Speleology and Magnetobiostratigraphic Chronology of the GD 2 Locality of the Gondolin Hominin-Bearing Paleocave Deposits, North West Province, South Africa

Journal of Human Evolution 51 (2006) 617e631

Speleology and magnetobiostratigraphic chronology of the GD 2 locality of the Gondolin hominin-bearing paleocave deposits, North West Province, South Africa

Andy I.R. Herries a,e,*, Justin W. Adams b,c, Kevin L. Kuykendall d,c, John Shaw e

a Palaeoanthropology Research Group, Department of Anatomy, School of Medical Sciences, University of New South Wales, Kensington 2052, Sydney, Australia b Department of Anthropology, Washington University, Campus Box 1114, One Brookings Drive, St. Louis, MO 63130, USA c School of Anatomical Sciences, University of the Witwatersrand, Johannesburg, South Africa d Department of Archaeology, University of Shefﬁeld, S1 4ET, UK e Geomagnetism Laboratory, School of Archaeology, Classics and Egyptology, Oliver Lodge, University of Liverpool, L69 7ZE, UK Received 6 October 2005; accepted 7 July 2006

Abstract

Speleological, paleomagnetic, mineral magnetic, and biochronological analyses have been undertaken at the Gondolin hominin-bearing paleocave, North West Province, South Africa. Two fossiliferous but stratigraphically separate sequences, GD2 and GD1/3, which were once part of a large cavern system, have been identified. Although some comparative paleomagnetic samples were taken from the GD 1, 3, and 4 localities that are currently under investigation, the research presented here focuses on the fossil-rich, in situ deposits at locality GD 2, excavated by E.S. Vrba in 1979. The GD 2 deposits are dominated by normal-polarity calcified clastic deposits that are sandwiched between clastic-free flowstone speleothems. The lower flowstone has a sharp contact with the red siltstone deposits and is of reversed polarity. The capping flowstone shows a change from normal to reversed polarity, thereby preserving a polarity reversal. While the paleomagnetic work indicates that the GD 2 fossil material was deposited during a normal-polarity period, the shortness of the sequence made matching of the magnetostratigraphy to the geomagnetic polarity time scale (GPTS) impossible without the aid of biochronology. While lacking multiple time-sensitive taxa, the recovery of specimens attributable to Stage III Metridiochoerus andrewsi is consistent with a deposition date between 1.9 and 1.5 Ma. A comparison of the magnetostratigraphy with the GPTS therefore suggests that the fauna-bearing siltstone of GD 2 date to the Olduvai normal-polarity event, which occurred between 1.95 and 1.78 Ma, and that the reversal from normal to reversed polarity identified in the capping flowstone dates to 1.78 Ma. The main faunal layers therefore date to slightly older than 1.78 Ma. Deposits from the GD 1 locality are dominated by reversed directions of magnetization, which show that this deposit is not of the same age as the faunal layers from the GD 2 locality. Ó 2006 Elsevier Ltd. All rights reserved.

Keywords: Early Pleistocene; Paleomagnetism; Biochronology; Olduvai Event; Dolomite Paleocave; Speleology; Australopithecus

Introduction deposits from existing open chambers). The site is situated 34 km northwest of Johannesburg and 20 km from the Sterk- The fossil site known as Gondolin is a historically lime- fontein Valley, near the town of Broederstroom, in the North mined relict cave deposit (or ‘‘paleocave,’’ to distinguish the West Province of South Africa (Fig. 1). The site lies in the transition zone between the Mixed Bushveld and Rocky Highveld grassland biomes on the rocky slopes of the Skurweberg moun- * Corresponding author. Palaeoanthropology Research Group, Department tain range (Low and Robelo, 1996). A number of depositional of Anatomy, School of Medical Sciences, University of New South Wales, Kensington 2052, Sydney, Australia. units can be identiﬁed at the site. However, mining activities at E-mail addresses: [email protected], [email protected] the turn of the twentieth century essentially removed the entire (A.I.R. Herries). center of the paleocave deposit, making stratigraphic

Fig. 1. Location of Gondolin in relation to other hominin-bearing paleocave deposits.

correlations across the site very difficult. Further excavation Watson (1993) interpreted the excavated GD 2 fauna as the re- around the edges of the initial opencast excavations to remove mains of a carnivore accumulation, with the suid fauna sug- speleothem deposits has resulted in the two main fossil-bearing gesting that the deposits had formed around the same time localities (GD 1 and GD 2; Figs. 2, 3) that are now stratigraph- as those of Swartkrans Member 1, dated by faunal correlation ically discontinuous. (de Ruiter, 2003) and ESR (Curnoe et al., 2001) to approxi- As a result of this heavy mining, the Gondolin site today mately 1.6 Ma. preserves only remnants of the paleocave deposit, consisting Locality GD 3 represents an infilled rift cavity on the west- of thin coverings of heavily calcified, fine-grained sediments ern edge of the site and grades up into the GD 1 deposits. Lo- (siltstones), breccias, conglomerates, and abundant speleo- cality GD 4 represents the basal deposits exposed in the center thems on the walls of the ancient cave (Fig. 3). Locality GD of the site and can be traced to the northern wall, where the 1 is located in the northwest corner of the cave and represents deposits lie below GD 1. a series of interstratified speleothem, in-washed sediments, Interspersed with the in situ sequences are extensive ex situ and talus deposits covering a variety of time periods related breccia dumpsites that were produced during the initial mining to a vertical entrance to the system. Fossil materials from at the site. One of these breccia dumps was sampled in 1997 GD 1 excavated in 2003 have been described by Adams and yielded the first hominin material from the sitedan iso- (2006), and research on these deposits is currently ongoing. lated left M1 or M2 that is likely to represent the genus On the northern edge of the cavity, huge blocks of cave infill Homo and an unusually large left M2 attributed to the genus deposits have detached from the 6-m-high wall. Paranthropus (Kuykendall and Conroy, 1999; Menter et al., Locality GD 2 (Figs. 2, 4) is located on the eastern edge of 1999). The Paranthropus M2 exhibits morphological features the cave and consists of a stratified speleothem and siltstone unique for South African representatives of the genus, but it sequence containing dense fossil accumulations from caching is metrically more similar to Australopithecus (Paranthropus) activity via a lateral entrance. These deposits are now com- boisei specimens from eastern Africa (Kuykendall and Conroy, pletely separate from the other localities and no direct strati- 1999; Tobias, 2000). graphic links can be made. Initial excavations by E.S. Vrba Few absolute dating methods optimally cover the time range and D. Panagos in 1979 removed in situ fossils from the GD of the South African hominin paleocaves and consequently nu- 2 deposits that were initially described by Watson (1993; merous problems occur in their application. Paleomagnetic Fig. 4). In her preliminary faunal and taphonomic analysis, (McFadden, 1980), electron spin resonance (ESR; Curnoe A.I.R. Herries et al. / Journal of Human Evolution 51 (2006) 617e631 619

Fig. 2. Survey of the Gondolin site and location of localities (redrawn by Herries after Menter et al., 1999). et al., 2001), and isotopic dating (Partridge et al., 2003) have all grain sizes give similar results for certain standard mineral been attempted at other South African early hominin sites, with magnetic tests but can be distinguished by a more detailed mixed results. Previous paleomagnetic analysis of South Afri- analysis. In many cases, the siltstones and breccias are domi- can cave deposits has suffered from a number of problems. nated by secondary magnetizations carried by these ultrafine Work undertaken at Sterkfontein and Swartkrans by Jones grains. This is a function of their age, as well as depositional et al. (1986) suggested that the cave breccias were dominated mechanisms and sediment sourcing from the surrounding en- by large detrital grains of magnetite and were unsuitable for pa- vironment (Herries, 2003). New techniques and more sensitive leomagnetic analysis due to environmental conditions of depo- equipment have made it possible to identify a weak primary sition. More recently, however, positive paleomagnetic results remanence carried by magnetic grain sizes that are stable have been achieved for speleothem deposits at Sterkfontein over these geologic time periods (Thackeray et al., 2002; Her- (Partridge et al., 1999, 2000; Herries, 2003) and from ries, 2003; Herries et al., 2006). calcified siltstones, conglomerates, breccias, and speleothem A primary task at even the longer-studied South African deposits at Makapansgat (Herries, 2003; Herries et al., 2006), early hominin sites is to develop speleogenetic and develop- Gladysvale (Lacruz et al., 2002; Herries, 2003) and Kromdraai mental models with firmly established depositional sequences (Thackeray et al., 2002). on which a magnetostratigraphy can be based. A lack of such Detailed mineral magnetic analysis at these localities models can lead to confusion over the age and sequencing of (Herries, 2003) has shown that the main problem for the reco- deposits and fossils, and mistrust in well-established and valid very of stable, primary paleomagnetic directions is in most techniques. The study of the caves, including an interpretation cases not due to the presence of large detrital grains of magne- of the speleogenesis, developmental history, and infill, is there- tite, as suggested by Jones et al. (1986), but rather by fine, vis- fore paramount for understanding depositional sequences at cous grains that do not hold a stable remanence over the time the sites and in constructing a reliable composite sequence period since deposition of the deposits. These two distinct on which magnetostratigraphic analysis can be based (e.g.,

Fig. 3. Panoramic photo of the Gondolin paleocave looking north and showing the localities (GD 1e4) and topography of the site. 620 A.I.R. Herries et al. / Journal of Human Evolution 51 (2006) 617e631

Gondolin deposits in order to document formation and development of the cave system and to construct as continuous a stratigraphic sequence as possible; (2) to conduct paleomagnetic analysis of all recognizable layers in an attempt to identify the polarity of the various deposits and assess contem- poraneity in deposition and possible age from a comparison of polarity transitions on-site with the GPTS; and (3) to consider the biochronological signiﬁcance of in situ excavated fauna to aid in the dating of the site.

Speleology

Speleogenetic origin

The original description of the formation and development of Gondolin was made by Menter et al. (1999), based on Fig. 4. Photo of the in situ GD 2 faunal deposits showing the major stratigraphic Brain’s (1958) developmental model for low-topographic do- phases. lomite cavern development. However, the slightly mountain- ous region in which Gondolin lies caused different structural disposition than that which occurred at the Sterkfontein valley Latham et al., 1999, 2003). It is also important to understand cave site. This model is therefore unlikely to fully explain the more completely the potential sources of the deposits and development of the Gondolin paleocave deposits, which dis- therefore of the remanence-carrying magnetic minerals that play many features that argue for a much more complex spe- occur in the different types of deposits. Such analysis is aided leogenetic origin, development, and infill. by detailed studies of fossils excavated from the various dis- The Gondolin site is the remains of a small cave system that tinct in situ paleocave deposits. Much of the confusion over formed in the Precambrian dolomite of the Eccles Formation, the age of the fossil deposits and sites comes from examina- Malmani Subgroup, of the Chuniespoort Group carbonate-Bif tion of ex situ fauna that has come from a variety of deposits Marine Platform. The dolomite represents a heavily altered, of different ages. The paleocave deposits do not record just shallow mud-flat deposit consisting of well-bedded, recrystal- one moment in time, but a series of geological events covering lized, gray-blue magnesian limestone or dolomite. The dolo- the life history of the karstic system, from the first speleoge- mite formed on the edge of, and later covered, the Kaapvaal netic inception to modification, infill, erosion, and, in almost Craton (remnant of an ancient continent) around 2.6 billion every case, multiple reuse of the karstic aquifers by more re- years ago. Numerous intercalated bands of chert occur and cent groundwater. Deposits can often be identified from all the sequence is interbedded with shale formed during marine stages throughout this cycle. transgressions (Meyer and Robb, 1996). The sequence is cap- In a magnetic polarity analysis of any stratigraphic section ped by a banded iron formation, the basal part of which is for chronology, it would be most desirable to obtain a reversal characterized by siderite-rich, microbanded iron formation sequence that was sufficiently long and uniform that it could with minor magnetite and some hematite-containing units be matched reversal for reversal with the geomagnetic polarity (Beukes et al., 1990). time scale (GPTS; Ogg and Smith, 2004). Ideally, this match After deposition of the Transvaal Supergroup banded-iron should be made independently of arguments from fauna or formation and the retreat of the Precambrian sea, a major other dating methods; however, a comparison of the defined weathering phase occurred, with primary karstification, fault- paleomagnetic ages with other dating methods is paramount ing, and uplift of the dolomite before its burial under the rocks in confirming and refining the different dating methods used. of the Pretoria Group. In later phases, heavy mineralization At smaller cave sites that have only a short stratigraphy, occurred related to volcanic intrusives and hydrothermal activ- such as Gondolin, ascribing an age based purely on limited po- ity. Ancient mineral veins and fault cavities that formed during larity reversals, if any, can be difficult if not impossible. In this these periods act as preferential routes for water flow through case faunal dating based on in situ fossils can be used to create the dolomite and are extremely important for cave formation a broad age range for the site and act as a useful guide to fit- and development in the Gondolin region during later periods ting magnetostratigraphies to the GPTS. Such approaches have of karstification. been relatively successful at other South African sites, such as This major weathering phase also caused the formation of Buffalo Cave at Makapansgat (Herries et al., 2006). Where in a chert-lag breccia. Waste rubble from this and later phases situ fauna has not been recovered or is not temporally diagnos- of weathering still occur on the slopes and hilltops in the tic, problems with interpretation will persist in the absence of Gondolin region. This is the result of the very high proportion a continuous, long-scale sequence. of insoluble material weathered out of the Eccles Formation The aims of the research reported in this paper were: (1) to dolomite, which is characterized by thin but high-density chert conduct speleogenetic and stratigraphic analysis of the banding. This, in turn, caused heavy choking of the alluviated A.I.R. Herries et al. / Journal of Human Evolution 51 (2006) 617e631 621 karst valleys and developing cave systems by colluvial sedi- much later periods. This is supported by rapid changes in ments formed by weathering of waste rubble on the hillslopes the magnetic polarity of such deposits. Such stratigraphic and subsequent pedogenesis. The Gondolin area is also associ- complexities are not evident in the GD 2 deposit, where a con- ated with chert-breccia-filled dykes and syenite dykes. These tinuous and undisturbed stratigraphic sequence occurs. features provide the potential for speleogenetic routes and Wherever they are formed in the world, most limestone groundwater flow at dolomite boundaries, where the interface caves experience a life history that consists of a subaqueous between the deposits would have acted as primary areas for (phreatic) phase, a partially air-filled (vadose) phase, phases dissolution. Many caves form due to the close association of of sedimentary infilling and breakdown, and a phase of unroof- more or less resistant material often forming at the interface ing and erosion (see Ford and Williams, 1989). Some of these of different rock types. Water moving from one rock type to phases overlap and details may vary even within a single cave another or between layers of dolomite and limestone with dif- system. At many hominin paleocave sites, this sequence has ferent chemical composition will also cause changes in water occurred at least twice, with water levels rising and falling chemistry, causing increased acidity and solution. In the east- a number of times. At the Makapansgat Limeworks, layers ern Drakensberg, caves often develop at the interface between of subaqueous and aerial speleothems alternate, showing that more resistant stratigraphic layers, such as between over- and water levels in the cave fluctuated significantly during its early underlying quartzite layers. Chert layers within the dolomite life. At Gladysvale, a modern cave has formed within the more would also have acted as channels of water flow and dissolu- ancient paleocave deposit. tion. Shale bands, on the other hand, can provide both routes A true water table does not occur in limestone. Where there for increased cave development through the washing out of is a high degree of connection between various cavities and shale beds and the development of karstic routes along bed- pore spaces (e.g., at Sterkfontein and Swartkrans), the pie- ding planes and as barriers to water flow. zeometric surfaces (level of water) may be thought of as ap- Certain features suggest a partially hydrothermal origin for proximating a water table. In higher-relief areas (e.g., at the cave and the occurrence of a mineral spring. Hydrothermal Makapansgat), streams sink at altitude to emerge at springs activity would enhance development of caves in dolomite not or resurgences in an adjacent valley and the piezometric sur- only by condensation-corrosion and replacement-solution, but face can be highly variable, approximating an S-shaped curve also by providing routes through otherwise insoluble chert from hilltop to valley bottom. Gondolin lies in hilly terrain banding. Hydrothermal karst has an important control on the (Fig. 3), forming the northeastern edge of the ‘‘West Rand’’ formation of ore bodies (Ford and Williams, 1989), which Malmani dolomite outcrop; Sterkfontein, Swartkrans, and are numerous in the surrounding area. It is possible that the Kromdraai lie on the southern edge (Fig. 1), before it dips cave may have been formed by meteoric epigene waters ex- beneath exposures of Pretoria Group shale (Silverton Forma- ploiting much older mineralized vein and fault cavities rather tion). The modern hydrology causes water to run from the than due to direct hydrothermal activity. Such a mineralized dolomite hills onto the overlying shale and quartzite plains. fault zone can be seen in the area of the GD 1 excavations, As such, the caves of this area most likely acted as small and the fossil-bearing paleocave may have started as a rift cut-off, input, and drain systems for water as it made its cave that formed along this zone. way through the dolomite hills and resurged onto the slate The fossil-bearing deposits exposed at Gondolin today rep- plain to the north. The caves here tend to form as small, shal- resent at least a second phase of karstification and infill in the low systems, which act as both inlets and outlets for meteoric region. This process of karstification has been ongoing since water flowing through the dolomite kopjes and, consequently, the formation of these caves. Cave formation and infill has have very small watersheds, with run-off coming from the hill been multigenerational, often at the same site. The various summits. During periodic flash floods in the summer months, South African early hominin deposits are thus ‘‘relict’’ karstic the site serves as a direct conduit for water flow from the hill- features in that the original context of deposition has been al- top to the valley floor below. The entire system is covered by tered by fossilization, erosion, and other geological processes. vertical shafts in which recent water has washed away loose, However, they remain exposed to, and are modified by, pro- decalcified deposits. Some of these have been filled with mod- cesses operating during successive and present geomorphic ern sediments where solution tubes (Fig. 4) and avens meet systems, with new karstic routes being developed and infilled and form a complex speleogenetic mix of calcified deposits over potentially the entire life history of the cave system. This of a variety of ages. In such circumstances, very detailed strati- makes their stratigraphy extremely complex; an excavation graphic analysis is paramount in understanding the deposits could well cut through a variety of deposits of very different and relative ages of fossils recovered from them. ages that are at the same stratigraphic level. In many cases, the law of superposition is invalid and layer-cake-like stratig- Cave development and infill raphies on a system-wide scale are the exception rather than the rule. This is well illustrated in the GD 1 excavations, where The specific geology of the Eccles dolomite, which is dif- multiple phases of infilling, undercutting, and reworking of ferent from both the Sterkfontein and Makapansgat areas, material have caused inverted sequences due to the presence along with the topography (higher than Sterkfontein, but not of thick flowstone layers that acted as false floors, under which as high as Makapansgat) have had a marked effect on both sediment and breccias were eroded out and redeposited at the speleogenetic origin and subsequent development of the 622 A.I.R. Herries et al. / Journal of Human Evolution 51 (2006) 617e631

Gondolin System in comparison to other sites. Such develop- a single vertical entrance in the area of GD 1 with a high- ment will determine the depositional sequences that occur, energy environment of deposition, as originally suggested by with different depositional regimes potentially active within Menter et al. (1999). the same system. Moreover, this will cause different tapho- Large amounts of speleothem in the central area suggest nomic processes to be active in different caves, or areas within a wet depositional environment in which water was trickling one cave system, with important implications for the discovery in from the ceiling or down walls and running through basal- and understanding of fossil assemblages. collapse rubble and cementing it together. Today the cave Because of the topography and erosion of the hillside, only would have only a small catchment area, but there are numer- a shallow paleocave deposit was exposed at Gondolin, out- ous dry stream valleys that would have formed under a wetter cropping over a wide area of hillside (Fig. 3). Consequently, climate than today. In the northeast, rounded pebbles show that lime mining took place as an opencast operation and almost water ran through this part of the cave, possibly from the completely removed the extensive paleodeposits, as evidenced northwestern GD 1 entrance to exit on its eastern edge, per- by the large quantities of breccia blocks removed during min- haps out of the GD 2 entrance. At GD 2, a pure flowstone layer ing and left in the modern surface dumps. Speleothem does not both starts and caps the fossil-rich clastic deposits and indi- appear to have made up such a vast proportion of this material cates that any stream running in this area was intermittent as at other sites, such as Makapansgat, or as visualized by and allowed occupation by carnivores during certain periods. Menter et al. (1999). This is evident from the large quantities The most recent phase of karstification is represented by the of calcified sediment blocks removed during the extraction formation of solution tubes within the dolomite and calcified process. Because of the highly altered nature of the locality, paleodeposits across the site. Many of these represent solution relationships cannot be identified between the eastern GD 2 tubes that have formed around tree roots; these are known lo- and western GD 1 localities. cally as makondos. These features cause decalcification of the The recently excavated GD 1 breccias in the northwestern paleodeposits and are contemporaneously infilled with sedi- corner of the system contain a high proportion of in-washed, ments from the current weathering regime. The soft deposits weathered chert-lag debris from the surface. As Menter that infill these features include fossils and represent a time- et al. (1999) noted, this area of the cave represents deposits transgressive mixture of material. It is thus very important to that were accumulated close to a vertical entrance. Fossil ma- excavate fossil material directly from paleodeposits if possi- terials from the GD 1 area are highly fragmentary and heavily ble, rather than from these features where mixing occurs. weathered, which suggests that they were exposed to the elements for a considerable time before being introduced into the Summary of Gondolin cave development: sequencing cave (Adams, 2006). The entrance would have been close to of events the crest of the hill and would have permitted at least intermittent water flow into the system. The presence of cave pearls in The modern dolomite caves of South Africa generally tend deposits below the GD 1 material suggests that a relatively wet to develop along joints and bedding planes. The Gondolin pa- but low-energy environment with small standing pools existed leocave may have developed initially as two independent cav- in this section of the cave. These deposits also contain small, ities that were later joined by collapse and increased solution. rounded pebbles, suggesting that a low-energy stream entered It seems likely that the GD 2 locality was partially separated the cave, possibly where water ran over the lip of the dolomite from the main cavity but remained part of the same hydrolog- rear wall and into the rift below. ical system. The western cavity (GD 1/3) developed from The depositional environment at GD 2 was notably differ- a small fault cavity, which developed along a mineralization ent. Fossil specimens are relatively complete, well-preserved, vein. This rift provided the path of least resistance through and exhibit only moderate amounts of weathering. The matrix the rock and was enlarged by percolating water and was later deposits are consistent with a different environment and are captured and enlarged by small phreatic conduits to produce dominated by red, fine-grained, calcified clastic deposits (silt- a major conduit for water drainage through the hillside. stone) with very few large chunks of dolomite or speleothem, With a lowering of the local piezometric surface, the cave suggesting that the material has been winnowed by water from was moved into the vadose zone of the rock. Percolating water a primary location such as an entrance-collapse breccia. There would then have enlarged the routes directly to the surface. is little evidence of allogenic stone in the deposits. A large Large-scale collapse then occurred due to lack of support number of immature individuals among the faunal remains, from water, as during the phreatic phase, and the cave would as well as partially articulated limbs and a large number of as- have expanded laterally into a large cavern. This area of the sociated skeletal elements, occur. The density of the accumu- dolomite has a very high degree of chert banding. High- lation of bone (Fig. 4), as well as other taphonomic indicators, frequency, thin chert banding would have caused instability af- suggests that the GD 2 assemblage was primarily accumulated ter dissolution of the interlayered dolomite and would have by leopards through denning and caching activities (Adams, resulted in a chert-rich, small-scale, block-breccia deposit. 2006). These animals would have had to have been able to Such deposits are noted at the base of the sedimentary se- access the cave, and so there was likely a lateral entrance to quence, where large chert blocks are cemented by speleothem. the cave on the valley side. This taphonomic and sedimento- The large amount of basal-collapse rubble stands as evidence logical information is inconsistent with there only being of this process of enlargement and collapse that filled the A.I.R. Herries et al. / Journal of Human Evolution 51 (2006) 617e631 623 central area of the cavern in the area of locality GD 4. This system can be employed, as used at most hominin paleocave basal collapse forms a breccia mass that constitutes the base sites (e.g. Partridge et al., 1999, 2000, 2003), and a temporal upon which subsequent sedimentation took place. Speleothem framework of deposition can therefore be established. Where from this deposit are formed from primary calcite and arago- stratigraphic relationships are uncertain and temporality of de- nite (rather than secondary recrystallization) in a very open, position is not established, a Member system is invalid and, in evaporative environment (Hopley, 2004). most cases, it simply denotes sedimentological type. In the GD 2 locality, a thick, pure basal flowstone was de- Given the above discussion, the GD 1/3 and GD 2 localities posited at the base of the sequence. Due to the sharp contacts are considered to represent paleocave fills from two separate evident between speleothem and overlying clastic deposits, entrances to the same cave system and thus are treated as sep- a significant time gap may have occurred. Paleomagnetic mea- arate magnetobiostratigraphic entities. At this time, we have surements show that the basal flowstones in different areas of concentrated analysis on the GD 2 locality. Paleomagnetic the cave were not formed at the same period. On top of this analysis was undertaken in the other areas to assess whether sloping flowstone, red silt was winnowed from entrance- the deposits could be from different periods, but these analyses collapse breccias, filling the edge of the cavity. This produced were not used as part of the magnetostratigraphic interpreta- mud floors onto which bones were deposited, forming the GD tion of GD 2. 2 fossil accumulation. A number of taphonomic factors, in- As the GD 2 sequence is very short, a Member sequence is cluding faunal biases, articulation and preservation of skeletal not utilized, and designation is made based on phases of differ- elements, and element modifications in this deposit, suggest ent sediment types. Three main phases can be identified at GD that the accumulation resulted from primary carnivore den- 2: (1) GD 2 Phase 1 represents the basal speleothem; (2) GD 2 ning. This could only have occurred when a lateral entrance Phase 2 represents the clastic layers that include the fossil de- to the cave was present. Red silt was continually deposited posits; (3) GD 2 Phase 3 represents the capping speleothem. It for some time, with increased input from larger clasts to was initially assumed that basal flowstone throughout the cave, form a more brecciated deposit in certain areas. Clastic depo- at GD 2, 3, and 4 (Fig. 2), would be of a similar age and rep- sition then stopped and pure speleothem formation again oc- resent the base of the entire sequence. A sample was taken curred. This filled in the small space remaining in the cavity from basal-block breccia deposit with layers of thick flowstone and caused the calcification of underlying deposits. Later in- at GD 4 (GD 4-01). This is considered to be the base of the trusive material accumulated as the result of the development sequence. Localized pure flowstone deposits at the base of of solution cavities. It was this relatively simple sequence of the GD 3 (GD 3-01) and GD 2 sequences (GD 2-01) were fossiliferous GD 2 deposits that was excavated by Vrba in also sampled to test these hypotheses. At GD 2 (Fig. 4), this 1979 and was sampled for paleomagnetic and faunal dating. basal flowstone (termed GD 2 Phase 1) is overlain with about Analysis of the stratigraphically more complex GD 1, GD 3, 4 m of fine-grained, light to dark red siltstone deposits (sam- and GD 4 deposits is ongoing. ples GD 2-02, 03, 04, 06) termed GD 2 Phase 2. The variation in color is mainly a factor of variable calcification. The middle Paleomagnetism part of this siltstone sequence contains extensive bone deposits from denning activity by carnivores and this is bound by sam- Sampling ples GD 2-03 and GD 2-04. Block sample GD 2-04 samples the interface between calcified siltstone and flowstone. Sub- Preliminary paleomagnetic analysis and full mineral mag- sample GD 2-04CS is a heavily calcified siltstone that was so- netic analysis for Gondolin was presented by Herries (2003). lidified by the first phase of calcite-rich waters filtering Subsequently, a further series of samples has been taken at through the deposit. Subsample GD 2-04FS is a partially con- the site and more subsamples processed from the blocks taminated flowstone that caps the siltstone sequence. This is used for Herries’ (2003) analysis. Within the GD 2 area, block followed by a capping speleothem (GD 2-05) that is termed samples were taken that represent all recognized units with GD 2 Phase 3. secure stratigraphic positions. For comparative purposes, a At the GD 3 locality, the basal speleothem (GD-01) is over- further series of samples was taken from small sections within lain with about 5 m of reddish brown, pebble-rich siltstone de- the GD 1 excavation area, at locality GD 3, and at the base of posits, breccias, and flowstones (GD 3 02-05, 07). Above this is the central collapse zone, GD4. Because of the highly altered a range of deposits whose stratigraphic relationship is as yet un- nature of the site from mining and subsequent collapse, rela- known. In certain areas, pink siltstone occurs with cave pearls tionships among the GD 1, GD 2, GD 3, and GD 4 localities (GD 3-06) that suggest a low-energy wet environment. Above are difficult to establish and the sequence of deposits in the this are the GD 1 deposits; a series of localized and time-trans- GD 1 and GD 3 area is difficult to interpret. Because the GD gressive breccia-fills occurs, which represents a vertical shaft 1 deposits are stratigraphically above those of GD 3, the law opening (GD 1), perhaps the source of the water that formed of superposition would suggest that they are younger. However, the pools in which the cave pearls occur. it seems likely that some of the GD 1 deposits are, in fact, older Samples were oriented in situ using a Suunto clinometer than those in GD 3. The sequence here does not represent and magnetic compass; subsequent corrections were made a layer-cake as at GD 2. In short sections where a layer-cake- for the declination of the local field according to the Interna- like type sequence can be established with certainty, a Member tional Geomagnetic Reference Field (IGRF) accessed through 624 A.I.R. Herries et al. / Journal of Human Evolution 51 (2006) 617e631 the British Geological Survey (http://www.geomag.bgs.ac.uk/ sample after the removal of any secondary magnetization) was gifs/igrf.html). Blocks were drilled in a zero field cage at the undertaken by three methods: (1) stepwise alternating field de- Liverpool University Geomagnetism Laboratory so as to remo- magnetization in 2.5e5.0 mT stages was undertaken using ve the influence of the Earth’s magnetic field and stop the acqui- a laboratory-built alternating field demagnetizer capable of sition of a drilling-induced magnetization. For each block imparting fields as high as 100 mT; (2) an 11-step thermal sample (15 in total; Table 1), at least two 25-mm-diameter demagnetization was undertaken on sister specimens between cores (44 in total; Table 1) were drilled and subsequently 100 and 700 C using a Magnetic Measurements thermal de- cut into a series of 25 20 mm subsamples. Heavily brecci- magnetizer; (3) a further set of sister specimens was subjected ated samples were avoided because they can often give spu- to an 11-point thermal demagnetization, with an initial alter- rious directions. Drilling was undertaken at right angles to nating field cleaning step of 12 mT. This was undertaken so the layering to obtain maximum sample resolution in deposi- as to remove any viscous magnetization and decrease the tem- tion and, hence, time. As the majority of samples came from perature to which highly calcified samples or calcites needed mined exposures, the top core was discarded so as to limit to be heated during subsequent thermal demagnetization. the effects of weathering, modern detrital contamination, After magnetic cleaning, ChRMs were determined through and possible effects from the mining process. principal components analysis (Kirschvink, 1980) using vector and stereographic projections to determine declination (orien- Methods tation in horizontal plane) and inclination (orientation in the vertical plane). Declination was corrected for local magnetic A series of standard mineral magnetic measurements were secular variation (13) and the polarities of subsamples undertaken on sister subsamples from the various paleomag- were assigned to normal or reversed polarity according to netic block samples to determine the magnetic mineralogy, whether their declinations were within a 40 cone of the nor- magnetic grain size, and concentration of remanence-carrying mal (0) or reversed (180) field and had inclinations above þ/ minerals (for more detailed description of the methodology, 20. Those outside this range of variation were determined see Walden et al., 1999). This was important for establishing as intermediate normal or reversed if they fell within an 80 the origin of the magnetic remanence preserved within the cone of declination variation or if their inclinations fell below speleothems and sediments. þ/20. If the values fell outside this range but a stable direc- Paleomagnetic methods employed here followed the proto- tion of magnetization could still be isolated, the samples were cols established in Butler (1992). Measurements were made designated as intermediate. Fisher (1953) mean directions using a cryogenic, SQUID-based, spinner magnetometer were then calculated for each block sample. This produced (FIT) with a minimum sensitivity of 0.2 108 Am2 kg1 a sequence of polarity intervals and reversals that were then for weak speleothem samples and a Molspin minispin magne- correlated to the globally known dated record of polarity tometer for any samples containing significant proportions of changes, or geomagnetic polarity time scale (GPTS; Ogg sediment. Magnetic cleaning to identify the characteristic rem- and Smith, 2004), to produce potential age ranges for the anent magnetization (ChRM; main remanence preserved in the various deposits and site as a whole.

Table 1 Paleomagnetic and mineral magnetic results

Sample no. Type Dec ( ) Inc ( ) a95 K N Polarity KLF XFD%Tc(C) Hc (mT) RS GD 2-05u FS 203.2 25.4 37.2 47.2 2 R GD 2-05m FS 252 8.6 35.4 55.3 2 I GD 2-05l FS 29.9 45.3 31.7 64.2 2 N GD 2-04FS CFS 350.3 48.9 8.5 212.5 3 N GD 2-04CS CS 39.9 43.6 12.6 394.6 2 N 33 9.09 570 28.2 0.55 GD 2-03 CS 39.7 39.7 9 106.2 4 N 182 10.99 595 26 0.59 GD 2-02 CS 17.2 52.2 6.1 415.6 3 N 411 8.76 585 27.3 0.53 GD 2-06 CS 345.5 60.1 5.6 484.9 3 N 305 9.54 576 29.1 0.6 GD 2-01 FS 199 49.8 24.4 26.6 3 R GD 1L-03 FS 36.5 57.6 15.5 64.4 3 N GD 1L-00 FS 93.1 53.1 81.8 11.6 2 I GD 3-05 FS 202 28 31.3 65.8 2 R 1037 9.64 579 28.4 0.7 GD 3-04 FS 202.3 47.8 66 16.4 2 R 567 10.05 590 29 0.71 GD 3-03 CS 214.4 27.8 53.9 23.6 2 R 96 8.13 576 28.5 0.63 GD 3-07 CS 88.5 57.2 25 102.2 2 I 387 8.43 575 29.1 0.65 GD 3-02 CS 13.1 47 10.5 137.8 3 N 490 7.76 580 26.9 0.65 GD 3-01 FS 220.9 29 8.5 867.6 2 IR GD 4-01 FS 218.6 30.5 32.9 59.9 2 R Abbreviations are as follows: FS ¼ flowstone; CS ¼ calcified silts; CFS ¼ contaminated flowstone; Dec ¼ declination; Inc ¼ inclination; Tc ¼ Curie point; RS ¼ ratio of magnetic susceptibility at room temperature and 196 C; KLF ¼ low temperature magnetic susceptibility; XFD% ¼ frequency dependence of magnetic susceptibility; Hc ¼ coercive force in milliTesla (mT). A.I.R. Herries et al. / Journal of Human Evolution 51 (2006) 617e631 625

Results [RS values (ratio of room temperature susceptibility versus susceptibility at 196 C); Table 1]. Samples from the GD 1 and Paleomagnetic and mineral magnetic results are shown in GD 3 localities show slightly higher RS values than GD 2, sug- Table 1. Natural remanent magnetization (NRM; combination gesting a greater contribution from stable single-domain (SSD) of primary and secondary magnetization in the sample before grains and less from ultrafine viscous grains. This is reflected in any magnetic cleaning) intensities ranged from between the stability of remanence, with GD 2 samples showing slightly 26 106 Am2 kg1 in siltstone samples to 1.1 106 Am2 less stability of remanence to alternating field demagnetization kg1 in the clastic-free speleothem deposits. Measured sample and a greater viscosity. Such variations may point to different NRMs were found to contain a secondary, viscous remanent sourcing of material or deposition at different time periods. magnetization (VRM) component that was removed by alter- This is indicated by different polarities for some samples from nating field demagnetization to between 100 and 120 Gauss the GD 1 and GD 3 locality when compared to the GD 2 locality. (10e12 mT), or by thermal demagnetization to around 200 Lowrie and Fuller (1971) tests were undertaken, and to 250 C(Fig. 5). These samples all had median destructive median destructive fields ranged between 15 and 30 mT with fields (the point at which NRM intensity is reduced by half) greater stability of anhysteretic remanent magnetization of around 150 to 200 Gauss (15e20 mT). Further stepwise de- (ARM) than isothermal remanent magnetization (IRM) at magnetization to between 300 and 1000 Gauss (30e100 mT), low fields, which suggests that a predominantly stable single- or 200 and 550 C, identified a single stable characteristic domain (SSD) mineralogy carries the stable ChRM. Hysteresis remanent magnetization (ChRM) residing within higher coer- loops also show partial separation of the curve at low fields, civity minerals (Fig. 5). again suggesting that a more stable ferrimagnetic SSD miner- Remanence was completely removed in most samples by alogy dominates. The IRM demagnetization curves, however, 550 C, suggesting that magnetite is the main remanence car- show a distinctly concave shape suggestive of a low coercivity rier. This is partially confirmed by thermomagnetic analysis, mineralogy, which is considered to be due to SP/SD boundary with samples having Curie temperatures of around 580 C. grains, as shown by XLT and XFD%, rather than multidomain Some samples have slightly higher Curie temperatures and grains, as normally interpreted from Lowrie and Fuller a 50% drop in magnetization after heating, which suggests (1971) tests. No evidence of multidomain grains is seen in the presence of stable maghaemite. low-temperature magnetic-susceptibility measurements or Some very high magnetic susceptibility (KLF) values are other tests. There is thus no evidence of the large detrital seen (up to 1037 SI; Table 1). The mean frequency dependence grains of magnetite, which may have made paleomagnetic di- of magnetic susceptibility (cFD%) is 9.29%, suggesting a rections unstable at Sterkfontein and Swartkrans, as suggested high proportion of ultrafine, superparamagnetic, single-domain by Jones et al. (1986). (SP/SD) boundary grains. These fine, viscous magnetite grains Nonsaturation of IRM acquisition curves suggests that the are responsible for the secondary viscous remnant magnetiza- corresponding oxidized and hydrated altered minerals, hema- tion (VRM) component seen in the samples. Secondary re- tite, and probably goethite also occur. Hematite causes the manence acquisition is primarily due to the relaxation of red coloration of sediments but exists mostly in its amorphous low-coercivity, viscous magnetic grains that do not hold a stable state or as ultrafine-grained pigment minerals that do not con- remanence over the time period represented by the age of the tribute to the magnetic remanence. A range of minerals and deposits. Low-temperature magnetic susceptibility (KLT) grain sizes occurs with finer, single-domain ferromagnetic curves have an SP tail below 150 C and a linear trend grains causing the VRM component and larger, stable single- back to room temperature indicative of a mixture of superpar- domain ferromagnetic grains (both magnetite and maghaemite) amagnetic grains and larger single-domain magnetite grains holding the ChRM.

Fig. 5. A comparison of vector and stereographic projections and intensity spectra for (THD; degrees centigrade) thermal and (AFD; units in Gauss [Gauss/ 10 ¼ mT]) alternating ﬁeld demagnetization for the same sample from GD 2-06.; intensity (106 Am2 kg1). 626 A.I.R. Herries et al. / Journal of Human Evolution 51 (2006) 617e631

Iron and manganese content of the local dolomite is fairly from the middle of the Bed 3 flowstone. Due to the slow de- consistent and relatively high (Veizer, 1978). The high NRM, positional rate of the flowstone, the record of the polarity KLF, and cFD% values are thought to be due to the high occur- change is not of sufficient resolution to determine much about rence of mineralization in the region with a greater supply of the behavior of the Earth’s geomagnetic fields during field re- Fe into the colluvial soil profiles from weathering of base do- versals. A single intermediate-polarity sample is bounded by lomite rock, interbedded shale, and overlying volcanics and normal and reversed samples. This is typical of polarity rever- ironstone that are high in Fe. The high proportion of ultrafine, sals recorded in speleothems from the South African hominin viscous-to-superparamagnetic grains is due to the continual sites, where speleothem deposition is seemingly too slow to burning of the landscape by natural fires that produce super- record high-resolution data on reversals during these time pe- paramagnetic grain-size populations. The sediments are riods (Herries, 2003). Due to the weak nature of the flowstone, thought to be of colluvial origin but have been winnowed smaller samples could not be used to increase the resolution. from a now-eroded entrance-collapse deposit. The colluvial Samples from GD 1 and GD 3 show a mixture of directions, soils in the area are minimally developed today, and the loca- with intermediate, normal, and reversed directions occurring. tion of the site at the crest of a hill means that colluvial soil Their stratigraphic relationship is undefined and so only gen- development is likely to have been limited in the past, and sed- eral conclusions can be made at present. iment input would therefore have been from a very small Some speleothem samples have more scattered orthogonal catchment of runoff from the hillcrest. plots due to the weak magnetization of the samples. Inclina- Grain sizes exist from ultrafine to large single-domain tions between 21 and 63.5 were also recorded for individ- grains, causing the remanence spectra of VRM and ChRM to ual subsamples in some instances. These were relatively overlap and cause greater difficultly in isolating the ChRM. consistent for all subsamples from the blocks, as shown by However, a stable primary ChRM can be identified in most K and a95 values, the highest a95 values and lowest K cases. Siltstone samples from GD 2 and GD 3 appear to be values being from weaker speleothem samples. This is also more stable to thermal (THD) than alternating-field (AFD) de- partly due to the fact that samples from speleothems include magnetization, while those from GD 1 are equally stable during multiple layers of flowstone deposition with lower deposi- both methods of demagnetization. Figure 5 shows a comparison tional rates than the siltstone deposits. Some samples pos- of THD and AFD for sample GD 2-06. Both samples show sessed inclinations that were shallow when compared to the equal stability and direction during both methods of analysis. average inclination, which is expected for this latitude. It is Figure 6 depicts examples of orthogonal plots for reversed- well known that clastic deposits can show shallow inclina- and normal-polarity samples and shows their bipolar nature. tions of several degrees due to particle settling, slope effects, Site means for reversed and normal samples were 210/33.9 and compaction (Brennan, 1993). Postdepositional effects and 23.1/46.8, respectively, which are not similar to the may also be caused by periodic flooding of the cave, al- present field direction for the site (341.2/64.6). though this appears to be of relatively low energy given Full polarity data are shown in Table 1. The sequence at GD the retained articulation of some fossils and lack of evidence 2 shows a change from reversed directions to normal direc- for winnowing of skeletal materials. Despite these problems, tions and back again. The Bed 1 (GD-2-B1) basal flowstone experiments suggest that lock-in times would likely be fairly has entirely reversed directions of magnetization. The fossil- rapid due to high depositional rates. The siltstone no doubt bearing Bed 2 (GD-2-B2) siltstone and breccias all have nor- represents a very short time period on the scale of thousands mal directions of magnetization. The calcite-rich interface of of years. In the case of speleothem, the magnetizations were Bed 2 and base of the Bed 3 capping flowstone (GD-2-B3) always easy to isolate and polarities safely assigned. This is also record normal directions of polarity. A change in polarity because, as with many speleothems, magnetic grains become from normal to reversed is then recorded in sample GD2-05 locked in by the calcite of the next layer, usually within a few years at most, and therefore do not show bedding er- rors and cannot be altered postdepositionally (e.g., Latham and Ford, 1993). The consistency in directions for the subsamples in each block and the supporting mineral magnetic analysis suggest that the ChRMs isolated in the samples are carried by stable single-domain magnetite and maghaemite grains and are primary (p)DRM reflecting the Earth’s ancient geomagnetic field direction at or very near to the time of their deposition. The polarity of the samples therefore reflects the age of their deposition.

Biochronology

Fig. 6. An example of stereographic projection and vector plots for normal (GD 2-02; negative inclination) and reversed (GD 3-04; positive inclination) Due to the multigenerational nature of the cave fills, faunal polarity samples using thermal demagnetization (THD; values in degrees analysis was only undertaken on fossils excavated by Vrba in centigrade). 1979 from in situ deposits that do not represent more recent A.I.R. Herries et al. / Journal of Human Evolution 51 (2006) 617e631 627 infills. The GD 2 fossil assemblage recovered by Vrba was Olduvai Gorge) and South Africa (Makapansgat, Bolt’s originally analyzed by Watson (1993), who offered a prelimi- Farm, Kromdraai, Swartkrans, the Vaal River Gravels, and nary report based on a portion of the 90,000þ specimens re- Gondolin) (Harris and White, 1979; Harris, 1983; White, covered from the Phase 2 siltstone deposits. Watson 1995; Turner et al., 1999). The first appearance datum suggested a tentative faunal age of between 1.9e1.5 million (FAD) for M. andrewsi in east Africa is in the Omo Shungura years for the assemblage based on an analysis of nine Metri- B10 stratum, dated to 2.95 Ma (Feibel et al., 1989), and in diochoerus andrewsi molar specimens examined by H.B.S. South Africa at Makapansgat Member 3, dated to between Cooke, who stated that the GD 2 specimens ‘‘resemble the 3.04 and 2.58 Ma (Herries and Latham, 2002; Herries, type of Broom’s ‘Notochoerus meadowsi’. This is now gener- 2003). White (1995) considered this FAD to be reliable be- ally regarded as a synonym of Metridiochoerus andrewsi,to cause older fossil-bearing units at Omo Shungura and Koobi which the Gondolin specimens may be referred. They are Fora lack any representatives of the species, which is the ear- also similar to specimens from Swartkrans Member 1 and liest known of the genus Metridiochoerus. The species persists lie at the upper limit of variation in the M. andrewsi material through the Pliocene and into the Pleistocene, giving rise to from Koobi Fora’’(Watson, 1993: 37). three further species within the genus before its last appear- Further taphonomic and paleoecological analysis of the GD ance datum (LAD) just above the Okote Tuff at Koobi Fora, 2 assemblage led to a comprehensive reexamination of all of which has recently been dated to 1.56 0.05 Ma (McDougall the specimens recovered during the 1979 excavation (Adams and Brown, 2006). In South Africa, M. andrewsi specimens and Conroy, 2005; Adams, 2006). This recent research recov- have been recovered from deposits as young as those in Krom- ered further suid specimens from the 1979 GD 2 collection, draai A (1.9e1.65 Ma; Delson, 1988) and Swartkrans Member such that the total sample of M. andrewsi craniodental remains 1 (1.63 0.16 Ma; Curnoe et al., 2001; 1.8e1.5 Ma; Brain, now consists of nine neurocranial and 21 identifiable dental 1993, 1994), Member 2 (1.8e1.0 Ma; Brain, 1993, 1994), specimens, representing at least three individuals. Reanalysis and Member 3 (1.8e1.0 Ma; Brain, 1993, 1994). did not, however, identify any additional biochronologically The roughly 1.4-million-year duration of M. andrewsi diagnostic species in the GD 2 assemblage. As such, M. across the radiometrically dated eastern African fossil local- andrewsi is the only species from the assemblage that can ities might appear to make this species inappropriate for es- be used to relate the magnetostratigraphic samples of the tablishing a biochronological context for the Gondolin GD 2 GD 2 deposits to the global polarity time scale. deposits. In a critical review of the Plio-Pleistocene Suidae, Both the originally described and recently recovered GD 2 however, Harris and White (1979) divided M. andrewsi metridiochoerine craniodental specimens exhibit morphologi- into three informal stages (IeIII) based on morphological cal features consistent with Cooke’s (Watson, 1993: 37) orig- changes in third molars across the late Pliocene and early inal assignment of the material to M. andrewsi. Detailed Pleistocene. Harris and White (1979) argued that assessment morphological and metrical descriptions of the entire Gondo- of the M3 characteristics of recovered M. andrewsi speci- lin GD 2 M. andrewsi sample are not presented here, as these mens across these stages provides an important tool for the have already been published elsewhere (Adams and Conroy, biochronological assessment of fossil sites, as Stage I M. an- 2005; Adams, 2006). Instead, a brief description of features drewsi specimens were found only in the oldest deposits that are highly diagnostic and evolved within the species (Omo Shungura Members BeD), Stage II specimens from (and lineage) are given for the Gondolin GD 2 M. andrewsi localities of intermediate age (e.g., Omo Shungura Members upper and lower third molars and compared with features in EeH; Olduvai Bed I), and Stage III M. andrewsi specimens M. andrewsi specimens recovered from other southern and in only the youngest deposits (Unit 4/‘‘Upper Member’’ eastern African Plio-Pleistocene sites. These comparisons Koobi Fora Formation [above the KBS Tuff]; Olduvai lower highlight two important factors for using this species in estab- Middle Bed II). lishing a chronological context of the GD 2 faunal assemblage. The M. andrewsi M3s from the Gondolin GD 2 deposits (G First, the Gondolin GD 2 M. andrewsi specimens are morpho- 10304, G 12086, and G 14015; see Fig. 7) all exhibit typical logically similar to those interred in deposits that formed dur- Stage III characteristics, as described by Harris and White ing the terminal Pliocene and earliest Pleistocene, which (1979), and contrast in these features from Stage I and II restricts the normal-polarity events during which the deposits M. andrewsi remains. On each M3, the labial skirt of the trigon could have accumulated. Second, both the tempo and mode lacks strong flare, which contrasts with the pronounced lateral of metridiochoerine evolution in southern Africa was broadly flare on Stage I and II specimens. The lightly worn G 10304 similar to that across eastern Africa, which is essential for specimen has an X-shaped protocone and paracone, which is using M. andrewsi specimens to calibrate the magnetostrati- a shared character of both Stage II and III M. andrewsi spec- graphic results. imens. Additional distinctive Stage II and III features of the M3 trigon on the GD 2 specimens include the presence of a sin- Chronology and morphology of gle median cusp separating the first and second pairs of lateral Metridiochoerus andrewsi pillars, and a large gap between the first and second pair of lateral pillars on the lingual surface. The talon morphology of the The extinct suid M. andrewsi has been recovered from fos- GD 2 M3s exhibits other Stage III characters that are not typ- sil sites in eastern Africa (Omo Shungura, Koobi Fora, ical of Stage I or II M. andrewsi specimens. The entire talon is 628 A.I.R. Herries et al. / Journal of Human Evolution 51 (2006) 617e631

Biochronological correlations

A primary assumption made in using M. andrewsi for biochronological comparisons is that populations of this species in different geographic regions followed the same evolutionary trajectory (Harris and White, 1979). This is a particularly important assumption to justify because of the geographic distance and ecological differences between the southern and eastern African fossil sites. Both of these factors could have led to genetic isolation and divergent evolutionary Fig. 7. Occlusal view of the G 12086 Metridiochoerus andrewsi right M3. trends among metridiochoerine populations during the Plio- Scale bar equals 1 cm. Pleistocene, thereby limiting accurate biochronological correlations between southern and eastern African sites. elongated, exhibits slight anterior apical curvature, and each Although it is difficult to test this assumption, Harris and tooth possesses four pairs of lateral talon pillars. These lateral White (1979) argued that there is no evidence for significant talon pillars are Y-shaped mesially, but diminish in size and disruptions in gene flow between M. andrewsi populations complexity distally. The labial lateral pillars are also strongly across sites that would confound biochronological compari- subdivided. The median talon pillars are tightly packed and sons between southern and eastern Africa. These authors circular, leading to a ‘‘jumbled’’ appearance of talon pillars point to two lines of evidence to support this position. First, when the tooth is worn. Finally, there is a moderate-to-thick gradual morphological changes are observed in M. andrewsi cementum layer over the entire corrugated enamel surface, across successive stratigraphic units within eastern African from the cervix to the occlusal plane (Harris and White, 1979). fossil sites (e.g., Omo Shungura Members BeG) and across The M. andrewsi M3s recovered from GD 2 (G 105, G geographically separated eastern African sites (e.g., Omo 9877, G 9878, and G 14017; Fig. 8) also exhibit typical Stage Shungura, Koobi Fora, Olduvai Gorge). This not only sug- III features. Recovered specimens are extremely hypsodont gests local genetic continuity among M. andrewsi popula- and preserve a slight lateral flexure near the cervical margin, tions during the Plio-Pleistocene, but also regional genetic which is absent or less notable in earlier stages. The talonid continuity during the evolution of the species across sites is expanded, with four or five pairs of parallel, flattened, and like Omo Shungura and Olduvai Gorge, which are over relatively tightly packed lateral pillars (which contrasts with 800 km apart. Second, recovered southern African Stage the lower number of more separated lateral pillar pairs on IeIII M. andrewsi craniodental specimens exhibit similar Stage I and II specimens). When viewed buccally, the M3s ex- morphologies to eastern African M. andrewsi remains from hibit a slight mesial apical curvature that gives these speci- a corresponding evolutionary stage. The shared, continuous mens a sigmoid appearance. The border between the trigonid evolution of M. andrewsi in both these regions supports an and talonid is marked by a single or double (G 14017) median interpretation of gene flow between even widely separated pillar, with a continuous row of median pillars running from M. andrewsi populations during the Plio-Pleistocene (Harris behind the first talonid lateral to the distal talonid. When in and White, 1979). occlusion, the mesial lateral pillars are Y-shaped, but are With direct comparisons of the Gondolin GD 2 M. andrewsi more circular in shape towards the distal part of the talonid specimens to other southern and eastern African M. andrewsi (with all lateral pillars circular in progressively worn speci- samples justified, a biochronological date for the GD 2 de- mens). Finally, as in the M3 specimens, the enamel surface posits can be suggested based on correlations to other, well- is corrugated and supports a thick cementum layer. dated fossil assemblages. Relative to the previously recovered southern African M. andrewsi specimens, the GD 2 M. andrewsi third molars are morphologically identical to the Stage III M. andrewsi third molars from Swartkrans Member 1 (SK 387, SK 388, and SK 392/2380; see Figs. 7, 8), which was recently dated using ESR to 1.63 0.16 Ma (Curnoe et al., 2001). The Swartkrans specimens were previously described by Harris and White (1979) as morphologically similar to Stage III specimens recovered from Unit 4 of Koobi Fora Areas 103 and 104 (‘‘Upper Member’’ specimens above the KBS Tuff; younger than 1.869 0.021 Ma; McDougall and Brown, 2006), Omo Shungura Members HeJ (1.88e1.6 Ma; Feibel et al., 1989), and lower Middle Bed II from Olduvai (Tuff IIA: 1.66 0.1 Ma; Hay, 1976; Manega, 1993). Given the similarities between the Gondolin GD 2 Stage III M. Fig. 8. Medial views of the G 105 and SK 387 Metridiochoerus andrewsi left andrewsi specimens and those from these southern and eastern

M3s. Scale bar equals 1 cm. African sites, and the LAD for M. andrewsi at 1.56 0.05 Ma A.I.R. Herries et al. / Journal of Human Evolution 51 (2006) 617e631 629

(McDougall and Brown, 2006), a maximum age of between into the base of the Phase 3 speleothem that caps the GD 2 1.9 and 1.5 Ma is suggested. sequence. The top layers of the Phase 2 clastic deposits consist of calcite-rich siltstone and show a progression from one re- Combined magnetobiostratigraphy gime to another. The sedimentary sequence and paleomagnetic results show no evidence of a significant hiatus between Phase The ideal requirements for magnetostratigraphic chronol- 2 and Phase 3, as suggested between Phase 1 and Phase 2. A ogy of a depositional sequence are that: (1) samples from sharp change is seen from normal to reversed polarity in the the deposits possess a primary remanent magnetization of suf- Phase 3 flowstone with one intermediate sample in between. ficient strength that, after the removal of any viscous compo- Due to the variation in depositional rates between speleo- nents, is measurable in a magnetometer; (2) the deposits were thems and sediments, compression and extension of polarity laid down at about the same depositional rate throughout the periods related to expected depositional rates need to be taken sequence, with no hiatuses or unmeasurable units, so that po- into account when assessing age. The samples from GD 2 are larity zones are not distorted or omitted; and (3) the sequence dominated by normal polarity and occur in siltstone, which is has a sufficient number of reversals such that its record can be expected to have a high depositional rate. They therefore cover tied uniquely to a part of the dated GPTS before evidence from a short time period compared to the reversed directions seen in other methods is considered. Fulfilling all these criteria in the the underlying and capping flowstones. Due to the unknown South African deposits and in caves in general is rare, as is the length of the hiatus between basal speleothem and the silt- case at Gondolin. The complex conditions that occurred un- stone, the task is, therefore, to identify this short period of nor- derground during cave infilling, subsequent weathering, and mal polarity and polarity transition, which occurs in the mining expressed by unconformities within, and between, capping speleothem of the GD 2 deposit. Due to the short the preserved and exposed sedimentary profiles can hide sub- sequence of the deposits, the dating of this normal-polarity stantial parts of the magnetostratigraphic record. Without period can therefore not be made without arguments for the some form of age constraints, even very broad ones, any age of the fauna recovered in situ from the GD 2 locality. correlation of the magnetic polarity sequence with the GPTS Given the broadest biochronological delineations for the cannot be explicit. age of the deposit between (maximally) 2.95/3.04 Ma (M. The magnetostratigraphy of the GD 2 deposits is shown in andrewsi FAD eastern Africa/southern Africa) and (mini- Figure 9. The sequence shows reversed polarity in the Phase 1 mally) 1.56/1.63 Ma (M. andrewsi LAD east Africa/southern basal speleothem. A sharp contact occurs between the Phase 1 Africa), the normal polarity recorded in the GD 2 Phase 2 silt- basal speleothem and the Phase 2 clastic deposits, suggesting stone and overlying Phase 3 flowstone can be related to three that a hiatus of some length occurred. The entire sequence of possible periods of deposition: either the end of the Gauss be- the Phase 2 siltstone and breccias, including the fossiliferous tween 3.03 and 2.58 Ma, the Reúnion events (2.14 0.03 Ma) deposits, is of normal polarity. The normal polarity continues (Ogg and Smith, 2004), or the Olduvai event between 1.95 and

Fig. 9. Magnetostratigraphy of the Gondolin paleocave GD 2 deposits. Suggested periods of deposition defined by depositional rates are outlined in gray. A reversed polarity is recorded in the basal speleothem deposits. A break in sedimentation then occurs before the deposition of fossil-bearing, normal-polarity calcified silts. The normal polarity continues into capping flowstone deposits before a reversal occurs, which is correlated to the end of the Olduvai event at 1.78 Ma. Faunal age ranges are also outlined (1.9e1.5 Ma). The fossils are estimated to date to slightly older than 1.78 Ma. Phases 1, 2, and 3 are denoted (P1, P2, P3). 630 A.I.R. Herries et al. / Journal of Human Evolution 51 (2006) 617e631

1.78 Ma. The morphological assignment of the suid fossils to costs for AIRH were funded by the Gondolin Research Pro- Stage III M. andrewsi makes the older Reúnion and Gauss pe- ject. Funding for faunal analysis was provided to JWA by riods unlikely candidates for assignment of the normal polarity the NSF (#0308014) and the Wenner-Gren Foundation. The identified in the fossil-bearing silts of GD 2. The recovery of SAHRA excavation permit for Gondolin during this research Equus sp. from GD 2 also eliminates assemblage deposition was held by KLK (No. 80/01/11/006/51). We also thank the during the Gauss event, as the genus only appears in Africa af- staff of the Transvaal Museum for access to the GD 2 faunal ter 2.36 Ma (Behrensmeyer et al., 1997; Adams and Conroy, materials, and Mr. Peter Fleming and his family for granting 2005). Moreover, where the Reúnion event has been noted access to his property and for their ongoing support of our in South Africa (Herries, 2003), it consists of a very short ep- research program at Gondolin. Thanks to Darryl Granger isode that is unlike the data from Gondolin. The normal-polar- and two anonymous referees for useful comments on an earlier ity siltstone and faunal beds of GD 2 are thus considered to version of this manuscript. date to the Olduvai normal event, between 1.95 and 1.78 Ma (Fig. 9), as it is the only normal-polarity period within the op- References timal faunal time range. The capping flowstone records a polarity transition from Adams, J.W., 2006. Taphonomy and paleoecology of the Gondolin Plio- normal to reversed polarity, and therefore represents the end Pleistocene cave site, South Africa. Ph.D. Dissertation, Washington Uni- of the Olduvai event at 1.78 Ma. The time period for the depo- versity in St. Louis. sition of the GD 2 siltstone is no doubt very short, covering Adams, J.W., Conroy, G.C., 2005. Plio-Pleistocene faunal remains from the perhaps as little as a few thousand years, with only a small Gondolin GD 2 in situ assemblage, North West Province, South Africa. In: Lieberman, D., Smith, R.J., Kelley, J. (Eds.), Interpreting the Past: amount of slowly deposited flowstone laid down before the po- Essays on Human, Primate and Mammal Evolution in Honor of David larity transition occurred at 1.78 Ma. The depositional age of Pilbeam. Brill Academic Publishers Inc., Boston, pp. 243e261. the fossils must therefore be close to the 1.78 Ma boundary. Behrensmeyer, A.K., Todd, N.E., Potts, R., McBrinn, G.E., 1997. Late Plio- This correlates well with the upper ESR age limit of cene faunal turnover in the Turkana Basin, Kenya and Ethiopia. Science e 1.79 Ma (Curnoe et al., 2001) and general age estimates 278, 1589 1594. e Beukes, N.J., Klein, C., Kaufman, A.J., Hayes, J.M., 1990. Carbonate petrog- (1.8 1.5 Ma; Brain, 1993) for Member 1 at Swartkrans, in raphy, kerogen distribution, and carbon and oxygen isotope variations in an which similar Stage III M. andrewsi specimens have been re- Early Proterozoic transition from limestone to iron-formation deposition, covered. Our analysis therefore suggests that the GD 2 fossil Transvaal Supergroup, South Africa. Econ. Geol. 85, 663e690. locality is contemporaneous with Swartkrans Member 1 and Brain, C.K., 1958. The Transvaal ape-man bearing cave deposits. Trans. Mus. at least partially contemporaneous with Member 2 of Krom- Mem. 11, Transvaal Museum, Pretoria. Brain, C.K., 1993. A taphonomic overview of the Swartkrans fossil assem- draai B (Thackeray et al., 2002) at roughly 1.8 Ma. blages. In: Brain, C.K. (Ed.), Swartkrans: A Cave’s Chronicle of Early Given that the GD 2, Phase 2 siltstone deposits date to be- Man. Transvaal Museum, Pretoria, pp. 257e264. tween 1.95 and 1.78 Ma, the underlying flowstone is likely to Brain, C.K., 1994. The Swartkrans palaeontological research project in per- date to the beginning of the Matuyama reversed chron between spective: results and conclusions. S. Afr. J. Sci. 90, 220e223. 2.58 and 1.95 Ma, perhaps the Matuyama C2r.1r chron Brennan, W.J., 1993. Origin and modification of magnetic fabric in fine- grained detrital sediment by depositional and post-depositional processes. between 2.13 and 1.95 Ma, but this is in no way conclusive. In: Aissaoui, D.M., McNeill, D.F., Hurley, N.F. (Eds.), Applications of The samples from the GD 1 and GD 3 localities are dominated Palaeomagnetism to Sedimentary Geology. Society for Sedimentary by reversed and intermediate polarities, suggesting that the Geology (SEPM), Special Publication 49, pp. 17e27. material does not cover the same time period as the faunal Butler, R., 1992. Palaeomagnetism: Magnetic Domains to Geological Ter- layers from GD 2. As a whole, these localities may date to ei- ranes. Blackwell Scientific Publications, Oxford. Curnoe, D., Grun, R., Taylor, L., Thackeray, J.F., 2001. Direct ESR dating of ther before and/or after the GD 2 deposits, in either the Ma- a Pliocene hominin from Swartkrans. J. Hum. Evol. 40, 379e391. tuyama C2r.1r chron between 2.11 and 1.95 Ma or the Delson, E., 1988. Chronology of South African australopith site units. In: Matuyama C1r.3r chron between 1.78 and 1.07 Ma. Currently Grine, F. (Ed.), Evolutionary History of the ‘‘Robust’’ Australopithecines. available data suggest that the same polarity transition at Aldine de Gruyter, New York, pp. 317e324. 1.78 Ma may be recorded in this sequence and that the GD de Ruiter, D.J., 2003. Revised faunal lists for Member 1-3 of Swartkrans, South Africa. Ann. Transvaal. Mus. 40, 29e41. 1/3 fossil deposits are at least, in part, younger than GD 2. Fur- Feibel, C.S., Brown, F.H., McDougall, I., 1989. Stratigraphic context of fossil ther paleomagnetic, stratigraphic, and faunal research of in hominids from the Omo Group deposits: Northern Turkana Basin, Kenya, situ deposits is currently underway to clarify these issues. Ethiopia. Am. J. Phys. Anthropol. 78, 595e622. Fisher, R., 1953. Dispersion on a sphere. Philos. Trans. R. Soc. Lond. A 306, e Acknowledgements 295 305. Ford, D.C., Williams, P., 1989. Karst Geomorphology and Hydrology. Unwin Hyman, London. The paleomagnetic and speleological work was funded by Harris, J., 1983. Koobi Fora Research Project, Vol. 2: The Fossil Ungulates: a grant made to AIRH from the Arts and Humanities Research Proboscidea, Perissodactyla, and Suidae. Clarendon Press, Oxford. Board (AHRB Studentship No. 99/3050). Paleomagnetic work Harris, J., White, T., 1979. Evolution of the Plio-Pleistocene African Suidae. e was undertaken at the Liverpool University Geomagnetism Trans. Am. Philos. Soc. 69, 5 128. Hay, R.I., 1976. Geology of the Olduvai Gorge. University of California Press, Laboratory. The Variable Field Translation Balance (VFTB) Berkeley, California. used in this work was funded by NERC JREI grant GR3/ Herries, A.I.R., 2003. Magnetostratigraphic seriation of South African homi- E0069 granted to John Shaw. Research at the site and travel nin palaeocaves. Ph.D. Dissertation, University of Liverpool. A.I.R. Herries et al. / Journal of Human Evolution 51 (2006) 617e631 631

Herries, A.I.R., Latham, A.G., 2002. Dating the depositional sequence and McFadden, P.L., 1980. An overview of palaeomagnetic chronology with australopithecine ‘‘Grey Breccia’’ of Makapansgat Limeworks using mag- special reference to the South African hominid sites. Palaeontol. Afr. 23, netostratigraphy. Am. J. Phys. Anthropol. 34 (Suppl.), 84e85. 35e40. Herries, A.I.R., Reed, K., Kuykendall, K.L., Latham, A.G., 2006. Speleology Menter, C.G., Kuykendall, K.L., Keyser, A.W., Conroy, G.C., 1999. First re- and magnetobiostratigraphic chronology of the Buffalo Cave fossil bearing cord of hominid teeth from the Plio-Pleistocene site of Gondolin, South palaeodeposits, Makapansgat, South Africa. Quat. Res. doi:10.1016/ Africa. J. Hum. Evol. 37, 299e307. j.yqres.2006.03.006. Meyer, F.M., Robb, L.J., 1996. The geochemistry of Black Shales from the Hopley, P.J., 2004. Palaeoenvironmental reconstruction of South African hom- Chuniespoort Group, Transvaal Sequence, Eastern Transvaal, South inin-bearing cave deposits using stable isotope geochemistry. Ph.D. Disser- Africa. Econ. Geol. 91, 111e121. tation, University of Liverpool. Ogg, J.G., Smith, A.G., 2004. The geomagnetic polarity time scale. In: Jones, D.L., Brock, A., McFadden, P.L., 1986. Palaeomagnetic results from the Gradstein, F., Ogg, J., Smith, A. (Eds.), A Geologic Time Scale 2004. Kromdraai and Sterkfontein hominid sites. S. Afr. J. Sci. 82, 160e163. Cambridge University Press, Cambridge, pp. 63e86. Kirschvink, J.L., 1980. The least-square line and plane and the analysis of Partridge, T.C., Granger, D.E., Caffee, M.W., Clarke, R.J., 2003. Lower Plio- palaeomagnetic data. Geophys. J. R. Astron. Soc. 62, 699e718. cene hominid remains from Sterkfontein. Science 300, 607e612. Kuykendall, K.L., Conroy, G.C., 1999. Description of the Gondolin teeth: Partridge, T.C., Latham, A.G., Heslop, D., 2000. Appendix on magnetostratig- Hyper-robust hominids in South Africa? Am. J. Phys. Anthropol 28 raphy of Makapansgat, Sterkfontein, Taung and Swartkrans. In: (Suppl.), 176e177. Partridge, T.C., Maud, R.R. (Eds.), The Cenozoic of Southern Africa. Ox- Lacruz, R.S., Brink, J.S., Hancox, J., Skinner, A.S., Herries, A., Schmidt, P., ford Monographs on Geology and Geophysics. Oxford University Press, Berger, L.R., 2002. Palaeontology, geological context and palaeoenviron- Oxford, pp. 126e129. mental implications of a middle Pleistocene faunal assemblage from the Partridge, T.C., Shaw, J., Heslop, D., Clarke, R.J., 1999. The new hominid Gladysvale Cave, South Africa. Palaeontol. Afr. 38, 99e114. skeleton from Sterkfontein, South Africa: age and preliminary assessment. Latham, A.G., Ford, D., 1993. The palaeomagnetism and rock magnetism of J. Quatern. Sci. 14, 293e298. cave and karst deposits. In: Aissaoui, D.M., McNeill, D.F., Hurley, N.F. Thackeray, J.F., Kirschvink, J.L., Raub, T.D., 2002. Palaeomagnetic analysis (Eds.), Applications of Palaeomagnetism to Sedimentary Geology. Society of calcified deposits from the Plio-Pleistocene hominid site of Kromdraai, for Sedimentary Geology (SEPM), Special Publication 49, pp. 149e155. South Africa. S. Afr. J. Sci. 98, 537e540. Latham, A.G., Herries, A.I.R., Kuykendall, K., 2003. The formation and sed- Tobias, P.V., 2000. The fossil hominids. In: Partridge, T.C., Maud, R.R. (Eds.), imentary infilling of the Limeworks Cave, Makapansgat, South Africa. The Cenozoic of Southern Africa. Oxford Monographs on Geology and Palaeontol. Afr. 39, 69e82. Geophysics. Oxford University Press, Oxford, pp. 252e276. Latham, A.G., Herries, A., Quinney, P., Sinclair, A., Kuykendall, K., 1999. The Turner, A., Bishop, L.C., Denys, C., McKee, J.K., 1999. A locality-based list- Makapansgat australopithecine site from a speleological perspective. In: ing of African Plio-Pleistocene mammals. In: Bromage, T.G., Schrenk, F. Pollard, A.M. (Ed.), Geoarchaeology: Exploration, Environments, Re- (Eds.), African Biogeography, Climate Change, and Human Evolution. sources. Royal Geological Society, London Special Publication 165, pp. Oxford University Press, Oxford, pp. 369e399. 61e77. Veizer, J., 1978. Secular variations in the composition of sedimentary carbon- Low, A.B., Robelo, A.G., 1996. Vegetation of South Africa, Lesotho and Swa- ate rocks. II. Fe, Mn, Ca, Ng, Si and minor constituents. Precambrian Res. ziland. Department of Environmental Affairs and Tourism, Pretoria. 6, 381e413. Lowrie, W., Fuller, M., 1971. On the alternating field demagnetization charac- Walden, J., Oldfield, F., Smith, J., 1999. Environmental Magnetism: A Practi- teristics of multidomain thermoremanent magnetization in magnetite. cal Guide. Quaternary Research Association Technical Guide 6, London. J. Geophys. Res. 76, 6339e6349. Watson, V., 1993. Glimpses from Gondolin: A faunal analysis of a fossil Manega, P.C., 1993. Geochronology, geochemistry and isotopic study of the site near Broederstroom, Transvaal, South Africa. Palaeontol. Afr. 30, Plio-Pleistocene hominid sites and the Ngorongoro volcanic highland in 35e42. northern Tanzania. Ph.D. Dissertation, University of Colorado. White, T., 1995. African omnivores: Global climatic change and Plio-Pleisto- McDougall, I., Brown, F.H., 2006. Precise 40Ar/39Ar geochronology for the cene hominids and suids. In: Vrba, E.S., Denton, G.H., Partridge, T.C., upper Koobi Fora Formation, Turkana Basin, northern Kenya. J. Geol. Burckle, L.H. (Eds.), Paleoclimate and Evolution with Emphasis on Hu- Soc. Lond. 163, 205e220. man Origins. Yale University Press, New Haven, pp. 369e384.