Avian community characteristics and demographics reveal how conservation value of regenerating tropical dry forest changes with forest age

Study sites

We studied bird communities during the northern winters of 2003–04 through 2007–08 in the buffer zone of Sierra de Bahoruco National Park, Dominican Republic where a growing human population and associated agricultural activities often conflict with park protection goals (Pasachnik, Carreras De León & León, 2016). Four study sites of >15 ha were established at elevations of ∼400 m within 2 km of the village of Mencia (18.176°, −71.737°). Sites were chosen based on age since disturbance (confirmed through interviews with landowners and long-term residents). The sites were originally cleared for agriculture (primarily beans, squash), and then utilized for occasional low-density grazing by small numbers of livestock. Remnant trees and living fences were characteristic of all sites. Each site contained two mist netting arrays, and contiguous blocks of habitat for at least 150 m on each side minimized possible edge effects. Because of their proximity to one another, all sites shared a similar landscape matrix including mature forest, shade coffee, mixed agriculture, pasture and village, so there were no differences in connectivity among these habitats for volant birds. All studies sites were within ∼200 m of a road and ∼50 m of intact forest (see Appendix S1 for an area map).

The study sites forming the chronosequence (with age since disturbance at initiation of study) were La Cueva (two years), La Caoba (five years), Morelia (ten years), and El Corral (20 years); see Appendix S2 for details of this chronosequence. A fifth site in mature dry forest studied as part of a prior project (Latta & Faaborg, 2002) during the winters of 1996–97 through 2000–01 consisted of two transects 2 km apart which were pooled in analyses. This site represented the native habitat of the Mencia sites prior to their deforestation and served as the endpoint in the chronosequence. This control site, Aceitillar (18.098°, −71.635°), was located ∼14 km east of Mencia and at a similar elevation (∼350 m). The Aceitillar site was also affected by occasional grazing by stray cattle and tree cutting for charcoal, but represents some of the best remaining dry forest in the region. Although logistical constraints prevented us from studying the Mencia and Aceitillar sites simultaneously, our extended, multi-year studies encompassed the range of weather conditions typical of the region, and likely ameliorated differences that may have arisen due to annual effects in shorter-term studies.

La Cueva (two years) and La Caoba (five years)

Vegetation at these sites is similar and dominated by native grasses, forbs, and woody shrubs in height categories <1.5 m, including Chrysophyllum oliviforme, Guettarda preneloupii, Chromolaena odorata, Trichilia hirta, Eugenia monticola, and Ehretia tinifolia. Trees are scarce but consist primarily of very young Senna spectabilis (an exotic species) and scattered remnant Bursera simaruba. These sites have the lowest diversity and density of shrubs of all sites, as well as the lowest mean canopy height (3.6 m; 3.9 m) and canopy cover (18%; 22%).

Morelia (ten years)

Vegetative cover occurs primarily in height categories <4.0 m. The diversity of native shrub species is the highest of all study sites, and includes Chrysophyllum oliviforme, Trichilia hirta, Psychotria berteroana, Piper aduncum, and Casearia aculeata. Trees are small and dominated by the exotic Senna spectabilis, but include scattered native Bursera simaruba, Ocotea coriacea, and Zanthoxylum martinicense. Mean canopy height is 5.2 m and mean canopy cover is 54%.

El Corral (20 years)

Shrub density and diversity decrease but include Trichilia hirta, Casearia aculeate, Eugenia monticola, Licaria triandra, and Nectandra hihua, all of which are native. Mean canopy height (6.4 m) and canopy cover (76%) increases with native trees that include Eugenia monticola, Nectandra hihua, Licaria triandra, Acacia farnesiana, Bauhinia divaricata, and Trichilia pallida, as well as the introduced species Leucaena leucocephala.

Aceitillar

The mature dry forest site has a mostly closed canopy (94%) with a mean tree height of 10.6 m, a few emergent trees, and an understory dominated by broadleaf shrubs. Trees include Capparis ferruginea, Zizyphus rignoni, Bursera simaruba, Cameraria angustifolia, Cordia buchii, and Plumeria obtuse, which are all native.

Sampling birds

Protocols were approved by the Institutional Animal Care and Use Committee of the National Aviary/Pittsburgh Zoo and PPG Aquarium (#NA06-004), and field work was authorized by the Secretaria de Estado de Medio Ambiente y Recursos Naturales. We conducted mist-netting in early- (November), mid- (December–January), and late-winter (late-January early-March). At the Mencia sites we used 16 nets (12 m × 30 mm mesh) in each net array for three days: 3 h the first afternoon, all day the next day and 3 h on the final morning. At the Aceitillar site, we placed nets in two lines of 31 and 39 nets with each array covering 700 m. Weather permitting, netting began ∼15 min after sunrise and ended at 16 h 00. All mist-netted birds were identified to species, sexed and aged (juvenile: HY/SY, or adult: AHY/ASY) using plumage or molt limits (Pyle, 1997; Latta et al., 2006). Birds were weighed with a Pesola spring balance (±0.5 g), and we measured wing chord (±0.5 mm). Most birds were banded with a numbered metal band and three color bands to identify each individual. For birds too small to band (hummingbirds, todies) we clipped the outer primary feather so recaptures within a netting session could be identified. Abundances of birds are summarized as birds captured per 1,000 mist-net hrs (mnh), where one 12-m mist net opened for 1 h = 1 mnh.

Mist nets are subject to several biases (Remsen & Good, 1996). This study minimized some of these problems: most vegetation in these sites is relatively low, except in the mature dry forest, yet all sites contain some remnant resulting in similar height ranges. We limited analyses of net capture frequencies to within-species comparisons and assumed equal capture probabilities within species among sites; finally, our mist-netting schedule (2–3 days every six weeks) minimized net shyness of birds.

We classified birds captured in mist nets into groups based on diet on the basis of principal food items consumed in optimal habitats (Wunderle & Latta, 1996). Groups included insectivores, nectarivores, granivores (including frugivores), carnivores, and omnivores. We grouped birds based on migratory status (Latta et al., 2006), including permanent residents and latitudinal migrants. We also classified birds by primary habitat, assigning all species to a single preferred habitat (secondary forest, deciduous forest, evergreen forest) based on Stotz et al. (1996).

Demographics and site persistence

Abundance data alone can be a misleading indicator of survival and habitat quality (James, 1971). Moreover, among over-wintering migratory birds, dominance interactions result in many species segregating by sex and age class (Holmes, Sherry & Reitsma, 1989; Latta & Faaborg, 2002), so age- and sex-specific data on persistence of individual birds is needed to assess habitat quality (Faaborg et al., 2010). Site persistence has been widely used as a proxy for survival in studies of over-wintering migrants (Holmes, Sherry & Reitsma, 1989; Latta & Faaborg, 2002; Faaborg et al., 2010). Over-winter site persistence is defined as the proportion of birds detected (by resighting or recapture) at any time >24 h after initial capture. Following banding, each plot was systematically searched 56.5 ± 6.2 h (SE) for color-banded individuals, with search areas extending ∼100 m beyond plot boundaries. Although resighting ability may vary with habitat, we minimized this potential bias with the use of multiple expert field personnel with many years of experience resighting color-banded birds in these habitats, and by continuing resighting within a site until all previously identified site-faithful birds were relocated, or until no more newly resighted banded birds were identified after 10 person-hours of searching a site. While a few color-banded individuals may have remained unidentified, consistent resighting effort among sites and years insured comparability of results among habitats.

Food resources

Assessment of potential food items was focused on insect and spider abundance (hereafter referred to as insects) as we sought to explain the distribution, in particular, of over-wintering migratory birds. Insects were sampled in December, January and February of years 1 and 2 with five yellow sticky traps placed at breast height for 48 h along a transect in each habitat. All insects 0.1–5.0 mm length were counted and pooled across sampling periods; the few larger insects were excluded because most wintering migrants consume primarily small arthropods (Greenberg, 1995; Poulin & Lefebvre, 1996). Leaf litter samples were collected in mid-winter (January) from n = 30, 25-cm diameter circles in each site and placed into a sorting pan. All insects <5.0 mm were counted and identified to order.

Statistical analyses

Comparing communities

We used rarefaction to compare species richness among sites, pooling data from across years. Rarefaction produces idealized species-accumulation curves that allow direct comparison of results among groups that differ in patterns of abundance or are sampled using different techniques (Gotelli & Colwell, 2001). Rarefaction calculates the expected species richness of the different groups for a constant sampling effort, but does not provide an estimate of asymptotic richness. Rather, for each accumulation curve we calculated a Chao 1 non-parametric estimator of richness with its variance and 95% confidence interval (Chao, 1984).

We calculated species richness and Shannon Diversity for each site and each year of the study. We converted diversity to the “effective number of species” by exponentiating species richness (Jost, 2006) to represent true diversity with mathematical properties allowing comparison among groups. We also calculated numerical dominance of species captured within each site, and Pielou’s evenness index (Magurran, 1988; Pielou, 1966) of the distribution of individuals among taxa (absolute evenness = 1.0). We modeled how these indices varied as pastures aged using regression and how they varied across the chronosquence using one-way ANOVA. We log-transformed species diversity and logit transformed evenness because it is a percentage (Warton & Hui, 2011) and modeled both using a linear mixed model (LMM). Species richness was modeled using a Poisson GLMM with an observation-level random affect to account for overdispersion (Harrison, 2014). All test statistics are reported for transformed data and analyses were re-run on untransformed data for plotting.

For each site we pooled data from across years and used Jaccard’s index to compare the similarity of communities based on presence/absence of species, and Sorenson’s measure to compare the similarity of sites based on the proportional abundance of recorded species. Both indices were calculated using the SimilarityPair() function in the SpadeR package (Chao et al., 2016). We also used non-metric multidimensional scaling (NMDS) with a Bray-Curtis distance matrix in the vegan package (Oksanen et al., 2017) to visualize the overlap in community composition among the sites. Due to the repeated sampling of the same sites and variation in effort these results are only exploratory.

Species-specific comparisons among habitats

Species-specific trends in capture rates were estimated using a random-slopes Poisson GLMM. To examine non-linear trends indicated by these models we also built separate GLMMs for certain species that incorporated a quadratic time term. We also built separate one-way ANOVA models for each species to examine differences between the pasture sites and the mature reference forest.

We modeled the site persistence of birds captured during the first year of the study at pasture sites (2003–2004) and mature forest (1996–1997). A bird was considered site persistent if it was re-sighted or recaptured during the year it was first captured or any subsequent year. We modeled site persistence using a one-way ANOVA style binomial GLMM where the number of “trials” was the number of individuals of each species captured, and the number of “successes” was the number of site-persistent birds. We calculated mean site persistence of migrants and residents for each site using species and month of capture (November, January, or February/March) as random effects. We included an observation-level random effect to account for binomial overdispersion (Harrison, 2015). We tested for linear and quadratic trends for both migrants and residents.

We examined differences among sites in the proportion of individual birds that were adult, male, migratory, and endemic. We analyzed all of these variables as binary data with logistic regression GLMMs. As for capture rates, we examined species-specific trends as regeneration time increased using a single random-slopes GLMM, and variation between the pasture and reference forest using one-way ANOVA. The species used in a given model depended on how accurately they could be aged or sexed and sample size. For models of the proportions of migrants and endemics only bird ID (for recaptures) and year were used as random effects.

To determine how the abundance of flying and litter arthropods changed as pastures underwent succession we modeled total arthropod abundance versus the initial time since pasture abandonment using a nested Poisson GLMM, with site as a random effect and an observation-level random effect for overdispersion. We tested for an overall difference between pasture sites and the mature forest using a t-test like Poisson GLMM with an indicator variable for “pasture” and “non-pasture”, with site treated as a random effect; confidence intervals around this difference were calculated using parametric bootstrapping. We explored overall differences in the arthropod assemblage by treating each sample as an independent replicate in a multivariate analysis of variance (MANOVA) with site as a factor. We examined trends in individual taxonomic groups across the chronosequence by plotting mean percent compositions of each taxa with bootstrapped 95% confidence intervals.

To determine if body condition varied among sites we calculated the Scaled Mass Index (SMI or ${\hat{M}}_I$; Peig & Green, 2009). We used log-transformed body mass and wing chord measurements as our body size variables and the smatr package (Warton et al., 2012) to calculate the scaling exponent b_SMA. SMI was only calculated for individuals that were classified as site persistent. In contrast to our analysis of site persistence, we used site-persistent birds captured any time during the study for this analysis. If a bird was captured more than once, we only used mass and length data from its first capture. We first built a regression-style linear mixed model (LMM) to estimate species-specific changes in SMI as regeneration time increased. We then examined change across the chronosequence using one-way ANOVA. Because of limited sample sizes we used only linear models without random effects for these ANOVA analyses.

Patterns of bird distribution

Through 40,723 mnh we recorded 7,315 captures of 60 species (Table 1). Twenty-five species of birds with >30 captures each composed 96.9% of all captures, but capture rates were higher in the two-year and five-year-old sites (231.5 and 236.0 birds/1,000 mnh, respectively) than in the ten-year and twenty-year-old sites (152.5 and 132.9 birds/1,000 mnh, respectively), or the mature dry forest (155.4 birds/1,000 mnh).

Rarefaction curves indicated that sampling intensity was appropriate, with curves with pooled data from all years for each site approaching an asymptote (Fig. 1). Observed species richness was highest in the three youngest sites and lowest in the twenty-year-old habitat, with intermediate richness in the mature dry forest (Fig. 1; Table 2). Shannon Diversity Index was highest in the three early-successional sites; diversity was lowest in the mature dry forest (Table 2). Chao 1 non-parametric estimates of species richness indicate no significant difference in the size of the species pool among sites, although the Chao 1 point estimate of richness was highest in the youngest site (Table 2).

Regression analysis indicated that species richness (β_{site_age} =  − 0.02, ${\chi }_{4,5}^2=5.7$, P_LRT = 0.017) and diversity (β_{site_age} =  − 0.18, ${\chi }_{4,5}^2=9.6$, P_LRT = 0.072) decreased significantly as habitat regeneration time increased (Fig. 2; Appendix S3). While the mature forest site was generally similar to the oldest regenerating pasture, ANOVA trend analysis indicated that these declines were consistent across the entire chronosequence (species richness trend: Z =  − 2.4, P = 0.02; diversity trend: Z =  − 5.6, p < 0.001). In contrast, evenness increased (β_{site_age} = 0.24, ${\chi }_{4,5}^2=4.2$, P_LRT = 0.04) with regeneration time, but declined to its lowest value at the mature forest site (quadratic trend: Z =  − 2.4, P = 0.017).

For data pooled across years, the extent of numerical dominance of species within a site also suggests the degree of evenness (Table 3); the five dominant species in each site accounted for 47.9–52.9% of captures in all early-successional habitats, whereas five species accounted for 61.1% of captures in mature dry forest. Although Greater Antillean Bullfinch (scientific names of all birds appear in Table 1) appeared as one of the five most abundant species in all habitats, other species contributed to among-habitat differences in numerical dominance. For example, Yellow-faced Grassquit and Common Ground-Dove were among the most common species in the two-year and five-year-old sites, with Ovenbird most abundant in the ten-year and twenty-year-old sites, and Bananaquit, Black-crowned Palm-Tanager, and Green-tailed Ground-Tanager reaching highest densities in oldest sites.

Similarity indices based on species presence/absence pooled across years were only moderately high with scores of 0.50–0.69 for habitat pairs (Table 4); scores tended to be higher between similarly-aged sites, especially among the three youngest pastures. Sorenson similarity based on the proportional abundances of individuals mist-netted (Table 4) had a broader range (0.67–0.81). Exploratory ordination with NMDS also indicated that the youngest sites had highly overlapping species pools, and that the mature dry forest reference site was very distinct; the oldest abandoned pasture was also somewhat distinct from the younger sites (Fig. 3).

Of the 60 species recorded, 13 (22%) were endemic residents, and 22 (37%) were latitudinal migrants (Tables 1 and 2). The proportion of endemic individuals increased as succession in the pastures progressed, but this trend was not significant (β_{site_age} = 0.06, SE = 0.04, Z = 1.4, P_LRT = 0.15), yet across the entire chronosequence there was a significant linear trend in the frequency of endemic individuals (Z = 5.28, P < 0.001). Though their community composition was different as discussed above, the frequency of endemic individuals was similar in the oldest pasture (25%) and the mature forest (23%).

The proportion of migrant individuals varied from 26–46% in the pasture sites (Tables 1 and 2). The overall proportion of migrants increased as regeneration time increased (β_{site_age} = 0.01, SE = 0.03) but the trend was not significant in regression models (${\chi }_{4,5}^2=0.25$, P_LRT = 0.61). Capture rates of migrants declined from the second oldest to the oldest site but there was no significant quadratic trend (${\chi }_{5,6}^2=0$, P_LRT = 1). The proportion of migrants in the mature forest was lower than any other site (15%), resulting in a significant negative quadratic trend across the chronosequence (Z =  − 5.25, P < 0.001).

Presence/absence data pooled across years suggest that most species occurring at our study sites were habitat generalists. Of the 60 recorded species, 21 of 25 species with n > 30 captures were recorded in all five sites (Table 1). Hispaniolan Lizard-Cuckoo and Black-crowned Palm-Tanager appeared to be true habitat generalists with no overall differences in capture rates among sites. However, others showed differences in abundance among sites (Table 1). Among residents, six species appeared to prefer the early-successional sites. These included Common Ground-Dove, Hispaniolan Emerald, Northern Mockingbird, Hispaniolan Spindalis, and Yellow-faced and Black-faced grassquits. In contrast, three species showed higher capture rates in more mature, forested sites. These included Key West Quail-Dove, Green-tailed Ground-Tanager, and Greater Antillean Bullfinch.

Linear trends as pastures age

Neotropical migrants varied in their response to forest regeneration. Capture rates of Ovenbird, Black-throated Blue Warbler, and Black-and-white Warbler all increased significantly as regeneration time increased (Fig. 4). Prairie Warbler, Palm Warbler, Common Yellowthroat, and Cape May Warbler all showed evidence of significant overall declines in capture rates as sites regenerated, while American Redstart showed a marginal decline. Among resident species, the Key West Quail-Dove and Hispaniolan Lizard-Cuckoo had higher capture rates as regeneration time increased, while captures of the Green-tailed Ground-Tanager and Black-crowned Palm-Tanager increased marginally. Capture rates of Hispaniolan Spindalis, Gray Kingbird, Black-whiskered Vireo, and Bananquit declined significantly with regeneration time.

Trends including mature forest reference

Across the entire chronosequence, the Black-and-white Warbler was the only Neotropical migrant which increased consistently in abundance with habitat regeneration time (ANOVA trend test: P_linear = 0.009, P_quadratic = 0.98; Appendices S4 and S5). Most other species increased in abundance over part of the chronosequence but declined in the oldest pasture and/or the mature forest site as indicated by significant quadratic trends (American Redstart: P_linear < 0.001, P_quadratic = 0.083; Black-throated Blue Warbler P_linear < 0.001, P_quadratic < 0.001; Ovenbird P_quadratic < 0.001). The Cape May Warbler (P_linear = 0.001), Prairie Warbler (P_linear < 0.001), and Palm Warbler (P_linear = 0.0013) exhibited consistent declines across the entire sequence of habits. Most Neotropical migrants reached peak abundance at the La Caoba site which represented five years of regeneration when the study began (Appendices S4 and S5).

Because many of the migratory species had non-linear trends in abundance across sites, we fit additional regression models to data for individual species that included a quadratic term age of regeneration. Ovenbird (P = 0.01), Common Yellowthroat (P = 0.06), Black-throated Blue Warbler (P = 0.06), and Palm Warbler (P = 0.01) all had significant quadratic terms in these regression models, indicating peaked abundances at a single pasture site with subsequent declines in abundance (Appendix S6).

Age ratios

Overall, the age ratio of permanent resident species remained constant as forest regenerated (β_{site_age} =  − 0.002, SE = 0.014, Z =  − 0.17, P_Wald = 0.9), while for over-wintering migrants the proportion of older individuals increased as regeneration progressed (Fig. 5A; residency × site-age interaction: β_{site_age} = 0.035, SE = 0.02, χ_9,10 = 4.48, P_LRT = 0.03). The age ratios of individual resident species increased or decreased with regeneration time to various degrees (Fig. 5B), while the trend for all migrant species was for a larger proportion of older birds as regeneration progressed. Increases in the proportion of older birds with regeneration time were significant for Cape May Warbler and marginally significant for Ovenbird and American Redstart (Fig. 5B).

Similar to the regression analysis, across the entire chronosequence there was no evidence for any trend for age ratios of residents (linear trend: P = 0.23; quadratic trend: P = 0.87). For migrants, there was a significant quadratic trend (p = 0.006) driven by very low mean proportions of AHY migrants in the reference mature forest at Aceitillar (33%) relative to the abandoned pasture sites (43%–61%). Mean proportions of AHY birds at each site for the most abundant species are summarized in Appendix S7.

Sex ratios

We reliably determined sex of six species of over-wintering migrants that were mist-netted with adequate sample sizes for analysis, including Common Yellowthroat, American Redstart, and Cape May, Black-throated Blue, Black-and-white, and Prairie warblers (Fig. 6A). We also determined sex ratios of three permanent resident species: Greater Antillean Bullfinch, Yellow-faced Grassquit, and Black-faced Grassquit (Fig. 6B). For most individual migrant species there was no change in the sex ratio as regeneration time increased. Two exceptions were significant increases in the proportion of male Cape May Warblers, and a marginal increase in American Redstarts. Overall, the sex ratio of residents tended to be more male-biased than migrants (58% male vs. 36% male; β_residency = 0.92, SE = 0.40, χ_7,8 = 4.6, P_LRT = 0.03). Considering the entire chronosequence, the percentage of males captured did increase for migrants (linear trend Z = 2.3, P = 0.022; Appendix S3), but not for residents (Z =  − 0.5, P = 0.6). Mean sex ratios birds for the most abundant species are summarized for each site in Appendix S7.

Preferred diet

Exploratory multivariate analysis with NMDS indicated that a major axis of variation among sites was the number of omnivorous individuals. The proportion of individuals of omnivorous species increased over the first 15 years of succession (β_site∗age = 0.016, SE = 0.0068, χ_4,5 = 4.34, P_LRT = 0.037). Across the entire chronosequence, the smallest proportion of insectivorous individuals occurred in the most recently abandoned pasture (23%, bootstrapped 95% CI [17%–32%]; Fig. 7A) and the largest occurred in the mature forest (43%, 95% CI [35%–52%]). This resulted in an overall significant linear increase across the chronosequence (Z = 3.6, P = 0.004).

Preferred habitats

The proportion of individuals of species which preferred secondary forests declined over the first 15 years of succession (β_{site_age} =  − 0.07, SE = 0.03, χ_4,5 = 3.73, P_LRT = 0.054), but increased somewhat at the oldest regenerating site (quadratic site age effect: β_{site_age^∧2} = 0.005, SE = 0.002, χ_5,6 = 10.4, P_LRT = 0.001). Across the entire chronosequence, the highest proportion of secondary-forest birds occurred in the most recently abandoned pasture (38%; Fig. 7B), and the least occurred in the mature forest (13%). This resulted in an overall significant linear trend across the chronosequence (Z =  − 6.6, P < 0.001). At the mature forest site, most individuals were of species that preferred evergreen forests (75%).

Site persistence

Within the regenerating pastures, mean site persistence of all migrant species increased from 0.44 at the youngest site to 0.62 at the oldest, but dropped to 0.15 at the mature forest site. This resulted in an overall significant negative quadratic trend (Z =  − 4.2, P < 0.001). For resident species, site persistence declined significantly overall (Z =  − 2.6, P = 0.009) from the youngest site (0.38) to the mature forest (0.13), though persistence was highest in the oldest regenerating site (0.43). Mean site persistence rates for the most abundant species are summarized in Appendix S8.

Arthropod resources

Site persistence may be related to food availability; flying insects caught on sticky traps were least abundant in the two-year-old habitat and most abundant in the mature forest site (Fig. 8A), with flying insects 1.7 times (bootstrapped 95% CI [1.02–2.6]) as abundant in the mature forest as compared to the pastures (nested-GLMM: ${\chi }_{3,4}^2=4.67$, P_LRT = 0.03). Across just the pasture sites, flying insects increased consistently with age of the regenerating site (Fig. 8A), although the trend was not significant (β = 0.013, SE = 0.017, ${\chi }_{3,4}^2=0.62$, P_LRT = 0.43). Exploratory MANOVA analyses treating samples within sites as independent indicates variation in the composition of insect communities among sites (F_32,164 = 1.98, P = 0.003). The two most recently abandoned pastures (<10 years since abandonment) tended to have more Coleoptera and Homoptera, and fewer Diptera, than the two older sites (>10 years) and the mature forest (Appendix S9a).

Leaf litter arthropods were most abundant in the oldest abandoned pasture, and least abundant in the mature forest (Fig. 8B). Leaf litter insects were 0.55 times (bootstrapped 95% CI [0.22–1.32]) as abundant in the mature forest compared to the pastures, although the difference was only marginally significant (${\chi }_{3,4}^2=2.69$, P_LRT = 0.10). Across pasture sites, litter arthropod abundance increased marginally with age (β = 0.039, SE = 0.022, ${\chi }_{3,4}^2=4.03$, P_LRT = 0.045). Exploratory analyses indicate marginally significant variation in the composition of insect communities among sites (F_28,168 = 1.48, P = 0.0.07). Ants tended to increase with time since abandonment, but had relatively low abundance in the forest, while spiders tended to decline with time but were relatively abundant in the forest (Appendix S9b).

Body condition

Body condition of site-persistent individuals of 15 species was compared across pasture sites by analyzing variation in the scaled mass index (SMA). As sites regenerated, regression analysis indicated that condition increased significantly for Hispaniolan Lizard-Cuckoo, and decreased for Red-legged Thrush (Fig. 9). For species that occurred in both abandoned pastures and the mature forest site, ANOVA trend analysis indicated that Bananquit peaked in body condition at moderately regenerated sites but declined in condition in the oldest regenerating habitat and in the mature forest (linear trend: t =  − 2.65, P = 0.009; quadratic trend: t =  − 2.51, P = 0.013; Appendices S10 and S11). Black-crowned Palm-Tanager did not show a significant decline in regression analysis (Fig. 9), but did when considering the mature forest site (linear trend: t =  − 3.4, P = 0.001). Similarly, the abundance of the Greater Antillean Bullfinch was fairly constant as regeneration time increased (Fig. 9), but declined significantly at the mature forest site (linear trend: t =  − 3.4, P = 0.001; quadratic trend: t =  − 2.3, P = 0.023). Green-tailed Ground-Tanager showed a similar pattern, with a relatively constant condition as regeneration progressed, but condition declined for birds at the mature forest site (linear trend: t =  − 2.32, P = 0.021; quadratic trend: t =  − 1.7, P = 0.09). Among over-wintering migratory birds, adjusted body mass did not vary significantly across sites (Fig. 9, Appendices S8 and S9).

Conservation implications

Many previous studies of the impact of anthropogenic habitat change on birds have focused on the impact of deforestation and fragmentation of primary forests on species and communities and the associated loss of biodiversity from these landscapes (Stouffer & Bierregaard, 1995; Sekercioglu et al., 2002). Relatively few studies, however, have looked at the inverse questions: i.e., what is the pattern of regained bird diversity as agricultural lands are abandoned, when do regenerating forests attain equivalent ecological value for birds found in mature forest, and what components of the agricultural matrix support higher biodiversity?

Unfortunately, our results are not particularly encouraging in terms of immediate impacts from creating new habitat for birds in Hispaniolan dry forest landscapes. That the avian community in our mature dry forest is highly distinctive as compared to forests regenerating for up to 24 years, emphasizes that attaining ecological equivalency takes decades or even longer. Regenerating forests appear to have more value to over-wintering migratory birds, which occurred in large numbers, but age and sex ratios, and site persistence data, also suggest that early-successional habitats may be of relatively low quality for most species. Some optimism may be seen, however, in that sites occurring later in our chronosequence tended to have older migrants present, and thus may provide appreciably better habitat for a few migrant species.

Results from this study suggest some components of the agricultural matrix may be of particular importance in contributing to avian diversity in regenerating dry forest sites. While tree cover in the context of the agroecological matrix may typically be thought of in terms of remaining blocks of primary forest, individual trees may also be retained for fruit, shade, or other functions (Harvey et al., 2006). These remnant trees may then serve birds as dietary resources (Sekercioglu et al., 2007), shelter (MacGregor-Fors & Schondube, 2011), stepping stones between forest patches (Graham, 2001; Sekercioglu et al., 2007), and microclimatic refuges (Greenberg, Bichier & Sterling, 1997; Sekercioglu et al., 2007). Despite the difference in age of regenerating pastures, a variety of bird species were likely attracted to the abundant flowers and fruit of remnant trees remaining in all of the abandoned pastures, contributing to diversity indices. A variety of trees have been shown to be important for birds in an agricultural countryside (Harvey et al., 2006; Estrada, Coates-Estrada & Meritt, 1997; Sekercioglu et al., 2007), but leguminous Acacia like Senna (Greenberg, Bichier & Sterling, 1997), and Bursera (Greenberg, Foster & Marquez-Valdelamar, 1995; Latta, Gamper & Tietz, 2001), both of which occurred in our sites, have been shown to be particularly valuable to birds. This suggests that agricultural efforts and restoration methods that promote the retention of remnant trees in the landscape can have a very positive impact on many bird species.

Finally, although protecting natural vegetation is essential for the preservation of major elements of biodiversity (Bruner et al., 2001; McKinney, 2002; Sekercioglu, Daily & Ehrlich, 2004), our results suggest that the agricultural matrix can provide habitat for some birds in the buffer zone of forested protected areas. Discussions of buffering the effects of agriculture on protected areas have often focused on providing a mature forest matrix to facilitate the movement or dispersal of forest species (Vandermeer & Perfecto, 2007; Burkey, 1989; Vandermeer & Carvajal, 2001). Until now much less has been written on how the agroecological matrix itself, and early successional regeneration, can provide habitat for birds (but see MacGregor-Fors & Schondube, 2011). However, care should be taken to examine body condition and demographic rates as we do here in assessing the value of these early successional habitats.

Study limitations

For this study we have opportunistically made use of a chronosequence of abandoned pastures that could be surveyed in parallel over five years, and combined this information with a previously studied mature forest site. Making inference from chronosequences necessarily requires assuming that the successional trajectories that resulted in conditions at the surveyed sites will be similar at other places and times. In our study we have additionally assumed that our reference forest site, which has never been logged, represents a comparable endpoint to pasture succession. Numerous factors could result in different successional trajectories in this landscape, such as the availability of propagules of forest species (Holl, 1999), the presence of invasive species (Aide et al., 2000; Sobanski & Marques, 2014), and altered soil properties (Nagy et al., 2017; Lozano-Baez et al., 2018). However, these factors are often most important during the initiation of woody cover regeneration, and our sites had all begun at least the initial development of woody cover. In addition, despite the recognized limitations of chronosequences, they have generally been considered invaluable for understanding long-term ecological dynamics (DeWalt, Maliakal & Denslow, 2003; Myster & Malahy, 2008; Walker et al., 2010).