Bird community response to fruit energy


Correspondence author. E-mail:


1. The abundance and predictability of food resources have been posited as explanations for the increase of animal species richness in tropical habitats. However, the heterogeneity of natural ecosystems makes it difficult to quantify a response of animal species richness to these qualities of food resources.

2. Fruit-frugivore studies are especially conducive for testing such ecological theories because fruit is conspicuous and easily counted. Fruit-frugivore research in some locations has demonstrated a relationship between animal abundance and fruit resource abundance, both spatially and temporally. These studies, which typically use fruit counts as the variable of fruit abundance, have never documented a response of species richness at the community level. Furthermore, these studies have not taken into account factors influencing the detection of an individual within surveys.

3. Using a combination of nonstandard approaches to fruit-frugivore research, we show a response of bird species richness to fruit resources. First, we use uniform and structurally similar, one-ha shade-grown coffee plots as replicated experimental units to reduce the influence of confounding variables. Secondly, we use multi-season occupancy modelling of a resident omnivorous bird assemblage in order to account for detection probability in our analysis of site occupancy, local immigration and local emigration. Thirdly, we expand our variable of fruit abundance, Fruit Energy Availability (FEA), to include not only fruit counts but also fruit size and fruit quality.

4. We found that a site’s average monthly FEA was highly correlated (0·90) with a site’s average bird species richness. In our multi-season occupancy model 92% of the weight of evidence supported a single model that included effects of FEA on initial occupancy, immigration, emigration and detection.

5. These results demonstrate that fruit calories can broadly influence the richness of a neotropical bird community, and that fluctuations of FEA explains much of the site occupancy patterns of component species. This study shows that in depauperate, managed landscapes fruit resource abundance supports more species and fruit constancy allows for higher levels of avian persistence, an important practical concept for conservation planning.


Forest birds are among the many groups of organisms that show distinct and consistent gradients in diversity, both between the tropics and temperate zone and within the realms. The underlying causes of these diversity gradients have been a major focus of ecological studies (Macarthur & Macarthur 1961; Pianka 1966; Karr 1968; Connell 1979; Rosenzweig 1992; McCain 2009). In the case of mobile organisms, such as birds, the abundance, diversity and predictability of food resources have frequently been invoked as potentially important in their contribution to increased diversity in tropical habitats (Schoener 1974; Karr 1976; Holmes, Bonney & Pacala 1979; Foster 1980; Levey 1988; Loiselle & Blake 1991; Poulin, Lefebvre & McNeil 1992; Christiansen & Pitter 1997; Kissling, Rahbek & Bohning-Gaese 2007; Kennedy, Nislow & Folt 2008). The specialized frugivory guild, for example, has been shown to contribute substantially to the increase in the number of birds in tropical vs. temperate forests (Fleming, Breitwisch & Whitesides 1987; Kissling, Bohning-Gaese & Jetz 2009), as well as forests of different rainfall regimes within the tropics. Fruit abundance alone may not account for the variation in the contribution of frugivores to overall diversity. For example, it has been posited that the major ecosystem-wide phenological gaps in fruit resource availability are responsible for the total absence of the exclusively frugivorous bird guild in the temperate zone (Stiles 1980, 1984; Kricher 1997). In addition to the specialized frugivorous bird guild, variation in food availability can affect a much larger number of omnivorous bird species that use substantial amounts of fruit. Unlike insectivorous bird species, the nutritional intake of partial frugivores needs to be balanced between fruit and arthropod resources and among different types of fruits. Fruit resources can be low in protein, lipids and minerals (Rode et al. 2006) requiring frugivores to seek these nutrients in animal matter in addition to locating fruit resources for essential fatty acids or minerals (O’Brien 1998; Nelson et al. 2000).

However, it is difficult to isolate the functional role that the abundance and predictability of a single resource type has and tease this apart from the many factors that vary at large biogeographical scales. It is, therefore, important to examine the role of food abundance and predictability in supporting increases in local diversity among assemblages experiencing similar climate and with a shared fauna. For some particularly complex ecosystems, such as moist tropical forests, ecological patterns and processes may be difficult to detect because of the inherent diversity of the systems (Wilson & Simberlo 1969; Wilbur 1976). Some of these processes can be more easily investigated in managed systems (such as agroforests) that share many of the ecological attributes, but are simplified in diversity and structure and more spatially homogeneous (Greenberg, Perfecto & Philpott 2008). Generally, ecological research within natural habitats has failed to demonstrate that the frugivore community does in fact respond to different levels of fruit resource availability (Moegenburg & Levey 2003; Saracco, Collazo & Groom 2004; Haugaasen & Peres 2007). However, the heterogeneity of the natural areas where these studies have been conducted may make it difficult to detect a community wide response to changes in food resources on a local scale.

Separating the role of energy availability and its predictability through time may be facilitated by studying simplified habitats with fewer resources. Island or mesocosm studies are two such simplified habitats, which have proven valuable in advancing ecological theory (Morin 1981; Thornton, Zann & Vanbalen 1993; Rowe & Dunson 1995; Schipper et al. 2001). However, island studies are often confounded by having distinct evolutionary histories at all trophic levels (Shanahan et al. 2001). Managed agricultural habitats that mimic some of the characteristics of natural forests provide another opportunity for examining the role of ecological factors controlling diversity in a simplified system (Cruz-Angon, Sillett & Greenberg 2008; Greenberg et al. 2008; Perfecto & Vandermeer 2008). Agricultural landscapes provide replicated, uniform patches where the confounding effect of natural heterogeneity can be held approximately constant while the regulating factor of interest is experimentally varied to study animal population or community response. In addition, the spatial and temporal distance between fruit resources in actively managed agricultural lands can be much greater than levels that existed prior to deforestation. Therefore, fruit resource or nutrient availability may be a limiting factor regulating bird populations in depauperate managed lands (Carlo, Collazo & Groom 2004; Rode et al. 2006).

Beyond contributing to our basic understanding of the effect of food resources on biodiversity in tropical forest systems in an abstract sense, addressing the ecological question of how fruit resource availability affects species richness patterns in the ever-increasing agricultural landscape is an urgent question to ask in the face of predicted species extinctions in regions of high biodiversity (Sekercioglu, Daily & Ehrlich 2004). Small scale agricultural lands in the tropics contribute to avian species survival because they are typically embedded within a patchwork of windbreaks, isolated trees and small forest patches or riparian zones (Graham 2001; Luck & Daily 2003; Sekercioglu et al. 2007; Tscharntke et al. 2008); however, only recently research has aimed to understand specifically how individual habitat components of these various patches contribute to local patterns of avian species richness or abundance (Graham 2001; Sekercioglu et al. 2007; Cruz-Angon et al. 2008). Modelling patterns of seasonal site occupancy by bird species within one widespread habitat type that is commonly part of the patchy agricultural landscape, i.e. shade-grown coffee plots, can provide insight into answering the question where within this patchwork do birds spend their time and why? Fruit-consuming birds with higher levels of frugivory often locate more fruit resources in forest patches than in actively managed agricultural lands, and therefore may rely more heavily on these fragments (Graham 2001; Sekercioglu et al. 2007). In Northern Latin America almost half (42%) of all permanent croplands are dedicated to the production of coffee (FAO 2000, 2005), and as such it is critical to conduct more ecological research to achieve a better understanding of the contribution towards species survival proffered by this extensive tropical agroecosystem. Due to the vast amount of land dedicated globally to the production of coffee as well as the high value that the coffee agroecosystem has for conservation of many animal species within fragmented landscapes (Greenberg et al. 2008), research aimed at understanding how this agroecosystem can most effectively support local species diversity should be a top priority for conservation science. Furthermore, high-quality coffee is increasingly being grown in an agro-forestry poly-culture system where higher numbers of individual tree species can be incorporated than that typically found in other agricultural production systems, such as sugar cane or cattle pasture. These features make the coffee agroecosystem the most appropriate location within the agricultural matrix to test animal community response to fruit resource availability within replicated, homogeneous plots.

Birds which rely on fruit to some degree may select resources, and therefore their location within the agricultural matrix, based on any or all of the following: the number of available fruits or crop size (Whelan & Willson 1994; Saracco et al. 2005); vital nutrients or secondary metabolites (O’Brien 1998; Cipollini 2000, Tsahar, Friedman & Izhaki 2002); or for their complementariness (Whelan & Willson 1994; Graham 2001; Witmer 2001). Previous research to assess the factors responsible for fruit choice, bird abundance or bird species richness has tended to use one of the above explanations alone, rather than formulating innovative ways to unite them. Here, we define and use a new predictor variable of fruit resource abundance, which we term gross Fruit Energy Availability, or FEA. The FEA of an individual tree can combine both the available fruit mass and an indicator of nutrient value (caloric value of an individual fruit). FEA may be a better indicator of the value of a tree or a site to fruit-consuming birds, as well as elucidate limiting factors responsible for constraining species richness of birds within the agricultural landscape. Additionally, if FEA influences bird seasonal occupancy of managed coffee plantations within the agricultural matrix, then these plantations may be managed to increase their capacity to contribute to the conservation of bird populations.

Here we study a generalized assemblage of fruit-consuming bird species and fruit-bearing plant species in replicated, structurally similar shade-grown coffee plantations to examine how accurately measurements of fruit abundance can predict bird species richness in managed landscapes. Furthermore, we use a multi-season occupancy modelling approach to examine the role that FEA has in predicting bird assemblage occupancy patterns (i.e. initial site occupancy, site immigration and site emigration). Specifically, our occupancy model allows us to test the performance of our predictor variable of fruit abundance, FEA, while taking into account daily and species-specific detection probabilities and to estimate the relationship between FEA and site occupancy patterns of unspecialised, omnivorous birds. We ask: (i) How well can fruit abundance measurements alone explain the variation in local bird species richness? (ii) Do site occupancy models more accurately approximate the truth when FEA is included in the model as a predictor variable for detection probability, site initial occupancy, site immigration and site emigration?

Materials and methods

Study site

Research was conducted at the University of Georgia’s San Luis Research Station in the San Luis valley of Northwestern Costa Rica (c. 10°17′N, 84°48′W) in the Municipality of Puntarenas. The San Luis Valley is the south-westernmost border of the Monteverde Cloud Forest Preserve, with an elevation between c. 500 and 1200 m a.s.l. Average monthly rainfall during dry season months (December through April) is c. 50 mm and during wet season months (May through November) is c. 550 mm. Due to conservation efforts in the region, the 56 ha ‘environmentally-friendly’ farming area, Finca La Bella, was created to provide formerly landless farmers with a 1 to 2 ha tract of land for subsistence farming. Finca la Bella (c. 900–1150 m asl) is a 56 ha matrix of community owned forest patches and windbreaks which subdivide c. 30 private landholdings of pastureland, sugar cane plantations, coffee plantations and other agricultural lands (Fig. 1). The small coffee plantations embedded within and around this matrix share similar management practices, history and surrounding matrix quality (Perfecto & Vandermeer 2002), and would be classified as Traditional Polyculture according to Moguel & Toledo’s (1999) classification.

Figure 1.

 Aerial photo of study area near Monteverde, Costa Rica, and study sites located in the San Luis Valley. Light gray patches are cattle pasture and darker areas are forest patches, coffee plantations or secondary growth forest. Black circles represent study sites.

A study site was defined as a one-ha coffee farm located within or near the farming area Finca la Bella. Sites were selected based on a visual assessment of their structural similarity, i.e. large trees (max height: c. 20 m, site mean height range: 8–10 m) interspersed throughout the entire coffee plantation as well as several single species (Croton niveus Jacq. or Montanoa guatamalensis B. L. Rob.) tree windbreaks of lower stature (height: <6 m). Additionally, farm structure was measured through percent canopy cover, counting the number of tree species as well as the number of tree species with fruits consumed by birds, and the number of individual trees per farm. A tree species was considered to bear fruits consumed by birds once a bird was observed foraging on fruits from the species. Site canopy cover was estimated along two parallel 100-m transects, reading canopy cover in the four cardinal directions with a convex densiometer at 10-m intervals. Although plantation management practices such as high diversity of tree species and low chemical input do not differ among farms, the amount of ripe fruit resources do differ among farms and among months, thereby creating a natural experiment of variable resources in a constant structural environment.

Bird sampling

Bird surveys were conducted during all months of 2008 except January and September in six coffee plantations, coded ALV, GIL, ODI, OLD, OLI and RAF, where site names represent the first letters in the names of the farm owners. One observer recorded all individual birds detected along fixed 100-m transects, conducted by walking down slope, through the middle of each coffee plantation. Species were recorded based on both sound and visual sightings, but primarily through visual sightings. The observer surveyed the same 100-m transect four times each morning and each survey lasted a total of 45 min. On each survey a bird species could either be detected or not detected. Each sampling day lasted c. 4 h, between the hours of 07.00 and 11.00, and the observer visited only one farm per day. Farms were randomly sampled for 3 days per month yielding a total of 12 sampling surveys per farm per month (four rounds × 3 days = 12 sampling surveys), except in April when a total of eight sampling surveys were conducted per farm. Before the start of each survey maximum and average wind velocity and ambient temperature were recorded using a Weather Kestrel (2500, Nielsen-Kellerman, PA, USA). Surveys were not conducted during heavy rains. We recognize that repeated surveys on the same transect are not independent data. Repeated observations, however, are required for the estimation of detection probability in occupancy analysis (see below). The time of day, transect number, bird species, activity and substrate for each individual observation were recorded. The bird’s activity was chosen from the following categories: Perching, Vocalizing, Foraging, Moving, Nest Building, or Flew In/Flew Out. The foraging activity category was only used when the bird was observed with a food item, and the food item was identified as Fruit; Nectar; Arthropod or Other. Substrate categories included: on ground, Coffee Plant, Croton Windbreak, Aster Windbreak, Tree or Shrub < 4 m, Tree > 4 m, Epiphyte or Vine. The plant species on which the bird was located was also identified and recorded.

Fruit sampling and FEA

Fruit Energy Availability for each coffee plantation was assigned on a monthly basis during all months of 2008 except January and September. Monthly FEA values incorporated the total number of ripe fruits per plant species, species-specific fruit calories and species-specific fruit size. The number of ripe fruits available to birds were counted for each individual tree or shrub in the coffee plantation, and then assigned into a corresponding index of fruit availability. The following Fruit Availability Index (FAI) was used for each individual plant bearing ripe fruit as 1 = 1–10; 2 = 11–25; 3 = 26–50; 4 = 51–100; 5 = 101–200; 6 = 201–500; 7 = 501–1000; 8 = 1001–10,000. All individual trees, shrubs, or epiphytes were classified into the FAI by counting all ripe fruits or, for larger trees, counting a random selection of branches and extrapolating this value to the number of branches in the entire tree. The un-harvested ripe fruits of tree species planted for human consumption, such as Psidium guajava L., Citrus spp. and Musa spp., were also included in the FAI. Individual plant FAI’s were assigned three times per month after concluding bird surveys on a farm, and the highest FAI for each individual plant was used as the monthly FAI for that individual. This allowed us to best approximate the total amount of ripe fruit available to birds throughout the month (i.e. accounting for fruits that may have only been available briefly, such as fruits harvested by farmers, fruits that became ripe in the latter part of the month, or fruits that were totally consumed by birds within the first half of the month).

The other two components of FEA, species-specific fruit calories and species-specific fruit size, were measured from c. 10 to 30 fruit samples, which were collected from each plant species. Seeds were removed from the fruit pulp to obtain the fruit species’ wet mass and for caloric content analyses. Fruits removed for caloric analysis were not included in the FAI as fruits were no longer available to birds. After fruits were weighed, fruit samples were dried at 55 °C for c. 12 h. Caloric values were measured on a per gram basis using a bomb calorimeter (Parr 1563, Parr Instrument Company, IL, USA).

Individual FEA values for each individual plant with ripe fruit was then calculated by multiplying the midpoint (N) of the assigned individual plant FAI by the species wet mass in grams and the per gram caloric value of the fruit species so that FEAi = NiwiCi where Ni is the number of ripe fruits on an individual plant, wi is the mean fruit wet mass in grams for a plant species, and Ci is the mean kcal per gram of fruit for a plant species. For each month, then, the assigned FEA value for a farm represents the sum of each individual plant FEAi; FEA = ΣNiwiCi. We realize that bomb calorimeter derived calories may not be equivalent to the caloric value of the fruit actually assimilated. For the purposes of this analysis we assume that the calories measured in ripe fruit pulp are a reasonable approximation of the food energy gained by the birds.

Community composition analysis

Bird species richness for each site was estimated and community similarity was compared between sites using summed species incidence frequencies in the program EstimateS (Version 8.0.0 Copyright R. K. Colwell: EstimateS provides a variety of species richness estimators based on either count or incidence based data sets. We used the approach described by Chao et al. (2005) in order to account for rare unseen species and because this measure is considered to be the most accurate for species rich communities, such as those of the tropics. The Chao’s Jaccard Index Based similarity estimator (corrected for unseen species) was used to estimate community composition similarity.

Multi-season occupancy model

A community of fourteen-bird species was selected to represent the bird community in our multi-season occupancy model. These birds were selected to evaluate a priori predictive models of seasonal occupancy, as well as monthly bird community immigration and emigration rates. Birds in the fourteen-bird assemblage were selected because they were year-round residents, were omnivorous and observed to consume fruit, were from a variety of different bird families, and had sufficient detections to calculate species-specific detectability.

The relationship between bird community turnover and FEA for the 14 resident birds was assessed using multi-season occupancy models (Mackenzie et al. 2003) in the program PRESENCE (available at The sampling design for multi-season occupancy models requires repeated surveys over time at each site in order to obtain species detection histories, a series of ones and zeros where one signifies the species was detected, and zero can signify either that the species was not present or that the species was present but not detected. Through this type of repeated sampling, estimates of detection rates for each species can be calculated and then those detection rates can be used to estimate other model parameters. Multi-season occupancy models use a simple application of likelihood theory to estimate the initial probability that a species is present (initial occupancy) at a site, then, in subsequent time periods, the probability that a species leaves a site that was previously occupied (emigration) and the probability that a species enters a previously unoccupied site (immigration). Multi-season occupancy models require repeated sampling to be done within primary sampling periods, or seasons (here we define a season as a month, which is a time-scale relative to a fruiting season) within the longer time-scale of the study to estimate local or site occupancy dynamics, and therefore estimated emigration and immigration rates are local or site related. A fundamental advantage of using multi-season occupancy models to analyse community turnover is that, with repeated surveys, these models can directly estimate species-specific detection probability to account for species present but undetected. In order to account for different rates of detection due to environmental factors, a Weather Kestrel 2500 was used to record temperature and maximum and average wind speeds prior to the start of each transect. We used our biological knowledge of the study system to select a set of five candidate models using the explanatory variables of Monthly FEA, wind speed, and species identity. Survey temperature was not used as an explanatory variable in the model because recorded daily morning temperature in the study area varied little, remaining steady between 18 °C and 21 °C (min: 17 °C; max: 24 °C). We examined support for the candidate models (see Table 1) using Akaike’s Information Criterion (AIC: Anderson, Burnham & Thompson 2000).

Table 1.   Multiseason occupancy model selection results using program PRESENCE. Response variables were psi (probability of initial occupancy), gamma (probability of immigration), eps (probability of emigration), and p (probability of detection). Explanatory variables were FEA (monthly fruit energy availability), wind (wind speed) and sp (species-specific estimates). When no explanatory variables were used for a particular response variable in a candidate model, a (.) follows the response variable. ΔAIC is the difference in Akaike’s Information Criterion betweeen the candidate model and the best model, and wi is the Akaike weight (support) for a particular model
psi(FEA), gamma(FEA), eps(FEA), p(FEA, wind, sp)0·00·91935860
psi(FEA), gamma(.), eps(.), p(FEA, wind, sp)5·00·07705869
psi(.), gamma(.), eps(.), p(FEA, wind, sp)11·50·00295877
psi(.), gamma(.), eps(.), p(FEA, sp)14·00·00085882
psi(.), gamma(.), eps(FEA), p(wind, sp)25·70·00005893


Site structure

The total number of tree species was similar across study sites (mean: 20·5 species per ha, range: 18–24 species), and the total number of plant species producing fleshy fruits consumed by birds varied from seven to 14 (Appendix S1 in Supporting information). Fruiting trees that were planted for human consumption were common on all farms with the most common species being Psidium guajava, Musa spp. and Citrus spp. For the purposes of this study, Musa spp. was considered as a ‘tree’ because of its large structure and the shade it provides in shade coffee farms. Although these fruit tree species are cultivated for human use, many bird species take advantage of extra fruits not harvested from these trees. Percent canopy cover was similar for all study sites, with an overall mean of 67·8%, range: 62–84% (F5, 95 = 1·68, = 0·15). Site mean tree height ranged from 8 to 10 m and no site had any tree taller than c. 20 m.

Fruit energy availability

Fruit caloric values were determined for 27 plant species bearing fruits consumed by birds within the six coffee plantations (Appendix S1 in Supporting information). Ficus spp. was the most common fruit-bearing tree species, occurring in all but one coffee plantation, although the number of individuals of Ficus spp. was different among farms. Other common fruit-bearing tree species included Inga punctata Willd., Cecropia obtusifolia Bertol., Acnistus arborensens L. and Sapium glandulosum L. Of the 27 plant species tested for caloric values, Ocotea monteverdenses W.C. Burger had the highest caloric value per gram (6·1 kcal g−1) and Musa spp./Citrus spp. had the lowest caloric value per gram (3·5 kcal g−1).

The abundance of mature fruits varied temporally both within and among farms. In the seasonal forests of the tropics plant phenology and the onset of reproduction has been linked to rainfall (Reich & Borchert 1984; Wright & Vanschaik 1994), however, we found no community wide pattern in the temporal variation of fruit abundance in the coffee farms (Fig. 2). Some plantations, therefore, had several monthly FEA values of 0 kcal while the single largest monthly FEA value on any given farm was obtained during July 2008 for a farm with over 25 individuals of fruiting Acnistus arborensens; 38 524·33 kcal. Mean monthly FEA for all months at all sites was 2692·53 kcal (SE 665·87).

Figure 2.

 Monthly Kcal and bird species richness per site surveyed in the San Luis Valley of Costa Rica from February 2008 to December 2008; gray bars represent monthly site FEA values, triangles represent monthly site bird species richness. Codes represent the first three letters of the last name of the farmer who owned the site.


The total number of transects conducted on each farm from February 2008 to December 2008 was 116. Species incidence frequencies were obtained by summing all transects out of 116 for which an individual was detected. For all transects conducted in all sites (n = 696), a total of 113 bird species was detected. Eighty species, or c. 71% of all species, were observed to consume fruit during this study (Appendix S2 in Supporting information). For all 113 species, the total number of detections per farm was as follows: ALV = 761, GIL = 809, ODI = 684, OLD = 1401, OLI = 795, RAF =1021 (n = 116). In order to assess bird community composition similarity across sites we used the yearly summed incidence frequencies (n = 116) for all 113 detected bird species in EstimateS. The Chao-Jaccard estimator predicted that bird community composition similarity between our six sites was high and ranged from 0·86 to 1·00.

FEA and bird species richness

Average monthly FEA per site varied significantly (Fig. 3; F5, 59 = 3·47, = 0·0086), ranging from 824 kcal to 3743 kcal. Additionally, average monthly bird species richness varied significantly among sites, both for the actual number observed (F5, 59 = 11·59, < 0·0001) and for the number estimated (F5, 59 = 6·18, < 0·0001), (Fig. 3). The average monthly FEA for each site was strongly and positively correlated with the average bird species richness per site for both observed and estimated values, Pearson’s product-moment correlation = 0·90 and 0·91, respectively (Fig. 3). Monthly totals of observed bird species richness were graphed against monthly FEA values for each farm (Fig. 2). Monthly totals of observed bird species richness were correlated with FEA for all months and all sites (Fig. 4a. Pearson’s product-moment correlation = 0·72, CI: 0·57, 0·82; t58 = 7·93, < 0·001). For comparison purposes and because the typical explanatory variable used in fruit-frugivore response studies has traditionally been the actual number of ripe fruits, or fruit counts, and not FEA, we also present the results of a correlation test between our monthly totals of observed bird species richness and monthly FAI, or the actual number of ripe fruits available per site per month (Fig. 4b. Pearson’s product-moment correlation = 0·55, CI: 0·35, 0·71; t58 = 5·05, < 0·001). The Pearson’s product-moment correlation between FEA and FAI was 0·86 (CI: 0·78, 0·92; t58 = 12·93, < 0·001). This strong correlation was expected, because FEA is calculated using FAI.

Figure 3.

 Correlation between average monthly FEA and average monthly bird species richness for farms surveyed in the San Luis Valley of Costa Rica from February 2008 to December 2008. (Estimated Species Richness; Pearson’s product-moment correlation = 0·91, t4 = 4·50, = 0·01/Observed Species Richness; Pearson’s product-moment correlation = 0·90, t4 = 4·20, = 0·01). Y-axis represents the mean monthly FEA ± SE per farm (F5, 59 = 3·47, = 0·008). X-axis represents mean monthly estimated bird species richness ± SE per farm (F5, 59 = 6·18, < 0·001).

Figure 4.

 (a) Monthly FEA and monthly bird species richness for all months surveyed in the San Luis Valley of Costa Rica from February 2008 to December 2008. Pearson’s product-moment correlation = 0·72, CI: 0·57–0·82; (t58 = 7·93, < 0·00001) (b) Monthly FAI and monthly bird species richness for all months from February 2008 to December 2008. Pearson’s product-moment correlation = 0·55, CI: 0·35–0·71; (t58 = 5·05, < 0·001).

FEA and bird assemblage stability

Ninety-two percent of the weight of evidence supported a single model that included effects of FEA on initial occupancy, immigration, emigration, and detection (Table 1). The probability of site immigration had a positive threshold response to FEA. When FEA reached 12 000 kcal, if a species was not present at the site during the previous month, there was almost a 100% chance that it would use the site during the current month (Fig. 5). The probability of site emigration steadily declined in response to FEA until by 12 000 kcal there was almost a 0% chance that a species present in the previous month would not also be present in the current month (Fig. 5). The initial occupancy probability had a positive threshold response to FEA and by 5000 kcal sites had almost a 100% chance of species occupancy.

Figure 5.

 Response of initial occupancy, immigration, and emigration to monthly available fruit kcal (FEA) up to 15 000 for birds surveyed in the San Luis Valley, Costa Rica during February 2008 through December 2008. Maximum observed value of FEA was 38 500. Dotted lines indicate 95% confidence intervals for predictions.

Detectability varied greatly by species (Table 2). Ortalis cinereiceps G. R. Gray Grey Headed Chachalacas had the lowest probability of detection (0·10; 95% CI: 0·07, 0·14) while Dives dives Deppe Melodious Blackbirds had the highest (0·90; 95% CI: 0·86, 0·93). There was also a positive effect of FEA (2·3e−5; 95% CI: 1·1e−5, 3·5e−5) and a negative effect of wind (−0·03; 95% CI: −0·06, −0·002) on detectability on the logit scale.

Table 2.   Species-specific detection probabilities from top community turnover model for 14 selected bird species from coffee farms in the San Luis Valley, Costa Rica. LCI and UCI indicate lower and upper 95% confidence intervals, respectively
Common nameScientific nameMeanLCIUCI
Grey headed chachalacaOrtalis cinereiceps0·100·070·14
Red billed pigeonPatagioenas flavirostris Wagler0·580·440·71
Orange-chinned parakeetBrotogeris jugularis Statius Muller0·680·560·77
Emerald toucanetAulacorhynchus prasinus Gould0·680·540·80
Keel billed toucanRamphastos sulfuratus Lesson0·800·710·86
Long tailed manakinChiroxiphia linearis Bonaparte0·360·200·55
Social flycatcherMyiozetetes similis Spix0·680·570·77
Great kiskadeePitangus sulphuratus Linnaeus0·650·530·75
Clay-coloured robinTurdus grayi Bonaparte0·600·450·73
Melodious blackbirdDives dives0·900·860·93
Golden-browed ChlorophoniaChlorophonia callophrys Cabanis0·620·500·72
Yellow throated EuphoniaEuphonia hirundinacea Bonaparte0·720·560·84
Blue-gray tanagerThraupis episcopus Linnaeus0·880·830·92
Buff-throated SaltatorSaltator maximus Statius Muller0·820·750·87


Fruiting phenology and bird response

Peak periods in fruit abundance typically occur in the neotropics and have been documented for various forest types in Costa Rica (Frankie, Baker & Opler 1974; – wet and dry forest, Wheelwright 1985; – cloud forest, Levey 1988- wet forest). We found no consistent months of peak fruit abundance in the agroforestry plots used in this study. However, our goal in this study was not to determine the ecological drivers, i.e. precipitation, soil quality, migrant arrival, disperser abundance, etc. of fruit phenology at our sites (Levey 1988); instead, our goal was to determine if birds responded to fruit availability. It is likely that the low tree species richness of our study sites (average = 20 spp. ha−1), when compared with tree species richness within natural forest in the same region (40–100 spp. ha−1; Kricher 1997) is responsible for the lack of a distinct pattern in fruit seasonality across all sites. Each site had its own unique pattern of fruit resource availability, with marked variation in the availability of fruit resources both within sites and among sites. Rather than the smooth seasonal peak or peaks reported from natural forest (Frankie et al. 1974, Wheelwright 1985; Levey 1988; Zimmerman et al. 2007), fruiting patterns in the coffee plantations were variable and often comprised of abrupt peaks and valleys. At one site, for example, five out of 10 (50%) of the months had an FEA between 0 kcal and only 250 kcal, while 2 months had >5000 kcal and fluctuations as great as 4500 kcal were measured for adjacent months. Bird species richness responded to the dramatic variation, with an average of 33 birds species detected in the farm during the 2 months of >5000 kcal, and an average of only 18 during the months of <250 kcal. Farms that showed less variation supported a more constant species richness of fruit-eating birds. For example, one farm maintained 8 months of the ten with >2000 kcal and 5 months of the ten with >4000 kcal. Species richness at this site never fell below 26 bird species, and maintained 32 or higher bird species for seven out of 10 months.

Community response

The bird community in our study responded to the temporal variation in FEA. Sites with a higher average of monthly FEA values had a higher average monthly overall bird species richness, demonstrating the importance of fruit resources for the largely omnivorous bird community of neotropical managed landscapes. In our occupancy model, FEA had a positive effect on the probability of immigration for our bird assemblage, which included bird species from many different bird families, including cracids, pigeons, psittacids, ramphastids, new world flycatchers, manikins, thrushes, thraupids, emberizids and icterids. A community wide response to fruit resources may be surprising given that no previous studies have found a response of species richness, even within the specialized frugivorous bird guild only (Herrera 1998; Moegenburg & Levey 2003).

We found that FEA explained much of the variance in overall bird species richness, and in fact, much more so (22%) than FAI, or the number of fruits alone. In addition to measures of fruit abundance, several fruit diversity predictor variables (i.e. the number of tree species with fruit, fruit colour variability, vertical stratification diversity of fruits and fruit size variability) have been used to explain frugivorous bird species richness (Kissling et al.2007). However, it was not our goal to evaluate all the potential hypotheses for bird species richness patterns in this study. Furthermore, we studied a non-specialized fruit-frugivore assemblage and therefore we expect that the subtleties of fruit specialization would not have as big an impact as simple availability. Additionally, although fruit diversity variables may explain frugivore richness on a broad spatial scale, this type of response would be difficult to investigate on a local scale in managed lands where low tree species diversity (<20 spp. ha−1) leads to a low monthly number of tree species producing fruits (<3 spp. ha−1), resulting in relatively uniform values of monthly fruit resource diversity. In our study, for example, over 65% of months sampled (n = 60) had three or fewer tree species with fruits. FEA alone explained 52% of the variation in overall bird species richness and therefore can be used as an excellent predictor variable, especially for local management decisions, while fruit diversity patterns may be more detectable over a continental scale.

In part, the effect of FEA on overall species richness was a result of a high proportion of omnivorous species in coffee plantations (Greenberg, Bichier & Sterling 1997). In our study 80 of 113 bird species were observed taking fruit, and this included bird species from virtually all bird families sighted during the study (See Appendix S2 in Supporting information). The bird family with the greatest number of species not observed to take fruit was the hummingbirds, which was comprised of 13 species. However, the two most common hummingbird species in our study sites were observed foraging on the fruits of Cecropia obtusifolia and Acnistus arborensens. The remaining 20 bird species were not observed to take fruit; however, at least ten of these species were observed less than five times in all sites combined, and therefore may not have been observed to forage on fruit due to the low number of observations and not because these species do not consume fruit. The classification of bird species into dietary categories or guilds is often not based on stomach content samples (Remsen, Hyde & Chapman 1993), and foraging observations may be biassed to more conspicuous manoeuvres such as catching insects. Attempts at classifying omnivorous bird species into guilds, such as obligate frugivores, partial frugivores, and opportunistic frugivores, have been made in the past based on the natural history literature, but currently quantitative estimates based on field studies are lacking for most omnivorous birds species and therefore most classifications are based on assumptions made from scant data. These assumptions may obscure the importance that fruit resources have in the diet of birds, such as woodpeckers or flycatchers, which are often classified into the omnivorous or insectivorous guilds. The results of this study underscore the lack of available information required to assign most neotropical bird species into precise dietary guilds, especially regarding guild classifications involving frugivory (Gomes et al. 2008).

Bird assemblage stability

Stable or persistent animal populations are often supported by more consistent food resources, regardless of what level or quality of resources are available (Karr 1971; Illius & O’Connor 2000, Tremblay et al. 2005; Robb et al. 2008). The bird community across all sites was highly similar. It is important to note here that in our study our response variable was monthly (=seasonal) values of bird species richness and so we were not concerned with evaluating the mechanisms responsible for an overall total or static condition of bird species richness at our sites, such as distance to forest. Furthermore, the high index of similarity assigned to the six sites demonstrates that the bird species pool using the sites is largely the same. Despite this similarity, there was a high degree of variation in bird species richness not only among the sites for different months, but more importantly within the sites when comparing different months (min: 15–25; max: 32–55).

Our best multi-season occupancy model predicted that in sites with a monthly FEA value below 500 kcal, bird species would be more likely to emigrate than immigrate. Although the coffee farms in this study region had much higher shade-tree diversity than the typical, more intensively managed, coffee farms, all but one farm had at least 2 out of 10 months with FEA values below 500 kcal, and two farms had more than 50% of monthly FEA values below 500 kcal. Our best model also predicted that above 12 000 kcal, the probability of immigration was at 100% and the probability of emigration at 0%; however, only four farms each had only one monthly FEA value exceeding 12 000 kcal. This study demonstrates that resource constancy allows for higher levels of avian persistence, an important practical concept for management decisions. Management guidelines for agro-forestry systems could include a minimum monthly FEA value which takes into account species richness as well as species persistence.

Conservation implications

Little is known about the home range size of many bird species in this region. However, of those species for which home ranges have been measured, several were only 2–4 ha (Sekercioglu et al. 2007). From a bird’s perspective, a 1-ha coffee plantation in the middle of the 2-ha home range of a fruit-consuming bird could be especially detrimental if the necessary food resources have been removed and only non-fruit bearing tree species remain. This effect would be especially negative during periods of low fruit abundance in surrounding forest patches. The bird species may perceive this habitat as high-quality because of the structural diversity; however, bird body condition and health could be affected through a lack of fruit resources providing essential macro or micronutrients (Rey & Valera 1999).

One third of resident avifauna in many neotropical forests are either frugivorous or granivorous and the number that occasionally consume fruits is even larger (Haugaasen & Peres 2007). Birds that consume fruit often provide the important ecosystem service of seed dispersal with c. 70% of woody tree species dependent upon the removal of their seeds by mobile organisms (Willson, Irvine & Walsh 1989; Ortiz-Pulido, Laborde & Guevara 2000; Haugaasen & Peres 2007). Dispersal of seeds away from the parent plant may result in greater seedling survivability if escape from seed predators is important (Sekercioglu et al. 2004). Unfortunately frugivorous birds, the major long-distance seed dispersers, are predicted to have a higher than average extinction rate over the next 100 years when compared to other bird guilds (Sekercioglu et al. 2004). However, this study demonstrates that bird community use of a particular component of the agricultural matrix can be increased with increasing levels of available fruit energy, as well as more constant with more constant levels of available fruit energy. Therefore, a central focus of agricultural management practices and restoration efforts tailored toward bird community conservation should include evaluating a plant species contribution to yearly fruit energy availability vs. contemporary approaches tending to focus more on preferred plant species or minimum levels of plant species diversity.


We thank the University of Georgia Research Station in San Luis for logistical support. We especially thank the farmers of the San Luis community Finca la Bella for allowing us to use their land to collect data. This work was supported by funding to V.E.P. from the EarthWatch Institute and Birders Exchange.