Norberto Asensio, Faculty of Environment and Resource Studies, Mahidol University, 999 Salaya, Puthamonton, Nakorn Pathon 73170, Thailand. Email: email@example.com
Core areas are thought to be critical parts of animal home ranges for sustaining the population, but few studies have tested this important assumption. We examined whether core areas of spider monkeys Ateles geoffroyi had better habitat quality than the rest of their home range (non-core areas). Habitat quality parameters, including density and diversity of food trees, degree of forest maturity and density of sleeping trees in core and non-core areas were analyzed using Moran eigenvector generalized linear model (GLM) filtering using spatial eigenvector mapping to control for spatial autocorrelation. The best fitting GLM revealed that spider monkeys' core areas had higher habitat quality than non-core areas. This study provides quantitative evidence supporting the concept of core areas including the most critical resources for an animal population. In this respect, spider monkeys' core areas are a key to understand their movement ecology and habitat preferences.
Core areas are defined as small areas of intense use within the home ranges of animals on which their sustainability may depend (Leuthold, 1977; Samuel, Pierce & Garton, 1985). Core areas are expected to contain critical resources for survival and reproduction, which implies that they are more ecologically relevant than other less-frequently used areas (Powell, 2000; Passinelli, Hegelbach & Reyer, 2001; Plowman et al., 2006). Individuals with better quality core areas may have better fitness as they have easier access to important resources (Emery Thompson et al., 2007). Quantitative analysis on whether core areas contain key biological features can provide a better understanding of habitat selection (Samuel & Green, 1988) and the potential role of core areas in establishing priorities for conservation (Bingham & Noon, 1997). While core areas are frequently reported as an aspect of animal ranging along with home ranges (e.g. Hellickson et al., 2008; Spehar, Link & Di Fiore, 2010), only a few published studies provide quantitative evidence for core areas containing the most critical resources in the home range (da Silva Júnior et al., 2009; Thompson, Chambers & McComb, 2009).
Spider monkeys (Ateles spp.) living in dry tropical forests, which is the most endangered ecosystem of the lowland tropics (Janzen, 1986), are an appropriate species to investigate the use and quality of core areas relative to non-core areas, as the extreme habitat fragmentation present in dry forests means that spider monkeys should use core areas in a more distinct manner than they would in more pristine forests. Spider monkeys tend to disappear from disturbed areas because they are especially susceptible to habitat destruction, fragmentation and hunting (Peres, 2001; Ramos-Fernández & Wallace, 2008). They are among the first species to disappear from over-exploited forests (Bodmer, Eisenberg & Redford, 1997) and are rarely found in small forest fragments (Gilbert, 2003) because they are generally found in low densities (but see Wallace, Painter & Taber, 1998), have large home ranges, reproduce slowly, are highly dependent on a fruit diet and have large body sizes (van Roosmalen & Klein, 1988; Gonzalez-Zamora et al., 2009; Di Fiore, Link & Campbell, 2010). They are also slower than sympatric primate species in returning to regenerating dry forest (Sorensen & Fedigan, 2000). Furthermore, their role as seed dispersers is critical for ecological processes in the Neotropical forests (Link & Di Fiore, 2006; Gutierrez-Granados & Dirzo, 2010; Chaves et al., 2011; Stevenson, 2011). Therefore, studying the characteristics of spider monkeys' core areas in dry forests may help highlight potential areas for conservation of the species in such an ecosystem.
Although several studies described the movement ecology (sensu Nathan, 2008) of spider monkeys and their use of core areas (Chapman, 1988; Symington, 1988; Nunes, 1995; Shimooka, 2005; Wallace, 2006; Spehar et al., 2010), there is no detailed information about the quality of their core areas in comparison with non-core areas. Our study aimed to compare the quality of spider monkeys' core and non-core areas in a tropical dry forest and discuss the results in light of the concept of core areas, animal movement and conservation.
Materials and methods
Study site and subjects
The study was carried out from January 2005 to December 2008 in the Santa Rosa Sector (10800 ha, 300-0 m elevation) of the Guanacaste Conservation Area, situated in north-western Costa Rica (10°50′N latitude, 85°38′W longitude). It is a highly seasonal forest with a severe dry season between December and May and a wet season during the rest of the year (900–2500 mm) (Janzen, 1986). A history of differential disturbance and subsequent restoration has resulted in a mosaic landscape with various stages of forest regeneration surrounding fragments of old evergreen mature and riparian forest (Arroyo-Mora et al., 2005; De Gama-Blanchet & Fedigan, 2006).
We investigated one community of individually recognized and well-habituated spider monkeys Ateles geoffroyi that varied in size (25–34 individuals) during the study period (5–8 adult and sub-adult males, 15–18 adult and sub-adult females, 3–7 juveniles and 2–9 infants) due to birth, immigration, dispersal or disappearance of its members. Spider monkeys at the site have a high degree of fission–fusion dynamics, which means that community members are rarely all together and instead split up and join in subgroups of variable size and composition (Asensio, Korstjens & Aureli, 2009).
An aerial satellite orthophoto of the field site was obtained from Digital Globe (http://www.digitalglobe.com; February 2004). The image had a ground sample distance of 1 m per pixel, with the spectral channels red, green and blue (RGB). We ran a supervised raster classification using the maximum likelihood method by linking 100 known ground sample points of four different vegetation types to the satellite image in ArcGIS 9.2 (ESRI, Redlands, CA, USA). Fifty extra ground sample points for each vegetation type were used for posterior verification of the final image. We identified four habitat types on the ground based on the maturity, average height and successional stage of the forest following Arroyo-Mora et al.'s (2005) classification: late mature forest (i.e. undisturbed old evergreen mature forest, areas of riparian forest or the latest successional stage forest with an average canopy height of 20 m), medium dry secondary forest (i.e. deciduous secondary forest with an average canopy height of 15 m), young dry secondary forest (i.e. earliest successional stage deciduous forest with an average canopy height of 5 m) and no forest (i.e. grasslands and pastures with or without acacia bush layers and highly scattered trees). We obtained a polygon shapefile coverage divided into the four habitat types that was vectorized via the ‘raster to vector conversion’ tool in ArcGIS 9.2. Due to the high resolution of the RGB image, individual pixels were smaller than some tree crowns, which sometimes produced small areas of shadows, gaps and thin edges with an incorrect habitat type. Thus, to improve the vector image ‘dissolve adjacent polygons’ extension for ArcView 3.x (Jenness, 2006) was used, which corrected the small ‘holes’ of less than or equal to 0.05 ha by incorporating them into the larger outer polygon class. The final image had an accuracy of 83.2%, according to the proportion of the number of verification points that correctly laid on the corresponding vegetation type.
Spider monkey habitat use
We collected data during full-day follows or balanced observations between mornings and afternoons when full-day follows were not possible. Spider monkeys' subgroups were followed throughout the 48 study months collecting data 3–5 days a week. Individuals were considered in the same subgroup when they were at a distance of ≤50 m from at least one other subgroup member (Asensio et al., 2009). We randomly selected the subgroup to follow after a fission.
The location of the followed subgroup was automatically recorded every 30 min using the track point setting on a handheld global positioning unit (Garmin GPSMAP 76CSX, Olathe, KS, USA) from roughly the centre of the subgroup. A total of 5381 30-min subgroup location points corresponding to 2691 sampling hours were collected during the study, with a mean of 1344 points per year (median = 1262; range: 1076–1776). Due to fission–fusion dynamics, individuals were not equally present in the followed subgroups. However, most community members contributed substantially to the dataset. For example, the individuals that were in the community for the entire 4-year study were sampled on average in 50.8 ± 4.3% [mean ± standard error (se); median = 50.2%; range: 31.7–69.1%] of the 5381 location points.
All sleeping and food trees were identified at the species level and georeferenced with the Global Positioning System (GPS) unit. Only food trees in which the subgroup fed on for at least 5 min were selected for analysis (cf. Link & Di Fiore, 2006). Given the 4-year span of the study, we considered these food trees as representative of the food sources available to the study community. In addition, the study area covered during subgroup follows did not change over these 4 years and a plateau was reached in the number of food trees used, emphasizing that our sampling effort was sufficient to infer the variability in the quality of the habitat the monkeys typically use. The mean (±se) GPS accuracy was 8.8 ± 0.14 m based on 493 circular error probability readings given by the GPS unit at locations throughout the field site. Geographical coordinates were collected using the coordinate system (datum) WGS84 and projected into Universal Transverse Mercator (Zone 16N) units.
Delineation of home ranges and core areas
We applied fixed-kernel estimators with least-squares cross-validation method to obtain the size of core areas within the 50% isopleths and the home range within the 95% isopleths using ‘Hawth's Tools for ArcGIS’ (Beyer, 2004). We calculated kernel areas based on data on the frequency of location use for the entire 4-year study period. A possible solution to reduce autocorrelation, that is, peudoreplication issues (Swihart & Slade, 1985), while having sufficient biologically realistic data, is to arbitrarily decide a minimum time interval when animals may likely switch locations (Willems & Hill, 2009). As the study monkeys are known to travel great distances rapidly (about 0.5 km h−1: Asensio et al., 2009) by setting the time interval between successive locations at 30 min, we reduced data autocorrelation while still maintaining biological validity. In addition to core areas based on frequency of location use, we calculated core areas based on intensity of location use by weighing location use for subgroup size (cf. Spehar et al., 2010). As core areas based on intensity of location use were similar to those based on frequency, our analyses focused only on core areas based on frequency of location use.
Core and non-core areas were divided into 1-ha hexagon cells using the Patch Analyst Extension for ArcGIS (Rempel & Kaufmann, 2003). Cell size may be less than 1 ha at the boundary of the home range and the boundary between core and non-core areas. We obtained 89 cells for core areas and 368 for non-core areas. For each cell, we calculated the value of the following variables of habitat quality. Sleeping tree density was the number of sleeping trees in a cell divided by cell size. Similarly, food tree density was the number of food trees in a cell divided by cell size. Food tree diversity was estimated with the Shannon index (Krebs, 1989) via PAST program (Hammer, Harper & Ryan, 2001). For each cell, we obtained the proportion of each of four habitat types: late mature forest, medium dry secondary forest, young dry secondary forest and no forest.
We fitted a binomial generalized linear model (GLM) to determine whether the categorization in core and non-core areas was explained by habitat quality variables. Given that habitat quality variables were correlated, we used a principal component analysis (PCA) with varimax rotation to obtain uncorrelated components using spss v. 17 (SPSS Inc., Chicago, IL, USA). A minimum eigenvalue of 1.0 was used to determine the number of components extracted from the PCA (Tabachnick & Fidell, 2007). Coefficients of correlation of each variable on the components greater or less than 0.6 were considered high loadings.
A first estimation of the GLM showed that residuals were highly spatially autocorrelated (Moran's I standard deviate= 16.1, P < 0.001). A variogram (estimated in r version 2.10 using geoR package v. 1.6–27) of the residuals showed a high variance of the residuals' semi-variance at short distance coinciding with a long-distance semi-variance smaller than the short-distance one (decreasing variogram). We interpreted that this resulted from the complex shape of the monkeys' home range (Fig. 1) coupled with the clumping of core areas. Furthermore, directional variograms showed that spatial autocorrelation was directionally dependent. This violated the isotropic assumption needed to incorporate spatial autocorrelation in most linear models (Lichstein et al., 2002).
Moran eigenvector generalized linear model filtering using spatial eigenvector mapping
Instead of incorporating spatial autocorrelation in a model relating our response variable to environmental variables, we decided to remove it using a spatial eigenvector mapping approach (SEVM) (Griffith & Peres-Neto, 2006; Dormann et al., 2007). In essence, SEVM attempts to reduce the number of dimensions needed to explain the observed autocorrelation by decomposing a matrix of relationships between sample points into eigenvectors where spatial relationship variance is ‘front-loaded’ in the first few eigenvectors. This matrix of selected eigenvectors can then be used in a GLM as an independent variable. This does not provide a mechanistic understanding of spatial autocorrelation (as there were directionality issues in the observed spatial autocorrelation), but attempts to remove its effects on the analyses. It is therefore possible for some of the variability that could be attributed to a habitat quality variable to be incorrectly attributed to an eigenvector instead. However, this technique has the advantage that the selected eigenvectors can provide information about the scale of spatial processes not accounted for by other independent variables that influence the response variable (Griffith & Peres-Neto, 2006).
The approach first defines a connectivity matrix W between sample points based on a Euclidean distance matrix d between cells:
wij = 1 − (dij/4t)2 and wij = 0 if dij < t. In our case, we classically defined the threshold value t as the maximum distance necessary between neighbouring points in order to maintain all sample points connected in one cluster (using a minimum spanning tree approach). We then eigen-decomposed a centred similarity matrix resulting from this connectivity matrix. We finally selected a given set of eigenvectors resulting from this decomposition to minimize spatial autocorrelation in the residuals of the original GLM. Starting with the original GLM, we added eigenvectors and recalculated Moran's I after each addition. The algorithm we used (implemented in R version 2.10 using spdep package version 0.5-4) permutes eigenvectors to find the set of eigenvectors that best reduces Moran's I, so that residuals of the resulting Moran eigenvector GLM (ME-GLM) are no longer significantly spatially autocorrelated (Griffith & Peres-Neto, 2006). We used Pearson's residuals. However, when we replicated analyses using deviance residuals, we found concordant results. We then assessed best fitting models using Akaike information criteria (AIC) and analyses of deviance between models.
Considering the 5381 30-min location points, spider monkeys used a 95% kernel home range of 304 ha in which there were five core areas for a total size of 46.1 ha (mean = 9.2 ha, range = 3.4–19.2 ha) accounting for 15% of the home range (Fig. 1). We identified 679 food trees and 41 sleeping trees. Although core areas represented only 13.2% of the home range, they contained 34% of food trees and 61% of sleeping trees.
When the seven habitat quality variables were entered into the PCA, sleeping tree density did not have a high loading on any component. Thus, we reran the PCA with the other six variables. Three components were extracted. Components 1, 2 and 3 explained 31.0%, 29.6% and 21.2% of overall variance, respectively, totalling to 81.7% (Table 1). Component 1 consisted of high positive loadings from per cent of young forest and per cent of no forest and high negative loadings from per cent of mature forest, and was labelled Young Forest and Open Areas. Component 2 showed high positive loadings for food tree diversity and food tree density, and was labelled High Food Quality Forest. Component 3 consisted of high positive loadings from per cent of medium forest and was labelled as Intermediate-aged Forest. The three components and sleeping tree density were used in the GLM.
Table 1. Varimax rotated component matrix from the principal component analysis. Values represent coefficients of correlation between each variable and each component. Values of >0.6 or <−0.6 (marked in bold) were considered high loadings
Food tree diversity
Food tree density
% mature forest
% medium forest
% young forest
% no forest
The best fitting GLM (GLMbest) incorporated PCA components Young Forest and Open Areas, and High Food Quality Forest, and sleeping tree density to explain the variance between core and non-core areas (Fig. 2). While the significance of the contribution of Young Forest and Open Areas was marginal, removing this term led to a significant decrease in variance explained [analysis of deviance between GLMs with and without Young Forest and Open Areas: χ21 = 4.3 P < 0.037; AIC(GLMbest) = 400.3 and AIC(GLMbest-Young Forest and Open Areas) = 402.7; Table 2].
Table 2. Results of the best fitting generalized linear model (GLM) (residual deviance 392.35 on 453 degrees of freedom, variance explained: pseudo-R2 = 12.9%) with estimates from the GLM and from the Moran eigenvector generalized linear model (ME-GLM) for coefficients and standard errors (se)
Young Forest and Open Areas
High Food Quality Forest
Sleeping trees density
The residuals of this model were spatially autocorrelated (Moran's I = 16.1, P < 0.001). We therefore fitted a ME-GLM. Firstly, the SEVM retained eight eigenvectors to remove spatial autocorrelation. Once these eight eigenvectors were added as independent variables, the residuals of the ME-GLM were no longer spatially autocorrelated (Moran's I = 0.84, P = 0.2). An analysis of deviance between the GLM and the ME-GLM showed that adding these spatial eigenvectors did provide significantly more information about the variance between core and non-core areas (χ28 = 214.5, P < 0.0001) (Table 3). Much of the pattern absorbed by the eigenvectors was correlation along a south-north axis (Fig. 3a–d). The last eigenvectors point towards processes taking place at a smaller spatial scale, particularly highlighting areas that were more similar than the rest of the home range (Fig. 3e–h).
Table 3. Analysis of deviance table for the Moran eigenvector generalized linear model (ME-GLM) with terms added sequentially. Variance explained by the ME-GLM: pseudo-R2 = 60.5%
d.f., degrees of freedom; ME, Moran eigenvector.
Young Forest and Open Areas
High Food Quality Forest
Sleeping trees density
Spider monkeys in the Santa Rosa sector used core areas containing higher habitat quality than the rest of their home range. Thus, our study provides quantitative evidence supporting the view that core areas contain critical resources for an animal population (Leuthold, 1977; Samuel et al., 1985). This study also corroborates findings in other species in which core areas have more biologically relevant features than non-core areas, such as decayed logs for voles (Thompson et al., 2009) and large trees for woolly spider monkeys (da Silva Júnior et al., 2009). Our results are in agreement with previous findings that spider monkeys prefer mature forest or forest with the latest successional stage of regeneration (Chapman, 1988; De Gama-Blanchet & Fedigan, 2006; Chaves et al., 2011). Indeed, we demonstrated that spider monkeys have preferences for areas including even more profitable habitat than the rest of their home range: spider monkeys' core areas contained a higher density and diversity of food trees, more mature forest and a higher density of sleeping trees.
Preference for higher quality areas within a matrix of high-quality habitat may explain why spider monkeys are especially vulnerable species when facing habitat fragmentation and disturbance (Ramos-Fernández & Wallace, 2008; Di Fiore et al., 2010). Habitat fragmentation forces spider monkeys to travel between distant high-quality core areas in order to meet their dietary requirements. In addition, given their highly arboreal lifestyle (van Roosmalen & Klein, 1988; Campbell et al., 2005), fragmentation can also eliminate critical arboreal routes to move between core areas (Laurance, 1994; Lindenmayer, Cunningham & Dunnelly, 1994). In contrast, other more generalist mammal species living in tropical dry forests, such as coatis Nasua narica, white-tailed deer Odocoileus virginianus and white-faced capuchin monkeys Cebus capucinus, may be less dependent on particular core areas with specific food resources as their broader diet allows them to survive despite the substantial seasonal changes in food abundance and fragmentation of this forest biome (Stoner & Timm, 2004).
Our findings indicate that high-quality habitat that can act as core areas is crucial for spider monkeys. However, just protecting the core areas is not sufficient when planning for spider monkey conservation, especially when their core areas consist of spatially separate nuclei within the home range (Fig. 1). At least in tropical dry forest undergoing regeneration, the matrix between core areas needs to be protected because it contains arboreal routes between critical resources and because barriers to dispersal would likely reduce population viability in the long term (Laurance, 2004). In addition, non-core areas included a large proportion of mature and last regeneration-stage forest and contained 66% of the food trees (Fig. 1). This means that core areas by themselves were insufficient in providing the minimum nutritional requirement for the study community. Furthermore, the level of use an area receives is not necessarily related to its importance during critical periods (Buchanan, Fredrikson & Seaman, 1998). For example, during this study, spider monkeys were observed to drink from two creeks just twice in the driest days of the year (pers. obs.). These water locations were outside the identified core areas, but they were likely crucial for the monkeys' survival. Thus, although we demonstrated that spider monkeys' core areas contain critical features and are a key to understand their movement ecology and habitat preferences, conservation initiatives in tropical dry forests need to focus on larger areas than spatially separate core areas.
We thank E. Murillo-Chacon and the staff from Santa Rosa sector, especially R. Blanco and M. M. Chavarria, for their support during field work; M. Luinstra, S. Wilson, A.M. Nuttall, E. Willems, F. Eigenbrod, C. Garcia, L. Maher, M.A. Veganzones and W.Y. Brockelman for valuable input on GIS; A. Douglas and T. Cornulier for insightful discussion; R. Espinoza and A. Guadamuz for botanical assistance; and A.C. Palma and an anonymous reviewer for helpful comments. This study was financially supported by the Leakey Foundation, the North of England Zoological Society and The British Academy. N.A. was supported by the Department of Political Science of the Basque Government (Zientzia Politikarako Zuzendaritza) and the Postdoctoral Fellowship program of Mahidol University, Thailand. Observations complied with current laws in Costa Rica.