The three karstic caves from North Vietnam (Blondel, 1929, Glazek, 1965 and Khang, 1991), that have been investigated belong to a series of a half dozen tower karsts about 100 m high, that emerge from the alluvial plain over a surface area of about 10 km² (Bacon et al., 2004, Bacon et al., 2006 and Bacon et al., 2008b). The Duoi U’Oi cave and Lower and Upper Ma U’Oi caves are located in the Man Duc village (Tan Lac District, Hoa Binh Province) about 85 km southwest of Hanoi, and 25 km south-southwest of Hoà Binh city (Fig. 1). This area forms part of the northeastern extremity of the Annnamitic Mountain chain and the western border of the Red River fault zone (Deprat et al., 1963, Ky et al., 2001, Long and Du, 1981, Luong, 1978a, Luong, 1978b and Zuchiewicz et al., 2004). The landscape of Tan Lac District is characterised by a typical and spectacular morphology of karst peaks (tower karst) overhanging a mainly argillaceous alluvial plain level (Fig. 1a). The caves are located within monotonous Carboniferous and Triassic limestone formations several hundred metres thick (Deprat et al., 1963, Luong, 1978a and Luong, 1978b).

In the study sites, the bedrock consists of poorly fossiliferous dark grey micritic marine limestone dated from the Lower to Middle Triassic (Luong, 1978a and Martini et al., 1998). A dense network of caves, galleries and fissures developed in limestone resulting from faults and fractures. Around Man Duc village, there are at least a hundred caves with exposed sections, which may be of great interest for understanding these karstic networks and their sedimentary infillings. Most of the caves are partly-filled with brown to red-brown argillaceous breccias, containing numerous iron oxide pisoliths, as well as bones and teeth. About 20% of the caves in the Man Duc area have this type of sedimentary infilling (mainly dated from the Late Pleistocene to the Holocene), and more a half contain fossil remains, even if most of the bone-bearing deposits are only partly preserved as relics on the floor, walls and roof. The majority of the breccia was removed as part of historic phosphate mining (Holocene bat guano accumulation) which complicates the understanding of the sedimentary history. The description of the cave infilling and its interpretations, as well as the faunal descriptions, can be found in Bacon et al., 2004, Bacon et al., 2006 and Bacon et al., 2008b.

Our initial observation that first sparked our interest is the observation that in the largest cliffs, the caves entrances/exits seemed to be aligned on a same horizontal plane. Along the walls of the high cliffs, we observed that the aligned cave entrances are separated vertical by tens of meters without caves. Under good outcrop conditions, such alignment of cave entrances can be traced over kilometres on a same horizontal level sometimes up to 100 m over the plain. In the Man Duc area, the entrances are characterised by well-rounded conglomerates made of allochthonous pebbles (sandstones, grauwackes, quartzites) of fluvial origin. Some conglomerates associated with the cave entrance/exit can be traced on same levels, 10 to 100 m high (Fig. 2e). What is surprising is that these conglomerates occur exactly at the entrance of the cave but do not penetrate more than one or two metres into the interior of the cave network. In fact, the conglomerates are rapidly replaced by typical karstic breccias in the interior of the caves. The conglomerates display typical alluvial features such as cross beds, channels fills, ripple laminations, whereas the breccias appear as immature deposits made of badly-rounded or sub-angular clasts. It is obvious that the outcrop displays an important boundary between endokarstic (breccia) and exokarstic (conglomerate) processes. The outcrop from Upper Ma U’Oi, plastered on the walls of the cliff at the same level than that of cavern entrance/exit, represents a typical example of such mixing between rounded alluvial pebbles and angular clasts of an endokarstic origin (Fig. 2e).

The horizontal alignment of the cave entrances, closely associated with typical alluvial conglomerates passing laterally to breccias into the network, is not coincidental. The observation in the same area of the modern water table level showing similar combinations of river and cave entrances, suggests that endokarstic (cave breccias) and exokarstic (alluvial conglomerates) developed together over time. Most parts of the karstic network are connected to the river. Thus, each alluvial terrace corresponds to an old alluvial plain, which gradually lowered over the time. The entrance of the galleries was therefore automatically readjusted to the new position of the alluvial plain. In fact, each conglomerate/cave entrance corresponds to higher alluvial plain in increasing age from base to top.

Several U/Th dating samples were taken on calcitic flowstones sampled inside the breccias from the Duoi U’Oi cave as well in Lower and Upper Ma U’Oi caves (Bacon et al., 2004, Bacon et al., 2006 and Bacon et al., 2008b). Some speleothems were dated twice. The flowstones from Duoi U’Oi give an age estimate of 66 ± 3 ka (Bacon et al., 2008b). Similar calcite from Lower Ma U’Oi reveals an age of 193 ± 17 ka (Bacon et al., 2006). The laminar speleothems from Upper Ma U’Oi suggest an age of around 600/800 ka or more, but due to some argillaceous pollution, this date has never been published, even though the result is coherent with the increasing ages from base to top of the tower karst. The stair-like geometry of the terraces, associated with cave (terrace/cave combination) summarized in Fig. 7 and Fig. 8, indicates at least three successive lowering of alluvial plain during the Late Quaternary: Upper Ma U’Oi (60 m), Lower Ma U’Oi (12 m) and Duoi U’Oi (4 m). The present-day alluvial plain corresponds to the last lowering of the water-table since the infilling of the Duoi U’Oi cave. According to the dating results, the last lowering of the alluvial plain since Duoi U’Oi (66 ka for 4 m), represents about 1 m every 16,500 years. If we apply this rate of lowering of the alluvial plain to the Lower Ma U’Oi terrace situated at 12 m over the alluvial plain, we obtain an age estimate of about 198,000 years which corresponds to the effective U/Th age (193 ± 17 ka). This suggests that the dating is coherent. The U/Th estimate from the Upper Ma U’Oi cave was unsuccessful partly due to detrital thorium pollution. Based on the average speed per meter, the age could be around 990 ka years what is not far from the estimation of the U/Th dating which gives a general estimate of 600/800 ka.

When the water table dropped due to tectonic uplifts and/or eustatic variations, a new alluvial terrace formed below the older terrace level by down cutting (Fig. 7 and Fig. 8). The karstic networks as well as the entrance/exit of the caverns are progressively adjusted to the new water table level (Fig. 7 and Fig. 8). It results a stair-like morphology, consisting of superposed alluvial terraces (Fig. 7 and Fig. 8). Such adjustments have been described in the literature (Palmer, 1991, Stock et al., 2005 and Wang et al., 2007, and others). In particular, these studies describe similar features on cave development within an incising landscape. Over the time, the bedrock incision and network development result in progressively older cave levels perched higher up canyon walls (Stock et al., 2005). Wang et al. (2007) describe and analyse the fossils contents of Chinese Plio-Pleistocene to Holocene cave infillings. According to these authors, long periods of uplift and river downcutting created new caves at different elevations. At the end of each phase of cave development, the sediments were introduced into the cavities through floodplain runoff and downward movement through openings in the limestone (Wang et al., 2007). Therefore, the caves at progressively higher elevations reflect progressively older periods of cave formation and infilling, and preserve fossil assemblages of increasing age, as we have also observed in Vietnam (Fig. 7 and Fig. 8).

One of the best examples about the spectacular water circulation inside the karst during the Messinian Salinity Crisis (MSC) is given by Mocochain et al. (2006). In this case study, the Mediterranean Sea was characterised by a short-lived (5.95–5.32 Ma) sea-level fall which reached 1500 m in some areas, before returning to current levels. During this alluvial plain lowering, new networks and new cave outlets were progressively developed inside the limestone. Surprisingly, Mocochain et al. (2006) note that cave outlets and general networks were also adjusted during the increasing rise of water-table, resulting from rising sea-levels. The flooding of the caverns led to the creation of incredible vertical galleries formed by strong upward groundwater flows called “chimney-schafts”.

In fact, the mechanisms of formation of the karstic terraces in Vietnam are similar in many respects to the fluvial terraces formed during the dropping of alluvial plains. Yang et al. (2011) report the development of at least four well-formed terraces dated to 239 ka in close relation to the adjacent karst. The referred rate of uplift/incision fluctuates between 0.1 and 0.5 m per ka, which is 2 to 10 times higher than those estimated for the Vietnamese examples (Fig. 7 and Fig. 8). In comparison, the terraces described in the border of the Maoxi River (China) by Yang et al. (2011) have exactly the same rate of alluvial plain lowering as what we have proposed here: that is around 0.06 m per ka. With respect to the sedimentary cave infilling, during each alluvial plain lowering, the breccias are partly (sometimes completely) eroded. If the amount of water is sufficient, each new lowering can rework the part of the remaining breccias leading to progressive disappearance. This may explain why the upper networks in the tower karsts are so often empty of sedimentary infilling. The presence of larger amounts of acidic water close to the soil covering the karsts, means that the speleothems are often badly developed in the upper part of the limestone (De Waele et al., 2009).

One important question remains regarding the main processes for the origin of the water table dropping. The influence of tectonics and neotectonics on the morphology of the cone and peak karsts from the Ha Long Bay, has been well-documented (Fenart et al., 1999). In surrounding areas (i.e. Dien Bien Phu basin, 150 km WNW from the studied area), the effects of a recent tectonic activity has been emphasised by Zuchiewicz et al. (2004), who described perched alluvial deposits dating to the Late Pleistocene and the Holocene. The combination of the sea level oscillations with an active tectonic background is considered as the most likely combinations of factors to explain the creation of the architecture of terrace deposits (stair geometry). Moreover, the proximity of Man Duc site to the Red River delta plain (less than 30 km apart) suggests that the Pleistocene sea level oscillations may have controlled the base level, and subsequently also the karstic and alluvial dynamics. The sea level pattern proposed by Molodkos and Bolikhovskaya (2002) for the last 600 ka, in northern Eurasia, allows us to establish a good correlation between the Middle Pleistocene rise and the high stand of sea level (240/180 ka) and the development of lower Ma U’Oi terrace (dated to 193 ka).

In conclusion, the origin and preservation of the fossil-bearing breccias are controlled in over shorter time-scales by the hydrology inside the karsts (Bacon et al., 2004 and Roussé et al., 2003). However, over longer time-scales, the deposit and preservation of the cave breccias or other fillings are directly linked to the combined effects of variations in tectonic uplift and sea level oscillations (Molodkos and Bolikhovskaya, 2002, Roussé et al., 2003 and Zuchiewicz et al., 2004).

The studied limestone crops out in the form of a tower karst called Pà Hang, on the southeastern side of the P’ou Loi Mountain. It is located in the northeastern part of Laos about 260 km NNE from Vientiane, in the Huà Pan Province (Fig. 1). The P’ou Loi Mountain belongs to the septentrional Annamitic Chain oriented broadly NNW-SSE. The basement consists of Palaeozoic granite and diorite (Saurin, 1961). These rocks are covered with a widespread sedimentary formation characterised by a grey to yellow argillaceous poorly cemented arkosic sandstone attributed to the Silurian, or perhaps partly Devonian (Saurin, 1961). This well-developed arkosic sandstone is capped by a thick limestone unit (Fig. 3a), which is badly dated at present, but which has been attributed by Saurin (1961) to an interval between the Late Carboniferous (Moscovian) and the Permian. This thick limestone unit forms a wide range of discontinuous tower karsts that emerge from the present rain forest. The region is dissected by numerous vertical faults that often create lateral contacts between the thick limestone unit and the arkosic sandstone, in the form of a major unconformity (arkosic sandstones at the base of the horizontal limestone unit are folded).

The limestone unit consists of a pluri-decametric massive sparitic dark-grey carbonate without marl intercalation. It crops out in the form of large to small tower karsts on a scale of hectometres to kilometres. Most of the mountain displays widespread karstic networks (Fig. 3a), in which bone-rich breccias can be found (Fig. 3b–e, Fig. 4b). The Nam Lot cave shows an interesting sedimentary history. The cave is a large cavity of about 60 m long divided by 5 to 6 superposed galleries all connected together, and with at least half of dozen entrances. The sedimentary deposits, located in both lower and upper part of the network, present a wide range of sediments (Fig. 9) comprising true conglomerates (Fig. 4a), conglomeratic sandstones (Fig. 3e), sandstones, limestone-rich breccias (Fig. 4b), sandy and silty clays (Fig. 3f). The conglomerates consist of rounded basement pebbles and cobbles up to 20 cm (Fig. 4a). The cave breccias contain mainly limestone clasts, in size from centimetres to decametres (Fig. 4b). Conglomerates occur only on the upper part of the cave, where they are plastered on the relics of the walls, whereas the breccias, clays and sandy/silty clays are present on the lower part. Except for the conglomerates, most of the sedimentary facies are rich in vertebrate remains, in particular, teeth. The majority of the remains derive from the lower part of the cave from the breccia and from the silty/sandy clays. Their distribution appears to be random, especially in the argillaceous facies.

The cavity reveals a complicated depositional history, consisting at least of three types of deposits, roughly superimposed above each over (Fig. 9). The oldest sedimentary layers consist of conglomerates (Fig. 9b) overlying sandy/silty clays (Fig. 9a), and covered by coarse-grained sandstones (Fig. 9c). The outcrops are discontinuous in the cave and mainly plastered on the floor and walls. From base to top, the deposits lie more than 10 m thick. The conglomerate displays an obvious pebble imbrication (Fig. 4a) suggesting a water flow from outside to inside, suggesting an infilling by the river. Moreover, the pebbles are made of basement rocks, thereby excluding an intra-karstic origin. The second deposit is formed by alterning sandstone and argillaceous sandstone. A major erosional surface separates the latter from the sandy/silty clays. The third deposit, made of coarse to very coarse-grained breccias, overlies the aforementioned sandstone (Fig. 9d). Both deposits, sandstone and breccias, display cross bedding and imbricated limestone clasts in several places, suggesting paleostreams from inside to outside, and thus an obvious endokarstic origin for the sediments.

The history of the sedimentary deposits in the cave can be presented as follows: in the first stage, the cavity is filled with three well-stratified units: sandy/silty clays at the base, followed by a plurimetric conglomerate that passes vertically to a coarse-grained sandstone (Fig. 10). The presence of this argillaceous layer without coarse-grained particles and clasts, suggests a very low water circulation inside the karst before the conglomerate deposition. The deposition of such a thick unit of fine-grained argillaceous materials strongly suggests that the cave was flooded, or had at least a highly reduced water circulation, that permitted the slow decantation of such sediment. The cave was undoubtedly situated under the level of the water-table (i.e. under the alluvial plain) (Fig. 10).

The excellent imbrication of the flattened pebbles (orientated inwards the cave) in the conglomerate suggests a cave infilling directly from the adjacent alluvial plain (Fig. 10). In fact, the cave entrances and alluvial plain are at the same level during this sedimentary stage. The coarse-grained sandstone is interpreted as well. Even if the alluvial infilling in the karsts constituted by rounded pebbles has been described in many caves, the thickness of the conglomerate in this case is very rare.

The second important stage of the cave history is the reworking of more than 90% of these first terrigenous deposits consecutively to a dropping of the alluvial plain. The drop of the latter caused a change in the water flow direction, now from inside to outside, as is demonstrated by the arrangement of the pebbles (Fig. 11). The two superposed sandstone and breccia units reveal at least two successive alluvial plain dropping events, illustrated by an obvious erosion surface (Fig. 12). The present-day situation (Fig. 13) displays a new reworking of the greatest part of sedimentary infilling, which remains as a breccias relic, or a remnant, on the floor, wall and roof. The alluvial plain occurs now around 25 m below the entrance of the cavern (Fig. 13).

The history of the Nam Lot cave shows many similarities to other Vietnamese examples. For the discussion about the successive water table changes, we refer the reader back to the depositional model through the time set out in the previous section. What is most remarkable in the Nam Lot cave is the intensity of the erosion between the two main infillings. This is especially true considering the conglomerate deposit, which was characterised by a major change in the direction of water flow (from outside to inside for the conglomerate, but the reverse for all subsequent infillings) (Fig. 10, Fig. 11, Fig. 12 and Fig. 13). Moreover, each sedimentary erosional surface represents a extended period of time and it is often difficult to quantify its importance in the absence of precise dating. In fact, the alternation infilling/reworking/infilling separated by obvious hiatuses is a frequent specific sedimentary process in the caves. Such mechanisms have been described in many caves, even if it is not so significant for the Nam Lot cave (Bacon et al., 2004, Bacon et al., 2006, Latham et al., 2007, Moriarty et al., 2000 and Zeitoun et al., 2010). Such alternations are often the result of both water table and climate changes with successive warn and cold periods, which together produce changes to the water supply. In many cases, the sediment accumulations are separated by alternating flowstone that allows U/Th dating to provide a chronological framework for the depositional history (Moriarty et al., 2000). Among the numerous caves we have visited, Nam Lot is the only one displaying such a huge conglomeratic infilling directly from the alluvial plain, while showing such clear evidence of flow direction from outside to inside.

The sequence of events from living animals to fossil assemblages in the karstic network is complex, as these events are numerous and diverse in nature. Most of the time, the assemblages are composed of isolated teeth with gnawed roots, deriving from middle to large-sized mammals. Indeed, one of the major selective factors is the post-mortem action of porcupines, which recover bones of carcasses after carnivores have finished (Nowak, 1999). Humans may also have an important role in the accumulation of remains, but that remains to be demonstrated. The deposit of bone and teeth remains is formed under strong to moderate water transport conditions inside the karstic network, as discussed in the previous section regarding the geological context. These characteristics are common to the Pleistocene sites of the Indochinese and Sondaic areas (Bacon et al., 2004, Bacon et al., 2006, Bacon et al., 2008b, Bacon et al., in press, Storm and de Vos, 2006, Storm et al., 2005, Tougard, 1998 and Zeitoun et al., 2005).

The Indochinese Subregion is a natural area within the Indomalayan Region (Corbet and Hill, 1992). Today, this Subregion includes three regional divisions: the Indochinese area, southern China, and central China transitional to the northern Palaeartic Region, with a “south-north” gradient from tropical to temperate climates (Fig. 14).

It is increasingly accepted that this division gradually took place at the beginning of the Early Pleistocene (∼2.6 Ma) under the influence of climatic changes, as well as major regional tectonic and geomorphologic events. In the transitional zone of central China, some of the oldest faunas, Renzidong (Anhui province) (Fig. 15) dated to between 2.4 and 1.9 Ma (Jin et al., 2003), reveals a mixed composition with “24 typical species of the Palaearctic Region, 23 species of Oriental Region and 20 widely distributed species” (Jin et al., 2000, p. 240). In central Burma, at the western border of the Indochinese area, one of the other oldest faunas, that of Irrawady also has in part a faunal composition of clear Indian affinity (Colbert, 1943). The composition of these two mammal assemblages, Renzidong and Irrawady, suggest that the Indochinese area did not yet have its typical faunas at the onset of the Pleistocene. Indeed, both show not only a mixed composition (Indochinese/Palaearctic, Indochinese/Indian) but also the last occurrence of some Tertiary taxa (Fig. 16). In the South of the Indochinese area, Early Pleistocene faunas could also have a sondaic influence, but this period is not documented. The appearance of clearly Indochinese mammalian faunas occurs soon after in the Early Pleistocene, as shows the Liucheng cave assemblage with Gigantopithecus blacki, Pongo pygmaeus, Ursus thibetanus, Stegodon orientalis, Megatapirus augustus among the most typical species (Kahlke, 1961 and Pei, 1957). Southeast Asia was at that time dominated by tropical rain forest.

The boundary between the Indochinese and Palaearctic faunal Regions, formed by the Qinling Moutains and the Huai River, remained fluctuating during the Pleistocene, as demonstrated by Tong, 2006 and Tong, 2007. According to this author, the faunal dispersal events in China, both southwards and northwards, occurred throughout the Pleistocene, in response to climatic fluctuations. The same observation has been made on the basis of Middle Pleistocene faunal compositions, especially for some rodents, throughout the boundary between the Indochinese and Sundaic faunal subregions (Chaimanee, 1998 and Tougard, 1998). It seems that the limit, now located at the Isthmus of the Kra, was fluctuating during the Middle Pleistocene. As for the boundary between the Indian and Indochinese Regions, the faunal dispersals are not documented.

Reconstructing the history of the faunas of the zoogeographical unit as a whole, including the three subregions, poses several challenges. The first one is the scarcity of well-documented assemblages, as most of them are only known by an inventory of the taxa (Cuong, 1985, Kahlke, 1961, Wang et al., 2007 and Wu and Poirier, 1995).

Other factors are linked to the normal practices in systematics of the scientific community. The most striking is the lack of stability in the systematics, especially the creation of species and subspecies of local value. Numerous species of middle to large-sized mammals have only been recorded from one site. The review of Rhinocerotids (Antoine, 2011 and Tong, 2001) is one among significant examples, but other groups are equally concerned (suids, cervids, large carnivores). Inversely, the validity of some species considered as a taxonomic “wastebasket” is questioned, and would also require a revision of the diagnosis, as their hypodygms probably encompass at least two species, and possibly even more. These problems in systematics have an important impact on our understanding of the evolution of the faunas, as we lack a clear view of faunal changes, especially the occurrence of new species and dispersal events, from the Indochinese Region as a whole, throughout the Pleistocene.

Fig. 16 presents the succession in the time of 14 selected mammalian assemblages. This biochronological timescale is far from complete as Pleistocene mammalian fossil records are rare, and their ages often imprecise. Nevertheless, the discontinuous sequence emphasizes that the first lineages of modern species appear during the Early Pleistocene, around 2.0 Ma, as shown by the composition of Liucheng, Mohui and Longgudong faunal assemblages (Kahlke, 1961, Pei, 1957, Wang et al., 2007 and Zheng, 2004). A “second wave” of modern lineages appearances occurs later during the Middle Pleistocene, with the Yenchingkuo (Colbert and Hooijer, 1953 and Matthew and Granger, 1923), Wuyun (Wang et al., 2007), Tam Hang South (Bacon et al., 2008a and Bacon et al., in press), Thum Wiman Nakin (Tougard, 1998) and Lower Pubu faunal assemblages (Wang et al., 2007). The first appearance of the modern lineages is naturally a subject for re-evaluation depending on new Early to Middle Pleistocene data. Indeed, it is highly probable that the modern lineages appear farther in the Pleistocene, considering the marked lacuna between the fossil faunas over this long period. The chronological range (Fig. 16) highlights that, in the Indochinese Subregion and central China, until the Lang Trang fauna, the lineages of modern species evolved together with those of archaic species (Sus xiaozhu, Nestoritherium sinensis, Tapirus sinensis, Elephas namadicus, Megatapirus augustus, Stegodon orientalis, among others). The relationships between these lineages are for the majority unknown.

Recently, we proposed a refined biochronological framework, using seven faunas dating from the Middle to Late Pleistocene (Bacon et al., in press), among which the faunas recovered during our own research in karstic sites, Duoi U’Oi, Ma U’Oi and Tam Hang. On the basis of the evolutionary stages of some species (subspecies defined on dental dimensions) and the occurrence of new taxa, our Fig. 17 indicates the presence of four distinct evolutionary levels.

Levels 2 and 3 contain diversified faunas with an association of extinct species (Megatapirus augustus, Stegodon orientalis), and subspecies (Arctonyx collaris cf. rostratus, Cuon alpinus cf. antiquus, Ursus thibetanus cf. kokeni, Muntiacus muntjak ssp.1, Sus scrofa ssp.1), as those of Tam Hang South (Bacon et al., 2008a and Bacon et al., in press), Phnom Loang (Beden and Guérin, 1973), and Thum Wiman Nakin (Tougard, 1998). The latter site is older than 169 ± 11 ka (based on U-series dates) (Esposito et al., 1998 and Esposito et al., 2002), and this provides a late Middle Pleistocene age estimate for these two levels.

Level 4, although not well documented, nevertheless shows an important change in the faunal composition. The oldest faunas from this level still retain some archaic species. This has been observed for the Lang Trang fauna (100–80 ka) with Stegodon orientalis, the possible Elephas namadicus, and the probable extinct forms of Tapirus indicus, Cuon alpinus and Rhinoceros sondaicus (de Vos and Long, 1993 and Long et al., 1996) showing that some archaic taxa could have persisted during the beginning of the Late Pleistocene between around 130 ka and 70 ka. Numerous other Chinese sites also have these faunal compositions with a mixture of modern and archaic species (Chen et al., 2002, Jin et al., 2009, Liu et al., 2010, Wang et al., 2007 and Wu and Poirier, 1995), but for most of them the ages are only rough estimations. The Duoi U’Oi (Fig. 18) and Ma U’Oi assemblages in northern Vietnam (respectively 66 ± 3 ka and > 47 ± 4 ka) (Bacon et al., 2004, Bacon et al., 2006, Bacon et al., 2008b, Demeter et al., 2004 and Demeter et al., 2005) are devoid of any archaic component marking the end of the “Stegodon-Ailuropoda era” around 70 ka in the region.

A major shift in the Duoi U’Oi fauna can be observed, in particular we note the extinction of the two last archaic species Stegodon orientalis and Megatapirus augustus, which were adapted to forested environments and warm climates. This, combined with the appearance of new subspecies in the lineages of modern middle to large-sized mammals, suggest an adaptation to new environmental and climatic conditions (Bacon et al., in press) (concerning the extinction of species in Southeast Asia, a possible combination of human induced and climatic factors is also proposed by Louys et al. (2007). The U/Th dating associated to the Duoi U’Oi fauna corresponds to the onset of the Late Pleistocene cooling period MIS4 (71–59 ka) (Imbrie et al., 1984). From the same period, located in the same area (Fig. 15), the Lower Pubu fauna is also characterised by the absence of Stegodon and Megatapirus, and documents the dispersal event of Equus hemionus recovered in South China so far (Fig. 16). Wang et al. (2007, p. 376) state that: “This extant species is, at present, distributed broadly in northern China, Mongolia, and central and western Asia; fossils of this species have been recovered in late Pleistocene contexts of Gansu, Inner Mongolia, in northern China (Deng and Xue, 1999), and also at 20–40 ka in Nanshan Cave, Guangxi, southern China (Wang and Mo, 2004). The shift in the faunal composition, together with the southwards dispersion of Equus hemionus, reveals the presence of increasingly open vegetation conditions and, as hypothesised by Wang et al. (2007, p. 376), the settlement of a temporary corridor between northern and southern China, which facilitated the migration of this horse species. Since the drop of the temperatures continued during MIS3 (59–24 ka), until the drier climatic conditions of MIS2 (24–12 ka) characterised by the Last Glacial Maximum (∼18 ka), one can wonder if this open environment around 70 ka, in South China and northern Vietnam may be the first local sign of the rain forest decline. Sun et al. (2000) observed from the cores of the South China Sea, during the Last Glacial Maximum, “an alternation of two types of pollen assemblages, namely herb-dominant and montane conifers-marked”. For the moment, as no Late Pleistocene faunas are known in the Indochinese Peninsula (Malaysia, Cambodia, Thailand, Vietnam), we have no idea of the southwards extent of the dispersion of Equus hemionus on the mainland. It is not possible as well to document, on the basis of the fossil faunas from the mainland, the hypothetical savannah-corridor passing “through the center of the Sunda Shelf, extending in an arc from south-Thailand to eastern Java” (Bird et al., 2005 and Heaney, 1991, p. 57). This question, far from being unresolved, is nevertheless essential as it is associated to the dispersion of modern Homo sapiens to Australasia during the Late Pleistocene (Barker et al., 2007, Bowler et al., 2003, Mijares et al., 2010, Storm, 2001 and Storm and de Vos, 2006).