SPATIAL ORGANIZATION OF A HERPETOFAUNA ON AN ELEVATIONALGRADIENT REVEALED BY NULL MODEL TESTS

ULRICH HOFER,1 LOUIS-FELIX BERSIER,2 AND DANIEL BORCARD3

1 Department of Vertebrates, Natural History Museum, Bernastrasse 15, 3005 Berne, Switzerland2 Zoological Institute, University of Neuchatel, rue Emile-Argand 11, 2007 Neuchatel, Switzerland3 Departement de sciences biologiques, Universite de Montreal, C.P. 6128, Succursale ‘‘Centre-Ville,’’ Montreal,Quebec H3C 3J7 Canada

Abstract. Five null model tests were applied to the herpetofaunal assemblage on thewestern slope of Mount Kupe, Cameroon. Based on the pattern of species range boundariesand abundances along the primary forest elevational gradient, ranging from 900 to 2000m, the relative importance of interspecific competition and ecotones in structuring theassemblage was assessed. Tests were run for (1) all species, (2) amphibians, (3) reptiles,(4) amphibians dependent on streams for reproduction, and (5) amphibians that do not usestreams for reproduction.

For three null models, the observed patterns did not differ from random expectations.The results indicated that there are very few species whose gradient distributions may belimited by interspecific competition between congeners. Significant discontinuities in abun-dance patterns and range boundary dispersion revealed zonations in all subsets analyzedbut neither indicated distinct species groups with sharp exclusion boundaries nor a strongresponse to vegetational ecotones. Physical factors varying in parallel with the gradientand specific habitat components, particularly water bodies suitable as amphibian breedingsites, are suggested to be the dominant factors limiting gradient distributions of amphibiansand reptiles on Mount Kupe. The zonations revealed suggest a pattern of three spatiallynonexclusive species groups: physical factors separate distinct lowland and montane specieslimited by physiological constraints and produce faunal discontinuities in the lower sub-montane forest around 1300 m; this boundary is encompassed by the range of a group ofanuran species, whose distributions on the gradient are centered at intermediate elevationand appear to be limited by specific habitat requirements. The response to predominantlyabiotic factors suggests a basic difference from endotherms, where biotic factors seem tobe of major importance in limiting elevational distributions.

Keywords: Amphibia; Cameroon; ecotone; gradient distribution; interspecific competition; null model; Reptilia;species groups; tropical forest.

INTRODUCTION

The pioneering approaches by Whittaker (Whittakerand Niering 1965, 1975, Whittaker 1967) and Terborgh(1971) provided the basis for a research domain thataimed at elucidating the structure of communities alongenvironmental gradients. The majority of faunal studiesfocused on diversity and endemism on elevational gra-dients. Declining species richness with increasing el-evation has been demonstrated for many taxa and isnow widely accepted as a general pattern (Rahbek1995), but attempts to establish further uniformity haveproven difficult. That species richness does not nec-essarily decline monotonically with altitude was dem-onstrated in birds and insects. In an analysis of all SouthAmerican tropical land birds, Rahbek (1997) examinedfour species richness/elevation models, two describinga monotonic relationship and two postulating hump-shaped patterns, one of the latter based on a null modelexpectation formulated by Colwell and Hurtt (1994).By factoring out area, Rahbek showed that species rich-

ness is not highest in the 0–500 m zone, but peaksbetween 500 and 1000 m. In insects, mid-elevationpeaks in species richness were observed by Janzen(1973) and Olson (1994), but not by Lawton et al.(1987) and Wolda (1987). In mammals, hump-shapedpatterns have been found in small mammals (Pattersonet al. 1989), whereas bats exhibit a monotonic decline(Graham 1990, Patterson et al. 1996). On a coarse scale,amphibians and reptiles essentially show a monotonicdecline in species richness (Heatwole 1982), althoughopposite trends have been observed in particular hab-itats (Heyer 1967).

Authors usually claim a complex interplay of factorsto explain the variability of observed gradient patterns,including nonbiological ones such as differences insampling regime (Wolda 1987, McCoy 1990), scaledifferences (Patterson et al. 1996), and species–areaeffects not adequately accounted for (Rahbek 1995,1997). Lawton et al. (1987) explain the decline of insectspecies richness with altitude by a decrease in habitat

Published in Ecology 80, issue 3, 976-988, 1999which should be used for any reference to this work

1

area, resource diversity, and primary productivity, em-phasizing the changes in host plant diversity and plantarchitecture. Vegetational habitat structure has alsobeen found to affect elevational distributions of smallmammals on Taiwan (Yu 1994) and of Andean birds(Terborgh 1985), although competitive interactionsseem to transcend the effect of vegetational ecotonesin the latter case (Terborgh and Weske 1975, Terborgh1977). From a macrogeographic perspective, the im-pact of phylogeny and speciation modes on diversitypatterns on local gradients has been repeatedly em-phasized (e.g., Duellman 1979, Cadle and Patton 1988,Patterson et al. 1996). For most regions, however, morehistorical data are required to elucidate these processes.In all, the various effects accounting for the distributionpatterns of assemblages on elevational gradients arewell recognized, but attempts to unravel them by ex-plicitly testing for the relative importance of singlefactors remain sparse.

Whittaker (1967) described four models of com-munity organization on gradients, distinguished on thebasis of whether or not species occur in discerniblegroupings and the extent to which boundaries betweenspecies are exclusive. The four models are: (1) distinctgroups of species with sharp exclusion boundaries; (2)sharp exclusion boundaries between competing spe-cies, but no natural groupings; (3) groupings of speciesthat are not exclusive; (4) no groupings and no exclu-sion. The models themselves are generated by fourmechanisms: (1) biotic interactions, (2) abiotic limits,(3) ecotones, and (4) dispersal constraints. An explicitattempt to test for the relative importance of the firstthree mechanisms in limiting species’ distributions wasmade by Terborgh (1971, 1985). By comparing the ele-vational distributions of bird species on one referenceand three control gradients in the Andes, Terborgh iden-tified direct and diffuse competitive exclusion as thefactor of ‘‘overriding importance’’ in limiting aviandistributions (Terborgh 1985). Evidence was based onobserved displacements of species boundaries in theabsence of potentially competing congeners on controltransects, and on the response of species to downwardor upward shifts of homologous ecotones on the variousgradients.

In this paper, we make use of null model tests toanalyze the gradient distribution pattern of a herpeto-faunal assemblage from Mount Kupe, Cameroon. Themain models have been developed by Pielou (1977,1978) and Dale (1984, 1986, 1988), and disregardabundances. Therefore, we also used chronologicalclustering (Legendre et al. 1985), a constrained per-mutation method that retains abundances. The appli-cation of some of Pielou’s and Dale’s models to sam-pling designs based on discrete points at regularlyspaced intervals required modifications that are alsodealt with in this paper. In combination, these tests helpto assess the relative importance of interspecific com-petition and ecotones in the gradient distribution of an

assemblage, based on its patterns of range boundariesand abundances. Potentially competing species pairscan be singled out and, at the assemblage level, non-random patterns can be identified.

The models for analyzing community structure onenvironmental gradients are well developed, but lackbroad application (Gotelli and Graves 1996). To obtainmeaningful results, the null model tests must meet threecriteria: (1) data on elevational ranges must be obtainedfrom single locations, (2) sampling must occur on acontinuous scale or at regularly spaced intervals, and(3) sampling must be done with equivalent intensity.In the case of tropical upland herpetofaunas, eleva-tional ranges of species, when given, were evaluatedby pooling data from several localities, e.g., from tran-sects on different mountainsides (Duellman 1979, Ca-dle and Patton 1988, Inger and Stuebing 1992, Duell-man and Wild 1993), or else they encompassed a smallelevational range (e.g., Raxworthy and Nussbaum1994). In studies explicitly addressing regional or localelevational patterns (Brown and Alcala 1961, Heyer1967, Scott 1976, Fauth et al. 1989), different samplingdesigns, small sample sizes, or confounded site andyear effects contributed to controversial conclusions(Fauth et al. 1989). Unfortunately, none of the exam-ined papers fulfilled all three criteria to an extent thata reanalysis of published data with null models is likelyto yield ecologically relevant results. A broader anal-ysis of gradient patterns and a comparison of perfor-mance of the various null model tests using data withdifferent sampling designs and degrees of resolutionare still desirable.

STUDY SITE

Mount Kupe, 48459 N, 98429 E, in the southwestprovince of Cameroon, is the first major peak inlandfrom Mount Cameroon, ;100 km northeast, but issmaller and only half as high at 2064 m. It forms partof an extensive volcanic mountain range running fromBioko Island to Mount Cameroon in the southwest, onto the Bamenda Highlands and the Adamawa Highlandsin the northeast, with the Obudu and Mambila Plateausextending into Nigeria. This range, known as the Cam-eroon Highlands, is the only highland area in tropicalWest Africa sufficiently high and extended to developlarge and distinct floral and faunal assemblages. Thehighest number of locally endemic amphibians onmainland Africa is found here, with .60 species beingrestricted to this region (Jenkins and Hamilton 1992)out of roughly 200 anuran species known from Cam-eroon alone (Amiet 1989).

In the Late Pleistocene, the Cameroon Highlandswere exposed to severe climatic fluctuations (Hamilton1992). Warm and humid conditions, alternating withespecially dry and cold phases, caused expansions andcontractions of montane biotopes and fragmentationsof lowland forest, repeatedly favoring allopatric spe-ciation and subsequent range extensions. During harsh

2

TABLE 1. Description of the elevational gradient sampled on the western slope of Mount Kupe, Cameroon, March–November1994, and man-hours of sampling time spent at each elevation.

Contour

Accessibletransect

length (m)

Accessibleriparian

zone length (m) Stream identity

Number of man hours†

dt nt dr nr

90010001100120013001400

240140230200220220

1502750503127

A, BBCCCC

444344433844

343636373132

910

9776

1211131210

9150016001700180019002000

240500340220400790

155

150

D

E

394242434141

383736353533

5

10

11

12

† Abbreviations: dt, day transect; nt, night transect; dr, riparian zone during day; nr, riparian zone at night. At the contours1600, 1700, 1800, and 2000 m, no watercourse was found.

climatic periods, many species survived in several‘‘refuges.’’ Modern distribution patterns of forest or-ganisms, centers of biodiversity, and gradients of spe-cies richness indicate three major centers of forest sur-vival in tropical Africa, Mt. Kupe being part of the‘‘Cameroon/Gabon refuge’’ (Maley 1987). Present-daypatterns in western Cameroon include species with rel-atively large distributions over widely separated moun-tains, and other species restricted to a single or smallgroup of mountains. As Amiet (1987) found no dif-ferentiation between isolated populations of montaneanuran species in Cameroon, he concluded that the lastextension of montane ranges to low altitudes was rel-atively recent and coincided with the last major coldphase, from 25 000 to 15 000 yr BP.

Mount Kupe is a steep-sided, cone-shaped mountainof horst uplifts and syenitic and granitic intrusionsformed by block faulting and bounded by structuraltroughs, within which volcanic activity has createdsmall cones (Tye 1986). Today the mountain is coveredby ;2100 ha of undisturbed, closed-canopy submon-tane forest, characterized by a fairly uniform structurewith a sparse ground layer and a thin understory (Tho-mas 1986). The canopy is closed and is ;30 m inheight, with a few scattered emergent trees. On ridges,the forest has a more open canopy #18 m tall and ahigher density of smaller understory trees. The statureof the forest gradually declines with elevation until thecanopy is at 10–15 m near the summit. The summitgives way to small areas of grassland. Although themountain is high enough to support afromontane forest,the typical montane vegetation is absent on Kupe.Above 1800 m however, there are a few montane plantspecies; this part of the forest is best regarded as tran-sitional between submontane and montane. Accordingto Lane (1994), the lower transitional zone on MountKupe, between submontane and lowland forest, extendsfrom 700 to 900 m. However, the primary forest below900 m has been logged or is severely degraded, exceptfor a few patches on the southwestern and southern

slopes of the mountain. Mount Kupe holds several per-manent streams, but, as some run partly underground,no watercourse was found within the study area be-tween 1600 and 1900 m.

Suchel (1972) gives an annual rainfall average of4891 mm on Mount Kupe, measured over a period of21 yr. The rainy season lasts for 7 mo. from April toOctober, with heavy rains almost daily and rainfallpeaking in August (878 mm). With no month receiving,70 mm, an appreciable dry season is absent. Reliabletemperature data on Kupe are not readily available. Theminimum temperature measured during the entire sam-pling period was 13.88C (1900 m, 13 March, at night);the maximum was 23.88C (900 m, 21 April, during theday).

MATERIAL AND METHODS

Data acquisition

Data on gradient distribution were acquired betweenMarch and November 1994. Fieldwork was restrictedto the primary forest between the village of Nyasoso(800 m) and the summit of the mountain (2064 m). Thelower altitudes up to 1200 m were sampled from thevillage. The work in the higher elevations started fromtwo field camps, set up at 1550 m and 1930 m, re-spectively. Seven periods of up to 10 d each were spentin the camps, alternating with stays at the village. InJuly and August, camping was suspended due to con-tinuous rainfall.

At 12 points between 900 and 2000 m (Table 1),separated by 100 m, transects were opened along thecontour line and were linked by two vertical main trails.To adequately sample species potentially confined towatercourses, we examined streams separately; ripariansampling zones could be located at eight of the 12elevations. Because a regular elevational spacing iscrucial for the null model analyses presented here, the100 m was maintained regardless of the suitability ofthe topography. This caused some transects and ripar-

3

TABLE 1. Extended.

No. sampling sessions in each month (March–November)

M A M J J A S O N

230000

004343

133443

434467

343200

633420

121112

322246

111100

0005

105

434263

542311

633000

000000

000000

323455

422000

000000

ian zones to fall at areas where a free extension inlength was severely hampered or impossible. Thus, themaximal transect and riparian zone lengths accessiblevaried from 140 to 790 m and from 27 to 150 m, re-spectively. The widths were constrained both by thetopography and the structure of the understory; wherepossible, a strip up to 20 m wide was examined alongtransects and up to 10 m wide on both streamsides atriparian zones. Yet, the topographic heterogeneityamong the elevational zones prevented a reasonablyaccurate delimitation of areas. We therefore based anequal sampling effort (ESE) at all elevations on a time-constrained technique (Campbell and Christman 1982).The basic schedule per elevation was eight day tran-sects (90 min each), five night transects (120 min each),and, for riparian zones, four day (30 min each) andfour night visits (45 min each). The sampling methodadopted was ‘‘cruising collecting’’ (Inger and Colwell1977), i.e., 3–5 people moved slowly along the transect,moving floor debris, turning logs and stones, rippingapart rotten wood, digging soil in the root system ofbig trees and under logs and inspecting the herb andshrub layer up to ;10 m; in riparian zones, the stream-bed was examined in addition. The amount of timespent at each elevation was counted in man-hours, i.e.,the time spans per sample were multiplied by the num-ber of workers involved. Given the indicated method,transect widths, and time spans, the crew covered upto 200 m on a transect and up to 40 m on a riparianzone sample. At the 1000-m elevation with a transectlength of only 140 m, we searched a broader strip,which could be extended to 45 m at this site. Heavyrains regularly slowed down or interrupted samplingsessions and required an adjustment. In general, timespans of individual samples were prolonged. In themiddle of the rainy season, however, differences in-creased to such an extent that a sixth night transect hadto be added at all elevations from 1500 to 2000 m tomaintain approximately equal regimes. Despite our ef-forts, we could not avoid some variability in the totalsampling time spent at each elevation (Table 1).

Between 14 March and 7 November 1994, the crewcompleted 226 samples totaling 1075 man-hours. Spec-imens encountered at odd times and during samplesbroken off due to heavy rains were added to the rowtotals of the Appendix, but were ignored in the analysis.Animals were either collected or were marked by toe-(skinks, geckos, frogs), gular-crest- (chameleons), orscale-clipping (snakes), and were released at the endof each sampling session. A collection of voucher spec-imens is deposited at the Natural History Museum ofBerne, with additional specimens at the Alexander Ko-enig Zoological Research Institute and Zoological Mu-seum in Bonn and in the collection of the Mount KupeForest Project in Nyasoso.

Analysis

To reveal nonrandom patterns on the gradient, thespecies 3 sample matrix has been subjected to severalnull model tests developed for the analysis of distri-butions along one-dimensional environmental gradi-ents. From the null models of Pielou (1977, 1978),originally applied to seaweed species on a latitudinalgradient, and of Dale (1984, 1986, 1988), originallyapplied to intertidal algae, we selected those applicableto our data and likely to yield meaningful information.For details of the methods, readers are referred to theoriginal papers. The five hypotheses tested are:

1) The species’ ranges, given their observed lengths,are located independently and at random within thetotal gradient length (Pielou 1977, conditional hypoth-esis H2). Each species pair is assigned a l value ac-cording to the pattern of overlap (0, no overlap; 1,partial; 2, complete; 1.5, partial or complete, i.e., thetwo upper, lower, or all boundaries coincide), the sumof l giving the observed overlap of the entire assem-blage, Ls. This value is compared to the expected over-lap for the entire assemblage E(Ls), which is the averageof expected amounts of overlap computed for all pos-sible species pairs. Strong deviations of Ls from E(Ls)indicate an unusually nested (Ls . E(Ls)) or nonover-lapping assemblage (Ls , E(Ls)).

4

2) The downslope boundary of a species is followedsignificantly often by the upslope boundary of another,i.e., the number of observed contiguities differs sig-nificantly from random expectation (Dale 1984). In theoriginal formulation of the test, boundaries are consid-ered contiguous regardless of the distance betweenthem, provided no other boundary intervenes (‘‘con-tiguities of sequence’’). However, ‘‘the contiguity hy-pothesis refers to ecological contiguities’’ (Dale 1984:94), where the upslope boundary of one species co-incides exactly with the downslope boundary of an-other. Significantly more such contiguities than ex-pected are consistent with a competitively structuredcommunity, with similar species replacing each otheron the gradient.

3) The observed gap (g) or overlap (y) length forany pair of species, given their observed range lengths,differs significantly from random expectation (Dale1986). Species pairs with significantly small gap sizesor overlaps may competitively interact. If a larger as-semblage of potential competitors is analyzed, the num-bers of significantly high or low g and y help to revealthe model of community organization. However, thelack of independence in the set of values does not allowsignificance testing at the community level (Dale1986).

4) The clumping (or out-spacing) of the species’range boundaries (either upper or lower) differs sig-nificantly from random expectation (Dale 1988). Thestatistic Wm measures the variability of interboundarydistances; the serial autocorrelation statistic hm mea-sures the degree of clumping of more than two bound-aries at a time. Significantly large values of hm indicatethe presence of boundary clumps. Because multipletests are performed simultaneously (all possible pairsof range boundaries are compared), we applied a Bon-ferroni correction to this analysis (Rice 1989).

5) Pielou’s and Dale’s tests assume sampling on acontinuous scale and simplify the community matrixby ignoring abundances and treating the species’ rangesas a ‘‘sheave’’ of line segments. However, many datasets from local gradients (including our own) consistof species abundances measured at discrete points ona gradient. Because the shapes of the species’ ampli-tudes may add essential information on the mechanismsof distributional limitation (Terborgh 1971), tests thatinclude abundances are desirable. We know of no suchtest operating at the level of pairwise species compar-isons. For an entire assemblage, the chronological clus-tering of Legendre et al. (1985), applicable on temporalor spatial scales and to abundances as well as to binarydata, tests for discontinuities in species compositionalong a gradient. Although the previous tests operateon the species ranges, here the objects in the raw datamatrix are the samples from different points on a gra-dient, under the single constraint that they appear intheir original spatial or temporal succession. A simi-larity matrix is built, using an appropriate index, and

is submitted to a constrained intermediate-link linkageclustering, where only contiguous samples can begrouped. Each fusion is submitted to a permutation testin which samples are randomly reallocated amonggroups. The clustering stops at a preset level of prob-ability of fusion between adjacent groups of samples.Despite its entirely different approach, this test is suit-able for the detection of species groupings, either ex-clusive (Whittaker’s model 1) or nonexclusive (model3). The chronological clustering tests whether or notwithin-group similarities between samples are signifi-cantly higher than among-group similarities.

Correction for ties

Dale’s and Pielou’s tests were designed for rangesmeasured on a continuous scale. They are, however,suitable for ranges measured on a discrete scale, pro-vided that some small modifications are introduced. Inhypotheses 1 and 3, range, gap, and overlap length canbe expressed in standard units (e.g., in numbers of pointintervals), and no modification is required. However,the tests of hypotheses 2 and 4 are biased towardclumping, because more boundaries will coincide ifranges are measured on a discrete rather than a con-tinuous scale. This requires an appropriate handling of‘‘ties,’’ i.e., of potentially artificial concentrations ofrange boundaries on sampled points. Pielou and Rout-ledge (1976) and Underwood (1978) provided tests forthe clumping of range boundaries in data sets based ondiscrete sampling. However, they do not fully replaceDale’s test of hypotheses 2 and 4, as they treat upperand lower boundaries in separate analyses, thereby pre-venting the identification of competitively structuredcommunities (Gotelli and Graves 1996). Consequently,we retained Dale’s tests (1986, 1988) and corrected forties in the following ways.

The test of hypothesis 2 deals with contiguous rangesof species, that is, cases in which the ending of a range(event E) is followed by the beginning of the range ofanother species (event B). Thus, on a sequence of, e.g.,BEBBEEBE, one will find two contiguities (events EB,are in italics). On a discrete scale, one could have, forexample, one ending (E) at a sampling point and twobeginnings (BB) at the following point. This configu-ration cannot be simply equated to the sequence BEB(one contiguity), because the ending and the actual be-ginnings can be located anywhere between the twosampling points. We have three possible sequences:EBB, BEB, and BBE, with, for each sequence, 1, 1,and 0 contiguity, respectively. Thus, the expected num-ber of contiguities is 2/3, that is two contiguities di-vided by three possible sequences. With t being thenumber of E’s plus the number of B’s, and m being thenumber of either E’s or B’s, the number of possiblesequences is , and it can be shown that the totalt( )mnumber of contiguities in these possible sequences is[(t 2 1)!/ (m 2 1)!] The corrected number oft 2 1( ).mcontiguities is the ratio of the total number of conti-

5

guities by the number of possible sequences, which,after simplification, is [m(t 2 m)]/t. Note that this cor-rection must be applied at each point of a gradientsampled on a discontinuous scale.

The test of hypothesis 4 deals with the clumping ofrange boundaries and uses the interboundary distancesas a parameter. We corrected potentially artificialclumps at sample points by maximizing the interboun-dary distances; e.g., with a 100-m interval, two bound-aries falling at 1300 m are transformed into 1283 mand 1316 m. This is a conservative correction, in thata regular spacing renders more difficult the detectionof significant clumps. An unbiased correction, using abroken stick distribution, is very difficult to apply heresince the ordering of the interboundary intervals, whichis arbitrarily set by the investigator, affects the hm sta-tistic.

Significance tests

Pielou (1977) provides a formula for the expectedLs value (hypothesis 1), but no significance test for thisstatistic. We resorted to a permutation test to evaluatethe probability that the observed Ls value is smaller orlarger than the expected value. In this test, we placedthe observed ranges at random positions on the gradientand computed the Ls statistic. We constrained the re-shuffling, in that the numbers of range boundaries co-inciding with the upper or lower end of the gradientwere the same as in the observed distribution of rangeboundaries. We performed 999 permutations. The po-sition of the observed Ls value in the distribution ofrandomized Ls is an estimate of the cumulative prob-ability that Ls deviates from E(Ls). Given that Pielou’sformula for E(Ls) does not correct for range boundariescoinciding with the endpoints of the gradient, we usedthe median of the permutation-based distribution as anestimate of the expected value of Ls.

Although Dale (1984) provides a table of criticalvalues for c (the number of contiguities in hypothesis2), we used the permutation test previously describedto assess the probability that the observed c differs fromrandom expectation. Dale’s critical values are com-puted such that beginnings and endings can occur any-where on the gradient, with the constraint that the firstevent is a beginning and the last one is an ending.However, when sampling distributions on a gradient,one is very likely to find more than one species rangestarting with the lowermost sampling point and endingwith the uppermost. These beginning and ending eventsare fixed and cannot be included in a permutation pro-cedure. This constraint is incorporated in the random-ization test that we have described, making it suitablefor the testing of c. Again, we used the median of thedistribution of randomized c values as an estimate ofthe expected c.

All programs except chronological clustering werewritten in Visual Basic and were tested with idealizedmatrices corresponding to the four model distributions.

They are available as Excel Macros by writing to thefirst author. Chronological clustering was performedwith program CHRONO from the R Package Version 3.0(Legendre and Vaudor 1991).

The five null model tests were applied to our originaldata from Mount Kupe. We ran the tests concerningthe assemblage level for (1) all species; and for foursubsets: (2) amphibians; (3) reptiles; (4) amphibianguild depending on streams for reproduction, i.e., tad-pole development in lotic water or lentic microhabitatsassociated with streams; and (5) stream-independentamphibian guild, i.e., species reproducing by direct de-velopment or breeding in ponds, puddles, and treeholes. To assess to what extent the results are affectedby rare species, the tests were rerun for all five setswith a reduced matrix, in which only the species witha total abundance of $10 individuals are retained. Intests concerning interactions at the species level (hy-potheses 2 and 3), we report significant results for con-generic pairs only, thereby disregarding diffuse com-petition involving heterogeners (sensu Terborgh andWeske 1975).

RESULTS

Composition and quality of the raw data

In all, 2734 amphibians and 596 reptiles were markedor collected, representing 64 species of 35 genera and12 families (Appendix). The amphibian fauna is dom-inated by species reproducing by direct development(Arthroleptis) or in streams. Species not reproducingin stream-associated water bodies belong to the generaWolterstorffina, Nectophryne, Acanthixalus, Chiro-mantis, and Hyperolius; all but the first are recordedrarely and very locally. About 17 of the 38 speciesbelong to the anurans endemic to the Cameroon High-lands (Gartshore 1986). The remaining are lowlandspecies confined to the western border (Petropedetescamerunensis, Astylosternus diadematus) or wide-spread in the Western Equatorial Forest (Amiet 1975).Chiromantis rufescens is the only species also reportedfrom the Upper Guinean Forest.

The reptile fauna is dominated by chameleons andscincids. The genera Chamaeleo and Leptosiaphos ex-hibit a considerable degree of endemism in the Cam-eroon Highlands, with the majority of the Kupe Lep-tosiaphos belonging to three new taxa (J.-L. Perret,personal communication, W. Bohme, personal com-munication), two with a submontane and one with apremontane distribution. Sightings of the single lac-ertid recorded, the diurnal, heliophilic Adolfus afri-canus, were restricted to a treefall at 1560 m. The sam-ple includes all lizard genera hitherto known to showmontane distributions in Cameroon. With 38 specimensof 11 species, snakes were rarely encountered, the onlyexception being the small, cryptozoic Buhoma depres-siceps. Several species known to occur at higher ele-

6

TABLE 2. Results of Pielou’s (1977) null model test on the random location of species’ ranges.

Group S m E(Ls) Ls P

Lambda frequencies

2 1.5 1 0

All species 4930

474650

1210.0511.5

1262.5496.5

0.8550.181

17341

499195

168122

33677

Amphibians 3021

507614

521.5256.5

504.0243.5

0.1910.113

2115

270117

5738

8740

Reptiles 199

421733

142.041.5

144.039.0

0.5710.110

314

388

2519

775

Stream-dependent amphibians 2415

442533

306.5115.5

312.0112.5

0.6290.189

94

18061

2413

6327

Stream-independent amphibians 6 767 17.5 18.5 0.765 0 9 5 1

Notes: In each group, the first row refers to analyses with all species of the respective group; the second row refers toanalyses with abundant species only ($ 10 individuals). All stream-independent amphibians had abundances . 10 individuals.Column heads are: S, number of species; m, mean span length of species in the assemblage; E(Ls), expected overlap basedon 999 permutations of range boundaries under Pielou’s hypothesis 2 (conditional hypothesis); Ls, observed overlap of testedassemblage; and P, cumulative probability. The lambda frequencies reflect the pattern of overlap (0, no overlap; 1, partial;2, complete; 1.5, partial or complete overlap).

vations in western Cameroon were only found in thefarm bush below 900 m.

To assess the quality of each elevational sample, spe-cies accumulation curves were plotted by adding upthe species appearing in the chronologically orderedsamples. On average, 76% of the species ultimatelyobtained were recorded in the first half of the samples(minimum 56%; maximum 100%), and 87% after three-quarters of the samples (minimum 67%). We estimatedthe maximal species richness at each elevation by fit-ting the Michaelis-Menten equation (Raaijmakers1987) and by computing Chao’s (1984) estimator (re-viewed in Colwell and Coddington 1994). Chao’s meth-od gave heterogeneous results, with the observed rich-ness varying between 39% and 100% (mean 73%) ofthe maximum estimated. The Michaelis-Menten pro-cedure yielded more consistent results, with the ob-served richness ranging from 70% to 90% (mean 78%).To what extent these inadequacies in the raw data affectthe direction of the null model results will be addressedin the discussion.

Null model tests

Table 2 lists the results of Pielou’s null model teston the random location of species’ ranges (hypothesis1). The permutations revealed no significant differ-ences between observed and expected overall overlap.High proportions of l values of 1.5 in all groups exceptthe reptiles undoubtedly favored this outcome, thestrongest in the stream-dependent species, where ob-served and expected overlaps are very close. Amongthe reptiles, the narrow ranges recorded for many spe-cies result in the smallest mean span length of allgroups and higher proportions of no or nested overlaps(l 5 0 or 2). As expected, the elimination of rare spe-cies increased the mean span lengths of species andreduced the proportion of zero l values, but the analysisof the reduced matrices again yielded no significant

differences between observed and expected parame-ters.

Table 3 presents the results of Dale’s null models.The test on contiguities (hypothesis 2) reveals a sig-nificant difference from random expectation only forthe entire assemblage, in which the observed numberof contiguities exceeds the expected one (P 5 0.97,cumulative probability). Congeners exhibiting an eco-logical contiguity are Arthroleptis adolfifriderici–A.adelphus and Chamaeleo quadricornis–C. montium at1250 m, and Leptosiaphos species A–L. species C at1350 m. The test of hypothesis 3 on gap sizes (g) andpartial overlaps (y) revealed no pairs with significantlysmall or large g, but large numbers of significantlysmall or large y. However, the test identified only twocongeneric pairs with significantly small y: Arthroleptisadolfifriderici–A. variabilis and A. adolfifriderici–A.species A, overlapping between 1250 and 1350 m, andbetween 1250 and 1450 m, respectively. Congenersrepresent only 3% of all significantly small y and 2.7%of all significantly large y, the majority of the latterconcerning the stream-dependent genera Petropedetes,Leptopelis, and Astylosternus. In all groups except thestream-independent amphibians, significantly small ycompose ,7% and significantly large y compose .56%of all partial overlaps observed. The test on the spacingof range boundaries (hypothesis 4) reveals a signifi-cantly higher variability of interboundary distances Wm

than expected by chance in all five groups. Except forthe stream-independent amphibians, the serial auto-correlation statistic hm indicates significant clumpingof boundaries in all groups. However, significantclumps are indicated at the level of the entire assem-blage only, between 1000 and 1500 m. The apparentlycontradictory result between the number of significantclumps and the hm statistic is due to the Bonferronicorrection. Results of the analyses with the reducedmatrix are largely convergent with those based on the

7

FIG. 1. Discontinuities in species composition and abundance pattern of the herpetofauna of Mount Kupe, Cameroon,revealed by chronological clustering (Legendre et al. 1985) along the elevational gradient. The raw data are log-transformed.The distance measure used is Steinhaus’s coefficient, with the connectance set at 0.5 and P set at 0.05. Dotted lines indicatethe streamless elevations.

complete set of species. Exceptions concern the rep-tiles, where significantly large partial overlaps disap-pear completely and Wm is no more significantly largerthan expected, and in the stream-independent amphib-ians, where no clumping is indicated by hm.

Fig. 1 presents results of the chronological clustering(hypothesis 5). At P , 0.05, the test reveals discon-tinuities in all five groups. Four discontinuities appearin the lower submontane forest, at 1250 m in the entireassemblage and in both reptiles and amphibians whenanalyzed separately, and at 1350 m in the stream-in-dependent amphibians. The remaining ruptures fartherup coincide with structural changes of the habitat: thestream-dependent amphibians respond to the change ofthe main watercourse from permanent to intermittentat 1450 m, the reptiles respond to the submontane–montane transitional zone around 1850 m, and all am-phibians are pooled to the lower end of the streamlesszone at 1550 m. The analysis of the reduced matricesyields the same results in all five groups, i.e., with theconnectance set at 0.5 and P set at 0.05, the numberand location of discontinuities remain unchanged.

DISCUSSION

The major outcome of the analyses is the dominanceof nonsignificant results. The null model tests suggestthat the elevational distributions of the majority of thespecies in the studied assemblage are limited by mech-anisms other than direct interspecific competition andvegetational ecotones. The tests on range boundary dis-persion and dissimilarities between contiguous eleva-tional samples indicate zonations in all subsets. How-ever, where significant boundary clumps appear, theyencompass a broad elevational range from 1000 to 1500m, thereby suggesting a scattered distribution of bound-aries; as revealed by chronological clustering, the localdiscontinuities at 1250 and 1350 m are expressed bychanges in abundances. Taken together, these findingsfit Whittaker’s model 3 of nonexclusive species groupsbetter than model 1 of groups with sharp exclusionboundaries. Model 3 was tentatively confirmed for theamphibians of Cameroon; Amiet (1975, 1989) recog-nized a group of anurans provisionally termed ‘‘faune

peripherique,’’ whose vertical distribution is centeredat intermediate elevations along the western and south-ern slopes of the Cameroon Highlands and extends intothe ranges of both montane and lowland species. OnMount Kupe, this group includes at least Arthroleptissp. A, Astylosternus perreti, Cardioglossa venusta,Conraua robusta, both Leptodactylodon, Leptopelismodestus, Petropedetes parkeri, and P. perreti, Phry-nodon sp. 2, Werneria preussi mertensiana, and Wol-terstorffina parvipalmata. The discontinuities around1300 m separate the two other groups consisting ofdistinct lowland and montane species.

The relatively wide transition zone coincides neitherwith a vegetational ecotone nor with any other obvioushabitat discontinuity, but nevertheless seems to marka change in environmental conditions that is limitingdistributions of leaf litter (e.g., Arthroleptis adelphus),stream-dependent (e.g., Cardioglossa gracilis, Petro-pedetes cameronensis), and probably arboreal (e.g.,Chamaeleo pfefferi, Cnemaspis koehleri, Hemidactylusechinus) species alike. Physical factors, such as tem-perature, precipitation, and evaporation, undoubtedlyaccount for most distributional limits among the low-land and montane species. Moreover, abiotic habitatcomponents such as water bodies suitable for repro-duction and specific microhabitats, often spatially andtemporally restricted in availability, are known to de-termine the local abundance of tropical amphibians andreptiles, and may also affect elevational distributions(e.g., Heyer 1967). For example, the anurans Acan-thixalus spinosus, Chiromantis rufescens, and Hyper-olius acutirostris were chance encounters associatedwith the single standing water body found during theentire sampling period, a puddle on a log in a treefallat 1560 m. Reduced availability of stream-associatedtadpole microhabitats at higher altitudes, suggested byInger and Stuebing (1992) to limit elevational rangesin Bornean anurans, should also affect stream-breedinganurans of Mount Kupe, in particular the ‘‘peripheral’’species. Among the reptiles, the heliophilic lizard Adol-fus africanus may, in higher elevations, depend on largetreefalls, grassy patches, and rocky outcrops to meetthe preferred microclimatic conditions, i.e., it may ul-

8

TABLE 3. Results of Dale’s (1984, 1986, 1988) null model tests.

Group S†

Contiguities‡

E(c z n) c Pc

Significant partial overlaps§

Small y Large y

All species 4930

9.84.0

11.54.9

0.9720.891

62 (7%)52 (16%)

481 (57%)104 (29%)

Amphibians 3021

3.52.2

3.52.0

0.3880.370

27 (8%)22 (13%)

205 (59%)72 (42%)

Reptiles 199

4.30.5

3.41.0

0.7160.856

6 (6%)6 (19%)

53 (56%)0

Stream-dependent amphibians 2415

2.11.7

2.11.3

0.440.079

00

147 (69%)38 (49%)

Stream-independent amphibians 6 0.5 0.5 0.248 5 (36%) 3 (21%)

Notes: In each group, the first row refers to analyses with all species of the respective group; the second row refers toanalyses with abundant species only ($ 10 individuals). All stream-independent amphibians had abundances . 10 individuals.

† Number of species.‡ E(c z n), no. expected contiguities; c, no. observed contiguities; Pc, cumulative probability based on 999 permutations of

range boundaries.§ Small y, no. significantly small partial overlaps; large y, number of significantly large partial overlaps. Percentages of

the total number of observed overlaps in tested group are given in parentheses.\ Clumps, no. significant clumps; E(Wm), expected variability of interboundary distances; Wm, observed variability; P,

cumulative probability; E(hm), expected degree of clumping, hm, observed degree of clumping. Larger values of Wm indicatemore variable interboundary distances; larger values of hm indicate more clumped range boundaries.

timately be limited by temperature. On the Mount Kupegradient, both mechanisms may be of approximatelyequal importance: physical factors probably separatethe vertical distributions of distinct lowland and mon-tane species at around 1300 m; specific habitat com-ponents are suggested as limiting the gradient distri-bution of the ‘‘peripheral’’ anuran fauna, whose rangeencompasses the transitional zone. The resulting pat-tern of overlapping species groups causes sharp zo-nations to disappear. Within-group variability in rangeextensions further accounts for the absence of distinctassemblage-level patterns and the weak response to thenull model parameters concerning zonations.

At the community level, the three ‘‘competition pa-rameters’’ exhibit no significant deviations in supportof Whittaker’s model 2: the numbers of contiguitiesnever significantly exceed random expectation, andproportions of significantly small partial overlaps arelow in all groups except the stream-independent am-phibians. For methodological reasons, the absence ofsignificantly small gap sizes cannot be assigned anyecological relevance. Bench tests showed that, giventhe degree of resolution within our data set, Dale’s teston gap sizes cannot yield significant results at P 50.05. In our data, gap sizes cannot be smaller than 100m, i.e., 1/12 of the total gradient length, whereas sig-nificantly small gaps appear only if the gap size issmaller than 1/40 of total gradient length. In practice,this means that the applicability of this test would re-quire at least 40 sampling points on a discrete scale.

The few gradient distributions probably affected bydirect interspecific competition, i.e., between conge-ners, are found among terrestrial (Leptosiaphos) andarboreal lizards (Chamaeleo) and anurans with directdevelopment (Arthroleptis). Among the latter, Arthro-leptis adolfifriderici and A. variabilis are the most sim-

ilar in morphology. The abundances of these two ef-ficiently sampled species show a marked decline to-ward the contact zone of their amplitudes (Appendix),a phenomenon termed ‘‘repulsion interaction’’ by Ter-borgh (1971: 27), whereby the abundances of presum-ably competing species replacing each other along agradient ‘‘fall off sharply in the zone of contact insteadof trailing off gradually.’’ In Chamaeleo, the pattern isless pronounced, and the low abundances of Leptosia-phos preclude comparable interpretations.

Congeners of the stream-dependent amphibians ex-hibit no contiguities or significantly small partial over-laps, and most of the significantly large partial overlapsamong congeners were found within this group (generaPetropedetes, Leptopelis, and Astylosternus). The lackof response to the null model parameters related tointerspecific competition suggests that this type of in-teraction does not affect the gradient distributions ofstream-dependent anurans. However, Inger and Green-berg (1966) demonstrated with a removal experimentin Sarawak that three syntopic species of stream-de-pendent frogs, congeners and similar in habits, did infact compete. For two species, the authors suggestedthat intraspecific competition fixes maximum popula-tion levels, thus allowing their coexistence. Intraspe-cific competition may also prevent spatial exclusion instream-dependent anurans on Mount Kupe, along withother factors depressing population levels and possiblyniche segregation.

With physical factors and specific, often abiotic, hab-itat components as dominant mechanisms limiting ele-vational distributions, the amphibians and reptiles onMount Kupe differ considerably from tropical endo-therms. In Andean birds, competitive interactions ac-count for about two-thirds of the limits, whereas ec-otones and unspecific factors varying in parallel with

9

TABLE 3. Extended.

Clumping of range boundaries\

Clumps E(Wm) Wm P E(hm) hm P

30

0.020.033

0.0660.081

.0.999

.0.9990.0100.016

0.0190.029

.0.999

.0.99900

0.0330.047

0.1170.139

.0.999

.0.9990.0160.023

0.0250.036

.0.9990.998

00

0.0520.111

0.0870.119

0.9970.720

0.0260.052

0.0330.068

0.9660.922

00

0.0420.067

0.1210.173

.0.999

.0.9990.0200.032

0.0290.027

0.9950.200

0 0.167 0.351 0.997 0.076 0.076 0.520

the gradient each account for about one-sixth of thelimits (Terborgh and Weske 1975, Terborgh 1977,1985). Replacements in elevational distributions alsohave been found in small mammals on Taiwan (Yu1994) and on Andean slopes (Cadle and Patton 1988),without addressing the relative importance of interspe-cific competition. Graham (1990) has suggested thatenergetic requirements, coupled with trophic resourceconstraints, determine many gradient distributions ofPeruvian bats and the rapid decrease in bat diversitywith elevation. Olson (1994) has also recognized re-source constraints and species interactions as potentialfactors limiting elevational distributions of Panaman-ian leaf litter insects, but emphasizes ecotones pro-duced by sharp physical clines or edaphic gradients toexplain pronounced drops in local insect diversity be-tween 1250 and 1500 m. Concerning physical factors,tropical insects and herpetofaunal assemblages may, infact, share some properties in their response to ele-vational gradients. Tropical amphibians are known tobe markedly sensitive to moisture (Toft 1980, Heatwole1982). Among lizards in the lowland forests of South-east Asia, Inger (1980) has found different responses.Arboreal diurnal species tend to be moisture sensitive,whereas terrestrial species, nocturnal and diurnal, re-spond more strongly to temperature. In the primaryforest on the western slope of Mount Kupe, temperatureis likely to exhibit a stronger elevational variation thandoes moisture. Thus, species tolerant to temperatureand independent from specific habitat components mayultimately be limited by interspecific competition intheir elevational distributions. In our assemblage, Ar-throleptis and Chamaeleo are presumably the only gen-era meeting these criteria.

As indicated in the Results, the differences betweenobserved and estimated maximal species richness re-vealed inadequacies in the raw data. The Michaelis-Menten estimates indicate that ;22% of the speciespresent at a site were overlooked. Despite the inac-curacies inherent in such extrapolations (Colwell andCoddington 1994), our conclusions may be affected bysampling bias. We tried to assess the effect of the rarespecies on our results by removing them from the dataand repeating the analyses. No additional community-

level patterns emerged, and the general results did notchange (Tables 2 and 3). This suggests that the ‘‘sig-nal’’ emerges from the more abundant species and islargely unaffected by the rarer (and possibly under-sampled) ones. Thus, we consider it unlikely that amore thorough sampling would result in conclusionsdifferent from ours. To further stay with the quality ofthe raw data, we restricted our statements concerninginterspecific competition and types of elevational re-sponses to abundant (assumed adequately sampled)species representing different adaptive zones and re-productive modes.

In all, we consider null models to be a valuable toolin gradient studies, a field in which a standard protocolis yet to be established and sampling designs rarelyfulfill the requirements for proper statistical analysis(Rahbek 1995; see also comments in Yu 1994). Withthe modifications suggested in this paper, most of thenull models based on presence–absence data becomeequally applicable to assemblages sampled on a con-tinuous scale or at regularly spaced intervals. However,despite the many topographic and logistic difficultiesusually encountered when sampling elevational gra-dients, the three criteria stated in the Introductionshould be accounted for. An appropriate adjustment ofsampling designs in future gradient studies would sub-stantially increase the potential for pattern identifica-tion at the community level.

ACKNOWLEDGMENTS

Fieldwork was supported by grants from the Swiss De-velopment Corporation, the Swiss Academy of Sciences, andthe Natural History Museum of Berne. The Mount Kupe For-est Project provided logistic help and housing in Nyasoso.E. J. Ebung, E. H. Njume, and N. S. Epie assisted the whole,C. Wild and I. C. Ojiawum parts of the fieldwork, and C.Wild introduced U. Hofer to the study area. U. Hofer greatlyappreciates the support of the Bakossi people in Nyasoso, inparticular, K. E. Epie and E. E. Ewang. J.-L. Amiet, J.-L.Perret, W. Bohme, A. Schmitz, and B. Hughes identified manyof the collected specimens. J.-L. Amiet made valuable tax-onomic and ecological comments. For constructive criticismof the manuscript, we thank C. Mermod, D. Meyer, B. D.Patterson, and an anonymous reviewer. This article is part ofa Ph.D. thesis by U. Hofer at the University of Neuchatel.

LITERATURE CITED

Amiet, J.-L. 1975. Ecologie et distribution des amphibiensanoures de la region de Nkongsamba (Cameroun). Annales

10

de la Faculte des Sciences du Cameroun, Yaounde 20:33–107.

. 1987. Aires disjointes et taxons vicariants chez lesAnoures du Cameroun: implications paleoclimatiques. Al-ytes 6:99–115.

. 1989. Quelques aspects de la biologie des amphi-biens anoures du Cameroun. Annee Biologique 28:73–136.

Brown, W. C., and A. C. Alcala. 1961. Populations of am-phibians and reptiles in the submontane and montane for-ests of Cuernos de Negros, Philippine Islands. Ecology 42:628–636.

Cadle, J., and J. Patton. 1988. Distribution patterns of someamphibians, reptiles and mammals of the eastern slope ofsouthern Peru. Pages 225–244 in W. R. Heyer and P. E.Vanzolini, editors. Proceedings of a workshop on neotrop-ical distribution patterns. Academia Brasileira de Ciencias,Rio de Janeiro, Brazil.

Campbell, H. W., and S. P. Christman. 1982. Field techniquesfor herpetofaunal community analysis. Pages 193–200 inN. J. Scott, Jr., editor. Herpetological communities: a sym-posium of the Society for the Study of Amphibians andReptiles and the Herpetologists’ League, August 1977.Wildlife Research Report 13. U.S. Department of the In-terior, Fish and Wildlife Service, Washington, D.C., USA.

Chao, A. 1984. Non-parametric estimation of the number ofclasses in a population. Scandinavian Journal of Statistics11:265–270.

Colwell, R. K., and J. A. Coddington. 1994. Estimating ter-restrial biodiversity through extrapolation. PhilosophicalTransactions of the Royal Society of London B, BiologicalSciences 345:101–118.

Colwell, R. K., and G. C. Hurtt. 1994. Nonbiological gra-dients in species richness and a spurious Rapoport effect.American Naturalist 144:570–595.

Dale, M. R. T. 1984. The contiguity of upslope and down-slope boundaries of species in a zoned community. Oikos42:92–96.

. 1986. Overlap and spacing of species’ ranges on anenvironmental gradient. Oikos 47:303–308.

. 1988. The spacing and intermingling of speciesboundaries on an environmental gradient. Oikos 53:351–356.

Duellman, W. E. 1979. The herpetofauna of the Andes: pat-terns of distribution, origin, differentiation, and presentcommunities. Pages 371–459 in W. E. Duellman, editor.The South American herpetofauna: its origin, evolution anddispersal. Monograph of the Museum of Natural History,the University of Kansas, Number 7. Lawrence, Kansas,USA.

Duellman, W. E., and E. R. Wild. 1993. Anuran amphibiansfrom the Cordillera de Huancabamba, northern Peru: sys-tematics, ecology and biogeography. Occasional Papers ofthe Museum of Natural History, the University of Kansas,Number 157. Lawrence, Kansas, USA.

Fauth, J. E., B. I. Crother, and J. B. Slowinski. 1989. Ele-vation patterns of species richness, evenness, and abun-dance of the Costa Rican leaf litter herpetofauna. Biotro-pica 21:178–185.

Gartshore, M. E. 1986. The status of the montane herpeto-fauna of the Cameroon Highlands. Pages 204–240 in S. N.Stuart, editor. Conservation of the Cameroon montane for-est. International Council for Bird Preservation, Cam-bridge, UK.

Gotelli, N. J., and G. R. Graves. 1996. Null models in ecol-ogy. Smithsonian Institution Press, Washington, D.C.,USA.

Graham, G. L. 1990. Bats versus birds: comparisons amongPeruvian volant vertebrate faunas along an elevational gra-dient. Journal of Biogeography 17:657–668.

Hamilton, A. 1992. History of forests and climate. Pages

17–25 in J. A. Sayer, C. S. Harcourt, and N. M. Collins,editors. The conservation atlas of tropical forests: Africa.Macmillan Publishers, Basingstoke, UK.

Heatwole, H. 1982. A review of structuring in herpetofaunalassemblages. Pages 1–19 in N. J. Scott, Jr., editor. Her-petological communities: a symposium of the Society forthe Study of Amphibians and Reptiles and the Herpetol-ogists’ League, August 1977. Wildlife Research Report 13.U.S. Department of the Interior, Fish and Wildlife Service,Washington, D.C., USA.

Heyer, W. R. 1967. A herpetofaunal study of an ecologicaltransect through the Cordillera de Tilaran, Costa Rica. Co-peia 1967:259–271.

Inger, R. F. 1980. Relative abundances of frogs and lizardsin forests of Southeast Asia. Biotropica 12:14–22.

Inger R. F., and R. K. Colwell. 1977. Organization of con-tiguous communities of amphibians and reptiles in Thai-land. Ecological Monographs 47:229–253.

Inger, R. F., and B. Greenberg. 1966. Ecological and com-petitive relations among three species of frogs (genusRana). Ecology 47:746–759.

Inger, R. F., and R. Stuebing. 1992. The montane amphibianfauna of northwestern Borneo. Malayan Nature Journal 46:41–51.

Janzen, D. 1973. Sweep samples of tropical foliage insects:effects of seasons, vegetation types, elevation, time of day,and insularity. Ecology 54:687–708.

Jenkins, M., and A. Hamilton. 1992. Biological diversity.Pages 26–32 in J. A. Sayer, C. S. Harcourt, and N. M.Collins, editors. The conservation atlas of tropical forests:Africa. Macmillan Publishers, Basingstoke, UK.

Lane, P. J. 1994. The vegetation of Mount Kupe, Cameroon.Report Number 8. Mount Kupe Forest Project, Yaounde,Cameroon.

Lawton, J. H., M. MacGarvin, and P. A. Heads. 1987. Effectsof altitude on the abundance and species richness of insectherbivores on bracken. Journal of Animal Ecology 56:147–160.

Legendre, P., S. Dallot, and L. Legendre. 1985. Successionof species within a community: chronological clustering,with applications to marine and freshwater zooplankton.American Naturalist 125:257–288.

Legendre, P., and A. Vaudor. 1991. The ‘‘R’’ package. Mul-tidimensional analysis, spatial analysis. Departement desciences biologiques, Universite de Montreal, Montreal,Quebec, Canada.

Maley, J. 1987. Fragmentation de la foret dense humide af-ricaine et extension des biotopes montagnards au Quarten-aire recent: nouvelles donnees polliniques et chronolo-giques. Implications paleoclimatiques et biogeographiques.Palaeoecology of Africa 18:307–334.

McCoy, E. D. 1990. The distribution of insects along ele-vational gradients. Oikos 58:313–322.

Olson, D. M. 1994. The distribution of leaf litter inverte-brates along a Neotropical altitudinal gradient. Journal ofTropical Ecology 10:129–150.

Patterson, B. D., P. L. Meserve, and B. K. Lang. 1989. Dis-tribution and abundance of small mammals along an ele-vational transect in temperate rainforests of Chile. Journalof Mammalogy 70:67–78.

Patterson, B. D., V. Pacheco, and S. Solari. 1996. Distri-bution of bats along an elevational gradient in the Andesof south-eastern Peru. Journal of Zoology (London) 240:637–658.

Pielou, E. C. 1977. The latitudinal spans of seaweed speciesand their patterns of overlap. Journal of Biogeography 4:299–311.

. 1978. Latitudinal overlap of seaweed species: evi-dence for quasi-sympatric speciation. Journal of Bioge-ography 5:227–238.

11

APPENDIXIncidence matrix of amphibians and reptiles recorded on the western slope of Mount Kupe, Cameroon, March–November

1994, as based on an equal sampling effort and subjected to the null model tests. The totals given in the last column includethe specimens recorded on the gradient outside sampling sessions and, thus, do not correspond to row totals. Column headers900–2000 m represent sampling zone contours.

Species

Elevation (m) of transect contour

900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 Total

AmphibaBufonidae

Bufo gracilipesBufo tuberosusNectophryne afraNectophryne batesiWerneria preussi mertensianaWolterstorffina parvipalmata

12

1 12

13 6 5 10 42 13 10 4 5 3

1512

10124

RanidaeConraua robustaPetropedetes cameronensisPetropedetes perretiPetropedetes newtoni

1047

1945

1943

1532

2

113

1

73

11

16473530

Petropedetes parkeriPhrynobatrachus cricogasterPhrynobatrachus werneriPhrynodon sp. 1†Phrynodon sp. 2†

25

11

1214

49

72

28

514

76

54

67

121

17

29

46

161419

145

1002319

322Hyperoliidae

Afrixalus lacteusAfrixalus laevisHyperolius acutirostrisLeptopelis brevirostrisLeptopelis calcaratusLeptopelis modestusLeptopelis rufus

66

4

31

2

1

1

8

2

8

9

213

1222

91512

RhacophoridaeAcanthixalus spinosusChiromantis rufescens

61

Pielou, E. C., and R. D. Routledge. 1976. Salt marsh veg-etation: latitudinal gradients in the zonation patterns. Oec-ologia 24:311–321.

Raaijmakers, J. G. W. 1987. Statistical analysis of the Mi-chaelis-Menten equation. Biometrics 43:793–803.

Rahbek, C. 1995. The elevational gradient of species rich-ness: a uniform pattern? Ecography 18:200–205.

. 1997. The relationship among area, elevation andregional species richness in neotropical birds. AmericanNaturalist 149:875–902.

Raxworthy, C. M., and R. A. Nussbaum. 1994. A rainforestsurvey of amphibians, reptiles and small mammals at Mon-tagne d’Ambre, Madagascar. Biological Conservation 69:65–74.

Rice, W. R. 1989. Analyzing tables of statistical tests. Evo-lution 43:223–225.

Scott, N. J., Jr. 1976. The abundance and diversity of theherpetofaunas of tropical forest litter. Biotropica 8:41–58.

Suchel, J. B. 1972. La repartition des pluies et les regimespluviometriques du Cameroun. Travaux et Documents deGeographie Tropicale, CEGET-CNRS (Centre d’Etude etde Geographie Tropicale-Centre National dela RechercheScientifique) 5:1–287.

Terborgh, J. 1971. Distribution on environmental gradients:theory and preliminary interpretation of distributional pat-terns in the avifauna of Cordillera Vilcabamba, Peru. Ecol-ogy 52:22–40.

. 1977. Bird species diversity on an Andean eleva-tional gradient. Ecology 58:1007–1019.

. 1985. The role of ecotones in the distribution ofAndean birds. Ecology 66:1237–1246.

Terborgh, J., and J. S. Weske. 1975. The role of competitionin the distribution of Andean birds. Ecology 56:562–576.

Thomas, D. W. 1986. Vegetation in the montane forest ofCameroon. Pages 20–27 in S. N. Stuart, editor. Conser-vation of Cameroon montane forests. International Councilfor Bird Preservation, Cambridge, UK.

Toft, C. A. 1980. Seasonal variation in populations of Pan-amanian litter frogs and their prey: a comparison of wetterand drier sites. Oecologia 47:34–38.

Tye, H. 1986. Geology and land forms in the highlands ofwestern Cameroon. Pages 15–17 in S. N. Stuart, editor.Conservation of Cameroon montane forests. InternationalCouncil for Bird Preservation, Cambridge, UK.

Underwood, A. J. 1978. The detection of non-random pat-terns of distribution of species along a gradient. Oecologia36:317–326.

Whittaker, R. H. 1967. Gradient analysis of vegetation. Bi-ological Reviews of the Cambridge Philosophical Society42:207–264.

Whittaker, R. H., and W. A. Niering. 1965. Vegetation ofthe Santa Catalina Mountains, Arizona: a gradient analysisof the south slope. Ecology 46:429–452.

Whittaker, R. H., and W. A. Niering. 1975. Vegetation ofthe Santa Catalina Mountains, Arizona. V. Biomass, pro-duction, and diversity along the elevational gradient. Ecol-ogy 56:771–790.

Wolda, H. 1987. Altitude, habitat and tropical insect diver-sity. Biological Journal of the Linnean Society 30:313–323.

Yu, H. T. 1994. Distribution and abundance of small mam-mals along a subtropical elevational gradient in cetral Tai-wan. Journal of Zoology (London) 234:577–600.

12

APPENDIX. Continued.

Species

Elevation (m) of transect contour

900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 Total

ArthroleptidaeArthroleptis adelphusArthroleptis variabilisArthroleptis ‘‘adolfifriderici’’‡

31134

2276

763

841 8

12 36 92 127 268 88 67 47

79377866

Arthroleptis sp. A§Arthroleptis sp. C§Astylosternus cf. montanusAstylosternus diadematusAstylosternus perreti

613235

3255

6

3591

16

2215

14

129

6

32

6

31

7 24 1 8 8176146

94

66Cardioglossa elegansCardioglossa gracilisCardioglossa venustaLeptodactylodon ornatusLeptodactylodon bicolorTrichobatrachus robustus

2086

1

3187

1

384

3

111

1

8702720

14

ReptiliaChamaeleonidae

Chamaeleo montiumChamaeleo pfefferiChamaeleo quadricornisRhampholeon spectrum

21

33

5

22

111

26

122

32

91

26

83

19

41832

82324

667

1013

11 1

554985

278Scincidae

Leptosiaphos rohdeiLeptosiaphos sp. A\

1 11 1 7 2 4 2

919

Leptosiaphos sp. B\Leptosiaphos sp. C\Panaspis chriswildiMabuya affinis 1

2 5 2 31 1 4 1 2 2 17

1311

LacertidaeAdolfus africanus 6

GeckonidaeCnemaspis koehleriCnemaspis spinicollisHemidactylus fasciatusHemidactylus echinus

11 2

1

1

3 3 3 1 1 13219

BoidaeCalabaria reinhardti 1 2

ColubridaeBoiga pulverulentaBothrolycus aterBuhoma depressicepsChamaelycus fasciatusDipsadoboa unicolorDipsadoboa sp.

2 11

12

1

1 2 1 2

13

18134

Mehelya guiraliRhamnophis aethiopissa

1 1 21

ViperidaeAtheris squamigerBitis gabonica 1

21

† Phrynodon sp. 1 and sp. 2 are distinguished by their mating calls (J.-L. Amiet, personal communication).‡ According to J.-L. Perret (personal communication), different from species described as adolfifriederici by Nieden from

Rwanda.§ Two small Schoutedenella-like taxa provisionally included in the genus Arthroleptis by J.-L. Amiet (personal commu-

nication).\ W. Bohme and A. Schmitz, personal communication.

13