In total, 13 short rock cores (5 cm in diameter) with length < 30 cm were drilled from the SYBH during May and June 2017 using a miniature underwater drilling rig carried by the FCV2000D remotely operated vehicle (ROV; Yu et al., 2022). Three-dimensional morphology was reconstructed using an unmanned sea-robot ship equipped with a multibeam system above 15 m and a professional-grade ROV equipped with a multibeam system below 15 m (Li et al., 2018). In addition to morphology, a high-resolution underwater camera was carried by the ROV to gather more detailed knowledge. This study focuses on a short core (length: 26 cm; latitude: 16.52°; longitude: 111.77°) consisting of speleothems and substrates from the depth of 116-m (Bh116; Figure 3 ). In Bh116, the speleothems broke into small pieces during the drilling processes due to their weak cements. Therefore, the relative positions for these broken pieces could not be accurately reconstructed though we carefully protected them during the collection, transportation, and rinsing.

We rinsed the subsamples one by one to remove covered slurries contaminated during the drilling process and put them back to their original positions. Substrates and the largest speleothem subsample ( Figure 3A ) were sliced for petrographic investigation under polarization microscopy. Unfortunately, the largest speleothem subsample of Bh116 was broken and lost after thin section grinding. In order to remove possible contaminants, we rinsed the subsamples by ultrasonic cleaning in deionized water for 12 hours and repeated three times before subsequent geochemical analyses. Generally, no debris was observed after the third rinsing. For the high-resolution surface structure analysis of calcite, samples of freshly broken pieces were studied with Hitachi-SU8010 (Japan) at Guangzhou Institute of Geochemistry, Chinese Academy of Sciences (Guangzhou, China). Calcium, carbon, oxygen, and magnesium concentrations of microporous calcite were determined by SAPPHIRE EDAX detector and Apollo X-SDD detector, respectively.

The carbonate mineralogy was determined by X-ray diffraction measurements. Two subsamples (~2 g) were milled from cleaned carbonates ( Figure 3A ) using a stage-mounted dental drill fitted with silicon carbide dental blade and measured using a Rigaku D/max-rB Theta-theta X-ray powder diffractometer at the First Institute of Oceanology, Ministry of Natural Resources. Copper K_α radiation (40 kV, 100 mA) was used as an X-ray source. Samples were packed into a rectangle cavity in an aluminum holder and scanned in a step-scan mode (0.02°/step) over the angular range of 3° to 70° (2θ). The mineralogical compositions of samples were constructed using the 104 reflection of calcite with 2θ from 29.25° to 29.80°, the 111 reflection of aragonite with 2θ between 25.5° and 26.5°, and the 104 reflection of dolomite with 2θ from 30.58° to 31.28°. Usually, the calcite 104 reflection branches into two asymmetric peaks, of which both high-magnesium calcite and low-magnesium calcite co-exist (Kontoyannis and Vagenas, 2000; Csoma et al., 2006; Zhai et al., 2015).

For elemental analyses, four samples (~50 mg) were milled after having been repeatedly rinsed with carbonates using a stage-mounted dental drill fitted with tungsten carbide dental burs ( Figure 3 ). Major elements, including Ca and Mg, were analyzed using an inductively coupled plasma optical emission spectrometer (ICP-OES), and minor elements were performed by inductively coupled plasma mass spectrometry (ICP-MS) at the First Institute of Oceanology, Ministry of Natural Resources. First, 50.00 mg carbonate powder was dissolved in 1.50 ml high-purity HNO₃ and 1.50 ml high-purity HF in a digestion bottle with Teflon liner at 190°C for 48 h. Solutions in the Teflon liners were evaporated to dryness on a heating plate. Next, the samples were added with 3 ml of 7 N HNO₃ and 0.5 ml Rh internal standard solution and heated at 150°C for 8 h. Finally, the samples were diluted to 50.00 g to be measured by ICP-OES (Icap6300, Thermo Fisher, USA). Exactly 10.00 g solutions were extracted from the 50.00 g solutions and diluted to 20.00 g to analyze by ICP-MS (X Series II, Thermo Fisher, USA). These four carbonate samples were measured together with 34 further carbonate samples from the SYBH. Up to three standard solutions and a laboratory blank were used to calculate the calibration curves for each element.

Stable carbon and oxygen isotopes were analyzed to discriminate the origins of deep submerged speleothems between seawater and air-filled conditions. Approximately ~10 mg carbonate powder was collected from 11 positions ( Figure 3 ) and measured on a Thermo Scientific MAT-253 Plus mass spectrometer coupled with an online carbonate preparation device (Kiel-IV) in the Isotope Laboratory of the Xi’an Jiaotong University. Carbonates were reacted with anhydrous phosphoric acid at 75°C. Results are reported relative to the Vienna Pee Dee Belemnite standard. The analytical errors (1 σ) are ±0.06‰ and ±0.03‰ for δ¹⁸O and δ¹³C (Zhao et al., 2021), respectively.

As relative positions for speleothems could not be accurately determined, we chose three subsamples with substrates to determine the possible onset and termination time of speleothems ( Figure 3 ). Approximately 100 mg of carbonate powder was collected from the speleothems ( Figures 3B, C ) using a stage-mounted dental drill fitted with tungsten carbide dental bur. The U and Th isotopes of these samples were determined by a plasma-sourced multi-collector mass spectrometer (MC-ICP-MS; Neptune-plus, Thermo-Finnigan) at the Isotope Laboratory of the Xi’an Jiaotong University. Chemistry procedures to separate U and Th followed Edwards et al. (1987). ²²⁹Th–²³³U–²³⁶U isotope dilution method was applied to correct for instrumental fractionation and to determine Th/U isotopic ratios and concentrations. U and Th isotopes were measured on a Mas-Com multiplier behind the retarding potential quadrupole in a peak-jumping mode. The instrumentation, standardization, and ²³⁰Th–²³⁴U half-lives were reported in Cheng et al. (2000); Cheng et al. (2013). Uncertainties in U/Th isotopic data, including corrections for blanks, multiplier dark noise, abundance sensitivity, and contents of the same nuclides in spike solution, were calculated offline at 2σ level (Cheng et al., 2013).

In addition to U–Th ages, one radiocarbon dating ( Figure 3A ) was determined by accelerator mass spectrometry at Beta Analytic, Inc. (Miami, USA) to provide additional geochronological identification. This radiocarbon date ( Table 1 ) was subsequently calibrated to the calendar age using the recently published IntCal20 curve (Reimer et al., 2020).

The morphologic features of SYBH, interpreted from 3-D topography data ( Figure 2B ) and underwater photographs ( Figure 8 ), provide meso-to-micro scale surface evidence to trace the deep submerged speleothems. The 3-D topography reveals a downward ceiling from -90 to -160 m along seaside walls ( Figure 2 ). Irregular and popcorn-like cave coatings extensively cover the ceiling between -90 and -120 m ( Figure 8 ). These coating-like speleothems presented in SYBH are small with centimeter dimensions. They prefer to cluster as a cauliflower-like shapes with downward growth direction from their substrates ( Figures 8B, C ). Most coatings, if not all, display a gravity-fed downward direction ( Figure 8 ), which probably suggests that these speleothems were formed through slow water dripping in air-filled conditions (Bian et al., 2019).

Petrographic investigations show the textures and fabrics of Bh116 including substrates and speleothems ( Figures 3 , 4 ). The inside substrates and outside speleothems can be identified by the naked eye depending on their distinct structures ( Figure 3A ). The obtained substrates (Bh116s1) are composed of gray and gray-white broken limestone characterized by loose, yellow, fine sediments. The speleothems (Bh116s2), however, consist of popcorn-like precipitates characterized by column fibers ( Figures 3B, C ). A thin section of the substrates shows that Bh116s1 consists of lagoonal sand with biologic debris consisting of coral fragments, foraminifers, echinoderms, and brachiopods ( Figures 4A, B ), reflecting a lagoon slope depositional environment, which corresponds to that found by Fan et al. (2019). While the speleothems are characterized by more complex textures ( Figures 4C–I ), they contain two recognizable fabrics: laminated microcrystalline fabric ( Figures 4B–D ) and microcrystalline and sparite crystals alternating with fabric microporous calcite ( Figures 4E–H ). Both of these two structures are highly porous, some of which have been cemented by sparite crystal precipitates ( Figure 4C ). Moreover, the laminated microcrystalline fabric calcites are surrounded by sparite crystal precipitates with dark terminations ( Figures 4C, D ) (Csoma et al., 2006; Bontognali et al., 2016; De Waele et al., 2018). In addition, brownish fine sediments and broken marine shells are observed at the margin of pores ( Figure 4H ). The SEM images show secondary calcite crystals ( Figures 5A–C ), dissolution points, and edges of calcites ( Figures 5D–F ). The secondary calcite crystals show an inner-core growth direction ( Figures 5B, C ), probably indicating the cements re-precipitated from the surrounding dissolved calcites (Tuccimei et al., 2010; Bontognali et al., 2016; De Waele et al., 2018). The geochemical compositions of surrounding calcites and cements are relatively homogeneous, in which MgCO₃ varied between 5% and 20% ( Figures 5G–R ). The presence of Mg in the deep submerged speleothems is also confirmed by the existence of high-magnesium calcite from bulk sample analysis ( Figure 6 ). Based on XRD, the substrates of Bh116 consist entirely of low-magnesium calcite ( Figure 6A ), while ~85% low-magnesium calcite and ~15% high-magnesium calcite constitute the speleothems in Bh116 ( Figure 6B ).

Both carbon and oxygen isotope values of speleothems and substrates are negative in the Bh116 ( Figure 7 ). The speleothems exhibit isotopically depleted values with respect to the substrates in both ¹⁸O and ¹³C. In addition, the oxygen and carbon isotope ratios of the speleothems from different subsamples are variable, and the outer one (Bh116s2c2) is more depleted in ¹³C and slightly enriched in ¹⁸O with respect to those of the internal subsample (Bh116s2c1).

The ²³²Th concentrations of substrates in two subsamples are 499.09 and 225.87 ppt ( Table 1 ). In contrast, the ²³²Th concentrations of speleothems are 0.18 and 4.55 ppt, which are much lower than the substrates. The ²³⁸U concentrations of speleothems are variable with concentrations of 6570.5 and 675.9 ppb. The U–Th ages of substrates are 188.21 ± 5.49 and 183.74 ± 2.78 ka BP, suggesting that they may have experienced deposition and secondary alterations and are not suitable for U–Th dating. The substrates are expected to have been deposited at about 1.4 Ma based on the depth of nearby borehole CK-2 (Fan et al., 2019). The U–Th ages of the speleothems, Bh116s2c1 and Bh116s2c2, are 29.16 ± 0.17 and 26.04 ± 0.18 ka BP, respectively ( Figure 3 and Table 1 ). The radiocarbon age of a subsample of speleothems (Bh116s2c3) is 18.64 ± 0.12 ka BP.

It is evident that deep submerged calcite coatings in SYBH are speleothems that precipitated under air-filled conditions during sea level lowstands. Substances of marine origin, including marine overgrowth on speleothems and marine substrates, are excluded for their negative carbon and oxygen isotopes with values as low as -6.7‰ to -5.3‰ and -5.0‰ to -2.8‰ ( Figure 7 ), respectively. These isotopic features differ from the marine overgrowth on speleothem, marine corals, and substrates but are close to subaerial speleothems collected from other coastal caves (Surić et al., 2005; Tuccimei et al., 2010; Gischler et al., 2017). This suggests that the deep submerged calcite coatings developed in SYBH grew under aerial conditions and can be used to determine maximum sea levels (Richards et al., 1994). The calcareous coatings resemble mushroom-shaped speleothems and phreatic overgrowths on speleothems and are porous (Tuccimei et al., 2010; Bontognali et al., 2016; De Waele et al., 2018), probably suggesting that they exhibit the same aerial origin.

As our deep submerged speleothems initially grew at ~29 ka under the air-filled vadose environment through calcite precipitation in coastal karst cave (Fairchild et al., 2006; Tuccimei et al., 2010; De Waele et al., 2018), the maximum sea level at ~29 ka BP should be lower than -116 m relative to the present. Our results agree well with previous data ( Figure 9A ) from Barbados (Bard et al., 1990; Peltier and Fairbanks, 2006) and Bonaparte (Yokoyama et al., 2000), as well as the reconstructed global ice volume equivalent ( Figure 9B ) and mean sea levels ( Figure 9C ) (Lambeck et al., 2014; Yokoyama et al., 2018). Lambeck et al. (2014) documented a ~40-m sea level falling around ~31–29 ka BP according to compiled data. Our results confirm that this rapid fall in sea level reached as low as -116 m at ~29 ka BP, and LGM was initiated at ~29 ka BP. It is noted that the sea level data from Bonaparte, indicating the onset of LGM (Yokoyama et al., 2000), have large uncertainties of up to 7.8 m. Our data also refined large sea level errors.

The LGM was thought to have been terminated by a rise in sea level (Hanebuth et al., 2000; Yokoyama et al., 2000). It began at ~19 ka BP, and the sea level rose 10–15 m within 500 years based on the interpretation of mangrove muds from the Sunda Shelf and Bonaparte Gulf in which tectonics are stable ( Figure 9A ). Recently, however, both Lambeck et al. (2014) and Yokoyama et al. (2018) demonstrated that this 10–15-m sea level rise was slower and lasted for ~3,000 years from 21 to 18 ka BP ( Figures 9B, C ). The possible termination of the speleothems from the SYBH with a radiocarbon age of 18.64 ± 0.12 ka BP suggests that the sea level was lower than -116 m at ~18.5 ka BP. Our results agree well with those sea levels reconstructed from Sunda Shelf and Bonaparte Gulf within errors and suggest that the termination of LGM is slightly later than the previous estimation ( Figure 9A ). However, our obtained results are inconsistent with the data reported from Barbados and the Great Barrier Reef, which experienced a rapid tectonic uplift of up to 0.34 mm/year (Fairbanks, 1989). The divergence in the conclusions regarding the termination of the LGM based on by sea level data from the SYBH, Bonaparte Gulf (Yokoyama et al., 2000; De Deckker and Yokoyama, 2009), and Sunda Shelf (Hanebuth et al., 2000; Hanebuth et al., 2009), as well as those determined from Barbados (Bard et al., 1990; Peltier and Fairbanks, 2006) and the Great Barrier Reef (Yokoyama et al., 2018), could be due to diverse and aged carbon, local carbon reservoir effects, and contaminated cement (Nakada et al., 2016), as they were determined from different materials (Hanebuth et al., 2000; Yokoyama et al., 2000). In addition, we confirm that the presence of old carbon was not significant in our samples and would not affect our results. Conversely, subsequent cement observed in thin sections from our obtained submerged speleothems may cause the measured radiocarbon age to be less than its actual value. In addition, our results coincide with the constructed ice volume equivalent and global mean sea level, suggesting that the deep submerged speleothems may be a promising way to determine Pleistocene low sea levels.