Long-term monitoring reveals an avian species credit in secondary forest patches of Costa Rica

Study sites

This study was conducted on private plots near Las Cruces Biological Station (LCBS), Puntarenas province, Costa Rica (8°47.7N, 82°57.32W). Rainfall at LCBS averages ∼4,000 mm/year. Daytime temperatures range from 13 to 26 °C. The Las Cruces area had relatively intact premontane rainforest until the 1950s when immigration, economic development, and government policies led to deforestation and agricultural production. As a result, annual deforestation rates of 2.1% from 1947 to 1960 and 3.9% from 1960 to 1980 shifted forests to progressively smaller fragments (Zahawi, Duran & Kormann, 2015). Subsequent abandonment of agricultural plots resulted in some regeneration, and these secondary forests are the focus of this study.

All three monitoring sites were regenerating broadleaf forests >30 years old (Fincas Sofía, Cántaros, and Corteza described in Appendix S1; Fig. 1). Sites were selected based on similarity of vegetation composition and structure. Vegetation was described and quantified by Arce & Brenes (2007) and re-sampled by Pereira (2011) to assess structural and compositional changes. The relevé method, fully described by Ralph et al. (1996) was used, with vegetation data collected in 9–13 variable radius plots (25–50 m) per study site, with each non-overlapping plot centered on a mist net location. Shrub and tree diversity were determined in the plot, and the average height of the lower and upper bounds of the tree stratum (all vegetation ≥5 m) and shrub stratum (all vegetation ≥0.5 and <5 m) were measured. The cover of each stratum was estimated using five cover classes (0–5, 5–25, 25–50, 50–75, and 75–100%). Also for each stratum, the species and their diameter-at-breast-height was measured for the largest and smallest trees in each stratum.

Forest plots did not change dramatically between vegetation sampling periods (Pereira, 2011). In both Fincas Cántaros and Sofía, fewer shrub species were recorded in 2011, with Miconia spp. more common in 2011. In addition, canopy closure increased in the tree layer of Finca Cántaros in 2011. Otherwise, there were no significant differences in vegetation diversity, structure, or composition at any of the sites, including the average lower and upper bounds of the vegetation layers, estimated percent cover, and diameter-at-breast-height for the largest and smallest trees in each layer.

Sampling birds

In tropical forests, point counts tend to be more effective at sampling avifauna in mature forests, while mist nets are more effective in disturbed forests where more species utilize the understory (Blake & Loiselle, 2001). To limit bias, we endeavored to use both methods (Ralph & Scott, 1981; Ralph et al., 1996), sampling biannually 2005–2014 in the late breeding season (July–August, except in 2006 and 2012) and mid-winter (January), although sampling by point counts took place primarily in January (Appendix S2).

We used 15 mist nets (12 × 2.5 m, 30 or 32 mm mesh) in fixed locations at each site, and opened them from 0545 to 1045 on two consecutive days, so net hours were consistent between seasons and among years. All mist-netted birds were identified to species by plumage (Stiles & Skutch, 1989) and to age (juvenile or adult) by plumage or molt limits whenever possible. All birds, except hummingbirds, were banded with a numbered metal band. As capture effort remained constant, we expressed abundance as the number of birds captured/season. Handling of birds was under permissions 720017214 and 221946136 provided by the Ministerio de Recursos Naturales, Energía y Minas of Dirección General de Vida Silvestre (MINAE), and the Institutional Animal Care and Use Committee of the National Aviary and Pittsburgh Zoo/PPG Aquarium permit 2006-SL1.

We conducted 10 min, 50 m fixed-radius audiovisual point counts using the intensive point count protocol of Ralph et al. (1996), with the number of points established (Finca Sofía = 5, Finca Cántaros = 4, Finca Corteza = 2) dependent on the size and shape of each site. Points were placed along narrow foot trails, a minimum of 100 m from other points to help maintain independence. All points were counted once per season and were completed from sunrise—09:30; no counts were conducted in inclement weather. Birds counted from these points were combined into a single mean, so the distance between points is less critical than extensive point counts where each point is intended to be statistically independent (Ralph et al., 1996).

We eliminated fly-overs, and species that are primarily nocturnal, aquatic, or aerial foragers from analyses. We classified birds as permanent residents, latitudinal migrants, or elevational migrants based on AOU (1998), Blake & Loiselle (1991, 2000, 2001), Reid, Harris & Zahawi (2012), and Stiles & Skutch (1989). We assigned species to a single preferred habitat on the basis of Stotz et al. (1996). Habitats were either: (1) primary forest; (2) secondary forest, scrub or edge; or (3) other non-forest habitat. We also classified birds based on habitat breadth (defined as the number of habitats a species occupied across its range), sensitivity to disturbance (designated as high, medium, or low), and conservation priority (scored as 1 or 2 = medium, or 3 or 4 = low), with all data derived from Stotz et al. (1996). Birds were grouped into foraging guild or diet on the basis of principal food items consumed (Stiles & Skutch, 1989; Boyle & Sigel, 2015), and included carnivores, insectivores, frugivores, granivores, nectarivores, and omnivores. We also determined which of four non-mutually exclusive foraging strata were utilized: terrestrial, understory, mid-story, and canopy (Stotz et al., 1996). For some regression analyses, similar groups were pooled to balance factor levels and/or increase sample sizes.

Characterizing avian communities

We characterized the pool of species using secondary forests by building rarefaction curves from our entire dataset, combining samples across years and sites to increase sample size. We used iNEXT (Hsieh, Ma & Chao, 2016) in R 3.3.1 (R Core Team, 2016) to compare species richness in different seasons (January, August) for both counts and net captures. For each curve we calculated a Chao 1 non-parametric estimator of richness and Shannon diversity (Chao, 1984; Colwell & Coddington, 1994); we expressed Shannon diversity as the effective number of species (Jost, 2006).

We calculated species richness and Shannon diversity from the raw capture data using vegan (Oksanen et al., 2016) in R. We modeled changes in ln(richness) and diversity using linear-mixed models (LMMs; Gelman & Hill, 2007). We examined changes in both overall species richness and richness within subgroups (i.e., primary vs secondary forest species).

Modeling population trends

Using random-slopes Poisson generalized LMMs, we modeled population trends for all species observed during a given season for four or more years. This approach allows for the leveraging of information from groups with larger sample sizes to model rarer species (Gelman & Hill, 2007). An observation-level random effect was also included to correct for over-dispersion (Kéry, 2010). We determined whether trends in net captures depended on species traits by testing for significant year-by-trait and year-trait-season interactions; we modeled each predictor separately to avoid multicollinearity. We calculated the estimated trend for each level of a predictor variable by combining regression parameters and their standard errors (SEs) using the multcomp package (Hothorn, Bretz & Westfall, 2008).

To estimate species-specific trends we fit a model without any predictors and extracted species-level slopes (BLUPS; Robins, 1989), and estimated their SEs using the se.coef function in the arm package (Gelman & Su, 2015). For all models, data from both seasons were used except for those related to latitudinal migration for which only January data was applicable. All models included random-intercepts for year and season nested within year using lmer from lme4 (Bates et al., 2015) in R. For both species-richness and population-trend models, we checked whether inclusion of traits and year × trait interactions improved our models by comparing AIC and log-likelihood values to null models with no fixed effects and models with only year as a fixed effect.

Trends in species richness

Total species richness from mist netting did not change over time (year effect p = 0.72) in either season (year × season p = 0.69). Similarly, Shannon diversity did not change over time overall (p = 0.20) or in either season (p = 0.50). Within subgroups, species richness did change over time. In January, richness of residents had an upward though non-significant trend (slope = 0.03, SE = 0.04) while richness of latitudinal migrants declined (slope = −0.13, SE = 0.04; year × migrant effect: χ²_5,6 = 7.1, p = 0.008; Fig. 3A). In both seasons, richness of primary forest species increased marginally over time (slope = 0.052, SE = 0.03), while richness of secondary forest species declined (slope = −0.08, SE = 0.03; year × habitat effect: χ²_8,9 = 10.9, p < 0.001; Fig. 3B). Species richness within foraging guilds also changed over time (year × foraging guild effect: p = 0.015, χ²_10,12 = 8.4; Fig. 3C). Richness of omnivores displayed a downward though non-significant trend (slope = −0.10, SE = 0.053) while other guilds were constant (frugivores, nectarivores, and seedeaters: slope = 0.024, SE = 0.04; insectivores: slope = 0.04, SE = 0.052). Richness also changed with respect to disturbance sensitivity (Fig. 3D). When the study began species ranked as moderately to highly sensitive to disturbance were less abundant than those less sensitive to disturbance. However, over time, richness of species highly sensitive to disturbance increased (year × sensitivity effect: χ²_7,8 = 26, p < 0.00001, slope = 0.087, SE = 0.03), while those with low sensitivity declined (slope = −0.17, SE = 0.024). For these models inclusion of year × trait interaction generally improved model fit by 5–50 AIC units relative to a null model (Appendix S3).

All other potential predictors of trends in species richness were non-significant. Additionally, we did not find any significant season × year interactions, indicating that differences between seasons are due solely to differences in intercept terms of models.

Abundance trends

Trends in abundance were similar between seasons for all the traits examined; all season × year interactions were non-significant (all p > 0.45, Appendix S4), as were all three-way season × year × trait interactions (all p > 0.25, Appendix S4). Across seasons, species with different habitat preferences (p = 0.001, χ²_11,12 = 11.3; Table 3), disturbance sensitivity (p = 0.00001, χ²_11,12 = 20.14), foraging guilds (p = 0.042, χ²_11,12 = 4.13), and habitat breadths (p = 0.0002, χ²_8,9 = 14.25) displayed different trends (Table 3; Appendix S4). Latitudinal migrants in January also differed marginally from residents (p = 0.11, χ²_8,9 = 2.57). Inclusion of year × trait interactions generally improved the relative fit of the models. For models with p-values <0.01, inclusion of the year × trait interaction reduced AIC scores by 8–10 AIC units (Appendix S5). Our foraging guild models have marginal p-values and similarly small improvements in AIC (∼1).

For a given type of trait, species more characteristic of mature forest generally increased while those more characteristic of secondary forest or early successional habitats tended to decrease (Fig. 4; Appendices S6 and S7). Species preferring primary forest exhibited a positive trend (slope = 0.045, SE = 0.02, p = 0.026) while species preferring secondary forest exhibited a marginal decline (slope = −0.041, SE = 0.024, p = 0.074). Species ranked as moderately or highly sensitive to disturbance increased over time (slope = 0.067, SE = 0.021, p = 0.001) while birds with low sensitivity declined (slope = −0.043, SE = 0.021, p = 0.038). Specialist foragers generally increased (slope = 0.027, SE = 0.019) while omnivores decreased (slope = −0.033, SE = 0.024), though neither trend was significantly different from 1 (p = 0.17 and 0.23, respectively). Finally, species scored as using more habitats decreased in abundance (slope = −0.035, SE = 0.009, p < 0.0001).

Species-levels trends

Significant species-specific trends were found in either season for five species (Table 4; Appendix S8). Of these, four showed positive population trends and were primary forest species. Only one species (Variable Seedeater) showed a highly significant negative trend, and that species is an omnivore typical of scrub and edge habitat.

Another 19 resident species showed marginally significant population trends (Table 4; Appendix S8). Four species had significant trends in both seasons, and in each case the direction of population change was consistent between seasons. A total of 12 out of 16 species with marginally significant positive population trends were primary forest species, while six of six species with marginally negative trends were associated with secondary forest, scrub or edge. Two additional species with negative trends were over-wintering Neotropical migrants.

Study limitations

While we have demonstrated the value of older secondary forest patches to birds, a better understanding of population health requires investigation of avian vital rates. Even with long-term studies such as this taking into account rare species, inter-annual variation, and seasonality, abundance data alone can be a misleading indicator of population size and habitat quality (Van Horne, 1983). Furthermore, abundance cannot generally be equated with survival or productivity, so data on these demographic parameters are required to assess the quality of secondary forest habitat to these birds.

This level of analysis has seldom been accomplished for species in secondary tropical forest (Barlow et al., 2007). Only recently, Şekercioḡlu et al. (2007) determined nesting success of three avian species in Costa Rica in a landscape including secondary forest fragments. They showed the conservation value of the agricultural countryside and suggested that this can be enhanced with even a modest increase in tree cover in the landscape matrix. Ruiz-Gutierrez, Gavin & Dhondt (2008) used mark-recapture analyses to show that apparent survival of the White-ruffed Manakin (Corapipo altera) was lower in primary forest fragments than in the large forest at LCBS, but emphasized the need for population-level studies of other species to test for sources of mortality in forest fragments and surrounding matrix habitats. Assessing survival and population trends is particularly challenging though, because of the need for sampling populations on an annual or more frequent basis using standardized protocols, and this is seldom done (Latta, Ralph & Geupel, 2005; Blake & Loiselle, 2015).

Finally, it should be remembered that other factors extrinsic to forest patches or the landscape matrix may also be affecting local birds—although in general these impacts are expected to be negative. In particular, declines and even extirpations of bird populations in tropical areas have been attributed to changes associated with global warming (Latta et al., 2011; Blake & Loiselle, 2015). While a changing climate may be affecting bird populations at our study sites, it is not likely responsible for the gains in forest-associated species recorded in this article.

Conservation implications

These results support the importance of secondary forest patches for bird conservation, and emphasize the value of the vegetation in the surrounding habitat matrix. Because we found very few changes in vegetation characteristics of our older secondary forest plots, we suggest that observed changes in the avian community, resulting in a species credit of birds associated with primary forest habitat, are related to changes in vegetation in the broader landscape. As such, we do not suggest that secondary forest patches serve as a safety net per se for tropical biodiversity; in this landscape, the safety net is likely found in the large blocks of core forest where species associated with primary forest persist as source populations. Rather, we suggest that bird diversity increases in maturing secondary forest through a species credit reflecting immigration of primary forest species from these source populations.

Although Zahawi, Duran & Kormann (2015) warned of the continuing threat of an extinction debt in the Las Cruces landscape resulting in the extirpation of additional species, our study suggests that the secondary forests in the tropical countryside are contributing to increasing trends in richness and abundance of bird species associated with primary forest. These results support understandings gained from regional studies that have shown that in landscapes such as Coto Brus, where low-intensity agriculture is a significant part of the land-use matrix, forested riparian corridors (Şekercioḡlu et al., 2015), clusters of trees as small as 20 m wide (Mendenhall et al., 2011), as well as secondary forest patches, can all contribute to biodiversity (Mendenhall et al., 2014), and affect resilience, stability, and ecosystem services (Karp et al., 2011).

Recognition of the value of secondary forests to birds, and perhaps other wildlife (Mendenhall et al., 2014), may impact decision-making on the value of acquiring and protecting secondary forests for conservation planning in these landscapes. This is not to suggest that conservation measures should not be taken to reverse the continuing loss of primary forest in the tropical countryside. While this study offers hope that in some landscapes, maturing secondary forest can provide habitat for a number of primary forest bird species, it should be remembered that secondary forests may differ systematically in vegetation composition and forest structure from the original primary forests (Chazdon, 2003; Lugo & Helmer, 2004), and successional trajectories are affected strongly by initial conditions and the surrounding landscape (Chazdon, 2003; Chazdon et al., 2009). As a result, not all birds associated with primary forests will benefit equally in these landscapes.