Contrasting changes in the abundance and diversity of North American bird assemblages from 1971 to 2010

Bird monitoring data

The North American Breeding Bird Survey (BBS) is a cooperative effort of the United States Geological Survey, the Canadian Wildlife Service, and, since 2008, the Mexican National Commission for the Knowledge and Use of Biodiversity, with the aim to monitor the status and trends of North American bird populations (Sauer et al., 2013). Initiated in 1966, the BBS monitoring program has been collecting yearly population counts from an increasing number of roadside routes. Each roadside route has a length of 24.5 miles (approximately 40 km), along which fifty regularly spaced sites are monitored each year in June. At each site, the observer records all birds heard or seen within a 400 m radius and within 3 min (Sauer et al., 1997; Rittenhouse et al., 2012). For this study, we used monitoring data from the conterminous United States from 1971 through 2010, thus excluding the early years of the survey, when relatively few routes were visited (Illan et al., 2014). The BBS data comprise raw counts, which are a function of both population size and detection probability, the latter being dependent on characteristics of routes, observers, and meteorological conditions (Phillips et al., 2010; Rittenhouse et al., 2010, 2012). To obtain a spatially consistent dataset covering the entire study period, we selected routes with at least one observation in each 5-year interval. Then, we reduced sampling variation in abundance induced by observed and weather effects by calculating 5-year average abundances per species per route (Illan et al., 2014). This resulted in a dataset of 768 routes each with eight 5-year average abundances of in total 519 species. A list of species is provided in the Supporting Information (Table S1).

Biodiversity metrics

We calculated 12 biodiversity metrics for each monitoring route and for each of the eight time periods. The set of metrics was selected to represent abundance, taxonomic diversity, functional diversity, and phylogenetic diversity (Table 1). Because temporal trends may differ between or be driven by particular species groups, we calculated the metrics not only for all bird species together, but also for four subgroups based on breeding habitat preferences (Table S1): grassland (27 species), woodland (129 species), wetland (83 species), and shrubland (81 species). Breeding habitat preferences were derived from the Patuxent Bird Identification Infocenter (Gough et al., 1998). This database also distinguishes species using urban breeding habitats; however, there were too few urban breeders in our dataset to perform meaningful metric calculations. Urban breeders (n = 13) were therefore not included in the separate breeding habitat group analyses, but they were included in the overall assessment. To assess the sensitivity of our findings to occasional observations of rare or vagrant species, we performed the metric calculations and subsequent analyses also based on only those 439 species that were observed consistently throughout all eight 5-year intervals.

Table 1. Overview of the biodiversity metrics calculated in this study

Biodiversity metric		Description a
a ai is the abundance of species i; pi is the proportional abundance of species i; n is the total number of species in the dataset; c is a constant.
Total abundance	TA	∑ i = 1 i = n a i
Geometric mean abundance	GMA	∏ i = 1 i = n ln ( a i + c ) 1 / n
Species richness	SR	Total number of species present
Shannon index	Shan	exp ( − ∑ i = 1 i = SR p i · log p i )
Simpson index	Simp	1 ∑ i = 1 i = n p i 2
Functional richness	FRic	The convex hull volume of the individual species in multidimensional trait space (Villéger et al., 2008).
Functional evenness	FEve	The regularity with which species abundances are distributed along the minimum spanning tree which links all the species in the multidimensional functional space (Villéger et al., 2008).
Functional divergence	FDiv	Species deviance from the mean distance to the center of gravity within multidimensional trait space, weighted by relative abundance (Villéger et al., 2008).
Functional dispersion	FDis	The weighted mean distance in multidimensional trait space of individual species to the centroid of all species. Weights are species’ relative abundances (Laliberté & Legendre, 2010).
Proportion of carnivores	PC	The abundance of carnivorous species (i.e., species with a diet consisting of at least 60% meat, fish, and/or carrion) relative to the total abundance of all species (%).
Community-weighted mean body mass	CWMm	The mean body mass of species weighted by the species’ abundances.
Phylogenetic diversity	PD	The total length of the phylogenetic branches connecting all species of a given assemblage (Faith, 1992).

Biodiversity changes

We quantified the temporal changes in the biodiversity metrics in two ways. First, we calculated the overall temporal change for each metric per route, that is the change from the first to the last 5-year interval, as

ESM,x=Mx,t=8−Mx,t=10.5·(Mx,t=1+Mx,t=8)(1)

where ES_M,x represents the effect size for biodiversity metric M at route x, and M_x,t=1 and M_x,t=8 are the values calculated for biodiversity metric M at route x for the intervals 1971–1975 and 2006–2010, respectively. This effect size measure yields a symmetrical index of decrease or increase between −2 and 2 (Böhning-Gaese & Bauer, 1996; Van Turnhout et al., 2007). Second, we estimated the temporal trends in the metrics across all routes throughout the eight 5-year intervals. To that end, we first calculated mean metric values over all routes for each 5-year interval (Table S2). To facilitate comparison among the metrics, we then standardized the metric values over time (zero mean and unit variance). Because we were interested in the overall direction and strength of the trends, we fitted ordinary least squares (OLS) regression models to the standardized metric values. Note that this approach results in slopes and intercepts identical to those resulting from an approach where OLS models are first fitted for each route and slopes and intercepts are then averaged over the routes. For comparison, we also fitted generalized least squares (GLS) models with temporally autocorrelated error structure (AR1), to account for possible nonindependence of observations closer in time. GLS models tend to have slopes similar to those of OLS regression models, yet GLS models have more conservative P-values because the temporal autocorrelation is accounted for (Dornelas et al., 2013).

Key metrics

To quantify the redundancy and complementarity among the changes in the various metrics, we applied a combination of principal component analysis (PCA), variable reduction, and cluster analysis. We did this based on the overall changes of the metrics across the routes (i.e., the effect sizes, Eqn (1)). First, we determined the dimensionality of the biodiversity changes with PCA, based on Spearman rank correlations among the effect sizes (Lyashevska & Farnsworth, 2012; Stevens & Tello, 2014). To determine the number of nontrivial or significant principal components, we randomly permuted the effect sizes per metric, conducted a PCA, retained the eigenvalues, and repeated this procedure 1000 times to create distributions of eigenvalues for each principal component that would be expected by chance. If the eigenvalue of a principal component based on the original dataset was larger than the 95th percentile of the eigenvalues based on the randomized data, then we considered that particular principal component significant (Peres-Neto et al., 2005). Next, we clustered the metrics based on the similarity of their loadings on the significant components using a hierarchical clustering algorithm (Ward's method) based on Euclidean distance and we identified a number of clusters equal to the number of significant components. Finally, we identified which of the metrics were most representative of the changes. A common approach to identify the most representative variables in a set is to select those variables that have high loadings on the first (few) principal components. However, by considering one principal component at a time, this approach may lead to a suboptimal or larger subset of the original variables than is strictly necessary (Cadima & Jolliffe, 2001). To avoid this, we used the ‘improve’ algorithm (with criterion ’Rm‘) from the ‘subselect’ package (version 0.12-5), which is specifically designed to identify which variables are most representative of the total variation in a dataset (Cadima & Jolliffe, 2001). As a benchmark for the number of key metrics to retain, we used the number of significant components. All statistical analyses were performed in r, version 3.0.3 (R Core Team, 2014).

Biodiversity changes

The biodiversity changes showed considerable spatial variation, as exemplified by the effect sizes ranging from −2 to 2 (i.e., the minimum and maximum values possible) for several of the metrics (Figs 1 and 2). However, when averaged over all routes, and when considering all species together, we found the majority of the metrics to increase over the study period (Fig. 3, Table S2–S4). Pronounced increases were observed in particular for the proportion of carnivores (which, on average, more than doubled from 1.3% in 1971–1975 to 2.8% in 2006–2010), the functional richness (which increased by nearly 60% over the study period), the community-weighted mean body mass (which increased from 100 to 150 g), and the total abundance of birds in quartile 3 (from 28 to 36 individuals). In contrast, distinct decreases were observed for the total abundance of all birds (from 901 individuals in 1971–1975 to 804 individuals in 2006–2010) as well as the abundance of the most common species (from 868 to 762 individuals). Highly similar results were obtained when excluding species that were not consistently observed throughout all 5-year intervals (Figs 1 and 3, Table S2–S4). Yet, changes were clearly different between the breeding habitat groups (Figs 2 and 3). Grassland birds showed, on average, decreases in various metrics, including the total abundance, geometric mean abundance, species richness, and phylogenetic diversity, whereas these metrics increased for woodland birds. Wetland birds tended to decrease in abundance, although not as sharply as the grassland birds. Yet, this group showed increases in various measures of both taxonomic and functional diversity, including Shannon index, Simpson index, functional richness, functional evenness, functional divergence, proportion of carnivores, and community-weighted mean body mass. For shrubland birds, the most pronounced change was a clear decrease in the community-weighted mean body mass.

Key metrics

The PCA yielded three significant principal components for all species together and for three of the four breeding habitat groups. For the shrubland breeders, four principal components were retained (Table S5). The cumulative proportion of variance explained by the nontrivial components was between 59% and 63%, with the first component explaining between 30% (shrubland breeders) and 37% (woodland breeders) of the total variance (Table S5). Clusters of metrics slightly differed among the species groups and with the data transformation applied to calculate the geometric mean abundance (Figs 4 and 5, Fig. S2). Yet, particular metrics were in all cases closely associated with each other, such as the total abundance and the abundance of the most common species, or the Shannon index and the Simpson index. Metrics depending on the number of species or traits without any abundance weighting (species richness, phylogenetic diversity, and functional richness) were always in the same cluster. In many cases, the richness metrics were also in the same group as the abundance of the less common species, reflecting that changes in the abundance of those species in particular coincided with changes in species composition. Finally, irrespective of the species group, the nonredundant set of metrics most representative of the overall changes included a metric of overall abundance (mostly total abundance), a richness metric (species richness, phylogenetic diversity, or functional richness), and a metric relying on both richness and evenness (Shannon index) (Table 2). These results did not change with the data transformation applied to calculate the geometric mean abundance.

Biodiversity trends

Based on long-term roadside monitoring data of the North American BBS, we found contrasting trends between the overall abundance of birds and the diversity of assemblages. Among the most pronounced trends was a distinct decrease in overall bird abundance, mainly driven by declines of the most abundant species. This finding is in line with the results of a recent study in Europe (Inger et al., 2015), indicating that declines of common bird species constitute a widespread phenomenon. In our dataset, we observed strong declines mainly for common grassland breeders, like eastern meadowlark (Sturnella magna), as well as highly abundant generalists, including the common grackle (Quiscalus quiscula), common nighthawk (Chordeiles minor), chimney swift (Chaetura pelagica), and house sparrow (Passer domesticus) (Table S1). Declines of generalists as well as farmland birds have been reported before, in both North America and Europe (Donald et al., 2001; De Laet & Summers-Smith, 2007; Reif, 2013). Agricultural intensification has been identified as a main driver, for example through increased drainage of grasslands, increased livestock densities, and increased use of pesticides, which reduce food availability for aerial insectivores in particular (Donald et al., 2001; Newton, 2004; Reif, 2013, North American Bird Conservation Initiative U.S. Committee, 2014). Trends may have been amplified by farmland abandonment in less productive or remote areas, which occurred across much of eastern North America (Flinn & Vellend, 2005). Forest regrowth in these abandoned farmlands may, in turn, explain why the overall abundance of woodland birds has increased, in contrast to the other habitat groups (Figs 2 and 3).

In contrast to the decline in overall abundance, we found various metrics to remain stable or increase over the study period (Fig. 1-3). Because species richness, phylogenetic diversity, and functional richness are derived from species composition without any abundance weighting, the overall increases in these metrics indicate that the assemblages have been subject to colonization by new bird species with new combinations of traits or distinct phylogenies (Mouillot et al., 2013). Underlying factors may include changes in habitat characteristics, shifts in species ranges, for example due to climate change, species recovery due to targeted conservation actions (e.g., forest restoration within the Conservation Reserve Program), or changes in observer skills (Böhning-Gaese & Bauer, 1996; Buckland et al., 2005; Van Turnhout et al., 2007; Rittenhouse et al., 2012; Reif, 2013; Inger et al., 2015). We cannot rule out the possibility that the proficiency of the observers has changed over the years. Yet, the BBS monitoring protocol is highly standardized, and given the large spatial and temporal scale of our analysis, we see no particular reasons for a directional observer bias.

All metrics other than overall abundance, species richness, functional richness, and phylogenetic diversity account for the proportional abundance of species or traits; hence, changes in these metrics may reflect species’ extinctions or colonizations, changes in abundance distribution, or a combination. For the grassland birds, we observed declines in species richness and the total abundance of the common species (TA.q4), yet no changes in the Shannon or Simpson index and even increases in functional evenness and functional divergence (Fig. 2, Table 2). This illustrates how the disproportionate decline of abundant species may yield positive trends in metrics that rely on evenness (Böhning-Gaese & Bauer, 1996). The overall increases in functional diversity and functional dispersion indicate that the declining common species are located toward the center rather than the edges of the functional trait space of the assemblages, whereas the reverse might hold for the less common species (Mouillot et al., 2013). Further, the increase in community-weighted mean body mass suggests a relative increase in species with slower life histories or larger body sizes over the past decades (Reif, 2013; Inger et al., 2015). Indeed, we found increases in abundance of various large-bodied species, including raptors and scavengers like hawks and vultures as well as wetland birds like geese, cranes, and cormorants (Table S1). This finding contradicts the generally positive correlation between body size and extinction risk (Gaston & Blackburn, 1995; Hilbers et al., 2016) and might at least partly be explained by targeted protection and conservation measures, of which these species may have benefited in particular (Van Turnhout et al., 2010, North American Bird Conservation Initiative U.S. committee, 2014).

In general, richness and total abundance are more likely to be positively than negatively associated (Bock et al., 2007; Hurlbert & Jetz, 2010). This seems at odds with the opposite overall trends in richness and total abundance as we observed for all bird species together. However, the negative association between richness and total abundance breaks down when looking at particular species groups (grassland or woodland breeders; Fig. 3) or at the monitoring route scale, where richness and overall abundance turned out to represent independent (uncorrelated) dimensions (Figs 4 and 5). These findings indicate that the declines in common species and increases in less common species occurred at different locations and in response to different possible drivers (agricultural intensification versus conservation measures and forest regrowth).

To summarize, our analysis of long-term North American BBS monitoring data revealed a considerable decline in the total number of birds over the past 40 years, which coincided with stable or increasing metrics of taxonomic, functional, and phylogenetic diversity. The stable or increasing diversity metrics, including increases in mean body size and the proportion of carnivores, indicate recovery of large-bodied and carnivorous species from previously low levels (‘rewilding’). Yet, the decline in total bird abundance may give rise to concern as a species’ contribution to ecosystem functioning is dependent not only on its traits, but also on its numbers (Inger et al., 2015). Given that the BBS is a roadside survey that covers birds only, our study does not allow to draw conclusions regarding interior habitats or other taxonomic groups. Nonetheless, our results on taxonomic diversity match three recent studies that found no net loss in local-scale taxonomic diversity based on large numbers of assemblage time series covering a variety of taxonomic groups (Vellend et al., 2013; Dornelas et al., 2014; Elahi et al., 2015). It remains to be investigated how these results relate to biodiversity changes occurring over larger spatial scales (gamma versus alpha diversity) as well as longer time frames.

Implications for monitoring biodiversity change

The contrasting changes we observed between various diversity metrics on the one hand and the overall abundance of birds on the other hand emphasize the relevance of a multifaceted approach to monitoring biodiversity change. Our results clearly show that an exclusive focus on richness and evenness metrics might not capture all relevant aspects of biodiversity change, because these metrics might simply miss out on or even respond positively to substantial losses of individuals (Böhning-Gaese & Bauer, 1996). Thus, increasing evenness should not be considered an unambiguous indicator of greater diversity, despite it being common to do so (Magurran, 1988; Purvis & Hector, 2000; Elahi et al., 2015). Our results further indicate that total abundance is more suited to capture losses of individuals than the geometric mean abundance, the latter being a composite measure of abundance and evenness and hence more sensitive to increases in the abundance or detectability of less common species.

Even a combination of metrics of total abundance, species richness, and the proportional abundance of species may not fully capture biodiversity changes, because species’ replacements may go unnoticed by these metrics (Buckland et al., 2005; Dornelas et al., 2014). Possible solutions are to consider changes in species composition (turnover) or to include metrics of functional and phylogenetic diversity, which might be more sensitive to environmental change (Winter et al., 2009; Mouillot et al., 2013). Indeed, for some species groups, the set of key metrics that we identified included functional richness or phylogenetic diversity rather than species richness, indicating that the former are, in some cases, more responsive to change (Table 2). Further, our results for the wetland and shrubland breeders suggest that the community-weighted mean body mass is also indicative of changes, as this metric may change considerably even when there is little change in species richness or evenness (Figs 2 and 3). Functional diversity metrics may become even more informative if more traits are included, in particular traits that are responsive to environmental change, such as migratory behavior (Van Turnhout et al., 2010). However, functional or phylogenetic diversity metrics require additional information (functional trait data, phylogenetic trees), which might be difficult to obtain in particular for taxonomic groups that are less well studied.

To summarize, we identified three main dimensions of biodiversity change (overall abundance, richness, and proportional abundance), consistent with McGill (2011), thereby observing opposing trends between overall abundance on the one hand and various diversity metrics on the other. This indicates that stable or even increasing metrics of richness or evenness may occur in parallel with substantial losses of individuals and supports the importance of population abundance as an essential biodiversity variable (Pereira et al., 2013). The abundance of each species at each site is a variable that can be used to derive all possible metrics of abundance and taxonomic diversity. If this essential biodiversity variable is combined with information on the species’ traits and phylogenetic positions, all other metrics used can be derived as well.