Introduction

The fossil record is fragmentary by nature, even for organisms that produce mineralized skeletal elements, and paleontologists therefore have to contend with the small amount of material that is available for study. This is particularly true for the terrestrial fossil record, because terrestrial sedimentation is temporally and spatially restricted and because a wide array of taphonomic factors hinder terrestrial organisms from entering the fossil record. Against all odds, organisms nevertheless do enter the fossil record and, occasionally, are preserved in exceptional quality or in exceptional quantity in localities known as Fossillagerstätten (Kidwell et al. 1986; Seilacher 1970). The taphonomic circumstances are varied that lead to the exceptional preservation of organisms at any given locality, but most Fossillagerstätten either yield fossils with exceptional preservation or in exceptional quantity and can therefore be classified readily as Konservatlagerstätten or Konzentratlagerstätten, respectively (Seilacher 1970). Although most depositional systems in principle have the ability to produce Fossillagerstätten, the vast majority of such localities are known from the marine realm.

Here we report a new Konzentratlagerstätte (informally referred to as “Mesa Chelonia”) from the Turpan Basin in Xinjiang Autonomous Province, China, that probably belongs to the late Middle Jurassic Qigu Formation of that preserves an aggregation of an estimate of >1,000 freshwater aquatic “xinjiangchelyid” turtles in a single layer within an area less than a hectare. Although much will be learned regarding the alpha taxonomy and systematics of “xinjiangchelyid” turtles from this site in the future, we herein only provide a preliminary description of part of the collected material and focus on the taphonomic settings that led to the formation of this rare terrestrial Konzentratlagerstätte and a comparison to other fossil and recent localities with turtle accumulations.

Materials and methods

The turtle Konzentratlagerstätte described herein was discovered in April 2008 by a team of the Sino-German Cooperation Project, but limited time did not allow for the excavation of fossils or study of the site. In April 2009 and October 2011, subsequent field teams revisited the locality, identified and documented the extent of the fossiliferous layer, and quarried several large turtle-rich blocks of matrix. The quarried fossils are currently being stored in the nearby city of Shanshan, Xinjiang, and will eventually be integrated into the municipal museum that is currently under construction. All specimens have been assigned Sino-German Cooperation Project numbers, which will be deposited with the specimens once the museum is operational. The detailed coordinates of the locality will be archived at the museum as well and will be disclosed to qualified researchers interested in studying the site.

To assess the number and distribution of turtles preserved at the site, the fossiliferous layer was fully exposed by removing loose, weathered sediment with brushes in an approximately 30 m long section along both flanks of the mesa. For practical purposes, the lateral extent of the fossiliferous layer was determined through the absence of weathered fossil turtle shell fragments within the 10 cm thick weathering zone. For estimating the number of preserved turtles, all hyo- and hypoplastral elements were counted that could be found within the weathering horizon closest to the surface. The hyo- and hypoplastron were chosen because the axillary and inguinal notches of these elements are greatly thickened in “xinjiangchelyid” turtles and can easily be identified, even if the element has otherwise disintegrated fully within the weathering horizon.

A half square meter block was removed from the most fossil-rich portion of the layer in 2009 using traditional plaster and burlap techniques and prepared using micro air scribes under makeshift conditions in the town of Shanshan. The distribution of turtles contained in the block was documented using photography and illustrated using a decimeter grid. The best-preserved material recovered from this block is the basis of the morphological descriptions and taxonomic assessments provided herein. The retrieval of several, still unprepared turtle-rich blocks in 2011 was facilitated by the use of a diamond chain rock saw. The resulting clean saw cuts were photographed and carefully examined for sedimentological features.

Geological settings

The bone bed is located about 25 km ENE of the city of Shanshan in the Turpan Basin (Fig. 1), a small intermontane foreland basin that formed during the Late Permian (Shao et al. 1999; Hendrix et al. 1992). Today, the basin is geographically separated by mountain ranges from the Junggar Basin to the North and the Tarim Basin to the South. Jurassic to Paleogene sediments are exposed within the basin along a central ridge that was uplifted during the Neogene (Zhao 1980; Dong 1992; Wings et al. 2007). Published reports on the geology and stratigraphy of the Turpan Basin are sparse (Shao et al. 1999, and references therein) and many uncertainties still remain regarding the absolute age of formations and their correlation with similar units in adjacent basins. The Jurassic sediments are divided into the Early Jurassic Sangonghe Formation, the Middle Jurassic Xishanyao, Sanjianfang, Qiketai, and Qigu Formations (the latter was recently dated in the Junggar Basin with 164.6 Ma ± 1.4 Ma, Wang and Gao 2012), and the Late Jurassic Karaza Formation (Dong 1997). These Jurassic strata generally consist of variegated, coarse clastics, although lacustrine swamp coals are present in the Middle Jurassic as well (Shao et al. 1999; Wings et al. 2007) and piedmont-fluvial deposits in the Late Jurassic (Liu and Di 1997). Red-colored sediments, especially prominent in the Qigu Formation, indicate a reduction in the monsoonal circulation in Asia resulting in a paleoclimatic change from humid to seasonally dry during the late Middle and early Late Jurassic (Shao et al. 1999; Hendrix et al. 1992; Eberth et al. 2001; Wang and Gao 2012). Future stratigraphic research needs to prove if Late Jurassic strata are indeed mostly absent in the Turpan Basin, as has recently been suggested for the Junggar Basin (Wang et al. 2012).

Fig. 1
figure 1

The geographic location of Mesa Chelonia in the Turpan Basin of Xinjiang Autonomous Province, China (above) and a photograph of the eastern side of the new turtle Konzentratlagerstätte at Mesa Chelonia (below). The Pleistocene conglomerate that caps the mesa and the underlying Jurassic sediments that dip at an angle of 60° towards the viewer are apparent. The arrow indicates a broad surface along which the fossil rich inner zone is exposed and from where the turtle-rich block with matrix was collected

The Qigu Formation is well known in the Junggar Basin and has been also identified in the Turpan Basin, but no explicit justification has been given for this correlation (Shao et al. 1999) and the assignment of units to this formation is not transparent. The deposits that allegedly represent the Qigu in the Turpan Basin are characterized by alternating coarse and fine-grained sediment packages that often contain freshwater bivalves (Unionidae), reflecting the changing depositional conditions typical of river systems (Schwermann 2010; Shao et al. 1999). Fine-grained, often laminated sediments were deposited in flood plains, whereas coarse sediments that often contain reworked mud clasts were formed in channels with greater water energy. Mottled, purple and green mudstones combined with peds indicate the formation of paleosols as a result of temporary subaerial exposure. Dominating fine-grained sediments with only occasional coarser intercalations were deposited in the lower part of the Qigu Formation in the Turpan Basin in a meandering river system with relatively low water energy. In the upper part, however, coarser sediments and lenticular conglomerates appear, indicating a change in the depositional environment to a braided river system (Schwermann 2010).

The turtle bone bed is situated at the transition between a rock unit consisting of red and green striped sediments below and a massive unit consisting of deep red sediments above. Based on the above-mentioned sedimentological characteristics, we interpret the turtle bone bed as being situated in the lower part of the Qigu Formation in the Turpan Basin, because this interpretation is consistent with previous stratigraphic schemes. We explicitly note, however, that the classification of sediments in the Turpan Basin is still in its infancy and that more detailed stratigraphic work may arrive at other results in the future. All fossil turtle material is concentrated in a single horizon that is only 10–20 cm thick and laterally restricted to less than 100 m. The stratum dips with approximately 60° to the north and is exposed on the west and east sides of a mesa that is capped by Pleistocene alluvial deposits (Fig. 1; ESM Fig. 1). Observations in the field reveal that complete turtle skeletons occur in extremely high concentrations within a relatively small and well-defined area of approximately 10 m length near the top of the eastern side of the mesa. Towards the west and east, this high-concentration zone is replaced by a marginal zone containing fewer, disarticulated turtles (Fig. 2). Fossil remains from the marginal zone lack evidence of abrasion, transport, or sorting. From the center of the high concentration zone, the marginal zone stretches 10 m along the western flank of the mesa and 30 m along the eastern flank. Given that the fossiliferous layer is tilted at an angle of 60°, the exposed (and measured) flanks of the mesa stand roughly perpendicular to one another in the dipping direction of the turtle layer.

Fig. 2
figure 2

Stratigraphic overview of the Konzentratlagerstätte exposed at Mesa Chelonia, Xinjiang Autonomous Province, China. Along the western side of the mesa, complete turtles are apparent along the cliff face, but the bone-bearing layer quickly disappears within 3 m. Mesa Chelonia is approximately 7 m across at the level of the turtle layer. To the east, turtle shells are exposed along the cliff face of the mesa, the high concentration fossil zone is 3 m wide, and the laterally adjacent outer zone continues for at least another 25 m. Note the fluctuations in thickness and compositions of the sediments. Modified from Schwermann (2010)

Sedimentology

A reddish-brown vertisol horizon indicates influence of pedogenesis in the exposed section below the turtle-rich layer. The horizon has a lumpy texture and faintly shows light-colored, centimeter-thick layers and subvertical knoll lines. The vertisol is overlain by a light-colored horizon of fluvial siltstone with arc-shaped bands of lithoclasts, small-scale cross bedding structures, and obscured, internal small-scale ripple marks. There is no erosion to the lower bedding plane of the siltstone layer. However, the upper surface of the partially eroded siltstone layer situated directly below the turtles shows a high relief and is plastically deformed (ESM Fig. 2). The fossiliferous stratum is a grayish-green siltstone with a fine sand fraction that contains subrounded mudstone lithoclasts with a diameter of 1–10 mm, which are especially abundant in the highly fossiliferous zone. The thickness of the stratum, clast size, and turtle bone density rather abruptly decrease towards the marginal zone and gradually decrease toward the margins of the fossiliferous layer. The lower contact of the high concentration zone reveals that the fossiliferous layer is highly irregular in thickness and that it locally penetrates deeply into and/or mingles with the underlying siltstone. In addition to its erosive base, the turtle layer consistently shows irregular bedding planes. Contorted portions of the turtle layer are only a few centimeters wide and contain vertically embedded turtle shells. The turtle shells have a carapace length of 15–20 cm. Numerous carcasses are disintegrated, but abrasion or sorting of bones is not evident. A standard petrographic thin section (ESM Fig. 3) of the turtle layer reveals a poorly sorted siltstone that contains a large number of subangular quartz grains ranging from 20 to 100 μm in diameter. Rounded mudstone clasts from 0.1 to 10 mm in diameter are suspended in the siltstone matrix.

The prepared block of fossil turtles (dimensions: 1.0 m × 0.6 m × 0.1 m) was quarried from an exposed portion of the inner zone that was surficially weathered and therefore only included the stratigraphically deeper portions of the fossiliferous layer. The block contained the remains of at least 18 turtles, which were embedded in a weathered, soft matrix (Fig. 3). Several specimens were preserved with limbs and tails preserved in articulation, although the advanced degree of weathering and their fragile nature did not allow preserving these under the available conditions. The remains of at least four skulls were found during preparation. In parts of the block, turtle shells were stacked directly on top of each other, whereas in other portions, turtle shells are stacked against and next to each other. The turtles were therefore buried in a dense stacking pattern.

Fig. 3
figure 3

Field jacket within Annemys sp. from the late Middle Jurassic Konzentratlagerstätte at Mesa Chelonia, Xinjiang Autonomous Province, China: a partially prepared jacket photograph in slightly oblique view, b schematic diagram depicting the original orientation of all turtles found in the block. Significant differences are apparent between the photo and the reconstruction, as many turtles seen in the reconstruction were already removed during earlier phases of preparation and because weathered turtles as well as turtles found along the margins of the jacket are depicted as complete shells. Each grid element corresponds to a square decimeter

The immediate top of the turtle layer consists of intercalated yellow-reddish fine-grained mudstones interpreted as overbank deposits. Lumpy weathering and red-colored areas, often void of internal structures, indicate renewed pedogenetic processes. Several detailed profiles along both sides of the mesa (Fig. 2) reveal that lithoclastic conglomerate bands occur in lenses above and below the fossiliferous layer that are limited in lateral extent to several meters. No root structures, invertebrate traces, or evidence of subaerial exposure (e.g., desiccation cracks, caliche) are present in the entire sequence.

Systematic paleontology

Testudines Batsch 1788

“Xinjiangchelyidae” Nessov in Kaznyshkin et al. 1990

Annemys Sukhanov and Narmandakh 2006; (Sukhanov 2000)

Annemys sp.

Figure 4

Fig. 4
figure 4

Annemys sp. from the late Middle Jurassic Konzentratlagerstätte at Mesa Chelonia, Xinjiang Autonomous Province, China: a carapace and b plastron of SGP 2009/14; c carapace of SGP 2009/4; d carapace and e plastron of SGP 2009/9; f dorsal view and g ventral view of skull of SGP 2009/18; h position of depicted fossils within the block recovered from the fossil rich inner zone of the Konzentratlagerstätte. Abbreviations: Abd abdominal scute, An anal scute, bo basioccipital, bps basisphenoid, co costal, epi epiplastron, ex exoccipital, Fe femoral scute, fpccc foramen posterius canalis caroticum cerebrale, fpcci foramen posterius canalis caroticum interni, fpccl foramen posterius canalis caroticum laterale, fpp foramen palatinum posterius, fr frontal, Hu humeral scute, hyo hyoplastron, hypo hypoplastron, ju jugal, Ma marginal scute, mx maxilla, na nasal, ne neural, nu nuchal, op opisthotic, pa parietal, pal palatine, Pec pectoral scute, per peripheral, pf prefrontal, Pl pleural scute, pmx premaxilla, po postorbital, pro prootic, pt pterygoid, qu quadrate, so supraoccipital, sq squamosal, V vertebral scute, vo vomer, xi xiphiplastron

Referred Material: SGP 2009/4, SGP 2009/9, SGP 2009/14, SGP 2009/18 (Figs. 3 and 4).

Description

The four specimens herein referred to Annemys sp. represent three partial shells with various degrees of lateral crushing and an isolated, uncrushed skull. SGP 2009/4 was deposited horizontally and is heavily fractured (Fig. 4c). However, although sutures are difficult to discern, all sulci are well preserved and the outline of the shell appears most authentic. SGP 2009/9 preserves both the scutes and sulci well, but the shell is slightly crushed along the long axis of the animal (Fig. 4d, e). Finally, SGP 2009/14 came to rest on its right side and was heavily crushed along the long axis (Fig. 4a, b). Although all three specimens include plastral material, only the plastron of SGP 2009/9 is complete enough to warrant illustration. SGP 2009/18 is a beautifully preserved skull that shows no major damage (Fig. 4f, g). A brief description of the material is provided herein to support our taxonomic assignment. A more comprehensive description will follow elsewhere.

Carapacial bones

The nuchal is a rather large, trapezoidal element that is approximately two times wider than deep. At least seven neurals were present and the small size and diamond shape of neural VII in SGP 2009/9 is typical for an abbreviated neural column of only seven elements, or at least for a gap in the neural column, but this assertion remains speculative given the lack of specimens that sufficiently preserve this area. The neural column consists of regular, coffin-shaped elements with short sides facing anterolaterally except for neural I (facing posteriorly) and neural II (quadrangular). Eight pairs of costals are present. Costal fontanelles are clearly missing in SGP 2009/9, the largest specimen, but appear to be present in SGP 2009/4, the smallest specimen. No specimen possesses a complete peripheral ring, but considering that SGP 2009/4 appears to possess 12 pair of marginal scutes, it is reasonable to speculate that eleven pair of peripherals were present, as in most crown turtles (Joyce 2007). A distinct gutter runs in all specimens along the anterior rim of the carapace. The pygal area is present in SGP 2009/4, but the number and form of elements cannot be discerned.

Carapacial scutes

As in the vast majority of turtles, five vertebral scutes and four pairs of pleural scutes are present. Vertebrals II–IV are as long as wide and hexagonal. By contrast, vertebrals I and V are significantly shorter, but expand significantly anteriorly and posteriorly, respectively. As in most turtles, the vertebral I–II and II–III sulci are placed over neural I and III, respectively, but the vertebral III–IV sulcus is placed over neural VI. The surface of the bone below the vertebrals is sculpted in a radiating pattern that originates symmetrically from the medioposterior end of each vertebral. The vertebrals were thus likely sculpted in life as well. Between SGP 2009/4 and SGP 2009/9, it is apparent that 12 pairs of marginals were present.

Plastral bones

The plastron had a ligamentous connection with the carapace. The epiplastra are broadly rectangular elements that contact another medially, the entoplastron posteromedially, and the hyoplastra posteriorly along a transverse suture. The entoplastron has the shape of an anteroposteriorly stretched oval, is about the same size as an epiplastron, and sits distinctly inset between the hyoplastra. The anterior plastral lobe is significantly broader and shorter than the posterior plastral lobe and has a broad transverse anterior margin that is formed by the epiplastra only. Mesoplastra are absent. The bridge is anteroposteriorly longer than the anterior plastral lobe, but shorter than the posterior plastral lobe. The hyoplastron forms well-developed axillary buttresses that reach anterolaterally along the peripheral series to contact the suture between peripheral II and III. A contact of the axillary buttress with the costals is absent. In SGP 2009/9, the largest of the three specimens, plastral fontanelles are clearly absent. However, in SGP 2009/14, the smallest specimen, a tear-shaped central fontanel is developed, as are semi-lunate lateral fontanelles. The xiphiplastra form the posterior half of the posterior plastral lobe and tapers posteriorly into a blunt apex.

Plastral scutes

A pair of gulars and extragulars is arranged in a transverse row and adorns the anterior plastral margin. The extragular are mediolaterally stretched rectangles that only cover the epiplastra and that only contact the gular medially. The gulars form small rectangles, and their posterior sulcus with the humeral coincides with the anterior entoplastral suture. The humeropectoral sulcus is oriented transversely and located well posterior to the entoplastron. The pectorals and abdominals cover much of the bridge region and laterally contact four pairs of inframarginals. The femoral/anal sulcus is distinctly omega-shaped and crosses the transverse hypoplastral/xiphiplastral suture. The midline sulcus is slightly sinusoidal throughout the plastral series.

Skull

The skull SGP 2009/18 is typical for basal pancryptodires in that it is rather flat, tapers towards the anterior, and has deep upper temporal emarginations. Small nasals are present that contact another along the midline. The prefrontals are small, restricted to the orbits, and do not contact another along the midline. The frontals contribute broadly to the orbits and extend anteriorly to contact the nasals. The postorbitals are long bones that do not contribute to the upper temporal emargination. The triturating surfaces are narrow, and a low lingual ridge is developed in the posterior half. The foramen palatinum posterius is large. The pterygoids reach the posterior margin of the skull, contact one another along the anterior half, possess distinct external processes, but do not contact the basioccipital. The foramen posterius canalis caroticus internus is located at the back of the skull in an embayment surrounded by the pterygoid, basisphenoid, and the basioccipital. The foramen posterius canalis caroticus lateralis (sensu Brinkman et al. 2012) is formed anteriorly by the pterygoid and the basisphenoid. However, a residual interpterygoid vacuity still appears to be present just anterior to the basisphenoid that communicates with this foramen. The basisphenoid has a well-developed, laterally projecting “basipterygoid process” (sensu Evans and Kemp 1975) that is closely associated with the foramen posterius canalis caroticus internus and the foramen posterius canalis caroticus cerebralis. The quadrate forms the majority of the ear capsule in dorsal view. Fine crenulations along the anterior border of the quadrate indicative of a cartilage cap confirm the presence of an otic trochlea. The antrum postoticum is voluminous, the cavum tympani is kidney-shaped, and the incisura columella auris is open.

Discussion

Estimating a body count

The number of turtles deposited at Mesa Chelonia can be conservatively estimated based on field observations. The high concentration zone, tightly packed with turtle shells, has an estimated minimum size of 10 × 2 m, resulting in an area of at least 20 m2. The block of matrix that was prepared from this inner zone revealed that 18 turtles were preserved in an area less than a half square meter, which corresponds to a density of 36 turtles/m2. Calculated for the full extent of the high concentration zone, 720 turtle skeletons are preserved within this zone. It is plausible that the original extent of this layer may have been much larger, since the highest turtle density is now found on the eroded surface of the layer and given that we neither speculate how deep the irregularly shaped layer continues into the hillside nor how much of the inner zone was eroded.

Observations in the field indicate that the marginal zone extends 10 m along the western flank and 30 m along the eastern flank from the center of the inner zone (ESM Fig. 1). Given that both flanks are perpendicular to one another, the outer zone can be inferred to not have been circular. The extent of the marginal zone in the other two directions is unclear, as this part of the layer has already been eroded. To remain conservative, we therefore assume that the marginal zone surrounds the high concentration zone and has the shape of a quarter oval no more than 10 m wide and 30 m long, the observed values from the field (ESM Fig. 1). This results in a surface area of 216 m2 once the high concentration zone has been subtracted. We were able to document on average two hyo- or hypoplastral elements per meter in the 10 cm deep weathering horizon of the outer zone, which corresponds to 20 such elements, or a minimum of five turtles/m2. We therefore mathematically arrive at 1,080 disarticulated turtle skeletons in the outer zone for a total of approximately 1,800 turtles in an area of 236 m2. Although our estimates are crude, it appears safe to conclude that many hundreds, if not thousands of turtles were deposited in this single layer. In terms of complete individuals, this may well be one of the richest terrestrial fossil vertebrate localities worldwide and it therefore appears appropriate to address it as a Konzentratlagerstätte.

Regional climatic patterns

The preservation of many hundred freshwater turtles at the Mesa Chelonia bone bed indicates that the climate in the late Middle Jurassic of the Turpan Basin was wet enough to support large populations of aquatic turtles, but was periodically affected by extreme droughts (see below). This is most consistent with a monsoonal, subtropical climate, as is found today in the Sahel Zone of Africa or much of Australia. Climate change patterns suggest a complex long-term shift from monsoonal to seasonally dry climates throughout northwestern China in the Jurassic (Eberth et al. 2001; Hendrix et al. 1992). By the Oxfordian, seasonally dry climatic conditions were permanently established in the region (Eberth et al. 2001).

Depositional environment

Intermittent tectonics significantly influenced deposition across northwestern China throughout the Jurassic (Eberth et al. 2001). The petrologic immaturity and poor rounding of the sand and silt-sized grains contained in the turtle layer is typical for alluvial systems near newly uplifted orogens. The depositional environment in the Jurassic Turpan Basin can be characterized as a braidplain channel system with episodic flow conditions. Vertisols were deposited in a climate with alternating wet and dry seasons or climate cycles. Small sized, shallow channels, much wider than deep, as well as matrix-supported clasts with a diameter of 1–2 cm, sometimes arranged in arc-shaped bands, indicate bed load transport adjacent to major channels with alternating upper and lower flow regimes.

The turtle bone elements in the layer are very poorly size-sorted, show little evidence of preferred orientations, and some turtle shells exhibit very high dip angles. These features are consistent with a debris flow. The high silt content and matrix-supported mudstone clasts indicate that the flow was cohesive. Lack of stratification and poor sorting of the matrix-supported components within the deposit are characteristic of high-density, water-laden, debris flows (i.e., a subaerial mudflow) (Major 2003; Collinson et al. 2006). The region of high turtle density may represent the depositional front of a moving flow, with possible longitudinal sorting along the flow axis (Eberth et al. 2006; Major 2003). If this interpretation is correct, the presence of isolated skeletal elements along the eastern section may represent the flow tail and a general flowing direction to the southwest.

Taphonomic interpretation

Given the seasonally dry monsoonal climate, drought-induced accumulation and mortality processes appear to be the most plausible taphonomic causes for the Mesa Chelonia site. Under this scenario, a turtle population in a riverine environment with small, shallow channels was forced to retreat to water holes during a drought. Due to the severity of the drought (phase III sensu Shipman 1975), the Mesa Chelonia water hole dried up, leaving large numbers of dead animals concentrated in a small area. Once the rains returned, the water runoff stripped loose sediment from the denuded land surface (Shipman 1975) and rapidly buried the accumulated carcasses.

The sedimentological evidence to support this drought scenario is unfortunately limited. Among geological criteria for drought sediments (Shipman 1975), none are present at Mesa Chelonia. In particular, there is no evidence for mud cracks, evaporites, or caliche in any of the strata associated with the fossils. Given the preponderance of other data in favor of a drought-induced mortality event, however, we explain the lack of mud cracks through surface erosion due to recurring water. Lack of evaporites or caliche is interpreted as evidence for a deposition system that was generally wet and only dried out occasionally and with catastrophic consequences.

The very presence of many hundreds to thousands of turtle carcasses at a single site demands identifying an accumulation mechanism. A number of extant turtles are known to aggregate during the breeding season to lay their eggs (Ernst and Barbour 1989), but these aggregate are not known to coincide with the die off of hundreds to thousands of individuals. There are many reports, however, of extant turtles accumulating due to drought conditions (see ESM for details). It is not plausible that the carcasses accumulated exclusively after death. Turtle carcasses lack unique sedimentological properties that would allow enriching them relative to other large sedimentary particles.

The isolated turtle bones within both fossiliferous zones support the hypothesis that the turtles had accumulated in a water hole during a drought. Many turtles perished during early phases of the drought due to a lack of food resources, and the carcasses started to macerate and disarticulate. Despite of many hundred disarticulated turtles preserved in the both zones, there is no evidence of abrasion, scavenging, sorting, extended transport, or the formation of lags. Hence, the disarticulated remains were not accumulated by a river system over a longer period.

The presence of plastically deformed and/or rounded mud clasts reveals that the turtles were transported before final burial. After heavy rainfall ended the drought, water returned to the basin with great force, triggering a debris flow in the river bed that eroded and transported fine-grained flood plain sediments and large intraclasts. The vertisols flanking the turtle deposit indicate that the debris flow possibly coursed through inactive channel reaches.

There are two possible scenarios for the formation of this parauthochthonous deposit: (1) the debris flow caused the wash off of turtles from the water hole; and (2) the debris flow may have caused the collapse of the marginal wall of a nearby shallow river channel connected to the water hole and resulting in slumping of the cohesive unconsolidated rainwater-saturated mix of turtle remains and mudstone clasts. The cohesive debris flow caused high erosion and partially plastic deformation of the soft sediments in the underlying bed. Mud clasts and bones were picked up and incorporated into the base of the advancing mudflow front, and then were parautochthonously deposited at the Mesa Chelonia site. Vertically embedded turtles are interpreted as a result of turbulent movements. However, the overall density of turtle carcasses was apparently not reduced. This high density indicates very short transportation over the course of several meters, assumed that the hypothetical water hole did not give refuge to a considerably higher number of turtles (tens of thousands). The debris flow contributed to the further disintegration of carcasses, but fresher carcasses and possibly alive specimens were embedded fully articulated. Soon after the initial, unsorted debris flow wave had passed, finer-grained sediments buried the lithoclasts and bones, protecting them from destructive surface processes.

Taphonomic comparisons

The monospecific or paucispecific turtle deposit (the current taxonomic assignment is Annemys sp., see ESM for details) is highly unusual compared to other bone beds caused by debris flows (Rogers 2005; Eberth et al. 2006; Fastovsky et al. 1995; Weigelt 1989). The limited transport ability of the debris flow may be partly responsible for the sole occurrence of turtle remains (paleoecological reasons are discussed below). For example, water energy may have been too low for transport of large dinosaur bones.

Mass deposits of nonmarine fossil turtles have been reported from the Nemegt Formation in Mongolia at Bambu Khudag, Tsagaan Khushuu, and Bugiin Tsav (Khand et al. 2000), but the extent of these deposits and their taphonomic circumstances remain unclear. We are otherwise only aware of two described nonmarine turtle mass burial sites with more than 50 individuals, the Turtle Graveyard site in North Dakota, USA (Lyson and Joyce 2009a) and the Black Mountain Turtle Layer in Wyoming, USA (Brand et al. 2000).

The Turtle Graveyard site has yielded approximately 100 turtle shells and 50 turtle skulls in an area less than one hectare in sediments of the Late Cretaceous Hell Creek Formation (Lyson and Joyce 2009a). The diverse assemblage consists of at least three species of baenid turtles and two species of trionychid turtles, turtles generally associated with fluvial sediments (Lyson and Joyce 2009a, b; Joyce and Lyson 2011). There are signs of decay, as all skulls, shells, and various postcranial long bones are found in isolation. All specimens are embedded in a single, thick sandstone layer that contains large break-up clasts and logs at the base and that fines upwards. The material clearly lacks evidence of significant transport, although numerous shells are upturned or stacked against logs. The number of specimens preserved at this site is far smaller than at Mesa Chelonia, but the likely mechanism that led to the accumulation of turtles is also thought to be an extreme drought in a monsoonal, subtropical climate (Lyson and Joyce 2009a). In a fluvial sedimentary system, numerous riverine turtles appear to have perished in a drying oxbow lake and came to their final resting site when the channel was reactivated with the returning of the rain (Lyson and Joyce 2009a).

The Black Mountain turtle layer, which is an expansive sedimentary layer within the Eocene Bridger Formation of the Green River Basin, may easily exceed Mesa Chelonia in the number of turtles that it contains. Surficial collecting has yielded hundreds of turtles over a relatively short distance of a few kilometers and many thousands of turtles were likely embedded in the layer. The vast majority of material consists of articulated shells. All other remains are found in isolation (Brand et al. 2000). The taphonomic settings of the Black Mountain turtle layer contrast those at Mesa Chelonia in important ways. The Black Mountain turtle layer is thought to represent the bottom of an extensive, shallow lake that permanently filled the Green River basin during parts of the Eocene. The permanent availability of water and lush tropical vegetation allowed for the buildup of a dense turtle population in the lake. However, although the climatological conditions were stable, a great number of turtles perished abruptly, when the basin was inundated with a thick layer of volcanic ash. Dead turtles floated in the lake until decay destroyed the body cavities allowing the articulated shells to sink to the bottom (Brand et al. 2000). The great number of turtles therefore is not the result of an accumulating mechanism, such as the contraction of available habitat, but appears to faithfully reflect a high population density as seen in many modern turtle communities (see ESM Table 1). Given the size of the Green River Lake and the great number of turtles documented by Brand et al. (2000), it appears likely that more turtles perished during the Black Mountain event. However, the aforementioned lack of an appropriate concentration mechanism resulted in the rather even distribution of these turtles over a wide region, instead of the massive accumulation in a small area.

Paleoecology

The extremely high density of fossil turtles at Mesa Chelonia and the absence of other fossils with the exception of some small, isolated ganoid fish scales are puzzling. Neither any plant remains nor other vertebrate groups appear to be present in this layer. This may be due to the consumption of other vertebrates (i.e., fishes) as a food-resource by the turtles, the escape of more amphibious vertebrates such as crocodilians, or—less likely—may indicate that the habitat was too ephemeral to support a more diverse aquatic fauna. Comparable mass accumulations of extant turtles are discussed in the ESM.

Conclusions

Far more than 1,000 individuals tentatively referable to the “xinjiangchelyid” turtle taxon Annemys sp. are calculated to have been deposited at the Mesa Chelonia site in the Turpan Basin of Xinjiang Autonomous Province, China. The sedimentology and taphonomy of Mesa Chelonia is consistent with former reconstructions of the depositional settings of the Qigu Formation (Shao et al. 1999; Eberth et al. 2001), by predicting a flat, intermontane basin setting in a warm, monsoonal climate characterized by severe episodic droughts. The bone bed was formed during a high-energy flow event that trapped and moved a poorly sorted mixture of mud, silt, and clasts, in addition to isolated turtle skeletal elements and complete turtle skeletons. The monospecificity or paucispecificity of this turtle deposit is very unusual and may partly reflect special paleoecological conditions, similar to drought-induced mass accumulations of turtles in Australia today. Mesa Chelonia is one of the largest fossil turtle accumulations worldwide, a rare terrestrial Konzentratlagerstätte. It also offers new insights into bone beds formed by debris flows, which have been rarely reported (Eberth et al. 2006).