Large-scale context and local insight: songbird population trends in the Great Plains

1 Introduction

Songbirds serve as ecological indicators, reflecting changes in habitat quality and broader ecosystem trends (Rosenberg et al., 2019; Whelan et al., 2008). Their population dynamics provide crucial insights into ecosystem health, highlighting the importance of long-term monitoring for informed conservation planning (Lindenmayer et al., 2022). Empirical population trend estimates distinguish short-term fluctuations from long-term patterns, supporting more effective management. Monitoring programs offer a structured approach to track population trends, with conservation efforts often relying on two approaches: large-scale surveys and targeted studies.

Large-scale and targeted monitoring strategies differ in their design, scope, and objectives, but both contribute valuable insights into avian population trends. Large-scale efforts like the North American Breeding Bird Survey (BBS) are designed to assess population trends across extensive geographic areas and long-time frames. Specifically, the BBS provides a standardized framework through which volunteer observers have conducted annual surveys across the U.S. and Canada since 1966. Resulting data are used to generate trend estimates that guide conservation at regional and continental scales (Sauer et al., 2013; Rosenberg et al., 2017). To achieve broad geographic coverage, these programs must allocate effort across many sites, which often limits the amount and types of data collected at individual locations (LaSorte et al., 2018). In contrast, targeted studies operate at smaller spatial and temporal scales, allowing researchers to tailor survey design and use highly trained observers to collect detailed data on specific habitats, species, or population driving mechanisms (e.g., conservation actions). Thus, the two approaches offer unique perspectives on the conservation status of songbird species.

Large-scale and targeted monitoring approaches fill critical gaps in their capacity to detect and interpret population dynamics. Large-scale programs can reveal range-wide or metapopulation trends that prompt focused investigations into underlying causes or areas of concern. Conversely, targeted studies provide fine-scale information on population dynamics at management-relevant scales. Interpreting local trends within the broader context of large-scale monitoring helps determine whether changes reflect regional patterns or site-specific processes. Agreement in magnitude and direction of trend estimates reinforces confidence in conservation assessments, while discrepancies can reveal scale-dependent dynamics or differences in sampling design. This relationship is evident in studies comparing BBS and local monitoring results (Holmes and Sherry, 1988; Bianchini and Tozer, 2023). At Hubbard Brook Experimental Forest, 12 of 19 species declined in both types of datasets, suggesting regional habitat losses (Holmes and Sherry, 1988). In contrast, two species increased in BBS but declined in the local monitoring program, pointing to site-specific processes such as habitat selection, food availability, or mortality sources (Bianchini and Tozer, 2023). Taken together, these approaches offer a more comprehensive understanding of avian population dynamics across scales, which is essential for prioritizing and implementing effective conservation strategies. However, agreement between monitoring approaches is not guaranteed and depends strongly on how trends are defined, the spatial scale of inference, and the treatment of uncertainty. In particular, differences in sampling design and statistical estimation can lead to divergent conclusions about both the magnitude and direction of population change.

A combined approach that leverages both large-scale trend data and targeted, site-specific insights is especially critical for conserving species in regions experiencing steep population declines, such as the arid ecosystems of the North American Great Plains. Songbirds in this region have undergone some of the most dramatic and consistent population declines in North America (Rosenberg et al., 2019; NABCI, 2022). These losses are driven by a complex suite of threats, including habitat fragmentation and conversion from agriculture and development, woody vegetation encroachment, invasive species, and climate change (Knick et al., 2003; Davis, 2004). As a result, Great Plains ecosystems are now considered among the most imperiled in the world (Somershoe et al., 2018). Integrating broad-scale and targeted monitoring in this context offers the best opportunity to detect trends, understand their drivers, and evaluate the effectiveness of management strategies for songbird species and the ecosystems they represent in the Great Plains.

Here, we compared population trend estimates for seven songbird species in the Great Plains of Montana from 2013 to 2019 using two monitoring approaches: BBS and a targeted study. Statewide BBS trend estimates were obtained from publicly available data. Trends from the targeted study were based on abundance estimates using a multi-species abundance model (Golding et al., 2017). These approaches differ in spatial extent, sampling frame, survey protocols, and observer expertise. Our goal was to evaluate how well large-scale BBS trends reflect local population dynamics and to assess whether targeted monitoring provides additional insights necessary to inform conservation strategies in the Great Plains.

2 Focal songbird species

We focused on seven migratory songbird species representatives of arid ecosystems in Montana, USA spanning a gradient of habitat preferences from grassland to sagebrush. Grassland specialists included chestnut-collared longspur (Calcarius ornatus), horned lark (Eremophila alpestris), lark bunting (Calamospiza melanocorys), thick-billed longspur (Rhynchophanes mccownii), and western meadowlark (Sturnella neglecta; Vickery et al., 1999). Brewer’s sparrow (Spizella breweri; Paige and Ritter, 1999) represented sagebrush-dependent species, while vesper sparrow (Pooecetes gramineus) served as a generalist (Jones and Cornely, 2002). By including species with varied habitat use and life history strategies, we aimed to capture a broad perspective on population trends in the northern Great Plains ecosystem.

3 Monitoring programs - data collection

3.1 Large-scale monitoring: breeding bird survey

For this study, we retrieved publicly available BBS trend estimates (USGS, 2023) based on data collected in Montana for all focal species from 2013 to 2019, which is the timeframe overlapping the targeted study. There are a total of 103 BBS routes potentially surveyed across the state of Montana each year (Figure 1). A route is surveyed once per year following a standardized protocol in which a single volunteer observer conducts 3-minute point counts at 50 stops along a 39.4 km roadside route, beginning 30 minutes before sunrise and recording all birds seen or heard within a 400 m radius between late May and early July. BBS routes are distributed statewide, across multiple types of habitats (e.g., forests, grasslands, sagebrush steppe) and not designed to align with the targeted study area, spatial overlap with the targeted study is limited (Figure 1). Visual and auditory bird observations are recorded.

3.2 Target monitoring: local study

The targeted study area encompassed ~ 2160 km² near the town of Roundup in central Montana (Figure 1). The landscape is characterized by mixed-grass prairie and sagebrush habitat with Wyoming big sagebrush (Artemsia tridentata) interspersed with western wheatgrass (Pascopyrum smithii), needle-and-thread grass (Stipa comata), blue grama (Bouteloua gracilis), and prairie Junegrass (Koeleria macrantha). We randomly selected 66 sampling plots, each measuring 500 m X 500 m (25 ha), from public and private lands with pre-established access, ensuring a minimum spacing of 500 m between plots. Bird surveys were conducted from May through early July in 2013 to 2019, with each plot surveyed three times over the breeding season. Surveys followed the dependent double observer transect method (Nichols et al., 2000; Golding and Dreitz, 2017; Golding et al., 2017). The dependent double observer survey method has been shown to provide higher and more precisely estimated detection probabilities, and thus, abundance estimates, in arid systems relative to single-observer methods (Golding and Dreitz, 2016).

The dependent double observer approach requires two observers to walk a transect with the secondary observer walking 3–5 m behind the primary observer. Surveys were conducted along a U-shaped transect, allowing the two observers to survey up to 125 m on each side of the transect and effectively cover the entire plot (Golding and Dreitz, 2016). The primary observer identified each bird’s species and location and relayed this information to the secondary observer, who recorded detections and noted any additional observations missed by the primary observer (Nichols et al., 2000). Auditory bird detections during the survey required visual confirmation before being recorded. Observer roles were alternated between surveys to estimate and account for individual variation in detection. Each year from 2013 to 2019, eight observers conducted the avian surveys in four two-person teams, resulting in 56 observer-years. Additional details of dependent double observer can be found in Golding and Dreitz (2016, 2017) and Golding et al. (2017).

4 Trend analyses

We retrieved publicly available BBS trend estimates for our study species in Montana from 2013 to 2019 (USGS, 2023) and developed comparable trend estimates from our targeted study data over the same period. Both sets of estimates were generated using hierarchical models that separate biological and observational processes, allowing for estimation of abundance while accounting for detection probabilities. Each model was implemented within a Bayesian framework using Markov chain Monte Carlo (MCMC) methods. This iterative procedure generates a distribution of replicated values from which posterior estimates and credible intervals are derived. All data visualization, statistical analyses, model diagnostics, and comparisons were conducted in R (R Core Team, 2024). The targeted study analysis was implemented in JAGS (Plummer, 2003) and accessed through R.

Although both modeling approaches were implemented within a comparable statistical framework, the resulting trend metrics differ in how they summarize temporal change. BBS trends are derived from endpoint, whereas the targeted study trends are estimated from regression across all years, the two metrics are not directly equivalent. Endpoint-based estimates may be more sensitive to interannual variability, whereas regression-based estimates incorporate information across the full time series.

4.1 Breeding bird survey

BBS trend estimates we retrieved are derived using the best-fitting model variant selected from four standard BBS model types, based on model fit criteria. All BBS estimates used an overdispersed Poisson regression framework, as described by Link et al. (2020), that accounts for multiple sources of variation in bird count data, including year, observer, and overdispersion effects. Year effects represent relative abundance or trend, describing temporal patterns in bird populations without attempting to estimate absolute abundance. Observer effects capture differences among observers and changes within observers over time, while route-level variation is incorporated through random effects associated with observer-route combinations. Overdispersion effects capture extra variation in counts beyond what a standard Poisson model predicts and are modeled as a mean-zero random term added to the linear predictor. The publicly available annual BBS trend estimates are calculated as the ratio of annual indices corresponding to the first and last years of the selected interval, 2013 and 2019 for this study.

4.2 Targeted study

We used a multispecies dependent double-observer abundance model (Golding et al., 2017) to estimate absolute abundance for each of the seven focal species then estimated species-specific temporal trends. This model is a type of N-mixture model implemented in a Bayesian hierarchical framework to estimate detection probabilities and species-specific abundances. It includes species-specific fixed effects for year to capture temporal trends and a random effect for site to account for spatial variability and overdispersion in counts. The observation process uses the dependent double-observer method to estimate detection probability by comparing paired observer detections, including zero counts of a focal species, and correcting for imperfect detection with observer-specific detection histories. To estimate population trend, we calculated a species-specific linear regression slope by regressing the log of posterior-predicted abundance of each species on year. This slope represents the rate of change in log-abundance through time. We exponentiated the slope to express trend as an annual multiplicative rate of change and then converted this value to a percent change per year for ease of interpretation. Expressing trends as annual percent change places the targeted study estimates on a scale that is more directly interpretable and broadly comparable to the publicly available BBS trend estimates, despite differences in how trends are calculated. Specifically, we computed the ordinary least-squares slope of log-abundance across years for each species using a closed-form expression, then exponentiated this slope to express the trend as an annual multiplicative change and converted it to a percent change. We obtained 95% credible intervals directly from the posterior distribution of the annual percent change derived from this slope.

4.3 Trend comparisons

Recognizing that different monitoring approaches may yield different insights, we directly compared population trend estimates from the BBS and targeted study using complementary descriptive approaches. First, we conducted visual comparison of species-specific trend estimates and their associated 95% credible intervals using side-by-side plots for each species (Figure 2). Overlap in CRIs was used as an indicator of whether the two approaches produced estimates that were similar given their respective uncertainty.

Second, we evaluated agreement in trend direction by classifying each species-specific estimate as increasing, decreasing, or uncertain. Trends were classified as increasing or decreasing when the 95% CRI excluded zero and as uncertain when the CRI overlapped zero. We then summarized concordance in direction between the two monitoring approaches across species. We also compared the magnitude of the differences in trend estimates between methods by calculating the difference between targeted study and BBS estimates for each species. Positive values indicate more positive (or less negative) trends in targeted study relative to BBS, whereas negative values indicate more negative trends. This calculation allowed use to evaluate how the magnitude of inferred population changes varied between monitoring approaches.

Finally, we used a Wilcoxon signed-rank test to assess whether there was a systematic difference in trend estimates between BBS and the targeted study across species. Given the small sample size (n = 7 species), and recognizing the Wilcoxon signed-rank test assumes independent paired differences, we treated this test as exploratory and interpreted results cautiously, with primary inference based on descriptive comparisons of trend magnitude, uncertainty, and direction.

5 Results

5.1 Observational counts

The number of observations for the seven focal species varied considerably between the two monitoring programs between 2013 and 2019 (Table 1). The BBS recorded a total of 93,330 observations for all species in Montana, with annual counts ranging from 8,469 observations in 2013 to 16,784 in 2017. The number of completed BBS routes was lowest in 2013 (55 routes) and highest in both 2017 and 2018 (88 routes) which corresponded with the highest observation totals (16,784 and 16,777 respectively). In contrast, the targeted study recorded approximately half as many observations (46,869 observations), with annual counts ranging from 4,062 observations in 2015 to 8,753 observations in 2013. The number and location of sampling plots in the targeted study remained constant across years, with 66 fixed plots surveyed three times each breeding season.

Table 1. Annual observations of seven migratory songbirds associated with arid ecosystems in North America from 2013–2019 during a large-scale, state-wide bird monitoring program, Breeding Bird Survey, and a targeted, local monitoring program near Roundup, Montana, USA.

Species-specific observation totals also differed by year and monitoring program. In the BBS data, western meadowlark was consistently the most frequently recorded species, accounting for 42-53% of the annual observations. In contrast, vesper sparrow or Brewer’s sparrow were most observed each year of the targeted study, comprising 20-28% of annual observations. Thick-billed longspur and Brewer’s sparrow were observed more frequently in the targeted study than BBS across all years. Conversely, horned lark, lark bunting, and western meadowlark were consistently observed more in BBS. Observations of chestnut-collared longspur and vesper sparrow observations fluctuated between the two programs, with no consistent pattern.

5.2 Trend comparison

Annual trend estimates from BBS and the targeted study showed different patterns across species (Table 2; Figure 2). In several cases, point estimates suggested similar directional trends between methods. However, BBS estimates were generally less precise with 95% credible interval (CRIs) that frequently overlapped zero, whereas the targeted study estimates were typically more precise and more often indicated directional change. For example, Brewer’s sparrow and horned lark both showed declining trends in the targeted study, with BBS estimates suggesting similar but more uncertain declines. In contrast, chestnut-collared longspur exhibited a positive trend in the targeted study, while BBS estimate was near zero with a wide CRI overlapping zero.

Table 2. Comparison of annual population trend estimates from Breeding Bird Survey, and a targeted, local monitoring program for seven migratory songbirds associated with arid ecosystems in Montana, USA, 2013–2019.

Trend estimates are expressed as percent change. Difference (Targeted – BBS) were calculated as the targeted-study estimate minus the BBS estimate. CRI overlap indicates whether 95% credible intervals overlap. Trend direction (increase, decrease, or uncertain) was based on whether CRIs excluded zero and is shown as BBS/Targeted. Direction match indicates whether classifications were the same.

Differences in magnitude between monitoring programs where also evident (Table 2; Figure 2), with the targeted study generally producing larger point estimates of annual percent change than BBS. The clearest divergence in magnitude occurred with thick-billed longspur, for which BBS estimated a positive but highly uncertain trend (11.34%, 95% CRIs: -1.76 – 27.18%), whereas the targeted study estimated a strong decline (-13.95%, 95% CRIs: –14.50 – -13.43%). For horned lark, thick-billed longspur, vesper sparrow, and western meadowlark, the CRIs from the BBS and the targeted local study did not overlap, suggesting differences in estimated trends between the two datasets. In contrast, CRIs overlapped for Brewer’s sparrow, chestnut-collared longspur, and lark bunting, indicating that the estimates were not clearly distinguishable given uncertainty.

Overall, agreement between BBS and targeted monitoring approaches was conditional and depended on how trend magnitude, uncertainty, and direction were considered. When trends were summarized by direction including uncertainty, concordance between methods was limited (Table 2). BBS estimates were often uncertain due to CRIs overlapping zero, whereas the targeted study estimates more frequently indicated directional change. As a result, agreement in trend direction depended strongly on how uncertainty was incorporated into the interpretation of trends.

The Wilcoxon signed-rank test did not detect a systematic difference in trend estimates (V = 7, p = 0.30). However, given the small sample size (n = 7 species) and limited power to detect differences, this result should be interpreted cautiously and should not be interpreted as evidence of equivalence between approaches.

6 Discussion

Our comparison of population trend estimates from large-scale (BBS) and targeted monitoring programs showed that agreement varied across species, with some trends aligning and other diverging, and many depending on how uncertainty was interpreted. These patterns reflect both methodological differences and the contrasting spatial scales of the two approaches. Although point estimates from the two approaches often suggested similar directional tendencies based on their point estimates, agreement between approaches was strongly influenced by differences in uncertainty associated with those estimates. BBS CRIs frequently overlapped zero, resulting in most trends to be considered uncertain, whereas the targeted study estimates more often indicated directional change. As a result, agreement in trend direction between monitoring approaches was limited once uncertainty was taken into account (Table 2), despite some similarities in point estimates.

Differences between monitoring programs were evident across species. For several species, including Brewer’s sparrow, horned lark, lark bunting, and vesper sparrow, the targeted study estimates indicated declines, while BBS estimates suggested similar but more uncertain trends. In contrast, chestnut-collared longspur showed a positive trend in the targeted study but a near-neutral and uncertain trend in the BBS. The most pronounced divergence occurred for thick-billed longspur, for which the targeted study estimated a strong local decline, whereas BBS estimates suggested a positive but highly uncertain trend. More broadly, discrepancies between approaches reflected differences in both the magnitude and precision of estimated trends across species.

However, it is crucial to consider that some differences in trend estimates may also arise from methodological variation. The BBS, for example, is largely road-based and more routes are completed near human population centers (Bart et al., 2005). This design can under sample remote or specialized habitats, yielding less reliable estimates for habitat-specialist species due to both potential bias and reduced precision. Targeted studies, in contrast, offer finer ecological resolution and can better capture species-habitat associations and population mechanisms, but their narrow spatial extent and focus on intact or managed sites may overrepresent population status relative to degraded areas. For species like thick-billed longspur, whose habitat preferences are highly specific, these design differences can substantially affect the apparent trajectory.

Our findings align with other monitoring efforts that demonstrate substantial spatial heterogeneity in the trend of songbird populations. For example, BBS estimates from 2013 to 2019 at the national level suggest Brewer’s sparrows are broadly stable in the US (0.52%, 95% CRIs: -1.65 – 2.98%; USGS, 2023), while Montana-specific routes show more declining patterns (-3.33%, 95% CRIs: -8.26 – 0.90%). Similarly, other targeted studies in Montana have documented variable trends in songbirds depending on site-specific conditions, with some habitats functioning as populations strongholds (e.g., high nest densities and success in native grasslands for western meadowlark in eastern Montana; Fondell and Ball, 2004) and other reflecting sink dynamics (e.g., reduced productivity in cropland sites for thick-billed longspur in northern Montana; Swicegood et al., 2023). These comparisons highlight the spatial heterogeneity in population responses and the importance of placing trend estimates in ecological and geographic contexts.

Differences in trend estimates can also arise from analytical choices. For example, the BBS calculates trend using only the first and last year of the estimate, whereas the targeted trend uses a regression across all years. Using only endpoints can produce highly variable estimates, as the trend can change dramatically depending on the years sampled. This effect is particularly pronounced for species, like many songbirds, whose local abundances or distributions fluctuate substantially among years in response to changing environmental conditions (Holmes, 2007). We chose the regression approach because it incorporates information from all years, providing a more stable representation of trend while yielding broadly similar results.

Although our results show where BBS and targeted trends diverge, separating ecological drivers from methodological ones is challenging. Some mismatches are consistent with biological processes, whereas others more plausibly reflect differences in sampling design or the way trends are estimated. In most cases, both factors are likely to contribute, and our study can identify patterns consistent with each explanation but cannot definitively attribute divergences to a single cause. Future work designed to better separate biological patterns from methodological effects could use comparable survey designs and analytical approaches applied across multiple spatial scales, allowing differences in trend estimates to be attributed more confidently to ecological rather than procedural factors. Conversely, applying different survey or analytical methods to the same spatial area would help isolate methodological sources of divergence. Thus, for our study, interpreting divergences in trend estimates therefore requires joint consideration of possible ecological context, survey design, and analytical choices.

Future research could also benefit by formally integrating broad-scale and targeted surveys within a unified analytical framework such as an integrated population model (IPM). IPMs combine multiple data sources within a hierarchical structure, allowing separation of biological process variation from observation error while leveraging the complementary strengths of different survey designs (Schaub and Abadi, 2011; Kéry and Schaub, 2012). Integrating the extensive spatial coverage of the BBS with ecological specificity and demographic resolution of targeted studies could improve precision, reduce survey-specific bias, and provide insight into demographic mechanisms underlying observed trends. More broadly, formal data integration has been shown to strengthen inference about population trajectories and drivers of change compared to analyses conducted on single datasets in isolation (Zipkin and Saunders, 2018). While analytically more complex, data integration may provide robust and mechanistically informative conservation guidance than evaluating each dataset separately.

Overall, this study highlights the complementary roles of broad-scale and localized monitoring in understanding avian population change. Large-scale programs such as the BBS provide essential baselines and the statistical power to detect widespread shifts, whereas targeted studies offer greater ecological resolution and insight into species-specific responses and habitat quality. Apparent concordance between approaches strengthens confidence in inferred population trajectories, while divergences can reveal scale-dependent dynamics, methodological sensitives, or emerging ecological change that merits further investigation. However, interpretation of agreement between monitoring program trends depends strongly on how uncertainty is considered. Evaluating these monitoring program trends cautiously and concurrently improves interpretation of population change and supports more informed conservation planning than reliance on any single monitoring framework alone.

6.1 Conservation implications

By combining insights from both large-scale and targeted monitoring programs, researchers and managers can more accurately assess population status, identify priority habitat areas, and prioritize conservation actions in regions or with species at greatest risk. Linking regional population trajectories to local ecological conditions enhances adaptive management by clarifying where observed declines reflect widespread drivers or site-specific constraints. Maintaining habitat heterogeneity and landscape connectivity will be critical for sustaining viable populations under continued environmental change. Incorporating cross-scale monitoring into conservation policy frameworks can strengthen decision-making by grounding management actions in both statistically robust trend estimates and ecologically resolved local data.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Marco Zaccaroni, University of Florence, Italy

Reviewed by: Roi Dor, Open University of Israel, Israel

Stuart Pimm, Duke University, United States