INTRODUCTION

Conservation planning and implementation require the identification of high-quality habitats and geographic locations where management actions may improve the viability of target species. The conservation of nonbreeding habitat of many Nearctic–Neotropical migratory songbirds can be especially challenging, yet many studies have shown that winter habitat quality and availability limit populations of migratory songbirds, and that alteration of winter habitat can lead to population declines (e.g., Marra et al. 1993, Sherry and Holmes 1996, Norris et al. 2004).

The Bicknell's Thrush (Catharus bicknelli) is considered one of the Nearctic–Neotropical migrants at greatest risk of extinction and thus of highest continental conservation concern (Pashley et al. 2000, Rich et al. 2004, Wells 2007, Townsend et al. 2015). The species is classified as Vulnerable by the International Union for the Conservation of Nature (IUCN) and as Threatened in Canada (Stattersfield and Capper 2000, COSEWIC 2009). Bicknell's Thrush breeding populations are declining in the core and northern parts of the species' range. Breeding Bird Survey data from 16 routes in Canada from 1968 to 2008 showed a significant decline of 9% per year (COSEWIC 2009). Trail-based point count surveys in the White Mountains of New Hampshire, USA, revealed a 7% annual decline from 1993 to 2003 (Lambert et al. 2008). The High Elevation Landbird Program documented annual declines of 17% in New Brunswick, Canada, and 15% in Nova Scotia, Canada, between 2002 and 2009 (Whittam and Campbell 2010).

At both ends of its migratory range, the Bicknell's Thrush has a limited, highly fragmented distribution and faces multiple threats to its habitats that may affect populations (Atwood et al. 1996, Lambert et al. 2005, Townsend et al. 2015). These threats include climate change (Rodenhouse et al. 2008), acid ion deposition (Johnson et al. 1992, Graveland and van der Wal 1996, Hames et al. 2002), mercury contamination (Rimmer et al. 2005, 2010), mountaintop development (Rimmer et al. 2004, Strong et al. 2009, Townsend et al. 2015), forestry operations (Nixon et al. 2001, Chisholm and Leonard 2008, Aubry et al. 2011, McKinnon et al. 2014), interspecific competition (Freeman and Montgomery 2016), and loss and degradation of winter habitats (Rimmer 2005, Townsend et al. 2009b, 2015). Evidence indicates that a complex interplay of ecological and demographic factors on the wintering grounds, exacerbated by forest loss and degradation, may be a primary limiting factor across the annual cycle of the Bicknell's Thrush (Townsend et al. 2009a, 2009b).

Recent studies in the Dominican Republic have identified wet, broadleaf, montane forest at middle elevations as a potentially crucial habitat for the Bicknell's Thrush (Townsend et al. 2010, 2012). Bicknell's Thrushes exhibit sexual habitat segregation in winter (Townsend et al. 2011, 2012). Female thrushes occur at relatively high densities in moderately disturbed mid-elevation rainforests of the Cordillera Septentrional, compared with densities in the more pristine cloud forests of remote mountain ranges, such as Sierra de Bahoruco and Cordillera Central, where males are encountered 4 times more often than females (Townsend and Rimmer 2006, Townsend et al. 2011). While research is ongoing to elucidate the complex overwintering ecology of Bicknell's Thrushes, the species' primary limiting factor is thought to be winter survivorship of females, including possible seasonal carryover effects (Townsend et al. 2009b).

Despite considerable recent attention focused on Bicknell's Thrush conservation, surprisingly few empirical data exist with which to identify priority conservation areas on the species' Caribbean wintering grounds. At a range-wide scale, presence-only records from the past ∼40 yr have been used to quantitatively evaluate winter distribution and habitat selection across the Greater Antilles (McFarland et al. 2013). The Dominican Republic was estimated to contain half of all potential winter habitat, with total winter precipitation, slope aspect, and land cover (particularly moist, broadleaf forests) found to be important predictor variables of thrush occupancy.

To stabilize or increase Bicknell's Thrush populations, immediate action must take place to conserve winter habitat (IBTCG 2010). Our study aimed to (1) estimate the abundance of Bicknell's Thrushes, while accounting for imperfect detection probability, in threatened forests of the Cordillera Septentrional, an area believed to be of high importance for female thrushes, and (2) relate abundance to a suite of habitat, geographic, topographic, and climatic variables. We also endeavored to demonstrate how spatial models of abundance can directly inform conservation actions for Bicknell's Thrushes in the Dominican Republic through identifying an area that was developed into a private reserve.

METHODS

Statistical Analysis

All statistical analyses were performed using R statistical software (R Core Team 2012) with package Unmarked 0.11 (Fiske and Chandler 2011). We modeled Bicknell's Thrush population dynamics using the model of Dail and Madsen (2011). This model extends the Royle N-mixture model (Royle 2004) for data collected on open populations (i.e. those experiencing mortality, recruitment, immigration, and emigration).

For each site (i = 1, . . . , R), primary sampling occasion (i.e. the repeated visits to each point; j = 1, . . . , J), and secondary sampling occasion (i.e. the 8 2.5-min counts during each of the 3 repeated visits; k = 1, . . . , K), we recorded the number of individuals detected, denoted as {y_ijk}. Within each of the J primary periods, the population was assumed to have been closed with respect to movement and demographic processes, but closure was not assumed between primary periods.

Abundance in the initial time period (N_i1) was treated as an outcome of a Poisson distribution with a mean of λ_i. This can be thought of as the expected abundance of animals following an initial territory establishment process. Subsequent to this initial time period, individuals remained on the plot with probability ω, which is the apparent survival probability, that is, the probability of surviving and not permanently emigrating. New individuals entered the plot at a Poisson rate of γ. The sum of the remaining individuals (S) and recruitment (G) determined future abundance (N_ij). These individuals were then detected with probability p. Thus, the model can be described as follows:

Each of the 4 parameters (λ_i, γ_i, ω_i, and p_i) can be modeled as a function of covariates. We used a forward manual stepwise approach to select the variables that best explained both detection probability and abundance.

For the detection probability component of the model, we considered time of day, date, observer, playback, temperature, an index of wind speed, and an index of cloud cover as covariates. Date was represented as the day of the year starting on January 1. We began by running a global model that included all detection covariates, with abundance held constant. We used the P-values from the global model to determine the order in which potential covariates were added during the stepwise selection process (e.g., the first model included only the covariate with the smallest P-value, the second model included the covariates with the 2 smallest P-values, and so on). In subsequent abundance models, as per Burnham and Anderson (2002), we retained any detection variables present in models within 2 Akaike's information criterion points (ΔAIC) of the top model. We kept apparent survival constant in all models. Recruitment was also modeled as constant and was estimated to allow for immigration and to account for serial correlation.

Once we had determined the best set of covariates for describing detection probability, we moved on to identify the best set of covariates for characterizing abundance. Holding detection to be a function of the best set of covariates, we used the same stepwise procedure as described for selecting detection covariates to identify the best model of abundance. We considered the following covariates: (1) a quadratic effect of elevation (hereafter, Elev²), (2) the percentage of native forest cover within 1 km of the survey point (For1km), (3) aspect (Northness and Eastness), (4) canopy height (Ht), (5) canopy cover (Cov), and (6) understory vegetation density (Den). Variables 1–3 were derived from remote sensing data, and variables 4–6 were determined from on-the-ground vegetation surveys. We selected these variables based on previous field experience and range-wide modeling results (McFarland et al. 2013).

RESULTS

We visited 3 of the 99 survey points twice and 96 of the points 3 times, yielding a total of 2,352 secondary sampling occasions. We detected Bicknell's Thrushes at 30 of the 99 survey points. Eighty percent of detections during surveys were of single individuals, and we never detected more than 2 thrushes at a single point. Thrushes were detected during just 5% of the secondary sampling occasions.

Only 39 detections (36%) occurred before the first playback period, illustrating the importance of using broadcast vocalizations for detection of this cryptic species. Playback and air temperature at the start of a survey were present in all supported models for detection probability (Table 1) and were the most significant variables (Table 2). Detection probability was predicted to nearly double after broadcasting vocalizations (Figure 1A). At the coolest temperatures (∼16°C), detection probability was predicted to be ∼0.4, dropping linearly to <0.2 when temperatures reached 27°C (Figure 1B). Additionally, observer and wind speed were important in some plausible models, with predicted effects on detection probability of ∼0.1 between the 2 observers and for the lowest to highest wind speeds during surveys (Figures 1C and 1D).

TABLE 1.

Model selection results for Bicknell's Thrush detection probability (p). All models included expected initial abundance (λ) ∼ global, rate of immigration (γ) ∼ constant, and apparent survival probability (ω) ∼ constant. Data were collected at 99 survey points in the eastern Cordillera Septentrional, Dominican Republic, January–March, 2010. K is the number of model parameters, −2lnL is the maximized log-likelihood, ΔAIC is the difference from the top model in Akaike's information criterion, and w_i is the Akaike model weight.

TABLE 2.

Parameter estimates and P-values from the global detection (p) model for Bicknell's Thrush (Table 1).

FIGURE 1.

Detection of Bicknell's Thrush in relation to (A) playback use, (B) air temperature (°C), (C) observer, and (D) wind speed in the eastern Cordillera Septentrional, Dominican Republic, January–March, 2010. Error bands and bars are ± 1 SE.

Initial abundance varied among points as demonstrated by the inclusion of multiple covariates in all supported models (Table 3). The importance of 5 covariates was clear (Table 4). A quadratic effect of elevation (Elev²) was present in all models (Table 3), with abundance predicted to peak at 552 m (Figure 2A). As expected for a forest-interior songbird, an effect of the amount of forest cover within 1 km² of a survey point was strongly supported, with abundance predicted to rise cubically to >2 thrushes per survey point with >80% forest cover (Figure 2B). Thrushes were predicted to potentially be present at low densities even when forest cover was <20%. Mountain slopes with more easterly and northerly aspects were predicted to have higher thrush abundance (Figures 2C and 2D); these aspects generally correspond to higher precipitation (eastern and northern slopes intercept moist easterly trade winds, leaving west- and south-facing slopes in a substantial rain shadow), and possibly greater wind damage from tropical storms. Vegetation density, a local-scale habitat covariate, was also a good predictor of thrush abundance (Figure 2E), and indicated that Bicknell's Thrush abundance would be greater in naturally disturbed mature forest and advanced secondary forest.

TABLE 3.

Model selection results for Bicknell's Thrush initial abundance (λ). All models included detection probability (p) ∼ Playback + Temp + Obs + Wind + Time, rate of immigration (γ) ∼ constant, and apparent survival probability (ω) ∼ constant. Data were collected at 99 survey points in the eastern Cordillera Septentrional, Dominican Republic, January–March, 2010. K is the number of model parameters, −2lnL is the maximized log-likelihood, ΔAIC is the difference from the top model in Akaike's information criterion, and w_i is the Akaike model weight.

TABLE 4.

Parameter estimates and P-values from the global initial abundance (λ) model for Bicknell's Thrush (Table 3).

FIGURE 2.

Estimates of expected abundance of Bicknell's Thrush in the eastern Cordillera Septentrional, Dominican Republic, January–March, 2010, in relation to (A) elevation (m), (B) the percentage of native forest cover within 1 km of survey points, (C) slope aspect eastness, (D) slope aspect northness, and (E) understory vegetation density based on the top-ranked model (Table 3). Error bands are ± 1 SE.

Predicted apparent recruitment over the winter season was low, whereas predicted apparent survivorship was ∼98% (90% CI: 96–99%; Table 5). Estimates of total abundance for the survey points ranged from 50 (90% CI: 27–105) Bicknell's Thrushes during survey period 1, to 46 thrushes (90% CI: 28–81) during survey 2, down to 42 (90% CI: 25–72) during the final survey.

TABLE 5.

Parameter estimates and 90% confidence limits (CL) from the most general dynamic abundance model considered for Bicknell's Thrush surveyed at 99 point count locations in the eastern Cordillera Septentrional, Dominican Republic, January–March, 2010.

We used the best model that contained only spatially explicit covariates—λ(Elev² + Eastness + For1km + Northness) p(Playback + Temp + Obs + Wind) γ(constant) ω(constant)—to map predicted winter abundance of the Bicknell's Thrush (Figure 3). With the highest elevations in the study area (∼600 m) and the most intact forest cover, the 2 scientific reserves were predicted to contain the largest areas of high thrush abundance.

FIGURE 3.

Expected abundance of Bicknell's Thrush in the eastern Cordillera Septentrional, Dominican Republic, if a count were conducted at the centroid of each 30 m² raster cell, predicted using the model λ(Elev² + E + For1km + N) p(Playback + Temp + Obs + Wind + Time) γ(constant) ω(constant). See Tables 1 and 3 for parameter definitions and model details.

DISCUSSION

Understanding patterns of distribution and habitat use is fundamental to informing conservation actions, especially those aimed at rare and vulnerable species. Identifying high-quality habitat is crucial for effectively conserving species that are experiencing population declines as a result of habitat loss, such as the Bicknell's Thrush. This wet-forest specialist is exceedingly cryptic on its restricted Greater Antillean wintering grounds, which has constrained efforts to elucidate its habitat-specific abundance and evaluate its conservation needs. Yet, without comprehensive knowledge of the species' nonbreeding demographics and habitat use, conservation resources may be inappropriately applied and population declines may be exacerbated.

As an example of how predictive models of abundance can inform conservation, we used the spatially explicit nature of our models to identify locations in the study area predicted to support Bicknell's Thrushes. Applying criteria from our abundance model to a region-wide assessment of available private properties, we identified a 400-ha abandoned farm that contained environmental conditions suitable for Bicknell's Thrushes and that was situated along the Cordillera Septentrional between the 2 existing protected areas to the southeast (Loma Quita Espuela and Loma Guaconejo scientific reserves) and La Salcedoa Scientific Reserve to the northwest, just beyond our study area. Conservation of this area thus offered the possibility of increasing landscape connectivity and protecting potential habitat for the Bicknell's Thrush. Based in part on this assessment, the area was purchased as part of a larger land deal and has now been established as the Dominican Republic's first-ever private conservation reserve, Reserva Privada Zorzal (Figure 3). Named after the local moniker for Bicknell's Thrush, most (∼70%) of the reserve has been set aside for passive and active reforestation designed to create habitat for Bicknell's Thrushes, with the remaining area used for active cultivation of agricultural crops, including shade-grown cacao. Field surveys in 2016 detected Bicknell's Thrushes at 48 of 107 survey points in forest fragments and regenerating stands of secondary forest on this property (H. Almonte personal communication), where our model predicted that Bicknell's Thrushes would occur at low to moderate abundance. Although other enabling factors were critical, notably a legal framework for creating a private nature reserve and the possibility of generating income from part of the property, the potential value of the area for conservation of the Bicknell's Thrush was crucial for making the case for establishing the reserve. We believe that this provides a compelling example of how distributional models, when properly applied, can directly inform conservation action.

We suggest that using our model to identify and prioritize additional lands for conservation of the Bicknell's Thrush in the region is an important next step. The Cordillera Septentrional provides critical wintering habitat for females (Townsend et al. 2011), which are believed to experience disproportionate winter habitat limitations (Townsend et al. 2009a), and is experiencing ongoing deforestation (Kerchner et al. 2010), making the region a top priority for range-wide conservation efforts. Reforestation of degraded lands, by both passive (e.g., natural regeneration) and active means (e.g., planting native trees, controlling invasive species), will be a key component of an overall strategy to conserve winter habitat for this species. Our model provides an initial framework to identify parcels of suitable habitat, the first critical step in ensuring their protection.

More broadly, our findings reinforce the importance of mid-elevation and wet broadleaf forests for conservation of the Bicknell's Thrush. The reliance of Bicknell's Thrushes on a narrow range of biophysical conditions leaves the species highly vulnerable to ongoing deforestation (wet montane forest remains the most endangered habitat on Hispaniola; Perdomo and Arias 2008) and to climate change, which is predicted to yield a substantial drying trend throughout the Caribbean region (Neelin et al. 2006). Long-term conservation of the Bicknell's Thrush, as well as the many endemic and other resident fauna with which they co-occur during winter, will hinge on concerted habitat protection and restoration. Given the limited public resources available for management of protected areas, expanding conservation efforts by enlisting willing private landowners, as with the case study that we have described here, should be considered.