The Dominican Republic is experiencing increased forest loss due to the expansion of avocado farming. While agricultural lands can provide viable habitat for birds, knowledge of avian use of avocado farms relative to native forest is limited. We conducted surveys in fall 2021 to compare the bird communities and vegetation complexity of two avocado farm sites and two native dry forest sites in the northern Sierra de Bahoruco region of the Dominican Republic to assess avian habitat use. Overall, we found that habitat associations differed for bird species distinguished by contrasting geographic ranges. In particular, the abundance, richness, and diversity of bird species endemic to Hispaniola were significantly greater in forests than in farms, while non-endemic resident bird species had the opposite habitat associations. The abundance, richness, and diversity of Neotropical migrants exhibited fewer distinct differences between farm and forest sites. Vegetation structural complexity was reduced in avocado farms relative to forests and was positively associated with the abundance, richness, and diversity of endemic bird species but negatively associated with these community measures of non-endemic resident species. Collectively, our findings suggest that avocado farms could provide viable habitat for non-endemic resident and migrant bird species, but that endemic species could be negatively impacted by an increase in avocado farming in the region. We offer recommendations for forest preservation, farming practices, and future research that could help to inform the conservation of endemic species in the region.

Study Sites
The study area is located near the town of Puerto Escondido, Independéncia Province, in the foothills of the northern Sierra de Bahoruco in the southwestern Dominican Republic. Natural forests in this area are considered subtropical dry forests (Knudson et al. 1988) with elevations ranging from 421 to 522 m. Dominant vegetation in the region includes large trees such as baitoa (Phyllostylon brasiliensis), candelón (Acacia scleroxyla), and guayacán (Guaiacum officinale) and understory shrubs such as escobón (Eugenia monticola) and grenadilla (Eugenia ligustrina). The town of Puerto Escondido is situated near multiple limestone water outlets, providing water for community needs and agricultural irrigation. We conducted field work in four study sites within the area: two native forest sites and two avocado farm sites. Avocado farms of two sizes were chosen for the comparison to reflect the variety of such farms in the region, including potential differences in vegetation complexity. Specifically, we chose one farm site comprised of multiple, relatively small community parcelas; this is land near town that is divided into adjacent plots, each with a single owner who primarily cultivates one crop. All parcelas cultivating avocados were included in the small farm study site, which was located within 2 km of native dry forest dry forest along a portion of a well-known birding trail called Rabo de Gato (Figure 1). The other forest site was located in Reserva Loma Charco Azul, a federally protected 17,400-ha dry forest reserve (Figure 1), known for being the main habitat of the endemic Bay breasted Cuckoo. A second farm site was located in a large organic commercial farm characterized by groves of primarily Hass and Carla avocados. Due to restricted permissions by the landowner, we are unable to disclose the specific location of our large farm site. Both avocado farm sites border native forest, which minimized potential differences in climate, topography, and human disturbance as a result of proximity to town.

Bird Community Surveys
We used point-count surveys (Wunderle and Waide 1993, Wunderle and Latta 1996, Greenberg et al. 2000, Marsden et al. 2001, Latta et al. 2003, Kennedy et al. 2010) to assess overall, endemic, and migrant bird communities in the study sites. We designated survey points using a grid system in which all points were at least 150 m apart to reduce the possibility of double-counting individual birds. We also ensured that all points were greater than 25 m away from the boundary of the site so that only species within the site were detected during surveying. The size and elevation of our study sites and the number of survey points in each site are listed in Table 1. We conducted point count surveys from October 13 to December 1, 2021, to include overwintering migrant species. We visited each point three times during the field season, and alternated between forest and farm point counts on consecutive days to account for differences due to the effects of weather or other short-term temporal events. We began surveys at sunrise and ended no later than four hours after sunrise, encompassing the period when birds are most active and likely to be detected. We recorded all birds seen and/or heard within 25 m of each point for a 10-minute period. Using relatively small point-count radius ensured that we were recording bird species within the habitat type we were surveying, which was especially critical in smaller avocado farms. Birds that flushed upon arrival were included in the point count, but birds flying over without stopping in the site were not counted. We averaged the number of individual birds and species across the three survey dates at each point. To investigate habitat use by non- endemic resident, endemic, and Neotropical migrant species, in particular, we similarly calculated the mean number of individuals and bird species within each group per survey point. Endemic bird species were defined as those only found in Hispaniola and its satellite islands (Latta et al. 2006), and migrant bird species were defined as those that breed north of the Tropic of Capricorn but winter south of this location (Hagan and Johnston 1992). We used the term ‘non-endemic resident’ to describe more geographically widespread, non-migratory species observed in our study sites.

To characterize the diversity of the non-endemic resident, endemic, and migrant bird communities in each study site, we calculated the frequency of each species identified as the percentage of the total bird survey points in each site in which that species was observed across all three site visits. We used the Shannon Diversity Index (Shannon 1948) to calculate the diversity of the overall, non-endemic resident, migrant, and endemic bird communities in each site during each visit. We averaged the diversity scores of the three visits to calculate a mean diversity measure per point. Bird survey details, including the distance that birds were identified from survey points, method of detection, weather variables, and ambient noise are provided in a supplementary dataset.

Vegetation Complexity Measures
We assessed canopy cover, canopy height, and canopy thickness at each bird survey point and at four additional subpoints, one in each cardinal direction 25 m from the main survey point (see Van Bael et al. 2007). Canopy cover was estimated by using a tube densiometer at each survey point and subpoint. To determine canopy height and canopy thickness, we used a digital rangefinder to measure height of the highest and lowest foliage associated with trees at each survey point and subpoint. Canopy thickness was calculated as the upper height minus the lower height at each measurement location (see Muth and Bazzaz et al. 2002, Mas and Dietsch 2004, Lee et al. 2020). ​​We averaged the vegetation data collected at each survey point and its subpoints to calculate the mean canopy cover, canopy height, and canopy thickness for each survey point. To assess the diversity of vegetation in each study site, we identified the five most dominant woody species greater than 1-m tall in a 25-m-radius circle around each bird survey point and visually estimated the percent cover of each of these species (see McClaren et al. 2019). The five woody plant species with the highest vegetative cover within the circle were identified as dominant species at that point. For cultivated crops, we included horticultural varieties as ‘species’ in our analyses given that horticultural vareties are selected for distinct differences in traits that could potentially impact habitat preference, including leaf shape and size and fruiting phenology. To characterize the diversity of the vegetation in each study site, we calculated the frequency of each dominant plant species as the percentage of the total survey circles in each site in which that species was observed across all three site visits.

Statistical Analyses
We tested for the normality of our data distribution within groups (i.e., sites) with Shapiro-Wilk test and for homogeneity of variances with Levene’s tests. When the assumptions of normality and homogeneity of variances were both met, we used a one-way ANOVA to test for the effect of site on vegetation complexity (i.e., canopy height, canopy cover, canopy thickness) and bird community measures (i.e., number of individuals, species richness, species diversity) followed by least-significant difference LSD post-hoc tests when the effect was significant. When the assumption of normality was met but variances were not homogenous, we used Welch’s ANOVA followed by Games-Howell post-hoc tests when the effect was significant. When the assumption of normality was not met but variances were homogenous, we used non-parametric Kruskal-Wallis tests to test for the main effect of site followed by Dunn-Bonferroni post-hoc tests when the effect was significant. In all cases, differences in mean values were considered significant if p is less than or equal to 0.05. To examine associations between vegetation complexity and bird community measures, we performed linear regression analyses in SPSS. To account for the numerous associations analyzed, we calculated and utilized corrected p-values to control the false discovery rate for the group of tests for each bird community type (i.e., non-endemic resident, endemic, migrant) in accordance with methods described by Benjamini and Hochberg (1995). Results of regression analyses were considered significant if FDR-corrected p is less than or equal to 0.05. All data analyses were conducted with SPSS (version 29, IBM Corp., Armonk, New York, USA).