New Excavations at Umhlatuzana Rockshelter, KwaZulu-Natal, South Africa: a Stratigraphic and Taphonomic Evaluation

Umhlatuzana rockshelter has an occupation sequence spanning the last 70,000 years. It is one of the few sites with deposits covering the Middle to Later Stone Age transition (~40,000–30,000 years BP) in southern Africa. Comprehending the site’s depositional history and occupation sequence is thus important for the broader understanding of the development of Homo sapiens’ behavior. The rockshelter was first excavated in the 1980s by Jonathan Kaplan. He suggested that the integrity of the late Middle Stone Age and Later Stone Age sediments was compromised by large-scale sediment movement. In 2018, we initiated a high-resolution geoarchaeological study of the site to clarify the site formation processes. Here, we present the results of the excavation and propose a revised stratigraphic division of the Pleistocene sequence based on field observations, sedimentological (particle size) analyses, and cluster analysis. The taphonomy of the site is assessed through phytolith and geochemical (pH, loss on ignition, stable carbon isotope) analyses. The results indicate a consistent sedimentological environment characterized by in situ weathering. The analysis of the piece-plotted finds demonstrates semihorizontal layering of archaeologically dense zones and more sterile ones. There was no indication of large-scale postdepositional sediment movement. We show that the low-density archaeological horizons in the upper part of the Pleistocene sequence are best explained by the changing patterns of sedimentation rate.

geoarchaeological study of the site to clarify the site formation processes. Here, we present the results of the excavation and propose a revised stratigraphic division of the Pleistocene sequence based on field observations, sedimentological (particle size) analyses, and cluster analysis. The taphonomy of the site is assessed through phytolith and geochemical (pH, loss on ignition, stable carbon isotope) analyses. The results indicate a consistent sedimentological environment characterized by in situ weathering. The analysis of the pieceplotted finds demonstrates semihorizontal layering of archaeologically dense zones and more sterile ones. There was no indication of large-scale postdepositional sediment movement. We show that the low-density archaeological horizons in the upper part of the Pleistocene sequence are best explained by the changing patterns of sedimentation rate.

Introduction
Umhlatuzana rockshelter is one of the few sites in southern Africa with a continuous occupation sequence spanning the last 70,000 years. It was at the beginning of this period that "modern" human behaviors appeared, represented by the Still Bay and Howiesons Poort industries, both of which are represented at the site (Archer et al. 2016;Högberg and Lombard 2016a, b;Kaplan 1990;Lombard 2007;Mohapi 2013;McCall and Thomas 2009). Located in KwaZulu-Natal (eastern South Africa), the site contains important information on the techno-cultural sequence of the South African Middle Stone Age (MSA) and Later Stone Age (LSA). The Pleistocene archaeological sites of the region are less abundant and not well-known compared to southwest South Africa, which has dominated the techno-cultural study of the period (e.g., Deacon 1984;Volman 1981;Wurz 2002). Thus, Umhlatuzana occupies a critical position due to its geographical location and because its MSA and LSA deposits bridge the gaps in archaeological sequences at other important sites.
Recent work in KwaZulu-Natal and the surrounding areas has brought to light evidence of early sophisticated behavior from the MSA (Fig. 1). For instance, Sibudu Cave has yielded important MSA remains (Backwell et al. 2008;Hodgkiss 2013;Rots et al. 2017;Wadley 2007;Wadley et al. 2011). Its occupation commenced earlier than that at Umhlatuzana, but Sibudu lacks deposits from the MSA-LSA transition and the subsequent LSA (Wadley and Jacobs 2006). Another important site, Umbeli Belli, has so far yielded archaeological remains from the later parts of the MSA and the Pleistocene LSA (Bader et al. 2016(Bader et al. , 2018. Border Cave has an important MSA occupation sequence and has yielded the earliest ages for the MSA/LSA transition (d 'Errico et al. 2012;Villa et al. 2012). Its occupation ceased after the transitional phase, and no evidence from the later Pleistocene LSA has been recovered. At Sehonghong in Lesotho, occupation started with the Howiesons Poort and continued up to historical times (Mitchell 1994;Stewart et al. 2016). Sehonghong does not preserve pre-Howiesons Poort occupations and has an occupation hiatus during the Last Glacial Maximum (Pargeter et al. 2017). Finally, Waterfall Bluff shows human occupation starting at least as early as 37.6 ka and lasting through the Middle Holocene (Fisher et al. 2020).
The continuous sequence at Umhlatuzana has the potential to enable us to relate all these occupational pulses to each other. Umhlatuzana is one of the few sites with occupation across the MSA-LSA transition (Loftus et al. 2019;Mackay et al. 2014;Mitchell 2008;Villa et al. 2012). However, at present, transitional MSA-LSA assemblages are not well understood in southern Africa, both in terms of technological organization and chronology. The earliest LSA in the region has only been informally characterized Mackay et al. 2014). Technological variability appears considerable, and assemblages are spread across great distances. It is unclear how the earliest LSA assemblages relate to transitional MSA-LSA assemblages.
Rescue excavation of the site was conducted in 1985 to mitigate the site's potential destruction from highway construction (Kaplan 1989(Kaplan , 1990. Unfortunately, due to uncertainties about the stratigraphic integrity of Umhlatuzana's sedimentary sequence, the archaeological assemblages have been overlooked in the discourse of the MSA/LSA transition (cf. Deacon andDeacon 1999, Mitchell 2002). Kaplan (1990) reported that no visible stratigraphic boundaries were observed in the Pleistocene part of the sequence and that the deposits were excavated in artificial spits. Moreover, a lateral difference in sediment characteristics was observed within the upper MSA and MSA/LSA transitional spits, casting doubt on assemblage integrity (Kaplan 1990, p. 6). Subsequent optically stimulated luminescence (OSL) dating showed moderate amounts of overdispersion (intergrain scatter not accounted for by individual measurement uncertainties), which implied that sedimentary mixing of deposits of different ages was limited to relatively small-scale events .
In 2018, we initiated a geoarchaeological project to study the Pleistocene stratigraphy and depositional history of the site. This fits the trend of recent studies that focus on understanding the depositional history of MSA sites in South Africa (e.g., Backwell et al. 2018;de la Peña et al. 2019;Goldberg et al. 2009;Haaland et al. 2017;Henshilwood et al. 2014;Larbey et al. 2019;Lotter et al. 2016;Miller et al. 2013). Our fieldwork has three interrelated goals. The first is to determine whether large-scale sediment movement affected the Pleistocene sedimentary sequence and to arrive at a detailed stratigraphic subdivision of the sediments. The second is to assess preservation conditions at the site, using geochemical methods, and therefore evaluate the previous conclusion that organic preservation was poor in the Pleistocene deposits (Kaplan 1990). Third, we seek to assess whether, and to what degree, the shelter surface was sloped during deposition. Answering these questions is key to contextualizing the assemblages from the previous excavations, and to establishing whether assemblages from spits at the same horizontal level but different squares are contemporaneous.
We use cluster analysis of piece-plotted archaeological finds, and sedimentological, mineralogical, and geochemical analyses, to achieve the above objectives. Our results replicate many of Kaplan's (1990) observations but do not support his idea of the large-scale depositional movement of sediments. We propose a new division of the Pleistocene sequence into stratigraphic units and conclude that the shelter's surface appears to have been largely horizontal. This in-depth understanding of Umhlatuzana's sedimentary sequence highlights the importance of the site for studying the development of modern human technological behaviors in southern Africa.

Site Setting
Umhlatuzana rockshelter is located between Durban and Pietermaritzburg in KwaZulu-Natal, South Africa (Fig.  1). The site is a shallow, northeast-facing rockshelter 47 m long, 8 m wide, and at the point of the excavation, 17.5 m high. It is located in the Umhlatuzana River valley, approximately 100 m above the current riverbed, Fig. 1 Map of South Africa, showing the location of Umhlatuzana rockshelter and its relative position to other important Middle and Later Stone Age sites discussed in the text within Natal Group sandstones and quartzites (~490 Ma; Marshall 1994). The lithostratigraphic characteristics of the sandstones make them relatively resistant to erosion and weathering processes, resulting in a rugged landscape with abundant overhanging walls. The river has been incising a deep valley since the early Pliocene (King 1982). Based on river incision rates in South Africa (Erlanger et al. 2009), the site was always high above the river bed during its phase of human occupation.
The site is currently situated in a coastal scarp forest (dominated by C 3 vegetation), with grassland (dominated by C 4 plants) occurring on the flat table-lands nearby (Mucina and Rutherford 2006). Located in South Africa's summer rainfall zone, the region's annual rainfall ranges from 750 to 1,350 mm, with 75% of the total occurring between November and March (Nel 2009). Although sheltered by large trees, the rockshelter receives sunlight throughout much of the day (Fig. 2). During rain showers, we observed that the deposits are protected by the rockshelter overhang and remain dry.

Research History
The site was discovered in 1982 by Dr. Rodney Maud during the construction of the Johannesburg-Durban N3 highway. It was excavated in 1985 by Jonathan Kaplan (1989Kaplan ( , 1990, who opened a trench of 6 m 2 . He excavated 4 m 2 of this to the bedrock (~2.5 m below the shelter surface) and the other 2 m 2 to 1.5-m depth. He documented a Holocene sequence containing finely stratified sandy deposits that included combustion features. However, from~50 cm below the surface, stratigraphic divisions were no longer visible, and this part of the sequence was excavated in artificial spits of 5-10 cm.
The uppermost part of the occupation sequence is dated to the last 2,000 years. It is represented by final Later Stone Age assemblages, with indications for contact with Iron Age societies in the form of pottery sherds and Achatina beads (Kaplan 1990). The uppermost Pleistocene and lowermost Holocene sediments contain a Robberg industry. This industry, characterized by the production of standardized bladelets and a paucity of retouched pieces, is correlated with South African MIS 2 occupations, 12-18 kya ; for a perspective from eastern South Africa, see Bader et al. 2020). At Umhlatuzana, Kaplan obtained radiometric ages for the uppermost Robberg sediments that were younger than those known elsewhere in South Africa. He, therefore, called these assemblages late Robberg. Further, the radiometric ages for the lowermost Robberg levels at the site were older than those known from other sites in southern Africa. These were, therefore, dubbed early Robberg. Both the early and the late Robberg at the site fit the standard definitions of Robberg technology (Kaplan 1990, p. 85).
The transitional MSA and LSA lithic assemblages underlying the Robberg were also excavated. These assemblages are important as few sites in the region have the record of this transition. The MSA assemblages contain hollow-based points and are called Late MSA by Kaplan (1990). At Sibudu, assemblages containing hollow-based points are characterized as final MSA by Wadley (2005), while the underlying assemblages characterized by unifacial points are called Late MSA (Villa et al. 2005;Wadley 2005; also see Conard et al. 2012;. Below the Late MSA, Kaplan (1990) described assemblages belonging to the Howiesons Poort technocomplex, and below that, the Still Bay. Both of these technocomplexes have been associated with the early appearance of the so-called modern human behaviors (Henshilwood 2012). The Howiesons Poort in Umhlatuzana is characterized by the Fig. 2 View of Umhlatuzana rockshelter looking northwest production of backed segments that were likely hafted using complex recipes (Lombard 2007). The analysis of segments from Umhlatuzana and Sibudu led to the conclusion that bow-and-arrow hunting was practiced during the Howiesons Poort Phillipson 2010, also see Lombard 2011). This period has also been associated with the development of innovative foraging behaviors such as trapping (Lombard 2011;Wadley 2010; also see Dusseldorp 2012). Not only the foraging strategies were innovative, but the production of a bow-and-arrow set was even more complex than other weapon systems ever known before that time (Lombard and Haidle 2012). The Still Bay at Umhlatuzana is characterized by the production of bifacial leaf points (Högberg and Lombard 2016a, b; also see Villa et al. 2009). At Blombos Cave in southwest South Africa, the Still Bay is associated with engraved pieces of ochre and the production of jewelry, demonstrating the presence of early symbolic behavior (Henshilwood 2012). The Still Bay technocomplex is also associated with "high-tech" behaviors. At Umhlatuzana and Blombos Cave, pressure flaking was practiced to aid in the production of finely shaped edges of points and for thinning bifacial points (Högberg and Lombard 2016a). Further, at Blombos Cave, the silcrete used in leaf point production was heat-treated to improve knapping suitability (Mourre et al. 2010).
With its Howiesons Poort and Still Bay assemblages, Umhlatuzana contributes to understanding the development of both technological sophistication and symbolic behavior. Single grain OSL ages were obtained for the Still Bay, Howiesons Poort, and Late MSA sediments . The most populous of the equivalent dose (D e ) clusters identified from finite mixture modeling analyses produced ages consistent with those of other Howiesons Poort (6 0,000 BP) and Still Bay (~70,000 BP) sites .
Site inspections in 2014 revealed that the site had been vandalized; sandbags had been removed from the trench, and illegal excavations had taken place. The northern profile had been undercut by 70 cm up to 2.20 m depth, and the eastern section had been cut back by around 30 cm (Anderson 2014). The reports of vandalism, combined with the potential destruction of the site as the South African National Roads Agency prepared to widen the N3 highway, prompted the current investigation.

Stratigraphic and Taphonomic Issues
Kaplan documented a 2.5-m stratigraphic sequence (Kaplan 1990). Within the Pleistocene part of the sequence, one major stratigraphic boundary, between socalled Red Brown and Purple Brown Sand deposits, was observed. The Red Brown Sands (RBS) occupy the north side (i.e., shelter mouth) of the site, and the Purple Brown Sands (PBS) are present toward the back of the shelter, partially underlying the RBS (Kaplan 1990, p. 7). The contact between them was proposed to represent a single postdepositional event of rotational slipping, whereby during a wet phase, a large block of sediment shifted downwards along the talus slope (Kaplan 1990, p. 6-7). The archaeological materials excavated by Kaplan (1990) were collected in 5-10 cm-thick, horizontal spits.
The preservation of organic material in these deposits varies. Charcoal and bone are well-preserved in the Holocene deposits (Kaplan 1989(Kaplan , 1990. In the upper part of the Pleistocene deposits, faunal and charcoal remains are largely absent; they become slightly more numerous in some of the lowermost layers (Kaplan 1990, p. 68-70). An appraisal of the preservation conditions and the geochemical environment is needed to determine if the changing abundance of organic materials within the Pleistocene sediments is a product of changing human behavior or taphonomic processes.

Excavation Protocol
We conducted excavations in June-August 2018 and July-August 2019. Surface material was collected, and the profiles of the trench excavated by Kaplan (1990) were cleaned and documented. The geoarchaeological analysis of the deposits was then carried out; this focused on the western profile ( Fig. 3) for several reasons: & Detailed sediment descriptions and chronology of the western profile were provided in the initial excavation report (Kaplan 1990). & The proposed postdepositional sediment movement was recorded in this section. & The northern profile was in poor condition due to damage from vandalism.
& The eastern profile is poorly visible due to tree roots and vandalism. & The southern profile is stepped and therefore has limited exposure.
First, the excavation profiles were inspected for lateral and vertical differentiation. To achieve this, strong white artificial lighting was rigged to ensure optimal visibility and optimal description of the Pleistocene sediments. Sedimentary units were defined and characterized by their color (using the Revised Standard Soil Colour Charts 2010), texture, strength, plasticity, stickiness, and presence of inclusions (based on Kyrillidou [2006] and references therein). Additional observations on the nature of the stratigraphic boundaries and the presence of disturbances (i.e., roots, bioturbation) were recorded. The 1:10 scale stratigraphic drawings were created before and after the excavation. The thorough description of the sedimentary sequence aimed to define stratigraphic subdivisions and provide a further line of evidence as to the depositional and postdepositional history of the site. Visually identified stratigraphic entities were classified as units and identified by Arabic numerals.
Using a Sokkia Robotic Total Station (RTS), we identified the original grid system developed by Kaplan (1990). Following the original grid, three new squares of 50 × 50 cm 2 were projected and subsequently excavated on the western side of the previous trench, L2a, L2b, and L3a (Fig. 4). Square L3a was excavated up to 2.40 m depth (bedrock), and squares L2a and L2b were both excavated up to 2.04 m depth. Where possible, the natural stratigraphy was followed. Thick stratigraphic units were subdivided into 2-cm spits to ensure stratigraphic control over finds and sieved materials.
Natural materials and artifacts > 2 cm (henceforth, "finds") were piece-plotted. All excavated sediments were systematically dry-sieved using nested 1 mm, 2 mm, and 5 mm sieves and labeled by square, stratigraphic unit, and spit. This ensured the recovery of small materials. The flotation of the sieved material was implemented to recover macrobotanical remains. Fabric information was recorded for all finds > 5 cm using multiple measurements with the RTS. The dip and orientation of the finds were plotted using linear measurements for elongated finds (> 5 cm with an elongation index higher than 1:7, Bertan and Textier 1995).

Sampling Protocol
Twenty-three bulk sediment samples from the main stratigraphic units were collected for particle size and geochemical (pH, organic matter content) analyses (Table 2). Twenty-one archaeological sediment samples from the Pleistocene and Pleistocene/Holocene boundary deposits were collected from the western profile and analyzed for phytolith content and mineralogical composition. Five control samples were additionally collected to analyze their phytolith and mineralogical component: two from the top surface sediments and three modern surface soil samples from the vegetated areas outside the shelter's edge. Lastly, 28 samples were obtained for stable isotope analysis, consisting of 24 from the archaeological deposits and four from the sediments on the shelter's surface.

Sediment Analyses
This section covers particle size, pH, Fourier transform infrared spectroscopy (FT-IR), organic content, elemental, and stable carbon isotope analyses.

Particle Size Analysis
To determine the character of the sediments, their source(s), mechanism(s) of sedimentation, and depositional environment(s), we analyzed the particle size distribution of 23 sediment samples (Karkanas and Goldberg 2018). We used a Helos KR laser diffraction sensor following the preparation protocol of the Free University Amsterdam Sediment Laboratory for "middle coarse sediments" with < 30% carbonate. Approximately 5 g of material was first treated with 5 ml of 30% H 2 O 2 . Demineralized water was added to 100 ml, and the solution was boiled; 5 ml of 10% HCl was added, and the samples were boiled, diluted with water, and left standing overnight. After decantation down to 50 ml, the suspension was filled up to 100 ml with water. About 300 mg Na 4 P 2 O 7 ·10H 2 O was added, and the samples were boiled. Particle size distributions were then characterized following Folk and Ward (1957).

pH Analysis
We conducted pH analysis to comprehend variability in the preservation of organic materials. The levels of acidity versus alkalinity of the sediment samples are Fig. 3 a Overview of the western profile of the K squares excavated by Kaplan (1990), prior to the 2018-2019 excavation highlighting the difference between the higher-and lower-moisture sediments and Pleistocene/Holocene boundary. b West profile of squares L2a, L2b, and L3a. c Upper hearth feature (units H2, H3, and H4). d Unifacial point from the Late MSA  Kaplan (1990). Squares L2a, L2b, and L3a excavated in 2018-2019 important indicators of the preservation conditions for organic materials (Pollard et al. 2007). The measurements were carried out in a 1:2 sediment:water solution, using a Fisher Scientific Accumet AB150. Three measurements were averaged for each sample. pH values are highly affected by the type of sediment, anthropogenic activities, and postdepositional microbiological activity whereby bacteria and fungi produce organic acids that reduce the pH of the sediments (Weiner 2010). All of these influence diagenetic processes on materials like bone, wood, phytoliths, mollusks, etc. (Garrison 2003). High pH levels (alkaline conditions) are associated with increased bone preservation while low levels (acidic conditions) result in better preservation of silica materials such as phytoliths (Garrison 2003;Pollard et al. 2007;Weiner 2010).

FT-IR Analysis
Fourier transform infrared spectroscopy (FT-IR) analysis was used to identify the bulk mineral components of a total of 26 samples to understand (1) the conditions that may have affected the state of preservation of phytoliths and (2) the mineralogical composition of the sediments. FT-IR has been extensively used in archaeological research for the identification of both crystalline and amorphous minerals, including organic materials (i.e., bone), minerals and precipitates (e.g., clay minerals, carbonates, sulfates, phosphates, nitrates), wood ash (e.g., pyrogenic calcite, charcoal, and opaline phytoliths), and materials exposed to elevated temperatures (see Weiner 2010). Samples were ground with an agate mortar and pestle. Spectra were obtained using an ALPHA platinum ATR single reflection diamond module (Bruker Alpha series) spectrometer. Phase identification was performed using the software OPUS 7.5 from Bruker by consulting standard literature (e.g., Madejová 2003;Müller et al. 2014;Sherman Hsu 1997;Vahur et al. 2016). The reference collection of FT-IR spectra of standard materials provided by the Kimmel Center for Archaeological Science, Weizmann Institute of Science (http://www.weizmann.ac. il/kimmel-arch/infrared-spectra-library) was consulted.

Organic Content Analysis
To determine the organic content of the sediments, loss on ignition (LOI) was employed using a Leco TGA 701. The organic content was measured at 550°C.
Comparative measurements of the organic matter throughout the sequence yield information on the preservation of the site and the presence of organic-rich formations such as palaeosols (Goldberg and MacPhail 2006). Anthropogenic activities also usually resulted in deposits with high organic content (Karkanas and Goldberg 2018).

Elemental Analysis and Stable Carbon Isotope Analysis
The total nitrogen content (TN), total carbon content (TC), total organic carbon (TOC), and δ 13 C TOC were determined via elemental analysis. Since part of the carbon in sediment samples is potentially geogenic, samples were analyzed with and without acid treatment. The acid-insoluble carbon (TOC) is considered the best representation of the organic carbon present in the sediments. TOC speaks to organic input/preservation in the sedimentary sequence and provides a point of comparison to LOI results. The ratio of 12 C to 13 C isotopes is expressed through the δ 13 C TOC metric (Ambrose 1986) and was obtained from the acid-treated samples.
In archaeological contexts, organic carbon as recorded in TOC and δ 13 C TOC is derived from both natural as well as anthropogenic inputs, such as firewood, and plant materials brought to the site (e.g., bedding, lipids from food preparation). Human agency may, therefore, impact the isotopic signatures of the plant materials observed at the site. Given the broader questions concerning the present site's stratigraphy and postdepositional history, we use the TOC and δ 13 C TOC analyses to explore whether isotopically and/or geochemically distinct units can be identified (regardless of precise cause). These results are considered in light of data from contemporary soils in the immediate surroundings of the rockshelter.
Samples were sieved to < 2 mm, freeze-dried, and then ground in a ball mill. Measurements for TOC involved an additional treatment with dilute hydrochloric acid (centrifuged, decanted, and repeatedly rinsed in deionized water) prior to freeze-drying. The samples were then encapsulated in tin cups and analyzed using a SerCon ANCA GSL elemental analyzer interfaced to a SerCon Hydra 20-20 continuous-flow isotope ratio mass spectrometer. All analyses were carried out in triplicate with a typical precision of~0.05% for elemental concentrations and 0.1‰ for δ 13 C TOC .

Phytolith Analysis
Phytolith analysis focuses on the preservation of the phytolith assemblages and their implication for the taphonomic history of the deposits.

Phytolith Extraction
Phytolith extraction followed the Katz et al. (2010) fast extraction procedure at the palynology laboratory of the Evolutionary Studies Institute, the University of the Witwatersrand. An initial sediment weight between 30 and 50 mg was required. Carbonate minerals were dissolved adding 50 μl of hydrochloric acid (6 N HCl). Later, 450 μl of 2.4 g/ml sodium polytungstate solution [Na6 (H 2 W 12 O 40 )·H 2 O] was added. The tube was vortexed, sonicated, and centrifuged for 5 min at 5000 rpm (MiniSpin plus, Eppendorf). The supernatant was subsequently removed to a new 0.5-ml centrifuge tube and vortexed. For examination under the optical microscope, an aliquot of 50 μl of the supernatant was placed on a microscope slide and covered with a 24 × 24-mm coverslip. Quantification of the total phytoliths present in 1 g of sediment was based on 20 fields of view at × 200 magnification, whereas morphological identification of phytoliths took place at × 400 magnification using an optical microscope (Olympus BX51). A minimum of 200 phytoliths was counted for the morphological analysis (Albert and Weiner 2001).
When this was not possible, geometrical traits were followed. Descriptions and naming of the phytoliths follow the International Code for Phytolith Nomenclature 2.0 (ICPN 2.0; Neumann et al. 2019).

Taphonomical Analysis
To determine the degree of preservation of phytolith assemblages, correlation coefficients using the nonparametric Spearman's correlations and their p value were calculated between the phytolith concentration per gram of sediment and four taphonomic indicators: (1) the percentage of weathered morphologies (i.e., phytoliths with signs of chemical dissolution; Esteban et al. 2018); (2) the percentage of fragile morphologies (Esteban et al. 2018, and references therein); (3) the diversity of the phytolith assemblage (number of morphotypes identified; Madella and Lancelotti 2012); and (4) the percentage of broken grass silica short cell phytoliths (GSSCPs) bilobate for each unit. This method measures the strength and direction of the association between two ranked variables (phytolith concentration and the taphonomic indicators). All statistical procedures were performed with the JMP SAS14.2.1 software.

Cluster Analysis
Visual inspection of the excavated sequence suggested patterns in the distribution of artifacts. For instance, there is a readily apparent higher find density in the lower layers. To objectively assess spatial patterns of varying density, statistical cluster analysis of the pieceplotted finds was undertaken using the HDBSCAN algorithm (Campello et al. 2013). HDBSCAN or Hierarchical Density-Based Spatial Clustering of Applications with Noise is a clustering algorithm that calculates for each point (i.e., a piece-plotted find) the minimum distance that is required to reach a set amount of other points ("m pts "; Campello et al. 2013, p. 163). Subsequently, the algorithm determines the mutual reachability distance, the distance required for two points to reach each other as well as the specified number of further points (m pts ) needed to form a cluster. This, in turn, is used to define the minimum spanning tree, a network of points, connected through the minimum mutual reachability. HDBSCAN, therefore, defines spatial clusters and determines their stability throughout this process. In other words, clusters are defined as groups of points that are persistently interlinked. This algorithm employs only one user-defined input parameter (cf. DBSCAN; Ester et al. 1996) and OPTICS (Ankerst et al. 1999), namely the minimum amount of points needed to form a cluster (m pts ; see below), and it can recognize clusters of complex shape and varying density. For this study, we used the density-based clustering toolset in ArcGIS Pro version 2.3.3 to apply the HDBSCAN algorithm on the three-dimensional point cloud of the piece-plotted finds. We used a consecutive series of m pts values of 2-20, 25,30,50,100,200, and 500 to scrutinize the changing clustering results.

Stratigraphy and Sediments
We divided the deposits into two principal groups: the upper group H, which corresponds to the Holocene deposits documented by Kaplan (1989Kaplan ( , 1990, and group P (Pleistocene), below, characterized by defined stratigraphic boundaries and the presence of anthropogenic and biogenic features. The most prominent sources of anthropogenic sediments are in situ combustion features (Fig. 3), charcoal-rich stratigraphic layers, and ash/concreted ash layers. The color of the group H units ranges from dark to pale brown. In contrast, group P lacks discreet stratigraphic boundaries. We observe a few changes in sediment color and structure. Boundaries are gradual and difficult to pinpoint on the profile, and could not be accurately marked in our field drawings. One noticeable change in sediment color, a darker deposit, is present in the northern part of the profile, laterally bounding a lighter deposit (Fig. 3). This conforms to the disjunction between PBS and RBS identified by Kaplan (1990). These two sediment packages are distinguished by a change in moisture content; the darker sediments closer to the rockshelter wall are more humid than the lighter sediments located closer to the shelter's edge. Kaplan (1990, p. 4, 5) also reported the different moisture levels but did not connect these to the fact that the moisture level affects the color of the sediments. The color of the group P deposits ranges from brown and dark yellowish-brown to black in the high-moisture sediments.
The character of the group P sediments is not uniform, particularly in find density. To more reliably determine boundaries between high-and low-density units, we visually assessed the distribution of plotted finds and employed cluster analysis. The results led to the adjustment of unit boundaries, as well as the definition of a new lower-density unit P2. This unit was identified and described during the excavation, but its boundaries were not visible on the profile. Thus, it was not included in the initial stratigraphic drawing. Table 2 describes the stratigraphic units, and the synthesized stratigraphic assessment is discussed later in the article.

Sediment Analysis
Particle Size Analysis Granulometry analysis shows that the entire sequence consists of sand to sandy loam (Fig. 5). Additional grain size parameters of selected samples are presented in Table 3. The particle size distributions are highly comparable throughout the sequence (Fig. 6). The total range of grain size values is between 100 and 550 μm. The sample average (n = 12) median particle size is 286 μm or 1.8 Phi (σ = 18 μm). All of the samples are poorly sorted (mean sorting 1.8 Phi or 297 μm) and negatively skewed (mean skewness − 0.50). The Holocene sediments are marginally better sorted (1.59 Phi vs. 1.85 Phi; Mann-Whitney U = 4, p < 0.05) and possibly less negatively skewed (average − 0.46 vs. − 0.53, nonsignificant) than the Pleistocene units. Particle size analysis of the rockshelter bedrock was not conducted because disaggregation was difficult to achieve in the laboratory. The typical grain sizes of the Natal sandstones range between 100 and 490 μm, with an average of 250 μm (2 Phi) (Bell and Lindsay 1999). The Umhlatuzana sediments thus adhere closely to the grain size distribution of the bedrock (Fig. 6).
pH Analysis pH values range from 4.6 to 9.1 (Table 2). These are weakly patterned throughout the sequence. The pH of the uppermost units (group H) is close to neutral, with values ranging from 7.2 to 7.6. The values associated with combustion features have an alkaline pH, ranging from 8.1 to 9.1. The upper part of group P units (P1-P5) demonstrates nearly neutral pH, with the average at 7.3. In the lower part of group P, the pH values drop both in the higher and lower-moisture sediments (5.1 in higher-moisture unit P14 and 4.6 in lower-moisture unit P13). The average pH of the lowermost units (P12-P17) is 5.3.

FT-IR Analysis
Our analysis shows that the overall mineralogical composition of the shelter's sediments is very homogeneous. The major mineralogical components of the 26 samples analyzed are clay (i.e., kaolinite) and quartz. Calcite (calcium carbonate), a common mineral in many archaeological sites, was virtually absent in the Pleistocene samples. The control samples collected from modern surface sediments, near the rockshelter, have traces of apatite, probably phosphate mineral carbonate hydroxylapatite (also called dahllite).

Organic Content Analysis
LOI results (Table 2) demonstrate the samples contain between 1.6 and 6.1% organic matter. The organic content demonstrates some patterning throughout the sequence. The uppermost surface layer (H1) contains the highest organic content, ranging from 4.1 to 6.1%. The rest of the group H units average 3.5%. A reduction in organic content is observed for stratigraphic units underlying the Pleistocene-Holocene boundary. These drop to 1.6% in unit P1 and gradually increase in the lower sequence. The average organic content for units The mean grain size for the rockshelter (Natal sandstone) was derived by Bell and Lindsay (1999) P1-P5 is 1.8%. Organic content reaches 5.2% in unit P13. The average organic content for units P12-P17 is 4.1%. No clear lateral patterning is observed.

Elemental Analysis and Stable Carbon Isotope Analysis
Total organic carbon (TOC) content confirms that the organic content is generally low, with~1% range through the sequence, spanning from 0.3% (unit P4) to 1.2% (units P9-P12). The lowest and least-varied TOC contents are seen for P3 to P4b (average 0.41%). TOC content is higher for the lower units. This trend is also seen in the LOI results. TN is positively correlated (r = 0.70, p = < 0.0001) with TOC and tends to be higher (and more varied) values in units P6 and below. The average δ 13 C TOC values for units P3-P4 and P5-P8 are identical (− 22.2 ± 0.4‰ compared to − 22.5 ± 0.4‰). Units P9-P14 have a slightly more positive average value (− 21.8 ± 0.6‰) but are also more varied and are not statistically different from units P3-P8. The Holocene samples H9b and H10 have markedly more negative δ 13 C TOC values (− 24 ± 0.6‰) compared to the strikingly invariant group P values (average − 22.1 ± 0.7‰). Thus, the values for group P are more positive than the modern soils and sediments sampled around the site (average, − 26 ± 2‰; n = 4). We note that samples from within the closed woody (C 3 ) vegetation downslope of the rockshelter (i.e., samples 2-4 δ 13 C TOC − 26 ± 1‰) are akin to the group H measurements, implying largely C 3 vegetation inputs during the Holocene (Table 4).

Phytolith Analysis
All analyzed samples contained phytoliths in varying quantities, ranging from 148,000 (unit P8) to 2,780,000 (unit P8) phytoliths per gram of sediments (hereafter, /g sed) (Table 5). Control samples from the shelter surface had the highest phytolith concentrations, followed by the samples from unit P3, P8, and P5/P6. Control samples from modern soils outside the shelter are next in phytolith abundances. Three of the four examined taphonomic indicators showed no significant correlation with phytolith abundance. The exception is the proportion of weathered morphotypes, which showed a moderate negative correlation (r 2 , − 0.468; p value, 0.0159) with phytolith concentration (Fig. 7). This indicates that samples with lower phytolith concentrations contain a larger proportion of weathered morphotypes (Table 6). Nonsignificant correlations were found for the proportion of broken bilobates (weakly positive correlation; Table 6), the total number of morphotypes, and the percentage of fragile morphotypes (both moderately positive correlation; Table 6).

Cluster Analysis
The results of the cluster analysis indicate that four interfaces between dense and sparse zones of finds are significant. By using 2 or 3 as m pts values (i.e., very few finds required to form a cluster), the number of clusters exceeds 400. Values from 4 to 9 yield a similar result in which nearly all the finds belong to one allencompassing cluster, but with small (< 20 finds) additional clusters located in the back of the excavated area. From an m pts value of 10 onwards, the finds are persistently divided vertically into three dense areas, separated by sparse areas (Fig. 8). The uppermost dense area overlaps layers P1-P5; it disconnects into separate smaller pockets of high density at m pts value ranges beyond 14. This means the general area may be dense, but its individual pockets are poorly connected to one another. Within this, a less dense area visible mostly at 10 m pts corresponds approximately to unit P2 (Fig. 8). Below it is a sparse wedge-shaped area, which ends abruptly at the top boundary of the second dense area. This dense area has sharply defined boundaries, overlaps units P6-P11, and is denser and far more homogeneous than the layers above. Below is the sparse but well-defined zone, units P12 and P13, followed by the third dense area (P14-P17), which is similar in characteristics to units P6-P11. The results suggest that the sediment below P5 is relatively undisturbed as they correspond to a reverse arch horizontal layering. The arching of the boundaries between dense and sparse zones is more pronounced in the bottom, and less so in the top, perhaps illustrating the topology of the sedimentary surfaces. The more heterogeneous appearance of the higher located dense area may be the result of postdepositional disturbance through, for example, animal burrowing. Figure 9 shows the synthesis of the stratigraphy of the western profile based on the results of the above analyses. Due to unclear stratigraphic boundaries in the Pleistocene sequence, piece-plotted measurements of finds were employed to distinguish high-and low-density stratigraphic units. The stratigraphic units from both groups H and P demonstrate horizontal layering. A more detailed description of the two groups, starting from the youngest to the oldest, follows below.

Group H-the Holocene
This group of units coincides with the sequence that Kaplan (1990) dated to the Holocene. & The uppermost surface deposit (unit H1) is a darkbrown sand layer that covers the entire excavation area and has a total thickness of~20 cm. Three distinct facies can be recognized within the unit, namely H1a, H1b, and H1c. Facies H1b is characterized by the presence of white, millimeter-sized, and probably calcitic inclusions. Leaves and roots are found within unit H1, indicating a very recent age. & Underlying unit H1, a hearth/combustion feature (H2-H4) extends across most of the west and north profiles. The boundary between units H1 and H2 is sharp and clear. The hearth consists of three layers, from top to bottom: unit H2, a yellowish-gray ash layer; unit H3, a black layer; and unit H4, a darkbrown layer. Heated bone is abundant in ash-rich unit H2. This layering is typical of in situ combustion features with a white uppermost layer of ash, a black layer characterized by heated organic material, and a lowermost red/brown layer of rubified sediment (Mentzer 2014). The contacts between the three hearth layers are sharp and partially bioturbated by insect burrows (likely ant lions, currently present at the site) (Fig. 3). Fig. 9 Stratigraphic drawing of the western profile exposed in squares L3a, L2b, and L2a. On the right: selected calibrated radiocarbon (black) and OSL (red) ages in years BP (Kaplan 1990. Group H (Holocene) unit subdivisions are based on lithostratigraphic characteristics of the layers. Group P (Pleistocene) unit subdivisions are based on a combination of lithostratigraphic characteristics and find density mapping & A dark-brown loamy sand, unit H5, underlies the hearth feature. It is characterized by poorly sorted charcoal remains that comprise around 5% of the deposit. Unit H5 extends throughout the excavation area and has a thickness of 10-20 cm. In some areas, unit H1 directly overlies unit H5. Several dug-out and bioturbation features are present within unit H5. & A second hearth feature (unit H6) is articulated within unit H5. Animal burrows are visible in it, and it consists of three smaller layers. The uppermost unit H6a is a light-gray ash layer, a coarse dark-brown layer (H6b) underlies H6a, and the bottom layer (H6c) is characterized by a dull-orange sediment color. Animal burrows were visible within these layers. Compared to the H2-H3-H4 hearth, H6 is a smaller-scale hearth feature with around 4 cm thickness and 20 cm width. & Unit H7 underlies unit H5; the contact between them is clear, abrupt, and bioturbated with visible smallscale tunneling. Unit H7 is an indurated ash layer with a dull-brown color and thickness of~5 cm.
& Unit H8 is a brownish-black loamy sand layer. It underlies unit H5, and the boundary between them is gradual and bioturbated. The thickness ranges between 5 and 15 cm. Unit H8 contains very few charcoal and bone inclusions randomly distributed. At a corresponding depth, Kaplan (1990) obtained a Late Holocene radiocarbon date (see Table 1) from the unit he called fine brown sand with ash, likely corresponding to our unit H8. & Unit H9 is a layer of distinct aggregates of finegrained material, moderately cemented for the most part, and often loose and friable. The boundaries between unit H9 and both overlying unit H8 and underlying unit H10 are sharp, irregular, and discrete. The color of unit H9 is dull yellow-orange, and its thickness ranges from 5 to 12 cm. Unit H9 spans much of the excavated area and can be found throughout the western, northern, and part of the eastern profiles. It is also characterized by the presence of heated bone fragments. & The lowermost unit of the Holocene layers is unit H10. The boundaries of unit H10 are sharp and discrete with both unit H9 and Pleistocene unit P1. The color of the sediments is dark brown, and its thickness reaches up to 20 cm. Unit H10 contains both charcoal and bone inclusions but at a low frequency. An early Holocene radiocarbon date (Table 1) was obtained by Kaplan (1990) from what he described as the orange-brown sand with ash level. This level possibly corresponds to our unit H10.

Holocene-Pleistocene Boundary
The boundary between the group H and the group P deposits is clear and well-defined on the stratigraphic  Table 3 Particle size parameters of selected representative samples. Following Folk and Ward (1957) Mean  266  211  245  214  228  171  185  225  197  239  292  195   Median  303  262  282  281  283  263  276  284  284  283  326  308   Sorting  350  323  367  295  307  249  257  298  266  316  310 227   (Fig. 3). This was also clearly observed by the initial excavator (Kaplan 1990) who obtained a radiocarbon date of 10,379 ± 104 cal BP (Pta-4307) from his layer 4, corresponding to our unit H10, while the underlying layer 5 in Kaplan's original work (our unit P1) yielded a date of 16,326 ± 439 cal BP (Pta-4226). All of this suggest a hiatus in sedimentation between group H and group P deposits (Kaplan 1990;. The Holocene and the Pleistocene deposits contrast in several other aspects. The find density for the Holocene deposits is low, as shown in the find distribution plots (Fig. 8). Grain size analysis suggests different texture patterns within the two groups, with a slightly higher percentage of clay for the Pleistocene deposits (Fig. 5). Moreover, the results of pH and LOI indicate a clear pattern between the two groups (Fig. 10). Group H deposits are characterized by higher pH and lower organic matter values, while group P has lower pH-higher organic matter values. A difference between group H and group P deposits is also implied by the δ 13 C TOC results (Table 4).

Group P-the Pleistocene
This group of stratigraphic units dates to the Late Pleistocene (Kaplan 1990).
& The uppermost unit is P1, brownish-black loamy sand. It has a thickness of~12 cm and is characterized by a high density of finds. Patches of orange sediments were observed during the excavation. This unit corresponds to the red-brown sand with ash in the original excavation. A radiocarbon date of terminal Pleistocene age was obtained at this depth (Table 1; Kaplan 1990). & Unit P2 is very similar to P1. The contact between them is not visible in the profile, and this unit was not initially drawn in the field sketch. Its different sedimentological characteristics became clear during excavation. Unit P2 is 5-10 cm thick and characterized by a low find density. The sediment of P2 is more compact than P1 and the underlying unit P3. Data from the find density plots were used to specify the location of this layer for the final stratigraphic drawing. & Unit P3 is a dark-brown layer with a width varying from 16 to 18 cm. It is characterized by a higher density of finds and the presence of a few heavily weathered bone fragments. & Unit P4 underlies P3 with a gradual and smooth boundary. It has a dark-brown color, and the thickness is 15-18 cm. The unit is firmer compared to unit P3 and has a lower find density. & Unit P5 is also a low find density unit. The boundary with P4 is vague and was not clearly visible in the field. The sediments of unit P5 have a dark-brown color; its thickness varies greatly from 5 cm in the southern part to 20 cm in the northern part. Unit P4 is firmer than P5, which is also characterized by the presence of many modern rootlets.
Underneath unit P5, a lateral color difference was observed, mirroring the RBS and PBS highlighted by Kaplan (1990). The sediments on the south side of the profile appear darker in color and clearly contained more moisture, while the sediments on the north side were lighter and drier. The boundary between the dark and light sediments was diffuse. A discrete color difference was recorded in the field (i.e., P7: 10YR1,7/1 black, P8: 10YR3/3 dark brown). However, when dried, the colors were more alike (P7: 10YR4/3 dull yellowishbrown, P8: 10YR5/4 dull yellowish-brown).
& Unit P6 underlies unit P5 on the south side (high moisture). The sediment is loose and contains a high density of finds and a total thickness of around 20 cm. This unit likely corresponds to the purplebrown sands observed by Kaplan (1990). A radiocarbon date of 27,800 ± 780 BP (Pta-4389) was obtained at a corresponding depth (Kaplan 1990). & Unit P7 is black loamy sand with a high density of finds. It has a thickness of 18 cm and underlies unit P6. The exact location of their contact is not clearly visible, but the sediments in P7 are softer and darker in color. & On the northern side of the west profile, unit P8 is dark brown with a high find density. With thickness that is 18-32 cm thick, P8 underlies unit P5, is laterally adjacent to units P6 and P7, and overlies units P10 and P11. None of its contacts with the surrounding units were clearly visible in the profile. & P9 underlies unit P7. It is located in the south, highmoisture area, and has a thickness of 15-21 cm. It is characterized by a high find density and a dark reddish-black color.
& Unit P10 has a relatively lower moisture content than P9 and also has a high find density. It underlies unit P8 and is laterally differentiated from unit P9 to the south and from unit P11 to the north. Unit P11 is a dull yellowish-brown sandy loam with a high find density. Units P10 and P11 are both around 20-25 cm thick. They are distinguished by sediment characteristics: unit P10 is softer compared to the firmer sediments of P11. A radiocarbon age of 30,100 ± 1,800 BP (Pta-4228) was obtained at a corresponding depth from Red Brown Sands XIII (Kaplan 1990). & Unit P12 has high moisture content and a low find density. The thickness ranges from 15 to 18 cm. It is adjacent to unit P13, which is characterized by lower moisture, loamy sand of 18-20 cm thickness, and a low find density. The boundaries of units P12 and P13 with their surrounding units are diffuse. Unit P13 corresponds to Kaplan's Red Brown Sands XVI from which a suite of radiocarbon dates with calibrated ages between 39,000 BP and 44,000 BP (Kaplan 1990) and an OSL age of 42,000 ± 3,000 BP  were obtained. & Unit P14 is approximately 15 cm thick, underlying unit P13. It is characterized by dark reddish-black sand with a high find density. The contact between units P14 and P13 dips by~20°to the north and is diffuse. & Unit P15 has a high find density and is laterally adjacent to P14 and below unit P13. The lower contact of unit 15 has not been excavated; its maximum exposed thickness is 15 cm in the north part of square L2A.
& The high-density unit P16 consists of dark reddishblack sands of 20 cm thickness, and underlying unit P14. During excavation, white, centimeter-sized inclusions with a fibrous structure were observed in the south part of square L3a. Analogous to other sites, they are interpreted as gypsum deposits (cf. Pickering 2006). FT-IR and micromorphology analyses of these deepest layers will be undertaken to confirm this. At a corresponding depth in the Red Brown Sands, not reached in our excavations, an OSL age of 60,000 ± 4,000 BP was obtained . & The lowermost exposed unit is P17 and had a very high find density, especially on the lower part (contact with bedrock). It underlies unit P16 and has a thickness of 17 cm. & The lower parts of the Pleistocene sequence (units P14-P17) were suggested to represent lag deposits because of their sometimes very high find density (Kaplan 1990, 12). However, no clear indications for erosional events or truncations were observed in the sequence.

Archaeological Observations
The Holocene sediments yielded both Iron Age and Later Stone Age material culture (Kaplan 1989 Kaplan (1990) and the absence of backed segments are consistent with the attribution of these lowermost H units to the final Later Stone Age ). Both units H10 and P1 yield bladelets. Larger artifacts such as blades and a convergent scraper were observed. This is in line with earlier collections Kaplan (1990, p. 41-44) attributed to the late Robberg. A number of piece-plotted bone fragments were also recovered from unit H10. In unit P1, bone preservation was much poorer, and only a few severely weathered bones were observed. Units P3-P5 yielded materials consistent with the assemblages attributed to the early Robberg (Kaplan 1990). Bladelets and bladelet cores were recovered, but larger pieces such as large flakes and a few blades were also present. No large retouched artifacts were observed. Unit P6 yielded both bladelets and characteristic MSA forms, including a hollow-based point (cf. Mohapi 2013). In unit P8, laterally to the south of unit P6, a similar mix of bladelets and MSA elements such as large blades and a unifacial point was observed. Unit P8 also yielded a number of backed segments. A scan of the lithic materials suggests unit P7 contains a more typical MSA assemblage with unifacial points and has a noteworthy presence of large debitage products and backed segments. Bladelets, although rare, were also recovered. Kaplan (1989) notes the presence of segments in the Late MSA and MSA/LSA transition layers but states that their presence increases in the Howiesons Poort assemblages. In our excavations, unit P11 yielded a large number of segments. This unit corresponds to the Late MSA described by Kaplan. Segments were also recovered in substantial numbers from deeper units P15 and P16. These correspond to the Howiesons Poort layers from Kaplan's (1990) excavations. The lowermost unit P17, directly above bedrock in square L3a, yielded a broken part of a point, bifacially worked with a serrated edge. This fits with previously described Still Bay materials from the site Högberg and Lombard 2016a).

Discussion
Our renewed excavations confirm many of the observations previously made by Kaplan (1989Kaplan ( , 1990 and . That is, Umhlatuzana contains a clearly defined Holocene sequence that contrasts with homogenous Pleistocene deposits that are characterized by diffuse stratigraphic boundaries. The site also preserves a long MSA and LSA occupation history. We demonstrate that the absence of visible stratigraphic indicators does not mean that a stratigraphic subdivision cannot be made. Field observations indicate differences in the physical characteristics of the sediment (i.e., strength and plasticity) across the sequence. We supplement these sediment characteristics with an analysis of find density to arrive at stratigraphic subdivisions. This approach parallels that taken at the nearby site of Umbeli Belli, where the Pleistocene sediments are also largely undifferentiated (Bader et al. 2018, p. 734).
The granulometric analysis shows a homogeneous sediment composition throughout the sequence, characterized as loamy sand. This composition closely matches the grain size distribution of the Natal sandstones ( Fig. 6; Bell and Lindsay 1999), suggesting a mostly closed depositional environment in which geogenic inputs mainly originate from in situ weathering of the rockshelter sandstones. This may partly account for the gradual and diffuse stratigraphic boundaries. Important in situ weathering input is also seen at other sites, such as Pinnacle Point 5-6 (Karkanas et al. 2015) and Mwulu's Cave (de la Peña et al. 2019).
There are no indications for aeolian, fluvial, or gravity flow sediments introduced into the sequence. Fluvial inputs were not expected as the river level is 80-100 m downhill. Sheetwash processes from surrounding slopes also appear absent as the fine bedding produced in such sedimentary regimes was not observed. Aeolian sediments are characteristically well-sorted; their presence would have resulted in the second peak of fine-grained sediments (see de la Peña et al. 2019). The sediments are consistently poorly sorted and negatively skewed (fine tail) and show a single peak (Fig. 6). The absence of other sedimentological input results in a homogeneous matrix with an essentially identical texture and composition throughout the sequence. In comparison to nearby sites such as Umbeli Belli, angular roof spall is not abundant (Bader et al. 2018). No clast-supported layers were present, and the matrix overwhelmingly consists of loamy sand throughout the sequence. The FT-IR results also indicate the homogeneity of the group P deposits, confirming the granulometry results. The presence of apatite in the two control samples from the shelter's surface is likely the result of recent human activities. The present-day use of the area for overnight shelter was observed during fieldwork by the presence of hearths with food remains and makeshift tents.
Organic artifact preservation is poor in the Pleistocene sediments. Bone remains in Pleistocene deposits were poorly preserved and scarce (n = 239 out of a total of 8,210 measured finds in the Pleistocene). A few bone fragments located in square L2a, unit P11, appear to be unweathered, implying that postdepositional processes did not affect the whole sequence equally. The organic content of the sediments is higher in the Holocene deposits (average 3.88%), is lower in the upper part of the Pleistocene sequence (average 1.83%), and then increases again in the lower part of the Pleistocene sequence (average 3.98%). The elemental analysis (comparison of the carbon content of acid-treated vs. untreated samples) reveals that most of the carbon content of the sediments is organic. This confirms FT-IR data showing carbonates to be (largely) absent. The relatively low organic contents of the site's sediments implied by the LOI analysis are mirrored by the low overall TOC content. The rather invariant TOC (also sometimes reported as % C) contrasts with sediments of other South African rockshelters, some of which show far greater variability (e.g., Collins et al. 2017;Loftus et al. 2015;Roberts et al. 2013).
The pH values range from 9.1 within combustion features in the Holocene deposits, to as low as 4.6 in the Pleistocene deposits. This is a high variation in comparison to the other rockshelters of similar geoarchaeological contexts (the type of bedrock, climatic conditions, time span, etc.). For example, Gledswood Shelter 1, a rockshelter developed in a sandstone bedrock with a~2.5 m Pleistocene occupation sequence, has a much lower degree of pH value variation, ranging from 5 to 5.5 (Lowe et al. 2016). The pH results of Umhlatuzana raise two questions: what caused the variation of the pH values and how are they related to preservation conditions throughout the sequence? One possibility is that the pH values are influenced by the variant moisture content within the sediments (e.g., higher-vs. lower-moisture layers). Regarding the preservation conditions, there is indeed a connection between high pH values and better bone preservation at the Holocene layers and poor bone preservation and lower pH values in the Pleistocene. However, for nonosseous organic matter, the reverse relationship holds. Especially in the Pleistocene, the pH and LOI are negatively correlated in strong and statistically significant terms (r, − 0.95; p < 0.01), and the Holocene layers show a similar, but less significant correlation (r, − 0.77; p, 0.02) (Fig. 10). A more systematic sampling strategy in the future is projected to shed more light on the exact trend.
The modern soils immediately outside the shelter contained lower phytolith densities than some of the archaeological sediments. This supports an anthropogenic origin of the archaeological phytolith assemblage. Phytolith preservation is good in relatively acidic conditions (e.g., Piperno 1988), and here they are wellrepresented in the Pleistocene deposits. Samples with high and low phytolith concentrations are present in both the low-and the high-find density layers. The presence of up to 20% weathered morphotypes in some samples indicates chemical alterations did affect the phytolith assemblage and led to the partial dissolution in some samples. The taphonomical analysis suggests that although phytoliths underwent partial dissolution, it was not extensive enough to affect the whole dataset. Therefore, the phytolith assemblage at Umhlatuzana partially represents the initial plant composition. The anomalously high presence of broken bilobates in some samples is intriguing. One explanation is that these are the result of episodes of trampling.
Discrete high-and low-find density zones were observed in the field and confirmed by cluster analysis of the piece-plotted finds. These zones cross-cut the boundary between high-moisture sediments (PBS sensu Kaplan 1990) on the south side of the trench and low-moisture sediments (RBS) on the north side, suggesting that no sediment slump or rotational slip took place. A Kruskal-Wallis test shows phytolith content does not differ significantly between high-moisture and low-moisture sediments (H, 0.007; p, 0.94). However, a comparison of the percentage of weathered morphotypes shows that there is a significant difference between the samples from the dark and wet deposits (PBS) compared to those from the dry deposits on the north side of the profile (H, 9; p < 0.01). The samples from the dark sediments appear to be more weathered, possibly due to the higher moisture content.
Cluster analysis identified three main high density find zones, separated by zones of low-find density (roughly corresponding to P4/P5 and P12/P13, Fig. 8).
The lower two dense areas and the sparse zone that separates them indicate a stable trench-wide sedimentation environment. The wedge-shaped zone between the middle and upper dense areas agrees with the impression of trench-wide stability, but its sloped upper and lower boundaries suggest that the stratigraphy cannot be perceived as absolutely level layers of archaeological material. This should be critically considered when studying the archaeological material from the original excavation as they were collected in horizontal spits. Additionally, the finds in the dense upper area are less homogeneously distributed. Cluster analysis indicates that either the depositional processes for these layers were less stable or postdepositional processes disturbed a layered initial deposit.
The different high-and low-find density layers may reflect periods of higher and lower intensity of human occupation, but they could have also resulted from lower and higher sedimentation rates (cf. Reynard et al. 2016). On current evidence, the upper low-find density zone may be explained most parsimoniously by an increased sedimentation rate (compare ages and depths in Table 1 and Fig. 1). On average, the phytolith content of the lowest units is higher than that of the upper part of the Pleistocene sequence (note that no phytolith data are currently available for the lowermost units). As phytolith input is likely anthropogenic, this could suggest that occupation intensity decreased in the upper units. This hypothesis will be assessed in the future with new dating results and micromorphological analysis.
A preliminary typological analysis of the artifacts mirrors the interpretations of Kaplan (1989Kaplan ( , 1990. In the 2018/2019 excavations, the lowermost layers were only reached in square L3A. These yielded a bifacially worked serrated piece fragment, in line with previously described Still Bay materials from the site Högberg and Lombard 2016b). Segments are present throughout the Middle Stone Age sediments, occurring from units P16 and P15 to unit P7. We have also recovered a variety of unifacial points, mainly in units P11-P8. Some of these are hollow-based points, and similar artifacts have been recovered at Sibudu (Mohapi 2012). The artifacts in the youngest MSA deposits at nearby Umbeli Belli yielded abundant bifacial points (Bader et al. 2016(Bader et al. , 2018. In contrast, we did not recover bifacial points in the uppermost MSA deposits.

Conclusions
The objective of the renewed excavations at Umhlatuzana was to understand the site's depositional history and to establish the integrity of the Pleistocene lithic assemblages. Our results confirm previous observations, such as the presence of Still Bay, Howiesons Poort, and Robberg materials, as well as LSA and Iron Age material culture in the uppermost levels. However, our stratigraphic analysis shows that the disjunction between dark and light sediments in the MSA deposits resulted from differences in moisture content and is not related to large-scale sediment movement. The fact that higher moisture levels occur closer to the rockshelter walls suggests that the moisture could be maintained by minor water flow from the bedrock sandstones to the deposits. Visual inspection of piece-plotted finds combined with cluster analysis suggests the shelter surface was near-horizontal throughout most of the site's depositional history. This means that the assemblages collected in artificial spits during the previous excavations do not represent severe chronologically mixed lithic collections.
Geochemical and FTIR analyses confirm the poor preservation conditions for organic material and the largely homogenous composition of the sediments. The low-pH environment is possibly connected to the paucity of faunal remains in the Pleistocene sequence. The apparent absence of combustion features in the Pleistocene deposits is surprising. At nearby sites of comparable age like Sibudu, they are abundant. It appears unlikely that the lack of such features reflects the absence of fire use at the site. The absence of combustion features is likely due to the postdepositional destruction of traces of fire. This may have had the added effect of removing visible stratigraphic layering. A combination of sediment homogenizing bioturbation activities and the effects of changing hydrology may have caused postdepositional alterations. This is consistent with the mostly homogenous geochemistry and sedimentology of the site (e.g., δ 13 C TOC , lithology).
The exclusion of large-scale sediment movement demonstrates that the intermediate character of the MSA/LSA transition assemblages recovered from the site is not the result of postdepositional mixing of MSA and LSA materials. The virtual absence of sloping in the piece-plotted high-and low-density zones also confirms that assemblages recovered in horizontal artificial spits do not cross-cut original depositional surfaces and can be studied as internally consistent assemblages.
Micromorphological and geochemical analyses are planned in the future to further address the preservation conditions across the archaeological sequence. Additional OSL dating will be undertaken to improve the chronological resolution of the sequence, as well as to analyze the character and prevalence of small-scale sedimentary mixing events.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.