Home Is Where the Hearth Is: Anthracological and Microstratigraphic Analyses of Pleistocene and Holocene Combustion Features, Riwi Cave (Kimberley, Western Australia)

The manipulation of fire is a technological act. The identification of the archaeological signatures of the controlled use of fire has important implications not only for the estimations of the origins and functions of the first fireplaces but also for our understanding of prehistoric technological development and resource use. At Riwi (Kimberley region, Western Australia), excavations over two field seasons have revealed a discontinuous occupation sequence over the past 45 ka, showing numerous, different combustion features interspersed within the deposit. Anthracological and micromorphological investigations at Riwi Cave indicate that the combustion features at the site can be categorised into three types: flat combustion features (type A), dug combustion features (type B) and thick accumulations of mixed combustion residues (type C). These provide evidence for two kinds of combustion practice: (i) fires lit directly on the ground and most likely not re-used and (ii) ground ovens, the latter appearing some 10,000 years after the first evidence for occupation of the site. A comparison of the wood species identified within these combustion features with those from equivalent scattered context levels, enables an exploration of the potential factors influencing wood selection and fire use through time at the site. A detailed understanding of the relationship between wood charcoal remains and archaeological context yields significant information on changes to environmental context and site occupation patterns over time.

Abstract The manipulation of fire is a technological act. The identification of the archaeological signatures of the controlled use of fire has important implications not only for the estimations of the origins and functions of the first fireplaces but also for our understanding of prehistoric technological development and resource use. At Riwi (Kimberley region, Western Australia), excavations over two field seasons have revealed a discontinuous occupation sequence over the past 45 ka, showing numerous, different combustion features interspersed within the deposit. Anthracological and micromorphological investigations at Riwi Cave indicate that the combustion features at the site can be categorised into three types: flat combustion features (type A), dug combustion features (type B) and thick accumulations of mixed combustion residues (type C). These provide evidence for two kinds of combustion practice: (i) fires lit directly on the ground and most likely not re-used and (ii) ground ovens, the latter appearing some 10,000 years after the first evidence for occupation of the site. A comparison of the wood species identified within these combustion features with those from equivalent scattered context levels, enables an exploration of the potential factors influencing wood selection and fire use through time at the site. A detailed Introduction An essential component of the hunter-gatherer tool-kit, fire is a source of light, warmth, protection and an instrument for cooking, manufacturing equipment and altering the environment. The origins and functions of the first fireplaces have important implications for hominin evolution and form a key debate in Palaeolithic archaeology (e.g. Alperson-Afil and Goren-Inbar 2010;de Lumley 2006;Goudsblom 1986;Gowlett 2006;Gowlett and Wrangham 2013;James et al. 1989;Roebroeks and Villa 2011;Sandgathe et al. 2011;Wrangham 2009), with the first habitual, controlled uses of fire linked to increases in brain size and cognition (Brain 1981;Gowlett 2006;Pruetz and LaDuke 2010;Rolland 2004;Wrangham 2009) and the colonisation of the northern latitudes (Brace et al. 1987;Gowlett 2006;Oakley 1956;Preece et al. 2006;Rolland 2004;Stratus 1989;Weiner et al. 1998;Wrangham et al. 1999). Evidence for anthropogenic fire can be contextually variable and, in the case of the earliest examples, highly contentious (e.g. Berna et al. 2012; for a recent review on the evidence of human use and control of fire, see Goldberg et al. 2017;Stahlschmidt et al. 2015, pp. 181-183). The most unambiguous signature for the habitual, controlled use of fire is the structured hearth, with the earliest evidence found in Qesem Cave in Israel, dated approximately to 400 ka (Karkanas et al. 2007;Shahack-Gross et al. 2014).
The identification of the archaeological signatures of hearth-building processes has important implications not only for the estimation of the first controlled uses of fire but also for understanding of prehistoric technological development and resource use. A suite of techniques for both the macro-and micro-scale analyses of combustion structures are currently employed, ranging from the in situ description of hearth structures (e.g. Metcalfe and Heath 1990;Solé et al. 2013;Vaquero and Pastó 2001), to the physical and chemical analysis of charred components and sediments with the application of geophysical (e.g. Barbetti 1986;Bellomo 1991Bellomo , 1993, geochemical (e.g. Karkanas et al. 2002;Rudner and Sumegi 2002), micromorphological (e.g. Mallol et al. 2013a;Mentzer 2014;Schiegl et al. 2004;Wattez 1992; cf. review in Aldeias 2017) and anthracological analyses (e.g. Beauclair et al. 2009;Henry and Théry-Parisot 2014;Scheel-Ybert et al. 2014;Vidal-Matutano 2016). The study presented in this paper combines the latter two approaches, anthracology and micromorphology, to explore building processes of hearths from an Australian Indigenous archaeological context.
Anthracological investigations of wood charcoal assemblages follow the premise that hearth features and concentrations of dense charcoal are episodic archaeological contexts. They are the primary refuse of the last few firing events, whereas dispersed wood charcoal from scattered contexts are secondary refuse, potentially accrued over a more protracted period of time (Asouti and Austin 2005;Byrne et al. 2013;Chabal 1990Chabal , 1992Chabal et al. 1999;Dotte-Sarout et al. 2015;Théry-Parisot et al. 2010). Where wood charcoal analyses are employed for palaeoenvironmental reconstruction, hearths and concentrated charcoal features must be avoided, and scattered charcoal from occupation contexts must be examined in order to represent the accumulation of multiple fuel wood collection events, so that more of the site's surrounding environment will be represented (Badal-García et al. 2012;Chabal et al. 1999;Dufraisse 2012Dufraisse , 2014Théry-Parisot et al. 2010;Scheel-Ybert 2002). Concentrated features, such as charcoal lenses and hearths, should be analysed in conjunction with scattered contexts to tease out the cultural factors that influence the creation of a charcoal assemblage at a site (Asouti and Austin 2005;Byrne et al. 2013;Théry-Parisot et al. 2010). However, in spite of this premise which is essential to the discipline, very few anthracological studies have explicitly employed micro-scale techniques to define the combustion context of a site in order to understand combustion processes better (e.g. Allué et al. 2017;Damblon et al. 1996;Damblon and Haesaerts 2002;Vidal-Matutano 2016).
In the Australian archaeological context, various geoarchaeological and archaeological analyses have explored the formation and post-depositional alterations of hearth features at open sites, particularly at Sturt National Park in arid western New South Wales (Fig. 1;Fanning and Holdaway 2001;Fanning et al. 2008Fanning et al. , 2009Holdaway et al. 2017) and Olympic Dam in northeastern South Australia ( Fig. 1; Sullivan et al. 2012;Sullivan and Hughes 2013). Analysis of sedimentary processes within Australian rock shelters has concentrated largely on preservation potential (Ward 2004;Ward and Larcombe 2003;Ward et al. 2006) and the vertical movement of artefacts (Allen and O'Connell 2003;Bird et al. 2002;Hiscock 1985Hiscock , 1990. With the exception of micromorphological analyses conducted at Carpenters Gap 1 rockshelter (Fig. 1;Vannieuwenhuyse 2016;Vannieuwenhuyse et al. 2017), which revealed the presence of combustion features' rake-out zones in the Late Pleistocene and Holocene archaeological levels, few attempts have been made in Australian archaeology to understand how fireplaces were built, maintained and used by Indigenous hunter-gatherer populations. Similarly, very few anthracological investigations have been conducted in Australian contexts, although the number of statistically viable analyses is starting to improve (see Byrne et al. 2013;Carah 2010;Dotte-Sarout et al. 2015;King 2015;Whitau et al. 2016a).
In this paper, we present the results from combined anthracological and micromorphological analyses of combustion features at Riwi Cave in the southern Kimberley region of northern Western Australia. At Riwi, excavations have revealed a discontinuous occupation sequence over the past 45 ka showing numerous different combustion features interspersed within the deposit (Balme 2000;Vannieuwenhuyse 2016;Whitau et al. 2016a;Wood et al. 2016). This sequence represents an exceptional opportunity to undertake a combined and detailed geo-anthracological analysis in order to explore the depositional and post-depositional factors that have affected the creation and preservation of hearths. We propose a typology of these features based on their sedimentological and anthracological characteristics. The wood species identified within the hearths are compared with the spectrum of charcoal identified from contemporaneous scattered contexts, which are more representative of the broad vegetation changes in the vicinity of the site (see anthracological analysis in Whitau et al. 2016a). The combined anthracological and micromorphological approaches enable an investigation of the possible factors influencing wood selection, fire-use and deep-time combustion features related practices in an Australian archaeological context.

Site Description, Context and Antiquity
Riwi Cave is located in the southern Kimberley region of Western Australia (Fig. 1), within the traditional lands of the Indigenous Gooniyandi people, on the northern edge of the Great Sandy Desert. Located in the South Lawford Range and formed within Devonian Pillara and Sadler limestone facies (Playford et al. 2009, p. 251), the cave is situated within a valley enclosed by low-range outcrops, through which an ephemeral creek flows during the wet season. Receiving a sub-tropical to semi-arid climate within the 500-mm isohyet of the Australian Summer Monsoon (Bureau of Meteorology 2015), the skeletal soils of the Riwi valley support a low tree steppe: hummock grassland (Triodia bitextura) with scattered bloodwood (Corymbia dicromophloia/ Corymbia opaca) and snappy gum (Eucalyptus brevifolia). A variety of dry rainforest associated taxa, including Celtis strychnoides, Dodonaea polyzyga and Flueggea virosa grow along the limestone range and outliers (Whitau et al. 2016a). The cave is composed of two chambers, and a channel running through the north-west side of the front chamber, where a yellow ball flower tree (Mallotus nesophilus) grows, signifies water circulation in the cave during the wet season (for site review and plan, see Vannieuwenhuyse 2016;Wood et al. 2016;Whitau et al. 2016a). Vestigial evidence of past human occupation includes rock art on the cave walls and lithic artefacts scattered across the floor of the cave's entrance.
Riwi Cave was first excavated in 1999 (Balme 2000). In 2013, the original 1 m 2 test pit (square 1) was emptied and the excavation area expanded, with the addition of 2 × 1 m 2 test pits inside the cave (squares 3 and 4) and one 1 m 2 test pit at the entrance of the cave (square 5). All squares were taken to bedrock in 50 cm quadrants in arbitrary 2 cm units, reaching an approximate maximum depth of 115 cm in squares 1, 3 and 4 (Figs. 2 and 3). With the exception of certain features, which were removed separately, and bulk sediment samples, which were collected for each excavation unit, all excavated materials were sieved though nested 5 and 1.5 mm screens. Because the Holocene units were desiccated, flotation was avoided. A detailed description of both the cave and the excavation specifics can be found in Whitau et al. (2016a) and Wood et al. (2016).
A precise radiocarbon chronology, coupled with a detailed optically stimulated luminescence (OSL) chronology, provides one of the most accurately dated archaeological sequences in Australia (Wood et al. 2016). The two chronologies are largely consistent throughout the sequence, both identify earliest occupation of the site around 46.4-44.6 ka cal BP (95.4% probability range) at the top of SU12, and both confirm the presence of several chrono-stratigraphic hiatuses in the upper levels of the sequence (Vannieuwenhuyse 2016;Wood et al. 2016). The radiocarbon chronology was mainly based on charcoal sampled from the numerous combustion features interspersed within the Riwi archaeological levels (Figs. 2 and 3). Detailed information about the provenance and stratigraphic context of the dated charcoal was gathered by the geoarchaeological observations undertaken in the field. The anthracological analysis provided background information on charcoal wood species and the context of charcoal production (see Table 1). These data are usually limited or not available in the construction of radiocarbon chronologies in Australian archaeological contexts (Ward et al. 2016), which are often only based on the age of charcoal fragments with little reference to the context of that charcoal's production (Wood 2015;Wood et al. 2016). The limited reporting of these data is in spite of recent work conducted elsewhere that characterises the spatial relationship between charcoal and charred seeds selected for radiocarbon analysis and associated archaeological materials and features (Asscher et al. 2015;Boaretto 2015;Boaretto et al. 2009;Rebollo et al. 2011;Toffolo et al. 2012).

Sampling
The Riwi anthracological and micromorphological samples were collected from the July 2013 excavations and from the archaeological material recovered and sorted in the  A short description is given for each feature with associated combustion features types, stratigraphic context (SU) and related excavation units (XU) from where anthracological (A) and micromorphological (M) samples were collected. When available, radiocarbon dates directly associated with the feature or same stratigraphic level are given. Radiocarbon dates are mentioned calibrated against SHCal13 (Hogg et al. 2013) in OxCal v.4.2 (Bronk Ramsey, 2009) following radiocarbon chronology done for the site by Wood et al. (2016) laboratory. Twelve combustion features from Riwi square 3 were selected for anthracological analysis (Fig. 2) in conjunction with the analysis of scattered contexts from squares 3 and 4 presented in Whitau et al. (2016a). Six micromorphological samples targeting combustion features were extracted from the eastern sections of square 1 (Fig. 3). The proximity of the sampling and the similarities in type of combustion features observed between the two squares allows the comparison and incorporation of both anthracological and micromorphological results to build our analysis. Table 1

Microstratigraphic Methods
Oriented micromorphological sediment samples were extracted using plaster bandages Macphail 2003, 2005), which were then prepared following the standard fabrication processes for soil thin sections (Camuti and McGuire 1999;Courty et al. 1989;FitzPatrick 1984). Resin impregnation of these monoliths was undertaken at the geotechnical facilities of the School of Earth and Environment at the University of Western Australia. Small pre-cut chips (54 × 63 × 10 mm) were sent to Spectrum Petrographics (Vancouver, Washington, USA) for thin sectioning. Thin sections were digitally scanned for archival and publication purposes (Arpin et al. 2002;De Keyser 1999). Thin section observations and microphotography were carried out using a Nikon polarising petrographic microscope in the School of Earth and Environment at the University of Western Australia. Observations were made under different magnifications (×10, ×25, ×50, ×100, ×500) using both plane polarised (PPL) and cross polarised light (XPL). Descriptions follow the terminology standardised by Stoops et al. (2010). Identification and interpretation of components and pedofeatures are based on the available micromorphology literature (primarily Bullock et al. 1985;Courty et al. 1989;Goldberg and Macphail 2005;Stoops et al. 2010) and case studies as cited in the text.

Anthracological Methods
All anthracological analysis was conducted at the Department of Archaeology and Natural History at the Australian National University. Charcoal was identified by snapping fragments along the transverse, radial and tangential longitudinal sections, with the aid of a scalpel where necessary (following Leney and Casteel 1975). An Olympus BH-2 reflected lightfield/darkfield microscope was used to examine exposed sections at magnifications of ×20-500. Rare types and archetypal examples of taxa were selected for further observation and imaging with a JEOL JCM-6000 Neoscope scanning electron microscope (SEM). Following Chabal (1990Chabal ( , 1992 and Théry-Parisot et al. (2010), quantification was conducted by count, rather than weight. All of the charcoal fragments over 2 mm were examined from each feature (see feature list in Table 1), except F1-C and F2-C, which are Riwi's two Holocene stratigraphic units (SU1 and SU2 in Whitau et al. 2016a). F1-C and F2-C are composed of an accumulation of combustion features mixed with other natural inputs, wherein features could not be sampled separately due to their thin morphology and the palimpsest nature of their deposition. All of the excavated sediment from F1-C and F2-C was collected during excavation within squares 3 and 4, with all of the > 1.5 mm charcoal fragments transported to the laboratory for anthracological analysis. A minimum of 300 identifiable fragments was sampled for the Holocene units, and a riffle-box was used to split the samples to ensure an unbiased coverage of size differentiation. Following Whitau et al. (2016a), charcoal fragments are described as indeterminate when type-level identification cannot be positively assigned and indeterminable where fragments cannot be positively identified due to degradation. Archaeological material was compared with the reference material housed at the Australian National University described by Whitau et al. (2016a). Wood identification keys including Hope (1998), Ilic (1991) and two online databases: Inside Wood (http://insidewood.lib.ncsu.edu/) and the University of Queensland Online Archaeology Collections (http://uqarchaeologyreference. metadata.net/archaeobotany/list) were also employed to aid identification.

Results
Based on excavation, sediment and anthracological observations, combustion features in the Riwi sequence have been grouped into three different types ( Fig. 4; Table 1): flat combustion features (type A), dug combustion features (type B) and palimpsest of combustion features (type C). The features analysed in this study are numbered from top to bottom, with a suffix for the combustion feature type. For example, F1-C is the highest feature in the sequence and a type C combustion feature. The sedimentary and anthracological characteristics of each type of combustion feature are detailed and compared in the following sections. Each type, with the exception of F3-A, appears to be found in subsequent chronostratigraphical levels, as demonstrated by the site's sequence (Figs. 2 and 3).

Description of Combustion Features in Riwi Sequence
Interdisciplinary analyses conducted on the site have demonstrated that Riwi was occupied on a regular basis from 45 ka up to the European contact period, despite a visibly discontinuous stratigraphic record (Vannieuwenhuyse 2016;Wood et al. 2016). The combustion features comprise the main non-material evidence for anthropogenic inputs within the Riwi sequence, and as such, provide exceptional insights into past behaviours over time.
The Pleistocene layers (SU12 to SU5, from the bottom to approximately 20 cm below the surface) have an excellent integrity and reveal the presence of numerous flat combustion features (type A; F8 to F14) appearing at the top of SU12, approximately 40 cm from the bedrock (Figs. 2 and 3). Several dug combustion features (type B; F4 to F7) are visible at the top of the Pleistocene sequence and at the bottom of the Holocene layers. In most of the observed sections, a sharp disconformity places Holocene ash-rich deposit (type C; F1 and F2) directly above the orange Pleistocene layers dated to 34-31 ka. However, discrete episodes of sedimentation have been preserved in some places, in particular, SU3, which includes charcoal and hearths (F3-A) dated to the Last Glacial Maximum (LGM, two charcoal dated at 20620-20040 cal BP, see Table 1) and is only visible in the south-east corner of square 3 (Fig. 2, see also figures in Vannieuwenhuyse 2016; Wood et al. 2016). The Holocene layers (F1-C and F2-C) are dated to 7 ka and 0.8-0.7 ka and are mainly composed of an ash-rich accumulation (type C) that encompasses compacted combustion residues (ash, charcoal) and other vegetal organic remains. Preservation of organic material via desiccation within the Holocene layers is exceptional and includes seeds, fruit fragments, paperbark fragments (Melaleuca spp. bark that had and still has a myriad of uses across Aboriginal groups of Australia, e.g. Wynjorrotj et al. 2005;Yunupingu et al. 1995), wood shavings and two wooden artefact fragments (Dilkes-Hall 2014; Langley et al. 2016;Whitau et al. 2016b).

Microstratigraphic Results
The Riwi natural sequence is composed of a mix of geogenic, botanical and animal bone fragments in various proportions (detailed results of the Riwi archaeo-stratigraphical sequence geoarchaeological analysis and a full description of the micromorphological thin sections analysed can be found in Vannieuwenhuyse 2016). While the Pleistocene layers are predominantly composed of geogenic particles, giving the sediment a strong orange hue within which combustion features are easily distinguishable, the Holocene deposit is grey because of the predominant proportion of combusted botanical residues (Figs. 2 and 3). Across the five combustion features sampled for micromorphology (three type A: F11-A, F13-A, F14-A; one type B: F6-B; and one type C: F2-C, see Table 1; Fig.  3), the main combustion by-products observed are vegetal residues (phytoliths, ash, seeds, wood charcoal) and some animal bone fragments (Fig. 5), observed at various burning stages (partially burnt, charred, turned to ash) and in varying states of preservation (depending on syn-and post-depositional modifications, e.g. decomposition, alterations).

Anthracological Results
Across the features sampled (seven type A: F3-A, F8-A, F9-A, F10-A, F11-A, F12-A, F13-A; three type B: F4-B, F5-B, F7-B; and two type C: F1-C, F2-C, see Table 1; Fig. 2), a total of 2824 charcoal fragments were analysed, 1861 of which were positively identified to varying taxonomic ranks. Table 2 lists the positively identified taxa, with their relative frequencies for each feature expressed in terms of absolute fragment counts and the percentages of the total number of fragments analysed. A total of 19 taxa, including two family level identifications (Lamiaceae sp. and Myrtaceae sp.), were determined from nine family groups. A full description of the archaeological charcoal types is presented in Whitau et al. (2016a, Appendix 1). Results are discussed in more detail in 'Types of Combustion Features-Micromorphological and Anthracological Characteristics'.
Saturation curves are presented in Fig. 6, which plot the number of identifiable charcoal fragments against the number of identifiable taxa for each feature. Each plateau indicates     Grevillea/ Hakea sp.
Absolute fragment counts

Types of Combustion Features-Micromorphological and Anthracological Characteristics
The following sections describe the sedimentary characteristics and wood charcoal assemblages identified for each of the combustion feature types.
With the exception of F3-A and F9-A, charcoal preservation from each of the analysed type A features is very poor. The few charcoal recovered from F10-A, F11-A, F12-A and F13-A are soft and rounded, with all of the indeterminable fragments unable to be examined due to brittleness, disintegrating to ash particles and providing no clean sections. Of the 45 charcoal recovered from across these four features, only 9 were identifiable, predominantly Eucalyptus sp. type A, and all of the Myrtaceae family. By contrast, in terms of both composition and preservation, the F8-A feature produced six identifiable fragments, three were identified Erythrophleum sp., two Corymbia sp. and one Flueggea sp., while eleven of the indeterminable fragments were too brittle and six were too vitrified.
The two type A features which produced reasonable anthracological assemblages are F9-A (N i = 158) and F3-A (N i = 101). F9-A, which is located at the bottom of SU7 (Fig. 2) is the only Riwi anthracological assemblage that is not dominated by Myrtaceae: Erythrophleum sp. (39.8%) dominates with subsidiary Vachellia sp. (12.1%), Corymbia sp. (5.2%) and Eucalyptus sp. (3.8%). F9-A is also anomalous in that it is the only Pleistocene unit to produce an assemblage with one individual taxon count (Erythrophleum sp., N i = 92) that is higher than the indeterminable count (72 fragments) for that unit. Observed within SU3-a stratigraphic layer dated from the LGM timing (Table 1) that shows evidence of postdepositional bioturbation (Fig. 2)-F3-A is the youngest type A feature analysed and sits above the type B contexts developed in the next section. The dominant taxon for this unit is Corymbia sp., with subsidiary Brucea sp. and Eucalyptus sp. charcoal. Comprising 8.5% of the total charcoal examined within the F3-A unit, Brucea sp. charcoal were identified here in their highest proportion of the Riwi anthracological assemblage.

Type B Dug Combustion Features (F4-B, F5-B, F6-B, F7-B)
Around 34 ka, type B combustion features are observed in section (Figs. 2 and 3). These dug features sometimes cut through the flat combustion features immediately below, disturbing the orientation of particles (Fig. 8b). Type B combustion features contain large fragments of charcoal embedded within a fine matrix of geogenic sands and/or random oriented ash particles (Fig. 8a).
The three type B features selected for wood charcoal analysis (F4-B, F5-B, F7-B) each have a relatively low percentage of indeterminable fragments, ranging from 33.8% for F5-B to 40.8% for F4-B. Corymbia sp. is the dominant taxon across the three type B features, with subsidiary Eucalyptus sp. and Erythrophleum sp. The feature F4-B, which has the lowest proportion of Eucalyptus sp. (3.9%) compared with F5-B and F7-B (17.6 and 15.3%, respectively), also has a higher proportion of Flueggea sp. (5.0%) than these other type B features, a proportion which is not duplicated in the other Pleistocene contexts.

Type C Palimpsest Combustion Features (F1-C, F2-C)
Type C features are only present in the Holocene deposits and are accumulations composed mainly of by-products of combustion (predominantly ash and charcoal), along with a mix of non-burnt vegetal parts and a minor proportion of geogenic sands (Fig. 9). The high proportion of combustion residues gives the Holocene deposits its grey colour (Figs. 2 and 3). Under a microscope, the microfacies show a very bright calcitic crystallic b-fabric as a result of the high proportion of ash particles (pseudomorphs of plant tissues) (Fig. 9b, d). These particles are found both in anatomic connection (Fig. 9c, d) and disturbed (Fig. 9a, b), which indicates some mixing in this level, resulting from maintenance activities such as reworking (cleaning rake-out of hearths but also human and animal trampling and turbation, see Fig. 4).

Type A Flat Combustion Features (F3-A, F8-A, F9-A, F10-A, F11-A, F12-A, F13-A)
The micromorphological evidence demonstrates that the flat combustion features found in the Pleistocene layers were constructed in the same way, where the fire was lit directly onto the ground surface (Fig. 4). The underlying substrate is affected by heat in two ways. The dark colour of the concave dark brown layer (1) is the result of the carbonisation of vegetal organics already present in the geogenic matrix. Their carbonisation was induced by the heat radiance from the fire lit on the ground surface above the sediments. The orange-pinkish layer (2) corresponds to the surface where the fire was lit and where temperatures were high enough to induce both the total combustion of organics already present in the sediment and the typical reddening of the sediment resulting from the oxidation of iron components (Aldeias et al. 2016;Canti and Linford 2000). In comparison with layer 1, the lighter hue of layer 2 is also a result of the presence of ash from the layers above.
The top of layers 1 and 2 are the remnants of a 'past occupation surface' where the fire was lit (as formulated by Mallol et al. 2013b), where people walked and lived in the cave at a certain time in the past. The layer with charcoal (3) corresponds to the combustion itself and contains big charcoal fragments and carbonised organics. Observation of the charcoal sampled for the anthracology analysis shows that they tend to be rounded and brittle in all of the analysed Pleistocene type A flat combustion features (with the exception of the two youngest features F3-A and F9-A). The poor preservation of the charcoal structure is mainly explained by the properties of the wood charcoal itself (differential preservation depending on species, use of wet or dry wood) as already suggested by Théry-Parisot et al. (2010). The top whitish layer (4) is composed of ash particles that are mostly still in anatomic connection. The anatomic connection of the ash particles indicates that the firing event was in situ, and that these structures were not used repeatedly (Mentzer 2014;Miller et al. 2010). The excellent preservation of these structures was permitted by a relatively rapid burial by aeolian natural sedimentation after each firing episode with evidence of only minor post-depositional processes, as demonstrated by the steady sediment deposition rate of Pleistocene levels (Wood et al. 2016) and the geoarchaeological analysis of the sequence (Vannieuwenhuyse 2016). Aeolian sedimentation, in conjunction with the post-depositional processes observed in the sequence, in particular the formation of pedogenic gypsum, points to relatively dry conditions over the Pleistocene period. Ash particles were not affected by dissolution, which indicates a relative dryness of the sequence through time (Vannieuwenhuyse 2016).
The fact that only few charcoal fragments are found in most type A combustion features (F8-A, F10-A, F11-A, F12-A, F13-A, see Table 2), and that the combustion by-products associated are mostly ash seems also to indicate that the wood burned in these contexts was mostly completely combusted. As such, this type of combustion feature probably indicates short lived activities or short visits to the cave, where the open air fire was left to burn until the fire died (cf. Chabal et al. 1999). Two type A features yielded more than 11 identifiable charcoal fragments: F3-A and F9-A (see saturation curves in Fig. 6). If the type A features represent short-lived, episodic combustion events, then these assemblages would be expected to have a lower diversity than their coincident scattered contexts, since the former are hypothesised to represent one or two fuel wood collection trips, the latter many (Asouti and Austin 2005;Byrne et al. 2013;Chabal 1990Chabal , 1992Chabal et al. 1999;Dotte-Sarout et al. 2015;Théry-Parisot et al. 2010). The F9-A feature, which yielded 158 fragments of identifiable charcoal, has a low taxa diversity (8 taxa). This low diversity is coupled with the lowest proportion of indeterminable charcoal (31.6%) of any of the Pleistocene units, indicating that this low diversity is not an effect of taphonomic factors acting on the assemblage; supporting the hypothesis that the type A hearth structures are episodic archaeological contexts. The F3-A feature, which sits above the type B features in between the Pleistocene and Holocene sediments, is an exception to this pattern as it yields a relatively high number of identifiable fragments (N i = 101) for a type A feature and a diverse assemblage (12 taxa). This feature is discussed in further detail in 'Fuel Wood Management: Comparing Concentrated and Scattered Charcoal Contexts' in relation to its context. Altogether, the micromorphological and anthracological analyses of the type A features indicate that these could have been small open-air hearths typically used as single-firing episodes, for heating, lighting and cooking purposes during short term visits to the site, where the fire was left to burn until the fuel wood was totally combusted. The type A combustion features (excluding LGM F3-A that shows a different pattern and is discussed in further detail in 'Fuel Wood Management: Comparing Concentrated and Scattered Charcoal Contexts' in relation to its context) were all rapidly and successively buried by aeolian sedimentation (Vannieuwenhuyse 2016) over a period of 14,000 years (from the first occupation level dated around 45 ka to the top of SU5 around 31 ka), indicating that visitation to the site was episodic and recurrent over many generations. (F4-B, F5-B, F6-B, F7-B) The type B dug combustion features, which are found in the upper Pleistocene levels (SU7) and date to around 34 ka (Table 1; Wood et al. 2016), present completely different sedimentary characteristics to the flat type A combustion features. The type B features contain many more charcoal than the type A flat hearths (Table 1; Fig. 5c, d), indicating a more incomplete combustion process. At the micromorphological scale, the ash particles are generally not observed in anatomic connection, which indicates that the combustion residues have been displaced. Geogenic sand is intimately mixed with the combustion by-products (Fig. 8a) within the structures and could be related to the covering of the fire with sediment (Fig. 4). Baking food in ground ovens is a cooking practice frequently recorded ethnographically amongst some Aboriginal groups (e.g. Gould 1968;Harney 1951) and was experienced by one of the authors (RW) during her fieldwork with Gooniyandi traditional owners. The sedimentary facies observed in the structure reflect an earth oven functioning, from the scraping preparation step to the abandonment of the structure in a disrupted state. Indeed, if the type B features were used as cooking pits, which were covered with earth as a ground oven, both the restricted exposure to oxygen and the extinction of the fire at the expected cooking time might explain why combustion of the fuel was incomplete (Antal and Grønli 2003). Such conditions would typically produce larger, more solid charcoal than the type A hearths, which were exposed to oxygen and left to burn until resulting in complete combustion of the wood. Similar dense and dug features have been interpreted as possible cooking pits in Holocene and terminal Pleistocene layers of rockshelter excavations in western and central Australia (Byrne et al. 2013;Smith et al. 1995), but here, our interpretation is supported by both micromorphological and anthracological lines of evidence.

Type C Palimpsest of Combustion Features (F1-C, F2-C)
The Holocene layers are mainly composed of a compact ash-rich deposit, which represents a palimpsest of in situ fire places and secondary contexts of ash accumulations (hearth rake-out and ash dumps) in conjunction with material evidence for diverse activities taking place in the cave (botanical remains and artefacts). Often individual features are difficult to distinguish but remnants of types A and B combustion features were both observed throughout these levels. This absence of clear layering is also observed at the microscale, with most of the particles in the Holocene deposits observed in random patterns, with both fresh and carbonised organic matter mixed with geogenic sands and vegetal organic debris, fresh or carbonised (Fig. 9a, b). These observations confirm that intense processes such as trampling, mixing and bioturbation due to the presence of humans and animals in the cave have disturbed the primary organisation of the upper levels of the deposit (Fig. 4).
Based on the microstratigraphic and anthracological analyses as well as archaeological material evidence from the site (Balme 2000;Vannieuwenhuyse 2016;Whitau et al. 2016a, b;Wood et al. 2016), the Holocene units seem to reflect a more consistent occupation at Riwi than the earlier Pleistocene layers. In other Aboriginal archaeological sites around Australia, the Mid-to Late Holocene is often described as a period of intensification, with various interpretations of changes observed in the archaeological record extrapolated from single site contexts to encompass regional and pan-continental spatio-temporal narratives (see summaries in Langley et al. 2011;Ulm 2013). In the Kimberley region, increased abundance in charcoal and lithic artefacts has been argued to signify a more intense occupation during the Holocene (e.g. Dortch and Roberts 1996;O'Connor 1995O'Connor , 1999Veitch 1996). In terms of lithic artefacts, recent reductionbased analyses (e.g. ) have demonstrated that peaks in lithic artefact discard directly correlate to distinct technological innovations, in line with observations of other assemblages in northern Australia (e.g. Clarkson 2008). Table 3 and Fig. 10 present the anthracological results from seven combustion features (F1-C, F2-C, F3-A, F4-B, F5-B, F7-B and F9-A) alongside the stratigraphic contexts SU7, SU8, SU9 and SU10 (reproduced from Whitau et al. 2016a) in relation to vegetation type, where taxa have been grouped into five categories: bloodwood/ eucalypt savanna, non-eucalypt savanna, dry rainforest, riparian and indeterminable (Fig. 11). Bloodwood/eucalypt savanna is comprised of all the Myrtaceae except for Melaleuca sp., which is the sole charcoal type in the riparian category (e.g. Fig 11f). Non-eucalypt savanna is composed of those arid-adapted, sclerophyll taxa that colonise the valley floor (e.g. Fig. 11g), and the dry rainforest taxa, traditionally associated with monsoonal vine thicket, are those which inhabit the limestone range and outliers (Fig. 11b). The dry rainforest component is comprised of Indo-Malayan taxa, which vary considerably at the family level, in comparison with the deep-time Australian flora (as represented by the other vegetation types) that comprise low familial and high species-level diversity. The dry rainforest taxa are important economically to huntergatherer populations (Dilkes-Hall 2014; Whitau et al. 2016a).

Fuel Wood Management: Comparing Concentrated and Scattered Charcoal Contexts
In Whitau et al. (2016a), we argue that firewood was predominantly collected from the valley floor throughout the occupation sequence (Fig. 11a), where the open vegetation would have allowed for easy collection in close proximity to the cave, following the Principle of Least Effort (Shackleton and Prins 1993). Across all features, with the exception of F9-A, firewood was most commonly collected from the bloodwood/eucalypt savanna species of the valley floor (Fig. 11a, f), which was dominated by Eucalyptus species, until the time of SU7 deposition, when the vegetation shifted to Corymbia sp. dominance, coupled with an increase in shrubby, noneucalypt savanna taxa (Erythrophleum sp. and Vachellia sp.) (Fig. 11g) (Whitau et al. 2016a). From SU7 onwards, Corymbia sp. maintains its dominance in the anthracological record throughout type B and the anomalous F3-A feature until the Early Holocene. The F1-C and F2-C Holocene units reveal that Riwi's highest charcoal diversity (13 and 14 taxa respectively, see Table 1) is mostly created by larger proportions of dry rainforest taxa and might reflect a different firewood collection strategy, with more fuel collected from the dry rainforest taxa of the limestone outcrops (Fig. 11b). This higher charcoal diversity could also be the result of better preservation of dry rainforest taxa in the upper units of the sequence. The latter seems likely to have been a significant contributing factor, with the exceptional organic preservation of the Holocene units producing the lowest proportions of indeterminable charcoal (F1-C = 28.0% and F2-C = 20.2%, Table 2).
At Riwi, F9-A is the sole context where non-eucalypt savanna taxa dominate the anthracological assemblage, in this case principally represented by Erythrophleum sp. (Fig. 11g). Even the nine charcoal identified from the F10-A, F11-A, F12-A and F13-A, type A features are all Myrtaceae (Fig. 11c, f) and predominantly Eucalyptus sp., which reflects, albeit with very poor assemblage numbers, the composition of the Pleistocene units (SU8-SU10) in which these features were located. The F9-A hearth Table 3 Anthracological results expressed in percentages of number of analysed fragments from combustion features F1-C, F2-C, F3-A, F4-B, F5-B, F7-B and F9-A compared with those of scattered contexts SU7, SU8, SU9 and SU10; the latter (and F1-C and F2-C) reproduced from Whitau et al. (2016a) Type Feature Bloodwood/eucalypt savanna    is positioned at the limit between SU8 and SU7, where a shift in taxa ranks is visible (Fig. 10). SU7 shows an increase in Corymbia sp., while Eucalyptus sp. types decrease in both abundance and diversity, and Erythrophleum sp. and Vachellia sp. increase to proportions that are not replicated elsewhere amongst the scattered charcoal assemblages (Whitau et al. 2016a). The fuel wood that comprises the F9-A anthracological assemblage was collected before or during the ecological shift observed in the SU7 assemblage, and it is interesting to note that the dominance of Erythrophleum in the feature clearly echoes the taxonomic composition shift represented in SU7 (Fig. 10). F9-A is a type A hearth structure, which was not re-used and so the dominance of Erythrophleum sp. is most likely an effect of a single firing episode. It could be that Erythrophleum sp. was collected in relation to its increased availability in the surrounding landscape where non-eucalypt savanna newly dominated or that it was specifically targeted as a fuel wood, without these hypotheses needing be mutually exclusive.
The Erythrophleum genus is composed of some eight (Ross 1998) or nine (Dunlop et al. 1995) taxa, one of which, Erythrophleum chlorostachys or ironwood, is endemic to Australia, where it grows from northeastern Queensland to the Kimberley region of Western Australia. Aboriginal groups in Queensland and the Northern Territory are recorded using ironwood in a variety of ways: an infusion of the bark is used to treat stomach pains, root infusions for cuts and leaf infusions for scabies; smoke from the wood and leaves is used to relieve constipation and smoke from the bark to produce sterility in women; resin from the roots is an adhesive and gum from the bark is edible and produces a red dye; and the wood is used for carvings, spears, music and cooking sticks (Brock 1988;Dunlop et al. 1995;Woinarski et al. 2002). While there are no ethnographic accounts of ironwood use in the Kimberley region, at the very least, ironwood is one of the densest of Australia's native timbers (1200 kg/m 3 ) (Boland et al. 1984;Woinarski et al. 2002) and might provide a different type of fire from the bloodwood/eucalypt species.
The flora represented in the type B features are broadly comparable with those in SU7 in terms of taxonomic diversity, relative frequency and rank of taxa, which is contrary to the expectation that combustion events have low taxonomic diversity, since they represent the last few firing events (Chabal et al. 1999;Dotte-Sarout et al. 2015;Théry-Parisot et al. 2010). The more likely explanation for the similarity between SU7 and the type B dug features (Fig. 10) is that both were used several times and the assemblages represent multiple collecting trips (Whitau et al. 2016a). Indeed, F5-B is a series of densely packed combustion features showing concave pits, ash and abundant charcoal; therefore, it securely represents several firing episodes. Such an explanation would also make sense with the proposed identification of these features as ground or earth ovens, with features being re-used several times during an occupation period. Moreover, the voluntary cessation of the combustion process involved in earth oven cooking is an additional factor for a better representation of the original fuel wood taxonomic diversity in comparison with open-air hearth features.
Between the types B and C features sits the F3-A type A hearth. Unlike the other type A contexts, F3-A has a high diversity (12 taxa) in addition to a high proportion of indeterminable charcoal (49.2%) ( Table 3). Bloodwood savanna dominates the assemblage, with the high proportion of dry rainforest taxa comprised largely of Brucea sp.  (Figs. 10 and 11e). The sole species of the genus endemic to Australia, Brucea javanica prefers secondary forest, sandy dunes and limestone rock. Generally growing as a small shrub or tree, B. javanica produces edible roots and fruit that can aid in the treatment of dysentery and fever (Kulip and Wong 1995). Its relatively high abundance in F3-A is not readily explicable and could relate to socio-environmental changes affected by the LGM. The position of the F3-A hearth shows that type A hearth structures continued to be built after the appearance of type B hearths in the record.
While the individual features of the type C units are more difficult to distinguish, both types A and B combustion structures were produced during this time, which is characterised by an increased taxonomic diversity and representation of dry rainforest taxa (Fig. 11b) in the archaeobotanical record, including non-woody remains (Dilkes-Hall 2014; Whitau et al. 2016a).
All anthracologically analysed contexts from Riwi cave show that firewood was typically collected from the valley floor and, with the exception of F9-A, bloodwood/ eucalypt savanna taxa were the favoured fuels, supporting arguments presented by Whitau et al. (2016a). The shift in taxon dominance from eucalypt to bloodwood observed in the scattered contexts, which is likely associated to environmental changes at around 34 ka (Whitau et al. 2016a), is reflected in the composition of all the features analysed in this paper.

Methodological and Archaeological Implications
The study presented here clearly supports important taphonomic studies (e.g. Dussol et al. 2017;Théry-Parisot et al. 2010) in showing that the factors affecting charcoal preservation are complex. Table 4 summarises the anthracological and micromorphological data alongside a summary of the discussion presented above. Fires which were lit directly on the ground surface, like the type A hearth structures, are far less likely to produce quantifiable charcoal within the structure itself, as the direct exposure to oxygen will often combust the fuel entirely. Increase in charcoal abundance in itself should not then necessarily be correlated with an intensification of occupation (Ward et al. 2016), as the prevalent type of combustion structure, its formation and length of use as well as its degree of preservation must also be considered. Similarly, changes in deposition dynamics (e.g. deficit of natural sedimentation) can lead to the creation of palimpsest type deposits in archaeological sequences (Bailey and Galanidou 2009;Mallol and Mentzer 2015;Vannieuwenhuyse 2016). At Riwi, the changes observed between combustion features, the increased abundance of charcoal dense features within F1-C and F2-C and the high trampling of surface deposits all point towards a change in site use during the deposition of the Holocene units, where occupation of the cave was intensified in terms of the number of site visits, coupled with a reduction of time between visits and lesser accumulation of natural inputs as pointed out in several Australian archaeological studies (cf. Ulm 2013;Ward 2004;Vannieuwenhuyse et al. 2017).
This study also signifies the importance of understanding site formation processes; as already stated by several Australian researchers Langley et al. 2011;Vannieuwenhuyse et al. 2017;Ward and Larcombe 2003;Ward et al. 2016). The exceptional preservation of flat type A hearths in Riwi provide an archaeological case study for how heating can impact underlying sediments in this type of fine sand substrate, in the Riwi example: several centimetres below the surface where the (1) Dark brown layer (concave shapes visible in stratigraphy) composed of geogenic sands, with a high proportion of microcharcoal and carbonised vegetal particles (Fig. 7c).
(4) Whitish layer (Fig. 7c) mostly composed of ash particles (pseudomorphs of plant calcium oxalate) and phytoliths that are often still in anatomic connection (articulated). Type B combustion features are characterised by the digging of a pit or depression where the fire was lit and frequently reused. These dug features sometimes cut through the flat combustion features immediately below, disturbing the orientation of particles ( Fig. 8b). Type B combustion features contain large fragments of charcoal embedded within a fine matrix of geogenic sands and/or random oriented ash particles (Fig. 8a).
Type C features are accumulations composed mainly of by-products of combustion (predominantly ash and charcoal), along with a mix of non-burnt vegetal parts and a minor proportion of geogenic sands (Fig. 9). The high proportion of combustion residues gives the Holocene deposits its grey colour (Figs. 2 and 3). Under the microscope, the microfacies show a very bright calcitic crystallic b-fabric as a result of the high proportion of ash particles (pseudomorphs of plant tissues) (Fig. 9b, d). These particles are found both in anatomic connection (Fig. 9c, d) and disturbed (Fig. 9a, b) which indicates some mixing in this level.

Hypothesised function
Small, open-air, single-use hearths which were lit directly onto the ground surface and used for heating, lighting, or cooking purposes during short term visits to the site. Earth or ground ovens which were constructed by digging a pit or depression into the substrate. A fire was lit within the depression and subsequently buried during a hypothesised underground cooking process. A palimpsest of numerous type A and B features coupled with an increase of combustion residues resulting from maintenance activities such as reworking (cleaning rake-out of hearths but also human and animal trampling and turbation, see Fig. 4).

Economy of wood
The charcoal within the type A combustion features is mostly completely combusted with the exception of the F3-A and F9-A assemblages. F9-A is the only anthracological sample from Riwi Cave which was not dominated by The taxonomic diversity of the type B dug features is similar to the associated scattered contexts and might indicate re-use of the combustion features. The dominance of bloodwood savanna taxa indicates that firewood was collected predominantly from The savanna taxa of the valley floor continue to be well represented alongside a higher proportion of dry rainforest taxa, which indicates a potentially more extensive use of the surrounding landscape and its various ecological niches. Frequent occupation of longer duration during the Holocene sequence producing a palimpsest of archaeological deposition, which forms this type C secondary context. fire was lit. This depth of effect has several implications in how to sample such features and how to interpret the archaeological material (botanical remains, bones, lithic artefacts, ochre pigments, but also sediment samples for luminescence dating) from the affected underlying substrate, as already pointed out by microstratigraphic experimental studies by Aldeias et al. (2016) and Mallol et al. (2013b).

Conclusions
Starting 45 ka years ago and for over 14,000 years, Pleistocene occupation at Riwi was intermittent and potentially episodic. The cave would be visited for short periods: a fire lit directly on the ground (type A), fuelled by wood collected from the eucalypt savanna of the valley floor, and abandoned, with these hearth structures most likely not re-used. Occupation was intermittent, with enough time between site visits to allow for these flat, open-air, single use hearths to be covered by natural, rapid aeolian deposition. From around 34 ka to the onset of the LGM, a new combustion feature (type B) is observed, which potentially reflects a different type of occupation at the site. Instead of lighting the fire directly on the ground, the fire was lit in a dug pit, covered with earth during combustion and extinguished prior to the completion of the combustion process; in a pattern similar to a ground oven. The anthracological composition of these type B structures is most similar to the scattered context of SU7, demonstrating the collection of fuel wood from the bloodwood-dominated savanna newly established on the valley floor. The taxonomic diversity of the type B dug features might illustrate that these ground ovens were re-used multiple times and potentially signify more frequent and/or longer occupation at the site during this time.
Type A and B combustion features continued to be produced during the LGM and through to the Holocene, when the site seems to have been visited more frequently. Change in deposition patterns and higher frequency of occupation produced a palimpsest of archaeological deposition, which forms the type C secondary contexts. This period also records an increase in use of vegetation resources located in the dry rainforest areas. The savanna taxa of the valley floor continue to be well represented, indicating a potentially more extensive use of the surrounding landscape in its various ecological niches. This aligns with an intensification of occupation, where intensification refers to an increase in the number of site visits and the duration of these visits.
Combined anthracological and micromorphological analyses of combustion features at Riwi have allowed us to propose a typology of features, to define the chronology of their appearance in the record and to document changes in site occupation patterns and landscape use over time. Our results show that interpretations of anthracological spectra should be adapted to the type of combustion structure recovered; the relationship between charcoal preservation and context is far too complex to warrant the direct association of charcoal abundance with intensification of site use and/or population increase.
Future experimental studies which explore both hearth building processes and fuel wood selection strategies using traditional Aboriginal methods will enable a deeper understanding of how fire was manipulated in the past and strengthen the functional interpretation of the different types of combustion features identified in our study. Precise and multiproxy studies allow a few tangible glimpses into the lives of a site's past inhabitants, building a home 'where the hearth is'.