A recent reference sample for reconstructing paleo-environments
Model design
For this study we designed a reference dataset consisting of environmental and ecological variables for 30 national parks in southern Africasouth of 15°S latitude. The dataset includes variables for climate (temperature and precipitation), vegetation, and the diversity of specialized herbivore communities.
In order to quantify diversities of specialized herbivore communities we analyzed species lists retrieved from the Biological Inventories of the World’s Protected Areas Database (Meese 2005). We assigned every macromammal species to a diet category, a size class and a specific locomotion type. Those ecological variables are based on measurable features and can be recorded in fossil assemblages by a similar approach. We used a system consisting of five diet categories (frugivores–HFR, folivores/browsers–HXB, mixed feeders–HMB, and two different types of grazers–HMG, HXG), four size classes (1-10 kg; 10-100 kg; 100-1000 kg; >1000 kg) subdivided in threesub-ranges (a, b, and c), and five locomotion types (ungulate–UNG; ubiquitous–UBI;graviportal–GRA; amphibian–AMP; arboreal–ARB). Score values are either based on own measurements or retrieved from the literature (e.g.,Estes, 1991; ForteliusandSolounias 2000;Gagnon and Chew 2000; Kingdon 1997; Skinner andChimimba 2005; WerdelinandSanders 2010). The three ecovariables are combined into species-specific ecoprofiles (Tab.ESM-4). We then counted the number of species per ecoprofile in each assemblage.
Specialized herbivore communities depend in occurrence and distribution on vegetation with a specific structure, the distribution of which is in turn controlled by climate. Although all species in a community were scored, we restricted the analysis to the specialized herbivore subsample. Therefore, we excluded carnivores, omnivores and unspecialized herbivore taxa feeding on any kind of plant matter.
We performed a cluster analysis (complete linkage, Chi-square distance measure) in order to distinguish six different communities. The number of the groups corresponds to the number of biomes in the region (Olson et al. 2001).The areas inhabited by specific specialized herbivore communities were mapped. The communities are then analyzed with respect to their ecological composition and diversity (Tab. ESM-5).
Ecological diversities
Specialized herbivore communities are furthermore characterized by two different kinds of variables. We calculated ecological diversities deco in analogy to Margaleff’sd (Hayek andBuzas 2010; Magurran 2011). Margaleff’sd has been suggested as a measure for biological diversity in given communities. It relates species numbers to the logarithm of the number of individuals. We calculated a modified version by relating numbers of ecoprofiles with numbers of species:
deco represents a direct measure for the comparison of ecological diversities in the communities. In our reference sample, ecological diversities range from 0.910 (three species representing two ecoprofiles) and 6.115 (31 species representing 22 ecoprofiles). Additionally, we calculated relative ecological diversities by relating deco to the potential maximum value dmax. Relative ecological diversities then represent the relation between deco and dmax for any given assemblage.
While deco varies with the number of species and the number of ecoprofiles alike, drel provides a relative measure of the diversity in an assemblage independent of its numerical size. In the recent reference sample values vary between 0.50 (three species representing two ecoprofiles) and 1.00 (each species displays a unique ecoprofile).
Ecological variables
In order to examine the structure of the communities, we grouped the ecoprofiles into eight categories according to ecologically relevant variables (frugivores, folivores and browsers <20kg, frugivores, folivores and browsers ≥20 kg, mixed feeders <20kg, mixed feeders ≥20kg, less specialized grazers <200 kg, less specialized grazers ≥200kg, specialized grazers <200 kg, specialized grazers ≥200kg) and calculated the frequencies by species numbers. Frequencies permit to examine the ecological structure of the assemblages and help to identify ecological features of the specialized herbivore communities in the sixgroups (Tab. ESM-7).
In order to test the validity of the ecological categories we performed a discriminant analysis including the grouped ecological categories (n=8) plus deco and drel, in total ten predictor variables. The groups resulting from the cluster analysis areusedas grouping variable. The classification functions are given in Tab. ESM-8. F-statistics are applied to identify a set of variables which permits to statistically reproduce the clusters. The rank in F-statisticsfor all the variables is listed in Tab. ESM-7. The most significant variable is the proportion of small-bodied frugivores, folivores or browsers (HFRHXB<20). Overall ecological diversities (deco) are less important than redundancy in the organization of the communities (drel). Besides, grazing profile groups dominate mixed feeders and large browsers with the exception of large specialized grazers, which are insignificant for the distinction of the communities. In the descriptive summary, we focus on the more significant variables. The entire set is given in Tab. ESM-7.
Climate variables
We characterized climate and vegetation in the national parks. To obtain the range of values for variables of climate and vegetation density for each national park, each park was georeferenced in GIS using the program ESRI ArcView, and transformed into shapefiles. Shapefiles were intersected with the global climatology dataset WORLDCLIM, which is a raster dataset with a resolution of 10 arcminutes ( For each climate parameter the mean value calculated from all grid cells covered by the national park shapefile as well as the maximum and minimum values of the extracted climate data were calculated and taken as the climatic characteristics of the national park. We based our characterization of the 6 specialized herbivore habitats particularly on the mean annual temperature (MAT) and the difference between the mean temperature of the warmest and coldest quarter (∆T = MTQwarm – MTQcold). The mean annual temperature in the national parks varies between 12.9 and 22.4 °C. Temperature shifts between summer and winter quarter range from minimally 3.8 to maximally 14.7 °C. This variable is a measure for temperature seasonality. The area under study covers subtropical and temperate regions. Therefore, an increase in temperature seasonality with increasing latitude is observed. With respect to precipitation we recorded mean annual precipitation (MAP), shifts in rainfall patterns between wettest and driest quarter (∆P1 = PQwet – PQdry), and the difference in precipitation between warmest and coldest quarter (∆P2 = PQwarm - PQcold).Mean annual precipitation varies in our sample between 18 and 1266 mm. The sample therefore includes arid as well as humid habitats. The absolute difference between precipitation of wettest and driest quarter permits to check, whether pronounced rainy seasons occur.Differences vary between 6 and 541 mm. Positive values in ∆P2 indicate summer rain, while negative values indicate winter rain. In our sample values vary between -267 and 541 mm. Main characteristics of climate variables in the six areas are summarized in Tab. ESM-8.
Vegetation variables
The same procedure was repeated with a remotely sensed global raster dataset based on the ECOCLIMAP data of Masson et al. (2003) that was transposed to a resolution of 10 arcminutes to be comparable with the climate dataset. Here, mean, maximum, and minimum values for three vegetation parameters (leaf area index [LAI], vegetation cover and greenness) were determined as indicators of the size of biomass on the land surface (e.g., Wittich 1997). The vegetation cover and greenness parameters are given in values between zero (no vegetation) and one (complete vegetation cover and maximum photosynthesis activity, respectively). The LAI as a measure for canopy density gives the ratio of leaf area to per unit ground surface area (e.g., Kraus 2008; ZhengandMoskal 2009). In our dataset, the dimensionless variable ranges from zero (no leaves, i.e., no vegetation) up to 5.4 in tropical rainforests with a more than five times larger leaf area than ground area, e.g. with a dense multi-storey canopy. By measuring vegetation density, canopy density, and photosynthesis activity, the three parameters considered also give an estimate of the openness of the habitats. Main characteristics of vegetation variables in the 6 areas are summarized in Tab. ESM-8.
Specialized herbivore communities and habitats
This procedure resulted in the characterization of 6 types of specialized herbivore communities and related habitats.The groups are identified on the basis of faunal variables (Tab. ESM-7), but moreoverrepresent specific climatic regimes and characteristics of vegetation (Tab. ESM-8). This permits to distinguish types of ecosystems which are associated with the techno-complexes in focus of this study.Note that the descriptions of the groups are relative, i.e., communities and habitats are distinguished with respect to the reference sample only. Changes in the scope of the reference sample result in a new classification and the distinction of other classes. Depending on the purposes of a particular study, the reference sample needs to be adjusted.
Group A communities are presently restricted to forest mosaics in northern sections of the East Coast. The medium diverse communities are characterized by comparatively high diversities among frugivores, folivores and browsers with a body mass up to 20 kg and among fresh grass grazers. Fresh grass grazer communities are dominated by species with body masses below 200 kg. The communities quite redundantly ecological organized. Habitats are characterized by mean annual temperatures in a medium range, but comparatively high mean annual precipitation, which supports a dense vegetation cover. In frame of the discriminant analysis several reclassification schemes are carried out. In the direct version, no misclassifications occurred. However, in a jackknifed version, where the reclassified case is excluded from the model, 50% of all cases were misclassified with group F communities, occurring in geographically neighboring areas. Similar misclassifications occur with the set of habitat-related variables both in the direct and jackknifed version of the classification matrix. Apparently,misclassifications are due to low case numbers (n=2) which do not provide a very robust definition of the group. Yet, in spite of a suboptimal definition type A communities are better recognized than respective habitats.
Group B communities predominantly occur in coastal forests located on the southern coastal plain. Communities in this group are characterized by a highly diverse community of small bodied frugivores, folivores and browsers on one hand and low diversities among the grazing profile spectra. Fresh grass grazers over 200 kg and obligate grazers under 200 kg are absent. The communities are well recognized and misclassifications do not occur in spite of comparatively low case numbers (n=3). Mean annual temperatures are comparatively low. Because of the location of the habitats on the southern coastal plain, they are exposed to both, Mozambique and Agulhas currents. As an effect temperature seasonality is mitigated in these habitats. Mean annual precipitation is in a medium range and aseasonal.This leads to vegetation density and photosynthesis activity in a medium range with comparatively dense canopies. However, the habitat structure is quite variable resulting in misclassifications. Only 67% of all cases in the direct version are correctly identified. In the jackknifed version only one out of three cases is correctly identified. Although the communities are quite well defined, they are not specifically indicating a particular habitat type, but inhabit varying environments.
Group C communities occur at present mainly in forests and woodlands along the South African West Coast. Overall diversities of the communities fall in the medium range. Group C communities are not specifically redundant in their organization. With respect to their ecological structure, type C communities are dominated by mixed feeders, particularly species with a body mass under 20 kg. Although a higher number of cases is attributed to this community, 60% of all cases are misclassified in the jackknifed version of the classification matrix. This is caused by high variability among the cases and comparatively high standard variations (Tab. ESM-7). Misclassifications occur with communities of groups D and E, which occur in geographically neighboring regions. Habitats of group C communities are characterized by comparatively low mean annual temperatures, elevated temperature seasonality, mean annual precipitation in the medium range with clear winter rains. Vegetation and canopy densities as well as photosynthesis activity fall in a medium range, but are apparently highly variable among the habitats. In combination with low case numbers this leads to a good recognition in the direct version of the classification matrix, but 80% misclassifications in the jackknifed version of the classification matrix. Habitats are predominantly misclassified as representing group D.
Group D communities are presently restricted to the Karoo and Albany thicket on the southern coastal plain. The communities are highly diverse. Diversity among small frugivores, folivores and browsers is in the medium range, but the proportion of grazing profile groups is increasing at the expense of diversities among mixed feeders. Diversity among fresh grass grazers with a body mass over 200 kg are increasing in comparison to smaller forms. In the direct version of the classification matrix group D habitats are well recognized. In the jackknifed version figures drop to 75%. A single case is misclassified as representing a community of group E. Habitats of group D communities are characterized by the lowest mean annual temperatures in our sample linked with pronounced temperature seasonality. The precipitation regime is characterized by low mean annual precipitation, but a pronounced rainy season in summer.Such a climate regime produces medium dense vegetation with low canopy densities and photosynthesis activity in the medium range. Based on climate and vegetation the jackknifed version of the classification matrix illustrates misclassifications in half of the cases. With respect to the variable set we applied for habitat description, the habitats of group D communities are erroneously associated with group C habitats.
Group E communities occur at present in the Kalahari, the Etosha Pan and coastal deserts along the Atlantic coastline. Diversities are in the medium range, but the communities are highly diverse in their organization. Diversities among frugivores, folivores and browsers up to 20 kg are comparatively low, while the proportion of grazing species approaches 40%. The majority of the grazing taxa are represented by fresh grass grazers of either body mass category. Specialized grazers with body masses below 200 kg occur, but only in low diversities. Misclassifications in the direct version of the classification matrix do not exceed 12% and occur only with adjacent group F communities. In the jackknifed version, the figure increases to 37% and misclassifications cover also group D. Group E communities can be considered quite well characterized by the set of ecological variables. Habitats of group E communities are characterized by mean annual temperatures in the medium range, but elevated temperature seasonality. Precipitation patterns are characterized by low mean annual rainfall linked with high precipitation seasonality. The boundary between regions with summer and winter rains apparently divides the distribution area. Such a climate regime supports only open vegetation with open canopies and low photosynthesis activity. Such open vegetation patterns do not mitigate regional climate. The set of variables we chose for description of the habitats misclassifies 25% of the cases in the direct and 37% of the cases in the jackknifed version.
Group F communities are presently restricted to the grasslands, savannas and shrublands in the northeastern part of our study area. Communities are extremely diverse and reach peak values in our model. Relative diversities, however, indicate a quite redundant ecological organization. The ecological structure of the communities shows small bodied frugivores, folivores and browsers in the medium range and maximum diversity among grazers of either degree of specialization and either size category. Misclassifications occur with groups D and E, both of them geographically neighboring. The climate regime is characterized by high mean annual temperatures, temperature seasonality in the medium range, medium precipitation with a pronounced rainy season in summer. The vegetation is comparatively dense, with canopy density and photosynthesis activity in a medium range. The habitats of type F communities are quite distinct. Misclassifications do not occur.
Classification of paleo-communities
This model is applied in our study in order to attribute fossil specialized herbivore communities to one of the 6 reference communities and to infer the range of habitats which are associated with the techno-complexes. For that purpose, the classification functions given in the first part of Tab.ESM-8 are applied. Ecological profiles for specialized herbivore taxa are assessed, the set of community related variables is determined and then the classification functions are applied to identify the community class represented by a particular fossil community.
The sample of assemblages we examine in our present study includes data on 69 fossil assemblages. We excluded those assemblages in which diversities were too low to provide a reliable signal (deco = 0.000) and restricted our study to those assemblages which are clearly associated with either one of the techno-complexes. The list of localities and the number of assemblages is stated in Tab. ESM-6. Note that several assemblages are associated with more than one techno-complex. Therefore, the total number of assemblages per techno-complex (n=73) exceeds the number of fossil assemblages (n=69).
References
Estes, R.D. (1991). The behavior guide to African mammals.Berkeley: University of California Press.
Fortelius, M., Solounias, N. (2000). Functional characterization of ungulate molars using the abrasion-attrition wear gradient: A new method for reconstructing paleodiets.American Museum Novitates, 3301, 1–36.
Gagnon, M., Chew, A.E. (2000). Dietary preferences in extant African bovidae. Journal of Mammalogy, 81, 490–511.
Hayek, L.-A.C.,Buzas, M.A. (2010). Surveying natural populations: Quantitative tools for assessing biodiversity,2nd edition.New York: Columbia University Press.
Kingdon, J. (1997). The Kingdonfield guide to African mammals.London: A & C Black.
Kraus, T. (2008). Ground-based validation of the MODIS leaf area index product for East African rain forest ecosystems. Ph.D.thesis, Erlangen: Universitätsbibliothek der Universität Erlangen-Nürnberg.
Magurran, A.E. (2011). Measuring biological diversity.Oxford: Blackwell.