Vegetative predictors of primate abundance - Wiley Online Library

American Journal of Primatology 69:1242–1256 (2007)

RESEARCH ARTICLE Vegetative Predictors of Primate Abundance: Utility and Limitations of a Fine-scale Analysis FRANCESCO ROVERO1 AND THOMAS T. STRUHSAKER2 1 Sezione di Zoologia dei Vertebrati, Museo Tridentino di Scienze Naturali, Trento, Italy 2 Department of Biological Anthropology, Duke University, Durham, North Carolina

Determining ecological predictors of primate abundance is important for both theoretical and applied conservation management. For forest primates, research has focused on comparisons of primate abundance and vegetation in different forest blocks or forest compartments with different management histories. However, great variation in primate abundance often occurs within single forests, especially in mountainous areas or in areas with habitat mosaics due to past disturbance. Here we assess, for the first time, the usefulness and limitations of small-scale, within-transect analysis of vegetative parameters as predictors of primate abundance in a very heterogeneous forest habitat in the Udzungwa Mountains of Tanzania. Relative abundance of four species of diurnal primates was recorded over a period of 2.5 years by walking three census transects 48 times each. Tree size, density, species composition, and food plants were measured along the same census lines. The fine-scale relationship between primate abundance and vegetative variables was analyzed through generalized linear modeling applied to 58 segments of these three census lines. Each segment was 200 m in length. For all four primate species, we found significant associations between their abundance and selected vegetative variables. The abundance of the endemic and endangered Udzungwa red colobus Procolobus gordonorum was positively related to mean basal area of large trees (diameter at breast height greater than 20 cm) and to the species richness of their food plants. Considering the very great variation in primate abundance that was recorded among segments of the census lines, our approach proved useful in predicting the relationship between primate abundance and small-scale habitat differences. The main limitation of this study, however, was the relatively low-predictive power of the models for some species, especially the Angolan colobus Colobus angolensis. We discuss

Correspondence to: Francesco Rovero, Sezione di Zoologia dei Vertebrati, Museo Tridentino di Scienze Naturali, Via Calepina 14, I-38100, Trento, Italy. E-mail: [email protected]

Received 16 October 2006; revised 16 January 2007; revision accepted 17 January 2007 DOI 10.1002/ajp.20431 Published online 14 March 2007 in Wiley InterScience (www.interscience.wiley.com).

r 2007 Wiley-Liss, Inc.

Primate Abundance and Vegetation / 1243

the potential reasons for this problem and suggest possible improver 2007 ments for future studies. Am. J. Primatol. 69:1242–1256, 2007. Wiley-Liss, Inc.

Key words: determinants of abundance; habitat quality; monkeys; Procolobus gordonorum; Udzungwa INTRODUCTION Understanding ecological parameters that determine animal abundance is increasingly important in animal ecology because of the implications for conservation [Morrison et al., 1998; Wasserman & Chapman, 2003]. For nonhuman primates, this research has focused primarily on vegetation [Cowlishaw & Dunbar, 2000; Mbora & Meikle, 2004; Medley, 1993; Skorupa, 1986]. Most diurnal primates are excellent subjects for the study of ecological correlates of abundance because they can be readily counted and the relevant vegetation, usually approximated by the community of large trees (i.e. trunk diameter Z10 cm at breast height), can be easily sampled using well-developed protocols [Struhsaker, 1997]. A common approach has been to correlate estimates of primate abundance with various vegetative parameters, by contrasting sites that are widely separated [Davies, 1994; Ganzhorn, 1992; Oates et al., 1990]. However, great variation in primate abundance can occur within a single forest [Chapman & Chapman, 1999; Chapman et al., 2002; Marshall et al., 2005; Onderdonk & Chapman, 2000; Struhsaker, 1997; Struhsaker et al., 2004]. The advantage of comparing subpopulations within the same or neighboring forests is that it controls for intrinsic, interspecific differences and reduces the likelihood of introducing other, unknown variables, such as predation pressure [Chapman & Chapman, 1999]. Examples of this approach include the studies of Skorupa [1988], who compared primate abundance with vegetation in logged and unlogged areas of the same forest in Kibale, Uganda and Medley [1993] and Wieczkowski [2004] who used the same method within several forest patches along the Tana river, Kenya. These studies found a number of significant correlations between primate abundance and vegetative parameters, such as tree basal area, canopy cover and amount of forest gaps. An important limitation to this kind of study design is that it only compares results between transects rather than within them. Accordingly, this assumes a relative degree of uniformity in habitat and primate density within the different study sites (i.e. forest patches or line transects) being contrasted. However, in cases where habitat heterogeneity is very pronounced, such as mountainous or heavily disturbed areas, it is often not possible to establish study transects of 3–4 km length in relatively uniform habitat [Warner, 2002; Rovero et al., 2006]. Variations in vegetation patterns typically occur both along the altitudinal gradient and between valleys and ridges. Moreover, the effect of past human impact often occurs in patches that are interspersed with more mature forest. In such cases, habitat variation within a study unit (transect or plot) might be very pronounced. Consequently, comparisons of transects, even when near one another in the same forest may be confounded by important differences in the nature and extent of habitat heterogeneity. In this study, we assess the usefulness of a finer scale analysis of vegetative correlates of primate abundance within a very heterogeneous forest habitat in the Udzungwa Mountains of south-central Tanzania. The Udzungwa Mountains are

Am. J. Primatol. DOI 10.1002/ajp

1244 / Rovero and Struhsaker

one of the most important sites in Africa for primate conservation, primarily because of the presence of the endemic Udzungwa red colobus Procolobus gordonorum, the endemic Sanje mangabey Cercocebus galeritus sanjei and the recently discovered Rungwecebus kipunji [Davenport et al., 2006; Dinesen et al., 2001; Jones et al., 2005]. METHODS Study Area and Subjects We conducted the study in the Mwanihana Forest (179 km2) of the Udzungwa Mountains National Park (UMNP, 1,990 km2, centered on 71460 S, 361510 E). UMNP covers the northeastern portion of the Udzungwa Mountains and Mwanihana Forest is situated on the slope of the east-facing escarpment. It has forest cover from 300 to 2,100 m a.s.l. Mean annual rainfall is about 2,000–2,500 mm. Details of vegetation zones and plant species composition of Mwanihana Forest are reported in Rovero et al. [2006] and Lovett et al. [1988, 2006]. The diurnal primates found in Mwanihana Forest include Udzungwa red colobus, Sanje mangabey, Angolan black and white colobus Colobus angolensis, Sykes’s monkey Cercopithecus mitis and yellow baboon Papio cynocephalus. Red colobus generally have much larger groups (mean of 41 individuals in the Mwanihana study area) than Angolan colobus (range 2–12) and Sykes’ monkeys (2–14), whereas group size for baboons and Sanje mangabeys can be greater than 40–50 [Ehardt et al., 2005; Rovero et al., 2006; Struhsaker et al., 2004]. Data Collection Primate census data were obtained through line transect methods [Struhsaker, 1981; Whitesides et al., 1988]. Three line transects (T1–T3), 4, 4 and 3.6 km long, respectively, were placed along existing pathways that run mainly from East to West and are positioned about 6 km apart. Altitude increases from 300 to 600–1,000 m a.s.l. depending on the transect, and habitat progressively changes from deciduous and semi-deciduous (i.e. predominantly deciduous), to semievergreen (i.e. predominantly evergreen) and evergreen forest (see Rovero et al. [2006] for more details). We used both a subset of data presented in Rovero et al. [2006], collected by F. Rovero and A. Mtui between July 2002 and August 2003 and new data collected by A. Mtui between February and December 2004, for a total of 48 census repetitions of each line. The number of censuses conducted per month varied from 1 to 4. Census lines were walked beginning at 07:00–07:30 h at an average speed of about 1 km/h recording all sightings of primates. For each primate sighting we recorded time, species, number of individuals, horizontal distance from the observer to the first animal seen, and position of the observer along the transect. Position on the transect was estimated by referring to numbered tags placed every 50 m along the trails. See Rovero et al. [2006] for further details on primate censuses. All trees with trunks at least 10 cm diameter at breast height (DBH) located within a strip width of 2.5 m each side of the trail edge along the transects were enumerated. This sampling protocol followed earlier studies [Butynski, 1990; Skorupa, 1988; Struhsaker, 1975]. The following measures were taken: DBH to the nearest 10th of a centimeter, tree height (visually estimated after calibration of estimated distances with actual values measured by using a range finder), tree species, proportion of tree stems Z10 cm DBH with presence of lianas. Samples of tree species that could not be identified in the field were identified later with help



from C. Ruffo (National Tree Seed Agency, Tanzania) and Q. Luke (National Museums of Kenya). Data Analysis We used the primate social group as the basic unit for data analysis [Whitesides et al., 1988]. We used only groups seen within an estimated distance of 40 m of the observer, to ensure that group positions were within or near the area sampled for vegetation. Frequency distribution of estimated distances showed that the majority (80%) of sightings are within this distance, with most values falling in the interval of 20–40 m [see also Rovero et al., 2006]. Using a smaller cut-off value would reduce the sample size considerably and therefore the power of the analysis. The relative abundance of each species was first approximated by the mean number of primate groups seen per km walked [Seber, 1982]. Secondly, for the purpose of correlation analysis with vegetation parameters, relative abundance was computed as the cumulative number of sightings of primate groups seen in each 200 m section of the transects. This spatial unit was chosen as the basis for correlation analysis because it was considered to be appropriate for computing vegetation variables to have a large enough sample of tree stems for analysis while maintaining an acceptable level of uniformity of habitat types within each portion. We did not use absolute density of primates for the analysis because the relatively small number of sightings recorded in 200 m sections did not represent an adequate sample size for computing density. To verify that the number of census replicates produced adequate, asymptotic precision of sampling for the 200 m segments, we analyzed sampling precision profiles for five randomly selected segments where the total number of red colobus sightings varied from 2 to 10. Precision was computed as the 95% confidence limits of the mean number of groups seen per portion of the transect walked and expressed as the percentage of the mean [Struhsaker, 1981]. Precision was judged adequate for our analysis as it reached nearly asymptotic levels after a variable number of census replicates (30–40) depending on the number of sightings (Fig. 1). The 20 variables obtained from tree measurements are listed in Table II, where they are divided into seven variables used for multiple regression models and 13 redundant variables that were not used in the analysis (see below for further details). We also used elevation (the altitude of transect portions) as a non-vegetative variable that may affect or predict primate abundance. Basal area was computed as both mean value (m2) and total value (m2/ha) per segment of transect. Although total basal area gives a proxy of potential food available from the trees sampled, mean basal area should reflect the size of trees that bear food. Diversity was computed using the Shannon–Wiener Index [Magurran, 1988]. The amount of gap in the forest cover was estimated as the percentage of the transect that crosses areas with no trees for at least 25 m on each side of the transect. The index of liana coverage was computed as the proportion of stems covered by lianas on the total number of stems. The four red colobus food plant variables were computed using a list of food plants obtained during a preliminary study of red colobus diet that involved follows of four focal groups. These groups lived within the transect areas in Mwanihana Forest and ranged in a variety of habitat types representative of the transect habitats. These groups were followed over a period of 3–5 months for an effective observation effort of 10–20 days for each group, yielding feeding records ranging in number from 120 to 700 per group (A. Pucci & F. Rovero, unpublished data). Feeding records were defined as food items the red


1246 / Rovero and Struhsaker 140 120

Precision (%)

100 80 60 40 20 0 4

8

12

16

20 24 28 32 Cumulative number of censuses

36

40

44

48

Fig. 1. Examples of sampling precision profiles of the number of red colobus group sightings recorded on five randomly selected segments of 200 m within the line transects censused in Mwanihana Forest, Udzungwa Mountains National Park, Tanzania. The total number of sightings scored on these transect segments was 2 (open circles), 4 (closed circles), 6 (open squares), 8 (closed squares) and 10 (open triangles), respectively. Precision is plotted against the cumulative number of census repetitions, lumped into groups of four census walks. Precision is computed as the 95% confidence limits of the mean number of groups seen per km walked and it is expressed as the percentage of the mean. The lower this value, the higher the precision of the sample. Asymptotes are generally reached after 28–40 repetitions.

colobus fed upon as recorded during scan observations conducted every 30 min. We computed both mean and total basal area for all food plant species in the diet (71 species) and the 15 most commonly eaten species. We also computed food species richness and the proportion of stems that were food species. We did not compute the Shannon Index of diversity for food plant species because the number of food stems was small and this index is not recommended for small samples [Magurran, 1988]. We used the following procedure to analyze the relationship between primate abundance and vegetation. As differences in primates’ relative abundance might be related to transects sampling relatively different areas, we considered transect a potentially confounding co-variable. Thus, we pooled data from the three transects after assessing that there were no significant differences for any primate species abundance among the transects (ANOVA: F(2, 54) 5 0.95, 0.79, 1.28, 1.76 and P 5 0.40, 0.46, 0.29, 0.18 for red colobus, Angolan colobus, Sykes’s monkey, and baboon, respectively). The effect of vegetative variables on primate abundance was assessed by means of generalized linear models (GLM, McCullagh & Nelder, 1989) implemented in the computer package R (version 2.1.0; http:// cran.r-project.org). We used GLM regression with Poisson error distribution that is appropriate when the response variables are counts [Maindonald & Braun, 2003]. To avoid multicollinearity effects due to redundancy between explanatory variables, only a subset of variables was fitted to the model (Table II). Redundancy between variables was investigated by building a correlation matrix between all variables and then carefully scrutinizing each pair of variables that was significantly correlated at Po0.05 [Spearman rank two-tailed correlation tests; Sokal & Rohlf, 1995]. Variables obtained from the subset of large trees (i.e., DBHZ20 cm) were preferred to the variables related to all trees measured because large trees represent canopy trees and food sources for the primates and



because this approach reduces the number of variables to be measured in the field. We acknowledge, however, that most of these primates also feed on smaller, non-canopy trees. Variables from large trees (DBHZ20 cm) were preferred to variables from tall trees (estimated height Z20 m) because height was based on visual estimations while DBH was actually measured. Diversity was preferred to richness because the Shannon–Wiener index considers both richness and evenness [Magurran, 1988]. The amount of gap was not retained, in preference to basal area to which it was negatively and significantly correlated, as gaps were only visually estimated and not measured. Although mean and total basal area were correlated for all trees (r 5 0.67, n 5 58, Po0.001) and for large trees only (r 5 0.52, n 5 58, Po0.001), we retained both variables as we considered it important to investigate the relative effects on primates of both measures because a large total basal area can be obtained from either many small trees or a few large trees. Elevation did not correlate significantly with any of the vegetative variables. We then built GLM models through a backward stepwise procedure [Crawley, 1993] whereby all selected variables are presented to the model, extracted one at a time, and the changes in model deviance assessed by a w2 test. This procedure retains the model with significant variables, if any, and the highest predictive power (least residual deviance). The fitting of logistic model in GLM is done by minimizing deviances. A deviance has a role very similar to sum of squares in regression (in fact, if the data are normally distributed, the two quantities are equivalent [Maindonald & Braun, 2003]). Variables were not transformed because in GLM this transformation is done through the link function [Maindonald & Braun, 2003; J. Birks, personal communication]. We also indicate the significant variables in the full, unreduced model [Marshall, 2007] because of controversy in the use of stepwise modeling [Whittingham et al., 2006]. As transect segments may not be independent sampling units because they are spatially sequential, any potentially confounding effect on primate abundance was assessed by repeating the regressions on every second segment and assessing whether the resultant models differed from the original models in the variables retained and their significance levels. RESULTS The rank order of primate species’ relative abundance was: red colobus (0.57 group sightings/km walked), Angolan colobus (0.42), Sykes’s monkeys (0.29), and baboons (0.09). The total number of primate group sightings was 735, of which 586 (80%) were within 40 m of the observer. These 586 sightings were used in the analysis. Distribution of total group sightings along 200 m portions of the transects was highly variable (mean7SD 5 2.5373.21, range 0–16; Table I). Sanje mangabeys were not seen frequently enough to be considered in the analysis (39 group sightings, 0.07 groups/km walked). Moreover, they spend most of their time foraging and traveling on the ground (T. Jones & F. Rovero, unpublished data) and thus it seems likely that the vegetative variables we measured would not be strong predictors of mangabey abundance. A total of 2,486 tree stems belonging to 158 species were measured along the three transects. Fifty-nine trees, corresponding to 2.4% of the total number sampled could not be identified. Red colobus abundance was positively related to mean basal area of large trees (i.e., DBHZ20 cm; z 5 1.8470.46, Po0.001) and to the richness of their food plant species (z 5 0.1470.02, Po0.001; Table III). Angolan colobus abundance was positively related to the total basal area of large trees (z 5 0.0270, Po0.001)



SD D D D SD SD SD O O EV EV EV EV EV EV EV EV EV EV EV

200 400 600 800 1,000 1,200 1,400 1,600 1,800 2,000 2,200 2,400 2,600 2,800 3,000 3,200 3,400 3,600 3,800 4,000 Total Mean SD

6 0 2 0 6 2 4 0 0 16 8 0 2 2 2 6 10 2 4 2 74 3.7 4.1

P. g.a

2 0 0 0 2 2 0 2 4 4 10 6 2 2 0 6 2 4 2 0 50 2.5 2.6

C. a.b

2 0 2 4 6 6 2 0 4 4 0 2 4 0 4 0 2 0 0 0 42 2.1 2.1

C. m.c 4 4 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 10 0.5 1.3

P. c.d D D D D D D D SD SD EV EV EV EV EV EV EV EV EV EV EV

Hab 0 0 2 6 0 0 0 4 2 12 10 6 4 4 6 0 2 0 6 0 64 3.2 3.6

P. g. 0 0 4 2 2 0 2 2 2 6 10 6 4 6 4 0 2 2 2 2 58 2.9 2.6

C. a. 4 0 6 6 0 0 2 0 6 4 6 4 2 0 0 2 0 0 0 0 42 2.1 2.5

C. m.

Transect 2 (‘‘Mwanihana peak’’)

2 6 4 2 2 2 0 2 0 0 0 0 0 0 0 0 0 0 0 0 20 1.0 1.7

P. c. D/SD D/SD D/SD D/SD EV EV EV EV EV EV EV EV EV EV EV EV EV EV

Hab 0 0 16 2 6 2 6 0 4 4 6 0 2 0 8 10 2 2

70 3.9 4.3

88 4.9 3.7

C. a.

4 10 8 10 2 6 8 2 4 2 8 0 0 0 8 8 8 0

P. g.

64 3.6 4.3

12 10 12 4 8 6 4 4 0 2 0 0 0 0 2 0 0 0

C. m.

Transect 3 (‘‘Sanje waterfalls’’)

4 0.2 0.6

2 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

P. c.

Left column shows the length along the transects, that correspond to increasing elevation. The habitat type (Hab) is indicated as D 5 deciduous, SD 5 semi-deciduous, O 5 open areas, EV 5 semi-evergreen. Primates are indicated by initials of Latin names. a Procolobus gordonoroum. b Colobus angolensis. c Cercopithecus mitis. d Papio cynocephalus.

Hab

m

Transect 1 (‘‘Camp site 3’’)

TABLE I. Total Number of Sightings of Primate Groups Recorded Within 40 m of the Observer in 200 m Sections (Left Column) of Three Transect Lines Censused 48 Times in the Udzungwa Mountains of Tanzania (Transect 1–3 with, in Parenthesis, Local Names of Respective Trails)


Primate Abundance and Vegetation / 1249 TABLE II. Vegetative Variables Obtained from Measurements Made Along Three Transects Used for Primate Censuses in the Udzungwa Mountains of Tanzania Variables presented to regression models with primate abundance 1. Species diversity (Shannon–Wiener index)

2. Mean basal area of trees with DBH Z20 cm

3. Total basal area of trees with DBH Z20 cm

4. Index of liana coverage (frequency of stems with lianas) 5. Mean basal area of all red colobus food plantsa 6. Total basal area of top 15 red colobus food plantsa 7. Species richness of red colobus food plantsa

Correlated variables not presented to the models (correlation coefficient) 8. Species richness (0.85) 9. Species richness of trees with DBH Z20 cm (0.66) 10. Mean basal area (0.76) 11. Mean basal area of trees taller than 20 m (0.52) 12. Stem density (0.30) 13. Total basal area (0.98) 14. Total basal area of trees taller than 20 m (0.94) 15. Stem density of trees with DBH Z20 cm (0.66) 16. Stem density of trees taller than 20 m (0.75) 17. Gap amount (0.50)

18. Total basal area of all food plants (0.65) 19. Mean basal area of top 15 food plants (0.84) 20. Proportion of red colobus food plant stems per total stems (0.44)

Left column lists the sub-set of variables used for multiple regression with primate abundance, the right column lists the redundant variables, not used for modeling, and, in parenthesis, their Spearman rank correlation coefficient (n 5 58) with the variables in the left column. a Variables used only for red colobus model. **r significant at Po0.01. *r significant at Po0.05.

and to the proportion of trees with lianas (z 5 0.9170.27, Po0.001). Sykes’ monkey abundance was positively related to tree species diversity (z 5 0.5170.19, Po0.01) and the proportion of trees with lianas (z 5 0.7770.33, Po0.05). Elevation was also weakly correlated, but negatively, with the abundance of Angolan colobus (z 5 0.00470, Po0.001). Baboon abundance was strongly, negatively related to mean basal area of large trees (z 5 10.6973.85, Po0.01) and weakly, negatively related to the proportion of trees with lianas (z 5 1.4570.77, P 5 0.058). Elevation had also a weak, negative effect on baboon abundance (z 5 0.0270, Po0.001). The deviance explained by the four models was 24.2, 12.3, 38.4 and 79.6% for red colobus, Angolan colobus, Sykes’ monkey and baboon, respectively. Significant variables in the full, unreduced models for the two colobines and Sykes’ monkeys were the same as those retained by the reduced model, whereas for baboons only elevation was significant in the full model. Models did not change in the variables selected or the significance levels when re-built using every second 200 m interval only, with the exception of the model for the red colobus that included two additional variables: the index of liana coverage and the mean basal area of food plants.


1250 / Rovero and Struhsaker TABLE III. Vegetative Predictors of the Relative Abundance of Four Primate Species Censused Along Three Line-Transects in the Mwanihana Forest of the Udzungwa Mountains National Park of Tanzania Variables retained by the modela

z7SE

P(z)

b

Udzungwa red colobus Mean basal area of trees with DBHZ20 cm Richness of red colobus food plant species Angolan colobus Total basal area of trees with DBHZ20 cm Index of liana coverage Sykes’ monkeyc Diversity of trees Index of liana coverage Baboond Mean basal area of trees with DBHZ20 cm Index of liana coverage

1.8470.46 0.1470.02

o0.001 o0.001

0.0270 0.9170.27

o0.001 o0.001

0.5170.19 0.7770.33

o0.01 o0.05

15.0573.75 4.1770.93

o0.001 o0.001

The multiple regression models were built on n 5 58 segments of transects of 200 m in length through a GLM with Poisson error distribution following a backward elimination procedure (see text). a Variables presented to the model: (1) mean basal area of trees with DBHZ20 cm, (2) total basal area of trees with DBHZ20 cm, (3) species diversity of trees, (4) index of liana coverage (frequency of stems with lianas), (5) elevation. b Additional food plant variables presented to the model for red colobus only: (1) total basal area of top 15 red colobus food plants, (2) mean basal area of food plants, (3) richness of red colobus food plant species. c Elevation was also retained by the model (z 5 0.00470, Po001). d Elevation was also retained by the model (z 5 0.0270, Po001).

DISCUSSION Habitat heterogeneity, especially related to vegetation structure, has been stressed as an important factor to consider when attempting to understand variation in primate abundance [Brugiere et al., 2002; Oates et al., 1990]. An idea of the heterogeneity of the Mwanihana transects in our study area is clearly demonstrated by the number of tree stems sampled in the 200 5 m plots along the transects, which ranged from 5 to 72 (coefficient of variation 5 32%). Coincident with this habitat complexity, is the fact that primate sightings were often highly clustered along the transects. For example, sightings of primates in contiguous 200 m sections of the transects varied by as much as 16-fold (Table I). Differences in visibility are unlikely to account for these striking contrasts in primate abundance because we only used sightings of primate groups that were relatively close to the observer. Moreover, Rovero et al. [2006] found that visibility, as indicated by primate sighting distances, did not vary significantly between habitat types. The potential lack of independence between transect segments is not considered to affect the relationship between primate abundance and vegetative variables, as, with the partial exception of the red colobus, the regression models built on alternated transect segments were virtually identical to the original models. The regression models explained about 12–80% of the deviance and is particularly low for the Angolan colobus. That the models did not explain more of the deviance for some species might be explained by the many potentially confounding factors associated with the analysis. First, even if it is a widely used protocol, the vegetation sample was from a relatively narrow strip of 5 m along the transects. This represents only 6.25% of the area where primate sightings were recorded (up to 40 m either side of the transects). Secondly, some of the sightings



may have been of primate groups moving between foraging areas that were well beyond the study transects. Third, there might be pronounced seasonal variations in the distribution of primates along the transects that were not taken into account. For example, red colobus are found in deciduous habitats mainly when new leaves are available during about 2 months every year and use it very little during the rest of the year. Finally, data on food plants were only available for red colobus. With these deficiencies and caveats in mind, our results provide insight into the utility and limitations of the methods we used in attempting to develop predictive models of primate abundance. Vegetative Predictors of Udzungwa Red Colobus Abundance That red colobus were positively associated with mean basal area of large trees indicates selectivity for forest areas with large, mature trees. Mean basal area is significantly larger in semi-evergreen portions of the transects than in both deciduous and semi-deciduous parts (ANOVA testing semi-evergreen versus combined deciduous and semi-deciduous: F(1, 54) 5 5.72, Po0.05). This is consistent with analysis of gross habitat selectivity indicating that the Udzungwa red colobus is more frequently found in mature, evergreen forest [Marshall et al., 2005; Rovero et al., 2006; Struhsaker et al., 2004]. In Kibale National Park, Uganda, red colobus (Procolobus rufomitratus tephrosceles) abundance was associated with basal area of all trees and canopy cover of trees taller than 15 m [Skorupa, 1988]. Furthermore, most of their food comes from tree species that are large, which is also reflected by their feeding selectivity [Struhsaker, 1975]. In the forest patches along the lower Tana River, Kenya, red colobus (Procolobus rufomitratus rufomitratus) abundance was positively associated with canopy height and basal area coverage of canopy trees [Medley, 1993]. Mbora and Meikle [2004] found that most of the variance in the density of P. r. rufomitratus in 20 forest patches along the Tana River was explained by basal area of food trees (per ha), basal area per food tree, density of food trees, and basal area per tree for all trees. The significant correlation between food plant species richness and Udzungwa red colobus abundance may reflect the importance of food variety and food quality in the diet of this species. Red colobus appear also to prefer areas with a large proportion of food stems, as this variable was correlated with food plant richness. The density of P. r. tephrosceles among six sites in Kibale was related to cumulative DBH of important food trees [Chapman & Chapman, 1999; Chapman et al., 2002]. Similarly, Skorupa [1986] found that the abundance of this species across three transect lines was positively correlated with total basal area of trees that contributed to the top 80% of diet. The lack of a similar relationship in our results suggests that either our sample of diet was inadequate or that food abundance per se is not a major constraint to red colobus abundance, whereas food variety, as expressed by food plant species richness, is a more important variable. This is consistent with chemical analyses of red colobus’ diet in Kibale National Park [Chapman et al., 2004; Wasserman & Chapman, 2003]. Several studies of primates suggest that in addition to quantity, quality and seasonal availability of food are the most important proximate factors that limit primate populations [Brugiere et al., 2002; Milton, 1982; Struhsaker & Leland, 1979]. Vegetative Predictors of the Abundance of Other Primates Interpretation of the significant effect of total basal area of large trees on the relative abundance of Angolan colobus is not straightforward because total basal



area of large trees is a function of both the size of trees and density of stems. However, data on gross habitat use show that C. angolensis select semi-evergreen portions in two of the three transects more so than does P. gordonorum [Rovero et al., 2006]. Total basal area was greater in semi-evergreen portions than in the combined deciduous and semi-deciduous portions of the transects (ANOVA: F(1, 54) 5 5.57, Po0.05). The association between forest areas with a large proportion of trees with lianas and Angolan colobus is also consistent, to some extent, with their selection for semi-evergreen forest. The proportion of trees with lianas was far greater in semi-evergreen than deciduous forest (F(1, 44) 5 26.04, Po0.001), but it was even greater in the semi-deciduous than semi-evergreen portions of the transects (F(1, 44) 5 4.50, Po0.05). Among the most common liana species in the study area is Entada rheedii whose seeds the Angolan colobus often eat (T. Struhsaker, unpublished data). Studies on other species are consistent with our results: the abundance of the congeneric Colobus guereza in Kibale National Park was most strongly correlated with large stem density and also with basal area and the percentage of canopy cover [Skorupa, 1988]. Tree species diversity and the proportion of trees with lianas were the two vegetative variables with greatest predictive value in terms of the relative abundance of Sykes’ monkeys. Although incorporation of altitude as a variable influencing Sykes’ monkeys’ abundance increased the percentage of the deviance explained by the model from 13 to 38%, this is surely due to a spurious effect. This is because Sykes’ monkey is abundant over a wider range of altitudes throughout East Africa than any other non-human primate. Furthermore, the altitudinal range covered by our study is not great enough to result in appreciable differences in energetic demands between the lowest and highest points on our transects. The effect of altitude in our transects is most likely due to its correlation with vegetation. Sykes’ monkeys were rarely found in semi-evergreen forest, being more abundant in semi-deciduous habitats and portions of the transects with secondary and degraded habitat [Rovero et al., 2006]. These latter habitats are more abundant in the lower altitudes of our transects, which, in turn, explains the effect of elevation in the model. Tree species diversity was highest in semideciduous portions of the transects then in deciduous and semi-evergreen, the difference being significant only between semi-deciduous and deciduous (ANOVA: F(1, 18) 5 33.91, Po0.001). Although liana cover was high in semi-evergreen habitats, especially in areas with gaps and regenerating vegetation, it was even higher in semi-deciduous habitats (see above) and in the large open area found along the first transect. In contrast, Skorupa [1986] found that the abundance of the conspecific blue monkey Cercopithecus mitis stuhlmanni in Kibale National Park appeared to vary independently of the 11 botanical variables that were tested. Despite the small sample size, the regression model for the yellow baboon had the highest predictive power probably because this species is confined to the lowland, miombo (deciduous) parts of the transects, which are characterized by reduced tree species diversity (see above) and, on average, reduced basal area of large trees in comparison with semi-evergreen portions (see above). The possible causal effects of these vegetative variables on baboon abundance are not apparent. Elevation had a very important effect, as the percentage of deviance explained by the model raised from about 43–80% with the inclusion of this variable. As discussed for Sykes’ monkeys, this is surely a spurious relationship because baboons occur at much higher elevations in numerous locations elsewhere in East Africa. Other variables that correlate with altitude are more likely to explain the variance in baboon abundance. For example, the miombo



habitat most frequented by baboons was at the lowest elevations of the transects and closest to the agricultural fields outside and adjacent to the park boundaries. Baboons frequently fed in these fields, particularly so on maize. Consequently, the strong association between baboons and miombo woodland may have been significantly influenced by proximity of the agricultural fields. Although there might be competitive exclusion between baboons and Sanje mangabeys, as shown by an inverse pattern of occurrence whereby mangabeys were more confined to evergreen habitats at higher elevations [Rovero et al., 2006], this is unlikely to have affected baboon abundance at higher elevations because baboons are so much larger than Sanje mangabeys. Furthermore, in an equivalent species-pair situation along the Tana River, Kenya, yellow baboons are clearly dominant to and regularly supplant the Tana River mangabeys (T. Struhsaker, personal observation), considered to be conspecific with the Sanje mangabey [Grubb et al., 2003]. If there is competitive exclusion, then it is likely that the baboons have a negative effect on mangabey abundance at lower elevations in the miombo. Conclusions and Recommendations The predictive powers of this study were generally low, especially for the two colobine monkeys. This raises the question as to how future studies that address issues of micro-habitat selectivity might be improved. Although we adopted the standard procedure of sampling trees Z10 cm DBH, results indicate that it would have been adequate to sample only trees Z20 cm DBH. Future efforts should test the generality of this finding. If this proves to be widely applicable, then sampling effort could be drastically reduced (in our case by 60% of the stems measured), as could the associated financial costs of field work. Alternatively, as in our case, it would have allowed us to sample larger areas of habitat, such as by increasing the strip width of the vegetation plots. This may have produced stronger predictive models. Our study also points out the importance of having data on food habits and range use from detailed studies of social groups. This kind of information is generally not available for census studies such as ours. Had this kind of information been available for all of the species in our study, our predictive models would likely have been stronger. It is important to note that the strongest, positive predictors of abundance were derived for the red colobus, the only species for which we had feeding data. The model derived for baboons also had strong predictors but these were negative, thus indicating avoidance. Our study also points out how difficult it is to predict the abundance of a species like the Udzungwa red colobus whose diet appears to be at least as dependent, if not more so, on food quality rather than quantity. We speculate that for this species dietary complement may be of paramount importance. Despite the predictive limitations of our results, the scale of sampling used in our study was adequate to detect small-scale habitat variations along the transects including forest disturbance as a result of past human activities. For example, total basal area and stem density were negatively correlated with the amount of gap in forest cover. These gaps varied in area from less than 100 m2 to several hundreds m2 and were largely derived from past pitsawing and human habitation. Our sampling technique also provides pragmatic predictive models that help explain the differences in occurrence of primates among gross, qualitative habitat types. Especially useful is the model for the endangered and endemic Udzungwa red colobus that stresses the importance of both mature habitat structure and variety of food plant trees. In addition, the methods we employed are particularly relevant in the evaluation of ecological variables that



might predict primate abundance in heterogeneous habitats. Our approach is consistent with the view of Chapman & Chapman [1999] that ‘‘yexamining small scale variation in primate density and ecological parameters will prove to be a profitable approach for understanding ecological determinants of primate abundance.’’ There are a number of other variables that affect primate abundance and should also be considered in future studies, such as predation [over hunting by humans and chimpanzees, Mitani & Watts, 1999; Mitani et al., 2000; Peres, 1999; Stanford, 1998; Struhsaker, 1999; Watts & Mitani, 2002] and disease [Pope, 1998, 2000; Struhsaker, 2000a]. Similarly, loss of habitat elsewhere can lead to artificially high and unstable densities of primates due to population compression through immigration [Decker, 1994; Siex, 2003; Struhsaker, 2000b]. Future research needs to consider not only habitat quality, but also other variables such as these because most ecosystems and animal populations are probably in nonequilibrium states. In terms of conservation implications, the results of our study clearly indicate the importance of protecting old growth, evergreen and semi-deciduous forest for maintaining the two colobus populations. Forests with numerous and large areas of open gap are unsuitable for the colobus. The endangered and endemic Sanje mangabey is also most common in the same habitat (86% of sightings in semievergreen and semi-deciduous forest). ACKNOWLEDGMENTS Financial support for our research was kindly provided by the Margot Marsh Biodiversity Fund and by the National Geographic Society (grant number 7433-03). FR was also supported through post-doctoral funding from the Provincia Autonoma di Trento through the Museo Tridentino di Scienze Naturali. Thanks to Arafat Mtui for invaluable help in data collection and the wardens and staff at the Udzungwa Mountains National Park for logistic support. Thanks to Alessandro Pucci for data on red colobus’ diet, Quentin Luke and Chris Ruffo for plant identification. The analysis gained from discussions with John Birks, Andy Marshall and Alessandra Petrucci. Research permits were granted by the Tanzania Wildlife Research Institute, the Tanzania Commission for Science and Technology and the Tanzania National Parks. We are grateful to Colin Chapman, Andrew Marshall, Tim O’Brien, Paul Struhsaker and two anonymous reviewers for comments on earlier versions of this manuscript.

REFERENCES Brugiere D, Gautier JP, Moungazi A, GautierHion A. 2002. Primate diet and biomass in relation to vegetation composition and fruiting phenology in a rain forest in Gabon. Int J Primatol 23:999–1023. Butynski TM. 1990. Comparative ecology of blue monkeys (Cercopithecus mitis) in high- and low-density subpopulations. Ecol Monogr 60:1–26. Chapman CA, Chapman LJ. 1999. Implications of small scale variation in ecological conditions for the diet and density of red colobus monkeys. Primates 40:215–231.


Chapman CA, Chapman LJ, Bjorndal KA, Onderdonk DA. 2002. Application of protein-to-fiber ratios to predict colobine abundance on different spatial scales. Int J Primatol 23:283–310. Chapman CA, Chapman LJ, Naughton-Treves L, Lawes MJ, McDowell LR. 2004. Predicting folivorous primate abundance: validation of a nutritional model. Am J Primatol 62:55–69. Cowlishaw G, Dunbar R. 2000. Primate conservation biology. Chicago: University of Chicago Press. 498p.

Primate Abundance and Vegetation / 1255 Crawley MJ. 1993. GLIM for ecologists. Oxford: Blackwell Science. 304p. Davenport TRB, Stanley WT, Sargis EJ, De Luca DW, Mpunga NE, Machaga SJ, Olson LE. 2006. A new genus of African monkey, Rungwecebus: morphology, ecology and molecular phylogenetics. Science 312:1378–1381. Davies AG. 1994. Colobine populations. In: Davies AG, Oates JF, editors. Colobine monkeys: their ecology, behaviour and evolution. Cambridge: Cambridge University Press. p 285–310. Decker BS. 1994. Effects of habitat disturbance on the behavioural ecology and demographics of the Tana river red colobus (Colobus badius rufomitratus). Int J Primatol 15:703–737. ˚ Dinesen L, Lehmberg T, Rahner MC, Fjeldsa J. 2001. Conservation priorities for the forests of the Udzungwa Mountains, Tanzania, based on primates, duikers and birds. Biol Cons 99:223–236. Ehardt CL, Jones TP, Butynski TM. 2005. Protective status, ecology and strategies for improving conservation of the Sanje mangabey Cercocebus sanjei in the Udzungwa Mountains, Tanzania. Int J Primatol 26: 557–583. Ganzhorn JU. 1992. Leaf chemistry and the biomass of folivorous primates in tropical forests. Oecologia 91:540–547. Grubb P, Butynski TM, Oates JF, Bearder SK, Disotell TR, Groves CP, Struhsaker TT. 2003. Assessment of the diversity of African primates. Int J Primatol 24:1301–1357. Jones T, Ehardt CL, Butynski TM, Davenport TRB, Mpunga NE, Machaga SJ, De Luca DW. 2005. The highland mangabey Lophocebus kipunji: a new species of African monkey. Science 308:1161–1164. Lovett JC, Bridson DM, Thomas DW. 1988. A preliminary list of the moist forest angiosperm flora of Mwanihana Forest Reserve, Tanzania. Ann Missouri Bot Gardens 75:874–885. Lovett JC, Marshall AR, Carr J. 2006. Changes in tropical forest vegetation along an altitudinal gradient in the Udzungwa Mountains National Park, Tanzania. African J Ecol 44:478–490. Magurran AE. 1988. Ecological diversity and its measurements. London: Chapman & Hall. 192p. Maindonald J, Braun J. 2003. Data analysis and graphics using R–an example-based approach. New York: Cambridge University Press. 362p. Marshall AR. 2007. Disturbance in the Udzungwas: Responses of Monkeys and Trees to Forest Degradation. Ph.D. thesis, University of York, York. Marshall AR, Topp-Jørgensen JE, Brink H, Fanning E. 2005. Monkey abundance and social structure in two high elevation forest

reserves in the Udzungwa Mountains of Tanzania. Int J Primatol 26:127–145. Mbora DNM, Meikle DB. 2004. Forest fragmentation and the distribution, abundance and conservation of the Tana river red colobus (Procolobus rufomitratus). Biol Cons 118:67–77. McCullagh P, Nelder JA. 1989. Generalized linear models. London: Chapman & Hall. 511p. Medley KE. 1993. Primate conservation along the Tana river, Kenya: an examination of the forest habitat. Cons Biol 7: 109–121. Milton K. 1982. Dietary quality and populations regulation in a howler monkey population. In: Leigh EG, Rand AS, Windsor DM, editors. The ecology of a tropical forest: seasonal rhythms and long-term changes. Washington: Smithsonian Institution Press. p 273–290. Mitani JC, Watts D. 1999. Demographic influences on the hunting behavior of chimpanzees. Am J Phys Anthropol 109: 439–454. Mitani JC, Struhsaker TT, Lwanga JS. 2000. Primate community dynamics in old growth forest over 23.5 years at Ngogo, Kibale National Park, Uganda: implications for conservation and census methods. Int J Primatol 21:269–286. Morrison ML, Marcot BG, Mannan RW. 1998. Wildlife-habitat relationships: concepts and applications. Madison: University of Wisconsin Press. 435p. Oates JF, Whitesides GH, Davies AG, Waterman PG, Green SM, Dasilva GL, Mole S. 1990. Determinants of variation in tropical forest primate biomass: new evidence from West Africa. Ecology 71:328–343. Onderdonk DA, Chapman CA. 2000. Coping with forest fragmentation: the primates of Kibale National Park, Uganda. Int J Primatol 21:587–611. Peres CA. 1999. Effects of subsistence hunting and forest types on the structure of Amazonia primate communities. In: Fleagle JG, Janson CH, Reed KE, editors. Primate communities. Cambridge: Cambridge University Press. p 268–283. Pope TR. 1998. Effects of demographic change on group kin structure and gene dynamics of red howling monkey populations. J Mammal 79:692–712. Pope TR. 2000. The evolution of male philopatry in neotropical monkeys. In: Kappeler PM, editor. Primate males: causes and consequences of variation in group composition. Cambridge: Cambridge University Press. p 219–235. Rovero F, Struhsaker TT, Marshall AR, Rinne TA, Pedersen UB, Butynski TM, Ehardt CL, Mtui AS. 2006. Abundance of diurnal


1256 / Rovero and Struhsaker primates in Mwanihana Forest, Udzungwa Mountains, Tanzania: a multi-observer comparison of line-transect data. Int J Primatol 27:675–697. Seber G. 1982. The estimation of animal abundance and related parameters. New York: Macmillan. 654p. Siex KS. 2003. Effects of population compression on the demography, ecology, and behavior of the Zanzibar red colobus monkey (Procolobus kirkii). Ph.D. thesis, Duke University, Durham. Skorupa JP. 1986. Responses of rainforest primates to selective logging in Kibale Forest, Uganda. In: Benirschke K, editor. Primates: the road to self-sustaining populations. New York: Springer-Verlag. p 57–70. Skorupa JP. 1988. The effect of selective timber harvesting on rain-forest primates in Kibale forest, Uganda. Ph.D. thesis, University of California, Davis. Sokal RR, Rohlf FJ. 1995. Biometry. New York: Freeman WH & Co. 896p. Stanford CB. 1998. Chimpanzee and red colobus: the ecology of predator-prey ecology. Cambridge MA: Harvard University Press. 296p. Struhsaker TT. 1975. The red colobus monkey. Chicago: University of Chicago Press. 311p. Struhsaker TT. 1981. Census methods for estimating densities. In: National research council, editor. Techniques for the study of primate population ecology. Washington DC: National Academy Press. p 36–80. Struhsaker TT. 1997. Ecology of an African rainforest: logging in kibale and the conflict between conservation and exploitation. Gainesville: University Press of Florida. 434p. Struhsaker TT. 1999. Primate communities in Africa: the consequence of long-term evolution or the artifact of recent hunting? In: Fleagle JG, Janson CH, Reed KE, editors. Primate communities. Cambridge: Cambridge University Press. p 289–294. Struhsaker TT. 2000a. The effects of predation and habitat quality on the socioecology


of African monkeys: lessons from the islands of Bioko and Zanzibar. In: Whitehead PF, Jolly CJ, editors. Old world monkeys. Cambridge: Cambridge University Press. p 393–430. Struhsaker TT. 2000b. Variation in adult sex ratios of red colobus monkey social groups: implications for interspecific comparisons. In: Kappeler PM, editor. Primate males: causes and consequences of variation in group composition. Cambridge: Cambridge University Press. p 108–119. Struhsaker TT, Leland L. 1979. Socioecology of five sympatric monkey species in the Kibale Forest, Uganda. In: Rosenblatt JS, Hinde RA, Beer C, Busnel MC, editors. Advances in the study of behaviour. New York: Academic Press. p 159–228. Struhsaker TT, Marshall AR, Detwiler KM, Siex K, Ehardt CL, Lisbjerg DD, Butynski TM. 2004. Demographic Variation Among the Udzungwa Red Colobus (Procolobus gordonorum) in Relation to Gross Ecological and Sociological Parameters. Int J Primatol 25:615–658. Warner MD. 2002. Assessing habitat utilization by Neotropical primates: a new approach. Primates 43:59–71. Wasserman MD, Chapman CA. 2003. Determinants of colobine monkey abundance: the importance of food energy, protein and fiber content. J Anim Ecol 72:650–659. Watts DP, Mitani JC. 2002. Hunting behavior of chimpanzees at Ngogo, Kibale National Park, Uganda. Int J Primatol 23:1–28. Whitesides GH, Oates JF, Green SM, Kluberdanz RP. 1988. Estimating primate densities from transects in a West African rain forest: a comparison of techniques. J Anim Ecol 57:345–367. Whittingham MJ, Stephens PA, Bradbury RB, Freckleton RP. 2006. Why do we still use stepwise modelling in ecology and behaviour? J Anim Ecol 75:1182–1189. Wieczkowski J. 2004. Ecological correlates of abundance in the Tana mangabey (Cercocebus galeritus). Am J Primatol 63:125–138.