Avian Community Density and Distribution Patterns among Nicaraguan Conventional and Organic Shade-Coffee Plantations

We used a distance-sampling survey method and multivariate statistics to obtain a unique estimate of bird species density and seasonal variation in shade-coffee plantations. Our aim was to determine which cultivation practices among plantations contribute most to bird abundance. We conducted avian species counts at 200 points distributed across 10 shade-coffee plantations bordering the lower slopes of the Mombacho Volcano Natural Reserve, in western Nicaragua. We measured vegetation structure (coffee plants and overstory). We used principal components analysis (PCA) among 14 habitat variables to derive a single phyto-geoclimate summary measure (PGSM). We also used PCA to derive an avian abundance summary measure (AASM) from three bird survey variables, which proved to be a good predictor of bird density. We found higher bird species densities (AASM) in organic and traditional polyculture shade coffee plantations whose structurally complex and diverse overstory could be verified by PGSM. However, this finding was true only for birds that were habitat specialists. Our results provide further evidence


Abstract
We used a distance-sampling survey method and multivariate statistics to obtain a unique estimate of bird species density and seasonal variation in shade-coffee plantations. Our aim was to determine which cultivation practices among plantations contribute most to bird abundance. We conducted avian species counts at 200 points distributed across 10 shade-coffee plantations bordering the lower slopes of the Mombacho Volcano Natural Reserve, in western Nicaragua. We measured vegetation structure (coffee plants and overstory). We used principal components analysis (PCA) among 14 habitat variables to derive a single phyto-geoclimate summary measure (PGSM). We also used PCA to derive an avian abundance summary measure (AASM) from three bird survey variables, which proved to be a good predictor of bird density. We found higher bird species densities (AASM) in organic and traditional polyculture shade coffee plantations whose structurally complex and diverse overstory could be verified by PGSM. However, this finding was true only for birds that were habitat specialists. Our results provide further evidence for promoting organic coffee cultivation practices that maintain a structurally diverse overstory and help retain avian species richness and abundance in coffee plantations.

Introduction
Elevated species richness, abundance and spatiotemporal use of traditional and diverse shade-coffee plantations by resident and migratory birds have been reported throughout the Neotropics [1]- [11]. However, not all coffee plantations provide similar quantities of food, shelter, and anti-predator benefits [12]. Prior research suggests that more diverse and structurally complex organic and traditional polycultures provide for a greater species diversity, abundance, reproduction, survivorship, and dispersal compared to newer, monocultural systems [5] [7] [13]- [18]. Coffee cultivation techniques that affect the structural and floristic diversity of the vegetation (e.g. pruning, application of chemicals) affect faunal populations [8]. Density, basal area and diversity of overstory trees in shade coffee and cacao plantations affect avian diversity and abundance and are often significantly correlated with numbers of forest specialist species [8] [19] [20]. Avian species richness is often the highest in low-intensity management sites near natural forest and forest remnants, and the lowest in high-intensity management sites far from natural forest and forest remnants [7] [21]- [27]. The size of forest remnants and coffee plantations also influences avian species richness, composition and site persistence [7] [19] [28] [29] [30] [31]. Broad-leaf and natural pine forest harbor more habitat specialists and endemics, especially insectivores, relative to coffee plantations [4]. However, shade-coffee plantations often harbor more forest and overwintering generalists, especially insectivores and nectarivores, owing to flowering overstory trees such as Inga spp. with their extrafloral nectaries and their ancillary insects, as well as species of open habitats such as pastures and grasslands [4] [20] [30] [32] [33].
In this study, we replicated several previous, mostly qualitative studies (but see exception [34]) to corroborate their results. But our primary goal was to better quantify the link between vegetational and avian ecological parameters due to coffee cultivation practices. To this end, we derived a phyto-geoclimate summary response measure (PGSM) and an avian abundance summary measure (AASM). These two novel metrics quantify vegetation structure and complexity in association with geoclimate variables, and bird density estimates from species richness and abundance measures, respectively. We predicted higher bird species richness and density in the more vegetatively complex organic and traditional polyculture shade coffee plantations than in coffee plantation monocultures and used PGSM and AASM to test this prediction. Our study objectives were, therefore, to: 1) use of multivariate statistics to characterize vegetation structure and complexity, as well as bird density estimates from species richness

Study Area and Site Selection
We conducted our study within the Mombacho Natural Reserve, Grenada, Nicaragua (Figure 1), by surveying 10 highland shade-grown coffee plantations located between 400 and 800 m.a.s.l. (mean elev. = 558.4 m ± 124.89 SD) within a 7-km radius inside a 6644-ha buffer zone on the lower slopes of the volcano bordering the Reserve (physical, topographical, edaphic, and vegetational characteristics of the volcano are summarized in [35]). All four organic farms studied were certified under the same standards established by the Organic Crop Improvement Association (OCIA International) under a project implemented by the Cooperative League of the United States of America (CLUSA) [35]. The four conventional and two transitional plantations were under a "commercial polyculture" management system (management types, tree species, physiognomy, and husbandry activities are summarized in Philpott et al. [10]).

Habitat Measurements
From each avian survey point (hereafter "plot" as described below), we ran four 25-m transects to the north, south, east, and west to establish a 25-m plot radius encompassing ≈ 0.2 ha in which 14 vegetation-habitat (phyto-geoclimate) variables were measured (Table 1). Because the average plantation was about 62 ha in size, the average number of points was 20 per plantation and the overstory was homogeneous in each plantation, our sampling area is representative of the area used by the birds.
Elevation was measured using an altimeter into four strata (≤10 m, 11 -20 m, 21 -30 m, >30 m); we weighted each stratum by tree density to estimate total leaf volume (≤10 m = density × 1; 11 -20 m = density × 2; 21 -30 m = density × 3; >30 m = density × 4) [10]. We measured maximum coffee plant height with a telescopic meter stick and height of the four tallest coffee plants measured along each of the four cardinal transects. We estimated coffee plant density by counting all vertical coffee stems within an approximate 1.5-m lateral distance along the length of each transect [7]. We measured percent overstory cover with an ocular tube at 12 locations along each transect, starting 1 m from center point and then every 2 m along transects. We pointed the tube vertically into the shade overstory at the center of each point. We measured diameter of all stems ≥ 3 cm 1.3 m from the base (DBH); and counted the number of stems < 3 cm in 1.7-m wide belt transects in the four cardinal directions from the plot center. We recorded the total number of species of overstory trees within the plot; trees were  identified to species where possible, whereas 17 were classified simply as "morpho" species. We calculated the maximum height of overstory shade trees 100-m from their base with a clinometer. We measured total leaf volume of trees separated to determine if foliage (leaves only) was present or absent at a point. To meet the criteria of foliage present, at least 25% of the tube's sighting area had to be covered by foliage. We calculated average percent overstory cover for each of the two-hundred 0.2-ha plots ([(3.14 × 25 m 2 )] × 0.0001). We recorded fruit and flower abundance of overstory trees on a scale from 0 to 4, representing percen-

Avian Surveys
We conducted 10-min, unlimited radius surveys at 200 points that were spaced ca.100 m apart and distributed randomly across the 10 shade-coffee plantations.
Only detections from within a plantation were recorded. We surveyed 83, 63 and 54 plots in 4 organic, 4 conventional, and 2 transitional plantations, respectively (described and tabular summaries in [35] detected during the count, estimated distance to detected birds or to the center of the cluster, i.e., groups of two or more [36], and categorized the observation as occurring in the coffee plants within the understory or the shade-tree overstory. The chronological order of plot visits was randomized to minimize temporal bias in bird detectability throughout the morning. Aerial birds were not recorded unless they alighted in vegetation during the 10-min count. Because cloud cover can be a major hindrance to avian surveys at high elevation cloud forest sites, we estimated percent cloud cover on a scale from 0 -100 in increments of 25 (0 = clear skies, 100 = overcast). Similarly, high winds can drown out bird vocalizations on exposed slopes. Therefore, we recorded wind speed using the Beaufort scale (http://www.spc.noaa.gov/faq/tornado/beaufort.html).

Avian Species Richness, Abundance and Density Estimation
We compared species richness and abundance by ANOVA statistical methods among plantations grouped by the three coffee cultivation practices: Conventional, Organic and Transitional. We estimated bird density per hectare using distance models [37] for species with ≥30 observations in program Distance 6.0 [36]. This method accounts for birds present but not detected by fitting a detection function, P, to observed counts for a given distance from the observer. We filtered the data by species and distances ≥ 0 to account for missing distance values. We truncated data to detections with distances ≤ 40 m and grouped observations by 0 -10 m, 10 -20 m, 20 -40 m, and 40 -60 m, except for raucous species that were audible at long distances: Yellow-throated Euphonia (Euphonia hirundinacea) and Hoffmann's Woodpecker (Melanerpes hoffmannii) (truncated at 85 m) and Keel-billed Toucan (Ramphastos sulfuratus) (truncated at 100 m).
The truncation distance for each species was determined by the maximum distance at which the species was audible, except for the toucan, which, although audible at distances greater than 100 m, was truncated at our designated maximum distance of 100 m. We did not include birds detected more than 40 m in the direction of points behind us as not to count the same individuals in areas of overlap. We used a half-normal detection function with a cosine series expansion to fit the data. Because each point was visited four times, we included survey effort as a multiplier, which allowed us to divide density by effort. Because some species, e.g., parrots, were often detected in groups, designated as clusters, we included the total number of individuals detected per point at each distance and specified the size-bias regression method to estimate group size. For species with >50 detections, we compared support for the global model internal to the Distance 6.0 software, with models including observer, period, and observer + period as covariates in the multiple covariate distance sampling (MCDS) engine. The most supported model had the lowest Akaike's Information Criterion (AIC c ) value (tabulated as ΔAIC c = 0). We report ΔAIC c from models evaluated for each species (Table 2). For species with fewer than 50 detections, we estimated density per hectare using a global detection function. We report detection probability (P), effective detection radius (EDR), cluster size, and density for each management type based on the top model (Table 3).

Statistical Design and Modeling
To examine patterns of seasonal bird density per hectare, we report density estimates for each species by period (Table 4), which are explained below.
Habitat variables not otherwise quantified on a measurement scale, but estimated as percentages (e.g., canopy cover and trees with flowers or fruits) were normalized via the arcsine-square root transformation, a common variancestabilizing technique. From the bird species observation data, we selected only those species with >= 30 detections for further analysis. These species were considered to have sufficient sample size for comparison. The intent of the design was to estimate the effects of shade-coffee cultivation practice (management system) and sampling period on a bird species diversity-richness response measure using simple linear regression (SLR) [38]. Species with a larger number (≥50) of detections allowed us to model the co-variation in this response with phyto-geoclimate variables per cultivation practice as well using the method of analysis of covariance (ANCOVA) [38]. The multivariate method of principal components analysis (PCA) [39] was used as a dimensionality reduction tool among the 14 phyto-geoclimate variables (i.e. dimensions). Each of the 14 principal components is ordered in magnitude (PC1, PC2, …, PC14) in accounting for the largest to the smallest percentage of total variation explained among the 14 original variables. We employed the first principal component (PC1) exclusively as a univariate summary measure. It characterizes in one dimension the largest amount of environmental information at each of the 200 avian sampling plots among the 10 coffee plantations. This method partitions the information in a correlation matrix comprising 14 C 2 = 91 combinations of pairwise correlations among the phyto-geoclimate variables [40]. The first eigen value of that matrix is also an indication of the number of dimensions represented in our phyto-geoclimate summary response measure (PGSM) or PC1. Unlike the foliage height diversity (FHD) measure of MacArthur and MacArthur [41], PGSM characterizes the habitat on a more continuous scale instead of just 2 to 4 height classes.
In addition, multivariate analysis of variance (MANOVA) and discriminant function analysis (DFA) [39] were used to discriminate differences among the vegetation-habitat characteristics of shade-coffee plantations grouped by organic, conventional, and transitional coffee cultivation practices. MANOVA determined the strength of evidence in favor of the hypothesis that the multivariate phyto-geoclimate measurement response vectors differ significantly among the three coffee cultivation practices [39]. DFA also identifies those phyto-geoclimate variables that contribute most to the discrimination among these three groups of measurements, that is, among their multivariate mean vectors called centroids [42].
DFA was conducted via a stepwise forward-selection procedure [42]. Tree species richness and diversity estimates were obtained using EstimateS 8.2 software [43].
To summarize bird abundance at each of the 200 sampling plots, we performed a similar PCA using three bird abundance variables: total bird count, mean distance to detection, and the standard deviation of detection distance. This PC1 is referred to as the avian abundance summary measure (AASM).
Density estimates using the Distance 6.0 software [36] for each of 21 species were also calculated for each of the four sampling periods, for each coffee cultivation practice. Shannon-Wiener and Simpson's diversity indices [44] were used to compare species diversity among management types.
We used SLR by the standard least squares method to compare the AASM scores for bird abundance to the calculated bird density measurements among the four sampling periods for all bird species. SLR was also used to determine the dependence, if any, of AASM on PGSM. Since there is measurement error in both, an orthogonal regression [45] was also performed for an alternate and perhaps more accurate explanation of the relationship between the two summary measures.
The AASM was further aggregated by sampling plot for each of two species groups-one group with relatively steep (large valued) SLR slopes between AASM vs bird density and another group with relatively shallow (small valued) SLR slopes.
We used an analysis of covariance (ANCOVA) model [38] to discover the effects of coffee cultivation practices on the AASM after accounting for its expected relation with the PGSM. All statistical analyses were performed using the JMP® [46].

Vegetation and Habitat Analysis
We found that tree species diversity is higher under organic and transitional management types compared to conventional. The Shannon-Wiener diversity index was highest in organic farms (organic = 2.44, conventional = 2.28, and transitional = 2.29). Although the Simpson's Diversity index was higher in tran-

Bird Species Density vs. Coffee Cultivation Practice
We documented 6110 audio-visual detections of 98 bird species, of which 13 were identified only to genera (Appendix). The average number of species (richness) among the two transitional plantations (avg. = 57) and the four organic plantations (avg. = 51) was greater than that of the four conventional plantations (avg. = 47), although not significantly significant (p > 0.05) ( Table   5). All bird species were more abundant in the overstory than in coffee plants, except for two species of wren, Cantorchilus modestus and Thryophilus pleurostictus.    The PC1 (AASM hereafter) from this PCA explained 82% of the variation among its three constituent variables, viz., bird count, mean distance to the detected individual/cluster, and the standard deviation of that detection distance.
The AASM also summarized ~2.5 of the 3 dimensions (see Methods above). The The entire data set was then re-aggregated to produce a new response vector from three variables, viz., total count, mean detection distance, and the standard deviation of detection distance, for each of the two species groups defined in included an interaction effect. For the generalist species group, the results showed no effect due to cultivation practice, but a significant period effect (P < 0.001) due to low bird counts for these species in period 2 ( Figure 6(A)). Neither was there a significant cultivation practice by period interaction effect. In contrast, the specialist species group showed a significant (P < 0.001) cultivation practice effect (Figure 6(B)), but no period or interaction effect.
Subsequent ANCOVA models were also fit separately for AASM g and AASM s to test for a cultivation practice effect after accounting for the relationship with the covariate (i.e., the PGSM). Results showed no significant regression relation between the AASM g vs. PGSM scores, or any evidence that the slope of this relation differs from zero among cultivation practices (Figure 7(A)). However, results for the habitat specialists showed a significantly (P = 0.008) positive overall regression relation between the AASM s vs. PGSM. In addition, at the average value of the covariate (vertical dotted line at zero Figure 7(B)), the least squares mean bird abundance/density was significantly higher in organic coffee plantations, than in either conventional or transitional coffee plantations (P = 0.07 and P = 0.005, respectively). Note that we regard a 7% chance of Type I error as indistinguishable from 5%.

Factors Affecting Bird Species Richness, Abundance and Density
Bird density estimation using methods that incorporate detection probability, such as the distance sampling method [37], is generally viewed with preference to relative abundance measurement only; albeit, results are dependent on meeting strict model assumptions, which are often hard to achieve. Also, by our not including birds detected more than 40 m in the direction of already surveyed plots behind us, our bird densities (per hectare) may be slightly lower.
However, our use of PCA produced a dimensionality reduced measure (AASM) of bird abundance per sampling plot that also yielded predictive information about bird density estimates from distance sampling calculations. Although this relationship has only moderate precision (R 2 ≅ 0.50), it is nonetheless significantly positive. The slope of this relation was bird species, dependent as illustrated in Figure 4, and was markedly different between two groups of bird species, which groups corresponded to habitat generalists and specialists.
Among the five species with steep slopes for bird density vs. AASM, two are Nearctic-Neotropical migrant warblers, Yellow Warbler Setophaga petechia (SETPET) and Tennessee Warbler Oreothlypis peregrina (OREPER) wintering in Nicaragua. A third is a resident corvid, the White-throated Magpie Jay Calocitta formosa (CALFOR). All three species are well known habitat and resource ge- The two migratory species (SETPET, OREPER) had higher densities than any of the resident birds as expected because the study site lies within the Central Americas Flyway used by Nearctic-Neotropical migrants, many of whose numbers greatly increase during fall and spring passage through Nicaragua.
Note that the greater percentage of detections for SETPET and OREPER in conventional, rather than either organic or transitional plantations (~50% in We believe it is unlikely that coffee cultivation practices would affect all birds in the same way. We found, in fact, no relation between AASM and PGSM for 5 habitat generalist species of the 21 species most frequently counted among farms. The same finding was also true for the remaining 16 habitat specialists among farms in transition or using conventional coffee cultivation practices. Only among farms using organic cultivation practices, was bird abundance/density (AASM) significantly and positively related to PGSM among habitat specialists.
The ANCOVA results substantiate our hypothesis that SETPET, OREPER, CALFOR, AMASAU and HYLELI are generalists that do not necessarily discriminate among habitats, and also explain the lack of statistical evidence of a relationship between PGSM and AASM. However, among the remaining 16 species wherein no migrants were represented, these behaved as habitat specialists in organic shade-coffee plantations, meaning that there was substantial value in PGSM as a predictor AASM and a positive correlation between them. This may be interpreted ecologically much like MacArthur and MacArthurs' [41] and Morton et al.'s [61] findings identifying, respectively, foliage height density and stem orientation as measures of habitat structure. Birds in these cases, i.e., those that visually discriminate the structure of the vegetation may perceive a greater potential as refugia from predation and the possibility of greater niche partitioning of 3-dimensional spatial resources including foraging strata and favorable sites for nesting within a given physical location of the forest and within particular vegetation types.

Conclusion
Benefits of Organic Farms and Other "Biodiversity-Friendly" Coffee Systems Our study demonstrates that bird abundance and density per hectare are higher within farms using organic shade-coffee cultivation practices. We assert that this is due to their incorporation of more mature (e.g., taller, structurally complex) tropical forest ecosystems, especially among habitat specialists. Our results confirm that shaded organic coffee plantations, which are most similar to

Author Contributions
All of the authors contributed to the conception and design of the study, acquisition of data, or analysis and interpretation of data. Martínez-Sánchez and Zolotoff assisted in summarizing coffee management strategies among the farms: organic and traditional polyculture shade coffee plantations. Zolotoff assisted Arendt (among others-see Acknowledgments below) in collecting the field data. Arendt wrote most of the ms. and conducted some of the traditional statistical analysis (parametric, non-parametric). Tuckfield, Thompson, and Reidy performed specialized statistical procedures (modelling, Distance Sampling, R, among others). Tuckfield derived the PGSM and AASM metrics, two novel ways to quantify vegetation parameters and avian species densities per unit area sections, wrote the related sections, and made numerous edits, adding wording and constructive comments throughout the manuscript's entirety. All authors provided editorial suggestions on various sections and drafts. Table A1. Species codes, scientific and common name, residency status (R = resident; M = migrant), foraging guild (C = carnivore; F = frugivore; G = granivore; I = insectivore; N = nectarivore; O = omnivore; S = scavenger) and strata (O = overstory; G = ground; S = shrub; T = trunk; U = understory), number of observations (total number of points at which a species was detected), and total detections (total detections of individuals of a species) and detections by period and overstory and coffee at 200 count points visited four times inshade coffee plantations on Mambacho Volcano, Nicaragua, 1998Nicaragua, -1999