in years for which data were available. Stable hydrogen isotope ratios of feathers grown in parts of Alaska are isotopically similar to those expected in parts of the East stratum. However, because migrants captured west of Ontario can safely be assumed to have originated from the West stratum we fixed the West proportion for such stations at 100% without regard to isotope results. Elsewhere, we used isotope data from within 100 km of the station, though several stations had no data for estimating the proportion of migrants from West versus East. The values assigned to each site in each season and the basis for each selection are shown in Appendix 1.

Statistical model

We developed a hierarchical model to estimate temporal (i.e. multi-decadal) patterns of population change within discrete geographic strata from which birds arriving at migration monitoring stations originated. The model simultaneously estimates (1) annual indices of bird abundance at each migration monitoring station, (2) the proportion of birds at each station that have arrived from each geographic stratum within each year, and (3) population trends within each stratum. We fit models separately to data collected during pre-breeding migration (i.e., the northward migration of breeding birds during North American spring season) and post-breeding migration (i.e., the southward migration of adult and newly fledged juvenile birds during the North American fall season). Equations and priors underlying the statistical model are included in Figure 2 and Table 1, and we describe logic of these equations below.

Quantities in the model are indexed by geographic stratum (j), year (y), monitoring station (s), and day of year (d). The highest level of the model (equation 1) describes the temporal pattern of population change in each geographic stratum j starting from a baseline year y₀. Our model assumes that abundance within each stratum (X_j,y) changes according to a log-linear trend.

The next level of the model (equations 2–4) describes the arrival of migrants from each geographic stratum to each migration monitoring station during each year, and the temporal distribution of these migrants across days of the season. Equation 2 describes the annual number of migrants arriving at a station from a stratum (M_j,s,y), and is modeled as a product of annual stratum abundances (X_j,y) and station-level migration parameters (ρ_j,s) that describe the contribution of stratum j to station s in year y. The parameters ρ_j,s are station-specific “capture rates” that convert the indices of abundance in each stratum to number of birds arriving at each station in each year. In cases where a migration monitoring station is known (or assumed) to exclusively capture birds from a subset of strata, the relevant parameters (ρ_j,s) can be fixed to zero for the strata that are not monitored by the station. We note that stratum-level intercept terms in the model (X_j,yo), are not estimable with migration data alone because infinite combinations of X_j,yo and migration parameters (ρ_j,s) would be equally consistent with the observed data. Thus, we fix X_j,yo = 1 such that the model estimates change relative to the start year within each stratum, ensuring all parameters are theoretically identifiable. This also implies the model cannot estimate relative abundances between strata and is only capable of estimating changes in relative abundance within each stratum. However, independent estimates of stratum relative abundances can be used to rescale X_j,y in each stratum and calculate range-wide trends (see below).

Equation 3 models an index of total abundance of migrants at each station (T_s,y) as a lognormal random variable, with median equal to the sum of M_j,s,y across J strata, with additional annual variation described by . The term acknowledges that there is annual variation in the total number of migrants arriving at each station beyond that which is attributable to changes in X_j,y (e.g., additional variation could be driven by random year-to-year fluctuations in migration pathways).

Equation 4 distributes the T_s,y migrants arriving at the monitoring station among days of the season. This component of the model is necessary because some monitoring stations are only operational for a subset of days per season; the model can therefore accommodate missing data within a season. Migration of individuals past monitoring stations is assumed to follow a symmetric seasonal pattern around a peak date. We therefore described the seasonal pattern of counts at each station using a Normal probability density function ƒ (d, μ_s, σ_s) that integrates to 1 across a season. The parameter μ_s is the ordinal date of the seasonal peak of migration at station s, while σ_s describes the temporal dispersion of the migration period at that station (i.e., approximately 95% of the station’s migration period occurs within 1.96σ_s on either side of μ_s).

The final level of the model describes the observation process associated with daily count data (equation 5) and breeding origin estimates (in our case study, based on stable isotope data; equation 6). To describe migration count data, equation 5 models the number of birds counted on each day of the season (d) at each station (s) in each year (y) as an over-dispersed Poisson process with median equal to T_s,y × ƒ (d, μ_s, σ_s). We assumed log-normal variance for unexplained “noise” in daily counts at each station (e.g., owing to weather conditions that affect daily migration behavior). Equation 5 also includes an offset equal to log(net hours) to account for spatio-temporal variation in monitoring effort. This also implies that T_s,y can be compared to empirical count data (e.g., for goodness-of-fit evaluation) by dividing daily counts by effort and summing those quantities across a season. We use a multinomial distribution to model breeding origins of migrants at the stations where they were collected (equation 6). In this equation, Y_s,y is a vector (denoted by bolded font) containing the number of sampled birds assigned to each of the J strata at a station in a year. The vector of probabilities describing the multinomial distribution (i.e., M_j,s,y / Σ^J₁M_j,s,y) allows station-level dynamics to be linked to stratum-level dynamics.

Trend estimation

Estimates of percent population change in a stratum relative to a baseline year (e.g., in this study, relative to the year 1998) can be calculated as:

(7)

where X_j,yend is an index of abundance in the final year (y_end). Following the North American BBS (Smith et al. 2014), we defined population trend as the geometric mean rate of change between two points in time, where:

(8)

Because our application of the model requires that stratum-level intercept terms are fixed to a constant (such that X_j,yo and ρ_j,s are identifiable), independent estimates of relative abundance in each stratum are required to re-scale estimates of X_j,y and thereby appropriately weight changes in abundance within each stratum at a range-wide scale. This re-scaling of X_j,y is necessary because a large positive trend in a stratum that contains almost no birds will have a small effect on the continental trend compared to a small negative trend in a stratum that contains a large majority of the population. For Blackpoll Warbler, there are several relative abundance rasters that could be used to re-weight stratum-level trends within a continental context. To evaluate the sensitivity of continental trend estimates to the choice of relative abundance raster, we focused our analysis on two maps produced using different datasets and analytical methods. First, we used a spatial abundance map from the Boreal Avian Modeling Project’s (BAM) Generalized National Models version 4.0, produced using boosted regression trees fit to a large dataset of point counts across North America that correct for variation in protocols among surveys (Solymos et al. 2013, Solymos and Stralberg 2020). The raster presents estimates of relative density in 1 km x 1 km pixels, as a prediction for the year 2011 for Canada. We cropped the relative abundance raster to the boundary of each stratum, and summed pixel values to yield estimates of relative abundance. However, BAM’s raster omitted the Alaskan portion of the Blackpoll Warbler breeding range, and we therefore predicted the relative abundance in Alaska based on the strong empirical relationship between BBS counts and BAM’s relative abundance estimates (Appendix 2). This exercise indicated that Blackpoll Warbler was 2.43 times more abundant in the West stratum than in the East in the year 2011 (1.99 times more abundant without the Alaska portion of the range included). We therefore re-scaled estimates of X_j,y in each of the j strata based on these values, using X_j,y / X_j,2011 × RelAbund_j,2011, where RelAbund_j,2011 was 2.43 and 1 for the West and East, respectively. Second, we repeated this exercise for a raster obtained from the community science project eBird using the “ebirdst” package in R, which represents a prediction of relative abundance for the year 2022 (Fink et al. 2023; Appendix 2). Based on this map, the relative abundance of Blackpoll Warbler was 1.46 times larger in the East stratum than the West stratum in 2022 (i.e., the opposite pattern predicted by BAM for the year 2011). Because our analysis only encompassed from the year 1998 to 2018, we applied this correction to 2018 such that the posterior mean of the population index for the East stratum was 1.46 times larger the West.

We did not attempt to evaluate which raster was a more accurate representation of large-scale Blackpoll Warbler relative abundance across boreal North America, and we instead present continental trend estimates using each product to illustrate the importance of relative abundance information for continental migration trend analyses. Importantly, both the BAM and eBird rasters were reviewed by species experts prior to their publication, but only BAM provides goodness-of-fit metrics and cross-validation scores associated with their raster. The BAM dataset includes a large number of point counts distributed throughout the boreal zone of North America over several decades, while eBird relies heavily on checklists collected over a similar time frame. Each product relies on fundamentally different methods to correct for variation in survey effort within their dataset, and uses different machine learning algorithms to account for spatio-temporal sampling biases.

Parameter estimation and model diagnostics

We fit the statistical model in a Bayesian framework using JAGS version 4.3.0 (Plummer 2003), interfaced with the R programming language version 4.0.2 (R Core Team 2024) using the jagsUI library version 1.5.2 (Kellner 2021). We specified vague priors on most model parameters (Table 1) except σ_s where we specified priors to facilitate better model convergence that were moderately informative, based on a priori knowledge of the typical “migration period” at stations. After a burn-in of 5000 iterations, we stored every 50th iteration until we accumulated 2000 posterior samples from each of three MCMC chains. We assessed chain convergence by visual examination of MCMC traceplots and by evaluating that the Gelman-Rubin convergence statistic was close to 1 for parameters on which we based inference (i.e., X_j,y). We also confirmed that effective sample sizes for those parameters were large enough (> 1000) to adequately characterize posterior distributions. The model took 52 minutes to fit on a personal laptop for pre-breeding migration, and 162 minutes for post-breeding migration.

We assessed goodness-of-fit of model predictions by evaluating the correlations between observed seasonal totals at monitoring stations and the expected seasonal totals based on the fitted models (Appendix 3). We also conducted posterior predictive checks to confirm that the distribution of simulated counts based on the fitted statistical model were consistent with the distribution of observed counts at each station (Appendix 3). Using simulation, we also evaluated whether the migration model was able to generate unbiased estimates of regional population trends across a wide range of regional trend and station-level capture rate scenarios (Appendix 3).

Comparison to trend estimates from the North American Breeding Bird Survey

We compared regional trend estimates from our model to those derived from the North American Breeding Bird survey to evaluate similarities and differences between conventional breeding season analyses and our migration analysis. We fit a Bayesian hierarchical model to BBS time series for Blackpoll Warbler from 1998 to 2018 (the same period as our migration analysis), using analytical strata implemented by the United States Geological Survey (USGS) for continental analysis. We specified a “first difference” population process model (Link et al. 2017, Smith and Edwards 2021), which is widely used in standard continental analysis of BBS data. We fit the model and extracted output using the “bbsBayes” package in R (Edwards and Smith 2020), specifying a 50,000 iteration burn-in period, after which we stored every 100th posterior sample until we accumulated 2000 posterior samples from each of 3 MCMC chains.

To derive regional trend estimates from the BBS, we assigned BBS analytical strata into East and West categories based on geographic overlap with the strata we used for migration analysis (illustrated in Appendix 4). This allowed us to calculate “post hoc” synthetic estimates of regional population trends by calculating annual BBS indices from the fitted model across analytical strata that overlapped with the coarse East and West strata used for the migration monitoring analysis. Detailed methods for estimating population trajectories and trends within custom strata are described in the bbsBayes package (Edwards and Smith 2020).

RESULTS

Simulations confirmed that the statistical model was capable of producing regional trend estimates that are identifiable, unbiased, and with appropriate credible interval coverage across a wide range of simulated population trends, and in situations where capture rates (ρ_j,s) are highly variable between stations, and migration counts are highly variable among years and among days within a season at each station (Appendix 3). Trends were also recoverable with highly incomplete isotope data, even when no stations were known a priori to exclusively monitor a single stratum.

Models unambiguously converged for empirical analysis of Blackpoll Warbler migration data in both pre-breeding and post-breeding seasons. Goodness-of-fit metrics for analysis of Blackpoll Warbler indicated a reasonable fit to the empirical data for most stations (Appendix 2), though posterior predictive checks suggested a poor fit for several station-year combinations owing to overdispersion in daily counts.

For Blackpoll Warbler populations breeding in the West stratum, migration monitoring in both seasons indicated similar population trends (Table 2, Fig. 3a,b). We detected strong evidence of population increases in the western stratum based on both pre-breeding migration and post-breeding migration (> 0.99 probability of positive trend). In contrast, analysis of the BBS indicated a high probability that western populations had declined over the same period (0.97 probability of negative trend; Fig. 3c).

For the East stratum population, pre-breeding migration trends showed strong evidence of large declines (< 0.01 probability of positive trend; Fig. 3d), with a median trend estimate of -4.6% per year leading to a decline of greater than 40% between 1998 and 2018. BBS also suggested there was a high probability (0.97) the trend was negative, and the magnitude of the BBS trend estimate was similar to that from pre-breeding migration monitoring (Fig. 3f, Table 2). The eastern population trend estimated from post-breeding migration (Fig. 3e) was extremely imprecise because few stations captured exclusively eastern migrants during post-breeding migration (Appendix 1; Fig. A1.4), and signals of eastern population changes were largely swamped by migrants originating from the western stratum.

Estimates of continental population trends depended on the source of relative abundance estimates among strata that were used to weight trend estimates in each stratum in a continental context (Fig. 4). If using a relative abundance raster produced by the Boreal Avian Modeling Project (Solymos and Stralberg 2020) that suggested the population in the West stratum was 2.43 times larger than the East stratum in 2011, the stratum-level trends we detected would have led to population increases in the West counter-balancing declines in the East, and the continental population would have remained approximately stable (based on pre-breeding migration; Fig. 4a) or have increased (based on post-breeding migration; Fig. 4b, Table 3). However, if using relative abundance estimates from eBird, which suggested the East stratum is currently 1.46 times larger than the West (Appendix 2; Fink et al. 2023), the continental population would have likely declined since 1998 based on pre-breeding migration data (Fig. 4c) or remained approximately stable based on post-breeding migration (Fig. 4d) though this estimate was highly uncertain owing to large uncertainty in the Eastern trend. In contrast to migration monitoring, the BBS detected strong evidence of population declines in both strata, leading to strong evidence of continental declines (Fig. 4e, Table 3).

DISCUSSION

Our study fulfills a longstanding need for North American landbird monitoring by facilitating large-scale population trend assessments for migratory species, particularly for the approximately 80 species that primarily breed in the core of the boreal forest (Dunn et al. 2005). Because many species are difficult to survey during the non-migratory periods of their life cycle, migration monitoring can be crucial in assessing the status of those species until other data become available. Our model synthesizes information from multiple sites (over 20 locations for the Blackpoll Warbler case study), and therefore represents an important step beyond analysis of migration trends on station-by-station basis (Canadian Migration Monitoring Network 2021). Variation in the number of migrants among stations is partitioned by our model into contributions from differences in regional population trends, differences in station-level catchment, and multiple sources of unexplained spatial and temporal variation.

In general, confidence in a scientific finding is enhanced when multiple independent lines of inquiry converge on a shared conclusion through so-called “methodological triangulation” (Heesen et al. 2019). For North American landbirds, there are now multiple survey programs from which population trends can potentially be estimated and compared. Large-scale, long-term, semi-structured community science programs such as the BBS and the Christmas Bird Count can provide valuable insights into population trends for species inhabiting the road-accessible portions of North America (Link et al. 2006, Sauer et al. 2020). Provincial and state breeding bird atlases (Dunn and Weston 2008), along with a suite of other regional monitoring programs (e.g., Pavlacky et al. 2017, Hill 2023) can provide finer-scale inference in locations that may otherwise be data-deficient. eBird also provides estimates of local and regional population trends over the last 10 years across portions of North America, based on unstructured volunteer data analyzed with machine-learning algorithms designed to minimize biases inherent in opportunistically collected datasets (Fink et al. 2023). Range-wide migration monitoring, in combination with stable isotope analysis, can provide a powerful complement to these other surveys because it relies on a fundamentally different methodology: the capture of migrating birds using standardized protocols, assignment of those individuals to regional strata, and integrated analysis of data using hierarchical models. Migration monitoring has an additional advantage in that pre- and post-breeding migration seasons can be analyzed independently from each other (as in our current analysis), providing a further check on consistency in estimates. Agreement between population trend estimates from each migration season and other relevant breeding and non-breeding surveys could greatly strengthen confidence in species status and trend assessments. Conversely, lack of agreement may help identify weaknesses in a particular survey approach or suggest new avenues for research.

For Blackpoll Warbler, multiple lines of evidence indicate that severe population declines have recently occurred in eastern North America. Independent analysis of pre-breeding migration and BBS data suggest that eastern populations have declined at rates of approximately -4% per year (Table 2). Congruent with these findings, Mountain Birdwatch, which monitors montane breeding bird communities in the eastern United States, also estimated Blackpoll Warbler population declines of approximately -4.5% per year (80% CI = -5.2% to -3.8) from 2010 to 2023 (Hill 2023). Strong declines in Blackpoll Warbler breeding occurrence were also detected in New Brunswick and Cape Breton Island, Nova Scotia, between the first (1990) and second (2010) Maritimes Breeding Bird Atlases (Stewart 2015). In contrast, eBird’s trend estimates suggest stable or increasing Blackpoll Warbler populations in eastern Canada between 2012 and 2022 (Fink et al. 2023); the cause of this discrepancy is unclear. Unfortunately, post-breeding migration data was unable to provide a precise trend estimate for the eastern stratum because all stations that received eastern birds simultaneously received larger numbers of western birds, which swamped eastern signals of population change (Fig. A1.4; Dunn et al. 2023). Provincial breeding bird atlases currently underway in Ontario and Quebec are collecting boreal avian data using a nationally standardized framework (Van Wilgenburg et al. 2020) and are expected to provide critical insights into the drivers of population changes that have occurred in eastern Canada since the early 2000s. Simultaneously, these efforts will provide another independent test of the ability of our model to track regional changes in population size.

Pre- and post-breeding migration analysis independently suggested that populations of Blackpoll Warbler west of Ontario, Canada have increased over a 20-year period. In contrast, the BBS suggested that populations have likely declined in this region over the same period. eBird also reports evidence of recent populations declines in western Canada and Alaska (Fink et al. 2023). However, data from BBS and eBird are heavily biased toward roadsides and human settlements and therefore sample an incomplete and biased portion of the western boreal forest (Machtans et al. 2014, Van Wilgenburg et al. 2015). BBS and eBird are also unable to estimate population trends in the Northwest Territories and Nunavut where long-term breeding season surveys are exceedingly sparse, but where a substantial proportion of the continental population of Blackpoll Warbler likely breeds (Appendix 2). The migration monitoring network likely captures substantial numbers of individuals from those regions, particularly during post-breeding migration when the entire continental population passes through eastern North America (Holberton et al. 2015). However, it is also possible that migration trends could be biased if sites monitoring “western” birds are poorly distributed such that the portion of the range they sample is unrepresentative of population dynamics within the western stratum.

It is often difficult (or even impossible) to infer the causes of regional variation in population trends for migratory species from single-season monitoring data alone, whether that data is collected during the stationary or migratory periods of the annual cycle (Hostetler et al. 2015). This is because seasonally specific pressures in one part of the species’ annual cycle can generate population trends in geographically distant regions when there is strong migratory connectivity between breeding and non-breeding areas. This phenomenon was reported in two populations of Golden-winged Warbler (Vermivora chrysoptera) that experienced divergent population trends in their North American breeding areas, hypothesized to be driven by different pressures in their distinct migratory pathways and Central and South American non-breeding areas (Kramer et al. 2018). Similarly, severe declines in eastern breeding populations of Canada Warbler (Cardellina canadensis) were attributed to reductions in recruitment caused by habitat loss in nonbreeding areas in the Andes (Wilson et al. 2018). Habitat change on the breeding grounds can also affect population trends, and for example, has been more strongly implicated in declines of Connecticut Warbler (Oporornis agilis) than habitat change in their South American nonbreeding areas (Hallworth et al. 2021). For Blackpoll Warbler, it is unclear why eastern populations have declined more rapidly than elsewhere in their breeding range. Recent climate change has been variable across the boreal forest, with more rapid warming occurring in the west but larger reductions in spring snow cover occurring in the east (Bush and Lemmen 2019). Yet, the demographic pathways by which climate affects the species remain poorly understood, and it is unclear how threats along migration routes and in nonbreeding areas differ for birds originating from eastern and western breeding populations. Ultimately, full-annual-cycle population models are needed to understand the cause of observed population changes (Hostetler et al. 2015), which require improved estimates of migratory connectivity between breeding and non-breeding areas in combination with improved monitoring throughout the life cycle.

Although our model can directly estimate stratum-level trends, continental trend estimates require independent relative abundance estimates to weight stratum-level trends appropriately in a broader context (see Methods; Appendix 2). Relative abundance rasters based on breeding season surveys can be used for this purpose (e.g., Solymos and Stralberg 2020, Fink et al. 2023), but the reliability of those maps depends on how representative the underlying data are spatially, temporally, and with respect to covariates that describe the species environmental niche. Unreliable abundance rasters could lead to biased estimates of range-wide trends in our analysis because stratum-level trends could be incorrectly weighted at the continental level. For example, the declines we detected in the East stratum based on pre-breeding migration could either represent a massive continental decline (i.e., if based on eBird relative abundance maps; Fink et al. 2023), or could have been largely counter-balanced by population increases in the West (i.e., if based on density rasters derived from a large-scale boreal point count dataset; Solymos and Stralberg 2020). Goodness-of-fit and cross-validation accuracy assessments were only available for the raster from the Boreal Avian Modeling Project, limiting our ability to determine which product is more appropriate for our purposes. However, because each raster was constructed using analytical techniques that make different assumptions and are based on different datasets with different spatio-temporal biases, it is likely that each raster performs better in some regions and worse in others. In some cases, ensemble approaches (i.e., averaging across different model results) can improve predictions of species distributions and relative abundance, though this is not guaranteed in general (Hao et al. 2020). Here, we instead reported predictions separately to illustrate the consequences of relative abundance weighting on continental trend estimates. Nevertheless, this emphasizes the critical need for rigorously designed, randomly selected, standardized surveys throughout the boreal zone to establish reliable baseline estimates of abundance (Van Wilgenburg et al. 2020), which will also greatly enhance the reliability of range-wide trend estimates from migration monitoring.

Our statistical model makes several additional assumptions that warrant consideration. First, migration counts at each station are driven by a complex product of both variation in regional population sizes and the probability that migrating individuals are observed at monitoring stations (Dunn 2005). Our model is unable to account for long-term directional changes in migration pathways and stopover behavior, which would impose spurious signals of population change. This occurs, for example, if migrants increasingly avoid particular monitoring stations when surrounding habitat changes considerably over time (Francis and Hussell 1998, Dunn 2005). However, network-wide trend analyses should be robust to this bias if many stations count birds from each region, reducing the influence of single stations on overall trend estimates. Additionally, site-specific migratory connectivity can be periodically reassessed to examine changes in the regions of the breeding range that are sampled by each station, as occurred for Blackpoll Warbler at some Great Lakes sites (Dunn et al. 2023). However, frequent analysis of feather isotopes across numerous stations could be logistically or cost-prohibitive so alternative methods for estimating station catchment should also be considered (e.g., Meehan et al. 2022, Dunn et al. 2023). Second, migration counts are extremely stochastic and difficult to characterize parametrically (Crewe et al. 2016b). Unrealistic error distributions could result in poor goodness-of-fit metrics in some years at some stations (e.g., Appendix 3) with associated credible intervals on estimates that are too narrow. Nonetheless, this phenomenon is not unique to our migration monitoring analysis and iterative improvements in model structure are almost always necessary for hierarchical analyses of long-term, large-scale monitoring data (Link and Sauer 2016). Third, the current structure of our population process model only includes a log-linear trend within each stratum and does not attempt to estimate random annual process variation in regional population sizes. This assumption may be justified for a wide-ranging, highly abundant species like Blackpoll Warbler, where large-scale regional trajectories derived from the BBS also resulted in relatively smooth trajectories (Fig. 3). However, this would not be true for nomadic or irruptive populations. Use of our model for standardized hawk migration counts should consider that confidence intervals on trend estimates for partial migrants, which vary both in proportion of population migrating and in distance traveled, may be too narrow because random temporal variation in population dynamics is not fully accounted for. Future applications of our model could also consider attempting to fit more flexible temporal trajectory functions within strata to describe curvilinear changes in trajectories over time (Smith and Edwards 2021).

Application of our method to a larger number of boreal breeding species could yield novel insights into the population status of boreal-breeding migratory species. Our analytical approach could potentially be extended to other groups of birds that are more reliably covered on migration or at stopover sites, such as shorebirds (Smith et al. 2023) and raptors (Farmer et al. 2007) that are complete migrants, provided that the origins of birds can be estimated at migration survey locations. Our model could also potentially be extended to incorporate non-standardized migration count data from community science networks such as eBird, especially if there was a well-distributed roster of sites frequently visited during migration seasons that could be targeted for daily coverage, conceptually adding thousands of migration monitoring “stations” across the continent. This would have numerous advantages including more comprehensive coverage of migratory populations and reducing the influence of individual monitoring stations on regional trend estimates, even though reliance on community science information requires careful screening of data and appropriate accounting of changes in observer effort over time (Fink et al. 2023). Continued efforts to implement and improve range-wide trend analyses from migration surveys holds considerable promise for North American landbird monitoring.

LITERATURE CITED

Bush, E., and D. S. Lemmen, editors. 2019. Canada’s changing climate report. Government of Canada, Ottawa, Ontario, Canada. https://doi.org/10.4095/314614

Canadian Migration Monitoring Network. 2021. The Canadian Migration Monitoring Network - Réseau canadien de surveillance des migrations: Researching Canada’s Landbirds for Twenty Years. CMMN-RCSM Scientific Technical Report 3. CMMN-RCSM, Port Rowan, Ontario, Canada.

Covino, K., S. Morris, M. Shieldcastle, and P. D. Taylor. 2020. Spring migration of Blackpoll Warblers across North America. Avian Conservation and Ecology 15(1):17. https://doi.org/10.5751/ACE-01577-150117

Crewe, T. L., D. Lepage, and P. D. Taylor. 2016a. Effect of sampling effort on bias and precision of trends in migration counts. Condor 118(1):117-138. https://doi.org/10.1650/CONDOR-15-78.1

Crewe, T. L., P. D. Taylor, and D. Lepage. 2016b. Temporal aggregation of migration counts can improve accuracy and precision of trends. Avian Conservation and Ecology 11(2):8. https://doi.org/10.5751/ACE-00907-110208

Cumming, S. G., K. Lefevre, E. Bayne, T. Fontaine, F. K. A. Schmiegelow, and S. J. Song. 2010. Toward conservation of Canada’s boreal forest avifauna: design and application of ecological models at continental extents. Avian Conservation and Ecology 5(2):8. https://doi.org/10.5751/ACE-00406-050208

DeLuca, W. V., B. K. Woodworth, S. A. Mackenzie, A. E. M. Newman, H. A. Cooke, L. M. Phillips, N. E. Freeman, A. O. Sutton, L. Tauzer, C. McIntyre, et al. 2019. A boreal songbird’s 20,000 km migration across North America and the Atlantic Ocean. Bulletin of the Ecological Society of America 100(3):e01551. https://doi.org/10.1002/bes2.1551

DeLuca, W. V., B. K. Woodworth, R. C. Rimmer, P. P. Marra, P. D. Taylor, K. P. McFarland, S. A. Mackenzie, and D. R. Norris. 2015. Transoceanic migration by a 12-g songbird. Biology Letters 11(4). https://doi.org/10.1098/rsbl.2014.1045

Dunn, A. M., and M. A. Weston. 2008. A review of terrestrial bird atlases of the world and their application. Emu - Austral Ornithology 108(1):42-67. https://doi.org/10.1071/MU07034

Dunn, E. H. 2005. Counting migrants to monitor bird populations: state of the art. Pages 712-717 in C. J. Ralph and T. D. Rich, editors. Bird conservation implementation and integration in the Americas: Proceedings of the Third International Partners in Flight Conference. 2002 March 20-24; Asilomar, California. Volume 2 General Technical Report PSW-GTR-191. U.S. Forest Service, Pacific Southwest Research Station, Albany, California, USA.

Dunn, E. H., B. L. Altman, J. Bart, C. J. Beardmore, H. Berlanga, P. J. Blancher, G. S. Butcher, D. W. Demarest, R. Dettmers, W. C. Hunter, et al. 2005. High priority needs for range-wide monitoring of North American landbirds. Partners in Flight Technical Series 2. Partners in Flight. https://www.partnersinflight.org/wp-content/uploads/2017/03/PIF-Technical-Series-02-Monitoring-Needs.pdf

Dunn, E. H., K. J. Kardynal, K. M. Covino, S. R. Morris, R. L. Holberton, and K. A. Hobson. 2023. Feather isotopes (δ²H_f) and morphometrics reveal population-specific migration patterns of the Blackpoll Warbler (Setophaga striata). Avian Conservation and Ecology 18(2):16. https://doi.org/10.5751/ACE-02539-180216

Edwards, B. P., and A. C. Smith. 2020. bbsBayes: An R package for hierarchical Bayesian analysis of North American Breeding Bird Survey data. Journal of Open Research Software 9(1):19. https://doi.org/10.5334/jors.329

Environment and Climate Change Canada. 2019. Blackpoll Warbler (Setophaga striata). Status of Birds in Canada 2019. https://wildlife-species.canada.ca/bird-status/oiseau-bird-eng.aspx?sY=2019&sL=e&sM=a&sB=BLPW

Farmer, C. J., D. J. Hussell, and D. Mizrahi. 2007. Detecting population trends in migratory birds of prey. Auk 124(3):1047-1062. https://doi.org/10.1093/auk/124.3.1047

Fink, D., T. Auer, A. Johnston, M. Strimas-Mackey, S. Ligocki, O. Robinson, W. Hochachka, L. Jaromczyk, C. Crowley, K. Dunham, A. Stillman, I. Davies, A. Rodewald, V. Ruiz-Gutierrez, and C. Wood. 2023. eBird status and trends, data version: 2022; Released: 2023. Cornell Lab of Ornithology, Ithaca, New York, USA. https://doi.org/10.2173/ebirdst.2022

Francis, C. M., and D. J. Hussell. 1998. Changes in numbers of land birds counted in migration at Long Point Bird Observatory, 1961-1997. Bird Populations 4:37-66.

Hallworth, M. T., E. Bayne, E. McKinnon, O. Love, J. A. Tremblay, B. Drolet, J. Ibarzabal, S. L. Van Wilgenburg, and P. P. Marra. 2021. Habitat loss on the breeding grounds is a major contributor to population declines in a long-distance migratory songbird. Proceedings of the Royal Society B: Biological Sciences 288:20203164. https://doi.org/10.1098/rspb.2020.3164

Hao, T., J. Elith, J. J. Lahoz-Monfort, and G. Guillera-Arroita. 2020. Testing whether ensemble modelling is advantageous for maximising predictive performance of species distribution models. Ecography 43:549-558. https://doi.org/10.1111/ecog.04890

Heesen, R., L. K. Bright, and A. Zucker. 2019. Vindicating methodological triangulation. Synthese 196:3067-3081. https://doi.org/10.1007/s11229-016-1294-7

Hill, J. M. 2023. The state of the mountain birds report: 2023. Vermont Center for Ecostudies, White River Junction, Vermont, USA. https://mountainbirds.vtecostudies.org/wp-content/uploads/2023/12/State-of-the-Mountain-Birds-Report-2023-v2.pdf

Hobson, K. A., S. L. Van Wilgenburg, E. H. Dunn, D. J. T. Hussell, P. D. Taylor, and D. M. Collister. 2015. Predicting origins of passerines migrating through Canadian migration monitoring stations using stable-hydrogen isotope analyses of feathers: a new tool for bird conservation. Avian Conservation and Ecology 10(1):3. https://doi.org/10.5751/ACE-00719-100103

Hobson, K. A., A. G. Wilson, S. L. Van Wilgenburg, and E. M. Bayne. 2013. An estimate of nest loss in Canada due to industrial forestry operations. Avian Conservation and Ecology 8(2):5. https://doi.org/10.5751/ACE-00583-080205

Holberton, R. L., S. L. Van Wilgenburg, A. J. Leppold, and K. A. Hobson. 2015. Isotopic (δ²H_f) evidence of “loop migration” and use of the Gulf of Maine Flyway by both western and eastern breeding populations of Blackpoll Warblers. Journal of Field Ornithology 86:213-228. https://doi.org/10.1111/jofo.12112

Hostetler, J. A., T. S. Sillett, and P. P. Marra. 2015. Full-annual-cycle population models for migratory birds. Auk 132:433-449. https://doi.org/10.1642/AUK-14-211.1

Kellner, K. 2021. jagsUI: A wrapper around rjags to streamline JAGS analyses. R package version 1.5.2. https://doi.org/10.32614/CRAN.package.jagsUI

Kirby, J. S., A. J. Stattersfield, S. H. Butchart, M. I. Evans, R. F. Grimmett, V. R. Jones, J. O'Sullivan, G. M. Tucker, and I. Newton. 2008. Key conservation issues for migratory land- and waterbird species on the world’s major flyways. Bird Conservation International 18(S1):S49-S73. https://doi.org/10.1017/S0959270908000439

Kramer, G. R., D. E. Andersen, D. A. Buehler, P. B. Wood, S. M. Peterson, J. A. Lehman, K. R. Aldinger, L. P. Bulluck, S. Harding, J. A. Jones, et al. 2018. Population trends in Vermivora warblers are linked to strong migratory connectivity. Proceedings of the National Academy of Sciences 115(14):E3192-E3200. https://doi.org/10.1073/pnas.1718985115

Link, W. A., and J. R. Sauer. 2016. Bayesian cross-validation for model evaluation and selection, with application to the North American Breeding Bird Survey. Ecology 97(7):1746-1758. https://doi.org/10.1890/15-1286.1

Link, W. A., J. R. Sauer, and D. K. Niven. 2006. A hierarchical model for regional analysis of population change using Christmas Bird Count data, with application to the American Black Duck. Condor 108(1):13-24. https://doi.org/10.1093/condor/108.1.13

Link, W. A., J. R. Sauer, and D. K. Niven. 2017. Model selection for the North American Breeding Bird Survey: a comparison of methods. Condor 119(3):546-556. https://doi.org/10.1650/CONDOR-17-1.1

Machtans, C. S., K. J. Kardynal, and P. A. Smith. 2014. How well do regional or national Breeding Bird Survey data predict songbird population trends at an intact boreal site? Avian Conservation and Ecology 9(1):5. https://doi.org/10.5751/ACE-00649-090105

Mahon, C. L., E. M. Bayne, P. Sólymos, S. M. Matsuoka, M. Carlson, E. Dzus, F. K. Schmiegelow, and S. J. Song. 2014. Does expected future landscape condition support proposed population objectives for boreal birds? Forest Ecology and Management 312:28-39. https://doi.org/10.1016/j.foreco.2013.10.025

Meehan, T. D., N. L. Michel, and H. Rue. 2019. Spatial modeling of Audubon Christmas Bird Counts reveals fine-scale patterns and drivers of relative abundance trends. Ecosphere 10(4):e02707. https://doi.org/10.1002/ecs2.2707

Meehan, T. D., S. P. Saunders, W. V. DeLuca, N. L. Michel, J. Grand, J. L. Deppe, M. F. Jimenez, E. J. Knight, N. E. Seavy, M. A. Smith, L. Taylor, et al. 2022. Integrating data types to estimate spatial patterns of avian migration across the Western Hemisphere. Ecological Applications 32(7):e2679. https://doi.org/10.1002/eap.2679

Morris, S. R., K. M. Covino, J. D. Jacobs, and P. D. Taylor. 2016. Fall migratory patterns of the Blackpoll Warbler at a continental scale. Auk 133(1):41-51. https://doi.org/10.1642/AUK-15-133.1

Niemi, G., J. Hanowski, P. Helle, R. Howe, M. Mönkkönen, L. Venier, and D. Welsh. 1998. Ecological sustainability of birds in boreal forests. Conservation Ecology 2(2):17. https://doi.org/10.5751/ES-00079-020217

Pavlacky Jr, D. C., P. M. Lukacs, J. A. Blakesley, R. C. Skorkowsky, D. S. Klute, B. A. Hahn, V. J. Dreitz, T. L. George, and D. J. Hanni. 2017. A statistically rigorous sampling design to integrate avian monitoring and management within Bird Conservation Regions. PLoS ONE 12(10):e0185924. https://doi.org/10.1371/journal.pone.0185924

Plummer, M. 2003. JAGS: A program for analysis of Bayesian graphical models using Gibbs sampling. Proceedings of the 3rd International Workshop on Distributed Statistical Computing 124(125):1-10.

R Core Team. 2024. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/

Rosenberg, K. V., A. M. Dokter, P. J. Blancher, J. R. Sauer, A. C. Smith, P. A. Smith, J. C. Stanton, A. Panjabi, L. Helft, M. Parr, and P. P. Marra. 2019. Decline of the North American avifauna. Science 366(6461):120-124. https://doi.org/10.1126/science.aaw1313

Rosenberg, K. V., J. A. Kennedy, R. Dettmers, R. P. Ford, D. Reynolds, J. D. Alexander, C. J. Beardmore, P. J. Blancher, R. E. Bogart, G. S. Butcher, et al. 2016. Partners in Flight landbird conservation plan: 2016 revision for Canada and continental United States. Partners in Flight Science Committee.

Sauer, J. R., W. A. Link, and J. E. Hines. 2020. The North American Breeding Bird Survey, analysis results 1966-2019. US Geological Survey data release.

Smith, A. C., and B. P. M. Edwards. 2021. 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. Ornithological Applications 123(1):duaa065. https://doi.org/10.1093/ornithapp/duaa065

Smith, A. C., M. A. R. Hudson, C. Downes, and C. M. Francis. 2014. Estimating Breeding Bird Survey trends and annual indices for Canada: how do the new hierarchical Bayesian estimates differ from previous estimates? Canadian Field-Naturalist 128(2):119-134. https://doi.org/10.22621/cfn.v128i2.1565

Smith, P. A., A. C. Smith, B. Andres, C. M. Francis, B. Harrington, C. Friis, R. G. Morrison, J. Paquet, B. Winn, and S. Brown. 2023. Accelerating declines of North America’s shorebirds signal the need for urgent conservation action. Ornithological Applications 125(2):duad003. https://doi.org/10.1093/ornithapp/duad003

Sólymos, P., S. M. Matsuoka, E. M. Bayne, S. R. Lele, P. Fontaine, S. G. Cumming, D. Stralberg, F. K. A. Schmiegelow, and S. J. Song. 2013. Calibrating indices of avian density from non-standardized survey data: making the most of a messy situation. Methods in Ecology and Evolution 4(11):1047-1058. https://doi.org/10.1111/2041-210X.12106

Solymos, P., and D. Stralberg. 2020. BAM generalized national models documentation, Version 4.0. Boreal Avian Modelling Project. https://doi.org/10.5281/zenodo.4018335

Stewart, R. L. M. 2015. Blackpoll Warbler. Pages 440-441 in R. L. M. Stewart, K. A. Bredin, A. R. Couturier, A. G. Horn, D. Lepage, S. Makepeace, P. D. Taylor, M.-A. Villard, and R. M. Whittam, editors. Second atlas of breeding birds of the Maritime Provinces. Birds Canada, Sackville, New Brunswick, Canada. https://www.mba-aom.ca/

Stralberg, D., S. Matsuoka, A. Hamann, E. M. Bayne, P. Sólymos, F. K. A. Schmiegelow, X. Wang, S. G. Cumming, and S. J. Song. 2015. Projecting boreal bird responses to climate change: the signal exceeds the noise. Ecological Applications 25(1):52-69. https://doi.org/10.1890/13-2289.1

Van Wilgenburg, S. L., E. M. Beck, B. Obermayer, T. Joyce, and B. Weddle. 2015. Biased representation of disturbance rates in the roadside sampling frame in boreal forests: implications for monitoring design. Avian Conservation and Ecology 10(2):5. https://doi.org/10.5751/ACE-00777-100205

Van Wilgenburg, S. L., and K. A. Hobson. 2011. Combining stable-isotope (δD) and band recovery data to improve probabilistic assignment of migratory birds to origin. Ecological Applications 21(4):1340-1351. https://doi.org/10.1890/09-2047.1

Van Wilgenburg, S. L., C. L. Mahon, G. Campbell, L. McLeod, M. Campbell, D. Evans, W. Easton, C. M. Francis, S. Haché, C. S. Machtans, et al. 2020. A cost efficient spatially balanced hierarchical sampling design for monitoring boreal birds incorporating access costs and habitat stratification. PLoS ONE 15(6):e0234494. https://doi.org/10.1371/journal.pone.0234494

Wilson, S., J. F. Saracco, R. Krikun, D. T. T. Flockhart, C. M. Godwin, and K. R. Foster. 2018. Drivers of demographic decline across the annual cycle of a threatened migratory bird. Scientific Reports 8(1):7316. https://doi.org/10.1038/s41598-018-25633-z

Wunder, M. B. 2012. Determining geographic patterns of migration and dispersal using stable isotopes in keratins. Journal of Mammalogy 93(2):360-367. https://doi.org/10.1644/11-MAMM-S-182.1