Frontispiece. Chandler S. Robbins counting birds at a North American Breeding Bird Survey stop, ca. 1985. Photograph by Barbara A. Dowell.
North American Fauna | www.fwspubs.org December 2013 | Number 79 | 1 Monograph The North American Breeding Bird Survey 1966–2011: Summary Analysis and Species Accounts
John R. Sauer*, William A. Link, Jane E. Fallon, Keith L. Pardieck, David J. Ziolkowski Jr. United States Geological Survey, Patuxent Wildlife Research Center, 12100 Beech Forest Road, Laurel, Maryland 20770
Abstract
The North American Breeding Bird Survey is a roadside, count-based survey conducted by volunteer observers. Begun in 1966, it now is a primary source of information on spatial and temporal patterns of population change for North American birds. We analyze population change for states, provinces, Bird Conservation Regions, and the entire survey within the contiguous United States and southern Canada for 426 species using a hierarchical log-linear model that controls for observer effects in counting. We also map relative abundance and population change for each species using a spatial smoothing of data at the scale of survey routes. We present results in accounts that describe major breeding habitats, migratory status, conservation status, and population trends for each species at several geographic scales. We also present composite results for groups of species categorized by habitats and migratory status. The survey varies greatly among species in percentage of species’ range covered and precision of results, but consistent patterns of decline occur among eastern forest, grassland, and aridland obligate birds while generalist bird species are increasing.
Keywords: hierarchical models; population survey; species groups Received: October 27, 2011; Accepted: July 12, 2013; Published Online Early: August 2013; Published: December 2013 Citation: Sauer JR, Link WA, Fallon JE, Pardieck KL, Ziolkowski DJ Jr. 2013. The North American Breeding Bird Survey 1966-2011: Summary Analysis and Species Accounts. North American Fauna 79: 1–32. doi:10.3996/nafa.79.0001 Copyright: All material appearing in North American Fauna is in the public domain and may be reproduced or copied without permission unless specifically noted with the copyright symbol ß. Citation of the source, as given above, is requested. The findings and conclusions in this article are those of the author(s) and do not necessarily represent the views of the U.S. Fish and Wildlife Service. * Corresponding author: [email protected] Subject Editor: Colleen Handel, United States Geological Survey, Anchorage, Alaska
Introduction United States, Canada, and Mexico. From their efforts, comprehensive summaries of population The North American Breeding Bird Survey (BBS) change have been calculated for .400 species of is a unique collaborative effort to increase our birds (Sauer et al. 2003). For most of these species, understanding of North American bird populations. the BBS forms the only source of information on the Started at a time when concern about bird popu- dynamics of the populations. BBS observers must be lations was focused on effects of pesticides, the BBS able to identify birds by sound and sight, and the now serves as the primary data source for estimating population changes of North American birds and for attention placed on competence of observers (at modeling the possible consequences of changes in first informally, and more recently through online land use, climate, and many other possible stressors training programs) greatly increased the rigor of the on bird populations. Jointly coordinated by the information, paving the way for additional bird United States Geological Survey, the Canadian survey programs that rely on volunteer observers. Wildlife Service, and, since 2008, the Mexican Although it has limitations, no other survey comes National Commission for the Knowledge and Use close in combining public participation and scientific of Biodiversity, the BBS incorporates the efforts of rigor to provide information for bird conservation thousands of volunteer bird counters across the and natural history.
North American Fauna | www.fwspubs.org December 2013 | Number 79 | 2 The North American Breeding Bird Survey 1966–2011 J.R. Sauer et al.
This summary of BBS results presents a synthesis implemented a pilot effort in Maryland and of information on bird population change and Delaware in 1965 and then inaugurated the survey distribution. We integrate habitat and migration in the eastern United States in 1966. Canadian status information, distribution maps depicting collaborators, particularly Anthony Erskine, imple- relative abundance, and current estimates and mented the survey in southeastern Canada in 1966 graphical illustrations of population change into as well. Survey routes were established in the central accounts for 426 species of North American birds. United States in 1967, and by 1968 the sampling The accounts provide a concise description of both framework had grown to include the continental the present efficiency of the survey, in terms of United States and much of southern Canada. precision of estimate of population change, and the Subsequently, additional survey routes have been population status for each species as described by the established, including efforts to get better informa- survey. We also present analyses of multispecies tion from remote regions of the western United groups to summarize patterns of population change States and northern Canada. Alaska also has many for groups of conservation interest. By providing a BBS routes along the limited road system (,70 survey-wide overview of species coverage by the routes sampled each year 1993 to present), but few BBS, this volume complements earlier comprehen- samples prior to 1993. Consequently, Alaska data sive summaries of the BBS (Robbins et al. 1986; cannot be analyzed for time periods comparable Sauer et al. 2003). Detailed regional information to other regions. Alaska trends for recent years regarding population change for all species is have been presented on the BBS Analysis and available on the BBS Summary and Analysis website Summary Website (Sauer et al. 2011), and research (Sauer et al. 2011). is underway to assess the feasibility of developing a combined on- and off-road survey to provide a more A Brief History of the North comprehensive survey of the state (Colleen Handel, American BBS U.S. Geological Survey [USGS], personal commu- nication). BBS-style surveys were conducted in The North American BBS was developed in Puerto Rico from 1997 to 2007, and an expansion response to a need for better information about of the survey into Mexico began in earnest with the population change in songbirds in North America. establishment of routes in northern areas starting Although legal requirements for management of in 2008. harvested species led to the development of conti- By 2010, there were 5,267 survey routes in the nent-scale surveys for waterfowl in the 1950s (Martin database and 3,140 of them were surveyed in 2010. et al. 1979), no such surveys existed for nongame Coverage varies across North America (Figure 1). birds. However, with the publication of Rachel Using routes per degree block as a summary statistic (as Carson’s Silent Spring (Carson 1962) and the ensuing calculated from degree blocks within the current BBS increased public awareness of threats to nongame analysis area, see Maps of Relative Abundance) bird populations, a need quickly became apparent. It comparatively fewer routes occur in the western was evident that the paucity of current information United States (x¯ = 2.73 [95% Confidence interval: about populations of most bird species and the lack 2.62, 2.83], N = 680 west of the 100th meridian and of a scheme by which to collect future data severely 4.53 [4.40, 4.67], N = 658 east of the 100th meridian) limited our ability to assess the consequences of with even fewer in northern Canada (1.64 [1.51, 1.77], pesticides and other stressors on these species. N = 310 north of the 49th parallel and 4.2 [4.11, Chandler S. Robbins, who was already a 4.31], N = 1,029 south of the 49th parallel). distinguished researcher in bird populations at the Patuxent Wildlife Research Center in Laurel, Methods Maryland, was able to convince his superiors in the U.S. Fish and Wildlife Service of the need for a Field methods of the BBS continent-scale monitoring program for nongame BBSs are conducted along roadside survey routes birds. Robbins realized that roadside-based survey composed of 50 independent stops, each stop located methods that he and colleagues had previously used 800 m (0.5 mi) apart. Routes are surveyed once for monitoring mourning dove Zenaida macroura and annually by an observer, generally a volunteer, who American woodcock Scolopax minor could be modified can identify birds by both sight and song. The to count many species simultaneously. He was observer selects a morning in June to conduct the already an influential member of the growing bird- survey (dates in late May are permissible in southern watching community and knew that a competent states and in early July in northern provinces); observer workforce could be recruited for such an observers are encouraged but not required to be effort. Robbins experimented with alternative ap- consistent among years in date chosen to survey the proaches to counting birds for several years before route. The survey commences at a pre-established settling on the protocol that is still used today. He start time (30 min before local sunrise) and the
North American Fauna | www.fwspubs.org December 2013 | Number 79 | 3 The North American Breeding Bird Survey 1966–2011 J.R. Sauer et al.
Figure 1. Map of route locations in the continental United States and southern Canada. observer progresses to each stop in sequence, survey are available on the BBS operations website parking in a safe pull-off at each. Route maps are (www.pwrc.usgs.gov/bbs/). provided to observers to assist them in finding stops, along with written stop descriptions and GPS Analysis of population change from locations on some routes (Figure 2). BBS data At each stop, the observer exits the vehicle and Constraints on estimation of population change and conducts a 3-min count of all birds heard and those historical analyses. Statistical analysis of BBS data can seen within a 400-m (0.25-mi) radius around the be controversial because complications associated point. The observers record data on data sheets with the nature of the count data and the scale of the provided to them (Figure 3). The same stop locations survey limit the application of standard sample are surveyed in subsequent years to ensure consis- survey methods (e.g., Sauer et al. 2003). The survey tency in sampling, although stops and routes are routes sample local populations, and inference occasionally relocated for reasons related to safety or regarding population change at regional scales access. Supplemental information including weather requires accommodation for 1) repeated counts of variables, disturbance (e.g., number of cars that pass the survey over years on routes, 2) inequities in the stop during the count period), and other data are numbers of samples among regions, 3) missing data also collected during the survey. on routes, and (4) changes in the BBS index After conducting the surveys, observers submit associated with changes in detectability. The issue their data to the respective national BBS offices, of detectability and attendant consequences for where the data are then processed into both raw inference regarding population change have been a and summarized formats for public release via the source of controversy from the start of the BBS. internet. These data, the protocols for conducting Because observers miss a portion of birds when the survey, and additional details regarding the counting, and no means exist in the present protocol
North American Fauna | www.fwspubs.org December 2013 | Number 79 | 4 The North American Breeding Bird Survey 1966–2011 J.R. Sauer et al.
Figure 2. Stop locations on field map used in Breeding Bird Survey. to estimate directly the fraction of birds missed, Innovations in analyses of BBS data reflect the controlling for factors that influence the proportion advances in computers and statistical methods over of birds counted is a critical component of the the period 1966–2011. In early years of the survey, analysis. Observer differences in counting ability are file storage and computational limitations con- well known to influence detectability (e.g., Sauer et strained analyses to methods such as route regression al. 1994; Kendall et al. 1996), and most analyses of that could be implemented on available computers. BBS data have controlled for observer effects on Modern computer-intensive approaches such as the counts (e.g., Sauer et al. 2008a). hierarchical models presented here provide many
North American Fauna | www.fwspubs.org December 2013 | Number 79 | 5 The North American Breeding Bird Survey 1966–2011 J.R. Sauer et al.
Figure 3. Breeding Bird Survey field data sheet. opportunities for analyses that more realistically and population change because missing data intro- model the multiscale, repeated-measure nature of duced spurious patterns in change in abundance the survey. (Geissler and Noon 1981; Robbins et al. 1986). To It was evident from the earliest BBS summaries control for effects of missing data, early summaries that comparisons of simple means of counts within of BBS data relied on methods that analyzed regions were an inadequate summary of abundance population change from subsets of data that were
North American Fauna | www.fwspubs.org December 2013 | Number 79 | 6 The North American Breeding Bird Survey 1966–2011 J.R. Sauer et al.
‘‘comparable’’ over time (i.e., the routes that were subject to these criticisms. Although a modified consistently surveyed by a single observer). Unfor- version of Poisson-based route regression (Link and tunately, methods that estimate change from ratios Sauer 1994) is used to estimate local trend for of counts at different times from small samples of mapping (see Population Change Maps), regional comparable routes produce biased estimates (Geiss- analyses results are now based on hierarchical models. ler and Noon 1981). Geissler and Noon (1981) Hierarchical model analysis. Hierarchical models suggested a route-regression analysis, in which linear provide a comprehensive framework for estimating regression is used to estimate change on individual population change and annual indexes of abundance survey routes and regional trends are computed from BBS data. The hierarchical models we use are as an average of these route-change estimates. In a class of generalized linear mixed models that early applications (Robbins et al. 1986; Geissler permit year, stratum, and observer effects to be and Sauer 1990), route regression was based on a governed by parameters that are random variables. normal regression with natural-logarithm trans- This hierarchical structure allows us to model the formed counts with a 0.5 constant added to influence of regions, observers, and other factors on accommodate zero counts. However, count data the distributions of the parameters influencing from the BBS are more naturally modeled as a counts, rather than on the individual counts generalized linear model, and Link and Sauer (1994) themselves. Using these models, we can formulate defined a Poisson regression with log links fit using regional summaries in terms of model parameters, estimating equations. In both analyses the goal was avoiding the ad hoc weightings used in the route- to estimate rate of change (the slope associated with regression approach. See Link and Sauer (2002) and time), with year and categorical observer data as Sauer and Link (2002) for details of these analyses predictors of the counts. The analysis controlled for and the conceptual issues associated with regional observer effects by allowing a different intercept for models of bird population change. each observer in the analysis (Link and Sauer 1994). Hierarchical models are often fit using Bayesian Interval-specific rate of change (trend) for a region methods, in which inference is based on the posterior was estimated as a weighted average of individual distributions of parameters. Bayesian analyses require route-slopes for the interval of interest, with mean specification of prior distributions of parameters and abundance and survey consistency used as route- the sampling distributions of the data. Although most specific weights and an area weight included to realistic Bayesian analyses are difficult to implement, accommodate regional differences in sample frames simulation-based Markov chain Monte Carlo meth- in multistrata analyses. Variances of these trend ods (MCMC; Lunn et al. 2000) can be used to estimates were estimated by bootstrapping. See approximate the posterior distributions used in Geissler and Sauer (1990) and Link and Sauer Bayesian inference. Using MCMC, posterior distri- (1994) for additional information about the route- butions and associated means, medians, and credible intervals (Bayesian confidence intervals) can be regression method and the weighting factors. calculated iteratively for parameters of interest from Additional analyses estimated annual indexes from many complicated models. Link and Barker (2010) residual variation associated with yearly data around discussed the methods and philosophical basis of the predicted trends (Sauer and Geissler 1990). Bayesian inference. Route regression proved to be a robust approach In the BBS hierarchical model, counts Y for estimation of population change (Thomas 1996) i,j,t (indexed by i for stratum, j for unique combinations but had clear limitations. Although convenient of route and observer, and t for year) are assumed computationally, the route-by-route summaries of to be independent Poisson random variables with change had no direct method to control for within- means l that can be described by log-linear interval variation in the quality of information. i,j,t functions of explanatory variables, Trends could be estimated for a region even if most ~ z { z of the route data were collected during the later log li,j,t Si biðÞt t (1) years of the interval. Because there was no way to v zc zgI ( j,t) ze determine whether route data in a region adequately j( i) i,t i,j,t represented the interval, a trend may have been which are stratum-specific intercepts (S) and slopes based on a preponderance of data from early or late (b), along with effects for observer/route combina- in the interval. The weightings, although constructed tions (v), year (c), start-up (g, with indicator I(j,t) to permit estimation of change for the total that takes value 1 for an observer’s first year of population, were criticized because the weights were survey on a route and 0 otherwise), and over- ad hoc approximations of unknown quantities (ter dispersion effects (e). The model requires specifica- Braak et al. 1994). Estimation in route regression tion of prior distributions for parameters. In our was focused on trend, and annual indexes were analysis, Si and bi are assigned diffuse (essentially flat) computed as residuals of the estimated trends (Sauer normal prior distributions, and other effects are and Geissler 1990), hence annual indexes were also specified as having mean zero normal distributions.
North American Fauna | www.fwspubs.org December 2013 | Number 79 | 7 The North American Breeding Bird Survey 1966–2011 J.R. Sauer et al.
Observer–route effects (v) are identically distributed predictor and log-transformed annual indexes as the 2 with common variance s v and overdispersion dependent variable. This definition of trend is easily effects (e) are identically distributed with common implemented as a derived statistic in the MCMC 2 variance s e. Variance of the year effects (c)is summaries by calculating a linear regression through the 2 allowed to vary among strata (s c,i). All variances annual indexes derived from each iteration and are assigned flat inverse gamma prior distributions. calculating the percentage change from the estimated From the model parameters, annual indexes of slope parameter associated with year. As with other abundance and trend are defined as derived posterior distributions, median slopes and credible parameters. Stratum-specific annual indexes of intervals are calculated directly from the MCMC results. abundance (ni,t, an index to the number of birds The program WinBUGS (Lunn et al. 2000) was per route in stratum i at year t) incorporate summed used to fit this model, to conduct the MCMC analysis, and exponentiated stratum effects, stratum-specific to evaluate summary statistics to determine when the trend and year effects, and associated variance Markov Chains are converging, and to summarize components added to accommodate asymmetries results. MCMC produces Markov chains that sample in the lognormal distribution: the posterior distribution once they become station- ary. We conducted the analysis for at least 20,000 n ~ i,t iterations to ensure stationary results, and then ð2Þ 2 2 conducted another 20,000 iterations to obtain results exp S z b { z c z 0:5s z 0:5s i i(t t ) i,t v e for estimating the posterior distributions. Some species required additional iterations before usable Stratum totals are Ni,t = Aini,t, where Ai is the area of estimates were obtained; for others, the large data sets the stratum. To obtain indexes for larger areas proved difficult to manage and we could only (groups of strata, e.g., states or countries), we sum summarize 10,000 replicates. For summary, we the Ni,t over the relevant i. For presentation, ‘‘thinned’ the resulting chains by a factor of 2 (i.e., composite indexes Nt are scaled by the total areas, using every second observation for calculating obtaining a summary on the scale of birds per route, estimates and credible intervals). We also output the ,X ~ MCMC replicates and used them to compute nt Nt i Ai . The strata are the regions formed by the intersection of Bird Conservation Regions regional summaries using FORTRAN programs. (BCRs; e.g., Sauer et al. 2003; see Population Change Summary) within states or provinces. For Population change summary summary, results from these areas are combined to Trend and annual index results are the primary provide indexes at the scale of states, provinces, summary information for species from BBS analyses. BCRs, and the entire survey area. Here, we present long-term composite trend results We define trend as an interval-specific geometric (B¯ ), for the intervals 1966 to 2011 and 2001 to 2011. mean of proportional changes in population size, Since the survey was implemented at different times expressed as a percentage (cf., Link and Sauer 1998). in different regions, trend estimates and indexes Thus the trend from year ta to year tb for stratum i is were computed using the first year in which data 100(Bi 2 1)%, where were available for the species; that is, western species