Detection of macro-ecological patterns in SouthAmerican hummingbirds is affected by spatial scaleCarsten Rahbek1* and Gary R. Graves21Zoological Museum, University of Copenhagen, Universitetsparken 15, DK-2100 Copenhagen O, Denmark (crahbek@zmuc.ku.dk)2Department of Vertebrate Zoology, National Museum of Natural History, Smithsonian Institution,Washington DC 20560, USA(graves.gary@nmnh.si.edu)Scale is widely recognized as a fundamental conceptual problem in biology, but the question of whetherspecies-richness patterns vary with scale is often ignored in macro-ecological analyses, despite theincreasing application of such data in international conservation programmes. We tested for scaling e?ectsin species-richness gradients with spatially scaled data for 241 species of South American hummingbirds(Trochilidae). Analyses revealed that scale matters above and beyond the e?ect of quadrat area. Speciesrichness was positively correlated with latitude and topographical relief at ten di?erent spatial scalesspanning two orders of magnitude (ca. 12 300 to ca. 1225 000 km2). Surprisingly, when the in?uence oftopography was removed, the conditional variation in species richness explained by latitude fellprecipitously to insigni?cance at coarser spatial scales. The perception of macro-ecological pattern thusdepends directly upon the scale of analysis. Although our results suggest there is no single correct scalefor macro-ecological analyses, the averaging e?ect of quadrat sampling at coarser geographical scalesobscures the ?ne structure of species-richness gradients and localized richness peaks, decreasing thepower of statistical tests to discriminate the causal agents of regional richness gradients. Ideally, the scaleof analysis should be varied systematically to provide the optimal resolution of macro-ecological pattern.Keywords: hummingbirds; latitude; scaling; species-richness gradients; topography; Trochilidae
1. INTRODUCTIONPatterns of species richness along latitudinal (Rohde1992; Rosenzweig 1995), elevational (Lawton et al. 1987;McCoy 1990; Rahbek 1995, 1997) and longitudinal gradi-ents (Cotgreave & Harvey 1994; Blackburn & Gaston1996a) have become prime exemplars of large-scalepattern in macro-ecology (Brown 1995). The importanceof scale in the resolution of biodiversity patterns has beenwidely recognized for decades (Hutchinson 1953; Whit-taker 1977; Ricklefs 1987; Wiens et al. 1987; Wiens 1989;Orians & Wittenberger 1991; Cornell & Lawton 1992;Levin 1992; Schneider 1994; Andersen 1997; Taylor &Gaines 1999). In practice, however, most studies ofspecies-richness gradients have tacitly assumed thatpatterns and generating mechanisms were similar atarbitrarily de?ned scales of analysis. Only recently haveinvestigators tested this hypothesis at relatively smallecological scales (0.02 m2 to 36 km2) by systematicallyvarying the cell size of the sampling grid (Palmer &White 1994; B?hning-Gaese 1997; Ault & Johnson 1998;Peltonen et al. 1998; Angermeier & Winston 1998; Carroll& Pearson 1998; Karlson & Cornell 1998; Ohmann &Spies 1998). Attempts to extend multiscale analyses toentire continents or oceans have been largely thwartedby the paucity of spatial data of su?cient quality andresolution, as well as by methodological and statisticalobstacles. One notable exception was Lyons & Willig?s(1999) analysis of species-richness gradients of SouthAmerican bats and marsupials, based on nested quadratsof ?ve sizes ranging in area from 1000^25 000 km2. Thelargest of their quadrats, however, was more than an
order of magnitude smaller than the equal-area grid ofca. 611000 km2, which has provided the spatial templatefor more than 50 papers on macro-ecology and conserva-tion biology since 1994 (papers based on the data set ofBlackburn & Gaston (1996a,b) and those listed inWilliams (1999)). This immediately raises the question ofwhether macro-ecological patterns revealed using quad-rats of this size or other arbitrary coarse-scale spatialformats change signi?cantly with scale. More impor-tantly, can the coarse-scale analyses now in vogueadequately characterize species-richness gradients andelucidate their underlying causes?We addressed those questions by investigating thecorrelative relationships between the species richness ofSouth American hummingbirds (n ? 241 species) andgeospatial variables (latitude and longitude) and topo-graphical relief (hereafter topography) at ten spatialscales spanning two orders of magnitude (quadrat sizevarying from ca. 12 300 to ca. 1225 000 km2). Theseindependent variables have been widely employed assurrogates for abiotic and biotic factors that in?uencespecies-richness gradients (e.g. Currie 1991; Rohde 1992;Cotgreave & Harvey 1994; Currie 1991; Kerr & Packer1997). The strength of this approach lies in the straight-forward nature of compiling the number of species inlatitude^longitude quadrats (hereafter abbreviated aslat^long). We asked three hierarchical questions: (i) Isspecies richness correlated with latitude and longitude ?(ii) To what extent does topography in?uence spatialpatterns of species richness? (iii) Does the pattern ofcorrelation between species richness and latitude, longi-tude and topography vary according to quadrat size?Finally, we address the implications of scaling for macro-ecological analyses and conservation programmes.Proc. R. Soc. Lond. B (2000) 267, 2259^2265 2259 ? 2000 The Royal SocietyReceived 18 May 2000 Accepted 21 August 2000
doi 10.1098/rspb.2000.1277
*Author for correspondence.
2. METHODS(a) Choice of taxaHummingbirds constitute one of the most speciose and argu-ably the most spectacular evolutionary radiation of birds. Thisnectarivorous^ insectivorous clade (Bleiweiss et al. 1997) iseminently suitable for testing macro-ecological scaling hypothesesbecause the geographical ranges of species are relatively wellknown and the taxonomic inventory is believed to be nearlycomplete, as few new species have been discovered since 1950(Graves1993). Hummingbirds occur from Alaska (618 N) toTierradel Fuego (558 S), but reach their greatest species richness in theequatorial Andes (?gure 1). Seventy-six per cent of the 319 speciesrecognized in Sibley &Monroe (1990) occur in South America.(b) Distributional dataDistributional data were compiled from primary sources (i.e.museum specimens and documented sight records; museumsources listed in acknowledgements) for each of the 241hummingbird species that occur in South America at a resolu-tion of 18? 18 lat^long quadrats. Final maps for each speciesrepresent a conservative extrapolation of e`xtent of occurrence?based on con?rmed records and the spatial distribution ofpreferred habitat (for description of the methodology andsources see Fjelds? & Rahbek (1997, 1998)). We used theWORLDMAP computer program (v. 4.19.06, Williams 1998) toaccommodate and overlay the hummingbird distributional dataand to generate ?gure 1. Species richness was calculated for lat^long quadrats aligned at the equator and prime meridian at tenspatial scales (18? 18, 28?28, 38? 38, . . .108?108). Quadratcentroids were used as spatial coordinates.(c) TopographyWe used topographical relief (maximum minus minimumelevation recorded in each quadrat) as a surrogate for topo-graphical heterogeneity (Rahbek 1997). Elevational data,rounded to the nearest 100 m, were compiled from OperationalNavigation Charts (1:1000 000; published by the United StatesDefense Mapping Agency Aerospace Center, St Louis, MO,USA). These data are the most reliable estimates of elevationalheterogeneity currently available in the public domain for theentire South American continent. Previous studies have consid-ered elevational range as a surrogate for a broader measure ofhabitat heterogeneity (Currie 1991; Kerr & Packer 1997). Ouruse of topographical relief is conservative in that it tends tounderestimate the true topographical heterogeneity at coarserspatial scales. We have avoided using extrapolation models oftopographical relief because of the unacceptably high error asso-ciated with point estimation (Olsen & Bliss 1997).(d) AreaArea per se has an indisputable in?uence on species richnessthat in principle must be dealt with in any analysis of species-richness gradients (Connor & McCoy 1979; Palmer & White1994; Rosenzweig 1995; Rahbek 1995, 1997; Pastor et al. 1996;Lyons & Willig 1999). Within our region of analysis, the area of18? 18 lat^long quadrats decreases from ca. 12391km2 at theequator to ca. 7036 km2 at 558 S. However, only 6% of the 241species of hummingbird occur south of 358 S (quadrat area,ca. 10 273 km2), meaning that the analytical results of ouranalyses were relatively similar whether or not quadrat area wastaken into account. Nevertheless, because collinearity betweenquadrat area and latitude is signi?cant, the e?ect of area on
species richness cannot easily be standardized at each scale ofanalysis without simultaneously obscuring the e?ect of latitudeon species richness. Equal-area gridded maps circumvent thisproblem. The projection used to generate such maps, however,di?ers signi?cantly from those used to produce traditional lat^long maps. This makes transfer of coordinate data obtainedfrom specimen labels and published specimen records ontoequal-area overlays more di?cult than onto traditional lat^longmaps. Moreover, the transfer of published range maps (lat^long)to equal-area grids is even more problematic, particularly fromtiny distributional maps in ?eld guides and other secondaryliterature sources (e.g. the approach used by Blackburn &Gaston (1996a,b) to compile their geographical database onNewWorld birds). These problems are especially prevalent in theAndean arc region where dense aggregations of geographicalrange boundaries are commonplace. The use of equal-area mapsalso makes comparisons with previously published studies moredi?cult because of frame and scale shifts. For these reasons webelieve the potential trade-o?s, at the present time, argue forthe continued use of lat^long quadrats for biodiversity studies.The treatment of coastal quadrats is a far more seriousproblem in the analysis of species-richness gradients, whether ornot lat^long or equal-area quadrats are used. One method-ological option is to omit coastal quadrats from analyses(Rosenzweig 1995; Lyons & Willig 1999), but this procedureeliminates much of the important biological signal in SouthAmerica, where biologically rich mountain ranges parallel theCaribbean coast of Venezuela and Colombia, the Atlantic coastof south-eastern Brazil, and the entire Paci?c coastline. Infor-mation loss under such a data-censoring scheme is highly scale-dependent. For example, omission of coastal quadrats wouldresult in sample size reductions of 15% for 18? 18 quadrats,54% for 58? 58 quadrats, and 83% for 108? 108 quadrats.In this study, we classi?ed each 18? 18 lat^long quadrat aseither continental (contained some land area) or oceanic(contained no land area, i.e. did not intersect the continentalshoreline at a 1:1000 000 map scale). The land area withinlarger-scale quadrats was estimated by summing the areas of thenested c`ontinental? 18? 18 quadrats. We then applied theArrhenius power function (Arrhenius 1921), which posits thatwhen sampling quadrats are positioned in a nested fashion, area(A) a?ects richness (S) in an exponential fashion, S ? cAz, andwhere the parameter c is the ratio of species richness to Az, i.e.c ? S/Az (Rosenzweig 1995). Given that z is similar betweensamples, one can standardize species richness to the same areaby substituting the observed area and species richness into thefollowing equation:Ssample1/Azsample1 ? Ssample2/Azsample2, (1)Ssample2 ? (Ssample1/Azsample1)/Azsample2, (2)Ssample2 ? Ssample1 ? (Asample1/A2)z. (3)In practice we used the equationSadj ? Sobs ? (Amax/Aobs)z, (4)where Sadj was species richness adjusted by the maximum area(Amax) of quadrat size within a particular scale of analysis, Sobswas the observed species richness, and Aobs was the actual areaof a given quadrat.Pastor et al. (1996) and Rahbek (1997) calculated the actualspecies^area curve of each sample and then used the derived
2260 C. Rahbek and G. R. Graves Scalar variance in species- richness gradients
Proc. R. Soc. Lond. B (2000)
statistics to evaluate and remove the e?ects of area. Thisspecies^area approach, however, assumes that area is notcorrelated with any of the independent variables at any scale.Violation of the non-collinearity assumption can lead to error inhypothesis discrimination (Francis & Currie 1998). In this case,the high correlation between quadrat area and latitude, espe-cially at our smallest scale of analysis (18? 18, r2 ? 0.92),prevents the application of a scale-speci?c approach. To amelio-rate this problem, we pooled species-richness data over all tenscales of analysis (ca. 12300 to ca. 1225 000 km2), and tested forequal means in residuals (ANOVA), di?erences among sets ofmeans in residuals (Tukey pairwise comparison test), trends inresiduals (Pearson product-moment correlation), and homo-geneity of variances in residuals (Sche?e?^Box test). All tests
were non-signi?cant ( p 4 0.05). We then used z ? 0.23, derivedfrom the pooled data and logS ? zlogA+ logc, to standardizespecies richness within each scale of analysis independent oflatitude (correlation between area and latitude for the pooleddata as low as r25 0.01).(e) Statistical analysesWe regressed species richness of quadrats on independentvariables (latitude, longitude and topography) and on thelatitude? topography interaction at each spatial scale (table 1).This procedure was repeated with partial regression analysis tofactor out the in?uence of other independent variables (table 2).All independent variables were entered into a multiple regres-sion model (species richness ? constant + latitude+ topography
Scalar variance in species- richness gradients C. Rahbek and G. R. Graves 2261
Proc. R. Soc. Lond. B (2000)
Table 1. Spatial and topographical determinants of hummingbird species richness
quadrant latitude longitude topography latitude? topographyscale n size (km2)a F r2 F r2 F r2 F r218 grid 1689 12 308 1387.39 0.45***b 0.36 0.00 69.70 0.04*** 159.88 0.09***28 grid 457 49 225 277.09 0.38*** 4.96 0.01 50.8 0.10*** 29.71 0.06***38 grid 216 110 729 114.12 0.35*** 4.57 0.02 43.70 0.17*** 12.50 0.06*48 grid 129 196 784 69.44 0.35*** 4.68 0.04 23.15 0.15*** 8.35 0.0658 grid 90 307 338 51.59 0.37*** 2.35 0.03 15.79 0.15*** 17.79 0.17**68 grid 66 442 325 39.62 0.39*** 3.69 0.06 17.77 0.22** 4.97 0.0778 grid 49 601 674 26.48 0.36*** 3.64 0.07 7.51 0.14 7.77 0.1488 grid 40 785 268 22.81 0.38** 1.44 0.04 10.57 0.28* 3.35 0.0898 grid 35 993 019 19.73 0.37** 0.39 0.01 12.46 0.27* 2.34 0.07108 grid 29 1 224 797 17.33 0.39*** 0.63 0.02 3.00 0.10 4.94 0.16mean r2 ? ? ? 0.38c ? 0.03 ? 0.16 ? 0.10c.v. of r2 (%) ? ? ? 7.6 ? 75.4 ? 47.1 ? 45.3
aMaximum quadrat size by which species-richnessvalues were standardized at each spatial scale (see ? 2).b Probability that the observed F-value is greater than or equal to the simulated F-value (9999 iterations in which species richness wasrandomly chosen from the pool of available values at each spatial scale). p-values were adjusted for error rate per variable:*p 5 0.05/10 ? 0.005; **p 5 0.01/10 ? 0.001; ***p 5 0.001/10 ? 0.0001.cIndependent variable has signi?cant in?uence on species richness at all spatial scales.Table 2. Partial correlation analysis factoring out the in?uence of other independent variables in the model
quadrant latitude longitude topography latitude? topography model with allvariablesscale n size (km2)a F r2 F r2 F r2 F r2 F r218 grid 1689 12 308 472.50 0.22***b 0.02 0.00 527.22 0.24*** 314.41 0.16*** 616.46 0.60***28 grid 457 49 225 49.49 0.10*** 0.10 0.00 305.82 0.40*** 185.31 0.29*** 215.02 0.66***38 grid 216 110 729 8.66 0.04* 0.34 0.00 202.40 0.49*** 115.39 0.35*** 123.12 0.70***48 grid 129 196 784 3.07 0.02 0.12 0.00 151.86 0.55*** 93.55 0.42*** 89.28 0.74***58 grid 90 307 338 8.87 0.09* 0.03 0.00 110.54 0.56*** 69.05 0.44*** 66.55 0.76***68 grid 66 442 325 0.55 0.01 0.09 0.00 89.10 0.59*** 61.59 0.49*** 62.34 0.81***78 grid 49 601 674 1.12 0.02 0.01 0.00 62.54 0.57*** 49.76 0.51*** 40.57 0.78***88 grid 40 785 268 0.02 0.00 0.05 0.00 85.08 0.69*** 46.80 0.55*** 42.37 0.83***98 grid 35 993 019 0.23 0.00 1.67 0.05 85.60 0.72*** 46.07 0.58*** 42.85 0.85***108 grid 29 1 224 797 0.60 0.02 0.15 0.01 73.13 0.73*** 51.81 0.66*** 35.55 0.86***mean r2 ? ? 0.05 ? 0.01 ? ? 0.55c ? 0.44c ? 0.76cc.v. of r2 (%) ? ? 131.9 ? 262.9 ? ? 27.2 ? 33.2 ? 11.2
a Maximum quadrat size by which species-richnessvalues were standardized at each spatial scale (see ? 2).b Probability that the observed F-value is greater than or equal to the simulated F-value (9999 iterations in which species richness wasrandomly chosen from the pool of available values at each spatial scale). p-values were adjusted for error rate per variable:*p 5 0.05/10 ? 0.005; **p 5 0.01/10 ? 0.001; ***p 5 0.001/10 ? 0.0001.cIndependent variable has signi?cant in?uence on species richness at all spatial scales.
+ latitude? topography) to estimate their power to predictspecies richness at di?erent spatial scales.Distributional assumptions of parametric regression tests arerarely met by macro-ecological data sets, and p -values reportedin such studies are often unreliable. To avoid such problems, wereport regression coe?cients obtained by randomly permutingthe dependent variable 9999 times (tables 1 and 2) (Legendreet al. 1998; see also Legendre et al. 1994). Spatial autocorrelation,an inherent quality of biogeographical data (e.g. geographicalranges of 4 99% of South American hummingbird speciesoverlap several to many adjoining 18? 18 quadrats), increases theerror in estimating the degrees of freedom and multiplies therisk of making type II errors. Although the relative importanceof independent variables can be estimated by comparing thederived r2-values, we emphasized non-signi?cant p-valuesbecause the e?ects of spatial autocorrelation cannot be removedby simple permutation or randomization methods (Manly 1997).
3. RESULTSAnalyses revealed that the choice of quadrat size signif-icantly in?uences the correlation between species richnessand latitude, longitude, and topography (tables 1 and 2).Variation of r2-values at di?erent grid dimensions indi-cates that scale matters above and beyond the e?ect ofquadrat area. Species richness was negatively correlatedwith latitude (r 2 ? 0.35^0.45; coe?cient of variation (c.v.)of r 2 ? 7.6%) at spatial scales spanning two orders ofmagnitude (table 1). Longitude was an insigni?cantpredictor of species richness regardless of the scale ofresolution (r25 0.07; c.v. ? 75.0%). The in?uence of topo-graphy on species richness ?uctuated markedly with scale(r2 ? 0.04^0.28; c.v. ? 47.1%).
Partial correlation analysis revealed a complex rela-tionship between species richness, latitude, and topo-graphy (table 2 and ?gure 2). When the e?ects oftopography were controlled, the predictive power oflatitude decreased to insigni?cant levels (r 25 0.05) whenquadrat area exceeded ca. 110 000 km2 (table 2), perhapsdue to collinearity between latitude and latitude? topography. Intriguingly, given the other variables inthe model, both topography (r 2 ? 0.24^0.73) and the lati-tude? topography interaction (r 2 ? 0.16^0.66) explaineda signi?cant proportion of the conditional variation inspecies richness regardless of scale. A simple multipleregression model including latitude, topography and thelatitude ? topography interaction explained from 60 to86% of the regional variability in species richness forquadrat areas ranging from ca. 12 300 to ca. 1225 000 km2(table 2).Although topography and latitude were generallyuncorrelated (r 25 0.03; c.v. ?121.8%), hummingbirdspecies richness appeared to be strongly associated withthe interaction between topography and latitude (table 2).This in?uence was especially pronounced at low latitudes(118N to 208 S) in the Andean region, where high eleva-tions, rugged topography, and orographical precipitationpatterns have resulted in perhaps the most complicatedmosaic of distinctive terrestrial habitats on Earth, eachsupporting a characteristic hummingbird fauna. At theequator, hummingbird species are distributed from sealevel to the snowline (0^5000 m). The elevational ampli-tude of the habitable zone decreases monotonically withlatitude to Tierra del Fuego (558 S), where the singlespecies of hummingbird is restricted to coastal habitatsnear sea level.
2262 C. Rahbek and G. R. Graves Scalar variance in species- richness gradients
Proc. R. Soc. Lond. B (2000)
(a) (b)
70? W 70? W 70? W
70? W 70? W
50? W
50? W
40? S
40? S
40? S
20? S 20? S
20? S 20? S40? S
0? 
0? 0? 
0? 50? W
50? W 50? W20? S
40? S
0? 
80 species
140 species
604020
(i) (ii)
(iii) (iv)
1057035
140 species1057035
140 species1057035
140 species1057035
Figure 1. Spatial variation in species richness of South American hummingbirds (Trochilidae): (a) compiled at 18? 18 scale;(b) compiled at (i) 18? 18, (ii) 38? 38, (iii) 58? 58, and (iv) 108? 108 scales (di?erent colour scales depicted in (a) and (b)).Note the excessive loss of information and the spurious extrapolation of high species densities in species-poor localities at coarserspatial scales. Grey areas illustrate quadrats supporting zero species.
4. DISCUSSIONThe cause of latitudinal gradients in species richness isstill hotly debated (Rosenzweig & Sandlin 1997; Rohde1998). Hypotheses receiving the most attention followthree major themes: (i) energy-related variables corre-lated with latitude, e.g. available ambient energy,primary productivity, potential evapotranspiration,seasonality, solar radiation, temperature (Currie 1991;Rohde 1992; Wright et al. 1993; Francis & Currie 1998);(ii) area and the large- scale steady state betweenallopatric speciation and extinction within latitudinalbands exhibiting approximately homologous tempera-ture (Rosenzweig 1995; see also Terborgh 1973); and(iii) geometric constraints on geographical range size andplacement (Colwell & Hurtt 1994; Rahbek 1997;Willig &Lyons 1998; Lees et al. 1999; Colwell & Lees 2000).Bearing these proposals in mind, our ?ndings are consis-tent with the hypothesis that the latitudinal speciesgradient is determined synergistically by a combination ofvariables correlated with both latitude and topography.Di?erences in the relationship between hummingbirdspecies richness and latitude at di?erent quadrat sizes(tables 1 and 2) indicate that scale may not be easilyaccounted for by using species^area relationships asothers have claimed (Palmer & White 1994; Pastor et al.1996). Still, our results are robust and o?er substantialsupport for the hypothesis that patterns of species richnessas well as generating mechanisms are unlikely to be scale-invariant. The Andean region is the global centre of avianspecies richness, a pattern that becomes even morepronounced when the e?ects of area are accounted for(Rahbek 1997). More than half of all hummingbirdspecies occur here and many endemic taxa are restrictedto narrow elevational zones (Graves 1985, 1988). Thein?uence of topography on species richness diminishesrapidly south of 208 S latitude as habitat diversity andtimberline decrease. Though the in?uence of area on
species richness is one of macro-ecology?s few unquestion-able laws, biome area per se has a minor in?uence onhummingbird species richness. For example, theAmazonian tropical moist forest (ca. 5 million km2)(Dinerstein et al. 1995) constitutes the largest biome inSouth America. Yet, 18? 18 quadrats in central Amazoniasupport only 16^25 species of hummingbirds, whereasspecies densities of quadrats straddling the eastern versantof the Andes, at equivalent latitudes, exceed 60 species(?gure 1). In essence, hummingbirds exemplify the emer-gent biotic pattern in the Neotropics, in which speciationand ? -diversity appear to be facilitated by topographicalvariation (Graves 1985), narrow homothermous eleva-tional bands ( Janzen 1967; Graves 1988; Rahbek 1997)and area (Rosenzweig 1995; Rahbek 1997).(a) Scale in macro-ecology and biodiversityconservationBecause biodiversity data compiled at macro-ecologicalscales are increasingly used as the empirical basis ofglobal conservation programmes (e.g. Dinerstein et al.1995; Statters?eld et al. 1998), an assessment of the scalingbias in empirical and theoretical analyses is urgentlyneeded. Many macro-ecological patterns are robustenough to be (re)discovered at coarse scales of resolutioneven when distributional data are transcribed from crudemaps published in secondary literature sources. Althoughour results collectively suggest that there is no singlecorrect macro-ecological scale for the investigation ofspecies-richness gradients, ?ner geographical scales aregenerally preferred. Our data suggest that the speciesrichness of most tropical birds and the relative impor-tance of factors responsible for richness gradients cannotbe adequately characterized at the coarser scalescommonly used (e.g. Cotgreave & Harvey 1994; Eggletonet al. 1994; Blackburn & Gaston 1996a,b; Blackburn et al.1998; Chown et al. 1998). For example, geographicalranges of South American hummingbirds are relativelysmall, averaging only 2.1 times the area of the ca.611000 km2 quadrats employed in the aforementionedanalyses. More than 68% of species have ranges smallerthan a single such quadrat.The averaging e?ect of quadrat sampling at coarsemacro-ecological scales obscures the ?ne structure ofspecies gradients and localized richness peaks (?gure 1).Coarse- scale projections lead to predictions of extraordi-narily high species richness in species-poor localities morethan 500 km from true richness peaks. Such spuriousextrapolations increase the risk of in?uential statisticaloutliers, and, most importantly, decrease the statisticalpower necessary to identify the causal agents of regionalspecies-richness gradients (tables 1 and 2). Needless to say,coarse-scale maps of species richness are ine?ectual forpinpointing areas of high endemism and inadequate forcomplementarity analyses for conservation purposes.As a ?nal note, we agree that time for cataloguing andmapping the Earth?s biota is running out. Conservationprogrammes must rely, in large part, on macro-ecologicalanalyses to identify and prioritize biologically importantregions for protection (Dinerstein et al. 1995; Fjelds? &Rahbek 1997; Statters?eld et al. 1998). Nevertheless, themode and quality of data collection (Gotelli & Graves1996) and scale of analysis (e.g. Whittaker 1977; Wiens
Scalar variance in species- richness gradients C. Rahbek and G. R. Graves 2263
Proc. R. Soc. Lond. B (2000)
0
20
40
60
80
100
proport
ion of v
ariance
  
explain
ed (%)
1? 3? 5? 10?
latitudetopographylatitude ?  topography
size of quadratsFigure 2. Conditional variation in hummingbird(Trochilidae) species richness explained by latitude,topography and the latitude? topography interaction.r2-values are derived from partial correlation analyses(table 2).
1989; Cornell & Lawton 1992; Levin 1992; Angermeier &Winston 1998; Ohmann & Spies 1998; Lyons & Willig1999) have a direct e?ect on the value and relevance ofresults. If macro-ecology is to provide the means toameliorate and minimize the current biodiversity crisis, athorough and scienti?cally based understanding ofscaling e?ects must be obtained. We urge caution ingeneralizing from macro-ecological studies conducted atcoarse spatial scales. Ideally, scale of analysis should bevaried systematically to provide a better resolution ofpattern and of the interrelationship among possible causalfactors.We thank A. Balmford, W. J. Boecklen, R. K. Colwell,D. Currie, N. J. Gotelli, J. H. Lawton, S. Pimm, K. Rohde, andthree anonymous reviewers for comments on the manuscript.Thanks to R. K. Colwell, S. D. Gaines, D. C. Lees, S. K. Lyons,P. H. Taylor and M. R. Willig for providing manuscripts inpress. P. Williams kindly provided the WORLDMAP softwareused to manage the distributional data. Primary trochiliformdistributional data were derived from the collections of Academyof Natural Sciences (Philadelphia), American Museum ofNatural History (New York), Carnegie Museum of NaturalHistory (Pittsburgh), Coleccio?n Ornitolo?gica Phelps (Caracas),Delaware Museum of Natural History, Field Museum of NaturalHistory (Chicago), L?Institute Royal des Sciences Naturelles(Bruxelles), Louisiana State University Museum of NaturalSciences, Moore Laboratory of Zoology (Los Angeles), MuseoArgentino de Ciencias Naturales (Buenos Aires), Museo deHistoria Natural `Javier Prado? de la UNMSM (Lima), Museode Historia Natural Universidad de Cauca (Popaya? n), MuseoEcuatoriano de Ciencias Naturales (Quito), Museo Nacional deCiencias Naturales (Bogota? ), Museo Nacional de HistoriaNatural (La Paz), Museo Nacional de Historia Natural(Santiago), Museu de Zoologia da Universidaded de Sa? o Paulo,Museu Nacional (Rio de Janeiro), Museu Paraense Em|?lioGoeldi (Bele?m), Museum Alexander Humboldt (Berlin),Museum Alexander K?enig (Bonn), Museum of ComparativeZoology, Harvard University, Museum of Natural History ofLos Angeles County, Muse?um d?Historie Naturelle (Neucha? tel),Muse?um National d?Histoire Naturelle (Paris), NationalMuseum of Natural History (Washington, DC), Natural HistoryMuseum of Gothenburgh, Rijksmuseum van NatuurlijkeHistorie (Leiden), Royal Ontario Museum (Toronto), SwedishMuseum of Natural History (Stockholm), The Natural HistoryMuseum (London and Tring), Western Foundation of VertebrateZoology (Los Angeles), Zoological Museum, University ofCopenhagen. Support for G.R.G. to travel to Copenhagen andC.R. to travel to Washington was given by the SmithsonianResearch Opportunities Fund and the Danish Natural ScienceResearch Council (grant no. 11-0390), respectively.REFERENCESAndersen, A. N. 1997 Using ants as bioindicators: multiscaleissues in ant community ecology. Conserv. Ecol. 1, 8.Angermeier, P. L. & Winston, M. R. 1998 Local vs regionalin?uences on local diversity in stream ?sh communities ofVirginia. Ecology 79, 911^927.Arrhenius, O. 1921 Species and area. J. Ecol. 9, 95^99.Ault, T. R. & Johnson, C. R. 1998 Spatial variation in ?shspecies richness on coral reefs: habitat fragmentation andstochastic structuring processes. Oikos 82, 354^364.Blackburn, T. M. & Gaston, K. J. 1996a Spatial patterns in thegeographic range sizes of bird species in the NewWorld. Phil.Trans. R. Soc. Lond. B 351, 897^912.Blackburn, T. M. & Gaston, K. J. 1996b Spatial patterns in thespecies richness of birds in the New World. Ecography 19,369^376.
Blackburn, T. M., Gaston, K. J. & Lawton, J. H. 1998 Patternsin the geographical ranges of the world?s woodpeckers. Ibis140, 626^638.Bleiweiss, R., Kirsch, J. A. W. & Matheus, J. C. 1997 DNAhybridization evidence for the principal lineages of humming-birds (Aves: Trochilidae). Mol. Biol. Evol. 14, 325^343.B?hning-Gaese, K. 1997 Determinants of avian species richnessat di?erent spatial scales. J. Biogeogr. 24, 49^60.Brown, J. H. 1995 Macroecology. University of Chicago Press.Carroll, S. S. & Pearson, D. L. 1998 The e?ects of scale andsampling size on the accuracy of spatial predictions of tigerbeetle (Cicindelidae) species richness. Ecography 21, 401^414.Chown, S. L., Gaston, K. J. & Williams, P. H. 1998 Globalpatterns in species richness of pelagic seabirds?theProcellariiformes. Ecography 21, 342^350.Colwell, R. K. & Hurtt, G. C. 1994 Nonbiological gradients inspecies richness and a spurious Rapoport e?ect. Am. Nat. 144,570^595.Colwell, R. K. & Lees, D. C. 2000 The mid-domain e?ect:geometric constraints on the geography of species richness.Trends Ecol. Evol. 15, 70^76.Connor, E. F. & McCoy, E. D. 1979 The statistics and biology ofthe species^area relationship. Am. Nat. 113, 791^833.Cornell, H. V. & Lawton, J. H. 1992 Species interactions, localand regional processes, and limits to the richness of ecologicalcommunities: a theoretical perspective. J. Anim. Ecol. 61, 1^12.Cotgreave, P. & Harvey, P. 1994 Associations among biogeo-graphy, phylogeny and bird species diversity. Biodiv. Lett. 2,46^55.Currie, D. J. 1991 Energy and large-scale patterns of animal-and plant-species richness. Am. Nat. 137, 27^49.Dinerstein, E., Olson, D. M., Graham, D. J., Webster, A. L.,Pimm, S. A., Bookbinder, M. P. & Ledec, G. 1995 A conserva-tion assessment of the terrestrial ecoregions of Latin America and theCaribbean. Washington, DC: The World Bank in associationwithTheWorldWildlife Fund.Eggleton, P., Williams, P. H. & Gaston, K. J. 1994 Explainingglobal termite diversity: productivity or history? Biodiv.Conserv. 3, 318^330.Fjelds?, J. & Rahbek, C. 1997 Species richness and endemism inSouth American birds: implications for the design of networksof nature reserves. InTropical forest remnants: ecology, managementand conservation of fragmented communities (ed. W. F. Laurance &R. O. Bierregaard), pp. 466^482. University of Chicago Press.Fjelds?, J. & Rahbek, C. 1998 Continent-wide conservationpriorities and diversi?cation processes. In Conservation in achanging world (ed. G. M. Mace, A. Balmford & J. R.Ginsberg), pp. 139^160. Cambridge University Press.Francis, A. P. & Currie, D. J. 1998 Global pattern of tree speciesrichness in moist forests: another look. Oikos 81, 598^602.Gotelli, N. J. & Graves, G. R. 1996 Null models in ecology.Washington, DC: Smithsonian Institution Press.Graves, G. R. 1985 Elevational correlates of speciation andintraspeci?c geographic variation in plumage in Andeanforest birds. Auk 102, 556^579.Graves, G. R. 1988 Linearity of geographic range and itspossible e?ect on the population structure of Andean birds.Auk 105, 47^52.Graves, G. R. 1993 Relic of a lost world: a new species ofsunangel (Trochilidae: Heliangelus) from `Bogota? ?. Auk 110,1^8.Hutchinson, G. E. 1953 The concept of pattern in ecology. Proc.Natl Acad. Sci. USA 105, 1^12.Janzen, D. H. 1967 Why mountain passes are higher in thetropics. Am. Nat. 101, 233^249.Karlson, R. H. & Cornell, H.V. 1998 Scale-dependent variationin local vs regional e?ects on coral species richness. Ecol.Monogr. 68, 259^274.
2264 C. Rahbek and G. R. Graves Scalar variance in species- richness gradients
Proc. R. Soc. Lond. B (2000)
Kerr, J. T. & Packer, L. 1997 Habitat heterogeneity as a determi-nant of mammal species richness in high-energy regions.Nature 385, 252^254.Lawton, J. H., MacGarvin, M. & Heads, P. A. 1987 E?ects ofaltitude on the abundance and species richness of insectherbivores on bracken. J. Anim. Ecol. 56, 147^160.Lees, D. C., Kremen, C. & Andriamampianina, L. 1999 A nullmodel for species richness gradients: bounded range overlapof butter?ies and other rainforest endemics in Madagascar.Biol. J. Linn. Soc. Lond. 67, 529^584.Legendre, P., Lapointe, F.-J. & Casgrain, P. 1994 Modellingbrain evolution from behavior: a permutational regressionapproach. Evolution 48, 1487^1499.Legendre, P., Lapointe, F.-J. & Casgrain, P. 1998 Permut! v. 3.4alpha: multiple regression over distance, ultrametric andadditive matrices with permutation test. Available at http://alize.ERE.UMontreal.CA:80/?casgrain/en/labo/permute/index.html.Levin, S. A. 1992 The problem of pattern and scale in ecology.Ecology 73, 1943^1967.Lyons, S. K. & Willig, M. R. 1999 A hemispheric assessment ofscale dependence in latitudinal gradients of species richness.Ecology 80, 2483^2491.McCoy, E. D. 1990 The distribution of insects along elevationalgradients. Oikos 58, 313^322.Manly, B. F. J. 1997 Randomization, bootstrap and Monte Carlomethods in biology. London: Chapman & Hall.Ohmann, J. L. & Spies, T. A. 1998 Regional gradient analysisand spatial pattern of woody plant communities of Oregonforests. Ecol. Monogr. 68, 151^182.Olsen, L. M. & Bliss, N. B. 1997 Development of a 30-arc-second digital elevational model of South America. In ESRIuser conference proceedings. Redlands, CA: EnvironmentalSystems Research Institute, Inc., CD-ROM. Available athttp://edcdaac.usgs.gov/gtopo30/papers/olsen.html.Orians, G. H. & Wittenberger, J. F. 1991 Spatial and temporalscales in habitat selection. Am. Nat. 137, S29^S49.Palmer, M. W. & White, P. S. 1994 Scale dependence and thespecies^area relationship. Am. Nat. 144, 717^740.Pastor, J., Downing, A. & Erickson, H. E. 1996 Species^areacurves and diversity^productivity relationships in beavermeadows of Voyageurs National Park, Minnesota, USA. Oikos77, 399^406.Peltonen, M., Heli?vaara, K., V?is?nen, R. & Keronen, J. 1998Bark beetle diversity at di?erent spatial scales. Ecography 21,510^517.Rahbek, C. 1995 The elevational gradient of species richness: auniform pattern? Ecography 18, 200^205.Rahbek, C. 1997 The relationship among area, elevation, andregional species richness in Neotropical birds. Am. Nat. 149,875^902.
Ricklefs, R. E. 1987 Community diversity: relative roles of localand regional processes. Science 235, 167^171.Rohde, K. 1992 Latitudinal gradients in species diversity: thesearch for the primary cause. Oikos 65, 514^527.Rohde, K. 1998 Latitudinal gradients in species diversity: areamatters, but how much. Oikos 82, 184^190.Rosenzweig, M. L. 1995 Species diversity in space and time. NewYork: Cambridge University Press.Rosenzweig, M. L. & Sandlin, E. A. 1997 Species diversity andlatitudes: listening to area?s signal. Oikos 80, 172^176.Schneider, D. C. 1994 Quantitative ecology: spatial and temporalscaling. San Diego, CA: Academic Press.Sibley, C. G. & Monroe Jr, B. L. 1990 Distribution and taxonomy ofbirds of the world.Yale University Press.Statters?eld, A. J., Crosby, M. J., Long, A. J. & Wege, D. C.1998 Endemic bird areas of the world. Priorities for biodiversityconservation. Cambridge, UK: BirdLife International.Taylor, P. H. & Gaines, S. D. 1999 Can Rapoport?s rule berescued? Modelling causes of the latitudinal gradient ofspecies richness. Ecology 80, 2474^2482.Terborgh, J. 1973 On the notion of favorableness in plantecology. Am. Nat. 107, 481^501.Whittaker, R. H. 1977 Evolution of species diversity in landcommunities. Evol. Biol. 10, 1^67.Wiens, J. A. 1989 Spatial scaling in ecology. Funct. Ecol. 3,385^397.Wiens, J. A., Rotenberry, J. T. & Van Horne, B. 1987 Habitatoccupancy patterns of North American shrubsteppe birds: thee?ect of spatial scale. Oikos 48, 132^147.Williams, P. H. 1998 WORLDMAP iv WINDOWS: software andhelp document 4.1. London: The Natural History Museum,privately distributed.Williams, P. H. 1999 Biodiversity?measuring the variety ofnature and selecting priority for conservation: publica-tions. Available at http://www.nhm.ac.uk/science/projects/worldmap/refs/Refs. html.Willig, M. R. & Lyons, S. K. 1998 An analytical model of lati-tude gradients of species richness with an empirical test formarsupials and bats in the NewWorld. Oikos 81, 93^98.Wright, D. H., Currie, D. J. & Maurer, B. A. 1993 Energysupply and patterns of species richness on local and regionalscales. In Species diversity in ecological communities: historical andgeographical perspectives (ed. R. E. Ricklefs & D. Schluter),pp. 66^74. University of Chicago Press.
As this paper exceeds the maximum length normally permitted,the authors have agreed to contribute to production costs.
Scalar variance in species- richness gradients C. Rahbek and G. R. Graves 2265
Proc. R. Soc. Lond. B (2000)