The GP is characterized by an overall flat-top morphology with a surface area of 310 km2. This platform morphology extends over a distance of 45 km from northeast to southwest and over 12 km from northwest to southeast, and is characterized by an elongated shape following the SW–NE orientation of the XU margin ( Figures 3A, B ). The bathymetry of this platform top ranges from -800 to -536 m, whereas it rapidly decreases to -1,100 and -1375 m at the platform slope toe ( Figures 3C—E ). The platform top features several circular depressions of unknown origin. The overall morphology shows a pattern of stepwise shallowing from southwest to northeast. The southeastern platform margin is substantially more rugose than the northwestern edge, and has a higher declivity ( Figures 3C—E ). The flat-topped submarine morphology of the GP exhibits sharp and abrupt margins that are commonly incised by well-developed, 2.1- to 8.7-km wide, steep convex-bankward embayments. The embayments are more pronounced on the southeastern flank of the GP most likely due to the platform margin collapses and slope failures affecting that area and resulting in a typical “scalloped” geometry of the bank margin. Rugged morphologies, irregular reliefs, and numerous circular structures are evident at the slope toe of the platform ( Figure 3E ). The northwestern flank of the platform exhibits two drowned terraces at depths of approximately 1,315 and 1,217 m ( Figure 3E ).

Coupled analyses of high-resolution bathymetric grids ( Figure 3 ) and underwater videos ( Figure 4 ) were used to determine the nature of the geological features. The platform top is characterized by flat, porous, and dull gray rocky slabs that are generally overlain by superficial sediments ( Figures 4A, B ), exhibiting common carbonate rock characteristics. Underwater images of the rugged morphologies and irregular reliefs revealed outcrops of bright rocks associated with vitreous texture and rounded rocky formations resembling pillow lavas, and dense polygonal fracturing networks that are similar to the tensional/contraction cracks that typically form during submarine volcanic eruptions ( Figures 4C, D ).

Two carbonate rock samples (C1, C2) were retrieved from two carbonate terraces of the platform top ( Figures 4A, B ; Table 1 ). The samples are approximately 2–3 cm thick ( Figures 4E, F ). Both samples have a dull grey color, owing to manganese oxide incrustations. The rocks are perforated by borings, which enlarged the cavities and fractures in the carbonate. Both carbonate samples comprise packstones marked by various shallow-dwelling carbonate producers, such as LBF and coralline algae ( Figures 5A–D ). Among the LBF, Nephrolepidina spp., Amphistegina spp., Miogypsina sp., Planorbulinella larvata, Textularia sp., Operculina rectilata, Cycloclypeus sp. are evident. The genus Amphistegina occupies the largest volume, followed by the genus Nephrolepidina ( Figure 6 ). Among the identified coralline algae, Corallinales are the most abundant ( Figures 5A, C ). Hermatypic corals are generally rare in both samples. A facies of micritic packstone-bearing planktonic foraminifera overlies the Mn crust in sample C1, thereby recording the end of shallow-water carbonate production and the drowning of the carbonate platform below the euphotic zone ( Figure 5B ). The planktonic foraminiferal assemblages include Globigerina spp. and Globigerinoides spp.

A poorly lithified sample (C3) from the platform slope corresponds to skeletal floatstone-bearing coralline algae, planktonic foraminifera, and halimeda algae ( Figures 5E, F ). The planktonic foraminiferal assemblages include Globorotalia spp. and Pulleniatina spp. ( Figure 5E ). Among the identified coralline algae, Hapalidiales are relatively abundant ( Figure 5F ). Volcanic fragments are not evident in sample C3, suggesting the absence of exposed volcanic landforms on the southwestern flank of the platform at the time of deposition.

The stratigraphic evolution of the GP can be divided into three distinct stages: (1) platform initiation and structurally controlled retrogradation at the tilting margin ( Figure 12 , AU1-AU3; Figure 13A ), (2) platform aggradation and cessation of faulting ( Figure 12 , AU4, Figure 13B ), and (3) platform progradation and backstepping ( Figure 12 , PU1-PU3, Figures 13C, F, E ). Finally, the platform drowned ( Figure 13F ). The strata deposited during stage 1 feature an evident northwestern margin retrogradation at the southwestern portion of the platform, suggesting that faulting may have remained active during this phase of deposition. The deceleration in accommodation production caused by the decreased fault-induced subsidence resulted in the occurrence of aggradational geometries in most of the platform margin during stage 2 ( Figures 12 , 13B ). Reef mounds were commonly pronounced at the platform margin, indicating that frequent photozoan carbonate factory with greater growth potential likely dominated the platform during stages 1 and 2. The corresponding time periods were characterized by high sedimentation rates recorded by both wells XK-1 and CK-2 ( Figure 12 ) (Yi et al., 2018; Fan et al., 2019). The photozoan carbonate factory reflected by the platform geometry is consistent with the frequent occurrence of photozoan carbonates from the Early Miocene (stage 1) to early Middle Miocene (stage 2) observed in well XK-1 (Wu F. et al., 2019). Most ICPs throughout the Indian and Pacific Oceans ( Table 3 ), such as the Maldives, Yadana Platform, and Marion Platform, were dominated by photozoan carbonate factories with great growth potential during these two periods, and platform drowning generally occurred in active tectonic settings, such as the Yadana Platform ( Table 2 ) (Paumard et al., 2017).

Stage 3 documents the synchrony of platform backstepping and progradation, which contradicts the sea-level process of stratigraphic stacking patterns. Previous studies have also documented the synchrony of drowning and ongoing bank-margin growth, which is contradictory to a sea-level explanation (Aubert and Droxler, 1996; Betzler et al., 2013). It is proposed that the cessation of normal faulting may have instigated platform progradation during stage 3, and that these backsteps may have been driven by changing environmental conditions. Coincident with the progressive reduction in stratal thickness from PU1 to PU3, the sedimentation rates recorded in wells XK-1 and CK-2 steadily declined ( Figure 12 ). According to the combined analysis of carbonate samples and platform geometry, the reduction in carbonate production was caused by carbonate factory turnover from a photozoan to a heterozoan type. The platform shifted from aggradation to progradation, and a change in faunal association occurred at 15 Ma on the Middle Miocene Maldives Platform (Reolid et al., 2019; Reolid et al., 2020). Similar examples can also be found in other carbonate platforms, where platform drowning was accompanied by changes in the carbonate factory toward a heterozoan type ( Table 3 ). Furthermore, the platform backstepped more significantly on the southwestern side than on the northeastern side ( Figure 13E ). Therefore, platform growth has been asymmetric since the onset of platform drowning when the partial platform drowned. Most areas of the platform margins persisted, and growth continued through progradation and slight aggradation. Similarly, the asymmetric growth of the Maldives carbonate platform also occurred during the partial bank demise with the synchroneity of continued bank-margin growth (Betzler et al., 2013).

The GP carbonate platform nucleated on a structural high and formed a spindle-shaped drowned bank in the last step of platform drowning ( Figure 13F ). According to Zampetti et al. (2004), a buried, drowned Miocene carbonate bank offshore Malaysia has a similar geometry and transitioned from flat-topped during the active growth stage to mound-shaped during the predrowning stage. Betzler et al. (2009) also identified a spindle-shaped drowned bank that formed during the last drowning step of the partial drowning of the Maldives carbonate platform. This was viewed as the final phase of the subphotic mound development and submergence below the wave base. During the final drowning process of the GP, a similar evolution is thought to have generated a spindle-shaped drowned bank. Seismic stratigraphy and biostratigraphy suggest that the platform was submerged during the late Middle Miocene. The drowning surface of the GP was identified on the carbonate slope ( Figures 1 , 7 , 8 ), corresponding to sequence boundary T40 (top of the Middle Miocene), which was recognized in our previous study on the continental shelf and QDNB (Wang et al., 2013; Wu S. et al., 2014). The foraminiferal assemblage of the two carbonate samples from the platform top was dominated by larger benthic foraminifers, such as Nephrolepidina, Amphistegina and Miogypsina ( Figure 6 ), corresponding to the Middle Miocene foraminiferal assemblage also recorded in wells XK-1 and CK-2 (Ma et al., 2018; Meng et al., 2022). The last occurrence of Nephrolepidina and Miogypsina at 577.04 m depth in well XK-1 represents the top of the Middle Miocene ( Figure 12 ) (Ma et al., 2018). Boudagher-Fadel and Banner (1999) defined the top of the Middle Miocene using the last occurrence of Nephrolepidina. Furthermore, the water depth at the highest point of the submerged platform was 536 m, falling within the depth interval of the Middle Miocene tops (521–577 m) indicated by the wells ( Figure S3 ). The seismic profile crossing the highest basement of the GCP also shows that the depth of the Middle Miocene top was 562 m (Fyhn et al., 2013), implying that the last drowning phase of the GP occurred during the late Middle Miocene.

Inorganic nutrient availability, siliciclastic supply, sea-surface temperature, and relative sea level are the major factors controlling carbonate factory turnover, and platform growth and demise (Mutti and Hallock, 2003; Halfar et al., 2006; Fyhn et al., 2009; Betzler and Eberli, 2019; Wu et al., 2019). The aggradational and retrogradational platform margins reflect an increase in accommodation in stages 1 and 2. In such a scenario, relative sea-level changes appear to be the most probable major controlling factor of platform development. The backstepping of the platform margins toward topographically high areas was likely controlled by fault-induced subsidence during stage 1 ( Figure 7 ). The inclined fault block, which remained topographically high, may have been controlled by younger faults that continued to move. The rising global sea level is also thought to have played a dominant role in shaping the platform geometries. Both global and local relative sea level curves indicate a rapid sea-level rise during the Early Miocene (Haq et al., 1987; Shao et al., 2017; Miller et al., 2020), and seismic data also indicate that the carbonate platforms were typically restricted to topographic highs on the XU due to the sea level rise during the Early Miocene (Wu S. et al., 2014). High sedimentation rates were recorded in both wells XK-1 and CK-2, demonstrating that carbonate production was likely high enough to be impacted from the high-frequency relative sea-level rises during the Early Miocene and early Middle Miocene ( Figure 12 ). We suggest that optimal conditions for carbonate production and accumulation existed during these time intervals, compared with the growth rates of other carbonate platforms (Schlager, 2000; McNeill, 2005).

Platform progradation and backstepping were synchronous as a consequence of faulting cessation and summer monsoon-driven currents during stage 3. On the one hand, the faulting cessation may have triggered platform progradation. It has been proposed that the apparent chronological synchronicity between the faulting cessation and progression is partly due to a slowdown in accommodation production caused by a reduction in fault-induced subsidence. Previous studies have also documented that the faulting cessation is coincident with the initiation of the expansion and coalescence of the Segitiga Platform caused by progradation, thereby demonstrating that the deceleration of accommodation production driven by decreased fault-induced subsidence contributes to platform progradation (Bachtel et al., 2004). The shallow-water components were also likely transported off the platform by summer wind-driven currents, resulting in platform progradation during stage 3. According to Clift et al. (2014), the EASM intensified over long periods from 21.2 to 17.3 Ma and from 15.6 to 10.5 Ma, with some oscillations occurring in each era. The heaviest summer monsoon formed between 15.6 and 10.5 Ma (Clift et al., 2014) and dominated the entire stage 3. On the other hand, the previous fault-induced subsidence and summer monsoon-driven currents, are also believed to have been responsible for the pronounced backstepping distance and the area on the southwestern side of the platform ( Figures 13C–E ). Heterozoan carbonates were found on the terraces caused by the gradual backstepping of the platform ( Figure 5 ), indicating that favorable platform growth conditions deteriorated during stage 3. The stepwise drowning of the carbonate platform, coupled with a steady decrease in stratal thickness from PU1 to PU3 and carbonate factory turnover from a photozoan to a heterozoan type ( Figure 12 ), reflects the long-term worsening of platform growth conditions. According to data from well XK-1 analyzed by Wu et al. (2019), upwelling from the east coast of Vietnam caused by the EASM increased the nutrient influx, inhibited coral growth, suppressed the development of reef-related facies, contributed to the development of open bank facies and thus promoted carbonate factory turnover from a photozoan to a heterozoan type during the Middle Miocene in the Xuande Platform ( Figure 12 ). Strong summer monsoon-related upwelling is also an inhibiting factor of the continued growth of the Phan Rang Carbonate Platform near the east coast of Vietnam (Fyhn et al., 2009). Betzler and Eberli (2019) compiled a list of modern and ancient examples of ICPs and carbonate contourite drifts documenting monsoon-triggered currents as a common element of carbonate platform drowning during the late Middle Miocene. The GP is closer to the east coast of Vietnam relative to the Xuande Platform ( Figure 1 ); thus, the monsoon-induced upwelling zone could also reach the GP during the Middle Miocene. As a result, we believe that summer monsoon-triggered currents suppressed coral reef formation by generating topographical upwelling on the southwestern side of the platform, leading to the gradual occurrence of a heterozoan carbonate factory with low growth potential on the carbonate terraces. These carbonate terraces on the southwestern side of the platform usually do not get affected by the relative sea-level rise resulting from the fault-induced subsidence, thereby leading to stepwise drowning of the GP. In addition, because of an abrupt increase in the subsidence rate detected in the surrounding QDN Basin (Wu S. et al., 2014; Zhao et al., 2018) , the sustained coral reef ecosystem was probably completely destroyed by the subsequent enormous subsidence rate during the Late Miocene. The enormous Huaguang MTDs expanded synchronously into the examined area, indicating an unstable basement throughout this time period (Wang et al., 2013). Therefore, the shallow-water carbonate builders on the GP were unable to persist and became as visible as those on the Xuande and Yongle Platforms.

Although the oxygen isotope records from the Deep Sea Drilling Project and Ocean Drilling Project cores show warmer temperatures during the Early Miocene and a general decrease in temperature from the Middle Miocene (mid-Miocene climatic optimum) to present ( Figure 2 ) (Zachos et al., 2001; Zhao et al., 2001), the significant abundance of photozoan carbonates during the Quaternary suggests that the temperature in the Xisha region has been sufficiently high since the Early Miocene for the production of photozoan carbonates (Wu et al., 2019). Therefore, the development of heterozoan carbonate factories and the subsequent platform flooding during the late Middle Miocene were not considered to be caused by a cooling event. Furthermore, the scarcity or absence of planktonic foraminifera indicated that heterozoan carbonates were deposited in shallower waters ( Figure 5 ), implying that the occurrence of heterozoan-dominated carbonates during platform drowning was not caused by an increase in water depth during the late Middle Miocene. The drop in temperature and increase in water depth have also been recognized as less important drivers for the emergence of a heterozoan carbonate factory, as well as photozoan to heterozoan factory turnovers (Wu et al., 2019). Siliciclastic sediment was not present in the carbonate samples obtained from the GP. Seismic profiles show that the GP was far from siliciclastic sediment sources and has been separated from them by deep-water troughs since the Early Miocene ( Figures 1 ; S1 ). Insufficient siliciclastic input causes the GP to crop out at a depth of hundreds of meters, implying that terrigenous input is unlikely to be the primary cause of differences in the platform backstepping pattern.

Previous studies have documented that magmatic and hydrothermal activities were intense at 5.5 and 2.6 Ma in the flanks of modern carbonate platforms in the northern SCS, resulting in upturned reflectors extending upward to the contemporary sea bottom (Xia et al., 2018; Gao et al., 2019a; Gao et al., 2019b; Sun et al., 2020; Zhao et al., 2021; Gao et al., 2022; Liu G .et al., 2023). Submarine volcanoes, magmatic intrusions, and related upward strata extending to the modern seafloor have also been identified by submersible observations and seismic profiles on the flank of the GP ( Figures 4 , 11 ), suggesting that contemporaneous volcanic activity also occurred there. Interestingly, fluid flow structures, such as pipes and craters, developed in the platform interior, and these pipes penetrated the carbonate platform from the basement ( Figure 8 ). The pipe structure is regarded as a conduit for hydrothermal fluids on the Zhongsha carbonate platform (Huang et al., 2020). Craters on the platform top were created by stratal collapses triggered by hydrothermal fluid erosion. Similar craters have been discovered on the top of the Guangle carbonate platform (Fyhn et al., 2013). The formation and extinction of shallow-water carbonate platforms are assumed to be predominantly governed by tectonic and volcanic activity at volcanic margins; the drowning of ICPs in the Mozambique Channel is such an example (Courgeon et al., 2016). However, carbonate samples from the platform top and flank contain few volcanic fragments, implying that volcanic activity has a limited influence on the diagenetic processes of carbonate rocks and platform evolution in the GP. Volcanic morphologies, extensive faults, and fracturing networks that exhibit extensional deformation patterns are missing from the GP platform top, implying that platform drowning is unlikely to be caused by extensional deformation.

Previous research has indicated that the heat supplied by magmatism in the basins may induce early hydrocarbon maturation (Gao et al., 2019a). This viewpoint is supported by extensive magmatic intrusion and sills identified beneath the strata of the “Haima” cold seeps in the QDNB ( Figure S1 ). The carbonate sediment produced by the GP was most likely transported into the nearby “Haima” region during the Miocene, improving the reservoir quality of the Miocene strata. Therefore, studying the stratigraphic evolution of the GP could provide new insights into hydrocarbon exploration in the surrounding basins.