The Middle Pleistocene vertebrate fauna from Khok Sung (Nakhon Ratchasima, Thailand): biochronological and paleobiogeographical implications

Abstract The fluviatile terrace deposits of Khok Sung, Nakhon Ratchasima province, have yielded more than one thousand fossils, making this the richest Pleistocene vertebrate fauna of Thailand. The excellent preservation of the specimens allows precise characterization of the faunal composition. The mammalian fauna consists of fifteen species in thirteen genera, including a primate, a canid, a hyaenid, proboscideans, rhinoceroses, a suid, cervids, and bovids. Most species correspond to living taxa but globally (Stegodon cf. orientalis) and locally (Crocuta crocuta ultima, Rhinoceros unicornis, Sus barbatus, and Axis axis) extinct taxa were also present. The identification of Axis axis in Khok Sung, a chital currently restricted to the Indian Subcontinent, represents the first record of the species in Southeast Asia. Three reptilian taxa: Crocodylus cf. siamensis, Python sp., and Varanus sp., are also identified. Faunal correlations with other Southeast Asian sites suggest a late Middle to early Late Pleistocene age for the Khok Sung assemblage. However, the Khok Sung mammalian fauna is most similar to that of Thum Wiman Nakin, dated to older than 169 ka. The Khok Sung large mammal assemblage mostly comprises mainland Southeast Asian taxa that migrated to Java during the latest Middle Pleistocene, supporting the hypothesis that Thailand was a biogeographic pathway for the Sino-Malayan migration event from South China to Java.


Introduction
In the Pleistocene, mammalian faunas in mainland Southeast Asia as well as in South China are characterized by the occurrence of Ailuropoda (giant panda) and/or Stegodon (extinct proboscidean), also called "Ailuropoda-Stegodon faunal complex". This faunal association is a characteristic of the long period ranging from the Early to Late Pleistocene (Kahlke 1961, Rink et al. 2008. The Ailuropoda-Stegodon complex is different in composition from the Pinjor assemblage in the Indian Subcontinent (Marwick 2009) and likely originated in mainland China. In Java, mammalian faunas are characterized by the Stegodon-Homo erectus complex. This faunal association is supposed to have migrated via the Siva-Malayan route (Fig. 1A), from Indo-Pakistan to Java, during the Early Pleistocene (von Koenigswald 1935, de Vos 1995, 2007, de Vos and Long 2001, Delfino and de Vos 2010. Unfortunately, fossil records in mainland Southeast Asia do not allow the assessment of these early dispersal events because of the scarcity of Early Pleistocene sites. The only described Early Pleistocene mammal fossil assemblages are from terraces along the Irrawaddy River in Myanmar (Colbert 1943) and probably from the cave of Pha Bong in northern Thailand (Bocherens et al. in press). During the Middle Pleistocene, it has long been known that there were significant faunal exchanges that occurred between mainland Southeast Asia and Indonesian islands. Two migration routes, known as "Sino-Malayan", are hypothesized ( Fig. 1A): an insular pathway via the Philippines proposed by von Koeningswald (1938) and a continental pathway via Thailand, Myanmar, and Cambodia proposed by de Terra (1943). Recent studies on the paleogeographical affinities of Middle Pleistocene large mammals suggest that the latter route is most consistent (Tougard 2001, van den Bergh et al. 2001. The sea floor between Taiwan and the Philippine Archipelago was too deep for the emergence of a land bridge, and thus did not allow a dispersal route for large mammals during the Middle Pleistocene. This interpretation is also supported by a high number of endemic species that occur in Philippine Archipelago (Heaney 1985, Corbet andHill 1992).
Dating from the Middle to Late Pleistocene, there are numerous paleontological and archaeological sites with mammalian fossil faunas discovered in Southeast Asian mainland (Indochinese) and islands (Sundaic) (Fig. 1B). The fossiliferous localities in Southeast Asia include the Mogok caves in Myanmar (Colbert 1943), Tham Khuyen, Tham Om and Tham Hai in Vietnam (Olsen and Ciochon 1990), Phnom Loang and Boh Dambang (Beden andGuérin 1973, Demeter et al. 2013) in Cambodia, Tambun  (red), and Taiwan-Philippine Archipelago route (blue) and B the location of the Khok Sung sand pit (star) and other Middle (red circle) and Late (yellow circle) Pleistocene sites. The Sunda shelf boundaries at the sea level about 120 m lower than the present day are compiled from Voris (2000). Some Middle Pleistocene sites in South China and central Eastern China are shown in the map. Only Gua Cha (Peninsular Malaysia) is Holocene in age (Groves 1985, Bulbeck 2003 in Malaysia (Hooijer 1962, Medway 1972, and Trinil H. K. (Hauptknochenschicht) and Kedung Brubus in Java (van den Bergh et al. 2001), all regarded as being Middle Pleistocene in age. Thailand is a critical position because it is located at the intermediate zone between different mammal communitites from South China and from Java (Lekagul and McNeely 1988, Corbet and Hill 1992, Tougard 1998. Studies on Thai Middle Pleistocene faunas are therefore crucial to understand the distribution patterns of large mammals across Southeast Asia. However, the information regarding the species composition, chronology, and paleogeographical affinitites of Thai Pleistocene faunas is poorly known due to the inappropriate taxonomic identification, to the lack of radiometric dating, and to the scarcity of substantial fossil sites, compared to that of other Southeast Asian countries.
An ongoing survey of Pleistocene deposits in Thailand has led to the discovery of numerous mammalian fossils by the Thai-French paleontological team in limestone caves from the northern to southern part of the country. Several fissure-filling and cave deposits: Thum Wiman Nakin (Ginsburg et al. 1982, Chaimanee and Jaeger 1993, Chaimanee 1998, Tougard 1998), Thum Phra Khai Phet (Tougard 1998, Filoux et al. 2015, Had Pu Dai (Pramankij and Subhavan 2001), Kao Pah Nam (Pope et al. 1981), and Thum Phedan   (Fig. 1), have been dated to the Middle Pleistocene. However, only the Thum Wiman Nakin cave yielded a tooth of Homo sp. (Tougard et al. 1998). This fossil site was dated as older than 169 ka using U-series geochronology on the stalagmitic floor above the fossiliferous layers (Esposito et al. 1998(Esposito et al. , 2002. The ages of the other caves are solely established based on faunal assemblages. Unlike cave deposits, the Pleistocene terrace sequence in mainland Southeast Asia is rare. In 2005, the Khok Sung sand pit (Nakhon Ratchasima province, northeastern Thailand) was excavated (Fig. 2). This locality, an ancient fluviatile terrace, constitutes the richest Pleistocene vertebrate fauna of Thailand with thousand vertebrate remains. All vertebrate fossils were collected from layers of sand and gravels rich in organic matter, around 6-8 m below the surface ( Fig. 2C) (for the detailed sedimentology see Duangkrayom et al. (2014) and Suraprasit et al. (2015)). The Khok Sung fauna is tentatively attributed to the late Middle Pleistocene, either 188 ka or 213 ka, based on paleomagnetic data and the occurrence of the spotted hyaena Crocuta crocuta ultima (Suraprasit et al. 2015). The Khok Sung locality yielded a unique and diverse fossil assemblage of plant remains, fish, reptiles, and mammals. Plant remains (fruits, seeds, leaves, wood, tubers, ambers, and pollens) suggest the presence of tropical mixed deciduous and dry green forests (Grote 2007). However, most of these fossil plants were possibly transported by the river and might have corresponded to the surrounding upland vegetation (Suraprasit et al. 2015). Some reptilian fossils were also described including turtles: Batagur cf. trivittata, Heosemys annandalii, Heosemys cf. grandis, and Malayemys sp., soft-shelled turtles: Chitra sp. and cf. Amyda sp. (Claude et al. 2011), and a gavial, Gavialis cf. bengawanicus (Martin et al. 2012). The mammalian assemblage consists of rhinoceroses, pigs, bovids, cervids, and an extinct elephant Stegodon, whose taxonomic attribution in generic and specific levels is poorly known (Chaima- Figure 2. Locality of Khok Sung: A the sand pit during the paleontological excavation B the location of vertebrate fossils C the lithostratigraphic and paleomagnetic sections (modified from Suraprasit et al. 2015). nee et al. 2005). In this paper, we provide taxonomic descriptions of vertebrate fossils from Khok Sung, focusing mainly on mammals. The fauna is compared with other contemporaneous Southeast Asian sites, and the results are used to propose a biochronological and paleobiogeographic framework for the fauna.

Fossil collecting and material
The sand pit was open for the pond construction (approximately 50 m long × 50 m wide × 8 m deep) ( Fig. 2A). Fossils were collected, while the water was pumped out of the sand pit. All fossils observed in the field are macrofossils. Numerous vertebrate The dental terminology is modified from Heintz (1970), Gentry et al. (1999) and Bärmann and Rössner (2011). In the case of measurements of stegodontid cheek teeth, the methods and parameters used for molar and ridge dimensions were given in Fig. 5. The H/W index and the laminar frequency (LF) were calculated, using the formula proposed by van den Bergh (1999: p. 29-30). The ridge formula of stegodontids follows the original notation of Osborn (1942). Halfridges, whose width and height were 25% less than the succeeding or preceding ridge, at the anterior or posterior extremities of a stegodontid molar are not counted and abbreviated as "x". The measurements of the cranial, mandibular, postcranial elements of mammals were taken, using the methods of von den Driesch (1976) (for metrical abbreviations, see Tab. 1).

Body mass estimation
The body mass of ruminants was estimated using the equations of Janis (1990) based on the M2/m2 surface area ratio. The surface area of M2/m2 used here is the best body mass predictor according to the high correlations with the body mass for bovids (r 2 > 0.93) and cervids (r 2 > 0.95) (Janis 1990: table 16.8). . Dental nomenclatures of lower cheek teeth of ruminants: A lower fourth deciduous premolar B lower fourth premolar C lower third molar. The dental terminology is compiled from Heintz (1970), Gentry et al. (1999), and Bärmann and Rössner (2011). Table 1. Abbreviations for postcranial bones from von den Driesch (1976).

Scapula HS
Height along the spine DHA Diagonal height from the most distal point of the scapula to the thoracic angle Ld Greatest dorsal length SLC Smallest length of the Collum scapulae (neck of the scapula) GLP Greatest length of the Processus articularis (glenoid process) LG Length of the glenoid cavity BG Breadth of the glenoid cavity Long bones GL Greatest length GLl Greatest length of the lateral part GLC Greatest length from the caput (head) PL Physiological length (for radius only) Ll Length of the lateral part Bp Greatest breadth of the proximal end BFp Greatest breadth of the Facies articularis proximalis (for radius only) BPC Greatest breadth across the coronoid process (=greatest breadth of the proximal articular surface) (for ulna only) SD Smallest breadth of diaphysis Dp Depth of the proximal end Bd Greatest breadth of the distal end BFd Greatest breadth of the Facies articularis distalis (for radius only) Dd Greatest Inner length of the foramen obturatum GBTc Greatest breadth across the Tubera coxarum-greatest breadth across the lateral angle GBA Greatest breadth across the acetabula GBTi Greatest breadth across the Tubera ischiadica SBI Smallest breadth across the bodies of the Ischia

Faunal similarity measures and cluster analysis
We compared differences in species compostion of Southeast Asian large mammal fauna during the Middle Pleistocene, using an analysis of the faunal similarity. According to unequal sampling conditions for our data, we applied two criteria for undertaking this analysis: localities are disqualified when they have fewer than 10 taxa identified at the species level and taxa are excluded when their appearances are still doubtful (i.e. poor taxonomic description or identification). We therefore selected Simpson's Faunal Resemblance Index (FRI) because it has the smallest influence of sample sizes and emphasizes faunal resemblances (Simpson 1943(Simpson , 1960. When fauna lists in several localities differ evidently in size, the Simpson's FRI is the most useful tool for eliminating the effect of size differences between two faunas, compared to other indices (Simpson 1960). The Simpson's FRI is also applied for analysing the faunal resemblances of vertebrate fossil records (e.g., Tsubamoto et al. 2004, Travouillon et al. 2006, Grossman et al. 2014). The formula of Simpson's FRI is expressed as FRI (%) = (Nc / N1) × 100, where Nc is the number of identified taxa shared by two faunas and N1 is the number Figure 5. Parameters and measurement methods used for the lower third molar of Stegodon. Lengths and widths of the molar ridges are abbreviated as "LR" and "WR", respectively. An illustration of two right m3 (lateral and occlusal views) of Stegodon orientalis is duplicated from the specimen IVPP V5216-15 (above) and IVPP V5216-13 (below).
of identified taxa in the smaller of the two faunas (Simpson 1960). A higher score indicates a greater similarity between the faunas. We performed a dataset, transformed into a similarity matrix, to generate the dendrogram using the "PAST"statistical software version 1.61 (Hammer et al. 2001). We selected an Unweighted Pair-Group Method with Arithmetic Mean (UPGMA) as cluster algorithms for our analysis because the dendrogram represents higher values of cophenetic correlation coefficient than the others.
Referred material. A right tibia, DMR-KS-05-04-04-1. Material description. The right tibia is complete ( Fig. 6A-D) and elongated (for measurements, see Appendix 1). On the proximal articular surface, the medial condyle is as large as the lateral one. The lateral condyle is convex anteroposteriorly (Fig.  6C). The posteromedial margin of the lateral condyle lacks a notch that indicates a single meniscus attachment. At the proximal end, the tibial tuberosity is developed. The shaft is elongated, anteriorly and laterally bowed, and not anteroposteriorly compressed (Fig. 6A, B). Distally, the trochlear surface is trapezoid in outline (Fig. 6D). The medial malleolus is well-developed and projects more anteriorly than posteriorly. The medial and lateral parts of the trochlear surface are equally separated by a weak median keel.
Taxonomic remarks and comparisons. Tibial morphology is relatively conservative within and among primates. Particularly, the morphological differences of tibiae among cercopithecoids are minimal (Turley et al. 2011). The distal part of tibiae of arboreal primates (including Hylobates and all arboreal cercopithecoids) is characterized by more rounded borders of the trochlear surface and a convex proximal border of the medial malleolus joining the trochlear surface (Tallman et al. 2013). The specimen DMR-KS-05-04-04-1 shows typical characters of the recent cercopithecoids whose tibial shaft is less mediolaterally compressed than those of great apes. However, the tibia from Khok Sung represents compatible dimensions with the tibiae of Hylobates (gibbon), Presbytis (surili), and Macaca (macaque). We suggest here to make a distinction between these genera based on the ratios of the greatest length of the tibia to the length or width of the proximal tibia (GL/Bp or GL/Dp). Based on these indices, the Khok Sung tibia falls within the range of recent Macaca (Tab. 2). According to the Table 2. Ratios of the greatest lengths of tibiae (GL) to the lengths and widths of proximal and distal tibiae (Bp, Dp, Bd, and Dd) of Khok Sung macaques compared to recent Southeast Asian primates.
well-developed. The upper margin of the olecranon is concave and possesses a slightly higher posterior part that extends laterally. The anconeal process is distinct. The medial and lateral coronoid processes diverge laterally (Fig. 6F). The trochlear notch is deep, forming nearly a semicircular surface for articulation (Fig. 6E).
The right femur preserves a complete proximal part and broken shaft (Fig. 6G, H). The greater trochanter is as high as the upper surface of the rounded femoral head. The intertrochanteric crest is straight and nearly oriented vertically (Fig. 6H). The upper border of the neck is flat. The lesser trochanter projects anteriorly and is situated at about 1.5 cm below the femoral head.
Taxonomic remarks and comparisons. The proximal ulna of canids is characterized by a bilobed and laterally compressed olecranon process, well-developed anconeal and lateral coronoid processes, and a laterally compressed shaft. The proximal crest of the olecranon is grooved anteriorly, but enlarged and rounded posteriorly (Tong et al. 2012). Pionnier-Capitan et al. (2011) suggested that in medial view the posteroproximal tuberosity of the olecranon of Canis is more proximally developed than in Cuon. The posteroproximal tuberosity of the Khok Sung ulna is as developed as that of Cuon. Furthermore, based on our comparisons with extant specimens, the Khok Sung canid ulna resembles that of Cuon alpinus because the olecranon bends more medially and the posterior border of the olecranon is straighter than those observed in Canis lupus. The Khok Sung specimen is slightly smaller than the recent Cuon alpinus (Tab. 3). However, it is much smaller than recent and fossil Canis lupus, as well as the paleosubspecies Cuon alpinus caucasicus (Tab. 3).
Living canids generally show a typical morphology of the proximal femur, characterized by their relatively vertical intertrochanteric crests, prominent lesser trochanter with the sharp crest extending downward along the shaft, moderately-sized greater trochanter, and slender shaft (France 2009, Tong et al. 2012. In Canis lupus, the lateral side of the caput femoris is obliquely prolonged towards the trochanteric fossa. The upper border of the neck is concave and shorter than those in Cuon alpinus (Ripoll et al. 2010). The femur DMR-KS-05-04-28-13 is canid-sized (Tab. 3) and is comparable in morphology to Cuon alpinus. For instance, the intertrochanteric crest is more oblique and straighter (nearly vertical and curved in Canis lupus), the caput femoris is round, and the upper border of the neck is long and flat (Ripoll et al. 2010). Table 3. Measurements (in millimetres) of ulnae and femurs of Khok Sung and other extant and fossil canids. * indicates a subadult individual. Metrical data of fossil canids are from Baryshnikov (2012Baryshnikov ( , 2015. Because the Khok Sung ulna and femur morphologically match better Cuon alpinus than Canis lupus, we identify these two postcranial specimens as belonging to Cuon sp. A fragmentary tusk (DMR-KS-05-03-15-2) contains dentine (outer and inner layers), cementum, and a pulp cavity (Fig. 7I-K). It is slightly curved upward and sub-rounded in cross-section for both the proximal and the distal section. A median longitudinal groove is present on the dorsal surface. The Schreger pattern commonly developed in elephantoid tusks is visible on the inner dentine layer. The maximum length of DMR-KS-05-03-15-2 is 159.2 mm and the mediolateral and dorsoventral diameters of the proximal cross-section are 73.88 and 70.56 mm, respectively. The outline of the tusk (DMR-KS-05-03-15-2) resembles S. trigonocephalus in its more medial-laterally than the dorso-ventrally compressed cross-section. The macroscopic distinctive features in cross-section are similar to S. sompoensis (van den Bergh 1999) but show the incremental lines more obviously. Table 4. Measurements (in millimeters) of cheek teeth of Khok Sung proboscideans, including a number of preserved ridges (NR), lengths (L), widths (W), heights (H), enamel thickness (ET), H/W indices (100 × H/W), and laminar frequencies (LF). The laminar frequencies are expressed as the following formula: LF=n*100/d l + n*100/d b / 2, where "d l " and "d b " are referred to distances at the lingual and buccal side of the tooth, respectively, and "n" is equivalent to the number of ridges between two measuring points (van den Bergh 1999). * indicates measurements of the maximum length of the preservation according to incomplete specimens. The H/W index is calculated for each ridge. The laminar frequency is measured based on the maximum number of preserved ridges. Lower dentition: DMR-KS-05-04-01-8 (dp3) is heavily worn and comprises three preserved ridges and an anterior cingulum ( Fig. 7D and Tab. 4). The buccal part of the third ridge is broken but it is presumably wider than the second ridge. The dp3 is subrectangular in outline or tapers towards the anterior part. The lateral sides between the first and second ridges are distinctly constricted.
Two hemi-mandibles of the same individual (DMR-KS-05-03-08-1 and DMR-KS-05-03-08-2) are moderately well-preserved (Tab. 4). The completely erupted m3 has eight ridges with small posterior cingulids (Fig. 7L, M). The symphysis and most of the ramus are broken away. The mandibular corpus is robust. We estimate the total number of ridges to be eleven based on the position on the corpus of the anterior root that supports two first lophs in Stegodon (Saegusa et al. 2005). The anteriormost preserved ridge is thus the third ridge, broken at its anterior and lateral parts in both specimens. The third to sixth ridges are strongly worn, whereas more posterior ridges are successively less damaged by abrasion. Valleys between the ridges are moderately filled with abundant cement. There is no median cleft. The m3 is much more elongated and contains five mammillae on the posteriormost ridge. The mammillae increase in size successively from the anterior to posterior ridge.
Taxonomic remarks and comparisons. The proboscidean cheek teeth from Khok Sung are assigned to Stegodon because there are more than five ridges or loph(id)s on molars, V-shaped valleys between ridges on molars , and step-like worn surface reliefs on the enamel layer (Saegusa 1996, Saegusa et al. 2005. The Khok Sung material shows well-developed cheek tooth features of derived Stegodon (e.g., a greater number of ridges and mammillae, high filled cements between the ridges, and a high angled cliff on the enamel surfaces (step-like structure "type 3", in Saegusa (1996)).
The morphologies and ridge sizes of upper molars from Khok Sung are congruent with Chinese S. orientalis (Tabs 5-7). However, we suggest that some comparative upper    third molars of S. orientalis (e.g., IVPP V5216-9) represent a total ridge number of ten (excluding anterior and posterior halfridges), different from the ridge formula (×11× for this species) given by van den Bergh et al. (2008: table. 3). The ridge formula of the M3 of S. orientalis therefore ranges from ten to eleven. The m3 of S. orientalis commonly has a total number of twelve ridges (excluding anterior and posterior halfridges). According to the fact that only a few comparative specimens of the m3 of S. orientalis are complete with the total ridge number of twelve, some of them (e.g., IVPP V1777 and IVPP V5216-16, based on our observations) display a total of 11 ridges (excluding anterior and posterior halfridges). In S. orientalis, the number of ridges on the m3 thus ranges from eleven to twelve. S. insignis has a total number of ridges ranging from eleven to thirteen (van den Bergh et al. 2008 (Fig. 9A, B). This tusk curves slightly upward and is dorsoventrally compressed and probably obovoid or oval in cross-section ( Fig. 9B, C). The Schreger pattern in the dentine is poorly developed or absent. The fractures of the cross-section are developed, perpendicular to the outer surface ("radiate cracking or fracture pattern") (van den Bergh 1999) (Fig. 9C). The maximum length of the preserved tusk is 196.1 mm and the mediolateral and dorsoventral diameters measured on the proximal cross-section are 71.3 and 49.1 mm, respectively.
Mandibles and lower dentition: a mandible (DMR-KS-05-03-00-126) preserves both sides of cheek tooth rows (right p2-m3 and left p3-m3), but most of its symphysis and entire ramus are broken off ( Fig. 10E-G) (for measurements, see Appendix 2). The posterior edge of the mandibular symphysis ends nearly at the middle part of p3. The ventral margin of the mandible is convex in lateral view (Fig. 10E). The mental foramen is situated below the p3. In ventral view, the small foramen is present at the central portion of the mandibular symphysis and the lingual mandibular outline is U-shaped (Fig. 10F, G). Only the basal part of a right tusk-like incisor is preserved in its socket. Another specimen DMR-KS-05-03-31-28 preserves a nearly complete mandibular symphysis and left p2 and p3 sockets (Fig. 10H, I). The left mandibular body behind the p3 is broken away. All lower cheek teeth are heavily worn and rectangular in occlusal outline ( Fig. 10F) (for measurements, see Tab. 9).
Nasal: a nasal bone (DMR-KS-05-03-00-56) is short and robust, bending downward and narrowing anteriorly towards the tip (Fig. 10J). The anterior surface is nearly straight in lateral view (Fig. 10K), whereas its ventral surface is flattened at the central suture. This nasal bone is most similar to Rhinoceros sondaicus (e.g., specimen MNHN-ZMO-1985-159), because its anterior part is pointed rather than rounded (Colbert 1942). In comparison, R. unicornis displays a convex anterior surface in lateral view and a well-developed horn protuberance of the nasal region. The maximum length and width of the nasal are 131.1 mm and 88.8 mm, respectively.
Taxonomic remarks and comparisons. Four isolated cheek teeth (P2, P3, M1, and M3) assigned to R. sondaicus are characterized by the following morphological features: a presence of the moderately developed crochet, sinuosity of the ectoloph, distinct parastyle fold, and deeper median valley compared to the posterior valley, and the absences of an antecrochet, protocone fold, and metacone bulge on M3. All of Figure 10. Cranial, mandibular, dental remains of Rhinoceros sondaicus from Khok Sung: A DMR-05-03-00-128, a left P2 in occlusal view B DMR-KS-05-03-22-17, a left P3 in occlusal view C DMR-KS-05-03-00-129, a left M1 in occlusal view D DMR-KS-05-03-00-127, a left M3 in occlusal view E-G DMR-KS-05-03-00-126, a mandible in lateral (E), occlusal (F) and ventral (G) views H-I DMR-KS-05-03-31-28, a fragmentary mandible in occlusal (H) and lateral (I) views J-K DMR-KS-05-03-00-56, a nasal in dorsal (J) and lateral (K) views. these characters coincide with the upper molars of R. sondaicus (Pocock 1945, Hooijer 1946, Zin-Maung-Maung-Thein et al. 2006, Groves and Leslie 2011. Large tusk-like incisors (i2) are notably typical of Asian rhinoceroses. The two small alveoli corresponding to the lost central incisors are autapomorphic of Rhinoceros (Groves and Leslie 2011). Our observations on the recent mandible iPHEP M05.5.001.B and MNHN-ZMO-1985-159 demonstrate that an alveolus extension of the lower incisors that reach posteriorly to the lingual side of the p2 is a characteristic of both living Javan (R. sondaicus) and Indian (R. unicornis) rhinoceroses (Tong and Guérin 2009). This feature efficiently distinguishes Rhinoceros from the Sumatran rhinoceros, Dicerorhinus sumatrensis, where the alveoli of the lower incisors do not extend as far (Tong and Guérin 2009). In the mandibles DMR-KS-05-03-00-126 and Table 9. Measurements (in millimeters) of cheek teeth of Khok Sung rhinoceroses, Rhinoceros sondaicus and Rhinoceros unicornis, compared to recent specimens (data from Guérin (1980)). "(i)" refers to an isolated tooth and "(m)" indicates a tooth attached to the mandible. DMR-KS-05-03-31-28, the lower incisor alveoli extend posteriorly into the mandibular symphysis, ventral to the lingual side of the p2 (Fig. 11H, I). The latter specimen also shares similar mandibular dimensions (Appendix 2) and morphology with the former specimen. Isolated lower molars of rhinoceroses from Khok Sung are difficult to assign to either R. unicornis or R. sondaicus due to heavy wear. In addition, there is a significant size overlap between these two species (Guérin 1980). The lengths of lower cheek teeth and molar rows provide a better distinction (little overlap in size) than those of isolated teeth. The lengths and widths of the cheek teeth on the mandible DMR-KS-05-03-00-126 fall almost within the range of R. sondaicus, with the exception of some specimens (p3, p4, and m3) that fit well with the larger-sized R. unicornis (Tab. 9). However, the lengths of the mandibular cheek tooth and molar rows of this specimen fall within the ranges of R. sondaicus (211.5-257 mm and 126.5-147 mm, respectively) and outside of the ranges for R. unicornis (Guérin 1980: table. 6). The two mandibles, DMR-KS-05-03-00-126 and DMR-KS-05-03-31-28, are thus assigned to R. sondaicus.  Mandible and lower dentition: a hemi-mandible (DMR-KS-05-03-17-13) is strongly compressed laterally and preserves a partial mandibular ramus and body with worn cheek teeth, except for the m3 which is unbroken (Fig. 12C-E) (for measurements, see Appendix 2). The lingual portion along the mandible is entirely broken. The mandibular depth below the m3 is higher than that of R. sondaicus. An isolated p2 is relatively worn and broken at its posterior part (Fig. 12B). At the lingual side of the p2, the anterior valley is slightly developed, whereas the posterior valley is prominent.
Mandible and lower dentition: DMR-KS-05-03-15-1 is incomplete, lacking the body and ascending ramus, broken posterior to the right p3 and to the left m2 (Fig.  13F, G) (for measurements, see Appendix 3). The mandible is inflated. The small mental foramen is present below the diastema between p1 and p2. Only the i3 and p1 are missing. The left p2 is not aligned along the cheek tooth row due to the deformation. The specimen DMR-KS-05-04-19-1 preserves a complete symphysis and a right body with the tooth row. The ramus is broken away (Fig. 13H, I). The mandibular body is successively inflated. The mental foramina are situated below the diastema between p1 and p2. For the specimen DMR-KS-05-04-19-1, the teeth are complete and moderately to heavily worn but the third incisors are missing. Lower incisors show a chisel-like appearance with long roots. The i2 is larger than the i1. Lower canines are slender and pointed, and curve backward. The lower canines of the mandible DMR-KS-05-03-15-1 belong to a male individual because of a more sharply triangular section (Hillson 2005) (Fig. 13F). The mandible DMR-KS-05-04-19-1 possesses a female canine characterized by more rounded cross-sections and well-developed roots (Hillson 2005) (Fig. 13H). The lower canines of the male specimen are more laterally inclined (about 30° from the cheek teeth) than those of the female individual (about 15°). The cross-section outlines of male canines (DMR-KS-05-03-15-1) are of the "verrucosic" type in which the posterior side is narrower than the labial one (Fig. 13F). All lower cheek teeth exhibit bunodont patterns with accessory tubercles, like in Sus. The lower cheek teeth increase in size from anteriorly to posteriorly (Tab. 10). Lower premolars are slightly to moderately worn. The p1 is unicuspid. Other premolars are tricuspid. All cuspids are sharp. The highest cuspid on the premolars is the metaconid. Lower molars are moderately to heavily worn and rectangular in outline ( Fig. 13F-J). The lower molars show complex occlusal patterns with well-developed main cuspids (protoconid, metaconid, hypoconid, entoconid, and pentaconid) and a bulky median column (hypopreconulid). The m2 is much larger and has a more developed posterior cingulid than the m1 (Fig. 13F, H). The m3 (DMR-KS-05-04-19-1) is elongated posteriorly (Fig. 13H). It has a well-developed talonid with bulky main and accessory cuspids (pentaconid, pentapreconulid, hexaconid, heptaconid). Another isolated posterior fragment (talonid) of the m3 (DMR-KS-05-04-19-3) is also elongated, as long as that of DMR-KS-05-04-19-1. This specimen exhibits smooth occlusal surfaces with wear and well developed main and accessory cuspids (Fig. 13J). The m3 is longer than the combination of m1 and m2 (Tab. 10).
Taxonomic remarks and comparisons. We compare our material to some Pleistocene Southeast Asian suid species, although only two distinct suid species, S. scrofa and S. barbatus, are known from many Pleistocene localities of mainland Southeast Asia. The sizes of the Khok Sung material are obviously larger than those of Pleistocene and extant Indonesian suids (S. brachygnathus, S. macrognathus, S. verrucosus, and S. celebensis) (Tab. 10). The Khok Sung suid material is comparable in size to S. scrofa and S. barbatus. The two suid mandibles from Khok Sung also show some distinctive taxonomic characters of S. scrofa and S. barbatus. For example, the mandible is not laterally enlarged or swollen and the diastema from p1 to p2 is longer than from c1 to  (Groves 1997). The lower premolar rows on the mandibles are aligned along the mandible, unlike S. verrucosus and S. celebensis in which the premolar rows diverge anteriorly (Groves 1997).
However, it is difficult to distinguish S. scrofa from S. barbatus only based on the cheek teeth because both species overlap in size (Tab. 10) and show almost similar dental patterns. The main differential characters between S. scrofa and S. barbatus are defined on the basis of the shape of lower canines in male individuals, whether the outline of the cross-section is of the "scrofic" (i.e. the posterior side is wider than the labial one (S. scrofa)) or "verrucosic" (S. barbatus) type (Badoux 1959, Hardjasasmita 1987. Similarly, this distinctive feature is demonstrated by the lower male canine index (the width of labial surface as a percentage of the width of posterior surface) (Groves 1981(Groves , 1997. The canine index ranges from 61.5 to 109.1 for recent S. scrofa and from 105.6 to 144.4 for extant S. barbatus (Groves 1981:  We also provide the canine index of the female specimen DMR-KS-05-04-19-1 in Tab. 11. A minor distinctive character between S. scrofa and S. barbatus is differences of the posterior accessory median cuspid (pentapreconulid) on the talonid. The pentapreconulid on the m3 is small or absent in S. barbatus (Badoux 1959). For other molar characters, S. barbatus shows more complex patterns with accessory tubercles and more rugose enamel than in S. scrofa (Tougard 1998, Bacon et al. 2011). However, the latter character is useless to make a distinction between both suid species according to our observations on the recent material of S. barbatus. The enamel surfaces of the molars in S. barbatus are often smooth or even sometimes smoother than in S. scrofa.
The female mandible (DMR-KS-05-04-19-1) and other isolated teeth are assigned to S. barbatus according to those described features. We also suggest that Pleistocene Sus barbatus probably shows evidence of sexual size dimorphism because the female specimen DMR-KS-05-04-19-1 is markedly smaller than the male specimen DMR-KS-05-03-15-1, as seen in the recent population. Table 11. Measurements (in millimeters) of lower canines of Khok Sung Sus barbatus. The canine index is expressed by the following formula: labial surface*100/posterior surface (Groves 1981  Material description. Crania and upper dentition: four crania are almost complete, lacking only the anterior portions (e.g., nasal, jugal, palatine, and maxilla) ( Fig.  14A-D). The specimen DMR-KS-05-04-18-50 shows nearly complete antlers, lacking only the left brow tine (Fig. 14A, B). The cranium DMR-KS-05-03-00-30 possesses a right antler portion preserving the complete brow tine but the broken main The shed antlers are characterized by three main tines, smooth surfaces, a short pedicle and brow tine, a long and slender main beam, a high angle (about 100-120°) between the main beam and the brow tine, and a well-developed burr (Fig. 14A, C, H-L). A small ornamented tine (or knob) is sometimes present along the dorsal surface of the brow tine or at the main beam-brow tine junction (Fig. 14C, J-L). The main beam is oriented upward, laterally, and posteriorly, and consists of forked tines apically. At the antlered crown, the inner tine is much shorter than the outer one ( Fig. 14A, H, I). The skull and antler exhibit a typical arrangement of recent Axis axis (e.g., the orientation of the main beam and brow tine, the bifurcation at the apical crown tine, and the shape of the basioccipital and basisphenoid) (for measurements, see Appendix 4).
P3 and P4 are similar to recent Axis, characterized by well-developed styles, medial cristae (more distinct on the P4), and posterolingual fossettes ( Fig. 15A) (for measurements, see Tab. 12). On the P4, the medial cristae join the postmetacrista and divide the fossa into two islands (Fig. 15A, C). Upper molars display distinct styles (particularly the mesostyle), entostyles, and anterior cingula (Fig. 15B, D, E). The metaconule fold is slightly developed. The M2 is slightly wider than the M3 (Tab. 12). The posterior lobe of the M3 is reduced in width (Fig. 15B).
Mandibles and lower dentition: twenty one mandibles range from fragmentary (preserving only the broken corpus) to nearly complete (lacking only the ascending ramus and coronoid process) individuals ( Fig. 15F-O) (for measurements, see Appendix 5). The mandibular symphyses are almost complete, but all incisors are missing. The protoconulid of the p2 is poorly-developed or absent ( Fig. 15F, H, J).
Lower third and fourth premolars exhibit a well developed metaconid which projects obliquely in occlusal view, posterior to the entoconid ( Fig. 15F, H, J) (for measurements, see Tab. 12). The latter conid joins the posthypocristid, forming a back valley on moderately worn teeth. The metaconid is bifurcated (two separated flanges: pre-and postmetacristids) on the p4. All lower molars are morphologically characterized by their brachyodont crowns and well-developed stylids (parastylid, metastylid, and entostylid), ectostylids (basal pillars), and anterior cingulids (also called "goat fold") ( Fig. 15F-Q). On the m3, the posterior ectostylid is absent (Fig. 15F, G, J-Q). The third lobe is ring-shaped as it is present on the recent specimens (e.g., MNHN-ZMO-1901-547, MNHN-ZMO-1988-153, ZSM-1951-70, and ZSM-1961 (Fig.  15F, P). But the third lobe is sometimes small and poorly-developed, as observed from the recent specimen ZSM-1963-27 (Fig. 15J, L, N). The back fossa is present on unworn to slightly worn teeth ( Fig. 15F, P), but absent on moderately to heavily worn ones (Fig. 15L, N). The posthypoconulidcristid is well-developed, a small crest protruding slightly more posterolingually (Fig. 15F).   Taxonomic remarks and comparisons. The antlers are useful to distinguish among the cervids, whereas the morphologies of lower cheek teeth are identical among Axis. The skulls, antlers, and teeth from Khok Sung are morphologically similar to those observed from recent A. axis. This suggests a morphological stasis in the evolution of antlers and teeth for this species.
Based on our observation on the extant comparative material of A. axis (e.g., the specimens MNHN- ZMO-1901-547, MNHN-ZMO-1988-153, ZSM-1951-70, and ZSM-1958, we thus demonstrate some dental morphological variation within species. The m3 of A. axis appears more morphologically variable than the other molars, such as the more or less developed posterior talonids and the presence/absence of back fossae. The cheek teeth of extant A. axis are relatively similar to those of A. porcinus (e.g., the specimens MNHN- ZMO-1904-60, MNHN-ZMO-1962-4188, ZSM-1968-493, and ZSM-1969. However, A. axis differs from A. porcinus in having less developed anterior cingulids on the lower molars and the presence of back fossae on the m3. Recent A. axis represents an intermediate size between A. porcinus and two cervid species (Panolia eldii and Rusa unicolor) (Tab. 14). A. axis from Khok Sung also follows the size tendency of recent populations (Figs 17 and 18).
Compared to other Pleistocene cervid species, the cheek teeth of A. axis from Khok Sung are smaller than those of A. shansius from Anhui and Yunnan (China) and of A. javanicus from Ngandong and Buitenzorg in Java and Carnul Cave in India, but are larger than those of A. lydekkeri from Trinil H. K. (Java) (Figs 17 and 18). Although, A. javanicus is closely related to or even synonymous with A. axis according to Meijaard Axis axis Bubalus arnee 3.09 5.88 3. Panolia eldii -- and , it is considered as a valid species due to studies of the geometric morphometric analysis performed on the teeth (Gruwier et al. 2015). According to the scatter diagrams of the dental sizes (Figs 17 and 18), Thum Wiman Nakin and Thum Prakai Phet fossil teeth assigned to A. porcinus (Tougard 1998, Filoux et al. 2015 are much larger than their extant populations and those from Khok Sung. Although the Pleistocene hog deer probably show clinal variation in size (Bergmann's rule) in re-  sponse to colder climates. The fossil teeth attributed to A. porcinus from Thum Wiman Nakin and Thum Prakai Phet, identified by Tougard (1998) and Filoux et al. (2015), possibly reveal a double size (or more) of the recent population. We suggest that these fossils likely belong to either other larger or new cervid species that lived during the Pleistocene across mainland Southeast Asia. We also cast doubt on the occurrence of A. porcinus in the Middle Pleistocene of Boh Dambang, Cambodia (Demeter et al. 2013). The existence of A. porcinus in Southeast Asia during the Middle Pleistocene is still doubtful.  Table 14. Body mass prediction of Khok Sung ruminants using second molar variables, compared to relative sizes of the recent population (Grzimek 1975, Lekagul and McNeely 1988, Nowak 1999. The predictive equations follow Janis (1990:  distal part), DMR-KS-05-03-00-119 (right distal part), DMR-KS-05-03-19-2 (right distal part), and DMR-KS-05-08-16-1 (left proximal part); three left metatarsi-DMR-KS-05-03-25-8, DMR-KS-05-03-28-17, and DMR-KS-05-03-15-15. Material description. Cranium and upper dentition: DMR-KS-05-04-20-4 is an incomplete cranium, lacking the whole anterior parts (nasal, jugal, palatine, and maxilla) ( Fig. 19A-C) (for measurements, see Appendix 4). This specimen is a juvenile individual according to the incompletely fused sutures. The basioccipital and basisphenoid are triangular in outline and have straight lateral edges (Fig. 19C), different from those of Axis, and as observed on the recent skull of Panolia eldii (e.g., MNHN-ZMO-1937-157, MNHN-ZMO-1944-307, MNHN-ZMO-2011-190, and NMW-2975. The foramina ovale of DMR-KS-05-04-20-4 are more circular and open more anteriorly than those of Axis. The right partial antler contains a half of the slender main beam, but lacks a brow tine entirely (Fig. 19A, B). The divergent angle between the main beam and the brow tine is of about 110°, similar to recent skulls of P. eldii (e.g., THNHM-M-125). The antler surface is smooth and the burr is poorly developed in relation to the ontogenetic stages. The preserved shed antler shows a typical character of P. eldii, whose main beams strongly project and curve laterally (Fig. 19A).

Genus
P2 exhibits a prominent medial crista which divides the fossette into two islands (Fig. 19D). The separated anterior fossette is larger than the posterior one. On the upper molars, the buccal styles, anterior cingula, and entostyles are distinct (for measurements, see Tab. 12). The entostyle is bifurcated ( Fig. 19E-H). The metaconule fold (spur) is poorly developed. The posterior lobe of the M3 is reduced in width (Fig. 19G, H). The buccal wall of the posterior lobe is oblique in occlusal view.
Mandibles and lower dentition: Two mandibles (DMR-KS-05-03-27-2: Fig.  19J, K and DMR-KS-05-04-9-5: Fig. 19L, M) are nearly complete, preserving the bodies with cheek tooth rows (for measurements, see Appendix 5). The first specimen also preserves a partial ramus and is more complete than the second one in which the mandibular body is broken.
An isolated i1 is spatulate (Fig. 19I). Lower premolars show more complex patterns compared to Axis (e.g., the bifurcation of the metaconid on the p3, the irregular shape of the posterior valley, and the presence of more developed pre-and postprotoconulidcristids) (Fig. 19K, L). Lower molars display well-developed anterior cingulids and stylids (for measurements, see Tab. 12). The m3 is characterized by the presence of a posterior ectostylid (Fig. 19K). The shape of the posterior lobe of the m3 resembles that of A. axis.
Postcranial remains: postcranial bones include a scapula (Fig. 20A, B), humeri ( They are almost complete. We identify these postcranial bones based on the correlation of size and proportion with the extant specimens of P. eldii (Tab. 13, and Appendices 1, 6-10, and 12).
Taxonomic remarks and comparisons. Several authors consider Eld's deer as belonging to either the genus Cervus (e.g., Lekagul and McNeely 1988, Tougard 2001, Gruwier et al. 2015 or Rucervus (e.g., Grubb 2005). However, Groves and Grubb (2011) suggested that placement of the Eld's deer in the genus Panolia is an acceptable alternative based on mtDNA analysis (Pitra et al. 2004).
The shed antler of the Eld's deer, Panolia eldii, is characterized by bow-or lyre-like shapes, long, noticeable, and laterally bending-main beams with a distal portion curving medially, and small ornamented branches of brow tines. The cheek teeth of P. eldii differ from those of A. axis in having a larger size, a more complex wear pattern of the mesolingual conids on the p3, more developed anterior cingulids on the lower molars, and a posterior ectostylid on the m3. The Khok Sung specimens assigned to P. eldii are similar in morphology to the extant specimens. As demonstrated by the body mass estimation (Tab. 14) and scatter diagrams (Figs 21 and 22), P. eldii from Khok Sung is also comparable in size to recent populations, to that from Thum Wiman Nakin, and to some fossil species (e.g., Cervus kendengensis from the Pleistocene of Bangle and Kali Gedeh in Java). However, we suggest that some isolated teeth of cervids from Thum Wiman Nakin (Tougard 1998) reveal an improper taxonomic identification. The P2 (TF 3371 and TF 4570), p2 (TF 3938, TF 3313, TF 3358, and TF 3983), p3 (TF 3373), and m2 (TF 4025), attributed to P. eldii, may belong to other cervids (possibly R. unicolor) due to their larger sizes. Our identification thus confirms the existence of P. eldii in Thailand during the late Middle Pleistocene. Material description. Antlers: DMR-KS-05-03-20-11 is a nearly complete antler, slightly broken at the middle part of the main beam (Fig. 23A). The fragmentary antler DMR-KS-05-03-26-2 comprises a burr, a broken brow tine, and a half of the main beam (Fig. 23B). The specimen DMR-KS-05-03-28-23 preserves the broken brow tine and main beam (Fig. 23C). The antler surface is rough. The shed antlers are morphologically characterized by three main tines, a long and slender main beam, a forked construction at the tip, and a well-developed burr ( Fig. 23A-C). On the apical bifurcation, the postero-internal tine is much shorter than the antero-external one. The main beam and brow tine are also much more robust, compared to the extant males of A. porcinus (e.g., the specimen MNHN- ZMO-1904-60 andNMW-2546). The divergent angle between the main beam and brow tine ranges from 50° to 90°. The shed antlers of Rusa unicolor are different from those of Axis axis in having slightly rougher surfaces, more divergent insertion relative to the frontal orientation, a shorter main beam, and a smaller angle between the main beam and the brow tine, and in lacking small-ornamented tines or knobs on the brow tine ( Fig. 23A-C). These characters match well the recent R. unicolor.
Upper dentition: upper molars are robust (Tab. 12) and show well-developed styles (particularly the mesostyle), anterior cingula, and entostyles (Fig. 23D, E). The entostyle is bifurcated, like in Panolia eldii, in relation to the moderately to strongly worn teeth. The fossettes are present at least in the middle stage of wear. The metaconule fold is poorly developed or sometimes absent. On the M3, the anterior lobe is wider than the posterior one (Fig. 23E). Mandibles and lower dentition: four mandibles are incomplete (for measurements, see Appendix 5). The specimens DMR-KS-05-03-13 (Fig. 23F, G) and DMR-KS-05-03-00-101 (Fig. 23H, I) preserve a partially broken mandibular body. The manidibles DMR-KS-05-03-31-2 and DMR-KS-05-03-27-4 are very fragmentary. All lower cheek teeth of R. unicolor are obviously larger than those of other Khok Sung cervids (Tab. 12). Lower molars display cervid-like patterns, such as well developed styles, anterior cingulids, and ectostylids (Fig. 23J, K). On the m3, the posterior lobe of the talonid in R. unicolor is more developed than those in Axis. Moreover, the posterior ectostylid is present (Fig. 23G  Taxonomic remarks and comparisons. According to Leslie (2011), we regard here Rusa as a separate genus within the family Cervidae. Four species are currently recognized: R. unicolor (sambar), R. marianna (Philippine deer), R. timorensis (rusa), and R. alfredi (Prince Alfred's deer).
Antlers of R. unicolor are characterized by its typical three tines and forked beams at the tip, similar in shape to those of Axis porcinus but much more robust. The sambar deer shares a similar dental morphology with the Eld's deer. But it differs from P. eldii as well as A. axis in being larger-sized and in having more developed anterior cingulids on lower molars. The sambar deer is much larger than A. axis (Figs 21 and 22). Based on the body mass estimated from the second molar sizes, Khok Sung large cervids fit well the size tendency of the modern populations of R. unicolor (Tab. 14). As demonstrated by the scatter diagrams (Figs 21 and 22), the recent sambar deer shows a wide range of size variation that sometimes overlaps with the Eld's deer. The cheek teeth of Khok Sung Rusa unicolor conform to the size variability of their recent population. They are also comparable in size and morphology to the fossil sambar deer from Thum Prakai Phet (Filoux et al. 2015), Phnom Loang (Beden and Guérin 1973), and Ma U'Oi (Bacon et al. 2004) (Figs 21 and 22). As is the case for P. eldii, some cervid specimens described from Thum Wiman Nakin are improperly identified. For instance, the P2 (TF 3371 and TF 4570) probably do not belong to R. unicolor according to their smaller sizes. The taxonomic revision of fossil cervids from Thum Wiman Nakin would lead to the recognition of either higher or lower diversity of cervids in Southeast Asia during the Middle Pleistocene. Material description. Upper dentition: DP3 (DMR-KS-05-03-29-8) is molariform and elongated, characterized by well-developed anterior and posterior cingula, buccal styles, and medial fossettes, a slightly-developed entostyle, and a reduction of the anterior lobe width and height compared to the posterior lobe (Fig. 25A). The P3 (DMR-KS-05-04-01-4) has distinct styles (particularly the metastyle), protocone, and hypocone and an irregular fossette. (Fig. 25B). Upper molars have a rectangular  nal groove runs along the lingual surface of the entostyle (Fig. 25D). The M3 is more rectangular in outline compared to other upper molars. The posterior lobe of the M3 is relatively reduced in width and the fossettes are large (Fig. 25E). Mandible and lower dentition: two mandibles, DMR-KS-05-03-9-1 (Fig. 25H, I) and DMR-KS-05-04-9-1 (Fig. 25J, K), are almost complete (for measurements, see Appendix 13). All incisors and premolars dropped out of the first specimen. The second specimen lacks all incisors and the p3. Another fragmentary mandible DMR-KS-05-04-29-1 preserves only a posterior lobe of the m3.
Postcranial remains: a humerus, DMR-KS-05-03-20-2(1), preserves the shaft and distal part ( Fig. L-N). We attribute this humerus to B. sauveli according to the proportional correlation with the extant specimens (Tab. 13 and Appendix 7). This specimen is also smaller than that of extant Bos javanicus and Bos gaurus (Appendices 1 and 7).
Taxonomic remarks and comparisons. Southeast Asian large bovids are accurately identified by differences in cranial features (especially horn cores), although they show sexual and ontogenetic variation in morphology. Lacking the cranial remains, it is difficult to make a distinction within the species of Bos. Due to the lack of cranial remains of koupreys (B. sauveli) collected from Khok Sung, we identify these fossils on the basis of dental features.
Based on our comparisons with some extant specimens (MNHN-ZMO-1940-51 and MNHN-ZMO-10801), the cheek teeth of koupreys are similar to those of other species of Bos, characterized by having hypsodont crowns, well-developed styles and stylids, a horse shoe-shaped infundibulum (anterior and posterior fossettes), and bifurcated or trifurcated entostyles depending on the wear stage. Among Southeast Asian   Beden and Guérin (1973), de Vos and Long (1993), Tougard (1998), Filoux et al. (2015, and Bacon et al. (2008bBacon et al. ( , 2011, respectively. large bovids, it differs from B. javanicus (banteng) and B. gaurus (gaur) in having a more developed postprotocristid on the p3 and p4, a metacentric chromosome-shaped molar in relation to the middle wear stage, a single large medial fossette on the upper molars, a flat lingual surface of the entostyle on the moderately to heavily worn molars. The M1 and M3 of B. sauveli are almost more square and rectangular in outline, respectively, compared to those of other Bos species. B. sauveli is usually smaller than B. gaurus and Bubalus arnee (wild water buffalo), but is often comparable in size to B. javanicus (Figs 26 and 27, and for the average of large bovid body mass, see Tab. 14).
According to the molecular phylogenetic analyses, the kouprey may have been domesticated in Cambodia (Hassanin et al. 2006) and they are probably a feral animal derived from hybridization between B. javanicus and B. taurus indicus (zebu) (Galbreath et al. 2006). However, the latter statement is not recently supported by the molecular sequences available for koupreys, bantengs, and zebus (Hassanin and Ropiquet 2007). These authors indicated that the mitochondrial sequences of Cambodian bantengs are divergent from those of Javan bantengs, but similar to those of koupreys. They also proposed that the mitochondrial genome of koupreys seems to have been transferred by natural hybridization into the ancestor of Cambodian bantengs. The taxonomic status of koupreys is currently under discussion and additional molecular analyses on Southeast Asian bantengs need to be examined in the future. However, our taxonomic identification of Khok Sung bovids suggests an existence of the Pleistocene kouprey in Thailand because of its high similarities in dental features with the type specimen MNHN-ZMO-1940-51 and the specimen MNHN-ZMO-10801. Material description. Horn core: a single horn core (DMR-KS-05-03-26-22) is small, curved upward (Fig. 28A, B) and slightly backward. The horn core base is oval in cross-section (Fig. 28A). A longitudinal ridge on the anterior surface of the horn core is present (Fig. 28B). This specimen belongs to a juvenile individual according to its very small size.
Mandibles and lower dentition: DMR-KS-05-04-3-1 is complete, posterior to the p2, with the exception of a small part of the angular region (Fig. 28K, L) (for measurements, see Appendix 13). Another mandible (DMR-KS-05-03-00-1) preserves only a portion of the ramus with the complete molar row ( Fig. 28M and Appendix 13). The isolated i1 (DMR-KS-05-03-00-27) is heavily worn, spatulate, and robust. Lower premolars have well-developed main cuspids and cristids (Fig. 28K, M). On the p2, the protocone is the highest cuspid and the posterior fossette is present. The p3 is elongated as long as the p4. The premetacristid is poorly developed. The postprotocristid on the p3 is larger than that on the p4. On the p4, the postprotocristid is narrow and anteroposteriorly constricted. The metaconid is most developed, compared to B. sauveli and B. javanicus as well as Bubalus arnee. For all lower molars, the ectostylid is slightly developed and not bifurcated (Fig. 28K, M-N) (for measurements, see Tab. 15). In lingual view, the metastylid is absent at the medium wear stage (Fig. 28K, M). In occlusal view, the entostylid is straight and short. The buccal outline of the protoconid and hypoconid is U-shaped in relation to the strong wear (Fig. 28M). The posterior talonid on the m3 is well-developed. The posthypoconulidcristid protrudes posteriorly.
Taxonomic remarks and comparisons. According to IUCN (2015), the wild forms of gaurs are considered as Bos gaurus, while their domestic forms are recognized as Bos frontalis (Gentry et al. 2004). We consider here the Pleistocene fossil gaurs as belonging to wild forms in terms of taxonomic nomenclature.
We assign the juvenile horn core (DMR-KS-05-03-26-22) to B. gaurus because the horn cores of gaurs are different from all other Bos species. They grow outward and curve upward, similar to those of Bubalus arnee, but their apical portion curves inward and slightly forward (Lekagul and McNeely 1988).
Mandibles and isolated teeth of B. gaurus are also observed. The cheek teeth of B. gaurus are distinguished from B. sauveli and B. javanicus by having two separate fossettes on the P2, more developed metaconids and more anteroposteriorly constricted postprotocristids on the p3 and p4, and more robust cheek teeth (Figs 26 and 27, and Tab. 15). The entostyles are usually bifurcated or sometimes trifurcated on the slightly to moderately worn upper molars (our observations on the comparative material of recent B. gaurus: e.g., ZSM-1972-5 andZSM-1961-313), similar to those of B. javanicus. But the entostyle is not bifurcated, when the molar is extremely worn, as seen on the specimen DMR-KS-05-03-00-20 (Fig. 28I). This character is therefore morphologically variable through wear. On the m3, the entostylid and posterior talonid in B. gaurus is almost more developed than that in B. javanicus. The angle between the posthypocristid and prehypoconulidcristid is slightly more divergent in B. sauveli than in B. gaurus. The size of Khok Sung B. gaurus falls within the range of the recent population (Figs 26 and 27, and Tab. 14). We elucidate here the co-occurrence of two Material description. Crania and upper dentition: DMR-KS-05-03-20-1 is undeformed and nearly complete (for measurements, see Appendix 14). Only the right maxilla, squamosals, and basicranium are damaged (Fig. 30A-C). The horn cores are broken at their middle portion. The cross-section of the horn core base is subtriangular and anteriorly flat (Fig. 30A). The frontals are narrow between the orbits and are flat or slightly convex at the region between horn core bases (Fig. 30A, C). The supraorbital foramina are large. The orbits face slightly forward (Fig. 30A, B), not laterally like Leptobos brevicornis and Bubalus teilhardi (Dong et al. 2014). The lateral margins of the premaxilla are concave (Fig. 30B).
DMR-KS-05-03-21-1, a juvenile cranium, is incomplete but slightly deformed. The posterior part of the skull is almost complete but the anterior part is broken (Fig.  30D, E). The cranium is likely elongated and laterally compressed (Fig. 30D). This specimen preserves two horn cores (broken at the right one) and a right tooth row with the M1, the P3 and P4 roots, and the unerupted M2 and M3 (Fig. 30E). The horn cores of DMR-KS-05-03-21-1 are slender, straight, and inclined upward and backward, and bend outward (Fig. 30D), similar to that of recent Bubalus arnee (e.g., MNHN-ZMO-1863-65). The horn cores are subtriangular in cross-section base, becoming subrounded toward the apex (Fig. 30D). The divergent angle between the horn cores is 105°. The frontals are short and narrow, forming an obtuse angle with the occipital plane. The parietals merged together. The occiput extends so far, posterior to the horn core bases. The basioccipital is laterally concave and triangular in outline (Fig. 30E).
Upper cheek teeth of Bubalus arnee are more robust, compared to those of Bos. P2 (DMR-KS-05-03-18-14: Fig. 30K) is elongated. The parastyle on the P2 is less developed than that on the P3 and P4. The molarized DP3 (DMR-KS-05-03-00-103: Fig. 30L) is characterized by a well-developed buccal styles, anterior cingulum, entostyle, and spur, and a larger posterior lobe. The P3 is subtriangular in outline and is marked by a distinct parastyle, paracone rib, and metastyle and a U-shaped fossette (Fig. 30G, I). The parastyle of the P3 often curves posteriorly. The DP4 (DMR-KS-05-04-29-8: Fig. 30M) is also molarized with the broken protocone. This specimen has well-developed buccal styles and two separate medial fossettes. The entostyle curves posteriorly in occlusal view and is positioned more lingually than the protocone and hypocone. The P4 is similar in morphology to the P3, but is more anteroposteriorly compressed. Upper molars display Bos-like patterns (e.g., the degree of the hypsodonty and selenodonty and the presence of distinct styles) but are more robust than most species of Bos (e.g., B. sauveli and B. javanicus) (Tab. 15). However, the mesostyles of upper molars of Bubalus arnee are more developed than those of Bos. The medial fossette between the anterior and posterior fossettes (infundibula) is well-developed, often separating into two or three islands with wear (Fig. 30G, I, N). The infundibula are U-shaped but sometimes become metacentric chromosome-shaped due to strong wear, like in B. sauveli (Fig. 30G, N). In occlusal view, the entostyle is long and straight or curves posteriorly, depending on the stage of wear, but is never bifurcated (Fig. 30G, I, N). The small fossette is sometimes present within the entostyle in relation to strong wear (Fig. 30N).
All lower cheek teeth are robust. All lingual stylids are distinct. The p2 has a welldeveloped postentocristid and posthypocristid (Fig. 31B, D, F, H). The metaconid is positioned more lingually than all of lingual cristids. The dp3 is elongated (Fig.  31F, H). The postprotocristid is large and the metaconid is well-developed. A small anterior fossette is present with wear. The p3 displays a well-developed preprotoconulidcristid and a posteriorly bending metaconid (Fig. 31B, D). The isolated dp4 (DMR-KS-05-03-00-4: Fig. 31M) is trilobed and elongated with a well-developed stylids (anterior and posterior ectostylid, parastylid, metastylid, and entostylid. On the dp4, the buccal outline of the protoconulid, protoconid, and hypoconid is V-shaped in occlusal view (Fig. 31F, H, M). The anterior ectostylid curves slightly posteriorly in contrast to the posterior ectostylid that bends anteriorly (Fig. 31M). A large fossette is present between the medial and posterior valley in relation to middle wear stage (Fig. 31M). On the p4, the metaconid is most lingually positioned (Fig. 31B, D). The premetacristid is more developed than the postmetacristids. The postprotocristid is very anteroposteriorly constricted. The postentocristid fuses with the posthypocristid beyond the middle stage of wear.
Taxonomic remarks and comparisons. According to IUCN (2015), the wild forms of water buffaloes are considered as Bubalus arnee, while their domestic forms are regarded as Bubalus bubalis (Gentry et al. 2004).
Although the cheek teeth of Bos and Bubalus are almost morphologically identical and often show highly variable occlusal morphologies in relation to the wear stages, they are distinguishable based on the dental morphology. Bacon et al. (2011) mentioned that Bubalus arnee is distinguished from Bos by several dental characters: more massive and voluminous cones, conids, and lingual stylids, more complex patterns of folded infundibula on the upper molars, U-shaped protoconids and hypoconids on the lower molars, and unbilobed entostyles and ectostylids. However, the latter two characters are highly variable with wear, as observed on many extant specimens of Bubalus arnee from MNHN, ZSM, and THNHM. Among the modern large bovids in Southeast Asia, some lower premolar (p3 and p4) and third molar features are more informative for the species identification than others (Thein 1974). Our comparisons suggest that the cheek teeth of Bubalus arnee differ from those of Bos in having more developed mesostyles, more complex shapes of the infundibulum at the similar stages of wear, less developed or smaller metaconids and narrower postprotocristids on the p3 and p4, a presence of the small fossette within the entostyle and an absence of the longitudinal groove on the lingual surface of the entostyle on upper molars, more distinct entostylids on the m3, and a presence of the back fossette on the m3. For the incisors, it is difficult to make a morphological distinction between Bubalus and Bos. However, we assign these isolated lower incisors to Bubalus arnee because they were found together with their molars at the same spot.
As demonstrated by the scatter diagrams (Figs 26 and 27), the cheek teeth of recent Bos and Bubalus populations are highly overlapping in size. The lower molar sizes of Bubalus arnee also overlap with some fossil species (Bubalus teilhardi and Leptobos brevicornis). However, tooth dimensions are informative to make an ongoing distinction among the Khok Sung large bovids. The largest bovid in this locality is Bubalus arnee, followed by B. gaurus and B. sauveli, respectively, similar to the size tendency of their recent population (Tab. 14). Material description. Isolated teeth are almost complete (for measurements, see Tab. 16), with the exception of the specimen DMR-KS-05-03-28-10 that preserves only a posterior lobe (Fig. 33G). Molars show typical features of Capricornis characterized by hyposodont crowns, smooth enamel, and distinct styles and stylids, and an absence of the ectostylids (Fig. 33). The parastyle, mesostyle, and metastyle on the M2 are perpendicular to the buccal wall (Fig. 33A). On the m3, the mesostylid is more developed than the other stylids and the posthypoconulidcristid protrudes posteriorly (Fig. 33C, E).

Genus
Taxonomic remarks and comparisons. We assign these isolated teeth from Khok Sung to Capricornis sumatraensis (Sumatran serow) because they are comparable in size and morphology to the extant specimens (Fig. 34). Among congeneric species, C. sumatraensis is larger than C. crispus as well as two goral species (Naemorhedus goral and Naemorhedus caudatus), but is smaller than C. milneedwardsi. In addition, it differs from C. crispus in having more developed metastylid and entostylid and a presence of   back fossettes on the slightly worn m3 and from C. milneedwardsi in having less developed metastylid and posthypoconulidcristid on the m3. Compared to other fossil records, C. sumatraensis from Khok Sung is smaller than that from the Late Pleistocene of Lang Trang in Vietnam (de Vos and Long 1993), Tam Hang South in Laos (Bacon et al. 2011), Padang Cave in Sumatra (Hooijer 1958), and Xianrendong in China Qi 1978, Chen andLi 1994) (Fig.  34) and from the late Middle Pleistocene of Guanyindong (Li and Wen 1986) in China. The Khok Sung material also matches morphologically that of the subspecies C. s. kanjereus from the Middle Pleistocene of Yenchingkuo in China (Colbert and Hooijer 1953) and from the late Middle Pleistocene of Thum Wiman Nakin in Thailand (Tougard 1998). However, C. sumatraensis from Khok Sung is larger than that from Thum Wiman Nakin and Naemorhedus from Thum Prakai Phet. It differs from C. s. qinlingensis described from the middle Early Pleistocene of Gongwangling in northern China (Hu andQi 1978, Zhu et al. 2015) in having its smaller size and less developed parastyle and metastyle on the M2. However, we do not assign the material to the subspecies level based on the few isolated teeth. Material description. Skull and dentition: DMR-KS-05-03-30-30 is a slightly deformed skull preserving a nearly complete premaxilla, maxilla, nasal, and palatine process (Fig. 35A, B), and a partial palatine at the ventral part. The minimum length of the skull is 315 mm. The external naris is wide, dorsally directed, and presumably subcircular in outline (Fig. 35A). The nasal becomes narrower at the nearly premaxillary-maxillary suture and tapers into a point at the posterior rim of the naris. The premaxilla is broken anteriorly at the hole for the reception of the first dentary alveolus. The premaxilla contains at least four teeth on each side. The second one is the largest tooth in the premaxillary rows, regularly corresponding to the position of a large alveolar hole in dorsal view. A short premaxillary process extends to the second maxillary alveolus centrally or the first interalveolus laterally in ventral view (Fig. 35B). The premaxillary-maxillary suture is characterized by distinct notches. A maxilla comprises 14 alveoli, with the largest tooth crown (44.3 mm high) positioned at the fifth dentary alveolus. The width of the skull at the fifth maxillary tooth is 171.8 mm (the maximum width of the preserved skull). The width of the skull at the diastema between the last premaxillary tooth and the first maxillary tooth (the minimum width of the preserved skull) is 98.9 mm. Many small foramina in front of the alveoli are situated on both the premaxilla and the maxilla. Along the anterior to posterior maxillary rims, the tooth row is slightly convex until ending at the eighth or ninth alveolus. Teeth are characterized by their conical forms and striated surfaces. However, they are highly variable in shape and size, in relation to the position along the tooth row. The teeth of crocodyles are either slender and pointed or short and blunt (Fig. 35C) but much more massive than those of gharials. Asymmetrical surfaces of the tooth are divided by two prominent longitudinal ridges that are positioned anteriorly and posteriorly.
Osteoderms: two nearly complete specimens (Fig. 35D-G) and one small fragment are characterized by rectangular shapes, wider than long (about 5-6 cm long and 7-8 cm width), and slightly flat to convex and irregular edges with small spiny outgrowths. A short median keel does not extend far anteriorly or posteriorly (Fig. 35D, F). The external surface has several large and rounded to elliptical pits on the dorsal part and fewer small foramina and striae with surrounding fibrous patterns on the ventral part (Fig. 35E, G). These specimens differ from Gavialis cf. bengawanicus (Martin et al. 2012) in the same locality by their more ornamented pits and more irregular surfaces on the dorsal surface. Taxonomic remarks and comparisons. The specimen DMR-KS-05-03-30-30 is a crocodilian cranium with a possible maximum length up to 50 cm. All morphological characters of the Khok Sung crocodiles are congruent with the extant fresh water crocodile, Crocodylus siamensis, as well as with its fossils recovered from the Early and Middle Pleistocene of Java (Trinil H. K., Kedung Brubus, and Kedung Lumbu) (Delfino and de Vos 2010). However, the Khok Sung cranium preserves only the anterior midway portion of the skull and does not allow some morphological access to other important parts (e.g., lacrymals, jugals, and pterygoids). We thus attribute this material to C. cf. siamensis.
Material description. Vertebrae are almost complete and represent a large-sized snake (for measurements, see Tab. 17). In anterior view, the cotyle is suboval in outline with the dorsoventral compression (Fig. 35H). The ventro-lateral margins of the cotyle are nearly straight. The neural spine is well-developed and steep. The neural canal is narrow. The dorsal margin of the zygosphene is convex. The tubercle is located at the junction between the base of the zygoshene and the top of the neural canal. In posterior view, the neural arch is high and massive. The zygantra are wide and deep. In dorsal view, the median tubercle at the base of the zygosphene is distinct and the interzygapophyseal constriction is well-developed. In ventral view, the haemal keel is high (Fig. 35I) and the subcentral groove is poorly developed.
Taxonomic remarks and comparisons. These four vertebrae are attributed to the family Boidae because of the following characters: a short, wide, and massive vertebral body (i.e., the widths of the centra are greater than the lengths, sensu Delfino et al. (2004)), a small prezygapophyseal process, paradiapophyses weakly subdivided into para-and diapophyseal surfaces, and an absence of spine-like hypapophyses on midand posterior-trunk vertebrae (replaced by haemal keels) Böhme 1996, Rage 2001). Vertebrae of pythonines are commonly identified by many distinct characters: a straight and posteromedially angled zygapophyseal bridge, a triangular-shaped neural canal, a prominent zygosphenal tuberosity, a steep anterior border of the neural spine, a posterior border of the neural spine overhanging posteriorly, an absence of the paracotylar foramina, a haemal keel of mid-and posterior-trunk vertebrae delimited laterally by subcentral grooves that reach the cotylar rim, and a haemal keel projecting below the centrum (Scanlon andMarkness 2001, Szyndlar andRage 2003). The Khok Sung snake vertebrae are identified based on overall similarities with extant taxa (from the original description by Hoffstetter (1964)): a relatively elongated centrum compared to the neural arch width and the vertebral height, a longitudinal ridge along the haemal keel, and a thick zygosphenal base. The Khok Sung specimens are comparable in size to recent (e.g., Python molurus bivittatus: the specimen NMW 17117) and fossil (e.g., Python sp.: the specimens RMNH DUB 5794, DUB 6951, and DUB 6952 recovered from Trinil H. K., Java) python vertebrae. According to the fact that the species-level distinction based on the vertebral morphology is poorly known, we therefore assign these vertebrae to Python sp.
Material description. The vertebra DMR-KS-05-03-08-36 is more complete than the specimen DMR-KS-05-03-29-36 (for measurements, see Tab. 17). The pre-and postzygapophyses are slightly broken at the second specimen. In both specimens, the neural spines are unfortunately broken away. In anterior view, the cotyle is oval in outline, dorsoventrally compressed, and ventrally oriented (Fig. 35J). The prezygapophyses lack a part of the prezygapophyseal process and are dorsally inclined about 45°. The neural canal is narrow. The neural arch lacks a part of the zygosphene. No paracotylar foramina are present. In posterior view, the condyle and the postzygapophyses show a mirrored morphology with the anterior part. No zygantrum is observed. In dorsal view, the prezygapophyseal facets are drop-shaped and project laterally. The interzygapophyseal constriction is also present. In ventral view, the synapophyses protrude laterally and the centrum is triangular in outline (Fig. 35K).
Taxonomic remarks and comparisons. We assign these two vertebrae to the the family Varanidae due to the following morphological characters: a centrum tapering posteriorly, a precondylar constriction, a ventrally facing cotyle, and a large and flared condyle (Romer 1956, Averianov andDanilov 1997). The Khok Sung vertebrae match well the genus Varanus because the condyle is much wider than the posterior end of the centrum and none of the articulatory surface is visible in ventral view. They are also similar in morphology to Varanus according to an amphicoelous centrum, Table 17. Measurements (in millimeters) of vertebrae of Python and Varanus from Khok Sung. Abbreviations: CL, centrum length (measured at the ventral midline); H, maximum height (measured from the tip of the neural spine to the ventral rim of the cotyle); WPP, width between pre-and postzygapophyseal processes; Wpre, width across zygapophyseal processes; Wpost, width across postzygapophyseal processes; Wcd, width of the condyle; Hcd, height of the condyle (measured from the dorsal to ventral rim); Wct, width of the cotyle; Hct, height of the cotyle. + refers to the measurement of two attached vertebrae and * indicates an incomplete preservation. condyles facing very dorsally (anterodorsal direction), an oval-shaped cotyle, a short neural spine, and an absence of the zygosphenes and zygantra (Lee 2005). Varanus sp. is reported from the Middle Pleistocene of Phnom Loang (Beden and Guérin 1973). Two varanid species, V. cf. komodoensis (larger) and V. salvator, are described from the Middle Pleistocene of Trinil H. K. (Hocknull et al. 2009). The Khok Sung specimens are comparable in size to the recent (e.g., Varanus salvator: NMW 39446/1) and fossil (e.g., Varanus sp.: RMNH DUB 3 and RMNH DUB 5792 recovered in Trinil H. K., Java) specimens. Identifying these vertebrae more precisely to the species-level, more detailed morphological comparisons need to be made in the future.

Faunal composition of Khok Sung vertebrate assemblage
Nine taxa: seven Testudines, an extinct gharial (Gavialis bengawanicus), and a spotted hyaena (Crocuta crocuta ultima), have been previously described from Khok Sung by Claude et al. (2011), Martin et al. (2012, and Suraprasit et al. (2015), respectively. In this paper, we studied other undescribed vertebrate fossils from Khok Sung. As a result, fourteen mammalian and three reptilian taxa are identified and added to the faunal list (Tab. 18). Overall, the Khok Sung fauna consists of at least fifteen mammalian (thirteen genera) and ten reptilian (nine genera) species. The mammalian assemblage comprises megaherbivores (> 1000 kg) of approximately 19% of the species (including proboscideans, rhinoceroses, water buffaloes) and other large species of about 37% (including artiodactyls, primates, and carnivores) of the vertebrate fauna (Fig. 36). The most abundant mammal group of the locality is represented by the artiodactyls (9 species). The non-mammalian species consists of about 44% of the total vertebrate fauna. The order Testudines is the most diverse group of non-mammalian taxa in the locality (22% of the fauna). In addition, other vertebrates such as birds and fish are tentatively observed. A single fragmentary cervical vertebra of the bird order Galliformes is also present (Fig. 35L, M). Numerous fish remains including vertebrae (e.g., the specimen DMR-KS-05-03-22-76: Fig. 35N) and pectoral spines (e.g., the specimen DMR-KS-05-04-11-20: Fig. 35P and DMR-KS-05-04-05-25: Fig. 35Q) are assigned to large silurids. Regarding our observations on the Khok Sung vertebrate collection, there are some complete reptile (e.g., carapaces of tortoises and soft-shelled turtles) and fish remains that have not been identified yet. The reptile and fish assemblages would probably indicate a higher diversity than those described from this study, if these undescribed specimens are taxonomically studied in the future. However, it is assumed that the identified mammal remains represent herein the whole mammalian fauna because we have already described almost all of vertebrate fossils (especially skulls and teeth) recovered from the Khok Sung sand pit during the excavation. Only few postcranial remains of mammals such as fragmentary or incomplete bones are unidentified according to the limitation of morphological accessibilities.
According to the fact that Khok Sung yields only large mammals (> 8 kg), the absence of medium-and small-sized mammal remains is likely due to taphonomic conditions and/or fossil collecting methods. Similarly to most of the Middle and Late  Pleistocene fossil sites in Southeast Asia, the biodiversity of Khok Sung large mammals is likely greater than that of present-day faunas (see Appendices 15-18 for the fossil and present-day fauna lists in South China and Southeast Asia) because the Southeast Asian fossil and present-day faunas mostly yield an average of approximately 13 species per site (Tougard and Montuire 2006) and of less than eleven species per area, respectively (Lekagul andMcNeely 1988, Corbet andHill 1992). It is obvious that the Khok Sung mammalian assemblage is characterized by genera and/or species that are similar to the living population in the same area and surrounding regions. However, some mammalian (Crocuta crocuta, Rhinoceros unicornis, Axis axis, and Sus barbatus) and reptilian (Batagur cf. trivittata) species in the Khok Sung fauna are no longer present in the region but occur far away from Thailand or even from Southeast Asia. Moreover, two taxa, Stegodon cf. orientalis and Gavialis cf. bengawanicus were present in the locality but became globally extinct later. The Khok Sung vertebrate fauna totally contains 19 of 27 identified taxa that are currently present in Thailand (Tab. 18 and Appendix 18).

Individual species distribution patterns
Past records and recent distribution patterns of large mammalian species present in Khok Sung are revealed in this work. Paleontological sites in Southeast Asia as well as South China are examined for the Early, Middle, and Late Pleistocene, compared with the modern distribution patterns. We only focus on mammalian taxa assigned to the species-level, including Stegodon orientalis and its co-occuring species, Rhinoceros sondaicus, Rhinoceros unicornis, Sus barbatus, Axis axis, Panolia eldii, Rusa unicolor, Bos sauveli, Bos gaurus, Bubalus arnee, and Capricornis sumatraensis.

Stegodontids and elephantids
The earliest records of derived Stegodon (e.g., Stegodon orientalis from Dayakou (Chen et al. 2013) and Stegodon trigonocephalus from Ci Saat (Sondaar 1984, van den Bergh et al. 2001) are likely from the Early Pleistocene. Fossils identified as Stegodon orientalis or S. cf. orientalis are recorded from South China (e.g., Daxin (Rink et al. 2008), Hejiang (Zhang et al. 2014), and Panxian Dadong (Han and Xu 1985, Bekken et al. 2004, Schepartz et al. 2005) and Vietnam (Tham Khuyen, Tham Hai, and Tham Om (Olsen and Ciochon 1990)). Another species, S. trigonocephalus, is reported from Javanese localities (van den Bergh et al. 2001). During the Middle to Late Pleistocene, Stegodon orientalis co-occurred with Elephas sp. or E. maximus in many localities throughout the Indochinese province (Fig. 37). The two species are found together from the late Middle Pleistocene of Khok Sung and the Late Pleistocene of the Cave of the Monk (Zeitoun et al. 2005(Zeitoun et al. , 2010 in Thailand, the early Late Pleistocene of Nam Lot and Tam Hang South (Bacon et al. 2008a(Bacon et al. , 2011(Bacon et al. , 2012(Bacon et al. , 2015 in Vietnam, and the Middle Pleistocene of Ganxian and Wuyun in South China (Chen et al. 2002, Rink et al. 2008, Wang et al. 2007, 2014. Stegodon orientalis is found in the Late Pleistocene of Luna (South China) and Keo Leng (northern Vietnam) caves (Olsen andCiochon 1990, Wang et al. 2014). Perhaps, this species survived until the Holocene in South China (Ma and Tang 1992, Tong and Patou-Mathis 2003, Tong and Liu 2004. The number of species of Stegodon lessens from the Early to Late Pleistocene, based on the fossil records of South Chinese localities (Louys et al. 2007). Although Stegodon orientalis is likely to have had a less widespread distribution in the Late Pleistocene than in the Middle Pleistocene (Fig. 37), the Pleistocene geographical distribution of this species is only based on a limited number of localities.
A fossil species of Palaeoloxodon is reported from several Middle Pleistocene localities in mainland Southeast Asia (Fig. 37), often co-occurring with Stegodon orientalis (e.g., the sites of Maba Xu 1985, Wu et al. 2011) and Tham Khuyen (Olsen and Ciochon 1990)). Palaeoloxodon is found in the Late Pleistocene fissure-filling deposits of Hum Hang, Lang Trang, and Ma U'Oi in northern Vietnam (Olsen and Ciochon 1990, Bacon et al. 2004, similar distribution to that of Elephas, but became extinct before the Holocene (Tong and Patou-Mathis 2003, Louys et al. 2007). The cause of global and local extinction of Stegodon orientalis and Palaeoloxodon is unknown at this time.
Elephas maximus is known from the late Middle Pleistocene of Thum Wiman Nakin (northeastern Thailand) (Tougard 1998(Tougard , 2001, and possibly reached the Indonesian islands of Sumatra, Borneo and Java during the late Pleistocene. Elephas is one of two living genera of elephants. The Indian elephant, E. maximus, is the only extant species. It is distributed throughout mainland Asia (including India, Nepal, Bangladesh, Bhutan, Myanmar, Thailand, Malaysia, Sumatra, Laos, Cambodia, and Vietnam) (Lekagul and McNeely 1988). The Indian elephant is not widespread throughout Southeast Asia as it is not found in central and northeastern Thailand and central southern Myanmar (Fig. 37). Those areas are mostly lowland or highland floodplains today, while Indian elephants prefer deep forest canopy (Lekagul andMcNeely 1988, Corbet andHill 1992). However, this preference for deep forests may be the result of humans encroaching and impacting their preferred habitats (Pushkina et al. 2010). It is possible that E. maximus became extinct locally in Java before 37 ka as it is absent Figure 37. The Middle (red) and Late (yellow) Pleistocene records of stegodontids and relative fossil elephants, and the current distribution (green) of Elephas maximus (Indian elephant). Stars indicate the co-occurrence of sympatric proboscideans. The current distribution of Indian elephants is compiled from Lekagul and McNeely (1988). from the locality of Wajak (dated to 37 ka, van den Brink (1982)). This local extinction is probably due to the drier and cooler climate beginning at 81 ka in Java (van der Kaars and Dam 1995) and/or the loss of rainforest habitats (Storm et al. 2005).

Javan and Indian rhinoceroses
The Early Pleistocene records of Asian rhinoceroses are poorly documented in Southeast Asia. Only R. sondaicus is reported from the upper part of the Irrawaddy Formation, near Pauk Township in central Myanmar (Zin-Maung-Maung-Thein et al. 2006) and from Sangiran in Java (Hooijer 1964) (Fig. 38).
The Middle Pleistocene record, especially the late Middle Pleistocene, includes numerous reports of Asian rhinoceroses (Figs 38 and 39). In the Indochinese subregion during the Middle Pleistocene, fossils of R. unicornis are found from Hsingan (Kahlke 1961) and Maba (Wu et al. 2011) (Beden and Guérin 1973). The only co-occurrences of these two species are from the late Middle Pleistocene of Thum Wiman Nakin (Tougard 1998(Tougard , 2001) and from our discoveries at Khok Sung. In the Sundaic subregion, fossils of Indian rhinoceroses have been described from the Middle Pleistocene of Tumbun (Malaysia) and Trinil H. K. (Java) (Hooijer 1962, Medway 1972, van den Bergh et al. 2001 and from the early Middle Pleistocene of Kedung Brubus where Javan rhinoceroses co-occurred (Hooijer 1946). In other biogeographic regions, R. unicornis occurred in Yenchingkou (central eastern China) (sensu Antoine 2012). According to original faunal descriptions, many Middle Pleistocene localities in China and Vietnam yielded fossil specimens of R. sinensis. This species was later synonymized with R. unicornis by Antoine (2012). However, R. sinensis is recently recognized as a valid species , so there remains some confusion about the presence of R. unicornis in many localities.
During the late Pleistocene, Javan and Indian rhinoceroses were widespread in Indochinese subregion (Figs 38 and 39). They co-occurred in the Cave of the Monk (Ban Fa Suai, northern Thailand) (Zeitoun et al. 2005(Zeitoun et al. , 2010, in Nam Lot and Tam Hang South (northern Laos) (Bacon et al. 2008a(Bacon et al. , 2011(Bacon et al. , 2012(Bacon et al. , 2015, and in Duoi U'Oi and Ma U'Oi (northern Vietnam) (Bacon et al. 2004(Bacon et al. , 2006(Bacon et al. , 2008b. Indian rhinoceros fossils were also found in the caves of Ham Hang and Keo Leng, northern Vietnam (Olsen and Ciochon 1990), while Javan rhinoceroses were recovered from Niah caves (Borneo, Malaysia) (Medway 1972, Harrison 1996 and several Indonesian localities: Lida Ajer and Sibrambang in Sumatra (de Vos 1983) and Punung, Gunung Dawung, and Wajak in Java (Badoux 1959, van den Brink 1982, Storm et al. 2005. Indian rhinoceroses seem to go extinct in Java after the middle Middle Pleistocene, as Figure 38. The Early (blue circle), Middle (red circle), and Late (yellow circle) Pleistocene records and the current distribution (green) of Rhinoceros sondaicus (Javan rhinoceros). The current distribution of the species is compiled from Groves (1967), Rookmaker (1980), and Groves and Leslie (2011). none are reported from Trinil H. K. (dated to ~540-430 ka, Joordens et al. (2014)) and early Late Pleistocene to Holocene sites.
Nowadays, the Indian rhinoceros is locally extinct from the Thai territory and several other countries in Southeast Asia. The species is restricted to Nepal and India and some parts of northernmost Myanmar (Laurie et al. 1983) (Fig. 39). The Javan rhinoceros survives across the Indochinese Peninsula and the Sundaic subregions (Groves and Leslie 2011) but became extinct in the island of Borneo during the Holocene (Medway 1960, Cranbrook 2000, Cranbrook and Piper 2007 (Fig. 38). The modern co-occurrences of the two species are restricted to a small area in eastern India (Antoine 2012). In the Holocene, the Javan rhinoceros likely co-occurred with the Sumatran rhinoceros, Dicerorhinus sumatrensis, but they are not sympatric today almost certainly because of human induced habitat loss leading to reduction of their geographic range during the last century (Groves and Leslie 2011).

Bearded pigs
During the Middle Pleistocene, Sus barbatus (bearded pig) is known from the caves of Thum Wiman Nakin and Thum Prakai Phet (Tougard 1998(Tougard , 2001 and the terrace Figure 39. The Middle (red circle) and Late (yellow circle) Pleistocene records and the current distribution (green) of Rhinoceros unicornis (Indian rhinoceros). "?" indicates the possible record of R. unicornis according to Antoine (2012). The current distribution of the species is modified from Laurie et al. (1983). deposit of Khok Sung (Fig. 40). Among these Thai localities, S. barbatus co-occurred with S. scrofa at least in Thum Wiman Nakin and Thum Prakai Phet.
In the late Pleistocene, S. barbatus is well-documented from many localities, extending its geographic distribution across Sumatra, Borneo, and Java. This species is likely more widespread in the late Pleistocene than the Middle Pleistocene (Fig. 40). In Indochinese and Sundaic subregions, the co-occurrence of S. barbatus and S. scrofa is known from the "Cave of the Monk" (Ban Fa Suai) in northern Thailand (Zeitoun et al. 2005(Zeitoun et al. , 2010, Tam Hang South in northern Laos (Bacon et al. 2008(Bacon et al. , 2011(Bacon et al. , 2015, Batu caves and Gua Cha (Holocene) in Peninsular Malaysia (Groves 1985, Ibrahim et al. 2013), Lida Ajer and Sibrambang in Sumatra (de Vos 1983, and Punung in Java (Badoux 1959). Only fossils of bearded pigs are collected from the latest Pleistocene of Niah Cave, Borneo (Medway 1972, Harrison 1996. Today S. barbatus is restricted to Peninsular Malaysia, Sumatra, and Borneo (Corbet and Hill 1992) (Fig. 40), in contrast with its widespread distribution across the Indochinese subregion during the Middle to Late Pleistocene. This species dispersed to Indonesian islands by the Late Pleistocene, as it is recorded from Punung of Java (Badoux 1959). After the land bridges submerged by rising sea level, some populations of S. barbatus were probably trapped on islands (Tougard 2001). Later on, S.  Corbet and Hill (1992).
barbatus went extinct in mainland Southeast Asia after the late Pleistocene. The cause of local extinction of S. barbatus in mainland Southeast Asia is unknown at this time. This taxon also became locally extinct later in Java as none is recorded from the Late Pleistocene of Wajak site (van den Brink 1982). The drier and cooler climates during the middle Middle Pleistocene or the reduction of rainforest habitats possibly explain the local extinction for bearded pigs in Java.

Chitals
Fossils of Axis axis (chital) have never been previously recorded from Thailand but were present in mainland Southeast Asia, at least in Khok Sung, during the late Middle Pleistocene (Fig. 41). Only Axis cf. porcinus is reported from the Late Pleistocene of the Cave of the Monk (Zeitoun et al. 2005(Zeitoun et al. , 2010. Other species of Axis are also described in Asia. A. shansius and A. rugosa are reported from the Early Pleistocene of China (Han and Xu 1985), whereas A. lydekkeri is recorded from the Early to Middle Pleistocene of Java (Gruwier et al. 2015). The Bawean deer, A. kuhli, is also reported in Java since the Holocene (van den Bergh et al. 2001, Moigne et al. 2004). Nowadays Axis axis is restricted to the Indian Subcontinent (India, Nepal, Sikkim, and Sri Lanka) (Fig. 41). Its habitat preferences are grasslands and open forests (Nowak 1999). The Pleistocene chital has a different geographical distribution as it was present in Khok Sung. The distribution range of A. axis in the Pleistocene is probably wider than in the present day. Rainforests became more dominant across Southeast Asia during the Late Pleistocene (Heaney 1991, Meijaard 2003, Louys et al. 2007). The local extinction of the chital in Thailand is likely caused by the reduction of open grasslands. In the future, additional fossil records of A. axis in Southeast Asia would allow to address some issues related to its local extinction, as well as its past distribution.
Nowadays, Panolia eldii is restricted to the Indochinese province (Fig. 42). Rusa unicolor is a widespread species native to the Indian Subcontinent, southern China, and Southeast Asia (both Indochinese and Sundaic subregions with the exception of Java (Fig. 43)) (Lekagul and McNeely 1988).

Koupreys, gaurs, and wild water buffaloes
Large bovids in Southeast Asia currently comprise four wild species: Bos sauveli (kouprey), Bos javanicus (banteng), Bos gaurus (gaur), and Bubalus arnee (wild water buffalo). Bantengs, gaurs, and koupreys presumably shared a common ancestor at 2.6 Ma (Plio-Pleistocene) and their lineages split in a short period of time (i.e., between 200 and 300 ka) based on the molecular estimations of divergence times (Hassanin and Ropiquet 2004). These molecular estimations are congruent with the fossil records of bantengs and gaurs in Asia. Fossil remains attributed to these species have been recorded in Southeast Asia since the Middle Pleistocene. The co-occurrence of these Pleistocene large bovids is reported from Thum Wiman Nakin (Tougard 1998(Tougard , 2001 and Khok Sung in northeastern Thailand (Figs 44-46). Fossil remains of gaurs are also reported from the Middle Pleistocene of Kao Pah Nam in northern Thailand (Pope et al. 1981), the middle Middle Pleistocene of Tham Khuyen and the late Middle Pleistocene of Tham Om in Vietnam (Olsen and Ciochon 1990), and the Middle Pleistocene of Yenchingkou in central eastern China (Colbert and Hooijer 1953) (Fig. 45). In addition, remains of fossil water buffaloes are described from the late Middle Pleistocene of Phnom Loang and Boh Dambang in Cambodia (Beden andGuérin 1973, Demeter et al. 2013).    Lekagul and McNeely (1988) and Hedges et al. (2008).
During the Late Pleistocene, the locality of the Cave of the Monk (Ban Fa Suai) yielded remains of these bovid species (cf.) (Zeitoun et al. 2005(Zeitoun et al. , 2010. Other localities yielded either only one species of Bos or the co-occurrence of two Bos species and Bubalus. Bubalus arnee occurred not only in Sumatra but also in Java during the latest Middle/early Late Pleistocene according to their fossil records in Sibrambang and Punung (Badoux, 1959, de Vos 1983, Storm and de Vos 2006, respectively (Fig. 46). Both taxa disappeared subsequently in Sumatra either after the early Late Pleistocene or during the Holocene. Neither koupreys nor gaurs are identified in insular Southeast Asia, thus most likely restricted to mainland Southeast Asia (Fig. 44).
The historical distribution of koupreys during the last century is restricted to Cambodia, southern Laos, southeastern Thailand, and western Vietnam (Lekagul andMcNeely 1988, Corbet andHill 1992). They become globally extinct today. Gaurs recently occur throughout mainland South and Southeast Asia and Sri Lanka (Lekagul andMcNeely 1988, Duckworth et al. 2008b) (Fig. 45). Nowadays, they are also present in South China where their fossils have never been found. Wild water buffaloes are currently native to Bhutan, Cambodia, India, Myanmar, Nepal, and Thailand (Lekagul andMcNeely 1988, Hedges et al. 2008). They become locally extinct in Vietnam (likely), Laos, Indonesia, Sri Lanka, and Bangladesh (Fig. 46).
Overall, the Pleistocene large bovid species in Southeast Asia is more widespread than the modern population. The anthropogenic impacts on the environments and land-scapes seem to have caused the reduction of large bovid population in several areas during the past decade. The koupreys is more widely distributed during the Pleistocene than today (Fig. 44). In addition to the human activity, the cause of reduction and extinction of koupreys is likely due to their high degrees of habitat specificity such as deciduous dipterocarp forests and especially in areas with extensive grasslands (Timmins et al. 2008), and/or according to high levels of niche competition with other large bovids.

Sumatran serows
The possible earliest records of Capricornis sumatraensis are from the middle Early Pleistocene site of Gongwangling (Hu andQi 1978, Han andXu 1985), dated to 1.63 Ma (Zhu et al. 2015), in central mainland China and from the Early Pleistocene of the Upper Irrawaddy Formation (Colbert 1938 in central Myanmar. C. sumatraensis during the Middle Pleistocene is widespread throughout mainland Asia and Southeast Asia (Fig. 47). It is known from the Middle Pleistocene of Yenchingkou in central eastern China (Colbert and Hooijer 1953), Wuming, Panxian Dadong, and Wuyun in South China (Han and Xu 1985, Chen et al. 2002, Bekken et al. 2004, Schepartz et al. 2005, Rink et al. 2008, Wang et al. 2007, 2014, Tham Om in Vietnam (Olsen and Ciochon 1990), Thum Wiman Nakin, Thum Prakai Phet, and Khok Sung in Thailand (Tougard 1998, Filoux et al. 2015, Boh Dambang in Cambodia (Demeter et al. 2013), and Badak Cave in Peninsular Malaysia (Ibrahim et al. 2013). Fossils of C. sumatraensis are also described from the latest Middle/early Late Pleistocene of Lida Ajer and Sibrambang in Sumatra and of Punung in Java (Badoux 1959, de Vos 1983, van den Bergh et al. 2001, Storm and de Vos 2006. However, no serows are recorded from Borneo. The Sumatran serow is a widespread species, native to mountain forests on the Himalayan range (northern India, Sikkim, and Nepal) of the Indochinese subregion (Southern China, Myanmar, Thailand, Laos, Cambodia, Vietnam, and Peninsular Malaysia) and on the island of Sumatra (Lekagul and McNeely 1988) (Fig. 47). C. sumatraensis became locally extinct in Java during the middle Late Pleistocene according to the lack of fossil records in Wajak (~37 ka). The advocated cause for the local extinction of serows is possibly related to the unfavorable climatic conditions. The drier and cooler climate that occurred after 81 ka in Java (van der Kaars and Dam 1995) probably affects significantly the niche preferences of forest-dwelling taxa.

Faunal comparisons of the assemblage with other penecontemporaneous assemblages
For the comparisons of vertebrate faunas between Khok Sung and other Pleistocene sites, we focus only on large mammals (for the mammalian fauna lists of the Middle to Late Pleistocene of Southeast Asian sites, see Appendices 16 and 17). The identification of the family level referred to "indet." and the species level designated "sp." are herein excluded from our comparisons. The Khok Sung large mammalian assemblage yields most extant and some extinct taxa, which are characteristic of the Ailuropoda-Stegodon assemblage. Compared to other Thai Pleistocene faunas, the Khok Sung mammalian assemblage shares 10 ten species with Thum Wiman Nakin (Tougard 1998(Tougard , 2001, six species with Thum Prakai Phet (Tougard 1998, Filoux et al. 2015, and nine species with the Cave of the Monk (Zeitoun et al. 2005(Zeitoun et al. , 2010. However, most of the mammalian taxa from the Cave of the Monk are assigned to "cf." (the open nomenclature) and the presence of fossil spotted hyaena, Crocuta crocuta, in this locality is still doubtful, i.e. only one fragmentary tooth is identified as belonging to Hyaenidae indet. by Zeitoun et al. (2005Zeitoun et al. ( , 2010) (Appendix 16). Compared to the surrounding Pleistocene faunas, the Khok Sung mammalian assemblage has taxonomic similarities of seven species with Nam Lot (Bacon et al. 2012(Bacon et al. , 2015, eight species with Tam Hang South (Bacon et al. 2008a(Bacon et al. , 2011(Bacon et al. , 2015, four species with Tham Khuyen (Olsen and Ciochon 1990), two species with Tham Hai (Olsen and Ciochon 1990), five species with Tham Om (Olsen and Ciochon 1990), four species with Hang Ham (Olsen and Ciochon 1990), five species with Keo Leng (Olsen and Ciochon 1990), four species with Lang Trang , three species with Ma U'Oi (Bacon et al. 2004(Bacon et al. , 2006, six species with Duoi U'Oi (Bacon et al. 2008b), four species with Boh Dambang (Demeter et al. 2013), and four species with Phnom Loang (Beden and Guérin 1973) (Appendices 16 and 17). The Khok Sung assemblage is more different from other Pleistocene faunas, especially from the Indonesian islands, which mainly yield endemic forms. According to the number of shared taxa, the Khok Sung mammalian assemblage more nearly resembles diversified faunas from Thum Wiman Nakin, Thum Phra Khai Phet, Nam Lot, and Tam Hang South than the others.
The Khok Sung assemblage shares at least one similar archaic mammal taxon such as Crocuta crocuta ultima and Stegodon orientalis, with these faunas. Crocuta crocuta is also recorded from Thum Wiman Nakin, Thum Prakai Phet, and Nam Lot, whereas Stegodon orientalis is reported from two Laotian sites: Nam Lot and Tam Hang South. By the way, most of forest dwelling and carnivorous taxa that are representatives of Middle Pleistocene mammalian assemblages such as Ailuropoda melanoleuca (giant panda), Ursus thibetanus (Asiatic black bear), Pongo pygmaeus (orang-utan), Muntiacus muntjak (Southern red muntjac), and Tapirus indicus (Malayan tapir) are absent in Khok Sung. The paleoenvironments of Khok Sung corresponded to a floodplain near the river channel (Duangkrayom et al. 2014, Suraprasit et al. 2015. The absence of most of these taxa in Khok Sung is likely explained by the local environments that are unfavourable to those species. Although some forest-inhabiting taxa (e.g., Elephas maximus and Capricornis sumatraensis) are found in the locality, these fossils (rare, fragmentary, or represented by isolated teeth only) were transported from the surrounding upland forests by the river.
The degree of the faunal similarity also depends on the number of identified taxa for each site. We further analyse the relationships between the geographic regions and faunas in Southeast Asia, using the Simpson coefficient of faunal similarity (Tab. 19) performed with the multivariate clustering analysis. The final dataset analysed for the similarity comprises 18 localities and 85 taxa. The analysis is based on the presence/ absence of mammalian taxa in the fauna lists complied from literatures (Appendices 15 and 16).
As a result, the Middle Pleistocene Southeast Asian taxa reveal two distinct associations (Javanese and mainland Southeast Asian faunas) (Fig. 48). Within the mainland Southeast Asian assemblages, the cluster analysis resolves two different groups between the Thai, Combodian, Vietnamese, and Chinese faunas (South China and Yenchingkou) and the central-eastern Chinese one (Koloshan) (Fig. 48). Among South Chinese localities, Hoshantung fauna is a distinct subcluster separated from other mainland Southeast Asian faunas. Hoshantung probably represents a different biochronological age from each other rather than high levels of endemism. The Thai and Cambodian faunas constitute a distinctive subgroup that is differentiated from the Vietnamese and Chinese assemblages. Within the Thai and Cambodian members, the Khok Sung fauna characterizes a distinct subcluster separated from three late Middle Pleistocene assemblages: Thum Wiman Nakin, Thum Prakai Phet, and Boh Dambang (Fig. 48), although the fauna of Khok Sung is most similar in composition to that of Thum Table 19. Wiman Nakin according to the Simpson's index (Tab. 19). This is likely due to the convention of the UPGMA method, which produces equal length branches from all nodes, and to the effects of higher faunal similarity between Thum Wiman Nakin and two other faunas.
Overall, this analysis suggests initially that the differences in species composition and distribution do not follow a trend of the latitudinal gradient north to south, but show spatial and time variability of large mammalian fauna in Southeast Asia. The main problems of mammalian fauna comparisons in Southeast Asia are likely due to the poorly-known species diversity and/or the imprecisely chronological determination in several localities.

Biochronology of Khok Sung fauna
According to the similarity analysis of the fauna, the mammalian fauna composition of Khok Sung is considerably different from the Early to early Middle Pleistocene assemblage of Java. This suggests an inconsistent age of the Early Pleistocene for Khok Sung. The Khok Sung assemblage is highly comparable in composition to three late Middle Pleistocene faunas: Thum Wiman Nakin (> 169 ka, Esposito et al. (1998(> 169 ka, Esposito et al. ( , 2002), Thum Prakai Phet (Tougard 1998, Filoux et al. 2015, and Boh Dambang (Demeter et al. 2013). However, our faunal comparisons suggest that the biochronological age of Khok Sung is possibly different, slightly older or younger, from those three localities according to some of the compositional dissimilarity. Two early Late Pleistocene sites: Nam Lot (≈86-72 ka, Bacon et al. (2015)), and Tam Hang South (≈94-60 ka, Bacon et al. (2015)) possibly remains contemporaneous according to the occurrence of several taxa sharing with Khok Sung (> 7 species).
On the basis of the previous paleomagnetic data analysed by Suraprasit et al. (2015), a short reversal polarity registered in a fine layer of silty mud lenses (Fig. 2) that occurs within the Brunhes normal chron could be presumably correlated to the geomagnetic excursions of either "Iceland Basin" (188 ka) or "Pringle Falls" (213 ka). In addition, the short reversal event of the paleomagnetic field in Brunhes normal chron is possibly correlated to "Blake" excursion (dated to around 120 ka, Lund et al. (2001)). However, we suggest a late Middle Pleistocene age rather than a Middle/Late Pleistocene transition according to the occurrence of several archaic taxa and to the closest faunal similarity with Thum Wiman Nakin.

Evolutionary and biogeographic affinities of Khok Sung fauna
Relationships of the Khok Sung vertebrate fauna for dispersal events from India to Java has been first proposed by Martin et al. (2012). Gavialis bengawanicus and Crocodylus siamensis as well as monitor lizards and pythons are known as typical taxa associated with the Stegodon-Homo erectus fauna, which presumably originated from the Miocene-Pliocene of Siwalik faunas in India and Pakistan (Head 2005, de Vos 2007, Hocknull et al. 2009, Delfino and de Vos 2010, Martin et al. 2012). These taxa migrated from mainland Southeast Asia to Java, via the Siva-Malayan route, by the Early Pleistocene as they are first recorded from the Early Pleistocene of Java (von Koenigswald 1935, de Vos 1995, 2007, de Vos and Long 2001, Delfino and de Vos 2010 (Fig. 1 and Appendix 16). According to the occurrence of Gavialis cf. bengawanicus in Khok Sung, Martin et al. (2012) hypothesized that this species reached Java through the fluvial drainages of Sunda shelf (rather than the dispersal by sea) during a low sea level event (with a minimum of about 170 m below the present day) of the Early-Middle Pleistocene transition (around 0.8 Ma) (Prentice and Denton 1988, van den Bergh et al. 2001, van der Geer et al. 2010 (Fig. 50). In the light of this scenario, G. bengawanicus might have appeared either earlier than or during the Early Pleistocene in Thailand.
However, in terms of faunal age, this scenario is no longer consistent because the Khok Sung fauna is recently attributed to a late Middle Pleistocene age (Suraprasit et al. 2015), younger than Gavialis and Crocodylus-bearing localities in Java. We propose that gharials and some other vertebrates (e.g., a freshwater crocodile, a large varanid, and a python) present in Khok Sung are possibly geographical remnants of the former Siva-Malayan fauna that survived until the late Middle Pleistocene as they occurred earlier in Java. Otherwise, these vertebrates possibly appeared either firstly or repeatedly (if the local extinction of those taxa previously occurred) in Thailand during the late Middle Pleistocene. Several cyclic occurrences of high amplitude glacial periods (~50 times since the last 2.7 Ma, Woodruff (2010)), related to the sea level lowering, during the Early to Middle Pleistocene (Prentice and Denton 1988, van der Kaars 1991, Zheng and Lei 1999 could provide high possibilities to facilitate faunal exchange between mainland and insular Southeast Asia (via the land bridges or Sunda shelf). The faunal exchanges by corridor and/or filter bridge dispersal between Thailand and Java might have occurred habitually during the glacial events.
Based on the occurrence of mammalian taxa in Khok Sung, we suggest an alternative relevance of this fauna for the "Sino-malayan" dispersal events from mainland Southeast Asia to Java (Fig. 1). This evidence is supported by the faunal turnover that occurred in Punung (Java), around 128 to 118 ka dated by luminescence and U-series analysis performed on the breccias (Westaway et al. 2007). The modern rainforest assemblage, known as the Pongo-Homo sapiens or Elephas-Homo sapiens fauna, has replaced the former Stegodon-Homo erectus faunal association in Java during since latest Middle Pleistocene (Westaway et al. 2007). The new faunal elements include Elephas maximus, Pongo pygmaeus, Symphalangus syndactylus (siamang), Macaca nemestrina (pig-tailed macaque), Panthera tigris (tiger), Dicerorhinus sumatrensis (Sumatran rhinoceros), Helarctos malayanus (sun bear), Capricornis sumatraensis, Bubalus arnee, Sus scrofa, and Sus barbatus. The Khok Sung mammalian assemblage consists of at least 4 of forest dwelling mammals: Capricornis sumatraensis, Bubalus arnee, Sus barbatus, and Elephas sp. (Appendices 16 and 17). These taxa presumably migrated from mainland Southeast Asian to Java and some of them are living today in the mainlafnd Southeast Asia (van den Bergh et al. 2001, van der Geer et al. 2010. The presence of exclusive tropical rainforest species in Punung indicates that their migration event could have occurred following the dry and open woodland environments of the penultimate glaciations at about 135 ka (de Vos 1983, de Vos et al. 1994. These mammals migrated southward to the exposed Sunda shelf that occurred during the late Middle Pleistocene (between 135 to 125 ka), when the sea level dropped about 150 m (van der Kaars 1991, Zheng and Lei 1999). The Sundaland was then covered partly by a savannah corridor, stretching from Thailand to the Lesser Sunda Islands (Morley andFlenley, 1987, Heaney 1991). This corridor served as a barrier to the dispersal of the rainforestdependent species. However, the forest-dwelling mammals survived in rainforest refugia for a while before reaching Java (van den Bergh et al. 2001).
On the other hand, the Khok Sung fauna lacks any evidence of taxa originating from Java. But the possible presence of Duboisia santeng in Tambun site (Peninsular Malaysia) may indicate the faunal exchange from Indonesia to the mainland Southeast Asia (Hooijer 1962, Medway 1972, Tougard 2001. D. santeng is described from the early Middle Pleistocene of Kedung Brubus and the middle Middle Pleistocene of Trinil H. K. (Hooijer 1958). This taxon presumably arrived on the island of Java via the Siva-Malayan route (von Koenigswald 1935, Tougard 2001. The poor record or absence of the Indonesian taxa in mainland Southeast Asia is likely due to the disappearance of the land bridge during the interglacial phase. This acted as a sea barrier that did not facilitate insular mammals to migrate out of the islands. The Khok Sung mammalian assemblage supports that Thailand was a biogeographic gateway of the Sino-Malayan migration event as the mainland forested faunal association replaced the earlier Siva-Malayan fauna (Stegodon-Homo erectus complex) subsequently in Java (von Koenigswald 1938, de Vos 1995. The glacial episodes are likely a key factor of southward onland dispersal of large mammals via the Sunda shelf. In addition, the occurrence of the Khok Sung reptiles is not truly representative of the early Siva-Malayan refugees but represents practically long-term survivors (e.g., Crocodylus siamensis, Heosemys annandalii, and Heosemys grandis) that evidently continued to exist up until today in Thailand. de Vos J (1983) fig.11). * indicates measurements of the maximum length of the preservation according to incomplete specimens. "es" refers to an estimated value of the full length due to incomplete specimens.     fig. 21). * indicates measurements of the maximum length of the preservation according to incomplete specimens. "es" refers to an estimated value of the full length due to incomplete specimens.

Bos gaurus Bubalus arnee
Mandible no.   fig. 8). * indicates measurements of the maximum length of the preservation according to incomplete specimens. "es" refers to an estimated value of the full length due to incomplete specimens.

Naemorhedus goral
cf. Fauna lists of extant large mammalian species from Southeast Asia and South China. The biogeographic affinities of the mammalian species are given and abbreviated as (I) for Indochinese taxa, (S) for Sundaic taxa, (O) for other geographic taxa (e.g., Palearctic and Indian regions), and (W) for widespread taxa. Data are compiled from Lekagul and McNeely (1988), Corbet and Hill (1992), and Nowak (1999). The binomial nomenclature is modified from categories of the IUCN red list of threatened species (2015) and from Groves and Grubb (2011).