We first would like to thank Clara Bolton and Camille Contoux for inviting us to publish this study. We would like to address a special thanks to Christine Steininger for all the discussions and debates which indirectly improved this paper. We thank the two anonymous reviewers whose comments greatly helped us to improve the manuscript.

1 - Introduction

1The transition between the Pliocene and the Pleistocene (2.58 Ma) is marked by the setting up of the glacial and interglacial cycle and the shift from warm-moist to cold-dry conditions (e.g. Zachos et al., 2001; DeMenocal, 2004, Maslin et al., 2015). This global climatic change resulted in an increase of arid environments and is often associated with a major faunal turnover (e.g. Coppens, 1975; Vrba, 1985, 1988, 1995; Behrensmeyer et al., 1997; Alemseged, 2003; deMenocal, 2004), and hominin evolution (e.g. Vrba, 1988, 1995; Potts, 1998; Bobe & Behrensmeyer, 2004; deMenocal, 2004; Behrensmeyer, 2006; Maslin & Christensen, 2007; Bonnefille, 2010; Levin, 2015; Maslin et al., 2015; Caley et al., 2018). Indeed, this period coincides with the emergence of two new hominin genera: Homo and Paranthropus. However, most of these relationships are based on the study of East African contexts (e.g. Potts, 1998; Bobe & Eck, 2001; Alemseged, 2003; deMenocal, 2004; Bobe, 2006; Trauth et al., 2009; Antón et al., 2014; Maslin et al., 2014). Despite the fact that previous comparisons between South and East African palaeontological records were made (Turner, 1988; Vrba, 1988; Turner & Wood, 1993a,b; Patterson et al., 2014), the geological context of South African caves has prohibited the use of absolute dating or geochemistry methods for a long time and the age of some deposits are still under debate (e.g. Berger et al., 2002; Clarke, 2002; Kramers et al., 2017; Stratford et al., 2017; Bruxelles et al., 2018). Thus, the comparison with the well chronological calibrated East African fossil record is challenging. However, the recent development of uranium-lead (U-Pb) and cosmogenic dating methods yielded absolute dates for speleothems and fossil enamel from South African deposits (e.g. Walker et al., 2006; Balter et al., 2008; de Ruiter et al., 2009; Granger et al., 2015; Pickering et al., 2018). Therefore, the South African faunas associated with hominin remains can now yield more accurate information about evolutionary patterns (e.g. Vrba, 1985, 1993; McKee, 1995), palaeoenvironments (e.g. Vrba, 1974a; Reed, 1997) or hominin evolution and subsistence behaviours (e.g. Brain, 1981; Pickering, 2006).

2The goal of this paper is to explore the evolution of the large mammal palaeobiodiversity and associations during the Late Pliocene and Early Pleistocene in South Africa. Indeed, the evolution of diversity patterns of large mammal communities can provide significant information concerning the ecological context of hominin evolution (Turner & Wood, 1993b; Du & Alemseged, 2018). We apply cluster and factor analyses in order to test if a shift occurred in the large mammal associations. Our analyses provide a framework to better understand the evolution of ecological structure of large mammal communities during the Late Pliocene and Early Pleistocene in South Africa.

2 - Material and methods

2.1 - Fossil deposits

3Our analyses included 25 Plio-Pleistocene deposits from South Africa (tab. 1, fig. 1). In order to accommodate the disparity of geological dates (tab. 2) and the absence of stratigraphic sequences, we decided to pool deposits into eight 0.5 Ma standard time-units (STU). Moreover, we excluded Sterkfontein member 4 and Stw53 Infill as well as Gladysvale deposits from all the analyses because there is no agreement about their geological ages. The fossils deposits of Bolt’s Farm were not considered here because we cannot be confident about their geological ages, the provenance of the material collected by Robert Broom or because they did not yield enough material for now (Gommery et al., 2014). Although the Minnaar’s Cave has been recently localized, inventory issues, the uncertainty about material provenance and geological age lead us to exclude the site from all the analyses (Gommery et al., 2012a). Finally, the layers 11 and 12 of Wonderwerk are not considered because they yielded only two genera, Eurygnathohippus sp. and Gazella sp. (Brink et al., 2016).

2.2 - Faunal data

4Based on literature, we recorded the large mammal fauna (greater than ~5 kg) of all the 25 fossil deposits in terms of presence and absence (tab. 3). We recorded 113 species within 63 genera of large mammals. All of them belong to 5 different mammalian orders (Primates, Carnivora, Cetartiodactyla, Perissodactyla, and Proboscidea) and 13 families. Following Turner and Wood (1993b), we included species identified as “cf.” and “aff.”. Furthermore, we took into account revised and synonymous taxa.

5Our knowledge of palaeobiodiversity can be driven by palaeontological bias rather than true diversity (Durham, 1967; Raup, 1972; Lyman, 1994; Smith, 2001; Smith, 2005; Smith & McGowan, 2007; Lloyd, 2012). Moreover, multiple taphonomical agents could be involved in the accumulation processes of cave fossil assemblages (e.g. Brain, 1981; Pickering et al., 2004b; Hopley & Maslin, 2010; Bountalis & Kuhn, 2014; Denys & Patou-Mathis, 2014). For that reason, we decided to use the presence of species rather than its abundance. Indeed, it permits to reduce the taphonomical biases affecting the abundance of taxa (e.g. Behrensmeyer, 1991; Reed, 2008).

2.3 - Cluster analysis

6In order to explore the similarity between fossil deposits, we applied cluster analysis on the presence – absence matrix at species level. We first produced a distance matrix by using the function vegdist from the vegan package in R (R Development Core Team, 2015), with the Jaccard Index method, which is based on the number of species shared by two samples and the number of species unique to each of them (e.g. Legendre & Legendre, 1997; Chao et al., 2004; Belmaker & Hovers, 2011). Then, we applied the hclust function on the distance matrix with the average method (UPGMA).

2.4 - Factor analysis

7We performed factor analysis in order to identify characteristic associations of mammal taxa through time (Sepkoski, 1981; Figueirido et al., 2012; Morales et al., 2015). We first calculated the number of species by genera within each STU. Then, the fa function from the psych package in R software was applied. As recommended by Morales et al. (2015), we used the “varimax” rotation method to maximize the sum of variances of the square loadings. The number of factors was fixed at 3 according to the scree plot produced by the function nScree from the nFactors package. According to Sepkoski (1981), the scores of each variables (i.e. the number of species within genera) are measures of the relative importance or diversity of each taxon within the factor. Thus, genera with a score larger than or equal to 0.5 were selected, since they depict the main contribution to the taxonomic diversity within each factor (Sepkoski, 1981). These factors are interpreted as representing “great evolutionary faunas”, which are defined as associations of taxa sharing time of origination, diversification and extinction (Figueirido et al, 2012).

3 – Results

3.1 - Cluster analysis

8The cluster analysis shows an interesting pattern (fig. 2). Indeed, most of the deposits were grouped by localities (Makapansgat, Gondolin, Swartkrans, Sterkfontein). The Jacovec Cavern (4.02 Ma) and Sterkfontein Member 2 (3.67 Ma), the two oldest deposits, are close to each other. Interestingly, Makapansgat member 3 (3.32-2.9 Ma) and 4 (3.0-2.8 Ma), which are grouped together, are also linked to Taung (2.8-2.4 Ma) when Makapansgat member 2 (3.32-3.06 Ma) is grouped with Drimolen Makondo (2.61 Ma). Haasgat, Drimolen MQ, Malapa and Kromdraai B, which are all three dated around 2.0 (tab. 1), showed similarity based on presence-absence data. Sterkfontein member 5 Oldowan (2.18 ± 0.21 Ma) and Acheulean (1.3-1.1 Ma) are grouped together, linked to Motsetse (1.64‑1.0 Ma). Gondolin 1 (1.78 Ma) and 2 (1.8 Ma) are closed to each other. Our cluster analysis shows a group composed of Swartkrans member 1 to 3 (from 2.3 Ma for the member 1 to 0.96 Ma for member 3) which are close to Cooper’s D (1.3 Ma), Elandsfontein (1.0-0.6 Ma), Kromdraai A (1.8‑1.6 Ma) and member 2 (> 2.2 Ma), to a lesser extent. Finally, the Cornelia-Uitzoek (1.07-0.99 Ma) is isolated and seems to have a specific taxonomic composition.

3.2 - Factor analysis

9Through the factor analysis, we could identify three characteristic large mammal associations (LMA) through the Late Pliocene and the Early Pleistocene in South Africa (fig. 3) that explain 68 % of the original variance. Each of these evolutionary faunas are composed of specific association of the most diverse genera (tab. 4). We should be cautious because the first two STU’s are composed of only one deposit.

3.2.1 - Large Mammal Association I (STU 1-2)

10This evolutionary fauna, covering the two first STU from 4.5 to 3.5 Ma and showing a peak of diversity between 4.0 and 3.5 Ma, seems to be characterized by a high diversity of carnivore taxa (tab. 4). Seven genera contribute to this LMA I (scores > 1). The most characteristic genus appears to be Chasmaporthetes. Indeed, it is represented by two species at that time, C. nitidula and C. silberbergi, which both occur at Jacovec Cavern and Sterkfontein Member 2 (Berger et al., 2002; Kibii, 2004). The genus Parapapio, and more specifically the species P. jonesi, is known from both deposits, while P. broomi occurs only at Jacovec Cavern. The presence of Acinonyx, Megantereon and Dinofelis as major contributing taxa are more questionable. Indeed, only one species of each of these genera occur at both deposits. This could be a result of a general low diversity through the assemblages (tab. 2) and/or taphonomical biases (see Discussion). The LMA I declines immediately after 3.5 Ma but continues to maintain low factor loadings until after 1.0 Ma (fig. 3). This indicates the persistence of genera through the next two LMA, with a minor factor loading increase during the STU 4, which could be due (see below), to the important contribution of Parapapio.

3.2.2 - Large Mammal Association II (STU 3-4)

11The second evolutionary fauna corresponds to the STU 3 and 4, from 3.5 to 2.5 Ma. It is characterised by the contribution of 20 genera (scores > 0.5). During that period of time, the pick of diversity of the genus Parapapio is the highest, specifically at Makapansgat, by the simultaneous presence of three different species, P. jonesi, P. broomi and P. whitei. Also, even if the genus Canis is already known at Jacovec Cavern (Kibii, 2004), the diversity increases during the LMA II, with the presence of three different species, C. brevirostris, C. adustus and the extant C. mesomelas. Finally, the genus Tragelaphus, which is absent from the two previous STU, is represent in this LMA by T. pricei and T. angasi at Makapansgat. The LMA II declines after the STU 3 and maintains low factor loadings until just after the STU 7. It means, like the LMA I, that there is a persistence of genera through the next evolutionary fauna.

3.2.3 - Large Mammal Association III (STU 5-8)

12The LMA III is the longest in time evolutionary fauna identified by the factor analysis, covering the STU 5 to 8, between 2.5 to 0.5 Ma (fig. 3). It is characterized by a major contribution of Tragelaphus and Antidorcas. Indeed, even if Tragelaphus is already known before, it reaches his peak of diversity during the LMA III by the presence of four different species during that time, T. oryx, T. strepsiceros, T. scriptus and T. angasi. This is also the case for Antidorcas, with the presence of A. recki, A. bondi and A. marsupialis between the STU 5 and 8, and also A. australis during the STU 8. Moreover, the genus Papio becomes a major component of the association, by the presence of at least four different species and/or sub-species during the STU 5 and 6, P. hamadryas robinsoni, P. hamadryas angusticeps, P. hamadryas and P. indens. The genus Metridiocheorus, which appears during the STU 5, also contributes to this LMA. Even if the genus Equus is recognized during the STU 4 at Drimolen MQ (Adams et al., 2016), it becomes more diversified after 2.5 Ma, by the presence of three different species, E. quagga, E. capensis and E. burchelli.

4 - Discussion

13The cluster analysis permitted to identify similar taxonomic composition between deposits based on species presence – absence. The Jacovec Cavern (4.02 Ma) and Sterkfontein Member 2 (3.67 Ma) are close, sharing a significant number of species, such as Papio izodi, Parapapio jonesi, Panthera pardus, Chasmaporthetes nitidula and Chasmaporthetes silberbergi. Kibii (2004), based on the composition of the faunal assemblage, suggests a mosaic of open and closed habitats for the Jacovec Cavern, which include riverine gallery forest, bushland and open environments. The same conclusions were obtained from the analysis of the Sterkfontein member 2 faunal assemblage (Pickering et al., 2004). Our cluster analysis supports the similarity between the two assemblages.

14One cluster is composed of Makapansgat members 3-4 and Taung. All the members of Makapansgat are comparable in terms of large mammal composition, notably by the presence of Theropithecus darti, Cercocebus, Phacochoerus shawi and Gazella vanhoepeni. However, Makapansgat member 2 is grouped with Drimolen Makondo, although they share only two taxa, Dinofelis sp. and Cercopithecoides sp. Even if the Drimolen Makondo (2.61 Ma) deposit could not provide useful information on palaeoecology until now, the results of our cluster analysis and the presence of Chasmapothetes ? nitidula, Cercopithecoides and Dinofelis sp. (Rovinsky et al., 2015), indicates a possible similarity with the Makapansgat member 2 faunal assemblage composition and, by extension, palaeoecological reconstruction. Reed (1996, 1997, 1998) proposed an environmental reconstruction for Makapansgat members 2-4 from closed woodland to shrubland/bushland or edaphic grassland. Palaeoecological studies of the Taung (2.8-2.4 Ma) deposit indicate a woodland or forest environment (Williams & Patterson, 2010).

15The Drimolen MQ (2.664 ± 0.392 - 1.789 ± 0.104 Ma) faunal assemblage has been interpreted as a mixed open-to-closed environments (Adams et al., 2016). Our cluster analysis indicates a close similarity between the Haasgat (2.3-1.95 Ma) and Drimolen MQ assemblages, sharing taxa such as Cercopithecoides williamsi, Dinofelis sp., Megalotragus sp., Connochaetes sp., Damaliscus sp., Tragelaphus sp., Oreotragus sp. and Equus quagga. Moreover, the position of Malapa (2.067 ± 0.161 Ma) and Kromdraai B (2.0-1.8 Ma), close to the Drimolen MQ-Haasgat cluster, could indicate a similar non-specific palaeoenvironment as it has been suggested for Drimolen MQ.

16The Sterkfontein member 5 Oldowan (2.18 ± 0.21 Ma) and Acheulean (1.3-1.1 Ma) faunal assemblages point out a more drier and open environments, but with still a moist and cover component, than older deposits (Kuman & Clarke, 2000), with shared taxa such as Antidorcas sp., Equus sp. and Damaliscus sp. In contrast to Kuman and Clarke’ study, we do not consider Panthera leo, which is present in both assemblages, as a grassland species since it can occur in open habitat (less than 20 % canopy cover) to mixed habitat (approximately 20 % canopy cover) (Lewis, 1997). No palaeoenvironmental interpretation has been provided until now for the Motsetse (1.64-1.0 Ma) assemblage. However, according to our cluster analysis, it seems that Motsetse is similar to Sterkfontein member 5 Oldowan and Acheulean.

17All the taxa (Antidorcas recki, Redunca sp., O. oreotragus and Equus sp.) recognized at Gondolin 1 (1.78 Ma) are present at Gondolin 2 (1.8 Ma), which are close to each other. According to his taphonomical analysis, Adams (2012b) identified the leopard (Panthera pardus or leopard-like felids) as the main accumulator of GD 2 and then postulated that the habitat surrounding Gondolin site should support this predator as well as their preys. Based on this assumption, Gondolin faunal assemblage indicates a dominance of open to wooded grassland with patches of rocky hillsides combined with seasonal to continuous water in valley floors.

18It is not surprising to observe a resemblance between the faunal assemblages of all Swartkrans members (from 2.3 Ma for the member 1 to 0.96 Ma for member 3). Indeed, they have a lot of taxa in common (de Ruiter, 2003) with a dominance of grassland component. However, taxa such as Hippopotamus (members 1-3), Elephas (members 1 and 3), Sivatherium (member 2), Kobus cf. leche (member 2 and 3) or the Redunca cf. arundinum (member 1) suggest the presence of a permanent source of water and an extensive woodlands surrounding the site. The majority of the Cooper’s D (< 1.375 Ma) faunal assemblage, which seems similar to Swartkrans according to our cluster analysis, has been interpreted as indicative of grassland environment with wooded component and a permanent water source (de Ruiter et al., 2009). The Elandsfontein (1.0-0.6 Ma) faunal assemblage indicates a more humid habitat, samples an interval of great vegetational productivity causes by an increase of moisture according to Klein et al. (2007). Our cluster analysis shows that the Kromdraai A assemblage (1.8-1.6 Ma) is close to Elandsfontein and by extension to Cooper’s D and Swartkrans. This illustrates a similar faunal composition between these sites.

19Finally, the Cornelia-Uitzoek (1.07-0.99 Ma) seems to be the most specific faunal assemblage compared to other deposits. Indeed, it has yielded a unique combination of taxa such as Aepyceros helmoedi, Megalotragus eucornutus, Metridiocheorus compactus, Kolpochoerus heseloni, Hippopotamus gorgops or Eyrygnathohippus cornelianus (Brink et al., 2012).

20The most interesting result of the cluster analysis was probably the distinction between two clusters. The first one is composed of all the deposits older than 2.4 Ma if the age interval of Taung is confirmed. The second one includes all the assemblages younger than 2.6 Ma, which corresponds to the lower limit of the Drimolen MQ deposit. Based on this observation, it seems that the cluster analysis supports the hypothesis of a faunal turnover between 2.61 and 2.4 Ma in South Africa. The only exception is the group formed by Makapansgat member 2 and Drimolen Makondo. Subsequent analyses will be necessary in order to test our interpretation. The factor analysis allows us to identify three characteristic associations of large mammal fauna between 4.5 and 0.5 Ma in South Africa (tab. 4, fig. 3 & 4).

21Only Carnivores (Chasmaportethes, Acinonyx, Panthera, Dinofelis and Megantereon) and Primates (Parapapio and Papio) taxa have the highest scores (> 2) within the LMA I (from 4.5 to 3.5 Ma). However, we should be cautious about the interpretation of this result. Indeed, as we mentioned before, this evolutionary fauna covers the two first STU which are both composed of only one deposit, the Jacovec Cavern (STU 1) and the Sterkfontein member 2 (STU 2). According to Kibii (2004), the Jacovec Cavern primate assemblage corresponds to a leopard refuse assemblage. The Sterkfontein member 2 was interpreted as a death-trap characterized by the presence of several carnivore species (Pickering et al., 2004). Thus, the pattern observed through the LMA I could be due to taphonomical biases in these two deposits rather than representing a palaeobiodiversity signal.

22The second LMA (from 3.5 to 2.5 Ma) is mainly composed of a mixed-environment signal by the presence of both woodland and grassland adapted taxa such as Tragelaphus, Oreotragus, Cercopithecoides, Syncerus, Notochoerus and Giraffa for the first category and Parapapio, Gazella, Theropithecus, Phacochoerus and Ceratotherium for the second.

23If we look at the composition of the LMA III, we notice the presence of Antidorcas, Tragelaphus, Metridiochoerus, Equus, Connochaetes, Hippotragus, Damaliscus, Redunca, Raphicerus and Cercopithecoides. Although most of them refer to open-grassland adapted taxa (Alcelaphini, Antilopini, Metridiochoerus and Equus), wetland genera (Reduncini and Hippotragini) and woodland (Tragelaphus, Raphicerus and Cercopithecoides) tribes are also present. The LMA III, which differs from previous LMA, could be a response of South African large mammals to a climatic shift around 2.5 Ma.

24If we look more closely at the large mammal associations, the Primate association is marked by the presence of Parapapio and Papio as high scoring taxa during the LMA I, follow by Parapapio, Cercopithecoides, Papio and Theropithecus in the LMA II. Then, the Primate association of the LMA III is only composed of Papio. Thus, Primates seem to be more diverse between 3.5 and 2.5 Ma. Also, after 2.5 Ma, we observe a decrease of the diversity inside the Cercopithecoides when the genus Papio become a more important component of the Primate association. This could indicate an opening of the environment.

25As for the carnivores it was noticed, that the genus Canis migrated from Eurasia and first appeared in Africa ca. 2.5 Ma, and that this taxon is a good indicator of the environment opening (Turner, 1995). However, in South Africa, we highlight the presence of this genus since at least at 4.0 Ma in Jacovec Cavern (Kibii, 2004), but also at 3.0 Ma in Makapansgat member 3 and member 4 (Reed, 1996). This can be consistent with the hypothesis of a long and gradual opening of the environment since 4 Ma (Reed, 1997). As pointed by Turner (1995), the success of felids (and notably pantherines) shows the changing to open landscape but their emergence in the fossil record are not correlated to any particular climate event. Moreover, we can notice that scores of machairodonts decrease between LMA I and III. It is also emphasized by the progressive decline of the LMA I through time, especially after the STU 7 (1.5-1.0 Ma). This is consistent with the extinction of machairodont species after 1.5 Ma (Turner, 1988, 1990), which is linked to the opening of the landscape, thus making the prey more difficult to catch (Turner, 1990). Concerning the hyaenids, the Pachycrocuta genus seems to have occupied a wide variety of environments (Turner, 1990). Its presence in the LMA III, even if it is relatively minor (score > 0.5), supports the hypothesis of a mixed wooded-open habitat during that time. As it was recognized before, South African caves display a high diversity of carnivore taxa, notably because most of the deposits have been interpreted as carnivore dens (Brain, 1981; Cruz-Uribe, 1991; Pickering, 2002). This could explain the huge frequency of carnivore genera as characteristics of large mammal associations through time, especially for the first LMA as it was mentioned before.

26Bovids are recognized to be representative of their environments and have therefore long been used in palaeoecology (Vrba, 1974a,b, 1975; Greenacre and Vrba, 1984; Alemseged, 2003). The first association (LMA I) does not display score larger than 0.5 for any bovid genera. However, we can notice that the second LMA is marked by more “wet-closed” adapted genera such as Tragelaphus, Oreotragus, Syncerus and Redunca but with Gazella as grassland adapted genus. After 2.5 Ma, during the LMA III, we observe that the bovid composition is more characterized by a mix of open environment and “wet-closed” adapted species such as Antidorcas, Connochaetes and Damaliscus for the first category and Tragelaphus, Redunca and Hippotragus for the second.

27As for the Suids, the presence of Notochoerus and Phacochoerus in the LMA II (score > 0.5) suggests the presence of some open area in the environment (Harris & Cerling, 2002). During the LMA III, the rise of Metriodiochoerus genus (score > 1) which is more adapted to abrasive diet, tends to indicate the opening of the landscape (Harris & Cerling, 2002).

28Therefore, we can identify a more woodland adapted fauna before 2.5 Ma (Parapapio, Tragelaphus, Syncerus, Oreotragus, and Cercopithecoides) followed by a more pronounced ‘open-grassland’ type of habitat with the indication of wooded components and the presence of a permanent water source. Indeed, grassland taxa are represented by the presence of Antidorcas, Metridiochoerus, Equus, Connochaetes and Damaliscus although Tragelaphus, Raphicerus, Hippotragus and Redunca are more representative of the humid-closed type of habitat.

29The identification of a “mosaic” habitat after 2.5 Ma requires additional comments. According to Reynolds et al. (2015), the concept of a heterogeneous landscapes is poorly defined and suffers from the lack of focus on temporal and spatial scale. Moreover, Hopley and Maslin (2010) introduced to concept of “climate-averaging” which is defined as the “amalgamation of individuals living under one or more climatic states within a single faunal deposit”. This is a crucial statement which highlights the artificial increase of biodiversity of broad range ecomorphologies and diets within the same faunal assemblage. It results from a poor definition of the duration of bone accumulation. In order to resolve this problem, Hopley and Maslin (2010) proposed to compare palaeoclimatic data set which is independent of faunal assemblages.

30Based on this statement, it has been previously observed that the proportion of C4 grasses seems to drastically increase at about 1.7 Ma in South Africa (Luyt & Lee-Thorp, 2003; Hopley et al., 2007), and reflecting drier environmental conditions and a more open habitat after that time. Our factor analysis, supported by the cluster analysis, place this shift between the STU 4 and 5 (3.0 and 2.0 Ma) by the presence of Equus, and more Alcelaphini-Antilopini bovid composition as well as large terrestrial primates and Metridiochoerus as characteristic of the third large mammal association after 2.5 Ma.

5 - Conclusion

31Our results confirm that a shift occurred in term of variability of biodiversity and large mammal associations during the Late Pliocene and Early Pleistocene. More interestingly, our results allow us to support the hypothesis of a faunal turnover identify between at least 2.61 and 2.4 Ma based on the cluster analysis. Based on the factor analysis, three evolutionary fauna could be identified between 4.5 and 0.5 Ma in South Africa. The first one, between 4.5 and 3.5 Ma, is characterized by high contribution of carnivore genera, probably driven by taphonomical biases. It is followed by faunal assemblages, between 3.5 and 2.5 Ma, which indicate a mosaic of open and closed environments. Finally, the third evolutionary fauna assemblage, between 2.5 and 0.5 Ma, corresponds to a “mosaic” type of habitat, a mixed of ‘open-grassland’ and woodland components with the presence of a permanent water source.

32The concept of “mosaic” habitat remains loosely defined and could be a consequence of the lack of focus on temporal and spatial scale (Reynolds et al., 2015). Therefore, future investigation is necessary in order to identify the probable climate-averaging of South African Late Pliocene and Early Pleistocene deposits. That should yield an improved framework for a better understanding of the evolutionary dynamics which driven large mammal communities during the Late Pliocene and Early Pleistocene and through the Plio-Pleistocene transition in South Africa.