Erosion of tropical bird diversity over a century is influenced by abundance, diet and subtle climatic tolerances

Human alteration of landscapes leads to attrition of biodiversity. Recommendations for maximizing retention of species richness typically focus on protection and preservation of large habitat patches. Despite a century of protection from human disturbance, 27% of the 228 bird species initially detected on Barro Colorado Island (BCI), Panama, a large hilltop forest fragment isolated by waters of Gatun Lake, are now absent. Lost species were more likely to be initially uncommon and terrestrial insectivores. Analyses of the regional avifauna, exhaustively inventoried and mapped across 24 subregions, identified strong geographical discontinuities in species distributions associated with a steep transisthmian rainfall gradient. Having lost mostly species preferring humid forests, the BCI species assemblage continues to shift from one originally typical of wetter forests toward one now resembling bird communities in drier forests. Even when habitat remnants are large and protected for 100 years, altered habitat characteristics resulting from isolation produce non-random loss of species linked with their commonness, dietary preferences and subtle climatic sensitivities.


Species Traits
For resident non-aquatic bird species in this study, we assigned preference for one of four habitat categories: open areas with little to no woody vegetative cover; edge habitat at the boundary of low, woody vegetation; the outer margins of forest of any age (forest edge); and interior of forest of any age. Residency status was classified as permanent resident or breeding migrant (i.e., seasonally occurring nesting species).
For species detected on BCI during any inventory period, we considered an additional six categorical and two continuous attributes previously associated with extinction risk in tropical birds 1,2 . Abundance on BCI was categorized based on estimated total island-wide population at the time of isolation 3 : common (> 100 individuals); occasional (10-100 individuals); or rare (<10 individuals). Regional abundance was considered but ultimately omitted due to strong correlations with several other variables.
Artificial nest experiments in tropical forest fragments, including on BCI, posited that the absence of large predators and subsequent increase in middle-sized nest predator abundance was responsible for avian extinctions, particularly among ground-dwelling species [4][5][6] . However, recent experimental and observational evidence suggests artificial nest studies lack sufficient external validity [7][8][9] and traditional theories about the identities of primary nest predators in tropical forests appear misinformed [10][11][12] .
To further evaluate the extent that nest attributes contribute to fragmentation sensitivity despite the lack of observational evidence in support of increased depredation rates by mesopredators, we included typical nest height and type as potential predictors of extinction risk on BCI. We assigned one of four possible categories for a species' typical nesting height: ground (<1m); understory (1-5m); mid-story (>5m but below the canopy); or canopy (the top level of vegetation regardless of height). We used three categories for nest type: open cup (bowl, platform, or scrape); enclosed, roofed nests with a single entrance (e.g. pendulum, pyriform, or pouch); or cavity nests in trees, burrows, or termite mounds. Obligate brood parasites were assigned the nest attributes of their most common hosts.
We established six dietary guilds according to a species' primary food source 13 : carnivores (vertebrates, snails, carrion, and occasional large arthropods); frugivores (fruits of any size); granivores (seeds and nuts); insectivores (insects, arthropods, and occasional small vertebrates); nectarivores (flower nectar); and omnivores (generalists which consume food from more than one category). We compiled four categories for the typical height at which a species searches for food: terrestrial (<1m); understory (greater than >1m but below midstory); arboreal (uppermost layers of vegetation, including midstory and canopy, regardless of height); and raptorial hunters that pursue prey across all forest strata. Aerially foraging birds, such as vultures, swifts, swallows, and nighthawks, were omitted because their daily foraging ranges extend well away from BCI.
Our two continuous species attributes were body mass and southern limit, an index of climatic tolerance. Body mass was the log-transformed mean individual weight across sexes from 14 . We used a species' southern distributional limit in the Canal area -the integer linear distance between a species' southernmost occurrence on the Panama isthmus and the Pacific entrance to the Panama Canal -as an index of climatic tolerance.

Outlier Analysis
Outlier analysis using Sørensen distance identified one sampling area, Nueva Providencia (PRO), with a distance value more than 2 standard deviations greater than the mean distance between sites (SD = 3.5). PRO is isolated on the northeastern edge of Gatun Lake and not adjacent to any other evaluated subregion (Figure 1). This subregion contained low avian species richness relative to the average for non-urban mainland areas (120 species vs. mean richness of 198) and exhibited peripheral placement with repulsion on at least one axis in preliminary ordinations. PRO did not possess extreme or unusual values for any environmental variables. We believe this subregion was identified as an outlier because it was the only sampling unit on the northeastern side of the Canal area and, without nearby avian communities of similar species composition, was not as easily ordinated in the context of the other subregions. Additional data from this general region would likely reduce the significance of PRO as an outlier. This sampling unit represents an area for which we have no other avian inventory data from predominantly forested habitat. Because we considered PRO to be within our target population, its bird inventory was as complete as other subregion inventories 15 , and we have no reason to doubt the accuracy of its data, we retained this subregion for analysis.

Likely Species
Several difficult to identify or detect species, including nocturnal birds and small upper-canopy passerines, were missing from earlier datasets, likely due to observer inexperience with tropical bird vocalizations 16 or lack of access to all parts of the island. To evaluate the effect of these common but poorly detected species (AKA "likely species"), we repeated ordination procedures on a reduced species matrix where 21 species we deemed were likely present during historical surveys but not reported were removed (Table S1). To quantify the effect of likely species on the ordination, we compared NMDS ordination scores between the primary and reduced species datasets using Mantel's asymptotic approximation method with a randomization test for 999 runs. This evaluates the null hypothesis of no correlation between distance matrices for the same sampling units. 17 . The standardized Mantel statistic (r) ranges from 0 to 1, with larger values representing higher correspondence between two ordinations. We calculated the percentage redundancy of the ordinations as r 2 multiplied by 100. We found strong agreement between ordinations including and omitting these species (r = 0.991). We conclude missed detections resulting from historical observer inexperience or unfamiliarity with a few tropical bird sounds did not appreciably influence NMDS results.

Urban Subregions
BCI has <0.02% urban land cover surrounding a scientific research station, and is otherwise forest, disturbed only by natural wind events. Mainland subregions along the Panama Canal range from <1% to over 97% urban land cover. Urbanized areas in this region have outsized effects on habitat and avian community composition. Sampling units with one or more large cities within their boundaries, despite having as little as 5% total urban cover, generally contained depauperate bird communities in small, remnant patches of degraded forest amidst large zones of species-poor anthropogenically altered habitat 15 . Our objective was to evaluate changes in the BCI bird community relative to comparable mainland forest habitats and bird communities. Because even lightly urbanized areas in central Panama appear to experience different structuring mechanisms than wholly forested areas, we removed subregions containing major cities with >5% urban cover from our analyses. Selective logging, fire, and other forms of disturbance common elsewhere in the tropics are rare within the forests of the Canal area subregions retained for analyses. Figure S1. Southern Panama Canal area range limits for extinct species on BCI. Grouped by date of last reported observation and habitat association, with best-fit linear regression lines between southern limit and decade of extinction (dashed lines). Small shapes are individual extinctions, large shapes represent individual decade means for each habitat type. Figure S2. Proportion of transisthmian and wet-forest birds extinct on BCI across all forest-associated species as well as broken down by core forest interior and forest edge association. Transisthmian birds occur in subregions along the entire Canal area, while wet-forest species occur exclusively in forests with >2000mm precipitation annually. Extinct birds are any species considered to have once been a permanent breeding resident that has gone undetected for a least a decade and has not demonstrated the capacity to re-establish breeding populations on the island post-isolation. Table S1. Common names of "likely species"difficult to detect species missing from species inventories, likely due to observer unfamiliarity with these species, lack of nocturnal surveying, or inability to access all parts of Barro Colorado Island. See Table S3 for corresponding scientific names.

Supplementary Tables and Figures
* Denotes species likely present on BCI at the time of isolation but not detected before 1950.  Table S2. Environmental factors used to characterize avian community structure along the Panama Canal corridor. Full descriptions of these environmental factors can be found in Rompré et al. 15 .

Variable Description Source
Forest Age Categorical. Represents dominant relative forest age (1=secondary; 2=mature secondary; 3=primary mature forests) estimated using ANAM (2003) criteria for species composition and disturbance history. 18 Altitude Maximum altitude per subregion in meters above sea level. 15 Area Total subregion area in km² 15 % Forest Proportional forest cover, calculated by dividing forested area by total area in each subregion. 15 % Unfragmented Degree of fragmentation, represented by percent total forest area included in 1 or 2 largest fragments within subregion. 15 % Urban Proportion urban cover, calculated by dividing urban area by total area in each subregion.

19,20
Plant Richness Plant and tree species richness within 1 ha plots. Plant data provided by Pyke et al. (2001). Trees defined as woody plants >10cm dbh. Kriging used to interpolate geographic plant richness for each subregion centroid 15 .

18,21,22
Precipitation Mean annual precipitation in mm for subregion, obtained from both ACP and atlases. For subregions without precipitation data, values obtained by interpolation from isohyets available for that period 15 .   Table S4. Residency status of non-aquatic, non-aerial species first detected on BCI after 1950 sorted by decade of first detection. Residency categories include "vagrant/flyover": species unlikely to maintain stable, resident breeding populations on BCI; "ephemeral": transitory breeders that maintain only intermittent breeding populations; "expanding": birds experiencing range expansions along the canal but not yet permanent residents of BCI -often urban associated; and "colonized": new species with stable, permanent breeding populations. See Table S3 for corresponding scientific names.