Late Pleistocene shell midden microstratigraphy indicates a complex history of human–environment interactions in the uplands of northern Vietnam

Abstract

Northern Vietnam is situated on a major route of Pleistocene hominin dispersal in East Asia, and the area's karstic caves preserve many prehistoric shell middens. Fossil and genomic evidence suggest a complex human history in this region and more widely across Southeast Asia and southern China, but related archaeological investigations are hampered by challenging site stratigraphies. Recent investigations of shell middens in other geographical settings have revealed the microstratigraphic complexity of these anthropogenic deposits. But caves promote distinctive site formation processes, while tropical climates may catalyse geomorphic and diagenetic changes. These environmental factors complicate the interpretation of northern Vietnam's shell middens and constraining their effects upon the formation, preservation and destruction of these sites is critical to understanding the archaeology of this region. We examine two archaeological cave sites, dated to the Late Pleistocene and located in the limestone uplands surrounding the Hanoi Basin. Each contains multiple shell midden layers associated with prehistoric occupation and burials. Using thin-section micromorphology (microstratigraphy), we reconstruct the depositional and post-depositional histories of these sites, presenting a geoarchaeological framework of interpretation that is applicable to shell middens in mainland Southeast Asia and tropical zones more widely. This work represents a further step towards improving our understanding of prehistoric human dispersals and adaptations in this region.

This article is part of the theme issue ‘Tropical forests in the deep human past’.

1. Introduction

By the marine isotope stage (MIS) 3 (57–29 ka) [1], Homo sapiens had dispersed throughout much of Southeast Asia [2–4], practising diverse subsistence strategies that included pig hunting in the rainforests of Sundaland [5,6] and pelagic fishing evidenced in the faunal record from Jerimalai Rockshelter, Wallacea [7]. Despite occupying a northerly position, the high rainfall and sheltered environments of the upland karst landscapes of northern Vietnam allow diverse forest ecosystems to flourish, including high-canopy tropical rainforest [8–10]. During glacial periods, when rainforest communities in lowland zones became fragmented [2,11], these upland regions sometimes formed forest refugia that were attractive to hunter–gatherer groups [12,13]. Foraging strategies in this seasonal zone focused on risk mitigation through the collection of staple resources [5], such as land snails, probably on a seasonal basis [14]. In the karst uplands of northern Vietnam, archaeological sites dating to MIS 2 and later are often associated with middens of mollusc shell that can form metre-thick deposits in caves [10].

These shell-rich deposits may be indicative of a specialist subsistence strategy that allowed communities to persist in the region throughout the environmental changes associated with the last glacial maximum (LGM; approx. 26.6–19 ka) and the Pleistocene–Holocene transition [15]. But such sedimentary records are difficult to interpret because of a range of issues, including an inadequate chronological framework, stratigraphic relationships that are obscured by the loose, clast-supported nature of the middens and the dynamic site formation processes that characterize the caves of the region [14,16,17]. These issues together ensure that the nature and rhythms of site occupation in this region, along with the long-term processes of adaptation that led to this environmental resilience, remain largely obscure.

Here, we present the results of microstratigraphic investigations of deposits infilling Diêm and Mang Chieng Caves, in Cúc Phu'o'ng National Park, northern Vietnam (figure 1a–c). This work was undertaken to assess the potential for geoarchaeological investigation of these inland shell midden deposits to reveal fine-grained shifts in the depositional environment over time. Such information may prove useful in understanding changing occupation practices and, in future, help to place the population replacements that have been inferred from the genetic data within an archaeological context [16]. We build upon previous microstratigraphic studies of archaeological and experimental shell midden deposits, including those in caves, which have revealed the complex, intertwined archaeological and environmental signals that may be preserved within such anthropogenic sediments [18–20]. These shell middens are deposited in tropical caves, resulting in further confounding factors that hamper their interpretation. Understanding the ways in which tropical environments can modify the archaeological record has been identified as a priority of modern geoarchaeological research [16,21], and a growing number of investigations have focused on constraining the geomorphic processes that operate in tropical cave mouths [22–29]. This work is, therefore, designed to further contribute towards the construction of a geoarchaeological framework to aid the interpretation of archaeological cave sediments in tropical regions.

Figure 1.

(a) Map showing Vietnam in its regional context. The extent of (b) is marked by a red rectangle. (b) Map of northern Vietnam. Tam Pa Ling, in Laos, is marked by a red star. Tràng An landscape complex, in northern Vietnam, is marked by a blue star. The extent of (c) is marked by a red rectangle. (c) Satellite photograph of Cúc Phu'o'ng National Park, with boundaries marked by a red dashed line, showing approximate positions of sites investigated in this study, and Con Moong Cave. (Online version in colour.)

1.1. Archaeological context of Cúc Phu'o'ng National Park and the surrounding area

Modern genetic evidence and palaeogenetic data suggest that the Pleistocene population history of mainland Southeast Asia (MSEA) was characterized by a complex series of dispersals, extirpations and population replacements [3,30–35]. The archaeology of the region has been difficult to link to these narratives, however, due to the regional lithic record's inherent unsuitability for traditional typological analysis [36], as well as the limited preservation of archaeological sediments that date to this period [16,17]. A generalized pattern of ‘Hoabinhian’ hunter–gatherer midden deposition in the Late Pleistocene, with midden reuse as burial locations by early farmers [37], likely masks finer-grained differences in human activity, both symbolic and economic, over time. But due to the nature of the sediments and the challenging conditions of preservation that are characteristics of the region, it is difficult to interpret site stratigraphy or assess the phasing of key features, such as burials, using standard methods [38].

By applying modern scientific techniques of investigation to the molluscan assemblages and archaeological sediments recovered from these upland sites, a number of recent studies, such as those at Hang Boi and Hang Trong in the Tràng An Landscape Complex, Vietnam (figure 1b), have generated new insights into inter-site variability in foraging strategy and site occupation [12,14]. Tràng An has been subject to repeated Quaternary marine inundations but is currently located approximately 45 km from the modern coastline [12]. While one study identified episodes of depositional and post-depositional changes that were likely related to a variety of biological, hydrological and atmospheric depositional agencies [14], the authors noted that they left those processes' inter-relationships unquantified. Microstratigraphic analysis is well suited to investigate the taphonomic effects of changing sedimentary environments and offers a way to build upon previous studies by reconstructing changing site-scale environments and occupation practices at an extremely high resolution [39].

Inland from Tràng An, in the foothills of the Annamite Cordillera, Cúc Phu'o'ng National Park, consists of two, densely forested limestone ranges (150–650 m.a.s.l.) that flank a broad, flat valley that narrows to a canyon in the east [40]. Lowland rainforest is found in Cúc Phu'o'ng's humid central valley, while limestone crests are only colonized by hardier, shorter plant species. Cave systems are present throughout the Park, preserving many archaeological sites [41–44], including Con Moong cave (CMC; figure 1c). This ecologically diverse and archaeologically important area provides a useful study region with which to examine the inland shell middens of MSEA.

Con Moong Cave contains evidence of human occupation stretching back to before 42 ka [22], though not associated with shell midden deposition during this early period. The record at this site indicates that the human settlement of these upland landscapes was not a linear process, as episodes of site abandonment corresponded with episodes of climate deterioration during the Late Pleistocene at approximately 43.5 ka and approximately 30 ka. The later shell middens at that site [45–47], like those at nearby Hang Trong, are associated with occupation that persisted through extreme climate instability [12,13,15]. A foraging economy based on mollusc collection may, therefore, represent a subsistence innovation that allowed a more persistent settlement of these challenging upland environments than preceding periods. The shell middens at Con Moong Cave are considered too fragile to safely extract bulk micromorphological samples, but Cúc Phu'o'ng National Park contains other Pleistocene archaeological sites that may provide useful case studies.

1.2. Study sites and their archaeology

Diêm Cave is situated overlooking a river valley at approximately 150 m.a.s.l. (figure 1c) [44]. The cave is a relict phreatic tube, subsequently weathered under a vadose streamway, with a northeast-facing entrance formed in a limestone cliff (electronic supplementary material, figure S1a,b). Excavations in 2012–2014 identified a series of shell middens (table 1) overlying archaeologically sterile lithostratigraphic units (LSUs A and B) that contain rounded quartz cobbles. Four distinct lithostratigraphic units were recognized in the midden deposits [44]. The lowest of these, LSU C, contained hundreds of flaked stone artefacts and several bone tools (details in table 1). This layer also contained an articulated, female crouched burial. LSU D, overlying, contained cobble and flaked stone tools with the fragmentary skeletal remains of two individuals, while overlying that, LSU E contained flaked artefacts, a bone awl and a decorative, bead-like item made from ochre. These layers are considered to contain material consistent with a hunter–gatherer lifestyle, but above them LSU F contained ceramics with incised decoration alongside a lithic assemblage that is similar to those of the underlying layers. Molluscan analysis is ongoing, but mollusc remains include terrestrial species such as Bradybaena jourdyi and Cyclophorous sp., and preliminary faunal analyses indicate that pig and deer hunting occurred throughout site occupation. A layer overlying this unit, LSU G, was interpreted as modern disturbance.

Table 1.

Summary of field observations, age estimates and micromorphological results for lithostratigraphic units observed at Diêm Cave. LSU, lithostratigraphic units; OSL, optically stimulated luminescence.

stratigraphic unit (LSU)	age estimates	field observations	micromorphological interpretation	archaeological materials
G		friable, dark brown silt containing whole mollusc shells	fluvially reworked occupation waste
F		very friable/loose clast-supported deposit of snail shells with grey-brown silt; capped by thin flowstone	intensive occupation, pyrotechnology and middening in a drier environment; becoming wetter as time progresses	429 artefacts: 15 pebbles. Seven split pebbles, two core-like fragments with the evidence of primary reduction
				365 debitage pieces, including 251 flakes: 16 primary flakes, 39 secondary flakes, 39 fragments, 7 spalls and 13 chips; two stone axe fragments; 39 ceramic fragments
				Human bones tentatively attributed to one individual, subject to carbonization and fragmentation
E	OSL (approx. 22 ka)	brown to reddish-brown friable silt containing shell fragments <1 cm	colluvial redeposition of occupation waste in moist environment; sporadic occupation continues	551 artefacts: two cobbles; five pieces with the evidence of primary flaking, including split cobbles and two core-like fragments; a double-platform bifacial transverse core
				537 debitage pieces, including 402 flakes, 20 primary flakes, 31 secondary flakes, 15 fragments, 16 spalls and 53 chips
				Two convex scrapers and two fragments attributed to convex scrapers; a leaf-shaped flake with the evidence of low-angle partial retouch on the ventral side; two fragmented disc-shaped scrapers (sumatraliths); a stone axe fragment; an awl-like bone tool. Decorative haematite object
D	shell Wk-47718 (23.30–22.95 cal. ka BP) OSL (approx. 22 ka)	friable clast-supported deposit of snail shells with grey-brown silt	increasingly intensive occupation and midden accumulation under drier conditions	477 artefacts: two cobbles, five split cobbles with the evidence of primary flaking; 459 pieces of debitage, including 300 flakes, 19 primary flakes, 40 secondary flakes, 22 fragments, 22 spalls and 55 chips
				Four convex scrapers (two intact and two fragmented), two choppers, two fragmented disc-shaped scrapers, retouched flake, fragments of a stone axe and an adze
C		friable red silt with shell fragments; capped by flowstone	low-intensity occupation in moist, colluvial environment	757 artefacts: 23 intact cobbles. 25 pieces, including 10 split cobbles and 12 core-like fragments. A single-platform unifacial transverse core manufactured from a rounded chunk of limestone and two core fragments that may show evidence of radial flaking
				709 pieces of debitage, including 416 flakes, 28 primary flakes, 32 secondary flakes, 157 fragments, 25 spalls and 49 chips
				Six tools, including end-scraper made on flake, spur-like tool, perforator, a scraper fragment. Two retouched flakes and two retouched lithic fragments. Of interest is an awl-like bone tool produced from a rib tip by planing
				Female burial in an anatomical order towards the top of unit. The skeleton shows that the deceased individual was buried on their side in a crouched position, with the head oriented into the cave. Overlain by flowstone/breccia
B		red, brown and black compact silt-clay	guano and diagenetically altered fluvial deposit
A		white compact silt-clay	fluvial deposit

Mang Chieng Cave was excavated in 2011–2012 [42,43] and is located at an elevation of approximately 203 m.a.s.l. at a greater distance from the river valleys than Diêm Cave (figure 1c). The southeast-facing, arched cave mouth is approximately 12 m wide and 3 m high. The accessible chamber extends horizontally and dips towards the east, covering a total area of approximately 160 m² with a maximum height of 5 m (electronic supplementary material, figure S1c,d). At the cave entrance, a 6 m² area was excavated in 2011, revealing a 50–60 cm thick layer of homogeneous sediments. A total of 345 lithic artefacts were distributed evenly throughout the layer, including grinding slabs associated with ochre processing. Fragmentary, charred skeletal remains attributed to nine individuals were recovered from the basal deposits and have been interpreted as possible evidence for ritual cannibalism [42]. A further 6 m² trench was excavated in the cave in 2012, exposing a further 1.2 m-thick section (table 2) with three LSUs recognized [43]. At the bottom of the sequence, LSU 1 contained fragmentary mollusc shells and a lithic assemblage of 82 artefacts that included split pebble tools. LSU 2, overlying, is a clast-supported shell midden deposit that contained 58 lithic artefacts, while the overlying LSU 3 is a similar shell midden deposit that contained 174 lithic artefacts including split cobble tools (sumatraliths), an arrowhead and notched tools (details in table 2). A polished bone tool was also recovered from LSU 3. Molluscan analysis is ongoing, but mollusc remains include terrestrial species B. jourdyi and Cyclophorous sp. Faunal remains throughout this sequence were carbonized, fragmentary and difficult to identify. In LSU 1, only a few fragments of deer and rodent bone were identified. LSU 2 yielded the most diverse faunal assemblage, including bird, rhinoceros and monkey bones. These species are absent from LSU 3, which is dominated by cervid and bird bones.

Table 2.

Summary of field observations, age estimates and micromorphological results for lithostratigraphic units observed at Mang Chieng Cave.

stratigraphic unit (LSU)	age estimates	field observations	micromorphological interpretation	archaeological materials
3		very friable/loose clast-supported deposit of snail shells with dark grey-brown silt	NA	174 artefacts: 6 cobbles (four hammerstones), 15 pebble fragments, 7 fragments of shale, some of which were modified into tools; 19 pieces show the evidence of primary flaking, including 12 split pebbles and 6 core-like fragments; single-platform unifacial core
				127 debitage pieces, including 71 flakes, 10 primary flakes and 20 secondary flakes; the collection also includes 12 fragments, 13 spalls and 1 chip
				25 tools: transverse, double and disc-shaped scrapers (sumatraliths), as well as those with a working edge that extends for ¾ of the perimeter length; adzes, nosed tool, notched tools, choppers and an arrowhead are present; bone tool with polished edges, showing an oval shape in plan
2	OSL (approx. 18 ka) shell Wk-47719 (18.26–18.09 cal. ka BP)	very friable/loose clast-supported deposit of snail shells with dark grey-brown silt	intensive occupation, pyrotechnology and middening in a drier environment	58 artefacts: two pebbles and two pebble fragments; 16 pieces (split pebbles) show the evidence of primary flaking; one pebble with retouch
				three parallel-sided cores; 38 debitage pieces, including 17 flakes; six secondary flakes and only one primary flake; ten fragments, 2 spalls and 2 chips
				14 tools, including the most representative group of scrapers: disc-shaped (sumatraliths), convergent pieces and those with a working edge that extends for ¾ of the perimeter length; adzes, chopping, axe and chisel-like tool are present
1		yellow-red silts containing shell fragments <1 cm	intensive trample zone with fluctuating moisture conditions and limited guano-driven diagenesis	82 artefacts, including pebble that has been used as a hammerstone, a pebble fragment and a slate fragment
				ten split pebbles (probable cores); 69 debitage pieces, including 38 flakes, 1 primary flake, 5 secondary flakes, 17 fragments, 1 spall and 1 chip
				nine stone tools, including six scrapers (disc-shaped, with a working edge extending for ¾ of the perimeter length); adzes and chopper are present

2. Methods

A range of complementary geoarchaeological and geochronological techniques were employed to study the sediments deposited in the two cave sites under investigation. LSUs were defined based on field sedimentary characteristics (texture, colour, compaction and nature of bounding surfaces) observed in the exposed and cleaned sections [48]. These definitions form the basis of site interpretation and relative chronology. Blocks of the undisturbed sediment were collected from each site, targeting stratigraphic transitions observed in the exposed profiles. Four blocks were taken from Diêm Cave, but only one block survived transport from Mang Chieng Cave to Australia. Block samples were dried, then impregnated with polyester resin under vacuum [49]. The cured blocks were cut into twelve 10 mm-thick wafers. Adelaide Petrographics (Adelaide, SA) produced 12 thin sections (50 mm × 75 mm), ground to standard geological thin-section thickness (approx. 30 µm). Thin sections were analysed using polarizing microscopes at a range of magnifications (8× to 200×). Descriptions of micromorphological features follow Stoops [50].

At both sites, sediment samples were collected for optically stimulated luminescence (OSL) dating, and since no charcoal or plant macrofossils suitable for radiocarbon dating were encountered in the exposed sections, we collected terrestrial snail shell samples instead. Radiocarbon dating of the aragonitic shells was carried out at the Waikato Radiocarbon Dating Laboratory using the same preparation procedures as described in McAdams et al. [22], including X-ray diffraction to confirm that shell recrystallization had not occurred. We used the IntCal20 dataset [51] to calibrate the radiocarbon ages using OxCal 4.4 [52]. We conducted OSL dating on individual sand-sized grains of quartz at the University of Wollongong using the same procedures as those used at the Con Moong Cave [22]. These OSL samples are still being analysed, and the impact of shell material on their external β dose rates is yet to be assessed (e.g. Cunningham [53]), so the final OSL ages will be reported separately.

3. Results

This section presents the summary result of microstratigraphic and geochronological analyses carried out at Diêm and Mang Chieng Caves. Flatbed scans of thin-section samples can be found in electronic supplementary material, figures S2–S13. Micromorphological results are presented in figure 2, and summary interpretative diagrams are presented in figure 3a,b.

Figure 2.

(a) Rounded quartz gravel, marked by red arrow, in a groundmass of phosphatized clays (MM4, PPL). (b) Authigenic phosphate mineral (presumably taranakite), marked by red arrow, precipitated in the zone of clay weathering (MM4, PPL). (c) Rounded, phosphatic clast with isotropic reaction rim, marked by red arrow (MM3A, PPL). (d) Micro-sparitic authigenic calcite crystals, marked by red arrows, precipitated on a bone fragment (MM3A, XPL). (e) Dendritic manganese staining of bone indicative of microbial activity, marked by red arrows, calcite coating marked by white arrow (MM3A, PPL). (f) Calcite coating on the upper surface of clastic materials, marked by white arrows (MM3A, PPL). (g) Calcite crystals with ash-rhomb-like morphologies, marked by red arrows, within the internal pores of a fish vertebra (MM3B, XPL). (h) Indurated fabric unit, marked by red arrows, containing shell fragments (MM2A, PPL). (i) Rounded soil aggregate, marked by white arrow, with ashy coating, marked by red arrows (MM2A, PPL). (j) Mollusc shell with compound coating of laminar calcite, marked by red arrow, overlain by silty clay, marked by white arrow (MM2B, PPL). (k) Fragmentary mollusc shell with thick coating of ashy silt, marked by red arrow (MM2C, PPL). (l) Rounded bone fragment altering to secondary phosphate minerals, marked by red arrow (MM1A, PPL). (m) Shell fragment altering to secondary calcite, marked by red arrows, along exterior (MM1B, PPL). (n) Multi-phase coatings on fragmentary clastic materials, marked by red arrow (MM1C, PPL). (o) Authigenic calcite, marked by red arrow, precipitated within the fracture of elongate bone fragment (MM1C, PPL). (p) Horizontally oriented fragments of mollusc shell (MC1A, XPL). (q) Flatbed scan of thin-section sample MC1A, showing horizontally oriented shell fragments within groundmass (as in electronic supplementary material, figure S11). (r) Fragmentary, horizontally oriented shell fragments, marked by red arrows, in a bioturbated, ashy groundmass (MC1B, PPL). (s) Phosphatic alteration of trampled bone fragment, marked by red arrow, at LSU1/LSU2 boundary (MC1C, PPL). XPL, cross-polarized light; PPL, plane-polarized light. (Online version in colour.)

Figure 3.

(a) Schematic interpretative diagram of Diêm Cave. (b) Schematic interpretative diagram of Mang Chieng Cave. (Online version in colour.)

3.1. Dating

Preliminary chronologies indicate that both sites were occupied during MIS 2, although the earliest occupational levels are undated at present. At Diêm Cave (figure 4a), the preliminary OSL ages of approximately 22 ka for two sediment samples collected from LSUs D and E are broadly consistent with the calibrated radiocarbon age range (95.4% confidence interval) of 23.30–22.95 cal. ka BP (Wk-47718) for a B. jourdyi (terrestrial gastropod) shell sample from LSU D. At Mang Chieng Cave (figure 4b), the preliminary OSL age of approximately 18 ka for a sediment sample from LSU 2 is similarly consistent with the radiocarbon age range of 18.26–18.09 cal. ka BP (Wk-47719) for a B. jourdyi shell sample from this LSU. Terrestrial snail shells from limestone caves are usually not ideal for radiocarbon dating, as snails can ingest ancient carbon from the limestone. However, the consistency between the radiocarbon ages and preliminary OSL ages suggests that this issue may not be significant at these two sites.

Figure 4.

(a) Photograph showing an exposed, northeast-facing section of the excavated trench in Diêm Cave (electronic supplementary material, figure S1a). LSUs are labelled as A–G, pink rectangles show positions of micromorphological block samples. Blue circles mark locations of sediment samples collected for OSL dating and their associated ages (in ka, uncertainties at 1σ). Pink circles mark locations of shell samples collected for radiocarbon dating and their calibrated age ranges (cal. ka BP, 95.4% confidence interval). (b) Photograph showing an exposed, southeast-facing section of the excavated trench in Mang Chieng Cave (electronic supplementary material, figure S1c). LSUs are labelled as 1–3, white rectangle shows the position of a micromorphological block sample. Blue circles mark the locations of an OSL sample and age. Pink circles mark the location of shell samples and calibrated radiocarbon age ranges. (Online version in colour.)

3.2. Diêm Cave

3.2.1. Stratigraphic observations and geochronology

A series of deposits were revealed in the northeast-facing section (figure 4a), corresponding well with published observations [44]. Field descriptions are presented alongside archaeological observations from previous studies in (table 1). The basal unit, LSU A, is a compact, whitish-grey sand-clay with a diffuse upper boundary. The overlying unit, LSU B, is a very firm sandy silt-clay with vivid red, brown and black lenses. This unit is overlain by a friable red silt, LSU C, containing shell fragments. This unit has a diffuse and partially indurated interface with the overlying LSU D, a friable, clast-supported sediment of mollusc shells with brown silt in interstitial spaces. To the right of the excavated profile, a flowstone-capped pedestal of LSUs C, B and A remains unexcavated. LSU D has a diffuse interface with the overlying LSU E, a brown to reddish-brown friable silt containing unoriented shell fragments less than 7 mm. LSU E's upper boundary with the overlying LSU F is a relatively sharp interface. LSU F is a loose, clast-supported deposit of mollusc shells with grey-brown silt in interstitial spaces. This layer is capped by a thin flowstone deposit, above which is a friable, dark brown silt, LSU G, which contains numerous shells.

3.2.2. Microstratigraphic results and interpretation

Block 4 (LSU A). Coarse sand- to fine gravel-sized rounded quartz grains are present within a groundmass of oxide-stained, weakly birefringent clays (figure 2a). Elsewhere, quartz is present as silt or fine sand-sized grains that form discrete concentrations within a largely undifferentiated groundmass. Weakly porostriated clays are distributed unevenly throughout the thin section, and greyish, authigenic mineral nodules (likely taranakite) are forming in areas of advanced clay weathering (figure 2b). Calcite coatings are present in some voids.

Interpretation. Despite severe post-depositional diagenesis, the rounding of the quartzose coarse fraction (electronic supplementary material, figure S10) and the well-sorted nature of the fine fraction indicate that the sediments recorded in this thin section were fluvially deposited, laid down when the stream that eroded the cave passage was active. These fluvial sediments were subsequently modified by guano-driven diagenesis in a moist and very acidic environment, evidenced by clay weathering and phosphate mineral authigenesis [54,55]. Calcite precipitation in pore spaces would have occurred subsequently when organic decomposition had ceased and the sedimentary pH had been buffered by carbonate-charged groundwater [56].

Block 3 (LSUs C and D). LSU C, visible in the lower portion of MM3A, consists of an indurated sediment containing a poorly sorted, chaotic mix of material. The groundmass is a porous arrangement of oxide-stained, reddish-brown, clay-silt excremental microaggregates. Subangular to sub-rounded quartz sand and a range of phosphatized aggregates with well-expressed isotropic hypocoatings are common (figure 2c). Bone and shell fragments have silt-clay coatings and bone fragments show a range of post-depositional modifications and coatings (figure 2d,e). The upper surface of this unit, the interface between LSU C and LSU D, is associated with an incipient flowstone coating, while calcite void infills are common and coarse materials, including bone and rock fragments, frequently have calcite coatings on the upper surface (figure 2f). The overlying unit, LSU D, is more porous, with little interstitial calcite precipitation. Fractured and weakly oriented shell and bone fragments are common. Above this, the groundmass is an oxide-stained, poorly sorted silt-clay. Passage features are ubiquitous and bone fragments and intact shells are common. But some bone and shell fragments have blackened surfaces and sand-sized bone fragments often have reddish coloration. Faunal remains are frequently coated with authigenic calcite crystals, some of which have ash-like morphology (figure 2g).

Interpretation. The heterogeneous material comprising LSU C, which was coated in a fine sediment and rounded during the transport, is indicative of low-energy colluvial sedimentation in a wet environment. The phosphatic materials in the groundmass are likely reworked, and there is little evidence that phosphatic diagenesis has affected the preservation of materials deposited in this sediment. The weak anthropogenic signal in this unit is, therefore, an indicator of low-intensity occupation in this part of the cave at the time of deposition. Carbonate precipitation is extensive, and the sediments are strongly indurated towards the upper boundary. LSU D has a much stronger anthropogenic signal. The horizontal orientation of shell fragments likely indicates a trample zone just above the LSU C/D interface, perhaps after a period of abandonment that allowed the incipient flowstone to form. Interestingly, while this shell midden is evidently an anthropogenic deposit, charcoal is absent and microstratigraphic evidence for food processing is equivocal. This may indicate that pyrotechnology and food processing were conducted elsewhere in the cave, or were not the focus of occupation activity at this site when this unit was deposited.

Block 2 (LSUs D and E). Within LSU D, at the base of MM2A, indurated sediments form a discrete fabric unit (figure 2h). This is overlain by spongey, reddish-brown clay-silt, containing a chaotic mix of whole and fragmentary shells, burnt and unburnt bone fragments, soil aggregates and charcoal. The groundmass grades greyer moving up the thin section, and clastic material often has calcareous silt coatings (ash-like in appearance) up to 1 mm thick (figure 2i). The interface between LSUs D and E is visible within MM2B. The lower unit, LSU D, consists of spongey, calcareous silt with complete shells and a variety of heated and unheated bone fragments (inferred from variations in colour). An overlying clay-silt unit, part of LSU E, contains fragmentary shell with rounded bone and speleothem fragments, all with approximately 1 mm thick clay-silt coatings. The only large shell fragment within this unit has a compound coating of laminar calcite overlain with clay-silt (figure 2j). The lower section of MM2C shows an extensively bioturbated silt-clay with a chaotic mix of poorly sorted subangular to sub-rounded sand-sized inclusions. Highly fragmented anthropogenic material is present throughout, including heated bone, associated with thick calcareous silt coatings (likely ash) (figure 2k). Porosity is visibly reduced moving up the thin section. Clastic material in an upper, greyish brown silt fabric unit, including burned bone, shows less fragmentation and moderate levels of sub-horizontal orientation. Carbonate cementation and precipitation in this unit are extensive.

Interpretation. The upper portion of LSU D is characterized by increasing proportions of anthropogenically deposited materials, including processed (heated) faunal remains, combustion residues and pedorelicts. The greyish, microcrystalline carbonate coatings on coarse material (likely recrystallized ashes, as discussed in Mentzer [57]) and limited fragmentation of mollusc shells suggest that the occupation waste was deposited directly after use, with little post-depositional disturbance or chemical diagenesis. We can infer that, during the time, LSU D was deposited, drier conditions prevailed and the site became the focus of increasingly intensive occupation, where a group of people, at least occasionally, carried out the daily activities that made up their lives.

The sharp upper boundary is consistent with an erosional event, probably relating to a change in hydrological conditions and the activation of either fluvial or sheetwash colluvial conditions. Overlying this abrupt contact, LSU E was deposited. In its lowermost extent, thick silt-clay coatings on fragmentary occupation waste indicate that material has been reworked from underlying midden layers during a return to humid colluvial deposition. But LSU E is a varied deposit, and compound coatings, comprising layers of silty clay and carbonate coatings (likely recrystallized ash), indicate that coarse materials within other microfacies probably had very complex depositional histories. It is, therefore, probable that human occupation continued despite the switch to wet colluvial conditions. The zone of reduced porosity and weak orientation of coarse material towards the top of thin-section MM2C, where fragmentary faunal remains are associated with greyish carbonate silt coatings, may indicate a subsequent period of more intensive human activity. But this fabric unit is also cemented with authigenic calcite, and the extent of carbonate diagenesis makes it difficult to assess whether the material was redeposited and fragmented by human or environmental agencies.

Block 1 (LSUs E,F and G). LSU E (thin-section MM1A) consists of a bioturbated, calcareous silt groundmass with rounded coarse components including common thermally altered bone fragments. Some bone and shell fragments exhibit silty clay or calcite coatings, while others are affected by phosphatization, mechanical weathering or bioerosion (figure 2l). Charcoal fragments are common and display varying degrees of humification and in situ fractures. In LSU F (thin-section MM1B), there is an increase in the proportion of whole shells, while burned and unburned bone fragments and fragmentary charcoal are distributed throughout the sediment, with the addition of possible seeds. Carbonate precipitation and diagenesis are associated with bone and shell surfaces (figure 2m), but there is little cementation of the spongey silt-clay groundmass. Ash is concentrated towards the top of the thin section, which is frequently associated with compound coatings on a clastic material. In the upper extent of LSU F, thin-section MM1C is a fabric unit with greatly reduced porosity, within which shells are fragmentary and show little evidence of arrangement or orientation. A range of multi-phase coatings are visible on coarse material within this unit (figure 2n). LSU G, visible as an overlying fabric unit, is associated with an increase in porosity and increased proportions of burned bone and shell are present. Silt-clay coatings are absent from clastic material in this unit, though some material is associated with carbonate coatings and surface recrystallization (figure 2o).

Interpretation. The upper part of LSU E results from ashy occupation waste being deposited in the midden, then reworked through trampling and colluviation under moist conditions, with the incorporation of reworked materials from other parts of the cave. Some coarse material is rounded, which is often an indicator of transportation over relatively large distances or with considerable energy. But given that charcoal fragments are common—albeit some showing signs of in situ fracturing—it is unlikely that a high-energy or prolonged period of colluvial transport was responsible for the deposition of this LSU. Indications of such varied depositional and post-depositional histories likely result from the myriad syn-depositional processes that can act upon deposited material on a cave floor, prior to burial [58].

In LSU F (thin-section MM1B), the increased proportion of whole shells indicates less post-depositional transport and trample of deposited materials, while ash preservation indicates relatively little chemical diagenesis or moisture transport. The chaotically arranged, silt-clay fabric unit at the bottom of thin-section MM1C, however, is an indicator of a return to wet colluvial conditions, and subsequent flowstone precipitation resulted from laminar flow of calcium carbonate-charged groundwater along this non-porous surface. LSU G contains chaotically arranged anthropogenic material that results from reworking of midden material by a reactivated streamway.

3.3. Mang Chieng Cave

3.3.1. Stratigraphic observations and geochronology

LSU 1 consisted of reddish to yellow-brown silts with stringers of finely fragmented mollusc shells and rounded limestone cobbles. LSU 2, overlying, is a loose, clast-supported deposit consisting of mollusc shells and occasional sub-rounded to subangular gravels, with dark grey-brown insterstitial silts. LSU 3 is a similar clast-supported deposit, distinguished by lighter grey-brown silts in its interstitial spaces.

3.3.2. Microstratigraphic results and interpretation

Block MC1 (LSUs 1 and 2). Visible in LSU 1 (thin-section MC1A), a poorly sorted fabric unit with low porosity is overlain by a spongey, oxide-encrusted silty clay with diverse, horizontally oriented coarse inclusions consisting primarily of elongated shell fragments (figure 2p,q), with infrequent charcoal and burned bone. A horizontally oriented limestone clast (greater than 5 cm) is present at the top of the thin section. This clast is overlain by a spongey clay-silt capping (visible in thin-section MC1B). Above this, LSU 1 continues and there is an increase in both porosity and concentrations of horizontally oriented shell fragments, which become increasingly fragmented and oriented towards the top of the thin section. The reddish groundmass of this upper portion of LSU 1 is much less oxide-stained, appearing more porous and containing increased proportions of charcoal and ash (figure 2r). An interface between LSU 1, a reddish silt-clay layer containing a finely fragmented shell, and LSU 2, an overlying grey layer containing more complete shells, is macroscopically visible in thin-section MC1C. Bone fragments in the underlying, reddish unit are partly isotropic on their upward-facing surfaces (indicative of phosphatic diagenesis) (figure 2s), and some horizontally oriented, elongate fragments at the interface display in situ fracture. The overlying unit of greyish silt has greatly increased proportions of fractured charcoal, along with sand-sized burned bone and rounded soil aggregates. Many clastic components have compound coatings of calcareous silt (ash-like in appearance) and silt clays.

Interpretation. The dense, clay-rich fabric unit at the base of LSU 1 (base of thin-section MC1A) remains of uncertain origin. Overlying, the oxide-stained unit containing many shell fragments that show an extreme degree of orientation may be assumed to be a trampled surface that is indicative of relatively intensive human occupation. Charcoal is present, but not in great abundance, possibly indicating that combustion/pyrotechnic activities were either taking place elsewhere in the cave (and being reworked to the study area in minor quantities) or were not a major occupation activity when this unit was deposited. While the reddish coloration and black coatings of some bone fragments suggest heating and burning [59], the extent of redoximorphism suggests that these features may also relate to post-depositional processes [22,60]. While the archaeologically sterile fabric unit resting on the flat rock may have been attached to the rock when it was deposited, it is interesting to note that, within LSU 1, shells in MC1B are less fragmented than those in MC1A. This suggests a period of less-intensive trampling and less-intensive human occupation in the study area at the time of deposition. Combustion residues, including ash, are more common in the upper part of LSU 1, but the lesser oxide staining may indicate that wet/dry cycling has been less intensive in this part of the sediment. The interface between LSU1 and LSU2, visible in MC1C, is associated with a trampled surface evidenced by the fragmentation and orientation of coarse components. Phosphatic alterations, such as those observed affecting bone fragments in this thin section, may indicate guano-driven diagenesis and periods of human abandonment [54–56]. The overlying, clast-supported unit contains a greater abundance of whole shells, with combustion residues intermixed with the interstitial silts. These have not been fragmented because this area of the cave was no longer a focus of trampling and occupation, rather an area where combustion residues and faunal remains (mostly mollusc shells) were deposited.

4. Discussion

Our geoarchaeological study has shown that, despite the geographical proximity and archaeological similarities between the two sites, the shell middens at Diêm and Mang Chieng Caves have contrasting depositional and post-depositional histories. Thin-section analysis has allowed the detection of microstratigraphic variations in a depositional environment that result from the distinct speleomorphological factors and occupation histories that characterize each of these sites. By exploring the relationships between these archaeological and environmental signals, we provide information that is key to understanding the taphonomy of archaeologically important materials in similar sites, but also to distinguishing site types and understanding the relationships between human occupation and environmental change in this understudied—but nevertheless important—region.

4.1. Shell midden formation and degradation processes in northern Vietnam

Some of the microfacies we observe at Diêm and Mang Chieng Caves, and the site formation processes we infer from them, are analogous to those observed in microstratigraphic studies of shell middens in open, coastal settings, including the preservation of directly deposited whole shells, reworking and fracture, pedogenesis affecting the fine fraction and pulsed aggradation [18–20]. But these middens were deposited in tropical caves, which open up a range of additional confounding depositional and post-depositional processes that hamper our ability to extract meaningful information about the subsistence strategies and lifestyles of their past inhabitants [21]. As a result, we recognize a suite of distinctive microfacies that relate to the interaction of tropical karstic geomorphic processes and the anthropogenic sediments that have been deposited in these cave mouths.

Mang Chieng Cave is stratigraphically the simpler of the studied midden sequences; nevertheless, the microstratigraphic evidence suggests that the sediments have been affected by a series of post-depositional changes in the burial environment. Of particular interest is LSU A, which contains oxide crusts that probably relate to cycles of wet and dry episodes [61] and intensive redoximorphic weathering under strongly seasonal, humid tropical conditions. Wet/dry cycling commonly increases the rates of chemical diagenesis and other taphonomic processes, which may have profoundly affected the preservation of archaeological materials in these layers. Ash is absent from this layer, but moist conditions may have encouraged ash crystal dissolution [56]. While charcoal has more preservation potential, it is relatively scarce in this layer. Recent studies have indicated its susceptibility to both mechanical weathering, as a result of wet/dry cycling [23], and diagenesis in alkaline sedimentary environments [62,63]. However, had this occurred we may expect to see a greater quantity of microcharcoal interspersed in the fine fraction. It is, therefore, likely that pyrotechnic activities and the deposition of combustion residue were limited in this area of the cave during the time of sediment deposition.

While shell survives in large quantities in this layer, showing little evidence of dissolution, bone is much less common than in overlying layers. The poor preservation of excavated bones overall was noted in previous studies of this site [42,43]. Bone is a porous material and its dissolution is greatly accelerated by wet/dry cycling [64,65]. Based on microstratigraphic investigations and taphonomic experiments [22,23], it has been suggested that bacteria may be more prominent taphonomic agents in the humid tropics than elsewhere. The phosphatic diagenesis that is recorded at the trampled surface at the upper extent of LSU 1 is likely to relate to guano deposition. Such processes often affect the preservation of faunal remains and combustion residues in archaeological sites in caves [56]. In this case, however, the resulting authigenic phosphate minerals are limited to amorphous apatite reaction rims on the upper surfaces of fragments of bone and carbonate rock, which indicates that phosphatization has not proceeded to a great extent [54,55]. Targeted geochemical studies of diagenetically altered sediments may give a clearer indication of past changes in the sedimentary environment [66].

The sediment record at Diêm Cave is more complex. Changes in the site's hydrology have clearly played a key role in governing site formation, driving both sediment deposition and the pace and degree of post-depositional change. Cultural material deposited under drier conditions is relatively well preserved, and diagenesis and fragmentation of bone and shell fragments tend to be limited in layers with little evidence of fluvial or colluvial redeposition. A similar taphonomic pattern was observed at Liang Bua on Flores, where cultural material was well preserved within the package of highly calcareous sediments that were deposited under drier conditions [67].

At Diêm Cave, increasing waterflow through the cave has resulted in erosional episodes, sediment redeposition and cementation. In midden deposits, layers of a highly fragmented shell may be considered evidence of redeposition related to maintaining space, deliberate construction or trample [20]. At Hang Boi, Rabett et al. [14] propose possible interpretations for the origin of similar layers of crushed shell and clay-silt, which included trampling, episodic mass movements of midden material and infiltration of clays from groundwater seepage. But at Diêm Cave, thick silt-clay sediment coatings are frequently adhered to fragmentary shell. Similar in appearance to the weakly oriented silt-clay ‘flood coatings’ that may infill pores in soils that have been flooded with sediment-laden water [68], these features are particularly prevalent in LSU E, but also in microfacies in LSUs D and F. Because of the position of these midden deposits, in the route of an intermittently active karst streamway, these features are likely evidence of intensive reworking of the midden deposits under hydrologically active conditions that promote wet colluviation. The environmental information recorded in the fine sediment fraction of these shell midden deposits is an important finding of this study. This corresponds to similar findings in recent studies of open-air middens, such as the freeze–thaw cycling inferred from acicular calcite pedofeatures in South American shell-matrix sites [19].

Because this explanation relates to microenvironmental factors at the Diêm Cave, it would be unwise to assume that macroscopically similar sediments elsewhere result from analogous processes. Even within the middens at the Diêm Cave, it is often difficult to assess the relative influence of human and environmental agencies on the deposition and redeposition of a sediment. We may assume that more complete shells with coatings of recrystallized ash relate directly to human pyrotechnology and economic activity, but when faunal remains are associated with compound coatings of grey microcrystalline calcite and clay-silt, it is less simple to reconstruct their depositional history. Episodes of increased hydrological activity also led to the formation of indurated sediments, which form due to carbonate precipitation in the matrix of porous sediments, and flowstones, which form on top of less porous sediments as a result of laminar flow. The complex taphonomy of indurated materials is highlighted by a growing body of studies from tropical caves [69–74], and the differential preservation of carbonate-coated faunal remains (figure 2j) in this study suggests that it may be a source of bias in the recovered assemblage.

The extent of bioturbation recognized at Diêm and Mang Chieng Caves is a further factor that complicates the investigation of sediment delivery and site-scale palaeoenvironments. All investigated sediments at these sites are affected by bioturbation, and the intensive bioturbation of archaeological cave sediments is a characteristic feature of MSEA more broadly [17,23]. Further research on the effects of MSEA's faunal communities, particularly invertebrates, on the stratigraphy of archaeological cave sites has the potential to significantly improve archaeological outcomes in the region. A priority should be to characterize the faunal communities in MSEA's caves, but experimental work can also help illuminate how faunal life cycles may impact stratigraphic and diagenetic processes [75,76].

It is interesting to note that, while previous studies have tended to focus on exploring contexts with either exceptional preservation [77,78] or complete destruction of shell-rich deposits in caves [79,80], this study and others in Vietnam [14] reveal sites with moderate or variable preservation, which suggests that intermediate processes are at work. While comprehensive discussion of midden formation processes across a wide range of environments is beyond the remit of this paper, these results suggest that future study of these questions may prove fruitful.

4.2. Archaeological implications of this work

By comparing the results of this study to similar sites in the surrounding region, we hope to assess variation in chronologies and inferred occupation practices, but several issues make inter-site comparisons difficult. First, geoarchaeological techniques are not commonly incorporated into research methodologies in this region and informal sedimentological descriptions can make it difficult to assess which deposits are actually clast-supported shell middens. Second, only a small portion of these midden sites have reliable dates associated with them.

At Diêm Cave, the earliest clast-supported midden LSU is dated as early as 23.42 cal. ka BP, while the earliest clast-supported midden layer at Mang Chieng Cave has been dated to 18.3 ka, meaning both were deposited around the time of the Last Glacial Maximum (approx. 22–19 ka [81]). At Con Moong Cave deposits of a similar age, ephemeral lenses of combustion residue indicate lower intensity occupation during this period [22], while the deposition of clast-supported midden layers at that site only began after 12 ka cal. BP [45]. The earliest excavated layer at Hang Boi, which contains whole mollusc shells, is dated to 12 362 ± 97 cal. BP [14], but midden layers above older archaeological deposits have been dated to 18 229–18 708 cal. BP at Hang Trong [12]. The earliest layers at Xom Trai, to the North, contain large quantities of mollusc shell alongside human burials and have been dated to approximately 22 000–19 000 cal. BP [82], while shell midden deposits have been dated as early as 39 960 ± 1050 cal. BP at Tham Lod in Thailand [83].

This sample of sites may indicate two distinct phases of the commencement of midden deposition in Pleistocene archaeological sites in Vietnam, one broadly concurrent with the LGM and the other with the Pleistocene–Holocene transition. In both cases, there may be occupation of a different character that occurs beforehand, including ritual activities such as the burials excavated at the Diêm Cave. The undated layer of highly fragmented shell at the Mang Chieng Cave means the possibility of a still-earlier phase of midden deposition persists, and the early dates from Tham Lod suggest that this behaviour may stretch far back into the Late Pleistocene. The reasons for such behavioural variability across a limited geographical area remain obscure. Ecology provides one potential explanation: Tham Lod is at a much higher altitude than the other sites (approx. 640 m.a.s.l. [83]) and its inhabitants may have needed to supplement their diet with mollusc collection at a much earlier stage, due to colder, drier conditions. But Con Moong Cave and Diêm Cave, and Hang Boi and Hang Trong, respectively, are at similar elevations and in similar ecological settings [12,14,22]. It is unclear why Con Moong Cave should see greatly reduced intensity of occupation at the LGM [22], while Diêm Cave, previously a focus of mortuary activity and low-intensity occupation, becomes the focus of midden deposition, resulting in much more intensive occupation and the continuation of ritual activities. This suggests we are observing sites that had different functions as part of a cultural landscape. The potential for labels such as ‘midden’ to obscure inter-site variability has been noted by previous researchers [14], but with the currently available dataset it is not possible to effectively tease apart the differences in economy and settlement practices between these different sites.

It seems that once mollusc consumption and deposition become established at these sites, it remains the dominant mode of sedimentation throughout the hunter–gatherer occupation phases. The midden deposits investigated in this study, like those at Tràng An [12,14], cannot be interpreted as resulting from an extended, continuous episode of human activity. Within single lithostratigraphic units, there may be microstratigraphic evidence for multiple, significant shifts in the mode of occupation, including episodes of site abandonment. This is a common characteristic of midden deposits across a range of settings [14,18–20,84,85]. They are often foci of complex occupation practices, involving intermittent pulses of aggradation and erosion related to natural processes or maintenance, as well as ritual activities such as pit digging [86].

5. Conclusion

This work represents an important step towards understanding how the later Pleistocene shell middens of Vietnam and MSEA relate to processes of adaptation, demographic shifts and environmental change because of the insights into changing occupation practice and site-scale environmental change that have been generated. Thin-section micromorphology is not routinely incorporated into archaeological research in this area, but it offers a powerful tool for understanding the formation of these archaeological features and an opportunity to move beyond the hunter–gatherer middening/early farmer funerary activity interpretative framework.

The archaeological materials excavated from these sites—particularly the evidence for human remains with varied and complex mortuary treatments—may offer useful insights into demographic shifts and processes of a cultural change in the Late Pleistocene uplands of MSEA. But our microstratigraphic results presented in this study highlight the complex, intertwined archaeological and environmental signals that are contained in these sediments at the microscale and demonstrate how these insights can be used to assess the relationships between past human activity and environmental change.

Lastly, our study and others that focus on past human relationships and interactions with tropical forests are important to contextualize modern-day Indigenous groups that thrive and survive in these environments [87]. Given that indigenous peoples' way of life is under increasing threat from climate change and land degradation, our work is important to feed into narratives that advocate for societies that depend on forest ecosystems [88]. It is also important to continue to document the intertwined natural and cultural heritage of these resilient, but increasingly threatened areas [89]. These form major sources of knowledge that are useful to the policymakers and ecologists who are striving to develop a sustainable future for our species.

Data accessibility

This article has no additional data.

Authors' contributions

C.M.: conceptualization, data curation, formal analysis, investigation, methodology, writing—original draft; M.W.M.: conceptualization, methodology, project administration, supervision, writing—review and editing; X.F.: data curation, formal analysis, investigation, writing—review and editing; A.V.K.: conceptualization, investigation, methodology; A.P.D.: conceptualization, investigation, methodology; D.T.N.: investigation; N.G.D.: investigation; R.G.R.: conceptualization, funding acquisition, investigation, methodology, project administration, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Competing interests

We declare we have no competing interests.

Funding

This study was funded by the Australian Research Council through Australian Laureate Fellowship FL130100116 (R.G.R.) and Future Fellowship FT180100309 (M.W.M.), by the University of Wollongong through a Postgraduate Award and an International Postgraduate Research Scholarship (C.M.) and by the Institute of Archaeology and Ethnography, Siberian Branch of the Russian Academy of Sciences (A.P.D.).