The Cai Beo site (N 20°43′8″, E 107°3′2″), nearly 4 m above current sea level, is located on Cat Ba Island, the largest island in Ha Long Bay in the Hai Phong Municipality (Figure 1). Cai Beo was reported first by French archaeologist Madeleine Colani who then conducted a small-scale test excavation in 1938 (Colani, 1938). Later, in 1972, 1973, 1981, 1986, and 2006, the site was excavated five times by Vietnamese archaeologists, opening a total excavated area of 449 m² (Nguyen, 2005; Nguyen, 2009). The known distribution area of Cai Beo is around 18,000 square meters.

Several hundreds of kg of large fish bones and marine shells were reported from the excavations at Cai Beo, indicating a sea-oriented economy (Nguyen, 2019). Additionally, rich terrestrial animal remains, such as masked civet, brown bear, deer, elk, wild boar, elephant and monkey, were reported from the site (Li, 2019), indicating an ancient forest environment at that time.

The deposit of Cai Beo contains three cultural layers, named as Pre-Halongian, Proto-Halongian, and Ha Long Culture during the 1973 excavation. However, since 1981 excavation, it has been generally suggested that the deposit of the site involves a local “Cai Beo phase” (subdivided into Cai Beo phases 1 and 2) and the later “Ha Long phase” (Figure 2). The lower two cultural layers of Cai Beo, phases 1 and 2, span about 7,000 through 5,000 years BP. This estimate is based on a radiocarbon date (ZK-328-0, see below) and comparisons with similar findings at another five Cai Beo cultural sites in Ha Long Bay (Nguyen, 2005; Nguyen, 2009).

The lowest layer (Cai Beo phase 1, see Figure 2: Cultural Layer I) represents the earliest occupation of the site. This layer is 2.4–3.2 m deep, consisting of sand deposits mixed with round gravels, debris, marine fish bones, oysters, and other seashells. Large numbers of stone tools were discovered, including many grinding tools (Figure 3A, all of which exhibited visible usage wear. Most stone tools are flaked chopping tools produced by the direct strike method. Additional items were disc-shaped tools, pointed tools, ¼ round pebble tools, stone hammers, stone anvils, and others. The so-called “polished stone tools” unearthed from this layer had shown sharpening only on the blade part. The discovered potsherds are small in number and size, characterized as thick and coarse. The middle cultural layer (see Figure 2: Cultural Layer II/Cai Beo Phase 2) is 1.2–2.4 m deep, and polished stone axes appeared in this layer.

The radiocarbon date (ZK-328-0) of 6,893–6,391 cal. years BP (5,810 ± 115 uncal. years BP, half-life: 5,730 years) (or 5,645 ± 115 uncal. years BP, half-life: 5,570 years) was obtained from fish bones (Laboratory of the Institute of Archaeology, 1977) excavated from the Cai Beo cultural layer. Other time indicators include the characteristics of human burial practice, craniometry, pottery and lithic remains (Nguyen, 2009). For instance, similar Haolizhuo (oyster picks) lithic tools were found at Cai Beo cultural layers widely distributed along the coastal southern China sites in Fujian, Guangdong, and Guangxi from 7,000 to 5,000 years BP (Hung, 2019). A flexed burial unearthed from the Cai Beo cultural phase has been identified as the Australo-Papuan affinity by Nguyen Lan Cuong (Nguyen, 2009). Both the flexed burial and the Australo-Papuan affinity are the key cultural indicators of hunter-gatherers in pre-Neolithic Southern China and Southeast Asia (Higham, 2013; Hung et al., 2017; Matsumura et al., 2019).

During the 1981 excavation, the lowest cultural layer was noticed in a position beneath and therefore pre-dating a marine transgression layer (about 20 centimeter thickness, a sterile gravel layer) that appeared in the depth of 2.4 meters under the square surface. Therefore, the earliest occupation at the site must have pre-dated this marine transgression (Nishimura, 2006) that had occurred at 6000 to 5500 years ago in the area of Ha Long Bay (Nguyen and Tran, 2009). This dating constraint further supports the proposed age of the Cai Beo cultural phase 1 as preceding the range of 6000–5500 cal. years BP.

The upper layer of the Cai Beo site represents the Ha Long cultural phase (Figure 2: Cultural Layer III), generally considered as starting around 4,500 years BP (Nishimura, 2006; Nguyen, 2009; Peng, 2018). At this site in particular, the associated layer is 0.2–1.2 m deep. New forms of wholly polished stone tools included well polished shouldered axes and quadrangular adzes. Other stone objects were chisels, hammers, sandstones with the so-called “Ha Long mark”, and stone rings. Compared with the findings from the preceding Cai Beo cultural layers, the pottery in the Ha Long cultural layer appeared in large quantities and in more varied forms and decorations, including the use of a wheel-thrown technique in some cases.

More than 38 sites around Ha Long Bay have been grouped into the Ha Long Culture in terms of the artifact association and time period (Nguyen, 2019). Among those Ha Long cultural sites, radiocarbon dates show an age post-dating the older “Cai Beo” context. The available C14 dates from several Ha Long cultural sites concentrated from ca. 4,500 through 3,500 cal. years BP (Chen, 2007; Nguyen, 2009; Nguyen and Tran, 2009; Nguyen and Clarkson, 2013; McColl et al., 2018; Peng, 2018). For instance, one of the C14 dates from Ba Vung site is 4,727–4,517 cal. years BP (4,100 ± 40 uncal. years BP, charcoal sample) (Chen, 2007), and another Ha Long cultural site of Hon Hai Co Tien is dated to 4,381–3,926 cal. years BP (3,755 ± 60 uncal. years BP, human bone sample) (McColl et al., 2018). At the Cai Beo site, the pottery and other artifacts from the Ha Long layer generally accord with the expectations of the early Ha Long Culture association.

During this project, we had submitted 12 samples from Cai Beo, including one human skull, one deer horn, one deer tooth, five mammal bones, and four marine fish bones, to Beta Analytic for AMS dating. Unfortunately, a reliable collagen fraction could not be isolated and purified from any of these samples. As a result, the bone material cannot be dated. However, the basic chronology and two major cultural phases of the site can be reconstructed through multiple lines of evidence as has been outlined here.

In total, 22 stone tools from Cai Beo were examined in this study for micro-plant remains. The samples include grinding stones, mullers/pounders, pitted stones, stone tools with the so-called “Ha-Long Mark”, scrapers, and pointed tools (Figure 3; Table 1). In addition, four control samples were collected from the non-use surface of stone tools and the storage in Hai Phong and Hanoi to detect modern contamination. All of the stone tools analyzed in this study had been excavated by Kim Dung Nguyen and colleagues from 1972 to 1986 from the Cai Beo site, and they were stored at the Hai Phong Museum and Institute of Archaeology in Hanoi.

The sediments and dust on each tool’s surface were first rinsed with distilled water and then cleaned in an ultrasonic bath with distilled water for 5 minutes to recover the residues. Next, the ultrasound mixtures were transferred to test tubes, which were processed to recover the micro plant remains, including starches and phytoliths, in the Department of Archaeology and Natural History laboratory, The Australian National University.

The extraction process of starch and phytolith followed the process of previous studies (Lu et al., 2008; Yang et al., 2013; Pearsall, 2016; Deng et al., 2017). First, the starch samples were isolated with LST (lithium heteropolytungstate) heavy liquid (density 1.9), mounted on a slide in a solution of 10% glycerine and 90% distilled water, then sealed with nail polish. For the preparation of the phytolith samples, the residues were rinsed three times with distilled water, and then the samples were separated with heavy liquid (density 2.35). After the above steps, the samples were rinsed twice with distilled water and then one more time with 30% ethyl alcohol. Finally, the phytolith samples were mounted on the slide with Canada Balsam. The slides of starch and phytolith were observed, measured, and counted under the optical microscope (Machine model: Olympus BX-51), respectively.

The identification of ancient starches is based on the modern reference collections from Vietnam by the authors of this study, the database (http://cmsgd.igsnrr.ac.cn/) built by the Institute of Geographic Sciences and Natural Resources Research (IGSNRR) (Yang et al., 2018), and other related published studies from tropical and subtropical areas of Asia and the Pacific (Fullagar et al., 2006; Lentfer, 2009). Furthermore, the identification of phytolith types was in accordance with the International Code for Phytolith Nomenclature and published documents (Wang and Lu, 1993; Katharina et al., 2019).

In total, 2,364 starch grains were recovered from the 22 stone tools (Table 1). Only very few starches were found in the control samples of non-use contexts, indicating that the starches recovered from the use surfaces of stone tools were indeed related with the ancient use. One hundred seventy-six starch grains could not be categorized due to the lack of diagnosable characteristics. In comparison, the other 2,188 starch grains can be classified into seven types and 17 subtypes based on their morphological features and biological attributes after comparison with modern references (Figures 4–6). The details of each type are described below.

Given the fact that most roots and tubers do not produce phytoliths, the phytoliths extracted from the surface of the stone tools are scarce. However, 80 phytoliths were recovered (Figure 7), the majority of which could be classified taxonomically only to the family level.

Fifteen globular echinate phytoliths are from palms (Aracaceae) (Figure 8A. Additionally, phytoliths from rice were discovered, including the scale decorated bulliform produced by rice leaf tissue, double-peaked phytolith produced by rice glumes, and parallel bilobate from stems (Figures 8B–E). The percentage of rice bulliform phytoliths with ≥9 fish-scale decorations can be used to differentiate wild rice from domesticated rice (Saxena et al., 2006; Huan et al., 2015). However, the numbers of rice bulliform phytoliths recovered in Cai Beo is small (n=9) and the decorations were obscure to observe, and therefore we could not ascertain whether they were wild or domesticated. One volcaniform phytolith was identified, originally from the leaf of banana (Musa sp.) (Figure 8F).

Edible aroids (family Araceae) comprise many underground food crops grown in tropical and sub-tropical regions. They are essential food crops in India, SEA, and the Pacific islands (Opara, 2003). At several sites in the ISEA and the Pacific, aroids often are discovered together with yams, in which Colocasia, Alocasia, and Cyrtosperma are reported (Barton and White, 1993; Barton, 2005; Fullagar et al., 2006; Barker et al., 2007; Loy et al., 2015). A study of ancient starch remains from the Niah cave in Borneo have identified grains of aroids (Alocasia spp.) from the upper layer of the Hell Trench sequence dating to less than 40,000 years BP (Barker et al., 2002; Barton, 2005).

Taro (Colocasia esculenta) is one of the oldest and most important cultigens in the Indo-Pacific region (Blench, 2012; Spriggs and Matthews, 2012). Nowadays, taro has persisted as one of the most significant staple diets and cultural elements in many parts of SEA. Still, the role of taro in Southeast Asian prehistory has been underestimated due to the lack of palaeobotanical study. Furthermore, the original domestication center of taro is still disputed (Oliveira, 2012). The greatest diversity of wild Colocasia species appears to extend from northeast India to southern China, within the Himalayan region of MSEA (Matthews, 1991). As mentioned, although some suggested independent domestication of taro occurred in Papua New Guinea during the early to middle Holocene (Matthews, 1991; Fullagar et al., 2006; Golson et al., 2017; Golson, 2020), a recent study of chloroplast DNA of taro does not support this hypothesis. Instead, it reveals that the cultivated taro was introduced into Papua New Guinea from SEA after an early or middle Holocene domestication (Ahmed et al., 2020).

Previously, a study in Guangxi of southwest China has recovered taro (Colocasia esculenta) starches from the Zengpiyan cave site (ca. 12,500 to 7,600 years BP). Starches were discovered on the excavated stone tools (Lu, 2003). Our new finding from Cai Beo confirms the early and long-term exploitation of taro in MSEA, which in accordance with archaeological, ethnobotanical, taxonomic, and genetic studies, helping to reconceptualize the management and domestication process of taro on a larger scale. A future archaeobotanical survey will concentrate on the remains found in older cave sites in Southern China and MSEA.

SEA and its adjacent areas are presumed to be one central origin and diversification center for yams (Arnau et al., 2017). In addition, several species of Dioscorea, such as D. alata, D. esculenta, D. pentaphylla, etc. commonly are consumed as regular dietary foods or as famine foods, fodder, and ethnomedicines in tropical Asia and the Pacific region (Maneenoon et al., 2008; Dutta, 2015; Andres et al., 2017). For example, an ethnobotany study in Peninsular Thailand reported that 15 species of Dioscorea were found in the living areas of the hunter-gatherer Sakai tribe at Banthad Range, in which eight species are consumed as main food sources by the Sakai people there (Maneenoon et al., 2008).

Previously, many archaeobotanical findings of yams have been reported from highland New Guinea, where modern humans possibly exploited yams around 40,000 years ago and integrated them into cultivation before 6,000 years BP (Fullagar et al., 2006; Summerhayes et al., 2010). In ISEA, related archaeobotanical evidence had been sporadic (Paz, 2001; Mijares, 2007; Oliveira, 2008). The starch granules and charred parenchyma from Dioscorea sp. discovered in the Niah cave presumably could date back to 40,000 years ago (Barton, 2005; Barker et al., 2007). Compounded with the lack of sufficient approaches of archaeobotanical works in ISEA, the roots and tubers naturally were preserved quite poorly in the humid tropical environments. Until now, with our new research, the evidence has been missing about Dioscorea consumption dating back to pre-Neolithic or even Neolithic MSEA. The study at Cai Beo has recovered a variety of yams, such as Dioscorea alata, Dioscorea esculenta, and Dioscorea bulbifera. These Dioscorea spp. plants require a set of knowledge and skill in harvesting, grinding, roasting, and processing (Maneenoon et al., 2008; Sharma and Bastakoti, 2010).

At least four genera of Fagaceae were recovered from Cai Beo, identified as Quercus sp., Cyclobalanopsis sp., Lithocarpus sp., and Castanopsis sp. As seen in other hunter-gatherer societies (Nguyen, 2008; Nguyen, 2014), the exploitation of acorns was an essential practice among the ancient hunter-gatherers in Ha Long Bay.

Several types of tree nuts, such as Juglans, Quercus, Castanopsis, Canarium, are found frequently at Hoabinhian sites (ca. 20,000–9,000 years BP) in Northern Vietnam (Nguyen, 2008; Nguyen, 2013), as well as in other parts of Mainland Southeast Asia (Higham, 2014). Pollen studies have shown that these trees, including Fagaceae plants, were prevalent during the middle Holocene in Northern Vietnam (Nguyen, 2014). The Fagaceae plant resources in Vietnam, rich with 45 species of the genus Quercus, have been identified. Among these 45 species, 38 belong to the subgenus Cyclobalanopsis, and seven species belong to subgenus Quercus (Binh et al., 2018a; Binh et al., 2018b). In addition, Castanopsis sp. and Lithocarpus sp., which belong to the evergreen forest type, can be found on rocky slopes in Cat Ba National Park on Cat Ba Island today (Thin, 1998).

The processing tasks of acorns as foods are diverse and labor intensive. Many species of acorns were used as tree crops in ancient societies across the world (Cao et al., 2007). The seeds of acorn trees could be consumed after being ground into flour and leached (Mason, 1995). Furthermore, acorn trees often were used as timber for floors, furniture, and traditional remedies (Dolai et al., 2012; Wan Omar et al., 2019).

Ancient acorns were reported from nearby coastal hunter-gatherer sites in the Pearl River Delta in Guangdong, dating to a similar age as the Cai Beo phase (Li, 2020). According to the study by Li (2020), acorns and oak-chestnuts undoubtedly were the essential starch sources for the indigenous communities in the Pearl River Delta during 6,000 through 4,500 years BP, and the use of acorns reduced after 4,500 years BP, about the time when rice farming started to appear. Similarly, our discoveries from Cai Beo suggest that acorns were part of the standard diet of ancient foragers in Ha Long Bay.

Micro-remains from Arecaceae are commonly found in ancient sites in Mainland and Island SEA (Yen, 1977; Paz, 2001; Barton, 2005; Oliveira, 2008; Castillo et al., 2020). At Cai Beo, two types of starches from palms were recovered. The Sugar Palm (Arenga pinnata) is an endemic plant to Southeast Asian countries (Haryoso et al., 2019). In Vietnam, it is grown on the highlands in the central or northern parts of Vietnam (Nguyen et al., 2014). It is a versatile plant, and almost all of its physical and production parts can be utilized (Ishak et al., 2013; Azhar et al., 2019).

The Kitul Palm (Caryota urens) naturally inhabits the understory tree stratum in tropical Asia’s moist lowlands and submontane forests (Rangabhashiyam and Selvaraju, 2015). It is a multipurpose palm. The starch extracted from its pith is known as “Kithul flour”, and it is claimed to have health benefits according to folklore and Ayurveda (Wimalasiri et al., 2016).

Moreover, the discovery of palm phytoliths in Cai Beo indicates that the plant naturally grew near the site. A previous study at Xincun (ca. 5,300 through 4,420 years BP) of the Guangdong coast, about 700 km to the north from the Ha Long Bay, demonstrated that the sago-type palms were a primary plant food before the rice in south China (Yang et al., 2013). Although the finding of starches from Arecaceae at Cai Beo is limited (N=15, 0.68% of the total), we cannot exclude the possibility that people had produced and consumed sago in Ha Long Bay.

The starches of a terrestrial fern (Angiopteris sp.) have been identified on the grinding stones excavated from Cai Beo. Edible ferns are among the most common wild food plants used by people worldwide. The stems, rhizomes, leaves, young fronds, shoots, and some whole plants of ferns can be used as food (Mannan et al., 2008). The rhizomes and stems of the Angiopteris sp. are rich in starch, and they are served as food in India and in China (Chen et al., 2010; Liu et al., 2012). In particular, in China, ethnic minorities in Yunnan, Guangxi, Guizhou, and Sichuan provinces of the southwest region consume much more fern species than the other parts of China (Liu et al., 2012).

The border region between the South-eastern Yunnan Province and Northern Vietnam is one of the areas with the richest biological diversity, including that of Angiopteris (Wang et al., 2020). During the historical period of SEA, the ferns of Angiopteris evecta (G. Forst.) Hoffm., Cibotium barometz (L.) J. Smith, Cyathea spp., and Pteridium aquilinum (L.) Kuhn were served as supplementary food sources or used for producing alcohol. These traditions have mostly been abandoned since modern times (De Winter and Amoroso, 2003). The recovered ancient starches of Angiopteris sp. from Cai Beo could reflect an early case of extracting and processing fern starch food in SEA more than 5,000 years ago.

Today, the findings of rice remain in MSEA date back to about 4,100–4,000 years BP (Weber et al., 2010; Barron et al., 2017; Nguyen, 2017; Castillo et al., 2018). The evidence for rice farming in the southern coastal areas of China dates back to about 5,000–4,500 years BP (Zhang and Hung, 2010; Yang et al., 2017; Li, 2020). A recent study confirms that the earliest rice and millet appeared together in the Pearl River Delta of Guangdong at 4,800–4,600 cal. years BP (Deng et al., 2022).

Sixteen rice phytoliths (N=16) were identified from the Ha Long cultural layer at Cai Beo. Although the discovered rice phytoliths from the site were too few to conclude whether they were wild or domesticated, importantly the rice phytoliths were restricted solely to the upper layer of this site within the Ha Long Culture association (Figure 7). These findings likely relate to the general context of agriculture that started in the northeastern coast of Vietnam around 4,500 years BP.

These findings emphasize the many new elements of cultural material remains that first appeared in the Ha Long cultural layer of this site. Indeed, further research will be needed to investigate this topic through more radiocarbon dating and a detailed chronology of the sites in this region.

The genus of Kaempferia is widespread and cultivated throughout SEA (Nopporncharoenkul and Jenjittikul, 2017). Nowadays, Rhizomes and leaves of Kaempferia galanga often are used as a flavoring in Vietnamese cuisine. However, because the discovered number of Type VI Starch in this study is small (N=8), the ancient use of these plants is unclear at this time.

Although only one starch granule and one phytolith of bananas (Musa sp.) were found in Cai Beo, this finding indicates the possibility that these coastal hunter-gatherers may have known about bananas and attempted to exploit them.