Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Environmental correlates of mammal species richness in South America: effects of spatial structure, taxonomy and geographic range

Environmental correlates of mammal species richness in South America: effects of spatial... Recent years have brought some consensus regarding the primary roles of hypotheses based on climate and energy (e.g. species‐energy hypothesis: Wright 1983 , water‐energy dynamic theory: O'Brien 1993, 1998 , O'Brien et al. 2000 ) and historical contingency (i.e., glaciation effects, differences in dispersal and speciation rates) in explaining much of the variance in species richness from the equator to the poles ( Whittaker et al. 2001 , see also Gaston 2000 , Kerr 2001 , Hawkins et al. 2003 ). The area of biomes ( Rosenzweig 1992, 1995 , Blackburn and Gaston 1997 ), topographic heterogeneity ( Simpson 1964 , Kerr and Packer 1997 , Rahbek and Graves 2001 , Jetz and Rahbek 2002 ), levels of precipitation ( Rahbek and Graves 2001 , Hawkins et al. 2003 ), and several measures of the availability of energy in local environments, such as mean and maximum temperature, net primary productivity, actual and potential evapotransporation, and solar radiation received per unit area (see e.g. Currie 1991 , Blackburn and Gaston 1996a , Kerr and Packer 1997 , Kerr and Currie 1999 , Currie et al. 1999 , Van Rensburg et al. 2002 , Hawkins et al. 2003 ) appear to be the best candidates to explain latitudinal variation in species diversity in both hemispheres (see also Pianka 1966 , Begon et al. 1986 , Rohde 1992 , Ricklefs and Schluter 1993 , Brown 1995 , Rosenzweig 1995 , Currie et al. 1999 , Gaston 2000 , Willig 2000 , Gaston and Blackburn 2000 , Kerr 2001 , Whittaker et al. 2001 , and chapters in the “species richness” section in Blackburn and Gaston 2003 for reviews). Despite the obvious importance of such conceptual progress, some issues still await comprehensive understanding. One issue is related to the kind of spatial heterogeneity that is inherent to ecological systems. Large‐scale physical processes occurring over the earth surface, history and ecological interactions among species can lead the environmental variables, the species richness and their interactions to have complex spatial structures; as a consequence, environmental variables and species richness are very often spatially auto‐correlated, with this implying that two sites located near one another are unlikely to be independent from each other (e.g. Legendre 1993 , Boone and Krohn 2000 , Lichstein et al. 2002 , Van Rensburg et al. 2002 , Diniz‐Filho et al. 2003 ). This complicates standard statistical testing of hypotheses because it can create false positive– or negative– results (see e.g. Legendre 1993 , Diniz‐Filho et al. 2003 ). Moreover, multi‐scale relationships between species richness and environmental variables are increasingly well recognized to affect gradients of species diversity, which complicate further ecological interpretation (e.g. Lyons and Willig 1999 , Rahbek and Graves 2001 , Arita and Rodríguez 2002 , Blackburn and Gaston 2002 , Jetz and Rahbek 2002 , Diniz‐Filho et al. 2003 ). On the other hand, dominant factors associated with richness may shift with latitude at continental and global scales (e.g. Kerr and Packer 1997 , Rahbek and Graves 2001 , Hawkins et al. 2003 ). However, they may also differentially affect species with different body sizes, belonging to different trophic or spatial guilds (e.g. Andrews and O'Brien 2000 ), having different dispersal capacities ( Kerr and Currie 1999 ), geographic range sizes (e.g. Jetz and Rahbek 2002 ) or evolutionary histories (e.g. Lathman and Ricklefs 1993 ). As a consequence, complex latitudinal diversity gradients may emerge from the effect of processes interrelated at different scales of time and space. In the present paper, we address some of these issues using South American mammals as a model system. We incorporate the effects of spatial structuring of ecological variables into tests of three non‐mutually exclusive hypotheses which are most often proposed to account for spatial patterns of variation in mammal species richness: species‐energy, environmental stability and habitat‐heterogeneity hypotheses. Note, however, that we will not address the possible role of stochastic factors within bounded domains in affecting gradients in mammal species richness in South America. Although the “mid‐domain effect” (e.g. Colwell and Hurtt 1994 , Willig and Lyons 1998 , Colwell and Lees 2000 ) has been considered a credible hypothesis to explain the latitudinal diversity gradients during the past decade, recent work suggests that the model is flawed in its assumptions and explains very little variation in species richness at macro‐scale (see Hawkins and Diniz‐Filho 2002 , Zapata et al. 2003 ). A critical assessment of mid‐domain models on the species richness patterns of south american mammals is out of the scope of the present analysis (see however: Willig and Lyons 1998 , Willig 2000 ). In the present analysis, we conduct tests separately for different taxonomic groups of mammals (i.e. marsupials, edentates, primates, rodents, artiodactyls, carnivores and bats) to explore different ways whereby two of these hypotheses (energy and heterogeneity) may interact to explain variation in mammal species richness at the continental scale. We also partition the total number of mammal species into range size quartiles to give a more complete picture of species’ distributions. Species with small geographic ranges may be more likely limited by environmental factors that vary on a local or regional scale (e.g. topography); in contrast, widespread species should be less affected by such regional factors (e.g. Brown and Maurer 1989 , Brown 1995 , Jetz and Rahbek 2002 ). Hence, in the present paper, we analyse whether determinants of overall species richness patterns are also representative for groups of species with different distributions (e.g. for narrow‐ranging and widespread species, as in Jetz and Rahbek 2002 ). We also show that differences in range size are not independent of taxonomy in this species assemblage. Hypotheses The species‐energy hypothesis proposes that richness is limited by the total or average amount of energy entering into an ecosystem. It predicts that high‐energy availability promotes the persistence of high species richness ( Wright 1983 , see also Brown 1988 , Brown and Lomolino 1998 , Currie 1991 , Gaston and Blackburn 2000 ). Here, we follow Hawkins et al. (2003) and distinguish between two versions of the energy‐hypothesis (i.e., productivity vs ambient‐energy versions): The productivity hypothesis ( Wright 1983 , see also Hawkins et al. 2003 ) states that the level of resource production in an ecosystem limits animal species richness. In this paper, we test this hypothesis using measures of energy closely related to the level of primary production in a region: annual evapotranspiration (AET) and annually Integrated Normalized Difference Vegetation Index (INDVI). AET is the amount of water that actually evaporates or is transpired from an area, depending on the joint availability of energy and water (see e.g. Currie and Paquin 1987 , Currie 1991 , Andrews and O'Brien 2000 ). INDVI provides an integrated index of ecosystem function through its strong correlation with aboveground net primary productivity and absorbed photosynthetically active radiation ( Kerr and Ostrovsky 2003 ). INDVI shows a non‐linear relationship with the leaf‐area index that defines canopy structure ( Waring and Running 1998 ), however, it remains the most commonly used and most intensively studied vegetation index (see Kerr and Ostrovsky 2003 for discussion). Both AET and INDVI can be considered indicators of the availability of energy (in chemical form) for primary consumers (see Whittaker et al. 2001 ). We predict a general positive relationship between AET, INDVI and mammal species richness at continental scale. The ambient‐energy hypothesis (see Turner et al. 1996 , Hawkins et al. 2003 and other references therein) proposed that the animal species richness of a region is directly controlled by the total or average energy available. Given that endotherms usually maintain themselves at a higher temperature than that of the environment, the lower the ambient temperature the more these animals must spend to maintain body temperatures, and consequently the less energy they devote to growth and reproduction. All other things being equal, higher temperatures will promote faster growth of individuals and populations; this greater biomass will, in turn, promote greater species richness (see Brown 1988 , Blackburn and Gaston 1996b , Turner et al. 1996 , Hawkins et al. 2003 for more detailed discussion). In this paper, we use potential evapotranspiration (PET) and minimum temperature of the coldest month (TMIN) to test this hypothesis. PET is the amount of water that evaporates from a saturated surface, depending mainly on the amount of energy available to evaporate water and, to a lesser degree, on the relative humidity ( Currie 1991 , see also Andrews and O'Brien 2000 ). TMIN is a partial index of the environmental energy regime that describes the minimum degree of heat at each particular site (e.g. Andrews and O'Brien 2000 ). It also relates the ambient‐energy hypothesis to the original climatic freezing‐hypothesis of von Humboldt ( Hawkins 2001 , Hawkins et al. 2003 ). We predict general positive relationships between PET, TMIN and mammal species richness. The environmental variability hypothesis proposes that temporally less variable environments usually permit a greater number of species to coexist because species are able to specialize more and to evolve narrower ecological niches. In contrast, more variable environments are expected to have low species richness as a consequence of fewer species being able physiologically to tolerate the stressful conditions of varying environments (for theoretical discussions see e.g. Pianka 1966 , MacArthur 1972 , Rohde 1992 , Turner et al. 1996 , Brown and Lomolino 1998 , Whittaker et al. 2001 ). Empirical analyses of mammal species richness patterns have offered equivocal evidence in support of this explanation (see e.g. Kerr 1999 , Andrews and O'Brien 2000 ). We recognize the variability in resource supply (e.g. Connell and Orians 1964 ) and climate as different versions of the environmental variability hypothesis. We used phenological seasonality (SEAS) and interannual variability or instability (INST) in INDVI to test the resource supply‐stability hypothesis (i.e., intra‐ and interannual variability in resource availability). The difference in temperature between the warmest and coldest month i.e., yearly temperature amplitude (AMPLIT), is used to test the climatic‐stability hypothesis. We predict that mammal species richness increases as climatic and energy supply stability increases. The habitat‐heterogeneity hypothesis proposes that high spatial heterogeneity promotes the persistence of high species richness because the limiting resources can be more readily subdivided in complex habitats. This promotes greater specialization and the coexistence of a great number of species (see Simpson 1964 , Pianka 1966 , MacArthur 1972 , McCoy and Connor 1980 , Rohde 1992 , Kerr and Packer 1997 , Brown and Lomolino 1998 ). In the Americas, the high habitat complexity and strong elevation gradient associated with the presence of mountains in the west of the continent are associated with high diversity in mammals ( Simpson 1964 , Patterson 1994 , Patterson et al. 1996 , Kerr and Packer 1997 ). In the present analysis, we use the spatial heterogeneity in NDVI (HETER) and elevation variability (EVAR) as two measures of habitat‐heterogeneity. We predict that the number of mammal species increases as HETER and EVAR increase. Methods Data on South American mammals Distributional information was compiled for 825 species of the following taxa: Marsupialia (MAR; N=61), Edentata (EDE; N=28), Chiroptera (CHI; N=189), Primates (PRI; N=78), Rodentia (Hystricognathi, HYS; N=155, Sciurognathi, SCI; N=247), Artiodactyla (ART; N=20), Carnivora (CAR; N=42), Perissodactyla (N=3), Lagomorpha (N=2). The basic data are the same used in Ruggiero (1994, 1999) and Ruggiero et al. (1998) . These previous papers should be consulted for details of the main sources of distributional information used and discussion of the potential problems associated with these data. In the present paper we added data on Sciurognathi and updated the original information for all mammal species based on the maps published in Eisenberg (1989) , Redford and Eisenberg (1992) , and Eisenberg and Redford (1999) . The geographic range of each species was drawn onto a cylindrical equal area (Peters’) projection map of South America, overlaid by a grid of 170 squares, each cell approximately covering 123 000 km 2 . The spatial resolution is the same used in Ruggiero (1999) and was maintained fixed throughout the present analysis. The presence (1) or absence (0) of each species was recorded in each cell of the grid map. Given the coarse‐scale resolution of our analysis, we adopt a conservative criterion to delimit species’ distributions; a species was assigned to be “present” when its geographic range covers at least 25% of any cell. Individual species grids were entered and further processed in IDRISI ver. 2.0 ( Eastman 1997 ). Species richness throughout the present analysis was the total number of species coexisting in each cell of a grid map. We analyse whether similar (or different) environmental determinants account for species richness patterns in narrow‐ranging, intermediate‐ranging and widespread species. Range size was estimated simply by counting the number of cells occupied by each species. Given the coarse‐scale resolution of our analysis, it was not possible to settle an exact limit for each range size quartile. Thus, each quartile actually proportionally represented 27.6% (Q 1 : narrow‐ranging species), 23.1% (Q 2 : intermediate‐ranging species), 24.2% (Q 3 : intermediate‐ranging species) and 24.9% (Q 4 : widespread‐ ranging species) of the total of 825 mammal species. Environmental data Remotely sensed vegetation data We used the NOAA/NASA Pathfinder AVHRR NDVI dataset (1981–1999) sub‐setted for South America (83°W to 33°W; 13.2°N to 57°S) ( Agbu and James 1994 ). The Normalized Difference Vegetation Index (NDVI) measures the proportion of the photosynthetically absorbed radiation and vegetation structure. It is calculated from atmospherically corrected reflectance (R) from the visible and near‐infrared AVHRR channels. Data of NDVI originally acquired at 1×1 km resolution were resampled into 8×8 km (4.35×4.35′) using the equal‐area Goode Interrupted Homolosine Projection ( Agbu and James 1994 ). Data were derived from compositing daily‐derived images over 10‐d periods to minimize cloudiness and smoke and/or fog effects. Ten‐day maximum NDVI composites retained considerable amount of anomalous reflectance caused by cloud and/or smoke contamination. Thus, ten‐day maximum NDVI images were composited to increasingly larger periods until clouds and/or smoke effects disappeared. Because complete loss of these effects was attained with six‐month maximum NDVI composites, semester NDVI composites were obtained for the periods October–March (OM i ) and April–September (AS i ). Unacceptable quality in images from the second semester of 1993 through the first semester of 1995, the OM 93–94, OM 94–95 and AS 94 NDVI images precluded their inclusion in the general dataset. Thus, the total number of images over which analyses were performed was n=15 and n=16 for OM i and AS i , respectively. Each semester was averaged over the years to obtain mean six‐month NDVI (OM and AS). We used six‐month composite NDVI images to estimate the following variables: annually Integrated Normalized Difference Vegetation Index (INDVI), phenological seasonality (SEAS), mean annual integrated NDVI (INDVI), interannual variability or instability (INST), and spatial heterogeneity in NDVI (HETER). We calculated phenological seasonality (SEAS) as: Annual integrated NDVI (INDVI i ) for each of the 15 years was obtained as: The degree of interannual variability or for instability (INST) in INDVI in each pixel had was estimated as the coefficient of variation: Spatial heterogeneity in NDVI (HETER) was quantified by calculating Shannon's diversity index over a 7×7 pixel (ca. 56×56 km) moving window on the INDVI image: where p is the proportion of each of 128 INDVI classes in the 7×7 window. NDVI cover‐class diversity was preferred over cover class richness as it accounts for the evenness of cover classes. Because of slight registration errors in the multitemporal dataset, contamination by pixels corresponding to ocean reflectance occurred. By applying a distance operator to the ocean mask, all pixels closer than 2 pixels (diagonals included) from water were excluded from pixel aggregation process for the estimation of INDVI, SEAS and INST. Similarly, to avoid the moving window that estimated heterogeneity to be contaminated by pixels corresponding to water, all pixels closer than 3 pixels (diagonals included) from water were excluded from pixel aggregation process for the estimation of HETER. Other physical variables We estimated elevation variability (EVAR) as the standard deviation in elevation calculated over a 20×20 pixel (ca 370×370 km) window on the 10×10′ Clark FNOC Elevation, Terrain, and Surface Characteristics Global Dataset ( Clark 1992 ). Minimum temperature of the coldest month (TMIN) and yearly temperature amplitude (AMPLIT; the difference in temperature between the warmest and coldest month) were obtained from the 30×30′ IIASA database for mean monthly values of temperature, precipitation, and cloudiness on a global terrestrial grid ( Leemans and Cramer 1992 ). Annual actual (AET) and potential evapotranspiration (PET) were obtained from the 30×30′ Ahn and Tateishi monthly potential and actual evapotranspiration and water balance dataset ( Ahn and Tateishi 1994 ). To quantify variation in species richness produced in coastal areas due to reductions in the amount of land, the proportion of (continental) land in each cell (PLAND) was calculated and incorporated as an independent variable. Geographic rectification and resampling All images corresponding to the environmental variables were projected to the Peters projection of the mammal dataset rubbersheeting or 20 uniformly distributed ground control points applying quadratic transformation and nearest neighbor resampling to a new resolution that was a multiple of the resolution of the mammal dataset. Finally, images were contracted to the resolution of mammal dataset by pixel averaging over fixed windows corresponding geographically to the cells of the mammal dataset. All operations were performed with IDRISI ver. 2.0 ( Eastman 1997 ). Analyses A number of methods have been proposed to assess the effects of spatial structuring of variables in statistical modeling (see Legendre 1993 , Lichstein et al. 2002 , Diniz‐Filho et al. 2003 ). We adopted the so‐called “raw data” approach, whereby the effects of environmental variables on species richness variation are examined by partial regression analysis; the effect of space in this kind of analysis is partitioned out by site variables as in trend surface analysis (see Borcard et al. 1992 , Legendre 1993 , Boone and Krohn 2000 , Lichstein et al. 2002 , Van Rensburg et al. 2002 for detailed explanation). We partialled out the spatial variation in mammal species richness into for components, representing the effects of a) local (“non‐spatial”: sensu Borcard et al. 1992 , Legendre 1993 ) environmental variation, b) regional (“spatially structured”: sensu Borcard et al. 1992 , Legendre 1993 ) environmental variation, c) spatial variation of mammal species richness that is not shared by the environmental variables; this component could reflect the effect of other unknown biotic or abiotic processes that are also spatially structured (e.g. species interactions within communities, social aggregation, etc.), interactions between variables considered in the present study that were not included in the statistical model (see Borcard et al. 1992 , Boone and Krohn 2000 ), and d) unexplained variation; this component represents the local effect of other unknown environmental factors. Multicollinearity among predictor variables may introduce serious distortions in standard multiple regression analyses (see e.g. Chattrjee and Price 1977 ). All environmental and spatial predictors used in the present analyses are significantly correlated (all r s >0.7, p<0.0001), with the highest Spearman correlation coefficients (r s ) obtained among energy descriptors. This complicates selection of the best environmental and spatial predictors of mammal species richness (see Diniz‐Filho et al. 2003 for discussion). To avoid final models with highly redundant and multicollinear data structures, we applied a forward selection procedure with the proviso that a tolerance threshold of 0.6 was used when entering a new variable into the model. Tolerance is equal to (1−R i 2 ), where R i 2 is the squared multiple correlation coefficient from the regression of the ith predictor variable on all other independent variables in the regression equation. Once the best environmental and spatial predictors have been selected by this procedure, we combined their effects in a standard multiple linear regression analysis. This allows us to evaluate the significance of the best environmental predictors in the presence of the best spatial predictors. In the section of results that follows, we report parameter estimates exclusively from these final regression models. Partial regression analyses performed in the present study involved several steps: 1) Forward elimination procedure, with a tolerance threshold of 0.6 to overcome the problems generated by multicollinearity, was used to select the best environmental predictors for each taxonomic group or quartile of species considered. The coefficient of determination from this step (r 1 2 ) measured the proportion of the total variation of mammal species richness explained by the local and regional spatially structured environmental variation (i.e. fraction (a+b) below). 2) A FORTRAN program (SpaceMaker: compiled version for DOS available at ) was used to describe the spatial structure of data by a third order polynomial function of the geographic coordinates of the cells in the grid map. To fit the trend surface to data on mammal species richness, we regressed species richness on all polynomials terms: Forward elimination procedure, with a threshold tolerance of 0.6, was used to select the individual terms of the spatial polynomial to be considered the best predictors of species richness for each mammal group or quartile of mammal species. The coefficient of determination from step 2 (r 2 2 ) measured the proportion of the total variation of mammal species richness explained by regional spatially structured environmental variation and spatial variation that is not shared by the environmental variables (i.e. fraction (b+c) below). 3) A standard multiple linear regression analysis of species richness applied on the previously selected (i.e. in steps 1 and 2) environmental and spatial variables allowed to estimate the total proportion of variance of species richness explained (r 3 2 ) by all (spatial and environmental) variables considered for each group of species (fraction (a+b+c) below). 4) As in Boone and Krohn (2000) , the different sources of variation were calculated as: Note that partial regression analysis, as applied in the present study, is only an exploratory way of examining the extent to which spatial autocorrelation may be influencing the effects of environmental factors operating on species richness because it only accounts for broad‐scale spatial trends; however, even after controlling for broad‐scale spatial patterns in the variables examined, the residuals in the fitted model may still be autocorrelated, indicating evidence of spatial dependence (see Lichstein et al. 2002 , Diniz‐Filho et al. 2003 ). We performed all the analyses twice. Preliminary analyses were based on raw (untransformed) variables. Then, we performed all analyses based on transformed variables because the assumption of a linear relationship between species richness and some of the predictor variables (EVAR, AMPLIT, SEAS, INST) was better conformed when these variables were log 10 ‐transformed; we also applied a square root transformation on the dependent variable to ensure normality and constancy of error variance (see e.g. Chattrjee and Price 1977 ). We evaluated qualitatively the extent to which the use of transformations changed our results, i.e. whether different predictor variables were selected based on transformed or raw models. Throughout the present analysis, the tests of hypotheses were performed first for all non‐flying mammal species taken together. Then, we repeat all the analyses separately for each mammal group or taxon considered and for each range size quartile. Note, however, that the sample size in each analysis is the number of grid squares, rather than the number of species, and so is fairly constant across all analyses: Marsupialia (MAR; N=154), Edentata (EDE; N=155), Chiroptera (CHI; N=166), Primates (PRI; N=124), Rodentia (Hystricognathi, HYS; N=165, Sciurognathi, SCI; N=165), Artiodactyla (ART; N=165), Carnivora (CAR; N=166), and range size quartiles (Q 1–4 ; N=165). Results Tests of hypotheses Results based on transformed and raw data differ slightly. Given that the main qualitatively trends are found in both analyses, we only report here the results of the analyses based on raw data. The species‐energy hypotheses These hypotheses predict a general positive relationship between productivity (AET, INDVI), ambient energy (TMIN, PET) and mammal species richness at the continental scale. When all non‐flying mammal species are analysed together, the productivity hypothesis is supported; actual evapotranspiration (AET) is selected as the best energy‐based predictor of mammal species richness ( Table 1 ); the significant positive effect of AET on species richness is maintained after including the effect of spatial structuring of data into the analysis (see Table 1 and Appendix 1). 1 Summary of key results obtained throughout the present analysis. Significant (p<0.05) positive or negative relationships between variables after including the effect of spatial structure. One relationship that turned non‐significant (p>0.05, in marsupials) after including the effect of spatial structure is within parentheses. TOT=all mammal species excluding bats, MAR=marsupials, EDE=edentates, CHI=bats (Chiroptera), PRI=primates, HYS=hystricognath rodents, CAR=carnivores, ART=artiodactyls, SCI=sciurognath rodents. Q 1 , Q 2 , Q 3 and Q 4 are quartiles as defined in the main text. See Appendix 1 for estimated coefficients and exact probability values. HYPOTHESES Species‐energy Environmental stability Habitat Land Productivity Ambient energy Resource supply Climatic Heterogeneity Area AET INDVI TMIN PET SEAS INST AMPLIT HETER EVAR PLAND Taxonomic groups TOT + + + MAR + (+) EDE + CHI + PRI − − HYS + + + CAR + + ART + + SCI + + Quartiles Q 1 + + + Q 2 + + Q 3 + + Q 4 + When eight mammal taxa are analysed separately, both the productivity and ambient energy versions of the energy hypotheses are supported. After including the effect of spatial structuring of data into the analyses, 4 out of 8 taxa show AET as a significant and positive predictor of mammal species richness. Similarly, the species richness of edentates increases as INDVI increases, and so do the richness of bats with TMIN and artiodactyls with PET ( Table 1 and Appendix 1); given the nature of multiple comparisons, and setting a p<0.05, 5% of the taxa would show a significant effect of the environmental variables considered by chance alone. However, the probability that 4 out of 8 taxa would show AET as a significant predictor of species richness (binomial p (4,8,0.05) =0.000014) indicates that the possibility of committing an overall Type I error is low (for INDVI, TMIN and PET, binomial p (1,8,0.05) =0.057). The partition of species into range size quartiles shows that the positive effect of energy on mammal species richness does not change substantially with range size. However, the narrowest‐ranging and widest‐ranging species show distinct biogeographic patterns ( Fig. 1 ). As a consequence, the productivity version of the hypothesis is supported by narrow‐ and intermediate‐ ranging species, with AET explaining just under 50% of mammal species richness variation. The ambient‐energy hypothesis is supported only by the widest‐ranging species, with TMIN explaining a greater proportion (ca 70%) of species richness variation ( Table 1 ; Appendix 1). 1 Geographic pattern of mammal species richness variation after species were assigned to range size quartiles. (a) Q 1 : first quartile, species’ range sizes covering from 1 to 3 cells in the grid map; (b) Q 2 : second quartile, species’ range sizes covering from >3 to 9 cells; (c) Q 3 : third quartile, species’ range sizes covering from >9 to 31 cells; (d) Q 4 : fourth quartile, species’ range sizes covering >31 cells. The environmental variability hypothesis This hypothesis predicts that mammal species richness will increase as climatic and energy supply stability increase, with this translating into a general negative relationship between phenological seasonality (SEAS), interannual variability in NDVI (INST), and yearly temperature amplitude (AMPLIT) and mammal species richness. When the effect of spatial structuring of data is included into the analyses only the primates support this prediction, showing a significant decrease in species richness as the yearly temperature amplitude (AMPLIT) increases ( Table 1 ; Appendix 1). Although inter‐annual variability in NDVI (INST) is selected as a significant predictor of mammal species richness variation for the narrowest ranging species (Q 1 ), the positive regression coefficient (b=0.19; Appendix 1) is against our original prediction. A region of high inter‐annual instability (INST) may explain this intriguing relationship, previously observed along the eastern slopes of the tropical Andes (see e.g. Fig. 4 in Fjeldså et al. 1999 ); in the present analysis, this region of high instability (INST) is included within cells where peaks in the richness of narrow‐ ranging species occurred. 4 Proportional representation of species for each taxonomic group in each range size quartile. The habitat heterogeneity hypothesis This hypothesis predicts that the number of mammal species will increase as spatial heterogeneity in NDVI (HETER) and elevation variability (EVAR) increase. When the effect of spatial structuring of data is included into the analyses, all non‐volant mammal species analysed together show a significant positive effect of EVAR on mammal species richness ( Table 1 ; Appendix 1). This positive effect is also shown in 3 out of the 8 taxa analysed separately (EVAR: binomial p (3,8,0.05) =0.00037). In contrast, the effect of spatial heterogeneity in vegetation (HETER) is less important and against our original prediction: it has a negative effect only on the species richness of primates, suggesting that species richness decreases in locally heterogeneous habitats (see Table 1 and Appendix 1). The partition of mammal species into range size quartiles confirms that elevation variability is an important predictor of mammal species richness for narrow‐ and intermediate‐ ranging species but not for the widest‐ranging species. The proportion of species richness variation explained by elevation variability decreases moderately with range size, from 48% in Q 1 to 34% in Q 3. Effects of spatial structure About 50% of the total mammal species richness variation at the continental scale is explained by the regional spatially structured component of environmental variation. However, there are clear differences across taxa in the proportion of variance explained by this component (e.g. between bats: 74% and primates or rodents: ca 20–30%; Fig. 2 ). Not surprisingly, this component accounts for a greater proportion of species richness variation in the widest‐ranging species (Q 4 ) rather than in intermediate‐ or narrow‐ ranging species ( Fig. 2 ). Effects of taxonomy and range size are also detectable on the proportion of mammal species richness variance explained by local environmental effects. In general, such effects explain <30% of mammal species richness variation. However, local effects of environment are higher for primates (42%) and rodents (hystricognath rodents: 22%, sciurognath rodents: 21%) and almost undetectable for edentates (0.04%), bats (0.04%) and artiodactyls (0.05%) ( Fig. 2 ). In general, local environmental effects on mammal species richness tend to decrease with range size ( Fig. 2 ). 2 Proportion (%) of species richness variance explained by environment in each mammal taxon (top) or range size quartile (bottom). Env. (local): variation in species richness explained by fine‐scale effects of environment. Env. (regional): variation in species richness explained by regional spatially structured variation of environment. Space: spatial variation of mammal species richness that is not shared by the environmental variables (unexplained spatially structured variation). Unexpl.: unexplained variation (=local effects of unknown environmental factors). The proportion of variance explained by component c) (spatially structured processes not considered in the present analysis) is <2% in all the analyses performed ( Fig. 2 ). This suggests that the spatial descriptors selected in the best‐fit models adequately describe the spatial structure in these ecological data. The proportion of variance that remained unexplained by the environmental factors considered in the present analysis is between 10 and 50% depending upon the taxonomic group considered; the partition of species into range size quartiles shows that a greater proportion of variance remains unexplained for narrow‐ and intermediate‐ ranging species than for widespread species ( Fig. 2 ). These trends suggest local effects of other unknown environmental descriptors on the mammal species richness pattern. Discussion Environmental effects Mammals in South America support the productivity version of the species‐energy hypothesis, and suggest actual evapotranspiration (AET) is the main energy predictor. This favors the idea that it is not direct energy per se but its transformation into different levels of resources available to mammals that is the mechanism underlying the species‐energy relationship at the continental scale (see also Van Rensburg et al. 2002 and other references cited therein). The ambient energy hypothesis receives less support throughout the present analysis. Our study contradicts previous evidence suggesting that potential evapotranspiration (PET) is the most important and least ambiguous predictor of species richness variation at continental (e.g. Currie 1991 , Kerr and Packer 1997 , Hawkins et al. 2003 ) or global scales (e.g. Whittaker et al. 2001 ). We also show that the effect of energetic constraints – as expressed by the minimum temperature of the coldest month (TMIN) – is detectable only for bats and/or for the widest‐ranging species. Indeed, bats contribute with a great proportion of species to the widest‐ranging quartile (see details below), which explains the parallel trend. Neotropical bats are generally considered to be eurytopic with their distributional limits being generally poorly constrained by phytogeographic zones (e.g. Willig and Mares 1989 ). Although most bats are insectivores, they have become adapted to obtain food from a variety of sources – e.g. insects, fish, fruit, nectar, small vertebrates, blood, and because of their specific capacities for echolocation and maneuverable flight they have been able to develop nocturnal habits (see e.g. Neuweiler 1989 , Arita and Fenton 1997 ). In general, Neotropical bats are poor thermo‐ regulators although the capacity to ensure effective temperature regulation varies among different taxa depending upon food habits ( Fig. 3 in McNab, 1982 ). The subtropics (between 30° and 35°S) represent the southernmost limit of distribution for several taxa (Stenodermatinae, Carollinae, Phyllostominae, Sturnirinae, Desmodontinae: McNab 1982 ), which also suggests their sensitivity to cold. As a consequence, our finding of a significant positive effect of TMIN on bat species richness patterns is not surprising and could explain the strong latitudinal gradient in species diversity previously reported for this taxon. Indeed, latitude alone was previously reported to account for ca 85% of the variation in bat species density (e.g. Willig and Selcer 1989 , Kaufman and Willig 1998 ). 3 Synergism between energy availability and habitat heterogeneity. Species richness is principally controlled by energy availability (no effect of heterogeneity; top left), habitat heterogeneity and energy availability contribute additively to species richness (top center), and heterogeneity increases in importance with the increase in energy availability (multiplicative effect; top right). Bottom graphs represent best fits to quadratic surfaces of mammal species richness (TOT=non‐flying mammals, and each taxon) as a function of elevational variability (EVAR) and annually integrated NDVI (INDVI). The present analysis confirms that continental patterns of variation in mammal species richness are explained by the synergism between energy (as expressed by AET or INDVI) and habitat heterogeneity. Elevation variability is the main heterogeneity predictor, mainly due to the presence of the Andes running along the west of South America (see also Kerr and Packer 1997 , Rahbek and Graves 2001 ). The present analysis suggest that elevation variability is more influential on the species richness patterns of rodents (representing ca 48% of analysed species) and artiodactyls, and this may be explained by the biogeographic history of these taxa being clearly associated to the Andean habitats (see e.g. Franklin 1982 , Reig 1986 , Eisenberg 1987 , Marquet 1989 , Smith and Patton 1993 for discussion). Conceptually, we recognize three main ways whereby energy and heterogeneity can interact to promote species richness ( Fig. 3 , top). 1) Heterogeneity may have no effect and species richness is principally controlled by energy availability. 2) Habitat heterogeneity and energy availability may contribute independently to species richness (additive effects), or 3) Heterogeneity may increase in importance with the increase in energy availability (multiplicative effect; Fig. 3 top). Empirical evidence in the literature provides evidence for the last case ( Kerr and Packer 1997 ). Model 3 is also supported for all non‐flying mammals in our study. However, when high order taxa are analyzed separately, it is clear that different groups display different energy‐heterogeneity synergism. Edentates and Chiroptera conform to model 1, with richness relatively independent of heterogeneity. Primates and hysticognath rodents conform to model 2. Here, heterogeneity increases richness in both high and low energy regions. Note, however, that primates span a narrower range of energy variation (mainly tropical and subtropical) than the other taxa. Finally, carnivores, artiodactyls, sciurognath rodents, and marsupials all conform to model 3, showing that heterogeneity‐richness relationships become markedly steeper when energy availability increases ( Fig. 3 ). Clearly, the main behavior for all non‐flying species is heavily influenced by these last 4 taxa, representing ca 58% of the analyzed species. The climatic variability hypothesis is not supported throughout the present study, thus confirming a previous observation for North American mammals ( Kerr 1999 ). In contrast, we find a positive effect of inter‐annual variability in resource supply on the richness of narrow ranging species. The species richness pattern in the narrow‐ ranging species shows peaks in species richness associated with the Andean mountain ranges. The high ecoclimatic instability observed in the eastern slopes of the tropical Andes and coastal deserts of the western slopes may be partly the response to the dominant signal of El Niño Southern Oscillation (ENSO) which causes major dislocations in the dominant circulation that mostly affect topographically controlled rainfall regimes. Continental‐scaled hydrological studies indicate strongest responses to ENSO in the tropical eastern slopes and Andean highlands where dry and wet periods correspond to warm and cold phases of ENSO, respectively ( Dettinger et al. 2000 ). Paleoecological and archeological evidence also point to the eastern tropical Andes as a climatically sensitive region at both the inter‐annual and century scale ( Paulsen 1976 , Thompson et al. 1992 ). Over these tropical mountain ridges high spatial and temporal variability in rainfall may induce sharp changes in vegetation. Over long time scales these environmental changes may have differentially affected speciation patterns of narrow‐ranged mammals that have little migration potential in tropical mountain ranges ( Janzen 1967 ). At this point, however, we must emphasize that the definition of regional climatic variability most often used in analyses of species diversity (e.g. Kerr 1999 , Andrews and O'Brien 2000 ) refers exclusively to intra‐annual climatic variability, or seasonality, in temperature and precipitation. In reality, climatic variability can affect species’ distributions – and hence species richness patterns–at different temporal scales. Although in the present analysis we have tried in part to overcome this problem by estimating a measure of inter‐annual variability in resource supply (INST), our window of time is still well within an ecological scale of analysis (15 yr). Other processes acting at longer time scales could underlie the spatial variation in species richness at the continental scale, although they may remain unnoticed at the relatively short‐time scale of our present analysis (see Brown 1995 , Rohde 1996 , Dynesius and Jansson 2000 , Clarke and Crame 2003 for examples and discussions). Effects of range size, taxonomy and spatial structure Mammals in South America suggest that differences in the range sizes of species may affect our perception of environmental determinants of species richness patterns at the continental scale. Different environmental descriptors account for the species richness patterns in the narrowest‐ and widest‐ranging species, as previously observed by Jetz and Rahbek (2002) for the sub‐Saharan avifauna. However, a difference between Africa and South America is clear. In Africa, the importance of habitat heterogeneity in accounting for variation in bird species richness increases markedly with decreasing range size, and the opposite is true for the effect of productivity ( Jetz and Rahbek 2002 ). In South America, determinants of overall species richness (i.e., elevation variability and productivity) consistently explain about the same proportion of variance across most of the range size classes considered (i.e. from Q 1–3 ). Only the widest‐ranging species (Q 4 ) show a different pattern, as they support exclusively the ambient‐energy hypothesis (TMIN). The stronger effect of elevation variability to account for species richness patterns in South American mammals is, in part, explained by the greater distributional extent and complexity of the Andean mountain ranges compared to African mountains. Another reason is that species richness patterns of narrow‐ and intermediate‐ranging species are rather similar in South America. Both groups of species suggest an increase in mammal species richness toward the west, in tropical Andean regions. In contrast, the richness of the widest‐ranging mammal species peak in tropical lowlands ( Fig. 1 ). Clearly, the main determinants of mammal species richness differ markedly only between those range size quartiles that show distinct biogeographic patterns. Mammals in South America confirm that species richness variation at continental scale is driven mainly by broad‐scale spatially structured components of environmental variation, as previously reported by Boone and Krohn (2000) and Van Rensburg et al. (2002) . However, South American mammals allow more detailed interpretation of the extent to which differences in taxonomy and range size affect the proportion of species richness variation explained by local and regional environmental effects. Major taxonomic groups differ in the proportion of variance explained by fine‐scale effects of the environment. The partition of mammal species into range size quartiles clearly shows that local environmental effects account for a greater proportion of species richness variation for narrow‐ and intermediate‐ ranging species, whereas regional effects increase markedly in importance for the widest‐ranging species. However, differences in range size are not independent of differences in taxonomy in this species assemblage (χ 2 =282.52, p (21) <0.001; Fig. 4 ). Only marsupials and artiodactyls contribute with equal proportions of species to each size range size quartile (marsupials: χ 2 =0.907, p (3) >0.05; artiodactyls: χ 2 =4.49, p (3) >0.05). The other taxonomic groups contribute with a significant greater proportion of species to either the narrow‐ranging quartiles (primates: χ 2 =16.26, p (3) <0.01; Hystricognathii: χ 2 =32.1, p (3) <0.01; Sciurognathii: χ 2 =60.79, p (3) <0.01) or to the widest‐ranging ones (edentates: χ 2 =11.39, p (3) <0.05; bats: χ 2 =127.34, p (3) <0.001; carnivores: χ 2 =29.25, p (3) <0.01) ( Fig. 4 ). This complicates distinguishing between ecological and historical components of mammal species richness variation at continental scale. However, one possible interpretation, on ecological grounds, is that the influence of range size on species richness patterns may be related to the spatial scale or “grain” at which different mammal taxa perceive the environment, according to their body size and/or dispersal capabilities. Suggestively, in the present study, the species richness patterns in the high mobile bats and in the large body‐sized terrestrial artiodactyls were better explained by the regional spatially structured components of environmental variation. The proportion of species richness variance explained by local effects of environment is very low in bats (<1%), approaching values previously reported for birds in South Africa (<2%: Van Rensburg et al. 2002 ). In contrast, the highest proportion of variance explained by fine‐scale environmental effects is found in primates (ca 40%), and in the smaller and less vagile rodents (ca 20%). Clearly, only further analyses that control for functional groups and evolutionary and biogeographic history will provide a definite answer around these issues. Acknowledgements Insightful comments and suggestions were provided by T. M. Blackburn, S. L. Chown, J. A. F. Diniz‐Filho, A. G. Farji‐Brenner, B. A. Hawkins, P. A. Marquet, A. Premoli and K. Roy, and several other colleagues attending the Friday Afternoon Seminars at Laboratorio Ecotono. CONICET (PEI: no. 317/98) and Universidad Nacional del Comahue supported this project. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Ecography Wiley

Environmental correlates of mammal species richness in South America: effects of spatial structure, taxonomy and geographic range

Ruggiero, Adriana; Kitzberger, Thomas
Ecography , Volume 27 (4) – Aug 1, 2004
Download PDF
16 pages

Loading next page...
 
/lp/wiley/environmental-correlates-of-mammal-species-richness-in-south-america-gZJiAbsa57

References (86)

Publisher
Wiley
Copyright
Copyright © 2004 Wiley Subscription Services, Inc., A Wiley Company
ISSN
0906-7590
eISSN
1600-0587
DOI
10.1111/j.0906-7590.2004.03801.x
Publisher site
See Article on Publisher Site

Abstract

Recent years have brought some consensus regarding the primary roles of hypotheses based on climate and energy (e.g. species‐energy hypothesis: Wright 1983 , water‐energy dynamic theory: O'Brien 1993, 1998 , O'Brien et al. 2000 ) and historical contingency (i.e., glaciation effects, differences in dispersal and speciation rates) in explaining much of the variance in species richness from the equator to the poles ( Whittaker et al. 2001 , see also Gaston 2000 , Kerr 2001 , Hawkins et al. 2003 ). The area of biomes ( Rosenzweig 1992, 1995 , Blackburn and Gaston 1997 ), topographic heterogeneity ( Simpson 1964 , Kerr and Packer 1997 , Rahbek and Graves 2001 , Jetz and Rahbek 2002 ), levels of precipitation ( Rahbek and Graves 2001 , Hawkins et al. 2003 ), and several measures of the availability of energy in local environments, such as mean and maximum temperature, net primary productivity, actual and potential evapotransporation, and solar radiation received per unit area (see e.g. Currie 1991 , Blackburn and Gaston 1996a , Kerr and Packer 1997 , Kerr and Currie 1999 , Currie et al. 1999 , Van Rensburg et al. 2002 , Hawkins et al. 2003 ) appear to be the best candidates to explain latitudinal variation in species diversity in both hemispheres (see also Pianka 1966 , Begon et al. 1986 , Rohde 1992 , Ricklefs and Schluter 1993 , Brown 1995 , Rosenzweig 1995 , Currie et al. 1999 , Gaston 2000 , Willig 2000 , Gaston and Blackburn 2000 , Kerr 2001 , Whittaker et al. 2001 , and chapters in the “species richness” section in Blackburn and Gaston 2003 for reviews). Despite the obvious importance of such conceptual progress, some issues still await comprehensive understanding. One issue is related to the kind of spatial heterogeneity that is inherent to ecological systems. Large‐scale physical processes occurring over the earth surface, history and ecological interactions among species can lead the environmental variables, the species richness and their interactions to have complex spatial structures; as a consequence, environmental variables and species richness are very often spatially auto‐correlated, with this implying that two sites located near one another are unlikely to be independent from each other (e.g. Legendre 1993 , Boone and Krohn 2000 , Lichstein et al. 2002 , Van Rensburg et al. 2002 , Diniz‐Filho et al. 2003 ). This complicates standard statistical testing of hypotheses because it can create false positive– or negative– results (see e.g. Legendre 1993 , Diniz‐Filho et al. 2003 ). Moreover, multi‐scale relationships between species richness and environmental variables are increasingly well recognized to affect gradients of species diversity, which complicate further ecological interpretation (e.g. Lyons and Willig 1999 , Rahbek and Graves 2001 , Arita and Rodríguez 2002 , Blackburn and Gaston 2002 , Jetz and Rahbek 2002 , Diniz‐Filho et al. 2003 ). On the other hand, dominant factors associated with richness may shift with latitude at continental and global scales (e.g. Kerr and Packer 1997 , Rahbek and Graves 2001 , Hawkins et al. 2003 ). However, they may also differentially affect species with different body sizes, belonging to different trophic or spatial guilds (e.g. Andrews and O'Brien 2000 ), having different dispersal capacities ( Kerr and Currie 1999 ), geographic range sizes (e.g. Jetz and Rahbek 2002 ) or evolutionary histories (e.g. Lathman and Ricklefs 1993 ). As a consequence, complex latitudinal diversity gradients may emerge from the effect of processes interrelated at different scales of time and space. In the present paper, we address some of these issues using South American mammals as a model system. We incorporate the effects of spatial structuring of ecological variables into tests of three non‐mutually exclusive hypotheses which are most often proposed to account for spatial patterns of variation in mammal species richness: species‐energy, environmental stability and habitat‐heterogeneity hypotheses. Note, however, that we will not address the possible role of stochastic factors within bounded domains in affecting gradients in mammal species richness in South America. Although the “mid‐domain effect” (e.g. Colwell and Hurtt 1994 , Willig and Lyons 1998 , Colwell and Lees 2000 ) has been considered a credible hypothesis to explain the latitudinal diversity gradients during the past decade, recent work suggests that the model is flawed in its assumptions and explains very little variation in species richness at macro‐scale (see Hawkins and Diniz‐Filho 2002 , Zapata et al. 2003 ). A critical assessment of mid‐domain models on the species richness patterns of south american mammals is out of the scope of the present analysis (see however: Willig and Lyons 1998 , Willig 2000 ). In the present analysis, we conduct tests separately for different taxonomic groups of mammals (i.e. marsupials, edentates, primates, rodents, artiodactyls, carnivores and bats) to explore different ways whereby two of these hypotheses (energy and heterogeneity) may interact to explain variation in mammal species richness at the continental scale. We also partition the total number of mammal species into range size quartiles to give a more complete picture of species’ distributions. Species with small geographic ranges may be more likely limited by environmental factors that vary on a local or regional scale (e.g. topography); in contrast, widespread species should be less affected by such regional factors (e.g. Brown and Maurer 1989 , Brown 1995 , Jetz and Rahbek 2002 ). Hence, in the present paper, we analyse whether determinants of overall species richness patterns are also representative for groups of species with different distributions (e.g. for narrow‐ranging and widespread species, as in Jetz and Rahbek 2002 ). We also show that differences in range size are not independent of taxonomy in this species assemblage. Hypotheses The species‐energy hypothesis proposes that richness is limited by the total or average amount of energy entering into an ecosystem. It predicts that high‐energy availability promotes the persistence of high species richness ( Wright 1983 , see also Brown 1988 , Brown and Lomolino 1998 , Currie 1991 , Gaston and Blackburn 2000 ). Here, we follow Hawkins et al. (2003) and distinguish between two versions of the energy‐hypothesis (i.e., productivity vs ambient‐energy versions): The productivity hypothesis ( Wright 1983 , see also Hawkins et al. 2003 ) states that the level of resource production in an ecosystem limits animal species richness. In this paper, we test this hypothesis using measures of energy closely related to the level of primary production in a region: annual evapotranspiration (AET) and annually Integrated Normalized Difference Vegetation Index (INDVI). AET is the amount of water that actually evaporates or is transpired from an area, depending on the joint availability of energy and water (see e.g. Currie and Paquin 1987 , Currie 1991 , Andrews and O'Brien 2000 ). INDVI provides an integrated index of ecosystem function through its strong correlation with aboveground net primary productivity and absorbed photosynthetically active radiation ( Kerr and Ostrovsky 2003 ). INDVI shows a non‐linear relationship with the leaf‐area index that defines canopy structure ( Waring and Running 1998 ), however, it remains the most commonly used and most intensively studied vegetation index (see Kerr and Ostrovsky 2003 for discussion). Both AET and INDVI can be considered indicators of the availability of energy (in chemical form) for primary consumers (see Whittaker et al. 2001 ). We predict a general positive relationship between AET, INDVI and mammal species richness at continental scale. The ambient‐energy hypothesis (see Turner et al. 1996 , Hawkins et al. 2003 and other references therein) proposed that the animal species richness of a region is directly controlled by the total or average energy available. Given that endotherms usually maintain themselves at a higher temperature than that of the environment, the lower the ambient temperature the more these animals must spend to maintain body temperatures, and consequently the less energy they devote to growth and reproduction. All other things being equal, higher temperatures will promote faster growth of individuals and populations; this greater biomass will, in turn, promote greater species richness (see Brown 1988 , Blackburn and Gaston 1996b , Turner et al. 1996 , Hawkins et al. 2003 for more detailed discussion). In this paper, we use potential evapotranspiration (PET) and minimum temperature of the coldest month (TMIN) to test this hypothesis. PET is the amount of water that evaporates from a saturated surface, depending mainly on the amount of energy available to evaporate water and, to a lesser degree, on the relative humidity ( Currie 1991 , see also Andrews and O'Brien 2000 ). TMIN is a partial index of the environmental energy regime that describes the minimum degree of heat at each particular site (e.g. Andrews and O'Brien 2000 ). It also relates the ambient‐energy hypothesis to the original climatic freezing‐hypothesis of von Humboldt ( Hawkins 2001 , Hawkins et al. 2003 ). We predict general positive relationships between PET, TMIN and mammal species richness. The environmental variability hypothesis proposes that temporally less variable environments usually permit a greater number of species to coexist because species are able to specialize more and to evolve narrower ecological niches. In contrast, more variable environments are expected to have low species richness as a consequence of fewer species being able physiologically to tolerate the stressful conditions of varying environments (for theoretical discussions see e.g. Pianka 1966 , MacArthur 1972 , Rohde 1992 , Turner et al. 1996 , Brown and Lomolino 1998 , Whittaker et al. 2001 ). Empirical analyses of mammal species richness patterns have offered equivocal evidence in support of this explanation (see e.g. Kerr 1999 , Andrews and O'Brien 2000 ). We recognize the variability in resource supply (e.g. Connell and Orians 1964 ) and climate as different versions of the environmental variability hypothesis. We used phenological seasonality (SEAS) and interannual variability or instability (INST) in INDVI to test the resource supply‐stability hypothesis (i.e., intra‐ and interannual variability in resource availability). The difference in temperature between the warmest and coldest month i.e., yearly temperature amplitude (AMPLIT), is used to test the climatic‐stability hypothesis. We predict that mammal species richness increases as climatic and energy supply stability increases. The habitat‐heterogeneity hypothesis proposes that high spatial heterogeneity promotes the persistence of high species richness because the limiting resources can be more readily subdivided in complex habitats. This promotes greater specialization and the coexistence of a great number of species (see Simpson 1964 , Pianka 1966 , MacArthur 1972 , McCoy and Connor 1980 , Rohde 1992 , Kerr and Packer 1997 , Brown and Lomolino 1998 ). In the Americas, the high habitat complexity and strong elevation gradient associated with the presence of mountains in the west of the continent are associated with high diversity in mammals ( Simpson 1964 , Patterson 1994 , Patterson et al. 1996 , Kerr and Packer 1997 ). In the present analysis, we use the spatial heterogeneity in NDVI (HETER) and elevation variability (EVAR) as two measures of habitat‐heterogeneity. We predict that the number of mammal species increases as HETER and EVAR increase. Methods Data on South American mammals Distributional information was compiled for 825 species of the following taxa: Marsupialia (MAR; N=61), Edentata (EDE; N=28), Chiroptera (CHI; N=189), Primates (PRI; N=78), Rodentia (Hystricognathi, HYS; N=155, Sciurognathi, SCI; N=247), Artiodactyla (ART; N=20), Carnivora (CAR; N=42), Perissodactyla (N=3), Lagomorpha (N=2). The basic data are the same used in Ruggiero (1994, 1999) and Ruggiero et al. (1998) . These previous papers should be consulted for details of the main sources of distributional information used and discussion of the potential problems associated with these data. In the present paper we added data on Sciurognathi and updated the original information for all mammal species based on the maps published in Eisenberg (1989) , Redford and Eisenberg (1992) , and Eisenberg and Redford (1999) . The geographic range of each species was drawn onto a cylindrical equal area (Peters’) projection map of South America, overlaid by a grid of 170 squares, each cell approximately covering 123 000 km 2 . The spatial resolution is the same used in Ruggiero (1999) and was maintained fixed throughout the present analysis. The presence (1) or absence (0) of each species was recorded in each cell of the grid map. Given the coarse‐scale resolution of our analysis, we adopt a conservative criterion to delimit species’ distributions; a species was assigned to be “present” when its geographic range covers at least 25% of any cell. Individual species grids were entered and further processed in IDRISI ver. 2.0 ( Eastman 1997 ). Species richness throughout the present analysis was the total number of species coexisting in each cell of a grid map. We analyse whether similar (or different) environmental determinants account for species richness patterns in narrow‐ranging, intermediate‐ranging and widespread species. Range size was estimated simply by counting the number of cells occupied by each species. Given the coarse‐scale resolution of our analysis, it was not possible to settle an exact limit for each range size quartile. Thus, each quartile actually proportionally represented 27.6% (Q 1 : narrow‐ranging species), 23.1% (Q 2 : intermediate‐ranging species), 24.2% (Q 3 : intermediate‐ranging species) and 24.9% (Q 4 : widespread‐ ranging species) of the total of 825 mammal species. Environmental data Remotely sensed vegetation data We used the NOAA/NASA Pathfinder AVHRR NDVI dataset (1981–1999) sub‐setted for South America (83°W to 33°W; 13.2°N to 57°S) ( Agbu and James 1994 ). The Normalized Difference Vegetation Index (NDVI) measures the proportion of the photosynthetically absorbed radiation and vegetation structure. It is calculated from atmospherically corrected reflectance (R) from the visible and near‐infrared AVHRR channels. Data of NDVI originally acquired at 1×1 km resolution were resampled into 8×8 km (4.35×4.35′) using the equal‐area Goode Interrupted Homolosine Projection ( Agbu and James 1994 ). Data were derived from compositing daily‐derived images over 10‐d periods to minimize cloudiness and smoke and/or fog effects. Ten‐day maximum NDVI composites retained considerable amount of anomalous reflectance caused by cloud and/or smoke contamination. Thus, ten‐day maximum NDVI images were composited to increasingly larger periods until clouds and/or smoke effects disappeared. Because complete loss of these effects was attained with six‐month maximum NDVI composites, semester NDVI composites were obtained for the periods October–March (OM i ) and April–September (AS i ). Unacceptable quality in images from the second semester of 1993 through the first semester of 1995, the OM 93–94, OM 94–95 and AS 94 NDVI images precluded their inclusion in the general dataset. Thus, the total number of images over which analyses were performed was n=15 and n=16 for OM i and AS i , respectively. Each semester was averaged over the years to obtain mean six‐month NDVI (OM and AS). We used six‐month composite NDVI images to estimate the following variables: annually Integrated Normalized Difference Vegetation Index (INDVI), phenological seasonality (SEAS), mean annual integrated NDVI (INDVI), interannual variability or instability (INST), and spatial heterogeneity in NDVI (HETER). We calculated phenological seasonality (SEAS) as: Annual integrated NDVI (INDVI i ) for each of the 15 years was obtained as: The degree of interannual variability or for instability (INST) in INDVI in each pixel had was estimated as the coefficient of variation: Spatial heterogeneity in NDVI (HETER) was quantified by calculating Shannon's diversity index over a 7×7 pixel (ca. 56×56 km) moving window on the INDVI image: where p is the proportion of each of 128 INDVI classes in the 7×7 window. NDVI cover‐class diversity was preferred over cover class richness as it accounts for the evenness of cover classes. Because of slight registration errors in the multitemporal dataset, contamination by pixels corresponding to ocean reflectance occurred. By applying a distance operator to the ocean mask, all pixels closer than 2 pixels (diagonals included) from water were excluded from pixel aggregation process for the estimation of INDVI, SEAS and INST. Similarly, to avoid the moving window that estimated heterogeneity to be contaminated by pixels corresponding to water, all pixels closer than 3 pixels (diagonals included) from water were excluded from pixel aggregation process for the estimation of HETER. Other physical variables We estimated elevation variability (EVAR) as the standard deviation in elevation calculated over a 20×20 pixel (ca 370×370 km) window on the 10×10′ Clark FNOC Elevation, Terrain, and Surface Characteristics Global Dataset ( Clark 1992 ). Minimum temperature of the coldest month (TMIN) and yearly temperature amplitude (AMPLIT; the difference in temperature between the warmest and coldest month) were obtained from the 30×30′ IIASA database for mean monthly values of temperature, precipitation, and cloudiness on a global terrestrial grid ( Leemans and Cramer 1992 ). Annual actual (AET) and potential evapotranspiration (PET) were obtained from the 30×30′ Ahn and Tateishi monthly potential and actual evapotranspiration and water balance dataset ( Ahn and Tateishi 1994 ). To quantify variation in species richness produced in coastal areas due to reductions in the amount of land, the proportion of (continental) land in each cell (PLAND) was calculated and incorporated as an independent variable. Geographic rectification and resampling All images corresponding to the environmental variables were projected to the Peters projection of the mammal dataset rubbersheeting or 20 uniformly distributed ground control points applying quadratic transformation and nearest neighbor resampling to a new resolution that was a multiple of the resolution of the mammal dataset. Finally, images were contracted to the resolution of mammal dataset by pixel averaging over fixed windows corresponding geographically to the cells of the mammal dataset. All operations were performed with IDRISI ver. 2.0 ( Eastman 1997 ). Analyses A number of methods have been proposed to assess the effects of spatial structuring of variables in statistical modeling (see Legendre 1993 , Lichstein et al. 2002 , Diniz‐Filho et al. 2003 ). We adopted the so‐called “raw data” approach, whereby the effects of environmental variables on species richness variation are examined by partial regression analysis; the effect of space in this kind of analysis is partitioned out by site variables as in trend surface analysis (see Borcard et al. 1992 , Legendre 1993 , Boone and Krohn 2000 , Lichstein et al. 2002 , Van Rensburg et al. 2002 for detailed explanation). We partialled out the spatial variation in mammal species richness into for components, representing the effects of a) local (“non‐spatial”: sensu Borcard et al. 1992 , Legendre 1993 ) environmental variation, b) regional (“spatially structured”: sensu Borcard et al. 1992 , Legendre 1993 ) environmental variation, c) spatial variation of mammal species richness that is not shared by the environmental variables; this component could reflect the effect of other unknown biotic or abiotic processes that are also spatially structured (e.g. species interactions within communities, social aggregation, etc.), interactions between variables considered in the present study that were not included in the statistical model (see Borcard et al. 1992 , Boone and Krohn 2000 ), and d) unexplained variation; this component represents the local effect of other unknown environmental factors. Multicollinearity among predictor variables may introduce serious distortions in standard multiple regression analyses (see e.g. Chattrjee and Price 1977 ). All environmental and spatial predictors used in the present analyses are significantly correlated (all r s >0.7, p<0.0001), with the highest Spearman correlation coefficients (r s ) obtained among energy descriptors. This complicates selection of the best environmental and spatial predictors of mammal species richness (see Diniz‐Filho et al. 2003 for discussion). To avoid final models with highly redundant and multicollinear data structures, we applied a forward selection procedure with the proviso that a tolerance threshold of 0.6 was used when entering a new variable into the model. Tolerance is equal to (1−R i 2 ), where R i 2 is the squared multiple correlation coefficient from the regression of the ith predictor variable on all other independent variables in the regression equation. Once the best environmental and spatial predictors have been selected by this procedure, we combined their effects in a standard multiple linear regression analysis. This allows us to evaluate the significance of the best environmental predictors in the presence of the best spatial predictors. In the section of results that follows, we report parameter estimates exclusively from these final regression models. Partial regression analyses performed in the present study involved several steps: 1) Forward elimination procedure, with a tolerance threshold of 0.6 to overcome the problems generated by multicollinearity, was used to select the best environmental predictors for each taxonomic group or quartile of species considered. The coefficient of determination from this step (r 1 2 ) measured the proportion of the total variation of mammal species richness explained by the local and regional spatially structured environmental variation (i.e. fraction (a+b) below). 2) A FORTRAN program (SpaceMaker: compiled version for DOS available at ) was used to describe the spatial structure of data by a third order polynomial function of the geographic coordinates of the cells in the grid map. To fit the trend surface to data on mammal species richness, we regressed species richness on all polynomials terms: Forward elimination procedure, with a threshold tolerance of 0.6, was used to select the individual terms of the spatial polynomial to be considered the best predictors of species richness for each mammal group or quartile of mammal species. The coefficient of determination from step 2 (r 2 2 ) measured the proportion of the total variation of mammal species richness explained by regional spatially structured environmental variation and spatial variation that is not shared by the environmental variables (i.e. fraction (b+c) below). 3) A standard multiple linear regression analysis of species richness applied on the previously selected (i.e. in steps 1 and 2) environmental and spatial variables allowed to estimate the total proportion of variance of species richness explained (r 3 2 ) by all (spatial and environmental) variables considered for each group of species (fraction (a+b+c) below). 4) As in Boone and Krohn (2000) , the different sources of variation were calculated as: Note that partial regression analysis, as applied in the present study, is only an exploratory way of examining the extent to which spatial autocorrelation may be influencing the effects of environmental factors operating on species richness because it only accounts for broad‐scale spatial trends; however, even after controlling for broad‐scale spatial patterns in the variables examined, the residuals in the fitted model may still be autocorrelated, indicating evidence of spatial dependence (see Lichstein et al. 2002 , Diniz‐Filho et al. 2003 ). We performed all the analyses twice. Preliminary analyses were based on raw (untransformed) variables. Then, we performed all analyses based on transformed variables because the assumption of a linear relationship between species richness and some of the predictor variables (EVAR, AMPLIT, SEAS, INST) was better conformed when these variables were log 10 ‐transformed; we also applied a square root transformation on the dependent variable to ensure normality and constancy of error variance (see e.g. Chattrjee and Price 1977 ). We evaluated qualitatively the extent to which the use of transformations changed our results, i.e. whether different predictor variables were selected based on transformed or raw models. Throughout the present analysis, the tests of hypotheses were performed first for all non‐flying mammal species taken together. Then, we repeat all the analyses separately for each mammal group or taxon considered and for each range size quartile. Note, however, that the sample size in each analysis is the number of grid squares, rather than the number of species, and so is fairly constant across all analyses: Marsupialia (MAR; N=154), Edentata (EDE; N=155), Chiroptera (CHI; N=166), Primates (PRI; N=124), Rodentia (Hystricognathi, HYS; N=165, Sciurognathi, SCI; N=165), Artiodactyla (ART; N=165), Carnivora (CAR; N=166), and range size quartiles (Q 1–4 ; N=165). Results Tests of hypotheses Results based on transformed and raw data differ slightly. Given that the main qualitatively trends are found in both analyses, we only report here the results of the analyses based on raw data. The species‐energy hypotheses These hypotheses predict a general positive relationship between productivity (AET, INDVI), ambient energy (TMIN, PET) and mammal species richness at the continental scale. When all non‐flying mammal species are analysed together, the productivity hypothesis is supported; actual evapotranspiration (AET) is selected as the best energy‐based predictor of mammal species richness ( Table 1 ); the significant positive effect of AET on species richness is maintained after including the effect of spatial structuring of data into the analysis (see Table 1 and Appendix 1). 1 Summary of key results obtained throughout the present analysis. Significant (p<0.05) positive or negative relationships between variables after including the effect of spatial structure. One relationship that turned non‐significant (p>0.05, in marsupials) after including the effect of spatial structure is within parentheses. TOT=all mammal species excluding bats, MAR=marsupials, EDE=edentates, CHI=bats (Chiroptera), PRI=primates, HYS=hystricognath rodents, CAR=carnivores, ART=artiodactyls, SCI=sciurognath rodents. Q 1 , Q 2 , Q 3 and Q 4 are quartiles as defined in the main text. See Appendix 1 for estimated coefficients and exact probability values. HYPOTHESES Species‐energy Environmental stability Habitat Land Productivity Ambient energy Resource supply Climatic Heterogeneity Area AET INDVI TMIN PET SEAS INST AMPLIT HETER EVAR PLAND Taxonomic groups TOT + + + MAR + (+) EDE + CHI + PRI − − HYS + + + CAR + + ART + + SCI + + Quartiles Q 1 + + + Q 2 + + Q 3 + + Q 4 + When eight mammal taxa are analysed separately, both the productivity and ambient energy versions of the energy hypotheses are supported. After including the effect of spatial structuring of data into the analyses, 4 out of 8 taxa show AET as a significant and positive predictor of mammal species richness. Similarly, the species richness of edentates increases as INDVI increases, and so do the richness of bats with TMIN and artiodactyls with PET ( Table 1 and Appendix 1); given the nature of multiple comparisons, and setting a p<0.05, 5% of the taxa would show a significant effect of the environmental variables considered by chance alone. However, the probability that 4 out of 8 taxa would show AET as a significant predictor of species richness (binomial p (4,8,0.05) =0.000014) indicates that the possibility of committing an overall Type I error is low (for INDVI, TMIN and PET, binomial p (1,8,0.05) =0.057). The partition of species into range size quartiles shows that the positive effect of energy on mammal species richness does not change substantially with range size. However, the narrowest‐ranging and widest‐ranging species show distinct biogeographic patterns ( Fig. 1 ). As a consequence, the productivity version of the hypothesis is supported by narrow‐ and intermediate‐ ranging species, with AET explaining just under 50% of mammal species richness variation. The ambient‐energy hypothesis is supported only by the widest‐ranging species, with TMIN explaining a greater proportion (ca 70%) of species richness variation ( Table 1 ; Appendix 1). 1 Geographic pattern of mammal species richness variation after species were assigned to range size quartiles. (a) Q 1 : first quartile, species’ range sizes covering from 1 to 3 cells in the grid map; (b) Q 2 : second quartile, species’ range sizes covering from >3 to 9 cells; (c) Q 3 : third quartile, species’ range sizes covering from >9 to 31 cells; (d) Q 4 : fourth quartile, species’ range sizes covering >31 cells. The environmental variability hypothesis This hypothesis predicts that mammal species richness will increase as climatic and energy supply stability increase, with this translating into a general negative relationship between phenological seasonality (SEAS), interannual variability in NDVI (INST), and yearly temperature amplitude (AMPLIT) and mammal species richness. When the effect of spatial structuring of data is included into the analyses only the primates support this prediction, showing a significant decrease in species richness as the yearly temperature amplitude (AMPLIT) increases ( Table 1 ; Appendix 1). Although inter‐annual variability in NDVI (INST) is selected as a significant predictor of mammal species richness variation for the narrowest ranging species (Q 1 ), the positive regression coefficient (b=0.19; Appendix 1) is against our original prediction. A region of high inter‐annual instability (INST) may explain this intriguing relationship, previously observed along the eastern slopes of the tropical Andes (see e.g. Fig. 4 in Fjeldså et al. 1999 ); in the present analysis, this region of high instability (INST) is included within cells where peaks in the richness of narrow‐ ranging species occurred. 4 Proportional representation of species for each taxonomic group in each range size quartile. The habitat heterogeneity hypothesis This hypothesis predicts that the number of mammal species will increase as spatial heterogeneity in NDVI (HETER) and elevation variability (EVAR) increase. When the effect of spatial structuring of data is included into the analyses, all non‐volant mammal species analysed together show a significant positive effect of EVAR on mammal species richness ( Table 1 ; Appendix 1). This positive effect is also shown in 3 out of the 8 taxa analysed separately (EVAR: binomial p (3,8,0.05) =0.00037). In contrast, the effect of spatial heterogeneity in vegetation (HETER) is less important and against our original prediction: it has a negative effect only on the species richness of primates, suggesting that species richness decreases in locally heterogeneous habitats (see Table 1 and Appendix 1). The partition of mammal species into range size quartiles confirms that elevation variability is an important predictor of mammal species richness for narrow‐ and intermediate‐ ranging species but not for the widest‐ranging species. The proportion of species richness variation explained by elevation variability decreases moderately with range size, from 48% in Q 1 to 34% in Q 3. Effects of spatial structure About 50% of the total mammal species richness variation at the continental scale is explained by the regional spatially structured component of environmental variation. However, there are clear differences across taxa in the proportion of variance explained by this component (e.g. between bats: 74% and primates or rodents: ca 20–30%; Fig. 2 ). Not surprisingly, this component accounts for a greater proportion of species richness variation in the widest‐ranging species (Q 4 ) rather than in intermediate‐ or narrow‐ ranging species ( Fig. 2 ). Effects of taxonomy and range size are also detectable on the proportion of mammal species richness variance explained by local environmental effects. In general, such effects explain <30% of mammal species richness variation. However, local effects of environment are higher for primates (42%) and rodents (hystricognath rodents: 22%, sciurognath rodents: 21%) and almost undetectable for edentates (0.04%), bats (0.04%) and artiodactyls (0.05%) ( Fig. 2 ). In general, local environmental effects on mammal species richness tend to decrease with range size ( Fig. 2 ). 2 Proportion (%) of species richness variance explained by environment in each mammal taxon (top) or range size quartile (bottom). Env. (local): variation in species richness explained by fine‐scale effects of environment. Env. (regional): variation in species richness explained by regional spatially structured variation of environment. Space: spatial variation of mammal species richness that is not shared by the environmental variables (unexplained spatially structured variation). Unexpl.: unexplained variation (=local effects of unknown environmental factors). The proportion of variance explained by component c) (spatially structured processes not considered in the present analysis) is <2% in all the analyses performed ( Fig. 2 ). This suggests that the spatial descriptors selected in the best‐fit models adequately describe the spatial structure in these ecological data. The proportion of variance that remained unexplained by the environmental factors considered in the present analysis is between 10 and 50% depending upon the taxonomic group considered; the partition of species into range size quartiles shows that a greater proportion of variance remains unexplained for narrow‐ and intermediate‐ ranging species than for widespread species ( Fig. 2 ). These trends suggest local effects of other unknown environmental descriptors on the mammal species richness pattern. Discussion Environmental effects Mammals in South America support the productivity version of the species‐energy hypothesis, and suggest actual evapotranspiration (AET) is the main energy predictor. This favors the idea that it is not direct energy per se but its transformation into different levels of resources available to mammals that is the mechanism underlying the species‐energy relationship at the continental scale (see also Van Rensburg et al. 2002 and other references cited therein). The ambient energy hypothesis receives less support throughout the present analysis. Our study contradicts previous evidence suggesting that potential evapotranspiration (PET) is the most important and least ambiguous predictor of species richness variation at continental (e.g. Currie 1991 , Kerr and Packer 1997 , Hawkins et al. 2003 ) or global scales (e.g. Whittaker et al. 2001 ). We also show that the effect of energetic constraints – as expressed by the minimum temperature of the coldest month (TMIN) – is detectable only for bats and/or for the widest‐ranging species. Indeed, bats contribute with a great proportion of species to the widest‐ranging quartile (see details below), which explains the parallel trend. Neotropical bats are generally considered to be eurytopic with their distributional limits being generally poorly constrained by phytogeographic zones (e.g. Willig and Mares 1989 ). Although most bats are insectivores, they have become adapted to obtain food from a variety of sources – e.g. insects, fish, fruit, nectar, small vertebrates, blood, and because of their specific capacities for echolocation and maneuverable flight they have been able to develop nocturnal habits (see e.g. Neuweiler 1989 , Arita and Fenton 1997 ). In general, Neotropical bats are poor thermo‐ regulators although the capacity to ensure effective temperature regulation varies among different taxa depending upon food habits ( Fig. 3 in McNab, 1982 ). The subtropics (between 30° and 35°S) represent the southernmost limit of distribution for several taxa (Stenodermatinae, Carollinae, Phyllostominae, Sturnirinae, Desmodontinae: McNab 1982 ), which also suggests their sensitivity to cold. As a consequence, our finding of a significant positive effect of TMIN on bat species richness patterns is not surprising and could explain the strong latitudinal gradient in species diversity previously reported for this taxon. Indeed, latitude alone was previously reported to account for ca 85% of the variation in bat species density (e.g. Willig and Selcer 1989 , Kaufman and Willig 1998 ). 3 Synergism between energy availability and habitat heterogeneity. Species richness is principally controlled by energy availability (no effect of heterogeneity; top left), habitat heterogeneity and energy availability contribute additively to species richness (top center), and heterogeneity increases in importance with the increase in energy availability (multiplicative effect; top right). Bottom graphs represent best fits to quadratic surfaces of mammal species richness (TOT=non‐flying mammals, and each taxon) as a function of elevational variability (EVAR) and annually integrated NDVI (INDVI). The present analysis confirms that continental patterns of variation in mammal species richness are explained by the synergism between energy (as expressed by AET or INDVI) and habitat heterogeneity. Elevation variability is the main heterogeneity predictor, mainly due to the presence of the Andes running along the west of South America (see also Kerr and Packer 1997 , Rahbek and Graves 2001 ). The present analysis suggest that elevation variability is more influential on the species richness patterns of rodents (representing ca 48% of analysed species) and artiodactyls, and this may be explained by the biogeographic history of these taxa being clearly associated to the Andean habitats (see e.g. Franklin 1982 , Reig 1986 , Eisenberg 1987 , Marquet 1989 , Smith and Patton 1993 for discussion). Conceptually, we recognize three main ways whereby energy and heterogeneity can interact to promote species richness ( Fig. 3 , top). 1) Heterogeneity may have no effect and species richness is principally controlled by energy availability. 2) Habitat heterogeneity and energy availability may contribute independently to species richness (additive effects), or 3) Heterogeneity may increase in importance with the increase in energy availability (multiplicative effect; Fig. 3 top). Empirical evidence in the literature provides evidence for the last case ( Kerr and Packer 1997 ). Model 3 is also supported for all non‐flying mammals in our study. However, when high order taxa are analyzed separately, it is clear that different groups display different energy‐heterogeneity synergism. Edentates and Chiroptera conform to model 1, with richness relatively independent of heterogeneity. Primates and hysticognath rodents conform to model 2. Here, heterogeneity increases richness in both high and low energy regions. Note, however, that primates span a narrower range of energy variation (mainly tropical and subtropical) than the other taxa. Finally, carnivores, artiodactyls, sciurognath rodents, and marsupials all conform to model 3, showing that heterogeneity‐richness relationships become markedly steeper when energy availability increases ( Fig. 3 ). Clearly, the main behavior for all non‐flying species is heavily influenced by these last 4 taxa, representing ca 58% of the analyzed species. The climatic variability hypothesis is not supported throughout the present study, thus confirming a previous observation for North American mammals ( Kerr 1999 ). In contrast, we find a positive effect of inter‐annual variability in resource supply on the richness of narrow ranging species. The species richness pattern in the narrow‐ ranging species shows peaks in species richness associated with the Andean mountain ranges. The high ecoclimatic instability observed in the eastern slopes of the tropical Andes and coastal deserts of the western slopes may be partly the response to the dominant signal of El Niño Southern Oscillation (ENSO) which causes major dislocations in the dominant circulation that mostly affect topographically controlled rainfall regimes. Continental‐scaled hydrological studies indicate strongest responses to ENSO in the tropical eastern slopes and Andean highlands where dry and wet periods correspond to warm and cold phases of ENSO, respectively ( Dettinger et al. 2000 ). Paleoecological and archeological evidence also point to the eastern tropical Andes as a climatically sensitive region at both the inter‐annual and century scale ( Paulsen 1976 , Thompson et al. 1992 ). Over these tropical mountain ridges high spatial and temporal variability in rainfall may induce sharp changes in vegetation. Over long time scales these environmental changes may have differentially affected speciation patterns of narrow‐ranged mammals that have little migration potential in tropical mountain ranges ( Janzen 1967 ). At this point, however, we must emphasize that the definition of regional climatic variability most often used in analyses of species diversity (e.g. Kerr 1999 , Andrews and O'Brien 2000 ) refers exclusively to intra‐annual climatic variability, or seasonality, in temperature and precipitation. In reality, climatic variability can affect species’ distributions – and hence species richness patterns–at different temporal scales. Although in the present analysis we have tried in part to overcome this problem by estimating a measure of inter‐annual variability in resource supply (INST), our window of time is still well within an ecological scale of analysis (15 yr). Other processes acting at longer time scales could underlie the spatial variation in species richness at the continental scale, although they may remain unnoticed at the relatively short‐time scale of our present analysis (see Brown 1995 , Rohde 1996 , Dynesius and Jansson 2000 , Clarke and Crame 2003 for examples and discussions). Effects of range size, taxonomy and spatial structure Mammals in South America suggest that differences in the range sizes of species may affect our perception of environmental determinants of species richness patterns at the continental scale. Different environmental descriptors account for the species richness patterns in the narrowest‐ and widest‐ranging species, as previously observed by Jetz and Rahbek (2002) for the sub‐Saharan avifauna. However, a difference between Africa and South America is clear. In Africa, the importance of habitat heterogeneity in accounting for variation in bird species richness increases markedly with decreasing range size, and the opposite is true for the effect of productivity ( Jetz and Rahbek 2002 ). In South America, determinants of overall species richness (i.e., elevation variability and productivity) consistently explain about the same proportion of variance across most of the range size classes considered (i.e. from Q 1–3 ). Only the widest‐ranging species (Q 4 ) show a different pattern, as they support exclusively the ambient‐energy hypothesis (TMIN). The stronger effect of elevation variability to account for species richness patterns in South American mammals is, in part, explained by the greater distributional extent and complexity of the Andean mountain ranges compared to African mountains. Another reason is that species richness patterns of narrow‐ and intermediate‐ranging species are rather similar in South America. Both groups of species suggest an increase in mammal species richness toward the west, in tropical Andean regions. In contrast, the richness of the widest‐ranging mammal species peak in tropical lowlands ( Fig. 1 ). Clearly, the main determinants of mammal species richness differ markedly only between those range size quartiles that show distinct biogeographic patterns. Mammals in South America confirm that species richness variation at continental scale is driven mainly by broad‐scale spatially structured components of environmental variation, as previously reported by Boone and Krohn (2000) and Van Rensburg et al. (2002) . However, South American mammals allow more detailed interpretation of the extent to which differences in taxonomy and range size affect the proportion of species richness variation explained by local and regional environmental effects. Major taxonomic groups differ in the proportion of variance explained by fine‐scale effects of the environment. The partition of mammal species into range size quartiles clearly shows that local environmental effects account for a greater proportion of species richness variation for narrow‐ and intermediate‐ ranging species, whereas regional effects increase markedly in importance for the widest‐ranging species. However, differences in range size are not independent of differences in taxonomy in this species assemblage (χ 2 =282.52, p (21) <0.001; Fig. 4 ). Only marsupials and artiodactyls contribute with equal proportions of species to each size range size quartile (marsupials: χ 2 =0.907, p (3) >0.05; artiodactyls: χ 2 =4.49, p (3) >0.05). The other taxonomic groups contribute with a significant greater proportion of species to either the narrow‐ranging quartiles (primates: χ 2 =16.26, p (3) <0.01; Hystricognathii: χ 2 =32.1, p (3) <0.01; Sciurognathii: χ 2 =60.79, p (3) <0.01) or to the widest‐ranging ones (edentates: χ 2 =11.39, p (3) <0.05; bats: χ 2 =127.34, p (3) <0.001; carnivores: χ 2 =29.25, p (3) <0.01) ( Fig. 4 ). This complicates distinguishing between ecological and historical components of mammal species richness variation at continental scale. However, one possible interpretation, on ecological grounds, is that the influence of range size on species richness patterns may be related to the spatial scale or “grain” at which different mammal taxa perceive the environment, according to their body size and/or dispersal capabilities. Suggestively, in the present study, the species richness patterns in the high mobile bats and in the large body‐sized terrestrial artiodactyls were better explained by the regional spatially structured components of environmental variation. The proportion of species richness variance explained by local effects of environment is very low in bats (<1%), approaching values previously reported for birds in South Africa (<2%: Van Rensburg et al. 2002 ). In contrast, the highest proportion of variance explained by fine‐scale environmental effects is found in primates (ca 40%), and in the smaller and less vagile rodents (ca 20%). Clearly, only further analyses that control for functional groups and evolutionary and biogeographic history will provide a definite answer around these issues. Acknowledgements Insightful comments and suggestions were provided by T. M. Blackburn, S. L. Chown, J. A. F. Diniz‐Filho, A. G. Farji‐Brenner, B. A. Hawkins, P. A. Marquet, A. Premoli and K. Roy, and several other colleagues attending the Friday Afternoon Seminars at Laboratorio Ecotono. CONICET (PEI: no. 317/98) and Universidad Nacional del Comahue supported this project.

Journal

Ecography – Wiley

Published: Aug 1, 2004

There are no references for this article.