Time trumps habitat in the dynamics of an avian community J. R. Curtis,1,† W. D. Robinson,1 and B. McCune2 1Department
2Department
of Fisheries and Wildlife, Oregon State University, Corvallis, Oregon 97331 USA of Botany and Plant Pathology, Oregon State University, Corvallis, Oregon 97331 USA
Citation: Curtis, J. R., W. D. Robinson, and B. McCune. 2016. Time trumps habitat in the dynamics of an avian community. Ecosphere 7(11):e01575. 10.1002/ecs2.1575
Abstract. Predictions of change in avian biodiversity often fail to account for the natural dynamism of
species assemblages. Information from historical datasets can be useful indicators of how avian communities may change in the future. However, simple comparisons of diversity and abundances over time may overlook additional changes in bird–habitat associations. In 2013, we revisited sites from a unique, highly detailed avian survey from the Willamette Valley, Oregon, conducted in 1952. Our objectives were to quantify and describe the extent of avian community change; relate observed species assemblages to changes in vegetation type and land-use class; and identify species with specific associations to either survey era. Using nonparametric ordination, we assessed whether the distribution of the avian community within the environmental space varied temporally. We used blocked indicator species analysis to identify species with specific association to historical or modern survey eras. Nonparametric permutation procedures identified both plot, each representing a different major habitat type, and year as significant factors defining communities. Year was more strongly related to the second ordination axis than the other environmental variables examined. Ordination of survey sites in species space confirmed modern avian communities were significantly different from their historical counterparts. Bird species richness increased and community composition changed by at least 50% despite the physiognomic characteristics of each plot’s habitat and the surrounding landscape remaining comparatively stable. Because of this, we conclude time trumped, or was more strongly associated with community change, than alterations in physiognomic habitat. These results suggest avian communities are naturally dynamic even in areas with relatively stable habitat conditions. Predictions that only address changes in climate, land use, or vegetation cover type may fail to predict changes in avian community composition. Historical datasets are a valuable means of understanding real-world changes in avian communities. We suggest additional factors, such as vegetation structure and microhabitat, may predict fine-scale shifts in avian species assemblages. Nevertheless, major changes in community composition occurred with apparently minor shifts in physiognomic habitat characteristics in our study area.
Key words: birds; community composition; habitat change; historical data; indicator species; non-metric multidimensional scaling; ordination; species assemblages; turnover; Willamette Valley. Received 19 September 2016; revised 20 September 2016; accepted 22 September 2016. Corresponding Editor: W. Alice Boyle. Copyright: © 2016 Curtis et al. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. † E-mail:
[email protected]
Introduction
extents (Wiens et al. 2009). Species–environment relationships, defined as the association between environmental characteristics and species distributions, possess a degree of dynamism influenced by the specificity of individual species’ habitat preferences and fluctuations in
Pressure from anthropogenic disturbance and global climate change demonstrate a growing need to understand how biodiversity will respond on multiple geographic and temporal v www.esajournals.org
1
November 2016 v Volume 7(11) v Article e01575
Curtis et al.
habitat characteristics over time. However, these dynamics often have not been characterized well. Historical datasets provide a means of examining changes in species–environment relationships over time. Paired with past environment and habitat characteristics, historical datasets expand our understanding of temporal variability in species distributions (Tingley and Beissinger 2009). Considering some ecological processes are only evident after multiple decades (Collins 2001), long-term datasets are instrumental to gauge ecosystem change adequately. Yet, few studies compare against data more than two decades old (see reviews in Igl and Johnson 2005, Tingley and Beissinger 2009). Even fewer studies utilize historical data from the western United States or Canada (Diamond 1969, Martin et al. 2004, Kotliar et al. 2007, Rotenberry and Wiens 2009, Shultz et al. 2012). Birds are useful indicators of environmental change because of their relative ease of detection, high mobility, and responsiveness to habitat variation (Temple and Wiens 1989, Crick 2004). Diversity indices and individual species abundance trends are frequently used to assess long-term changes in avian communities (Holmes and Sherry 2001, Parody et al. 2001, Catterall et al. 2010, Curtis and Robinson 2015). However, simple comparisons of abundance, richness, or diversity may overlook additional changes in bird–habitat associations. Evidence suggests a mix of ecological and biological factors creates dynamic, individualistic responses to habitat change that are frequently unpredictable (Kokko and López-Sepulcre 2006, Beale et al. 2008). Historical habitat conditions, site fidelity, and changes in resource availability influence avian responses to environmental variation (Knick and Rotenberry 2000, Hitch and Leberg 2007, Devictor et al. 2008). Because many factors contribute to the species and abundances of birds within an area, comparisons of diversity alone may not fully depict how bird communities change. Together, historical data and nonparametric ordination can provide new perspectives on long-term community change. Nonparametric comparisons of species assemblages grouped by year allow a holistic, similarity-based assessment of changes in community composition between periods. Modern ordination techniques allow v www.esajournals.org
ecologists to test whether groups of species distribute themselves differently within the environmental space along gradients of time and/or space. Blocked indicator species analysis (ISA) identifies species characteristic of a given survey period after accounting for differences between sites. Changes in indicator species between survey eras may exemplify shifts in the structure and composition of the regional community, as well as species with salient detection differences between observers. We evaluated the nature of avian community compositional change along environmental and temporal gradients using a rare, highly detailed, historical avian dataset (Eddy 1953). These bird survey data are uniquely valuable to understanding how avian communities in the Willamette Valley changed in the past 60 years. We resurveyed sites from this historical dataset to: (1) quantify and describe the extent of avian community change between 1952 and 2013, (2) relate observed changes to changes in vegetation type and land-use classes, and (3) identify species with specific associations to either survey era. Using nonparametric ordination, we assessed whether the distribution of the avian community within the environmental space varied temporally. We used blocked ISA to identify species that differentiate historical and modern surveys among sites. We also explored the effects of detection type and rare or poorly detected species on the observed species assemblages. Ultimately, this research provides insight into long-term variation in avian communities in the face of limited broad-level habitat change and anthropogenic activity.
Methods Study area and data collection
In 1952, Richard Eddy surveyed birds at six sites within 20 km of Corvallis, Oregon (Eddy 1953). His thesis was the first to provide detailed, repeated counts of local summer birds in the region. Eddy’s sites ranged from 8 to 20 ha in size and were non-randomly selected to represent primary habitats in the region (Table 1). Sites and habitats included coniferous forest, dominated by Douglas fir (Pseudtsuga menziesi) with an understory of maple (Acer sp.) and alder (Alnus sp.). During historical surveys, the southern
2
November 2016 v Volume 7(11) v Article e01575
Curtis et al.
Table 1. Overview of study sites, including primary habitat type, area in hectares, mean elevation in meters across each site, both historical and modern species richness, and beta diversity measured in half-changes (βD). Site
Primary habitat
Coniferous Marsh Mixed deciduous
Coniferous forest Emergent wetland vegetation Open mixed-species deciduous forest (oak and ash) Oak forest with some conifers Flooded riparian forest
Oak woodland Willamette
Area (ha)
Elevation (m)
1952 Richness
2013 Richness
βD (1952–2013)
21.5 23.3 7.4
322.6 78.7 88.0
32 34 32
19 51 36
2.16 0.99 1.22
16.3 10.4
130.0 62.2
25 27
31 36
1.48 1.26
Notes: Species richness is the total number of species observed across five survey visits using visual detections only. βD r epresents average dissimilarity between 1952 and 2013. A βD value of 1 indicates one half-change, or 50% change in avian community composition, between years.
portion of the site was “burned over” and contained low shrubs and young Douglas fir (Eddy 1953). The marsh site consisted primarily of emergent vegetation such as cattails (Typha latifolia) and sedges (Cyperaceae family) bordered by hardwoods including oak (Quercus sp.), willow (Salix sp.), and Oregon ash (Fraxinus latifola). The mixed deciduous site was characterized by several hardwoods including Oregon ash and mature Oregon white oak (Quercus garryana) with a small creek running through the length of the site. White oak dominated the oak woodland site, with small patches of Douglas fir and thickets of maple and poison oak (Toxicodendron diversilobum). Finally, the Willamette river site was a floodplain forest of alder, Black cottonwood (Populus trichocarpa), maple, and some Douglas fir, with tall grasses. We could not relocate a sixth site, referred to as “brushy,” using Eddy’s descriptions and so omitted it from modern surveys. Eddy walked through a given site for 2 h between 05:00 and 10:00, recording the number and species of birds observed. He repeated this method 10 times for each site, starting the second week of June through late August (Eddy 1953). Eddy’s thesis contains abundances for every species from each survey, as well as individual species densities for each site across 10 visits. To our knowledge, these data are the oldest and most extensively detailed bird surveys available from the Pacific Northwest. Eddy also provided descriptions of dominant tree and shrub species at each site. Because a few present-day common but difficult-to-see species were absent from his counts, we suspect Eddy-recorded visual detections only (Curtis 2015). v www.esajournals.org
Modern avian surveys
We relocated each historical site as accurately as possible using Eddy’s site descriptions and aerial photographs of Benton County from 1956 (USDA FSA 1956). Maps of the original survey sites were not available. While we were able to identify the general location of the five remaining historical survey sites, Eddy’s accounts were insufficient to designate exact plot boundaries for two areas (Willamette river and coniferous). We created buffers of “likely area” around these two selected survey sites that included adjacent land of the same habitat as our plots and could have been included in the original surveys. We extended a 200-m grid within the site into the “likely areas” and conducted five 5-minute stationary point counts at each grid intersection within both designated sites and “likely areas.” We used Student’s t-tests to compare mean Shannon–Wiener diversity across all counts within the designated site and “likely areas.” These t-tests failed to reject the null hypothesis of no difference in mean diversity within the plot and adjacent similar habitat (for full details see Curtis 2015). Therefore, for either site for which placement was uncertain, there was no evidence our selected site location resulted in a different avian community than the adjacent region. We used ArcGIS (ESRI 2013) to overlay a 200-m2 grid along the longest axis of the designated site boundaries. This grid was used to improve the precision of spot-mapping data and to designate point count locations for the site placement comparisons above. One observer (JRC) spot mapped the relocated sites five times throughout the 2013 breeding season (mid-May to mid-July) using the protocol described in Bibby et al. (2000), which reflects the general survey procedure described
3
November 2016 v Volume 7(11) v Article e01575
Curtis et al.
the first five visits from the historical data, and commencing modern surveys earlier in the year, unduly influenced our results (see Curtis 2015 for additional details).
by Eddy. Spot-mapping data are directly comparable to the historical area search methods, but collect supplementary observation information for every bird detected. Beginning within 10 min of sunrise, the surveyor systematically walked the entire site from grid point to grid point, recording the geographic location, number, species, sex, detection method (visual or auditory), and any relevant behavior of all birds encountered. Surveys did not occur in rainy or windy conditions. An initial evaluation of Eddy’s data suggested avian abundances remained relatively stable across the first five visits to each site during the historical surveys (Curtis 2015). These five visits cover approximately the same date range as the modern surveys and fall within the average breeding season for most Oregon birds. Eddy’s late summer surveys occurred after the main breeding season of many species, and his late- season counts included family groups and/or occurred after some species depart their breeding territories. There is no evidence to suggest the phenology of breeding birds in this region shifted over the past 60 years. Therefore, it is unlikely restricting the data comparisons to
Environmental variables
Eddy provided only superficial descriptions of vegetation in his study sites, but we located aerial photographs that depicted his study areas. We collected supplementary habitat data for each site (Table 2). In this study, we consider habitat characteristics on a broad, or physiognomic, level rather than fine detail of plant species and vegetation community composition. To evaluate differences in physiognomic habitat for each site over time, we quantified changes in percent cover of land use and vegetation types between historical and modern surveys using high- resolution digital scans of aerial photographs from Benton County (Scale 1:20,000; USDA FSA 1956) and 1-m resolution satellite imagery of Oregon (USDA FSA 2012). Scanned aerial photographs were georeferenced in ArcGIS (ESRI 2013) to align with satellite images. We assessed general vegetation maturation by comparing tree sizes and coverage, as well as how much of the
Table 2. Abbreviated names and descriptions of environmental variables used. Cover variable
Type
Description
Elev Habitat Size Year Deciduous Emergent Evergreen Grassland Mixed Pasture ShrubScrub WoodyWetland Urban Water Agri CoolForest Disturbed FloodForest GrassShrub WarmForest WetMeadow
General General General General Land use Land use Land use Land use Land use Land use Land use Land use Land use/veg Land use/veg Vegetation Vegetation Vegetation Vegetation Vegetation Vegetation Vegetation
Mean elevation in feet Categorical variable of habitat type (one per site) Area of site in square meters Year of survey Percent cover of deciduous forest Percent cover of emergent perennial herbaceous wetlands Percent cover of evergreen forest Percent cover of grass vegetation Percent cover of mixed forest types Percent cover of land used for livestock grazing purposes Percent cover of shrub-type and scrub vegetation Percent cover of forested marsh, swamp, and/or wetland Percent cover of urban development (buildings, roads, etc.) Percent cover of open water, both running and still Percent cover of land used for agricultural crops Percent cover of cool, upland, and coniferous forest Percent cover of recently disturbed open land Percent cover of riparian and flooded forest Percent cover of mixed grassland and shrub Percent cover of warm, temperate, and lowland forest Percent cover of freshwater wet meadow and marsh
Note: “Type” indicates whether the variable referred to a specific land-use or vegetation class, or was a general quantitative variable.
v www.esajournals.org
4
November 2016 v Volume 7(11) v Article e01575
Curtis et al.
underlying shrub and herbaceous vegetation was visible through the canopy on our aerial and satellite imagery. We used heads-up digitizing in ArcGIS to create a vector map of visually classified land use and vegetation types within 150 m of each site for both survey eras based on observable physical characteristics of the vegetation. We classified the vegetation type of each polygon to the formation level using the US Nati onal Vegetation Classification System (Federal Geographic Data Committee, Vegetation Sub committee 2008). Therefore, vegetation type represents “a combination of general dominant and diagnostic growth forms reflecting global macroclimatic factors” (Federal Geographic Data Committee, Vegetation Subcommittee 2008). We classified the land use of each polygon using classes from the National Land Cover Database (Homer et al. 2015). We used ArcGIS (ESRI 2013) to calculate mean elevation, exact area in square meters, and percent cover of each land use and vegetation type classification for each site (see Appendix S1). This approach circumvented the limitations associated with computer-based modeling of land cover over fixed cells. Computer-based land cover models, such as the National GAP Land Cover Dataset, are generated at coarse scales (0.4 ha minimum) with only moderate resolution (30 m; US Geological Survey 2012). Computer- based mapping of land cover can miss small extents of patchily distributed habitats including riparian areas and wetlands (US Geological Survey 2012), both of which were present in our study area and considered important for analyzing habitat change over time. In addition, because computer-modeled land use and vegetation information is not available for the 1950s, generating freehand polygons for both survey periods provided us with physiognomic habitat data that were directly comparable between years.
species in each survey period, using the best fitting model from historical species abundances across all site visits. We selected the best fitting model from three options (no covariates, date covariate, and habitat covariate) based on AIC values using package unmarked (Fiske and Chandler 2011). Detection probabilities for the modern dataset were modeled using visual-only detections as well as all detection types (Appendix S2; see Curtis 2015 for additional details). Detection probabilities were used to confirm there were no salient differences in species detectability between years. We did not adjust abundances by detection probabilities. Our response variable was the mean abundance of each species per site for a given year. Mean species abundance per site ranged from 0.1 (several species) to 213.4 (Red-winged blackbirds, Agelaius phoeniceus). The initial species dataset contained response variables for all species visually detected at 10 sampling units (five historical sites and five modern resurveys). A secondary species dataset contained the mean abundances for species at the 10 sampling units observed using all detection types. Because the inclusion of rare or insufficiently detected species could influence conclusions, we created a tertiary dataset in which we removed bird species that were poorly fit by the detection probability models (Appendix S3). The matrix of habitat variables contained values for mean elevation, area in square meters, and percent cover of each vegetation type and land-use class, as well as categorical variables for year and site. Data analyses and transformations were conducted in PC-ORD 6.12 (McCune and Mefford 2011). We first removed rare species, defined as those species detected in less than two surveys for a given year. This reduced noise produced by species that likely did not occupy the study sites during the breeding season. We used rank-abundance curves to evaluate changes in abundance and diversity of species at each site between survey years. Rank-abundance curves indicate changes in community composition through differences in the total number of ranks, or species, and the slope of the curve, which represents evenness. We then transformed the data to improve linearity of relationships and decrease emphasis on highly abundant species. We used a square
Statistical analysis
Differences in observer ability and changing survey conditions may influence the probability of detecting species (Tingley and Beissinger 2012). To evaluate changes in detection prob abilities between survey eras, we used package “unmarked” (Fiske and Chandler 2011) in program R (R Core Team 2013) to calculate simple estimates of detection probability for each v www.esajournals.org
5
November 2016 v Volume 7(11) v Article e01575
Curtis et al.
root transformation on the response variables, achieving a result similar to a log transformation but without requiring special treatment of zeros and mean abundances less than one (McCune et al. 2002). Relativizations were not performed for either rows or columns of the community matrix because sampling effort was considered approximately equal among sites and among years and because we wished to preserve the differences in abundance between common and rare species. To identify potentially influential outliers among the sampling units, we examined the average distances of sampling units from the grand mean of distances between sampling units using Sørensen distance (McCune et al. 2002). No sample units were identified as outliers. To test for within-site differences between years, we used permutational multivariate analysis of variance (PerMANOVA; Anderson 2001) using Sørensen distance, the fixed group as year, a random factor for sites as blocks (randomized block design), and 4999 permutations. PerMANOVA provides quantitative measures for the effects of site and year in nonparametric data via pseudo-F statistics and P-values from permutation tests. Because of the small sample size, we decided to seek additional evidence with a similar test: blocked MRPP (MRBP; Mielke 1984). Blocked MRPP answers the same questions as PerMANOVA, and is also nonparametric, but uses a different test statistic (the average within-group distance) and asymptotic evaluation of the P-value. Blocks were defined by sites, groups were defined by year, and we selected median alignment within blocks. As Sørensen distance is incompatible with MRBP, Euclidean distance was used (McCune et al. 2002). Additionally, we calculated the amount of compositional change in bird species for each site as half-changes (βD) corresponding to the average dissimilarity between years, using Sørensen distance as the dissimilarity coefficient, D. A βD value of 1 represents 50% dissimilarity in community composition between years at a single site. Average half-changes improve the scale for community dissimilarity, as they linearize the relationship between the selected distance coefficient and changes in diversity over space or time (McCune et al. 2002). We used blocked indicator species analysis (ISA; Dufrêne and Legendre 1997) to identify bird species that best represented individual sites v www.esajournals.org
within years. Blocked ISA utilizes proportional species abundances and frequencies between blocks (sites) within a priori groups (year). Results include indicator values (IV) between 0 and 1 for each bird species in the community. These IV scores reflect the strength of association between species and groups based on concentration of abundance and faithfulness within groups. We identified blocked indicator species for birds with visual detections only, after removing rare species and performing a square root transformation on the data. To test the statistical significance of the IV scores, we performed analyses for 4999 random permutations and obtained a P-value for each IV. Species with high observed IV scores and significant P-values (0.2 are displayed as a joint plot with vector lengths corresponding to r2 along both axes. See Appendix S3 for species common and scientific names and codes. v www.esajournals.org
8
November 2016 v Volume 7(11) v Article e01575
Curtis et al.
melanocephalus). Within the rank-abundance curves, these species generally experienced notable shifts in rank, abundance, or both rank and abundance across all sites (Fig. 1). Finally, we compared NMS scores to test whether removing poorly detected birds and non-visual detections influenced results. Mantel tests for three comparisons of NMS scores found strong redundancy among ordinations. (91.4% between ordinations of including and omitting non-visual detections; 91.6 between the ordinations of visual detections and the same matrix with poorly detected species removed.) Therefore, omitting rare or poorly detected birds, and non-visual detections did not significantly influence NMS results. Among NMS ordinations, the configuration derived from average abundances for visual detections with only rare species removed had the lowest final stress (final stress = 4.904, P = 0.02); therefore, we used that species matrix to draw conclusions.
Table 3. Strongest Pearson (r) and Kendall (τ) correlation coefficients between environmental variables and the two-dimensional NMS configuration of sampling units in species space. Cover variable
Axis
r
τ
CoolForest Elevation Evergreen Emergent Wetlands Grassland Year
1 1 1 1 1 2
0.788 0.773 0.743 −0.874 −0.533 0.886
0.762 0.707 0.609 −0.835 −0.365 0.745
Note: See Table 2 for a description of each cover variable above.
with shrub and forested habitat. Both the modern marsh and Willamette river sites shifted along the first axis in the opposite direction; in modern surveys, these sites decreased in percent cover of grass. Based on the second axis, the coniferous site experienced the largest amount of change in bird species assemblage, followed closely by the oak woodland site. Separation of sites was greater along the second ordination axis, characterized by time, than along the first axis, which differentiated sites based on habitat types. Thus, it appears time depicted long-term avian community change in our study better than the measured habitat variables. One of our objectives was to identify species that distinguished historical and modern survey periods. When we performed blocked ISA among sites grouped by year, we found several significant indicator species (P
