- Email: [email protected]
Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013
JGLR-00995; No. of pages: 10; 4C: Journal of Great Lakes Research xxx (2015) xxx–xxx
Contents lists available at ScienceDirect
Journal of Great Lakes Research journal homepage: www.elsevier.com/locate/jglr
Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013 Douglas C. Tozer ⁎ Bird Studies Canada, Box 160, 115 Front Street, Port Rowan, ON N0E 1M0, Canada
a r t i c l e
i n f o
Article history: Received 7 November 2014 Accepted 15 October 2015 Available online xxxx Communicated by Craig Hebert Index words: Citizen science Coastal wetland Great Lakes Multi-scale Marsh bird Multi-season site occupancy model
a b s t r a c t Using data from 21,546 point counts conducted by volunteers in Bird Studies Canada's Great Lakes Marsh Monitoring Program, I assessed whether occupancy of 15 breeding marsh bird species increased or decreased throughout the southern portion of the Great Lakes basin between 1996 and 2013. I accounted for differences in detection probability, addressed spatial autocorrelation, and assessed whether initial occupancy in 1996 and subsequent colonization or extinction at a site within and across species was influenced by site, wetland, and landscape scale covariates. Occupancy of 9 of 15 (60%) species significantly decreased, whereas occupancy of only 1 (7%) species significantly increased. The results show the power of citizen science and suggest that the largest number of decreasing marsh-dependent breeding bird species will benefit from conserving, restoring, or creating large wetlands surrounded by limited urban land use, and from addressing issues within International Joint Commission Areas of Concern. Plus, individual or smaller groups of decreasing species will also benefit from conserving, restoring, or creating robust-emergent-dominated but interspersed, purple loosestrife (Lythrum salicaria)-free, Phragmites-free wetlands surrounded by higher proportions of wetland cover in the surrounding landscape, and from addressing issues within Great Lakes coastal wetlands. These actions will help promote colonization or reduce extinction and help slow or maybe even reverse declining trends in occupancy among decreasing species across the southern portion of the Great Lakes basin. © 2015 International Association for Great Lakes Research. Published by Elsevier B.V. All rights reserved.
Introduction Monitoring marsh dependent breeding birds can be challenging due to the birds' secretive behavior, low numbers, infrequent occurrence, and inaccessible habitat (Conway and Gibbs, 2011; Steidl et al., 2013). For instance, the Breeding Bird Survey (BBS), a passive, long-term, roadside monitoring program (Environment Canada, 2010), which is often considered the best source of information on North American bird population trends (e.g., North American Bird Conservation Initiative Canada, 2012), detected secretive marsh bird species on less than 5 of 91 (5%) survey routes per species per year on average throughout the Great Lakes-St. Lawrence lowlands between 1996 and 2013 (Pardieck et al., 2014). Such low numbers of annual detections require N30 years of data to achieve adequate statistical power for estimating linear population trends (Steidl et al., 2013). This is not effective for conservation and management because declines may be more subtle or rapid, and identification of problems needed more quickly. To remedy the situation, Bird Studies Canada launched the Great Lakes Marsh Monitoring Program (GLMMP) in 1995, a volunteer citizen science program (Dickinson et al., 2010), which has operated every year since (Bird Studies Canada, 2014). To date, over 1500 trained ⁎ Corresponding author. Tel.: +1 519 586 3531. E-mail address: [email protected].
participants have performed ~28,000 call broadcast surveys at targeted wetlands, resulting in detections of secretive marsh birds at greater than 50 of 232 (21%) survey routes per species per year on average in the Great Lakes basin, many times more detections than achieved by the BBS in the same region. Using standardized broadcasts of secretive species, visiting survey points at least two times per season, and surveying only in suitable emergent marsh habitat ensures that large sample sizes are achieved by the GLMMP each year for estimating population trends (Tozer, 2013). With the detection rates achieved by the GLMMP, it requires b15 years, or about half the time compared to the BBS, to achieve adequate statistical power for estimating population trends of most secretive marsh bird species in the Great Lakes basin (Steidl et al., 2013). However, in order to garner large numbers of volunteer participants to achieve such large sample sizes, the program has largely allowed participants to select sampling sites, which makes drawing inferences to larger un-sampled sets of sites less robust. Nonetheless, if one assumes that the GLMMP sites are approximately representative of wetlands in the region, the high precision of trends from the GLMMP make it the most suitable long-term dataset for guiding conservation of secretive marsh birds in the Great Lakes basin. Previous analyses using the GLMMP dataset have identified numerous marsh bird population declines (e.g., Timmermans et al., 2008; Tozer, 2013, 2014). In a recent example, populations of 10 of 19 (53%) marsh breeding bird species declined between 1995 and 2012 across
http://dx.doi.org/10.1016/j.jglr.2015.10.015 0380-1330/© 2015 International Association for Great Lakes Research. Published by Elsevier B.V. All rights reserved.
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
2
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
the Great Lakes basin, including bitterns (e.g., Botaurus), shallow(e.g., Porzana) and deep-water rails (e.g., Gallinula), and marsh-nesting terns (e.g., Chlidonias; Tozer, 2013). One study suggested that marsh bird trends were more negative within than outside Great Lakes International Joint Commission (IJC) areas of concern (Tozer, 2013), and another reported that declining marsh bird trends were steeper in coastal marshes in Lake Erie, Lake Huron, and Lake Michigan than in Lake Ontario (Timmermans et al., 2008). However, none of the previous studies above adjusted for differences in detection probability, addressed spatial autocorrelation, or explored influences of habitat covariates on occupancy dynamics. In this study, I report extended and updated annual occupancy estimates and trends of marsh-breeding birds in the southern portion of the Great Lakes basin based on GLMMP data. But I do so while addressing each of the three important items left unattended by the earlier studies listed above, which adds tremendous value to the results for conservation as elaborated on below. The probability of detecting marsh birds during surveys varies with a number of factors including time of day, date, and weather (Conway and Gibbs, 2011). As a result, the probability of detecting marsh birds during surveys is almost never perfect (Mackenzie et al., 2003). Birds can be unavailable for detection for numerous reasons including being absent, or being present but never being visible or audible to the observer (Dénes et al., 2015). Birds can also be missed because observers fail to see or hear them, even though they were visible or audible, respectively (Dénes et al., 2015). Thus, detection probability must be accounted for in a way such as I accomplish in this study because if it varies systematically over time, unadjusted occupancy estimates may show trends where none exist, or fail to show trends where they do exist (Kéry and Schmidt, 2008). Marsh bird survey points tend to be clustered at various scales, such as within wetlands or within wetland complexes (e.g., Conway, 2011). This often results in responses being more similar among clustered compared to unclustered points simply because they are closer together in space (Dormann et al., 2007). Referred to as spatial autocorrelation, this common feature of survey data violates independence assumptions required by statistical modeling (Dormann et al., 2007). Thus, spatial autocorrelation must be accounted for in a way such as I accomplish here because if certain spatially clustered points are assumed to be independent they may unduly influence parameter estimates and associated confidence limits and lead to unfounded recommendations for conservation. Long-term marsh bird survey data provide important information on trends over time (e.g., Tozer, 2014). However, these data also harbor valuable information on habitat features associated with high colonization and low extinction (MacKenzie et al., 2003). This information then translates into conservation actions that will help promote colonization or reduce extinction, respectively, and in turn help slow or maybe even reverse declining trends. Thus, one should explore occupancy dynamics in a way such as accomplished here, particularly for species showing significantly decreasing occupancy over time (hereafter “decreasing species”) versus those that do not (hereafter “increasing or stable species”), because it ultimately leads to more informed beneficial conservation actions for decreasing species. My ultimate goal was to determine if occupancy for each of 15 marsh bird breeding species changed throughout the southern portion of the Great Lakes basin over the past two decades, and to generate conservation actions that will benefit decreasing species. In doing so, I was careful to account for differences in detection probability, address spatial autocorrelation, and assess whether initial occupancy and subsequent colonization or extinction at a site within and across species was influenced by site, wetland, and landscape scale covariates. Pursuing these outcomes seemed especially justified within the southern portion of the Great Lakes basin because intensive human land use and associated environmental stress has, and continues to have, a strong influence on remaining wetlands, in an area where up to 90% of wetlands have been converted to other land uses (e.g., Ducks Unlimited Canada,
2010; Wolter et al., 2006). I accomplished my goal using multi-season site occupancy models and an information theoretic approach to simultaneously evaluate variables potentially related to detection and occupancy dynamics using a large dataset consisting of ~ 21,000 GLMMP point counts conducted at ~2,100 sites over 18 years. Methods Study design The GLMMP expanded throughout the Great Lakes basin in 1995, although coverage in the initial year and in the northern portion of the basin was never as extensive as subsequent years and in the southern portion of the basin (Fig. 1). I included only data from 1996 onward from the southern portion of the Great Lakes basin because, from that point and onward, there was reasonably good geographic coverage for birds throughout the area (~588 sites surveyed across ~181 routes per year on average; Fig. 2). I defined the southern portion of the Great Lakes basin as the part of the basin within Bird Conservation Regions 13, 14, 22–24, and 28 (Bird Studies Canada and NABCI, 2014). GLMMP participants selected survey marshes or they were randomly assigned. Participants conducted surveys within 1–8 semi-circular 100 m-radius plots (hereafter “sites”) along survey routes within one marsh or across multiple marshes. Most sites (90%) were located on shore adjacent to marshes that were covered mostly by non-woody plants. Sites were N250 m apart to avoid double-counting individuals. Bird surveys Each site was surveyed for 10 min on 2–3 occasions at least 10 days apart between 20 May and 5 July in each year that the survey route was active. Surveys occurred in either the morning (30 min before local sunrise to 1000 local time) or evening (4 h before local sunset to dark) and only under weather conditions that were favorable for detecting all species and individuals present (no precipitation; wind: Beaufort 0–3, 0– 19 km/h, recorded during each survey). Participants broadcasted calls during surveys to entice individuals of especially secretive species to reveal their presence by approaching or responding vocally. The calls occurred in the first 5 min of each 10-min survey and consisted of 30 s of vocalizations followed by 30 s of silence for each of least bittern (Ixobrychus exilis), sora (Porzana carolina), Virginia rail (Rallus limicola), a mixture of American coot (Fulica americana) and common gallinule (Gallinula galeata), and pied-billed grebe (Podilymbus podiceps), in that order. The GLMMP bird survey protocol is described in more detail in Bird Studies Canada (2009a). For each survey, I assigned day-of-year (e.g., 20 May = 140; hereafter “date”), morning versus evening (defined above; hereafter “time of day”), and a Beaufort wind scale value (hereafter “wind”). Site covariates Visual estimates of percent areal coverage of major habitat and vegetation types were made annually within 100-m-radius semi-circular plots at each site. Multi-season site occupancy models can, in theory, incorporate these annual estimates directly into models, but as I describe under Analysis below, attempts to include these “yearly site covariates” (sensu Fiske et al., 2015) resulted in models that failed to converge. Instead, I took the first year value for each site to get coverage of open water (hereafter “% water”), robust emergent herbaceous vegetation (e.g., cattails [Typha], hereafter “% emergent”), common reed (Phragmites sp., hereafter “% Phragmites”), purple loosestrife (Lythrum salicaria; hereafter “% loosestrife”), and trees and shrubs (e.g., willows [Salix], hereafter “% trees and shrubs”), and I took the difference between the last-year value and the first-year value at each site to get overall change in coverage of open water (hereafter “change in % water”), robust emergent herbaceous vegetation (hereafter “change in
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
3
Fig. 1. Routes surveyed for birds by participants in Bird Studies Canada's Great Lakes Marsh Monitoring Program (GLMMP) throughout the Great Lakes basin between 1996 and 2013. In this study, I used data from only the southern portion of the basin, south of the Canadian Shield (solid gray; Great Lakes basin – south) to estimate detection probability and occupancy dynamics of marsh bird species because sampling and land use were more uniform throughout this region than to the north (hatched; Great Lakes basin – north). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.).
% emergent”), common reed (hereafter “change in % Phragmites”), purple loosestrife (hereafter “change in % loosestrife”), and trees and shrubs (hereafter “change in % trees and shrubs”). I included these latter
change measures as “site covariates” (sensu Fiske et al., 2015) in the colonization and extinction processes during modeling. The GLMMP habitat survey protocol is described in more detail in Bird Studies Canada (2009b).
Wetland and landscape covariates
Fig. 2. Number of sites or routes surveyed for birds per year (upper panel) and number of sites as a function of the number of years surveyed (lower panel) by participants in Bird Studies Canada's Great Lakes Marsh Monitoring Program throughout the southern portion of the Great Lakes basin between 1996 and 2013. Numbers above bars show the value of each bar as a percent of the total number of sites surveyed.
Previous studies show that marsh bird occupancy or abundance is influenced by land cover characteristics within a distance of at least 3 km from the wetland (e.g., Tozer et al., 2010), so I used land cover or wetland inventory data and the tools in ArcGIS 10.2 (ESRI, 2013) to delineate the proportion of wetland, agriculture, and urban land cover within 3 km of the sample point (hereafter “% wetland”, “% agriculture”, and “% urban”, respectively). I omitted from these calculations portions of 3-km buffers that were part of one of the Great Lakes. Percent wetland was characterized in Ontario using the Consolidated Wetland Layer (Ontario Ministry of Natural Resources) and in the US using the National Wetland Inventory (US Fish and Wildlife Service). All wetland types except open water (Ontario) and lakes and riverine (US) were included in % wetland. I included all wetland types partly because the Ontario data had polygons with unknown wetland types, so restricting the analysis to only emergent wetland polygons might have introduced unknown biases. I also determined the area of each wetland containing each site using these same wetland layers (in ha); and used the loge of wetland area to capture the effect of this covariate on marsh birds (e.g., Tozer et al., 2010; hereafter “log wetland area”). Percent agriculture and % urban were characterized in Ontario using the Ontario Land Cover Compilation (Ontario Ministry of Natural Resources) and in the US using the National Land Cover Database (United States Geological Survey). Percent agriculture in Ontario consisted of pasture and abandoned fields and cropland, whereas in the U.S. it consisted of pasture/ hay and cultivated crops. Percent urban in Ontario consisted of settlement and developed land, whereas in the U.S. it consisted of developed
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
4
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
open space and developed low, medium, and high intensity. Each site was also categorized as being within or outside the watershed of a Great Lakes IJC Area of Concern (hereafter “AOC”; IJC, 2012) and along a Great Lakes coast (hereafter “coastal”) or inland. Sites were defined as coastal based on a hydrogeomorphic classification system (Albert et al., 2005; Great Lakes Coastal Wetlands Consortium, 2004). Analyses The response variable was the occurrence within sites during individual surveys of each of 15 breeding marsh bird species. I used multiseason site occupancy models for each species (MacKenzie et al., 2003) and an information theoretic approach via Akaike's information criterion (AIC) to select best approximating models from among sets of candidate models (Burnham and Anderson, 2002). Multi-season site occupancy models consist conceptually of two coupled hierarchical logistic regressions, the first describing actual initial occupancy at each site in the first year and subsequent colonization and extinction in each of the following years, and the second describing observed occupancy during each survey, conditional on actual occupancy. When the two processes are combined and parameterized via maximum likelihood they jointly describe occupancy dynamics, i.e., probability of initial occupancy in 1996, probability of subsequent colonization, and probability of subsequent extinction, while accounting for uncertainties in detection (MacKenzie et al., 2003). At each of several stages of model selection described in more detail below, I selected the top ranked model, i.e., ΔAIC = 0, as the best model and passed it along to the next stage, except in cases where a lower ranked model existed within 2 AIC units of the top ranked model, i.e., ΔAIC ≤ 2, which included all of the covariates found within all of the higher ranked models. In these latter instances I selected the lower ranked but more complex model as the best model to avoid discarding important variables. I tested the overall fit of global models using a parametric bootstrap procedure and associated Χ2 statistic to ensure that covariates as a group reliably explained the response variable; P-values N 0.05 indicated adequate fit (Fiske and Chandler, 2015). I selected for analysis 21,546 point counts where covariate information was complete, discarding point counts with missing covariate data as recommended by Kéry and Chandler (2012). I conducted my analysis for each species in five stages to make the number of models manageable. First, I considered the influence of all single-variable (3 models) and all two-variable additive combinations (3 models) of date, time of day, and wind on detection probability, along with an intercept-only and global model, which included all of the covariates, a total of 8 models per species. I did not consider cloud cover or ambient temperature because most past studies have reported equivocal effects of cloud cover and no effect of ambient temperature on vocalization probability in marsh birds (Conway and Gibbs, 2011). I included the covariate or covariates from the best detection probability model in the detection process of all subsequent models for each species described below. Given that my main goal did not involve estimating detection probability or what influences it, I do not report or discuss this part of the analysis in any detail. Second, I examined the influence of the following 11 covariates on the probability of initial occupancy in 1996: % water, % emergent, % Phragmites, % loosestrife, % trees and shrubs, loge wetland area, % wetland, % agriculture, % urban, AOC, and coastal. I considered all singlevariable models (11 models) and additive combinations of only sitescale (% water, % emergent, % Phragmites, % loosestrife, and % trees and shrubs), wetland-scale (log wetland area, AOC, coastal), and landscape-scale (% wetland, % agriculture, and % urban) covariates (3 models), along with an intercept-only and global model, which included all of the covariates, a total of 16 models per species. I included the covariate or covariates from the best initial occupancy model in the initial occupancy process of all subsequent models for each species described below.
Third, I examined the influence of the following 11 covariates on the probability of subsequent colonization: change in % water, change in % emergent, change in % Phragmites, change in % loosestrife, change in % trees and shrubs, log wetland area, % wetland, % agriculture, % urban, AOC, and coastal. I considered all single-variable models (11 models) and additive combinations of site-scale (change in % water, change in % emergent, change in % Phragmites, change in % loosestrife, and change in % trees and shrubs), wetland-scale (log wetland area, AOC, coastal), and landscape-scale (% wetland, % agriculture, and % urban) covariates (3 models), along with an intercept-only and global model, which included all of the covariates, a total of 16 models per species. I treated all 11 of the covariates as “site covariates” (sensu Fiske et al., 2015). It is worth noting that information from the annual site-scale vegetation covariates could have been treated as “yearly site covariates” (sensu Fiske et al., 2015), but initial attempts to do so resulted in models that failed to converge, which is why I calculated the change between the last survey and the first survey at each site and treated the result as a site covariate. This is justified because the model estimates the average annual probability of colonization for a given level of change (positive or negative) in each covariate, and even though one estimate of change across all of the years in each time-series imposes a temporal smoothing on the resulting estimates of occupancy, it is still useful for assessing the influence of covariates on colonization, and similarly for extinction described below. I included the covariate or covariates from the best colonization model in the colonization process of all subsequent models for each species described below. Fourth, I examined the influence on the probability of subsequent extinction of the same 11 covariates listed for colonization above, via a similar candidate model set described for colonization above. I included the covariate or covariates from the best extinction model in the extinction process of all subsequent models for each species described below. Fifth, to account for spatial autocorrelation, I compared a model that included latitude, longitude, latitude2, longitude2, and latitude × longitude in the occupancy, colonization, and extinction processes to a model that did not include any latitude or longitude terms (e.g., Schuster and Arcese, 2013). For all species, the model containing the latitude and longitude terms fit much better than the model that did not contain these terms. So I included latitude, longitude, latitude2, longitude2, and latitude × longitude in the occupancy, colonization, and extinction processes of all final best models for each species, and I do not report any further details for these models or any further details for the latitude or longitude terms in each model. I note that I could have accounted for spatial autocorrelation earlier in the modeling process, but doing so would have assumed that any spatial component in the results was error, regardless of whether the spatial component was the result of a functional relationship between a spatially arranged covariate and the response, and in turn might have unduly discarded covariates useful for inference (e.g., Dormann, 2007). To assess trends in occupancy over time, I plotted annual occupancy estimates from final best models with the addition of a categorical effect of year in each of the detection, colonization, and extinction processes, i.e., a fully time dependent model as described by Weir et al. (2009) and Kéry and Chandler (2012). Annual occupancy was the proportion of sampled sites that was occupied in each year, rather than the occupancy rate of the entire population of hypothetical sites from which the sampled sites were selected; see Weir et al. (2009) for further details, where such estimates are referred to as “smoothed trajectories”. Confidence intervals for annual occupancy estimates were generated via nonparametric bootstrapping as described in Kéry and Chandler (2012). I estimated trends in occupancy over time as the geometric mean rate of change between annual occupancy in 1996 and in 2013 as described in equation 4 of Smith et al. (2014), and assessed the statistical significance of each trend via parametric bootstrapping. I accomplished the latter by writing an R-function that calculated the trend in occupancy (i.e., the geometric mean rate of change between annual occupancy in 1996 and in 2013), then I fed the R-function into the
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
“parboot” function in the unmarked package, which derived an estimate of the 95% confidence limits around the occupancy trend estimate (see below for software details; Kéry and Chandler, 2012). Bootstrapping in this manner was necessary to account for the varying precision of the annual estimates used to calculate each occupancy trend. Occupancy trend estimates with 95% confidence intervals that did not overlap zero were considered statistically significant. To compare parameter estimates of covariates within and across species for initial occupancy, colonization, and extinction, I plotted on the logit scale standardized parameter estimates and associated 95% confidence limits from final best models for each species. To compare relative importance of covariates across decreasing species and across increasing or stable species for initial occupancy, colonization, and extinction, I plotted for each covariate for each group of species the mean and associated 95% confidence limits of the species-level sums of Akaike weights from all models that contained the covariate of interest. To compare parameter estimates of covariates across decreasing species or across increasing or stable species for initial occupancy, colonization, and extinction, I plotted on the logit scale for each group of species the mean and associated 95% confidence limits of standardized species-level coefficients from final best models. I then used the plots of mean relative importance and mean coefficients to compare and contrast occupancy dynamics between decreasing versus increasing or stable species. I was especially interested in identifying habitat features associated with higher mean initial occupancy, higher mean colonization, or lower mean extinction across decreasing species, identified by 95% confidence intervals around means that did not overlap zero, because actions to increase such features will likely maximize conservation benefits. All analyses were performed using the R programming language (R Development core Team, 2014). To facilitate model convergence and estimation and comparison of parameter estimates across variables and species, I standardized all of the explanatory variables prior to analysis by subtracting the mean and dividing by the standard deviation (Kéry and Chandler, 2012). Models were fitted using package unmarked (Fiske et al., 2015) and AIC tables were generated using package AICcmodavg (Mazerolle, 2013). Results The dataset consisted of 21,546 point counts conducted at 2149 sites along 706 routes over 18 years between 1996 and 2013 (Figs. 1 and 2). The number of routes surveyed per year (181 ± 52 [mean ± SD]) ranged from 114 to 255 (Fig. 2). Correspondingly, the number of sites surveyed per year (588 ± 136) ranged from 410 to 769 and increased between 1996 and 2013 (linear regression: F1,16 = 23.9, P = 0.0002; Fig. 2). However, there was only low to moderate correlation between annual occupancy for each species and the number of sites surveyed per year (Pearson |r| across all species: mean = 0.3, range: 0.1 to 0.6), which suggested that the number of sites surveyed per year was not a confounding factor. Most (76%) sites were surveyed for 2 or more years, with 52% surveyed for 2–6 years and 24% surveyed for 7–18 years (Fig. 2). The number of sites as a function of covariates is given in Fig. 3. Global models for each species fit adequately (all P N 0.2), which suggested that the site, wetland, and landscape scale covariates as a group adequately explained occupancy dynamics for each species. However, confidence limits for initial occupancy in 1996 for pied-billed grebe were inestimable for % loosestrife, coastal, and % agriculture; parameter estimates and associated confidence limits were unreasonable, i.e., many times larger and wider, respectively, compared to all of the other estimates for all of the other species, or confidence limits were inestimable for colonization for American coot, common gallinule, and mute swan; and parameter estimates and associated confidence limits were unreasonable for extinction for American coot, common gallinule, least bittern, and mute swan. Thus, I do not report parameter estimates
5
or confidence limits in these instances, nor do I consider these covariates in subsequent analyses below. Model selection details for the detection, initial occupancy, colonization, and extinction possesses are given in Electronic Supplementary Material (ESM) Tables S1, S2, S3, and S4, respectively. Occupancy significantly decreased between 1996 and 2013 throughout the southern portion of the Great Lakes basin for 9 of 15 (60%) species, whereas occupancy significantly increased for 1 (7%) species and remained statistically unchanged for 5 (33%) species (Fig. 4). Across all species with significant trends, occupancy of black tern (Chlidonias niger) decreased the most, followed by American coot, common gallinule, sora, pied-billed grebe, American bittern (Botaurus lentiginosus), Virginia rail, least bittern, and Canada goose (Branta canadensis); whereas occupancy of common yellowthroat (Geothlypis trichas) showed the only significant increase (Fig. 4). Common grackle (Quiscalus quiscula), marsh wren (Cistothorus palustris), mute swan (Cygnus olor), red-winged blackbird (Agelaius phoeniceus), and swamp sparrow (Melospiza georgiana) showed no statistically significant changes over time (Fig. 4). Notably, occupancy of 100% of 7 especially secretive marsh-dependent species significantly declined during the study: American bittern, American coot, common gallinule, least bittern, pied-billed grebe, sora, and Virginia rail (Fig. 4). The relative importance, magnitude, and direction of covariates and their associated parameter estimates for explaining the initial occupancy, colonization, or extinction process varied greatly among species (ESM Tables S1, S2, S3; Figs. S1, S2, S3). As such, there were few consistent patterns across species, with the following exceptions. The mean relative importance of wetland scale covariates, i.e., AOC, coastal, and log wetland area, for the colonization and extinction processes across decreasing species and across increasing or stable species was higher compared to each of the rest of the covariates, although 95% confidence intervals overlapped among all of the covariates (Fig. 5). According to mean standardized parameter estimates from final best models for decreasing species, initial occupancy in 1996 decreased with increasing % loosestrife and % trees and shrubs, but increased with increasing % open and log wetland area; colonization decreased with increasing % urban, but increased with increasing log wetland area; and extinction was higher within AOCs than outside AOCs (Fig. 5). By contrast, according to mean standardized parameter estimates from final best models for increasing or stable species, initial occupancy in 1996 was higher at coastal compared to inland sites; colonization increased with increasing change in % trees and shrubs and was higher within AOCs than outside AOCs; and extinction was higher within AOCs than outside AOCs (Fig. 5). Notable species-level relationships for decreasing species included the following. Initial occupancy in 1996 decreased with increasing % Phragmites for Virginia rail, decreased with increasing % urban for American coot, black tern, common gallinule, and pied-billed grebe, and was lower at coastal compared to inland sites for American bittern; but increased with increasing % emergent for common gallinule and Virginia rail, increased with increasing % wetland for American bittern, American coot, black tern, and common gallinule, and was higher at coastal compared to inland sites for black tern and common gallinule (Fig. S1). Colonization decreased with increasing change in % emergent and change in % open for pied-billed grebe, decreased with increasing change in % loosestrife for sora, decreased with increasing change in % trees and shrubs for pied-billed grebe, decreased with increasing % urban and % wetland for Virginia rail, was lower within AOCs compared to outside for pied-billed grebe and sora, and was lower at coastal compared to inland sites for black tern and Virginia rail; but was higher at coastal compared to inland sites for pied-billed grebe (Fig. S2). Extinction decreased with increasing change in % emergent, change in % open, and change in % trees and shrubs for pied-billed grebe, decreased with increasing log wetland area for American bittern and sora, and decreased with increasing % agriculture, % urban, and % wetland for Virginia rail; but increased with increasing change in % loosestrife for Virginia rail, increased with increasing % Phragmites for pied-billed grebe, increased
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
6
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
Fig. 3. Number of sites as a function of covariates used to estimate occupancy dynamics of 15 breeding marsh bird species throughout the southern portion of the Great Lakes basin between 1996 and 2013. Variables are defined in the methods. Q = quartile.
with increasing log wetland area for Virginia rail, and was higher at coastal compared to inland sites for Virginia rail (Fig. S3). Discussion Occupancy of 60% of the marsh breeding bird species considered in this study significantly decreased across the southern portion of the Great Lakes basin between 1996 and 2013. Nearly all of the decreasing species nest and forage, almost exclusively, in emergent marsh habitat.
By contrast, all of the increasing or stable species, with the exception of marsh wren and swamp sparrow, frequently nest and forage in habitats other than emergent marsh, such as overgrown fields and open-water ponds nearly devoid of emergent vegetation. For increasing or stable species with more general habitat requirements, emergent marsh breeding habitat may have been less important for maintaining occupancy. Given that nearly all of the decreasing species were marshdependent may suggest that the quality of emergent marsh breeding bird habitat declined throughout, or in part of, the southern portion of
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
7
Fig. 4. Temporal trends in occupancy of 15 breeding marsh bird species throughout the southern portion of the Great Lakes basin between 1996 and 2013. Shown are annual occupancy estimates (circles) with associated 95% confidence limits (solid lines) from final best models. Naive annual occupancy estimates unadjusted for differences in detection probability or spatial autocorrelation are also shown (dashed lines). See methods for details. Trends in occupancy (%/year) and associated 95% confidence limits (lower, upper) are shown for each species; species with statistically significant trends are shown in bold. Note that the vertical axis differs among species, but always has an extent of 50.
the Great Lakes basin during the study. However, it cannot be ruled out that these patterns were instead, or in part, caused by mechanisms operating beyond the southern portion of the Great Lakes basin, such as mechanisms operating in migratory staging areas or on the wintering grounds or both (Hostetler et al., 2015). Nonetheless, by linking colonization and extinction to site, wetland, and landscape scale covariates, I was able to identify conservation actions that will likely benefit decreasing species within the southern portion of the Great Lakes basin. The tremendous variation in occupancy dynamics as a function of covariates within and among species in this study makes generating conservation actions to benefit most decreasing species especially challenging. However, the results suggest that wetland scale covariates were the most important overall drivers of colonization and extinction across decreasing species, with mean probability of colonization increasing with increasing log wetland area and mean probability of extinction being higher within AOCs compared to outside AOCs (Fig. 5).
By contrast, site and landscape scale covariates were less important drivers of colonization and extinction, with only mean probability of colonization decreasing with increasing % urban (Fig. 5). These general results suggest that the largest number of decreasing marshdependent breeding bird species will benefit from conserving, restoring, or creating large wetlands surrounded by limited urban land use, and from addressing issues within AOCs. Such actions will promote colonization and reduce extinction of the largest number of decreasing species, which will potentially help slow or maybe even reverse the declining trends in occupancy identified in this study. The above results and associated conservation actions are not surprising considering that similar patterns have been found before amongst at least some of the decreasing species. For instance, occupancy or abundance of at least some of the decreasing species increased with increasing wetland area in Iowa Prairie marshes (Brown and Dinsmore, 1986), Great Lakes coastal wet meadows (Riffell et al.,
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
8
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
Fig. 5. Mean relative importance (mean weight) and mean standardized coefficients (mean coefficient) and associated 95% confidence intervals of site, wetland, and landscape scale covariates for explaining initial occupancy, colonization, and extinction of decreasing species and increasing or stable species throughout the southern portion of the Great Lakes basin between 1996 and 2013. Mean weight calculated as the mean of the species-level sums of Akaike weights from all models that contained the covariate of interest. Mean coefficient calculated as the mean of standardized species-level coefficients from final best models. See text for details. chg = change; emerg = emergent; l-strife = loosestrife; Phrag = Phragmites; t + s = trees and shrubs; wet = wetland.
2001), and cattail-dominated wetlands in southern Ontario (Tozer et al., 2010); and decreased with increasing amount of surrounding urban land use in Lake Erie and Lake Ontario coastal wetlands (Smith and Chow-Fraser, 2010). Furthermore, abundance was lower, or time trends in abundance were more negative, within AOCs than outside AOCs (Tozer, 2013). But these previously known patterns and associated conservation actions have never before been supported by such extensive long-term data from throughout the southern Great Lakes region as is the case here. The conservation actions outlined above apply to the largest number of decreasing species. However, the results of the study suggest a number of additional actions that will benefit individual or smaller numbers of decreasing species. In fact, all but one (% agriculture) of the covariates were associated with lower probability of colonization or higher probability of extinction of at least one or more decreasing species (Figs. S1, S2, S3). It follows that each of these patterns forms the basis for a beneficial conservation action. Thus, along with the more general actions
listed above, the results suggest that it is also important to conserve, restore, or create robust-emergent-dominated but interspersed, purple loosestrife-free, Phragmites-free wetlands surrounded by higher proportions of wetland cover in the surrounding landscape, and to address issues within Great Lakes coastal wetlands. As above, these additional results and associated actions are, for the most part, not surprising considering that similar patterns have been found before amongst at least some of the decreasing species for most of the covariates. For instance, abundance of nearly half of the decreasing species found in this study increased with increasing open water-emergent vegetation interspersion in New York (Rehm and Baldassarre, 2007), and in at least one of the years of another study in the Midwestern US (Bolenbaugh et al., 2011). Pooled abundance of marsh-nesting birds decreased slightly with increasing Phragmites cover in Great Lakes coastal marshes at Long Point, Ontario (Meyer et al., 2010). Occurrence or density increased with increasing amount of wetland in the surrounding landscape for least bittern in Québec
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
(Jobin et al., 2011), least bittern and Virginia rail in eastern Ontario (Quesnelle et al., 2013), and various other deceasing species in Iowa (Fairbairn and Dinsmore, 2001). Abundance of some decreasing species was lower at Great Lakes coastal compared to inland sites (Tozer, 2013). But like the patterns identified above, these previously known patterns and associated conservation actions have never before been supported by such extensive long-term data as is the case here. By contrast, the results for colonization and extinction for % loosestrife from this study appear to be novel, seeing as a recent review of peer-reviewed literature on the topic failed to find any reference to the relationship between occupancy dynamics and the presence of purple loosestrife for any of the decreasing species (Lavoie, 2010). Each of the conservation actions described above is based on colonization and extinction dynamics. Focusing on these processes makes sense because they are most directly linked to changes in occupancy during the period of the study; and are, therefore, most useful for generating beneficial conservation actions. However, it is worth noting that for decreasing species mean probability of initial occupancy in 1996 decreased with increasing % loosestrife and % trees and shrubs, but increased with increasing % open and log wetland area (Fig. 5). Even though these patterns were presumably largely influenced by mechanisms operating prior to the period of study, they still likely provide useful information for generating beneficial conservation actions because they describe conditions that will promote occupancy. Thus, it may be justified that conserving, restoring, or creating robust-emergent-dominated but interspersed, purple loosestrife-free wetlands should also appear in the list of general conservation actions to benefit the largest number of decreasing species given above, rather than appearing only in the list to benefit individual or smaller groups of decreasing species. I used data collected over many years by a large network of citizen scientists to assess whether occupancy of breeding marsh bird species increased or decreased throughout the southern portion of the Great Lakes basin, and whether occupancy dynamics at a site within and across species was influenced by site, wetland, and landscape scale covariates. Conducting over 21,500 point counts across 18 years without a program such as Bird Studies Canada's GLMMP would not have been possible, which highlights the utility of using citizen science programs to monitor wildlife. The results suggest that the largest number of decreasing marsh-dependent breeding bird species will benefit from conserving, restoring, or creating large wetlands surrounded by limited urban land use, and from addressing issues within AOCs. In addition, individual or smaller groups of decreasing species will also benefit from conserving, restoring, or creating robust-emergent-dominated but interspersed, purple loosestrife-free, Phragmites-free wetlands surrounded by higher proportions of wetland cover in the surrounding landscape, and from addressing issues within Great Lakes coastal wetlands. These actions will help promote colonization or reduce extinction and help slow or maybe even reverse declining trends in occupancy among decreasing species across the southern portion of the Great Lakes basin. Acknowledgments I am grateful for the dedication of hundreds of GLMMP participants, whose collective effort made the study possible. I am also grateful for the coordination and promotion of the GLMMP by Kathy Jones and a large team of regional coordinators throughout the Great Lakes basin; and for past coordination of the program by Russ Weeber, Steve Timmermans, and Ryan Archer. Adam C. Smith, an anonymous reviewer, Myles Falconer, Tara Crewe, Becky Stewart, and Karla Falk provided helpful comments and statistical advice that improved the paper. Bird Studies Canada is thankful for ongoing or recent support of the GLMMP by Environment Canada, United States Environmental Protection Agency, Government of Ontario, Great Lakes Restoration Initiative, Eastern Habitat Joint Venture, Wildlife Habitat Canada, Ducks Unlimited Canada, TD Friends of the Environment Foundation, and U.S. Fish and Wildlife Service; and for past support by Brant Waterways Foundation,
9
Great Lakes Commission, Great Lakes Sustainability Fund, Great Lakes United, National Audubon Society, Ontario Trillium Foundation, Wetland Habitat Fund, and Great Lakes Protection Fund. Appendix A. Supplementary data Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.jglr.2015.10.015. References Albert, D.A., Wilcox, D.A., Ingram, J.W., Thompson, T.A., 2005. Hydrogeomorphic classification for Great Lakes coastal wetlands. J. Great Lakes Res. 31, 129–146. Bird Studies Canada, 2009a. Marsh Monitoring Program Participant's Handbook Getting Started. Bird Studies Canada in Cooperation with Environment Canada and the. U.S. Environmental Protection Agency, Port Rowan, ON. Bird Studies Canada, 2009b. Marsh Monitoring Program Participant's Handbook for Surveying Marsh Birds. Bird Studies Canada in Cooperation with Environment Canada and the. U.S. Environmental Protection Agency, Port Rowan, ON. Bird Studies Canada, 2014. The Great Lakes Marsh Monitoring Program. Bird Studies Canada, Port Rowan, ON (http,//www.birdscanada.org/volunteer/glmmp/index.jsp? lang = EN&targetpg = index Accessed 5 June 2015). Bird Studies Canada, NABCI, 2014. Bird Conservation Regions. (Published by Bird Studies Canada on behalf of the North American Bird Conservation Initiative. http://www. birdscanada.org/research/gislab/index.jsp?targetpg = bcr Accessed 5 June 2015). Bolenbaugh, J.R., Krementz, D.G., Lehnen, S.E., 2011. Secretive marsh bird species cooccurrences and habitat associations across the Midwest, USA. J. Fish Wildl. Manag. 2, 49–60. Brown, M., Dinsmore, J.J., 1986. Implications of marsh size and isolation for marsh bird management. J. Wildl. Manag. 50, 392–397. Burnham, K.P., Anderson, D.R., 2002. Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach. Springer-Verlag, New York, NY. Conway, C.J., 2011. Standardized North American marsh bird monitoring protocol. Waterbirds 34, 319–346. Conway, C.J., Gibbs, J.P., 2011. Summary of intrinsic and extrinsic factors affecting detection probability of marsh birds. Wetlands 31, 403–411. Dénes, F.V., Silveira, L.F., Beissinger, S.R., 2015. Estimating abundance of unmarked animal populations: accounting for imperfect detection and other sources of zero inflation. Methods Ecol. Evol. 6, 543–556. Development Core Team, R., 2014. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria 3-900051-07-0 (http://www.R-project.org/ Accessed 5 June 2015). Dickinson, J.L., Zukerberg, B., Bonter, D.N., 2010. Citizen science as an ecological research tool, challenges and benefits. Annu. Rev. Ecol. Evol. Syst. 41, 149–172. Dormann, C.F., 2007. Assessing the validity of autologistic regression. Ecol. Monit. 207, 234–242. Dormann, C.F., McPherson, J.M., Araújo, M.B., Bivand, R., Bolligar, J., Carl, G., Davies, R.G., Hirzel, A., Jetz, W., Kissling, D., Kühn, I., Ohlemüller, R., Peres-Neto, P.R., Reineking, B., Schrӧder, B., Schurr, F.M., Wilson, R., 2007. Methods to account for spatial autocorrelation in the analysis of species distributional data: a review. Ecography 30, 609–628. Ducks Unlimited Canada, 2010. Southern Ontario wetland conversion analysis. Final report, March 2010. Ducks Unlimited Canada, Barrie, ON ([online] URL: http://www. ducks.ca/assets/2010/10/duc_ontariowca_optimized.pdf Accessed 5 June 2015). Environment Canada, 2010. North American Breeding Bird Survey — Canadian Results and Analysis Website version 3.00. Environment Canada, Gatineau, QC (http,// www.ec.gc.ca/reom-mbs/default.asp?lang = En&n = 0D74F35F-1 Accessed 5 June 2015). ESRI, 2013. ArcGIS 10.2. Environmental Systems Resource Institute, Redland, CA. Fairbairn, S.E., Dinsmore, J.J., 2001. Local and landscape-level influences on wetland bird communities of the prairie pothole region of Iowa, USA. Wetlands 21, 41–47. Fiske, I., Chandler, R., 2015. Overview of Unmarked: an R package for the analysis of data from unmarked animals. http://cran.r-project.org/web/packages/unmarked/ vignettes/unmarked.pdf (Accessed 5 June 2015). Fiske, I., Chandler, R., Miller, D., Royle, A., Kéry, M., 2015. Package unmarked. http://cran.rproject.org/web/packages/unmarked/unmarked.pdf (Accessed 5 June 2015). Great Lakes Coastal Wetlands Consortium, 2004. Great Lakes Coastal Wetlands Consortium: Inventory and Classification. http://glc.org/projects/habitat/coastal-wetlands/ cwc-inventory/ (Accessed 5 June 2015). Hostetler, J.A., Sillett, T.S., Marra, P.P., 2015. Full-annual-cycle population models for migratory birds. Auk 132, 433–449. IJC [International Joint Commission], 2012. Great Lakes Water Quality Agreement, Protocol Amending the Agreement between Canada and the United States of America on Great Lakes Water Quality, 1978, as Amended on October 16, 1983, and on November 18, 1987, Signed September 7, 2012, Entered into force February 12, 2013. International Joint Commission, Windsor, ON. Jobin, B., Fradette, P., Labrecque, S., 2011. Habitat use by Least Bitterns (Ixobrychus exilis) in Quebec. Waterbirds 34, 143–150. Kéry, M., Chandler, R., 2012. Dynamic Occupancy Models in Unmarked. Swiss Ornithological Institute and USGS Patuxent Wildlife Research Center (29 March 2012. http:// cran.r-project.org/web/packages/unmarked/vignettes/colext.pdf Accessed 5 June 2015).
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
10
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
Kéry, M., Schmidt, B.R., 2008. Imperfect detection and its consequences for monitoring for conservation. Community Ecol. 9, 207–216. Lavoie, C., 2010. Should we care about purple loosestrife? The history of an invasive plant species in North America. Biol. Invasions 12, 1967–1999. MacKenzie, D.I., Nichols, J.D., Hines, J.E., Knutson, M.G., Franklin, A.B., 2003. Estimating site occupancy, colonization, and local extinction when a species is detected imperfectly. Ecology 84, 2200–2207. Mazerolle, M.J., 2013. AICcmodavg: Model Selection and Multimodel Inference Based on (Q)AIC(c). Version 2.00. R Foundation for Statistical Computing, Vienna, Austria (http://CRAN.R-project.org/package = AICcmodavg Accessed 5 June 2015). Meyer, S.W., Badzinski, S.S., Pettrie, S.A., Ankney, C.D., 2010. Seasonal abundance and species richness of birds in common reed habitats in Lake Erie. J. Wildl. Manag. 74, 1559–1567. North American Bird Conservation Initiative Canada, 2012. The State of Canada's Birds, 2012. Environment Canada, Ottawa, ON (http,//www.stateofcanadasbirds.org/State_ of_Canada's_birds_2012.pdf Accessed 5 June 2015). Pardieck, K.L., Ziolkowski, D.J., Hudson, M.-A.R., 2014. North American Breeding Bird Survey Data Set 1966–2013, Version 2013.0. USGS Patuxent Wildlife Research Center, Maryland (www.pwrc.usgs.gov/BBS/RawData/ Accessed 5 June 2015). Quesnelle, P.E., Fahrig, L., Lindsay, K.E., 2013. Effects of habitat loss, habitat configuration and matrix composition on declining wetland species. Biol. Conserv. 160, 200–208. Rehm, E.M., Baldassarre, G.A., 2007. The influence of interspersion on marsh bird abundance in New York. Wilson J. Ornithol. 119, 648–654. Riffell, S.K., Keas, B.E., Burton, T.M., 2001. Area and habitat relationships of birds in Great Lakes coastal wet meadows. Wetlands 21, 492–507. Schuster, R., Arcese, P., 2013. Using bird species community occurrence to prioritize forests for old growth restoration. Ecography 36, 499–507.
Smith, L.A., Chow-Fraser, P., 2010. Impacts of adjacent land use and isolation on marsh bird communities. Environ. Manag. 45, 1040–1051. Smith, A.C., Hudson, M.-A., Downes, C., Francis, C.M., 2014. Estimating breeding bird survey trends and annual indices for Canada: how do the new hierarchical Bayesian estimates differ from previous estimates? Can. Field-Nat. 128, 119–134. Steidl, R.J., Conway, C.J., Litt, A.R., 2013. Power to detect trends in abundance of secretive marsh birds, effects of species traits and sampling effort. J. Wildl. Manag. 77, 445–453. Timmermans, S.T.A., Badzinski, S.S., Ingram, J.W., 2008. Associations between breeding marsh bird abundances and Great Lakes hydrology. J. Great Lakes Res. 34, 351–364. Tozer, D.C., 2013. The Great Lakes Marsh Monitoring Program 1995–2012, 18 Years of Surveying Birds and Frogs as Indicators of Ecosystem Health. Bird Studies Canada, Port Rowan, ON (http,//www.birdscanada.org/download/GLMMPreport.pdf Accessed 5 June 2015). Tozer, D.C., 2014. Coastal Wetland Birds. In State of the Great Lakes 2011. Environment Canada and U.S. Environmental Protection Agency (Cat No. En161-3/1-2011E-PDF. EPA 950-R-13-002. [Online] URL, http://binational.net/wp-content/uploads/2014/ 11/sogl-2011-technical-report-en.pdf Accessed 5 June 2015). Tozer, D.C., Nol, E., Abraham, K.F., 2010. Effects of local and landscape-scale habitat variables on abundance and reproductive success of wetland birds. Wetl. Ecol. Manag. 18, 679–693. Weir, L., Fiske, I.J., Royle, J.A., 2009. Trends in anuran occupancy from northeastern states of the North American amphibian monitoring program. Herpetol. Conserv. Biol. 4, 389–402. Wolter, P.T., Johnston, C.A., Niemi, G.J., 2006. Land use land cover change in the U.S. Great Lakes basin 1992 to 2001. J. J. Great Lakes Res. 32, 607–628.
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
Contents lists available at ScienceDirect
Journal of Great Lakes Research journal homepage: www.elsevier.com/locate/jglr
Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013 Douglas C. Tozer ⁎ Bird Studies Canada, Box 160, 115 Front Street, Port Rowan, ON N0E 1M0, Canada
a r t i c l e
i n f o
Article history: Received 7 November 2014 Accepted 15 October 2015 Available online xxxx Communicated by Craig Hebert Index words: Citizen science Coastal wetland Great Lakes Multi-scale Marsh bird Multi-season site occupancy model
a b s t r a c t Using data from 21,546 point counts conducted by volunteers in Bird Studies Canada's Great Lakes Marsh Monitoring Program, I assessed whether occupancy of 15 breeding marsh bird species increased or decreased throughout the southern portion of the Great Lakes basin between 1996 and 2013. I accounted for differences in detection probability, addressed spatial autocorrelation, and assessed whether initial occupancy in 1996 and subsequent colonization or extinction at a site within and across species was influenced by site, wetland, and landscape scale covariates. Occupancy of 9 of 15 (60%) species significantly decreased, whereas occupancy of only 1 (7%) species significantly increased. The results show the power of citizen science and suggest that the largest number of decreasing marsh-dependent breeding bird species will benefit from conserving, restoring, or creating large wetlands surrounded by limited urban land use, and from addressing issues within International Joint Commission Areas of Concern. Plus, individual or smaller groups of decreasing species will also benefit from conserving, restoring, or creating robust-emergent-dominated but interspersed, purple loosestrife (Lythrum salicaria)-free, Phragmites-free wetlands surrounded by higher proportions of wetland cover in the surrounding landscape, and from addressing issues within Great Lakes coastal wetlands. These actions will help promote colonization or reduce extinction and help slow or maybe even reverse declining trends in occupancy among decreasing species across the southern portion of the Great Lakes basin. © 2015 International Association for Great Lakes Research. Published by Elsevier B.V. All rights reserved.
Introduction Monitoring marsh dependent breeding birds can be challenging due to the birds' secretive behavior, low numbers, infrequent occurrence, and inaccessible habitat (Conway and Gibbs, 2011; Steidl et al., 2013). For instance, the Breeding Bird Survey (BBS), a passive, long-term, roadside monitoring program (Environment Canada, 2010), which is often considered the best source of information on North American bird population trends (e.g., North American Bird Conservation Initiative Canada, 2012), detected secretive marsh bird species on less than 5 of 91 (5%) survey routes per species per year on average throughout the Great Lakes-St. Lawrence lowlands between 1996 and 2013 (Pardieck et al., 2014). Such low numbers of annual detections require N30 years of data to achieve adequate statistical power for estimating linear population trends (Steidl et al., 2013). This is not effective for conservation and management because declines may be more subtle or rapid, and identification of problems needed more quickly. To remedy the situation, Bird Studies Canada launched the Great Lakes Marsh Monitoring Program (GLMMP) in 1995, a volunteer citizen science program (Dickinson et al., 2010), which has operated every year since (Bird Studies Canada, 2014). To date, over 1500 trained ⁎ Corresponding author. Tel.: +1 519 586 3531. E-mail address: [email protected].
participants have performed ~28,000 call broadcast surveys at targeted wetlands, resulting in detections of secretive marsh birds at greater than 50 of 232 (21%) survey routes per species per year on average in the Great Lakes basin, many times more detections than achieved by the BBS in the same region. Using standardized broadcasts of secretive species, visiting survey points at least two times per season, and surveying only in suitable emergent marsh habitat ensures that large sample sizes are achieved by the GLMMP each year for estimating population trends (Tozer, 2013). With the detection rates achieved by the GLMMP, it requires b15 years, or about half the time compared to the BBS, to achieve adequate statistical power for estimating population trends of most secretive marsh bird species in the Great Lakes basin (Steidl et al., 2013). However, in order to garner large numbers of volunteer participants to achieve such large sample sizes, the program has largely allowed participants to select sampling sites, which makes drawing inferences to larger un-sampled sets of sites less robust. Nonetheless, if one assumes that the GLMMP sites are approximately representative of wetlands in the region, the high precision of trends from the GLMMP make it the most suitable long-term dataset for guiding conservation of secretive marsh birds in the Great Lakes basin. Previous analyses using the GLMMP dataset have identified numerous marsh bird population declines (e.g., Timmermans et al., 2008; Tozer, 2013, 2014). In a recent example, populations of 10 of 19 (53%) marsh breeding bird species declined between 1995 and 2012 across
http://dx.doi.org/10.1016/j.jglr.2015.10.015 0380-1330/© 2015 International Association for Great Lakes Research. Published by Elsevier B.V. All rights reserved.
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
2
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
the Great Lakes basin, including bitterns (e.g., Botaurus), shallow(e.g., Porzana) and deep-water rails (e.g., Gallinula), and marsh-nesting terns (e.g., Chlidonias; Tozer, 2013). One study suggested that marsh bird trends were more negative within than outside Great Lakes International Joint Commission (IJC) areas of concern (Tozer, 2013), and another reported that declining marsh bird trends were steeper in coastal marshes in Lake Erie, Lake Huron, and Lake Michigan than in Lake Ontario (Timmermans et al., 2008). However, none of the previous studies above adjusted for differences in detection probability, addressed spatial autocorrelation, or explored influences of habitat covariates on occupancy dynamics. In this study, I report extended and updated annual occupancy estimates and trends of marsh-breeding birds in the southern portion of the Great Lakes basin based on GLMMP data. But I do so while addressing each of the three important items left unattended by the earlier studies listed above, which adds tremendous value to the results for conservation as elaborated on below. The probability of detecting marsh birds during surveys varies with a number of factors including time of day, date, and weather (Conway and Gibbs, 2011). As a result, the probability of detecting marsh birds during surveys is almost never perfect (Mackenzie et al., 2003). Birds can be unavailable for detection for numerous reasons including being absent, or being present but never being visible or audible to the observer (Dénes et al., 2015). Birds can also be missed because observers fail to see or hear them, even though they were visible or audible, respectively (Dénes et al., 2015). Thus, detection probability must be accounted for in a way such as I accomplish in this study because if it varies systematically over time, unadjusted occupancy estimates may show trends where none exist, or fail to show trends where they do exist (Kéry and Schmidt, 2008). Marsh bird survey points tend to be clustered at various scales, such as within wetlands or within wetland complexes (e.g., Conway, 2011). This often results in responses being more similar among clustered compared to unclustered points simply because they are closer together in space (Dormann et al., 2007). Referred to as spatial autocorrelation, this common feature of survey data violates independence assumptions required by statistical modeling (Dormann et al., 2007). Thus, spatial autocorrelation must be accounted for in a way such as I accomplish here because if certain spatially clustered points are assumed to be independent they may unduly influence parameter estimates and associated confidence limits and lead to unfounded recommendations for conservation. Long-term marsh bird survey data provide important information on trends over time (e.g., Tozer, 2014). However, these data also harbor valuable information on habitat features associated with high colonization and low extinction (MacKenzie et al., 2003). This information then translates into conservation actions that will help promote colonization or reduce extinction, respectively, and in turn help slow or maybe even reverse declining trends. Thus, one should explore occupancy dynamics in a way such as accomplished here, particularly for species showing significantly decreasing occupancy over time (hereafter “decreasing species”) versus those that do not (hereafter “increasing or stable species”), because it ultimately leads to more informed beneficial conservation actions for decreasing species. My ultimate goal was to determine if occupancy for each of 15 marsh bird breeding species changed throughout the southern portion of the Great Lakes basin over the past two decades, and to generate conservation actions that will benefit decreasing species. In doing so, I was careful to account for differences in detection probability, address spatial autocorrelation, and assess whether initial occupancy and subsequent colonization or extinction at a site within and across species was influenced by site, wetland, and landscape scale covariates. Pursuing these outcomes seemed especially justified within the southern portion of the Great Lakes basin because intensive human land use and associated environmental stress has, and continues to have, a strong influence on remaining wetlands, in an area where up to 90% of wetlands have been converted to other land uses (e.g., Ducks Unlimited Canada,
2010; Wolter et al., 2006). I accomplished my goal using multi-season site occupancy models and an information theoretic approach to simultaneously evaluate variables potentially related to detection and occupancy dynamics using a large dataset consisting of ~ 21,000 GLMMP point counts conducted at ~2,100 sites over 18 years. Methods Study design The GLMMP expanded throughout the Great Lakes basin in 1995, although coverage in the initial year and in the northern portion of the basin was never as extensive as subsequent years and in the southern portion of the basin (Fig. 1). I included only data from 1996 onward from the southern portion of the Great Lakes basin because, from that point and onward, there was reasonably good geographic coverage for birds throughout the area (~588 sites surveyed across ~181 routes per year on average; Fig. 2). I defined the southern portion of the Great Lakes basin as the part of the basin within Bird Conservation Regions 13, 14, 22–24, and 28 (Bird Studies Canada and NABCI, 2014). GLMMP participants selected survey marshes or they were randomly assigned. Participants conducted surveys within 1–8 semi-circular 100 m-radius plots (hereafter “sites”) along survey routes within one marsh or across multiple marshes. Most sites (90%) were located on shore adjacent to marshes that were covered mostly by non-woody plants. Sites were N250 m apart to avoid double-counting individuals. Bird surveys Each site was surveyed for 10 min on 2–3 occasions at least 10 days apart between 20 May and 5 July in each year that the survey route was active. Surveys occurred in either the morning (30 min before local sunrise to 1000 local time) or evening (4 h before local sunset to dark) and only under weather conditions that were favorable for detecting all species and individuals present (no precipitation; wind: Beaufort 0–3, 0– 19 km/h, recorded during each survey). Participants broadcasted calls during surveys to entice individuals of especially secretive species to reveal their presence by approaching or responding vocally. The calls occurred in the first 5 min of each 10-min survey and consisted of 30 s of vocalizations followed by 30 s of silence for each of least bittern (Ixobrychus exilis), sora (Porzana carolina), Virginia rail (Rallus limicola), a mixture of American coot (Fulica americana) and common gallinule (Gallinula galeata), and pied-billed grebe (Podilymbus podiceps), in that order. The GLMMP bird survey protocol is described in more detail in Bird Studies Canada (2009a). For each survey, I assigned day-of-year (e.g., 20 May = 140; hereafter “date”), morning versus evening (defined above; hereafter “time of day”), and a Beaufort wind scale value (hereafter “wind”). Site covariates Visual estimates of percent areal coverage of major habitat and vegetation types were made annually within 100-m-radius semi-circular plots at each site. Multi-season site occupancy models can, in theory, incorporate these annual estimates directly into models, but as I describe under Analysis below, attempts to include these “yearly site covariates” (sensu Fiske et al., 2015) resulted in models that failed to converge. Instead, I took the first year value for each site to get coverage of open water (hereafter “% water”), robust emergent herbaceous vegetation (e.g., cattails [Typha], hereafter “% emergent”), common reed (Phragmites sp., hereafter “% Phragmites”), purple loosestrife (Lythrum salicaria; hereafter “% loosestrife”), and trees and shrubs (e.g., willows [Salix], hereafter “% trees and shrubs”), and I took the difference between the last-year value and the first-year value at each site to get overall change in coverage of open water (hereafter “change in % water”), robust emergent herbaceous vegetation (hereafter “change in
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
3
Fig. 1. Routes surveyed for birds by participants in Bird Studies Canada's Great Lakes Marsh Monitoring Program (GLMMP) throughout the Great Lakes basin between 1996 and 2013. In this study, I used data from only the southern portion of the basin, south of the Canadian Shield (solid gray; Great Lakes basin – south) to estimate detection probability and occupancy dynamics of marsh bird species because sampling and land use were more uniform throughout this region than to the north (hatched; Great Lakes basin – north). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.).
% emergent”), common reed (hereafter “change in % Phragmites”), purple loosestrife (hereafter “change in % loosestrife”), and trees and shrubs (hereafter “change in % trees and shrubs”). I included these latter
change measures as “site covariates” (sensu Fiske et al., 2015) in the colonization and extinction processes during modeling. The GLMMP habitat survey protocol is described in more detail in Bird Studies Canada (2009b).
Wetland and landscape covariates
Fig. 2. Number of sites or routes surveyed for birds per year (upper panel) and number of sites as a function of the number of years surveyed (lower panel) by participants in Bird Studies Canada's Great Lakes Marsh Monitoring Program throughout the southern portion of the Great Lakes basin between 1996 and 2013. Numbers above bars show the value of each bar as a percent of the total number of sites surveyed.
Previous studies show that marsh bird occupancy or abundance is influenced by land cover characteristics within a distance of at least 3 km from the wetland (e.g., Tozer et al., 2010), so I used land cover or wetland inventory data and the tools in ArcGIS 10.2 (ESRI, 2013) to delineate the proportion of wetland, agriculture, and urban land cover within 3 km of the sample point (hereafter “% wetland”, “% agriculture”, and “% urban”, respectively). I omitted from these calculations portions of 3-km buffers that were part of one of the Great Lakes. Percent wetland was characterized in Ontario using the Consolidated Wetland Layer (Ontario Ministry of Natural Resources) and in the US using the National Wetland Inventory (US Fish and Wildlife Service). All wetland types except open water (Ontario) and lakes and riverine (US) were included in % wetland. I included all wetland types partly because the Ontario data had polygons with unknown wetland types, so restricting the analysis to only emergent wetland polygons might have introduced unknown biases. I also determined the area of each wetland containing each site using these same wetland layers (in ha); and used the loge of wetland area to capture the effect of this covariate on marsh birds (e.g., Tozer et al., 2010; hereafter “log wetland area”). Percent agriculture and % urban were characterized in Ontario using the Ontario Land Cover Compilation (Ontario Ministry of Natural Resources) and in the US using the National Land Cover Database (United States Geological Survey). Percent agriculture in Ontario consisted of pasture and abandoned fields and cropland, whereas in the U.S. it consisted of pasture/ hay and cultivated crops. Percent urban in Ontario consisted of settlement and developed land, whereas in the U.S. it consisted of developed
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
4
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
open space and developed low, medium, and high intensity. Each site was also categorized as being within or outside the watershed of a Great Lakes IJC Area of Concern (hereafter “AOC”; IJC, 2012) and along a Great Lakes coast (hereafter “coastal”) or inland. Sites were defined as coastal based on a hydrogeomorphic classification system (Albert et al., 2005; Great Lakes Coastal Wetlands Consortium, 2004). Analyses The response variable was the occurrence within sites during individual surveys of each of 15 breeding marsh bird species. I used multiseason site occupancy models for each species (MacKenzie et al., 2003) and an information theoretic approach via Akaike's information criterion (AIC) to select best approximating models from among sets of candidate models (Burnham and Anderson, 2002). Multi-season site occupancy models consist conceptually of two coupled hierarchical logistic regressions, the first describing actual initial occupancy at each site in the first year and subsequent colonization and extinction in each of the following years, and the second describing observed occupancy during each survey, conditional on actual occupancy. When the two processes are combined and parameterized via maximum likelihood they jointly describe occupancy dynamics, i.e., probability of initial occupancy in 1996, probability of subsequent colonization, and probability of subsequent extinction, while accounting for uncertainties in detection (MacKenzie et al., 2003). At each of several stages of model selection described in more detail below, I selected the top ranked model, i.e., ΔAIC = 0, as the best model and passed it along to the next stage, except in cases where a lower ranked model existed within 2 AIC units of the top ranked model, i.e., ΔAIC ≤ 2, which included all of the covariates found within all of the higher ranked models. In these latter instances I selected the lower ranked but more complex model as the best model to avoid discarding important variables. I tested the overall fit of global models using a parametric bootstrap procedure and associated Χ2 statistic to ensure that covariates as a group reliably explained the response variable; P-values N 0.05 indicated adequate fit (Fiske and Chandler, 2015). I selected for analysis 21,546 point counts where covariate information was complete, discarding point counts with missing covariate data as recommended by Kéry and Chandler (2012). I conducted my analysis for each species in five stages to make the number of models manageable. First, I considered the influence of all single-variable (3 models) and all two-variable additive combinations (3 models) of date, time of day, and wind on detection probability, along with an intercept-only and global model, which included all of the covariates, a total of 8 models per species. I did not consider cloud cover or ambient temperature because most past studies have reported equivocal effects of cloud cover and no effect of ambient temperature on vocalization probability in marsh birds (Conway and Gibbs, 2011). I included the covariate or covariates from the best detection probability model in the detection process of all subsequent models for each species described below. Given that my main goal did not involve estimating detection probability or what influences it, I do not report or discuss this part of the analysis in any detail. Second, I examined the influence of the following 11 covariates on the probability of initial occupancy in 1996: % water, % emergent, % Phragmites, % loosestrife, % trees and shrubs, loge wetland area, % wetland, % agriculture, % urban, AOC, and coastal. I considered all singlevariable models (11 models) and additive combinations of only sitescale (% water, % emergent, % Phragmites, % loosestrife, and % trees and shrubs), wetland-scale (log wetland area, AOC, coastal), and landscape-scale (% wetland, % agriculture, and % urban) covariates (3 models), along with an intercept-only and global model, which included all of the covariates, a total of 16 models per species. I included the covariate or covariates from the best initial occupancy model in the initial occupancy process of all subsequent models for each species described below.
Third, I examined the influence of the following 11 covariates on the probability of subsequent colonization: change in % water, change in % emergent, change in % Phragmites, change in % loosestrife, change in % trees and shrubs, log wetland area, % wetland, % agriculture, % urban, AOC, and coastal. I considered all single-variable models (11 models) and additive combinations of site-scale (change in % water, change in % emergent, change in % Phragmites, change in % loosestrife, and change in % trees and shrubs), wetland-scale (log wetland area, AOC, coastal), and landscape-scale (% wetland, % agriculture, and % urban) covariates (3 models), along with an intercept-only and global model, which included all of the covariates, a total of 16 models per species. I treated all 11 of the covariates as “site covariates” (sensu Fiske et al., 2015). It is worth noting that information from the annual site-scale vegetation covariates could have been treated as “yearly site covariates” (sensu Fiske et al., 2015), but initial attempts to do so resulted in models that failed to converge, which is why I calculated the change between the last survey and the first survey at each site and treated the result as a site covariate. This is justified because the model estimates the average annual probability of colonization for a given level of change (positive or negative) in each covariate, and even though one estimate of change across all of the years in each time-series imposes a temporal smoothing on the resulting estimates of occupancy, it is still useful for assessing the influence of covariates on colonization, and similarly for extinction described below. I included the covariate or covariates from the best colonization model in the colonization process of all subsequent models for each species described below. Fourth, I examined the influence on the probability of subsequent extinction of the same 11 covariates listed for colonization above, via a similar candidate model set described for colonization above. I included the covariate or covariates from the best extinction model in the extinction process of all subsequent models for each species described below. Fifth, to account for spatial autocorrelation, I compared a model that included latitude, longitude, latitude2, longitude2, and latitude × longitude in the occupancy, colonization, and extinction processes to a model that did not include any latitude or longitude terms (e.g., Schuster and Arcese, 2013). For all species, the model containing the latitude and longitude terms fit much better than the model that did not contain these terms. So I included latitude, longitude, latitude2, longitude2, and latitude × longitude in the occupancy, colonization, and extinction processes of all final best models for each species, and I do not report any further details for these models or any further details for the latitude or longitude terms in each model. I note that I could have accounted for spatial autocorrelation earlier in the modeling process, but doing so would have assumed that any spatial component in the results was error, regardless of whether the spatial component was the result of a functional relationship between a spatially arranged covariate and the response, and in turn might have unduly discarded covariates useful for inference (e.g., Dormann, 2007). To assess trends in occupancy over time, I plotted annual occupancy estimates from final best models with the addition of a categorical effect of year in each of the detection, colonization, and extinction processes, i.e., a fully time dependent model as described by Weir et al. (2009) and Kéry and Chandler (2012). Annual occupancy was the proportion of sampled sites that was occupied in each year, rather than the occupancy rate of the entire population of hypothetical sites from which the sampled sites were selected; see Weir et al. (2009) for further details, where such estimates are referred to as “smoothed trajectories”. Confidence intervals for annual occupancy estimates were generated via nonparametric bootstrapping as described in Kéry and Chandler (2012). I estimated trends in occupancy over time as the geometric mean rate of change between annual occupancy in 1996 and in 2013 as described in equation 4 of Smith et al. (2014), and assessed the statistical significance of each trend via parametric bootstrapping. I accomplished the latter by writing an R-function that calculated the trend in occupancy (i.e., the geometric mean rate of change between annual occupancy in 1996 and in 2013), then I fed the R-function into the
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
“parboot” function in the unmarked package, which derived an estimate of the 95% confidence limits around the occupancy trend estimate (see below for software details; Kéry and Chandler, 2012). Bootstrapping in this manner was necessary to account for the varying precision of the annual estimates used to calculate each occupancy trend. Occupancy trend estimates with 95% confidence intervals that did not overlap zero were considered statistically significant. To compare parameter estimates of covariates within and across species for initial occupancy, colonization, and extinction, I plotted on the logit scale standardized parameter estimates and associated 95% confidence limits from final best models for each species. To compare relative importance of covariates across decreasing species and across increasing or stable species for initial occupancy, colonization, and extinction, I plotted for each covariate for each group of species the mean and associated 95% confidence limits of the species-level sums of Akaike weights from all models that contained the covariate of interest. To compare parameter estimates of covariates across decreasing species or across increasing or stable species for initial occupancy, colonization, and extinction, I plotted on the logit scale for each group of species the mean and associated 95% confidence limits of standardized species-level coefficients from final best models. I then used the plots of mean relative importance and mean coefficients to compare and contrast occupancy dynamics between decreasing versus increasing or stable species. I was especially interested in identifying habitat features associated with higher mean initial occupancy, higher mean colonization, or lower mean extinction across decreasing species, identified by 95% confidence intervals around means that did not overlap zero, because actions to increase such features will likely maximize conservation benefits. All analyses were performed using the R programming language (R Development core Team, 2014). To facilitate model convergence and estimation and comparison of parameter estimates across variables and species, I standardized all of the explanatory variables prior to analysis by subtracting the mean and dividing by the standard deviation (Kéry and Chandler, 2012). Models were fitted using package unmarked (Fiske et al., 2015) and AIC tables were generated using package AICcmodavg (Mazerolle, 2013). Results The dataset consisted of 21,546 point counts conducted at 2149 sites along 706 routes over 18 years between 1996 and 2013 (Figs. 1 and 2). The number of routes surveyed per year (181 ± 52 [mean ± SD]) ranged from 114 to 255 (Fig. 2). Correspondingly, the number of sites surveyed per year (588 ± 136) ranged from 410 to 769 and increased between 1996 and 2013 (linear regression: F1,16 = 23.9, P = 0.0002; Fig. 2). However, there was only low to moderate correlation between annual occupancy for each species and the number of sites surveyed per year (Pearson |r| across all species: mean = 0.3, range: 0.1 to 0.6), which suggested that the number of sites surveyed per year was not a confounding factor. Most (76%) sites were surveyed for 2 or more years, with 52% surveyed for 2–6 years and 24% surveyed for 7–18 years (Fig. 2). The number of sites as a function of covariates is given in Fig. 3. Global models for each species fit adequately (all P N 0.2), which suggested that the site, wetland, and landscape scale covariates as a group adequately explained occupancy dynamics for each species. However, confidence limits for initial occupancy in 1996 for pied-billed grebe were inestimable for % loosestrife, coastal, and % agriculture; parameter estimates and associated confidence limits were unreasonable, i.e., many times larger and wider, respectively, compared to all of the other estimates for all of the other species, or confidence limits were inestimable for colonization for American coot, common gallinule, and mute swan; and parameter estimates and associated confidence limits were unreasonable for extinction for American coot, common gallinule, least bittern, and mute swan. Thus, I do not report parameter estimates
5
or confidence limits in these instances, nor do I consider these covariates in subsequent analyses below. Model selection details for the detection, initial occupancy, colonization, and extinction possesses are given in Electronic Supplementary Material (ESM) Tables S1, S2, S3, and S4, respectively. Occupancy significantly decreased between 1996 and 2013 throughout the southern portion of the Great Lakes basin for 9 of 15 (60%) species, whereas occupancy significantly increased for 1 (7%) species and remained statistically unchanged for 5 (33%) species (Fig. 4). Across all species with significant trends, occupancy of black tern (Chlidonias niger) decreased the most, followed by American coot, common gallinule, sora, pied-billed grebe, American bittern (Botaurus lentiginosus), Virginia rail, least bittern, and Canada goose (Branta canadensis); whereas occupancy of common yellowthroat (Geothlypis trichas) showed the only significant increase (Fig. 4). Common grackle (Quiscalus quiscula), marsh wren (Cistothorus palustris), mute swan (Cygnus olor), red-winged blackbird (Agelaius phoeniceus), and swamp sparrow (Melospiza georgiana) showed no statistically significant changes over time (Fig. 4). Notably, occupancy of 100% of 7 especially secretive marsh-dependent species significantly declined during the study: American bittern, American coot, common gallinule, least bittern, pied-billed grebe, sora, and Virginia rail (Fig. 4). The relative importance, magnitude, and direction of covariates and their associated parameter estimates for explaining the initial occupancy, colonization, or extinction process varied greatly among species (ESM Tables S1, S2, S3; Figs. S1, S2, S3). As such, there were few consistent patterns across species, with the following exceptions. The mean relative importance of wetland scale covariates, i.e., AOC, coastal, and log wetland area, for the colonization and extinction processes across decreasing species and across increasing or stable species was higher compared to each of the rest of the covariates, although 95% confidence intervals overlapped among all of the covariates (Fig. 5). According to mean standardized parameter estimates from final best models for decreasing species, initial occupancy in 1996 decreased with increasing % loosestrife and % trees and shrubs, but increased with increasing % open and log wetland area; colonization decreased with increasing % urban, but increased with increasing log wetland area; and extinction was higher within AOCs than outside AOCs (Fig. 5). By contrast, according to mean standardized parameter estimates from final best models for increasing or stable species, initial occupancy in 1996 was higher at coastal compared to inland sites; colonization increased with increasing change in % trees and shrubs and was higher within AOCs than outside AOCs; and extinction was higher within AOCs than outside AOCs (Fig. 5). Notable species-level relationships for decreasing species included the following. Initial occupancy in 1996 decreased with increasing % Phragmites for Virginia rail, decreased with increasing % urban for American coot, black tern, common gallinule, and pied-billed grebe, and was lower at coastal compared to inland sites for American bittern; but increased with increasing % emergent for common gallinule and Virginia rail, increased with increasing % wetland for American bittern, American coot, black tern, and common gallinule, and was higher at coastal compared to inland sites for black tern and common gallinule (Fig. S1). Colonization decreased with increasing change in % emergent and change in % open for pied-billed grebe, decreased with increasing change in % loosestrife for sora, decreased with increasing change in % trees and shrubs for pied-billed grebe, decreased with increasing % urban and % wetland for Virginia rail, was lower within AOCs compared to outside for pied-billed grebe and sora, and was lower at coastal compared to inland sites for black tern and Virginia rail; but was higher at coastal compared to inland sites for pied-billed grebe (Fig. S2). Extinction decreased with increasing change in % emergent, change in % open, and change in % trees and shrubs for pied-billed grebe, decreased with increasing log wetland area for American bittern and sora, and decreased with increasing % agriculture, % urban, and % wetland for Virginia rail; but increased with increasing change in % loosestrife for Virginia rail, increased with increasing % Phragmites for pied-billed grebe, increased
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
6
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
Fig. 3. Number of sites as a function of covariates used to estimate occupancy dynamics of 15 breeding marsh bird species throughout the southern portion of the Great Lakes basin between 1996 and 2013. Variables are defined in the methods. Q = quartile.
with increasing log wetland area for Virginia rail, and was higher at coastal compared to inland sites for Virginia rail (Fig. S3). Discussion Occupancy of 60% of the marsh breeding bird species considered in this study significantly decreased across the southern portion of the Great Lakes basin between 1996 and 2013. Nearly all of the decreasing species nest and forage, almost exclusively, in emergent marsh habitat.
By contrast, all of the increasing or stable species, with the exception of marsh wren and swamp sparrow, frequently nest and forage in habitats other than emergent marsh, such as overgrown fields and open-water ponds nearly devoid of emergent vegetation. For increasing or stable species with more general habitat requirements, emergent marsh breeding habitat may have been less important for maintaining occupancy. Given that nearly all of the decreasing species were marshdependent may suggest that the quality of emergent marsh breeding bird habitat declined throughout, or in part of, the southern portion of
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
7
Fig. 4. Temporal trends in occupancy of 15 breeding marsh bird species throughout the southern portion of the Great Lakes basin between 1996 and 2013. Shown are annual occupancy estimates (circles) with associated 95% confidence limits (solid lines) from final best models. Naive annual occupancy estimates unadjusted for differences in detection probability or spatial autocorrelation are also shown (dashed lines). See methods for details. Trends in occupancy (%/year) and associated 95% confidence limits (lower, upper) are shown for each species; species with statistically significant trends are shown in bold. Note that the vertical axis differs among species, but always has an extent of 50.
the Great Lakes basin during the study. However, it cannot be ruled out that these patterns were instead, or in part, caused by mechanisms operating beyond the southern portion of the Great Lakes basin, such as mechanisms operating in migratory staging areas or on the wintering grounds or both (Hostetler et al., 2015). Nonetheless, by linking colonization and extinction to site, wetland, and landscape scale covariates, I was able to identify conservation actions that will likely benefit decreasing species within the southern portion of the Great Lakes basin. The tremendous variation in occupancy dynamics as a function of covariates within and among species in this study makes generating conservation actions to benefit most decreasing species especially challenging. However, the results suggest that wetland scale covariates were the most important overall drivers of colonization and extinction across decreasing species, with mean probability of colonization increasing with increasing log wetland area and mean probability of extinction being higher within AOCs compared to outside AOCs (Fig. 5).
By contrast, site and landscape scale covariates were less important drivers of colonization and extinction, with only mean probability of colonization decreasing with increasing % urban (Fig. 5). These general results suggest that the largest number of decreasing marshdependent breeding bird species will benefit from conserving, restoring, or creating large wetlands surrounded by limited urban land use, and from addressing issues within AOCs. Such actions will promote colonization and reduce extinction of the largest number of decreasing species, which will potentially help slow or maybe even reverse the declining trends in occupancy identified in this study. The above results and associated conservation actions are not surprising considering that similar patterns have been found before amongst at least some of the decreasing species. For instance, occupancy or abundance of at least some of the decreasing species increased with increasing wetland area in Iowa Prairie marshes (Brown and Dinsmore, 1986), Great Lakes coastal wet meadows (Riffell et al.,
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
8
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
Fig. 5. Mean relative importance (mean weight) and mean standardized coefficients (mean coefficient) and associated 95% confidence intervals of site, wetland, and landscape scale covariates for explaining initial occupancy, colonization, and extinction of decreasing species and increasing or stable species throughout the southern portion of the Great Lakes basin between 1996 and 2013. Mean weight calculated as the mean of the species-level sums of Akaike weights from all models that contained the covariate of interest. Mean coefficient calculated as the mean of standardized species-level coefficients from final best models. See text for details. chg = change; emerg = emergent; l-strife = loosestrife; Phrag = Phragmites; t + s = trees and shrubs; wet = wetland.
2001), and cattail-dominated wetlands in southern Ontario (Tozer et al., 2010); and decreased with increasing amount of surrounding urban land use in Lake Erie and Lake Ontario coastal wetlands (Smith and Chow-Fraser, 2010). Furthermore, abundance was lower, or time trends in abundance were more negative, within AOCs than outside AOCs (Tozer, 2013). But these previously known patterns and associated conservation actions have never before been supported by such extensive long-term data from throughout the southern Great Lakes region as is the case here. The conservation actions outlined above apply to the largest number of decreasing species. However, the results of the study suggest a number of additional actions that will benefit individual or smaller numbers of decreasing species. In fact, all but one (% agriculture) of the covariates were associated with lower probability of colonization or higher probability of extinction of at least one or more decreasing species (Figs. S1, S2, S3). It follows that each of these patterns forms the basis for a beneficial conservation action. Thus, along with the more general actions
listed above, the results suggest that it is also important to conserve, restore, or create robust-emergent-dominated but interspersed, purple loosestrife-free, Phragmites-free wetlands surrounded by higher proportions of wetland cover in the surrounding landscape, and to address issues within Great Lakes coastal wetlands. As above, these additional results and associated actions are, for the most part, not surprising considering that similar patterns have been found before amongst at least some of the decreasing species for most of the covariates. For instance, abundance of nearly half of the decreasing species found in this study increased with increasing open water-emergent vegetation interspersion in New York (Rehm and Baldassarre, 2007), and in at least one of the years of another study in the Midwestern US (Bolenbaugh et al., 2011). Pooled abundance of marsh-nesting birds decreased slightly with increasing Phragmites cover in Great Lakes coastal marshes at Long Point, Ontario (Meyer et al., 2010). Occurrence or density increased with increasing amount of wetland in the surrounding landscape for least bittern in Québec
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
(Jobin et al., 2011), least bittern and Virginia rail in eastern Ontario (Quesnelle et al., 2013), and various other deceasing species in Iowa (Fairbairn and Dinsmore, 2001). Abundance of some decreasing species was lower at Great Lakes coastal compared to inland sites (Tozer, 2013). But like the patterns identified above, these previously known patterns and associated conservation actions have never before been supported by such extensive long-term data as is the case here. By contrast, the results for colonization and extinction for % loosestrife from this study appear to be novel, seeing as a recent review of peer-reviewed literature on the topic failed to find any reference to the relationship between occupancy dynamics and the presence of purple loosestrife for any of the decreasing species (Lavoie, 2010). Each of the conservation actions described above is based on colonization and extinction dynamics. Focusing on these processes makes sense because they are most directly linked to changes in occupancy during the period of the study; and are, therefore, most useful for generating beneficial conservation actions. However, it is worth noting that for decreasing species mean probability of initial occupancy in 1996 decreased with increasing % loosestrife and % trees and shrubs, but increased with increasing % open and log wetland area (Fig. 5). Even though these patterns were presumably largely influenced by mechanisms operating prior to the period of study, they still likely provide useful information for generating beneficial conservation actions because they describe conditions that will promote occupancy. Thus, it may be justified that conserving, restoring, or creating robust-emergent-dominated but interspersed, purple loosestrife-free wetlands should also appear in the list of general conservation actions to benefit the largest number of decreasing species given above, rather than appearing only in the list to benefit individual or smaller groups of decreasing species. I used data collected over many years by a large network of citizen scientists to assess whether occupancy of breeding marsh bird species increased or decreased throughout the southern portion of the Great Lakes basin, and whether occupancy dynamics at a site within and across species was influenced by site, wetland, and landscape scale covariates. Conducting over 21,500 point counts across 18 years without a program such as Bird Studies Canada's GLMMP would not have been possible, which highlights the utility of using citizen science programs to monitor wildlife. The results suggest that the largest number of decreasing marsh-dependent breeding bird species will benefit from conserving, restoring, or creating large wetlands surrounded by limited urban land use, and from addressing issues within AOCs. In addition, individual or smaller groups of decreasing species will also benefit from conserving, restoring, or creating robust-emergent-dominated but interspersed, purple loosestrife-free, Phragmites-free wetlands surrounded by higher proportions of wetland cover in the surrounding landscape, and from addressing issues within Great Lakes coastal wetlands. These actions will help promote colonization or reduce extinction and help slow or maybe even reverse declining trends in occupancy among decreasing species across the southern portion of the Great Lakes basin. Acknowledgments I am grateful for the dedication of hundreds of GLMMP participants, whose collective effort made the study possible. I am also grateful for the coordination and promotion of the GLMMP by Kathy Jones and a large team of regional coordinators throughout the Great Lakes basin; and for past coordination of the program by Russ Weeber, Steve Timmermans, and Ryan Archer. Adam C. Smith, an anonymous reviewer, Myles Falconer, Tara Crewe, Becky Stewart, and Karla Falk provided helpful comments and statistical advice that improved the paper. Bird Studies Canada is thankful for ongoing or recent support of the GLMMP by Environment Canada, United States Environmental Protection Agency, Government of Ontario, Great Lakes Restoration Initiative, Eastern Habitat Joint Venture, Wildlife Habitat Canada, Ducks Unlimited Canada, TD Friends of the Environment Foundation, and U.S. Fish and Wildlife Service; and for past support by Brant Waterways Foundation,
9
Great Lakes Commission, Great Lakes Sustainability Fund, Great Lakes United, National Audubon Society, Ontario Trillium Foundation, Wetland Habitat Fund, and Great Lakes Protection Fund. Appendix A. Supplementary data Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.jglr.2015.10.015. References Albert, D.A., Wilcox, D.A., Ingram, J.W., Thompson, T.A., 2005. Hydrogeomorphic classification for Great Lakes coastal wetlands. J. Great Lakes Res. 31, 129–146. Bird Studies Canada, 2009a. Marsh Monitoring Program Participant's Handbook Getting Started. Bird Studies Canada in Cooperation with Environment Canada and the. U.S. Environmental Protection Agency, Port Rowan, ON. Bird Studies Canada, 2009b. Marsh Monitoring Program Participant's Handbook for Surveying Marsh Birds. Bird Studies Canada in Cooperation with Environment Canada and the. U.S. Environmental Protection Agency, Port Rowan, ON. Bird Studies Canada, 2014. The Great Lakes Marsh Monitoring Program. Bird Studies Canada, Port Rowan, ON (http,//www.birdscanada.org/volunteer/glmmp/index.jsp? lang = EN&targetpg = index Accessed 5 June 2015). Bird Studies Canada, NABCI, 2014. Bird Conservation Regions. (Published by Bird Studies Canada on behalf of the North American Bird Conservation Initiative. http://www. birdscanada.org/research/gislab/index.jsp?targetpg = bcr Accessed 5 June 2015). Bolenbaugh, J.R., Krementz, D.G., Lehnen, S.E., 2011. Secretive marsh bird species cooccurrences and habitat associations across the Midwest, USA. J. Fish Wildl. Manag. 2, 49–60. Brown, M., Dinsmore, J.J., 1986. Implications of marsh size and isolation for marsh bird management. J. Wildl. Manag. 50, 392–397. Burnham, K.P., Anderson, D.R., 2002. Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach. Springer-Verlag, New York, NY. Conway, C.J., 2011. Standardized North American marsh bird monitoring protocol. Waterbirds 34, 319–346. Conway, C.J., Gibbs, J.P., 2011. Summary of intrinsic and extrinsic factors affecting detection probability of marsh birds. Wetlands 31, 403–411. Dénes, F.V., Silveira, L.F., Beissinger, S.R., 2015. Estimating abundance of unmarked animal populations: accounting for imperfect detection and other sources of zero inflation. Methods Ecol. Evol. 6, 543–556. Development Core Team, R., 2014. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria 3-900051-07-0 (http://www.R-project.org/ Accessed 5 June 2015). Dickinson, J.L., Zukerberg, B., Bonter, D.N., 2010. Citizen science as an ecological research tool, challenges and benefits. Annu. Rev. Ecol. Evol. Syst. 41, 149–172. Dormann, C.F., 2007. Assessing the validity of autologistic regression. Ecol. Monit. 207, 234–242. Dormann, C.F., McPherson, J.M., Araújo, M.B., Bivand, R., Bolligar, J., Carl, G., Davies, R.G., Hirzel, A., Jetz, W., Kissling, D., Kühn, I., Ohlemüller, R., Peres-Neto, P.R., Reineking, B., Schrӧder, B., Schurr, F.M., Wilson, R., 2007. Methods to account for spatial autocorrelation in the analysis of species distributional data: a review. Ecography 30, 609–628. Ducks Unlimited Canada, 2010. Southern Ontario wetland conversion analysis. Final report, March 2010. Ducks Unlimited Canada, Barrie, ON ([online] URL: http://www. ducks.ca/assets/2010/10/duc_ontariowca_optimized.pdf Accessed 5 June 2015). Environment Canada, 2010. North American Breeding Bird Survey — Canadian Results and Analysis Website version 3.00. Environment Canada, Gatineau, QC (http,// www.ec.gc.ca/reom-mbs/default.asp?lang = En&n = 0D74F35F-1 Accessed 5 June 2015). ESRI, 2013. ArcGIS 10.2. Environmental Systems Resource Institute, Redland, CA. Fairbairn, S.E., Dinsmore, J.J., 2001. Local and landscape-level influences on wetland bird communities of the prairie pothole region of Iowa, USA. Wetlands 21, 41–47. Fiske, I., Chandler, R., 2015. Overview of Unmarked: an R package for the analysis of data from unmarked animals. http://cran.r-project.org/web/packages/unmarked/ vignettes/unmarked.pdf (Accessed 5 June 2015). Fiske, I., Chandler, R., Miller, D., Royle, A., Kéry, M., 2015. Package unmarked. http://cran.rproject.org/web/packages/unmarked/unmarked.pdf (Accessed 5 June 2015). Great Lakes Coastal Wetlands Consortium, 2004. Great Lakes Coastal Wetlands Consortium: Inventory and Classification. http://glc.org/projects/habitat/coastal-wetlands/ cwc-inventory/ (Accessed 5 June 2015). Hostetler, J.A., Sillett, T.S., Marra, P.P., 2015. Full-annual-cycle population models for migratory birds. Auk 132, 433–449. IJC [International Joint Commission], 2012. Great Lakes Water Quality Agreement, Protocol Amending the Agreement between Canada and the United States of America on Great Lakes Water Quality, 1978, as Amended on October 16, 1983, and on November 18, 1987, Signed September 7, 2012, Entered into force February 12, 2013. International Joint Commission, Windsor, ON. Jobin, B., Fradette, P., Labrecque, S., 2011. Habitat use by Least Bitterns (Ixobrychus exilis) in Quebec. Waterbirds 34, 143–150. Kéry, M., Chandler, R., 2012. Dynamic Occupancy Models in Unmarked. Swiss Ornithological Institute and USGS Patuxent Wildlife Research Center (29 March 2012. http:// cran.r-project.org/web/packages/unmarked/vignettes/colext.pdf Accessed 5 June 2015).
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015
10
D.C. Tozer / Journal of Great Lakes Research xxx (2015) xxx–xxx
Kéry, M., Schmidt, B.R., 2008. Imperfect detection and its consequences for monitoring for conservation. Community Ecol. 9, 207–216. Lavoie, C., 2010. Should we care about purple loosestrife? The history of an invasive plant species in North America. Biol. Invasions 12, 1967–1999. MacKenzie, D.I., Nichols, J.D., Hines, J.E., Knutson, M.G., Franklin, A.B., 2003. Estimating site occupancy, colonization, and local extinction when a species is detected imperfectly. Ecology 84, 2200–2207. Mazerolle, M.J., 2013. AICcmodavg: Model Selection and Multimodel Inference Based on (Q)AIC(c). Version 2.00. R Foundation for Statistical Computing, Vienna, Austria (http://CRAN.R-project.org/package = AICcmodavg Accessed 5 June 2015). Meyer, S.W., Badzinski, S.S., Pettrie, S.A., Ankney, C.D., 2010. Seasonal abundance and species richness of birds in common reed habitats in Lake Erie. J. Wildl. Manag. 74, 1559–1567. North American Bird Conservation Initiative Canada, 2012. The State of Canada's Birds, 2012. Environment Canada, Ottawa, ON (http,//www.stateofcanadasbirds.org/State_ of_Canada's_birds_2012.pdf Accessed 5 June 2015). Pardieck, K.L., Ziolkowski, D.J., Hudson, M.-A.R., 2014. North American Breeding Bird Survey Data Set 1966–2013, Version 2013.0. USGS Patuxent Wildlife Research Center, Maryland (www.pwrc.usgs.gov/BBS/RawData/ Accessed 5 June 2015). Quesnelle, P.E., Fahrig, L., Lindsay, K.E., 2013. Effects of habitat loss, habitat configuration and matrix composition on declining wetland species. Biol. Conserv. 160, 200–208. Rehm, E.M., Baldassarre, G.A., 2007. The influence of interspersion on marsh bird abundance in New York. Wilson J. Ornithol. 119, 648–654. Riffell, S.K., Keas, B.E., Burton, T.M., 2001. Area and habitat relationships of birds in Great Lakes coastal wet meadows. Wetlands 21, 492–507. Schuster, R., Arcese, P., 2013. Using bird species community occurrence to prioritize forests for old growth restoration. Ecography 36, 499–507.
Smith, L.A., Chow-Fraser, P., 2010. Impacts of adjacent land use and isolation on marsh bird communities. Environ. Manag. 45, 1040–1051. Smith, A.C., Hudson, M.-A., Downes, C., Francis, C.M., 2014. Estimating breeding bird survey trends and annual indices for Canada: how do the new hierarchical Bayesian estimates differ from previous estimates? Can. Field-Nat. 128, 119–134. Steidl, R.J., Conway, C.J., Litt, A.R., 2013. Power to detect trends in abundance of secretive marsh birds, effects of species traits and sampling effort. J. Wildl. Manag. 77, 445–453. Timmermans, S.T.A., Badzinski, S.S., Ingram, J.W., 2008. Associations between breeding marsh bird abundances and Great Lakes hydrology. J. Great Lakes Res. 34, 351–364. Tozer, D.C., 2013. The Great Lakes Marsh Monitoring Program 1995–2012, 18 Years of Surveying Birds and Frogs as Indicators of Ecosystem Health. Bird Studies Canada, Port Rowan, ON (http,//www.birdscanada.org/download/GLMMPreport.pdf Accessed 5 June 2015). Tozer, D.C., 2014. Coastal Wetland Birds. In State of the Great Lakes 2011. Environment Canada and U.S. Environmental Protection Agency (Cat No. En161-3/1-2011E-PDF. EPA 950-R-13-002. [Online] URL, http://binational.net/wp-content/uploads/2014/ 11/sogl-2011-technical-report-en.pdf Accessed 5 June 2015). Tozer, D.C., Nol, E., Abraham, K.F., 2010. Effects of local and landscape-scale habitat variables on abundance and reproductive success of wetland birds. Wetl. Ecol. Manag. 18, 679–693. Weir, L., Fiske, I.J., Royle, J.A., 2009. Trends in anuran occupancy from northeastern states of the North American amphibian monitoring program. Herpetol. Conserv. Biol. 4, 389–402. Wolter, P.T., Johnston, C.A., Niemi, G.J., 2006. Land use land cover change in the U.S. Great Lakes basin 1992 to 2001. J. J. Great Lakes Res. 32, 607–628.
Please cite this article as: Tozer, D.C., Marsh bird occupancy dynamics, trends, and conservation in the southern Great Lakes basin: 1996 to 2013, J. Great Lakes Res. (2015), http://dx.doi.org/10.1016/j.jglr.2015.10.015