Impact statements
This study provides the first radiocarbon-dated sea-notch-based chronology for the Song Hong (Red River) Delta, Vietnam. The results clarify the poorly resolved Late Pleistocene Vinh Phuc coastal inundation (transgression) and provide a robust Mid-Late Holocene sea-level curve. From our reconstructed curve, we identify three broad periods of sea-level stability: 6–5.4 cal. kBP, 5–4.5 cal. kBP (which represents the Mid-Holocene high-stand, as expressed in this record) and 3.9–3 cal. kBP, together with two intervals when sea levels declined (regression): a steep regressive period 4.5–3.9 cal. kBP and a shallow regression 3–1.7 cal. kBP. The mechanism(s) driving the periods of quiescence remain to be determined. For the periods of regression, we propose that variability in the strength of the Asian Summer Monsoon/El Niño-La Niña may have played a significant role. Our chronology and model provide local nuance to existing understanding of sea-level change within the northern South China Sea during periods and across timescales outside the range or reach of modern instrumentation. In light of growing recognition that socio-cultural upheaval during the Late Holocene coincided with extreme climate events, potential correspondence between these events and intervals of sea-level regression carries two implications for the future. First, given expected 21st century climate instability, it adds local detail to modelling how sea levels are expected to change under different climate scenarios. Second, it offers the possibility that future climate conditions may periodically slow or interrupt the trajectory of regional sea-level rise; an opportunity that mitigation strategies may be able to exploit in defence of this region’s coastlines and particularly its heavily populated mega-deltas.
Introduction
River deltas comprise subaerial landforms that cap underlying deposits with subaqueous extensions from a river feeding sediment directly into a standing body of water at a rate exceeding effective dispersal processes (Hori and Saito, Reference Hori and Saito2022; Syvitski et al., Reference Syvitski, Anthony, Saito, Zăinescu, Day, Bhattacharya and Giosan2022). Globally, major river deltas have become centres of urbanisation and hotspots of environmental degradation and risk (McGranahan et al., Reference McGranahan, Balk, Colenbrander, Engin and MacManus2023; Anthony et al., Reference Anthony, Syvitski, Zăinescu, Nicholls, Cohen, Marriner, Saito, Day, Minderhoud, Amorosi, Chen, Morhange, Tamura, Vespremeanu-Stroe, Besset, Sabatier, Kaniewski and Maselli2024; Anthony et al., Reference Anthony, Syvitski, Cohen, Saito, Zăinescu, Vespremeanu-Stroe, Nicholls, Marriner, Amorosi, Maselli, Minderhoud, Tamura, Day, Woodroffe, Preoteasa, Tatui, Sabatier, Morhange, Besset, Kemp and Chen2025). In 2017, almost 45% (c. 152 million people) of a total global deltaic human population of c. 339 million people resided within the mega-deltas (i.e., >10,000 km2; Woodroffe et al., Reference Woodroffe, Nicholls, Saito, Chen, Goodbred and Harvey2006) of the Ganges–Brahmaputra–Meghna, Irrawaddy, Chao Phraya, Mekong, and Song Hong rivers in South and Southeast Asia (see also Caldwell et al., Reference Caldwell, Edmonds, Baumgardner, Paola, Roy and Nienhuis2019; after Edmonds et al., Reference Edmonds, Caldwell, Brondizio and Siani2020, Supplementary Table 1). The ecological and socio-economic services that deltaic environments provide are central to the maintenance of core regional infrastructure, the pursuit of sustainable development and the development of strategies to mitigate exposure to storm surges and the diverse impacts of climate-change-induced sea-level rise (The World Bank, 2009; Woodroffe et al., Reference Woodroffe, Nicholls, Saito, Chen, Goodbred and Harvey2006; Neumann et al., Reference Neumann, Emanuel, Ravela, Ludwig and Verly2015; Yuen et al., Reference Yuen, Hanh, Quynh, Switzer, Teng and Lee2021; Cremin et al., Reference Cremin, O’Connor, Banerjee, Bui, Chanda, Hua, Van Huynh, Le, Murshed, Mashfiqus, Vu, Sebesvari, Large and Renaud2023; Scown et al., Reference Scown, Dunn, Dekker, van Vuuren, Karabil, Sutanudjaja, Santos, Minderhoud, Garmestani and Middelkoop2023; Chan et al., Reference Chan, Paszkowski, Wang, Lu, Mitchell, Tran, Warner, Li, Chen, Li, Pal, Griffiths, Chen, W-Q and Zhu2024; Mega-Deltas, 2024). Sea-level change can be driven by variations in oceanic mass or volume, or by changes in the relative position of land and the sea surface. In the first case, sea-level change is defined as ‘eustatic’, and in the latter it is defined as ‘relative’ (Rovere et al., Reference Rovere, Stocchi and Vacchi2016a, Reference Rovere, Raymo, Vacchi, Lorscheid, Stocchi, Gómez-Pujol, Harris, Casella, O’Leary and Hearty2016b).
With relative sea-level (RSL) rise increasingly at the forefront of scientific assessment and both local and international policy frameworks, emphasis is being placed on combining global coarse-scale data with fine-scale natural and social science data to deliver locally tailored effective responses to coastal vulnerability (e.g., Milne et al., Reference Milne, Gehrels, Hughes and Tamisiea2009; Verburg et al., Reference Verburg, Dearing, Dyke, Svd, Seitzinger, Steffen and Syvitski2016; Hutton et al., Reference Hutton, Nicholls, Lázár, Chapman, Schaafsma and Salehin2018; Mann et al., Reference Mann, Bender, Lorscheid, Stocchi, Vacchi, Switzer and Rovere2019; Noor and Abdul Maulud, Reference Noor and Abdul Maulud2022; Rabett et al., Reference Rabett, Morimoto, Kahlert, Stimpson, O’Donnell, Mai Huong, Manh, Holmes, Khanh and Coward2023; Orthy and Hasan, Reference Orthy and Hasan2024).
Recent integrated assessment models incorporating the effects arising from geological and human-induced deltaic subsidence – for example, through groundwater extraction and upstream dam development (McGowan et al., Reference McGowan, Large, Henderson and Wallerstein2023; Chan et al., Reference Chan, Paszkowski, Wang, Lu, Mitchell, Tran, Warner, Li, Chen, Li, Pal, Griffiths, Chen, W-Q and Zhu2024) – have significantly increased projected rates of RSL rise in Southeast Asia (Nicholls et al., Reference Nicholls, Lincke, Hinkel, Brown, Vafeidis, Meyssignac, Hanson, Merkens and Fang2021; Vasilopoulos et al., Reference Vasilopoulos, Quan, Parsons, Darby, Tri, Hung, Haigh, Voepel, Nicholas and Aalto2021). Coupled with increases in the frequency of storm surge events, as well as salinity intrusion and coastal erosion (Neumann et al., Reference Neumann, Emanuel, Ravela, Ludwig and Verly2015; Duc et al., Reference Duc, Yasuhara, Nguyen and Nguyen2017; Liu et al., Reference Liu, DeMaster, Nguyen, Saito, Nguyen, Ta and Li2017; Minderhoud et al., Reference Minderhoud, Erkens, Pham, Bui, Erban, Kooi and Stouthamer2017; Haq and Milliman, Reference Haq and Milliman2023; Do et al., Reference Do, Vu, Samadhiya, Kumar, Gupta and Jagtap2024), potential underestimation of sea-level projections (Seeger and Minderhoud, Reference Seeger and Minderhoud2026), and considering that the two principal Vietnamese deltas (the Mekong and Song Hong/Red River) account for 44% (42,240/96,000 km2) of the total regional deltaic area, effective and sustainable future coastal management will be critical. Palaeo-proxies of RSL within deltaic environments are an essential component in the establishment of a geological-scale context for coastal and sea-level evolution outside the reach of instrumentation, which is confined to the relatively stable conditions of recent centuries (e.g., tide-gauge measurements began c. 1700). Integration of processes operating across multiple spatiotemporal scales is therefore required to better constrain future extreme water level projections due to sea-level rise (Horton et al., Reference Horton, Kopp, Garner, Hay, Khan, Roy and Shaw2018; Lin et al., Reference Lin, Kopp, Xiong, Hibbert, Zheng, Yu, Kumar, Dangendorf, Yi and Zhang2025).
The last major marine transgression (locally, the Dong Da transgression) during the Mid-Holocene c. 6,000–5,000 years BP (before present, where ‘the present’ is taken as AD 1950) has become a valuable point of reference (e.g., Woodroffe and Murray-Wallace, Reference Woodroffe and Murray-Wallace2012; Meltzner et al., Reference Meltzner, Switzer, Horton, Ashe, Qiu, Hill, Bradley, Kopp, Hill, Majewski, Natawidjaja and Suwargadi2017; Caldwell et al., Reference Caldwell, Edmonds, Baumgardner, Paola, Roy and Nienhuis2019; Helfensdorfer et al., Reference Helfensdorfer, Power and Hubble2019; Barnett et al., Reference Barnett, Charman, Johns, Ward, Bevan, Bradley, Camidge, Fyfe, Gehrels, Gehrels, Hatton, Khan, Marshall, Maezumi, Mills, Mulville, Perez, Roberts, Scourse, Shepherd and Stevens2020; Rabett et al., Reference Rabett, Morimoto, Kahlert, Stimpson, O’Donnell, Mai Huong, Manh, Holmes, Khanh and Coward2023; Anthony et al., Reference Anthony, Syvitski, Zăinescu, Nicholls, Cohen, Marriner, Saito, Day, Minderhoud, Amorosi, Chen, Morhange, Tamura, Vespremeanu-Stroe, Besset, Sabatier, Kaniewski and Maselli2024; Larsen et al., Reference Larsen, Van Hoang, Tran and Pham2024). Although multiple studies have examined the development of the Mekong and Red River deltas – hereafter referred to as MRD and RRD – most reconstructions have emphasised sedimentological, hydrogeological or seismic survey analysis (e.g., Ta et al., Reference Ta, Nguyen, Tateishi, Kobayashi, Tanabe and Saito2002; Tanabe et al., Reference Tanabe, Hori, Saito, Haruyama, Doanh, Sato and Hiraide2003a, Reference Tanabe, Hori, Saito, Haruyama, Vu and Kitamura2003b; Hori et al., Reference Hori, Tanabe, Saito, Haruyama, Nguyen and Kitamura2004; Schimanski and Stattegger, Reference Schimanski and Stattegger2005; Tanabe et al., Reference Tanabe, Saito, Lan Vu, Hanebuth, Lan Ngo and Kitamura2006; Nguyen et al., Reference Nguyen, Ta and Saito2010; Fujimoto et al., Reference Fujimoto, Masatomo, Nguyen, Thi Kim Oanh, Kawase, Huynh, Nakamura and Schmidt2011; Proske et al., Reference Proske, Hanebuth, Gröger and Diệm2011; Duong et al., Reference Duong, Lieu, Cuc, Saito, Huong, Phuong and Thuy2020; Collins et al., Reference Collins, Nguyen, Ta, Mao, Ishii, Kitagawa, Nakashima, Vo and Tamura2021; Yen et al., Reference Yen, Nhan, Nghi, Toan, Khien, Lam, Long, Thanh, Hung, Trang, Dien, Tuyen, Truong, Dung, Thao and Quang Lan2021; Tran et al., Reference Tran, Hung, Thu, Shakirov, Syrbu, Toan, Khien, Huong, Lee, Thanh, Nhan, Trang, Thao, Dung and Dien2022). However, post-depositional distortions through subsidence and bioturbation, and the effect of compaction during sampling are complicating factors that may contribute to inconsistencies between datasets (Mann et al., Reference Mann, Bender, Lorscheid, Stocchi, Vacchi, Switzer and Rovere2019). In this article, we present a 14C-dated tidal notch chronology, providing an independent line of evidence for the timing and process of marine transgression and regression during the Late Pleistocene to Late Holocene and consider its relevance to extreme climate-ocean events predicted for the 21st century.
Recent advances in relative sea-level reconstruction have emphasised the importance of quantifying the indicative meaning and associated uncertainties of sea-level indicators (e.g., Rovere et al., Reference Rovere, Stocchi and Vacchi2016a, Reference Rovere, Pappalardo, O’Leary and Elias2025). In this framework, geomorphological and biological markers are interpreted relative to the tidal range within which they form, allowing both chronological and vertical uncertainties to be incorporated into regional sea-level reconstructions. The oyster-bearing tidal notches of Tràng An, northern Vietnam (Figure 1), provide a valuable archive in this respect because they preserve both geomorphological and biogenic evidence of former shoreline positions.
(a) Notch site at Tam Coc. Possibly a tidal notch that either formed under medium exposure conditions that led to the formation of a secondary infralittoral zone or sheltered conditions, which would account for its deep and narrow v-shape. In this case, the presence of two apexes may indicate periods of SL still stand (image: Kahlert). (b) Thanh Boat wharf notch site. Oyster shells were found concreted to the inside of a karst conduit. While no notch was present at the sampling location, its elevation is consistent with the upper notch sequences that are present throughout Thay Vi valley (image: Kahlert). (c) Oyster shell fragments were found in sediments concreted to the Thay Vi lower notch sequence (image: Kahlert). (d) Oyster shell fragments were found in sediments concreted to the Thay Vi upper notch sequence. (e) Co Vien Lau notch site. Oyster shells were concreted into recesses within the upper notch (image: Kahlert).
Figure 1. Long description
A multi-panel set of five photos labeled a through e.
* Panel a: A wide-angle view of a deep, v-shaped tidal notch in a limestone cliff. A red and white vertical survey pole stands in the center for scale.
* Panel b: A close-up of a karst conduit at a wharf. The rock face is textured and gray, with a small dark opening where oyster shells are concreted inside.
* Panel c: A view of the lower notch sequence at Thay Vi. A small rectangular color and scale calibration card lies on the sandy ground in the foreground. Above it, dark, bulbous sediment concretions containing oyster shell fragments are attached to the overhanging rock.
* Panel d: A vertical view of the upper notch sequence at Thay Vi. A red and white survey pole is positioned against a lower ledge. The rock face above shows vertical weathering patterns and horizontal indentations where shells are embedded.
* Panel e: A panoramic view of the Co Vien Lau notch site. A long, horizontal limestone overhang is covered in green vegetation. Below the notch, a dense row of lily pads grows in the water along the base of the cliff, bordering a paved path in the foreground.
Notch formation
Tidal sea notches (Pirazzoli, Reference Pirazzoli and Ovd1986; Rovere et al., Reference Rovere, Stocchi and Vacchi2016a, Reference Rovere, Pappalardo, O’Leary and Elias2025) offer an alternative way to reconstruct regional sea-level change allowing for associated processes, such as uplift (Maeda et al., Reference Maeda, Omura, Berdin, Hosono, Atsumi and Nakamura2004; Marfai et al., Reference Marfai, Sekaranom and Cahyadi2014; Surakiatchai et al., Reference Surakiatchai, Choowong, Charusiri, Ch, Chawchai, Pailoplee, Chabangborn, Phantuwongraj, Chutakositkanon, Kongsen, Nimnate and Bissen2018; Oliver and Terry, Reference Oliver and Terry2019). To date, no detailed sequence of sea notch-based assessment of RSL exists for either of Vietnam’s two principal deltas. In the MRD, the isolated granite monadnocks and limestone hills in the vicinity of Ha Tien–Kien Luong have been a focus of sea notch research (Fontaine, Reference Fontaine1970; Fontaine and Délibrias, Reference Fontaine and Délibrias1974; Kiernan, Reference Kiernan2011; Tran and Ha, Reference Tran and Ha2020), though the contribution of these data to understanding the evolution of the MRD has tended to be neglected in favour of studies into human and contemporary impacts (Kiernan, Reference Kiernan2011). In the north, incorporation of sea notch evidence into RRD evolution has received recent attention (Yao et al., Reference Yao, Harff, Meyer and Zhan2009; Tue et al., Reference Tue, Quan, Nguyen, Dung, Quy and Nhuan2018; Kahlert et al., Reference Kahlert, O’Donnell, Stimpson, Mai Hương, Hill, Utting and Rabett2021) through geological, archaeological and palaeoecological research undertaken in the Tràng An Landscape Complex World Heritage Site. The current article expands the northern portfolio of evidence by providing the first Accelerator Mass Spectrometry (AMS) radiocarbon-dated multi-phase Pleistocene–Holocene chronology for the RRD and considers the potential wider implications raised by its findings.
The processes and constraints affecting marine notch formation and the use of these features as markers of eustatic sea level have been presented elsewhere (Pirazzoli, Reference Pirazzoli and Ovd1986; Rovere et al., Reference Rovere, Pappalardo, O’Leary and Elias2025). The methodology applied specifically in the context of Tràng An is presented in detail in Kahlert et al. (Reference Kahlert, O’Donnell, Stimpson, Mai Hương, Hill, Utting and Rabett2021). In brief, erosional tidal or wave-cut notches comprise continuous or semi-continuous bands that form around the tidal apex on exposed cliff faces during periods of stable sea level. Chemical dissolution, mechanical erosion and bioerosion are all associated with the tidal range and gradually carve out horizontal bands along exposed rock faces (Trenhaile, Reference Trenhaile2015). Since notches require stable sea-level conditions to form, their presence and elevation above sea level can serve to estimate past relative sea-level changes (Georgiou et al., Reference Georgiou, Stocchi, Casella and Rovere2024). Where datable materials, preferably marine oysters, are found cemented to a notch, a sea-level chronology can be established (Boyd and Lam, Reference Boyd and Lam2004; Kahlert et al., Reference Kahlert, O’Donnell, Stimpson, Mai Hương, Hill, Utting and Rabett2021). Their morphology is determined by a complexity of factors, such as the biota that occupies a notch, water chemistry, wave action and tidal range. All these processes impact notch morphology and, therefore, sea-level reconstruction. Exposure to wave action, for example, can broaden the distance between notch base and roof and, under moderate to high wave action exposure, lead to the formation of infralittoral and surf notches within a notch, whereas a notch that formed in a sheltered location exhibits a narrower, v-shaped morphology (Pirazzoli, Reference Pirazzoli and Ovd1986). Over time, notches can undercut rocks by several meters, causing overhanging cliffs to slip or collapse. These processes, being integral to coastal karst evolution, potentially impact RSL reconstruction by offsetting notch elevation and shearing off geomorphological markers, particularly notch roofs.
Study location
The RRD preserves a geological archive spanning the Late Palaeozoic to the present (Yen et al., Reference Yen, Nhan, Nghi, Toan, Khien, Lam, Long, Thanh, Hung, Trang, Dien, Tuyen, Truong, Dung, Thao and Quang Lan2021). Its bedrock is composed predominantly of thickly bedded Carboniferous–Permian limestones, deposited around 250–300 million years ago in a shallow tropical marine environment on the margin of the Indochina Block (UNESCO, 2014a). Later subjected to tectonic deformation during multiple orogenic episodes and most notably during the Cenozoic, as a result of the India–Eurasia collision, which reactivated the Red River Fault (RRF) (Nghi et al., Reference Nghi, Ngo Quang, Do Thi Van, Nguyen Dinh and Nguyen Van1991; Phach et al., Reference Phach, Lai, Shakirov, Le and Tung2020).
The RRF is part of a major left-lateral strike-slip structure, the Ailao Shan-Red River shear zone, extending from Yunnan, China, into northern Vietnam, exerting strong control over regional uplift, subsidence, and drainage reorganisation (Zhang et al., Reference Zhang, Chen, Wang, Reiners, Cai, Speranza, Zhang, Zhong and Liu2022). Neotectonic uplift exposed the southern part of the Phan Si Pang Ninh Binh structure, of which the Tràng An massif (N 20 15 24, E 105 53 47) (Figure 2a) forms part (Do et al., Reference Do, Nguyen, Nguyen, Dam, Dinh, Tran and Trinh2013; UNESCO, 2014c).
(a) Red River Delta and locations of sampling sites in Tràng An and Halong Bay (image: Kahlert). (b) Location of sample sites within the Tràng An WHS property (image: Kahlert). (c) Location of dated oysters within the Thai Vi valley of the Tràng An massif (image: Kahlert).
Figure 2. Long description
Panel a is a regional map of the Red River Delta outlined in black. Major cities marked with black dots include Viet Tri in the Northwest, Hanoi in the center, Haiphong in the East near Cat Ba or Halong Bay, and Nam Dinh and Ninh Binh in the South. The Trang An area is highlighted with a red boundary South of Ninh Binh. A scale bar indicates 0 to 40 kilometers.
Panel b is a grayscale topographic map focusing on the Trang An W H S property, which is enclosed by a dashed black line. The terrain shows dense karst formations. Ninh Binh is labeled to the Northeast and Tam Coc to the Southeast. Two sample location symbols, a circle and a triangle, are clustered near Tam Coc. A scale bar indicates 0 to 6 kilometers.
Panel c is a high-resolution color topographic map of the Thai Vi valley. A red line marks a boundary in the South. Sample locations are marked with black dots and triangles. From North to South, labels include Wk-8269, Wk-8268, and a cluster containing Wk-8267, U B A-44781, U B A-45068, U B A-44782, U B A-54214, and U B A-44780. Further Southeast, another cluster includes U B A-45067, U B A-54212, U B A-45066, and U B A-54215. A scale bar indicates 0 to 1 kilometer.
Throughout the Quaternary, prolonged fluvial incision, combined with fracturing and jointing associated with fault activity, enhanced subsurface water circulation, accelerating karstification and the development of dolines, poljes, conical hills and underground landforms that are typical of Vietnamese karst (Do et al., Reference Do, Tran, Pham, Batelaan, Dusar, Masschelein, Vu, Tran and Nguyen2004) and its later evolution, which singles out Tràng An for particular recognition as a World Heritage (Do et al., Reference Do, Tran, Pham, Batelaan, Dusar, Masschelein, Vu, Tran and Nguyen2004; UNESCO, 2014a, 2014b). The present landscape comprises steep tower karst separated by flat-floored alluvial basins and seasonally flooded valleys, where ongoing dissolution processes continue to be moderated by dense forest vegetation and tropical monsoonal rainfall, as they have been for at least the last 20,000 years (e.g., Gillieson, Reference Gillieson and Gupta2005; Rabett et al., Reference Rabett, Ludgate, Stimpson, Hill, Hunt, Ceron, Farr, Morley, Reynolds, Zukswert, Simpson, Nyiri, Verhoeven, Appleby, Meneely, Phan, Dong, Lloyd-Smith, Hawkes, Blyth and Tâń2017).
Tràng An’s position on the southern margin of the RRD places it at the likely maximum extent of the Mid-Holocene (Dong Da/Flandrian) marine transgression (e.g., Tanabe et al., Reference Tanabe, Hori, Saito, Haruyama, Doanh, Sato and Hiraide2003a, Reference Tanabe, Hori, Saito, Haruyama, Vu and Kitamura2003b; Hori et al., Reference Hori, Tanabe, Saito, Haruyama, Nguyen and Kitamura2004; Tanabe et al., Reference Tanabe, Saito, Lan Vu, Hanebuth, Lan Ngo and Kitamura2006; Funabiki et al., Reference Funabiki, Haruyama, Quy, Hai and Thai2007; Rabett et al., Reference Rabett, Morimoto, Kahlert, Stimpson, O’Donnell, Mai Huong, Manh, Holmes, Khanh and Coward2023). Sea-level high stands during interglacial phases of the Quaternary, particularly during the Mid-Holocene, caused marine transgressions that inundated low-lying valleys and cave systems, depositing marine and estuarine sediments within the massif (UNESCO, 2014c; O’Donnell et al., Reference O’Donnell, Nguyen, Stimpson, Holmes, Kahlert, Hill, Vo and Rabett2020). These events left behind stratigraphic markers, such as marine abrasion notches, shell-rich sands and tidal flat deposits. Conversely, low stands exposed the coastal plain, allowing fluvial incision into the karst and delivery of riverine sediments inland (Lam and Boyd, Reference Lam and Boyd2001; Boyd and Lam, Reference Boyd and Lam2004; Kahlert et al., Reference Kahlert, O’Donnell, Stimpson, Mai Hương, Hill, Utting and Rabett2021). The preservation of wave-cut notches along the flanks of the property provides a unique window into deltaic sea-level change in the distal or downstream area of the RRD (Lam and Boyd, Reference Lam and Boyd2003; Boyd and Lam, Reference Boyd and Lam2004; Nguyen et al., Reference Nguyen, Nguyen and Dũng2012; Pham et al., Reference Pham, Van, Nguyen and Nguyen2013; Tran et al., Reference Tran, Nguyen, Vu and Trinh2013;UNESCO, 2014a; Kahlert et al., Reference Kahlert, O’Donnell, Stimpson, Mai Hương, Hill, Utting and Rabett2021).
Methods
Between 2017 and 2019, a total of 30 notches were surveyed as part of the SUNDASIA project, and a relative sea-level (RSL) model for Tràng An and the Red River Delta (RRD) was developed by Kahlert et al. (Reference Kahlert, O’Donnell, Stimpson, Mai Hương, Hill, Utting and Rabett2021). Multiple measurements were taken along sections of notches at discernible morphological features, including base, apex and roof (after Pirazzoli Reference Pirazzoli and Ovd1986). In the case of compound notches with multiple retreat points (apexes), measurements were taken of each apex and at the interface separating them. All measurements were taken with a Leica TS15 total station, which was aligned to georeferenced reference points measured with nRTK (network Real Time Kinematic) GNSS (Global Navigation Satellite System) rovers (Leica GS15 and later Trimble R12i). The GNSS rover was connected to a CORS (continuously operating reference station) at Nam Ðịnh using NTRIP (Networked Transport of RTCM via Internet Protocol). Measurement integrity between multiple field seasons was ensured through control measurements taken at a local datum point (National Benchmark 140,411, U Bo mountain) and at an arbitrary reference point at the project fieldwork base. The elevation measurements were also cross-referenced with the Hon Dau 1992 national datum (EPSG:5126) via direct measurement during previous fieldwork (Kahlert et al., Reference Kahlert, O’Donnell, Stimpson, Mai Hương, Hill, Utting and Rabett2021). Multiple, repeated control measurements were also taken at a subset of notch locations. Where RTK correction was not possible, measurements were post-processed in Leica Geo Office using RINEX data from the Nam Ðịnh reference station.
Tidal notches were treated as the primary sea-level indicator, while oyster shells provided chronological constraints on notch occupation. Vertical uncertainty was estimated conservatively from survey precision and notch morphology following the geomorphological framework developed by Kahlert et al. (Reference Kahlert, O’Donnell, Stimpson, Mai Hương, Hill, Utting and Rabett2021). Elevations were additionally adjusted for long-term uplift using:
where
$ {h}_c $
is uplift-adjusted elevation (adj. asl),
$ {t}_c $
is the median calibrated age (cal. BP) and
$ {U}_t $
is uplift.
The northwest of the massif is dominated by an elevated Pleistocene fluvial terrace, which marks the maximum extent of the Mid-Holocene high-stand (Tanabe et al., Reference Tanabe, Hori, Saito, Haruyama, Doanh, Sato and Hiraide2003a, Reference Tanabe, Hori, Saito, Haruyama, Vu and Kitamura2003b). Notches in this area are rare, and only fragments have survived near the fringe of the fluvial terrace (Kahlert et al., Reference Kahlert, O’Donnell, Stimpson, Mai Hương, Hill, Utting and Rabett2021). Several sites were identified along the eastern flanks of the massif, occasionally within the interior (e.g., Vung Tham – see O’Donnell et al., Reference O’Donnell, Nguyen, Stimpson, Holmes, Kahlert, Hill, Vo and Rabett2020). Isolated examples were recorded along the western flanks (e.g., So Lai and Thien Ha). The principal concentration of notches containing in-situ and potentially datable marine sediments was recorded in the southeast, near Tam Coc (Figure 2b,c), representing the area most directly influenced by past tidal processes (UNESCO, 2014a; Kahlert et al., Reference Kahlert, O’Donnell, Stimpson, Mai Hương, Hill, Utting and Rabett2021). This area comprises the most mature karst, characterised by wide valleys flanked by steep ridges and tower karst. Complex notch systems line the bases of these formations and are indicative of repeated inundations over long periods.
Selection of target sites for sampling was guided by two principal parameters: (1) elevation of the notch apex and (2) preservation status of the notch. Preference was given to well-preserved notches where sediments were more likely to survive in situ. Based on these parameters, a priority list of sites was compiled to maximise the chances of obtaining samples for all modelled phases.
Sampling sites and samples were photographically recorded, described in the field, and shipped under licence to Queen’s University Belfast for analysis. The location of each sample within the relevant notch was matched against existing digital records, and an elevation value was extrapolated. All samples were examined for organic material such as shells and charcoal. Potentially datable material was extracted, taxonomically examined (EH) and submitted to the Queen’s University Belfast 14CHRONO AMS dating facility. All shells were pretreated with acid etch (Reimer et al., Reference Reimer, Hoper, MacDonald, Reimer and Thompson2015) and graphitized using hydrolysis. The charcoal sample was treated with AAA (acid-alkali-acid; Reimer et al., Reference Reimer, Hoper, MacDonald, Reimer and Thompson2015) and graphitized via combustion.
Statistical analyses were performed in R (R Core Team, 2025) using the rice package (Blaauw and Reimer, Reference Blaauw and Reimer2026) for radiocarbon calibration, posterior sampling and marine reservoir correction. Monte Carlo simulation and locally weighted regression (LOESS) were implemented using base R functions.
Existing Holocene radiocarbon ages from oysters sampled in marine notches in Tràng An and Ha Long Bay were taken from Lam and Boyd (Reference Lam and Boyd2001) and recalibrated alongside our new samples to ensure direct comparability. Since the shell samples were estuarine, they were calibrated using a 50:50 mix of the northern hemisphere IntCal20 calibration curve (Reimer et al., Reference Reimer, Austin, Bard, Bayliss, Blackwell, Bronk Ramsey, Butzin, Cheng, Edwards, Friedrich, Grootes, Guilderson, Hajdas, Heaton, Hogg, Hughen, Kromer, Manning, Muscheler, Palmer, Pearson, van der Plicht, Reimer, Richards, Scott, Southon, Turney, Wacker, Adolphi, Büntgen, Capano, Fahrni, Fogtmann-Schulz, Friedrich, Köhler, Kudsk, Miyake, Olsen, Reinig, Sakamoto, Sookdeo and Talamo2020) and the Marine20 calibration curve (Heaton et al., Reference Heaton, Köhler, Butzin, Bard, Reimer, Austin, Bronk Ramsey, Grootes, Hughen, Kromer, Reimer, Adkins, Burke, Cook, Olsen and Skinner2020) with a fixed regional marine reservoir correction of ΔR = −190 ± 23 years (Yang et al., Reference Yang, Zhou, Cheng, Zhang, Wei, Ma, Zhou, Yan, Chen and Hou2025). The single charcoal sample was calibrated using IntCal20 without application of a marine reservoir correction. Owing to their temporal separation from the Holocene and the increasing uncertainty associated with radiocarbon ages approaching the effective dating limit, the Pleistocene samples were calibrated separately using the same calibration parameters but were not incorporated into the Holocene relative sea-level model.
Rather than using summary statistics from the calibrated age distributions, chronological uncertainty was propagated directly through the analysis by randomly sampling 2,000 calendar ages from each calibrated posterior probability distribution. Sea-level change was then reconstructed using a Monte Carlo ensemble of second-degree locally weighted polynomial (LOESS) regressions (Cleveland, Reference Cleveland1979), where
and z is elevation,
$ {f}_t+\varepsilon $
is the locally weighted regression as a function of calibrated age and
$ \varepsilon $
represents the combined chronological and vertical measurement uncertainty propagated through the Monte Carlo simulation. A smoothing span of 0.55 was used throughout. Predictions from the 2,000 model realisations were evaluated across a regular calendar-age grid, from which the ensemble median sea-level curve and the 95% uncertainty envelope (2.5th and 97.5th percentiles) were derived. Tidal notches constitute the primary sea-level indicator for this study, whereas oyster shells provide chronological constraints on the timing of notch occupation. Consequently, uncertainty associated with the surveyed elevation and geomorphological definition of each notch was propagated directly through the sea-level reconstruction, whereas the indicative meaning of oyster occurrence was considered separately during interpretation. This approach follows the distinction proposed by Rovere et al. (Reference Rovere, Stocchi and Vacchi2016a, Reference Rovere, Raymo, Vacchi, Lorscheid, Stocchi, Gómez-Pujol, Harris, Casella, O’Leary and Hearty2016b) between uncertainty associated with the vertical position of the geomorphic sea-level indicator and uncertainty associated with its indicative meaning, the latter being only partly constrained for Tràng An owing to the absence of a reconstructed palaeo-tidal regime.
Chronological uncertainty was propagated by randomly sampling 2,000 calendar ages from each calibrated posterior probability distribution using the r.calib() function. Vertical measurement uncertainty was propagated simultaneously by randomly sampling elevations from normal distributions centred on the measured notch elevations. For samples collected during this study, measurement uncertainty combined GNSS/total-station survey uncertainty (± 0.05 m) and uncertainty associated with defining the notch apex (± 0.10 m). For legacy samples from Boyd and Lam (Reference Boyd and Lam2004), survey uncertainty was conservatively increased to ± 0.20 m to reflect the absence of original survey metadata while retaining the same notch-definition uncertainty. Vertical measurement uncertainty was therefore calculated as
where
$ {\sigma}_v $
is total vertical measurement uncertainty,
$ {\sigma}_{GPS} $
is survey uncertainty and
$ {\sigma}_n $
is uncertainty associated with defining the notch apex. For each Monte Carlo iteration, a randomly sampled calibrated age was paired with a randomly sampled elevation, thereby propagating both chronological and vertical measurement uncertainty throughout the relative sea-level reconstruction.
Results and discussion
The model presented in Kahlert et al. (Reference Kahlert, O’Donnell, Stimpson, Mai Hương, Hill, Utting and Rabett2021) reconstructed palaeo-coastlines rather than sea levels. K-means cluster analysis was applied to account for vertical uncertainty across a range of notch sites caused by highly localised vertical displacement in this dynamic karst landscape (UNESCO, 2014b) as well as differing hydromorphological settings. The model indicated three, potentially four, distinct phases of RSL high stands during the Holocene. The modelled RSL values that describe these three phases were adopted for the current model and adjusted for isostatic and tectonic influences. The central region of the Red River Basin, particularly around the Vinh Ninh Fault, has experienced vertical land movement (VLM) of c. -50 m since the end of the Miocene (Phach et al., Reference Phach, Lai, Shakirov, Le and Tung2020). The area around Tràng An, on the south-west margin of the basin, has likely experienced uplift rather than subsidence in order to facilitate the development of fenglin karst. For this landform to evolve, planation and uplift must be in equilibrium, and uplift rates of 23 mm to 300 mm/ka have been cited by Waltham (Reference Waltham2009). A conservative approach was adopted to account for VLM, using minimum values based on the available evidence from Tràng An (Kahlert et al., Reference Kahlert, O’Donnell, Stimpson, Mai Hương, Hill, Utting and Rabett2021). The evolution of the Tràng An karst is believed to have commenced some 5 million years ago (UNESCO, 2014a), with continuous karstification and tectonic processes forming the complex fenglin and fengcong landscape observed today. The tallest surviving karst tower has a height of 240 m. Thus, 240,000 mm/5,000 ka = 48 mm/ka (
$ {U}_t $
). This value represents a minimum uplift estimate because it does not account for the erosion of mountain peaks or possible increased uplift rates since the beginning of the Holocene (Phach et al., Reference Phach, Lai, Shakirov, Le and Tung2020).
Twelve notches were identified as likely to contain concreted marine sediments. Potentially viable samples were extracted from eight of these sites. Four samples (Than Boat Wharf, Thay Vi lower, Tam Coc and Co Vien Lau) included datable material from oysters (Ostreidae). Sediments from the remaining sites did not produce datable material. While Ostreidae were deemed the most favourable species, the scarcity of datable material from some samples required that non-oyster molluscs were also initially submitted for dating. A total of 10 samples were dated (N = 6 in 2021 and N = 4 in 2024), of which 6 returned Holocene and 4 returned Pleistocene dates. Combined with existing dates from Boyd and Lam (Reference Boyd and Lam2004, N = 11), we used 17 data points to model a sea-level curve for the Mid-Late Holocene. Large chronological uncertainties associated with early dates near the limits of radiocarbon dating rendered these data unsuitable for sea-level modelling.
Pleistocene sequence
Pleistocene dates between c. 52 and 29 cal. ka BP were recorded at Co Vien Lau 1 and Tam Coc. At Tam Coc, the samples were obtained from an oyster bed that formed at the base and apex of the notch. The Co Vien Lau 1 sample came from a residual sediment layer stuck to the roof and apex of the notch. Repeat dating of different specimens from the notches put the age of the Co Vien Lau 1 oyster bed between c. 42.2 and 29.4 cal. ka BP and the Tam Coc oyster bed between c. 51.7 and 37.8 cal. ka BP (Table 1). Adjusting the recorded elevations for isostasy, and assuming a constant uplift rate of 48 mm/ka, the notches at Tam Coc and Co Vien Lau were 2.2 m below their current level 46.5 k yrs ago and uplifted to 1.4 m below their current level by c. 30 cal. ka BP (Table 1).
Calibration results (R – rice) of Pleistocene dates and elevations adjusted to isostasy
Table 1. Long description
The table contains 10 columns: Lab I D, Site, Species, 14 C B P, S D, 95 percent range (cal. B P), 95 percent h p d ranges (cal. B P), q 5.0, m asl, and m cal asl.
* Row 1: Lab I D U B A 45066, Site Tam Cốc, Species Ostreidae, 14 C B P 43,193, S D 1,480, 95 percent range 51,712 to 43,659, 95 percent h p d 50,636 to 43,134 (95 percent), q 5.0 45,370, m asl 4.5, m cal asl 2.3.
* Row 2: Lab I D U B A 54212, Site Co Vien Lau, Species Ostreidae, 14 C B P 36,591, S D 709, 95 percent range 42,270 to 40,297, 95 percent h p d 42,360 to 40,438 (95 percent), q 5.0 40,590, m asl 2.9, m cal asl 1.0.
* Row 3: Lab I D U B A 54215, Site Tam Cốc, Species Ostreidae, 14 C B P 34,348, S D 536, 95 percent range 40,699 to 37,753, 95 percent h p d 40,694 to 37,749 (95 percent), q 5.0 38,430, m asl 4.5, m cal asl 2.7.
* Row 4: Lab I D U B A 45067, Site Co Vien Lau, Species Ostreidae, 14 C B P 25,733, S D 188, 95 percent range 30,394 to 29,413, 95 percent h p d split into two ranges: 29,406 to 29,332 (1.2 percent) and 30,388 to 29,488 (93.8 percent), q 5.0 29,100, m asl 2.9, m cal asl 1.5.
Custom calibration curve.
The earliest dates from Co Vien Lau 2 (UBA-45067 and UBA-54212) and Tam Coc (UBA-45066 and UBA-54215) approach the upper limit of radiocarbon dating and should therefore be interpreted with caution because some may exceed the effective range of the method. Especially the earliest dates from UBA-45066 and UBA-54214 may well be beyond past that limit. The two later date ranges, however, fall towards the end of the poorly resolved Vinh Phuc transgression chronology, which is estimated to span c. 70–30 ka BP (Trung et al., Reference Trung, Van, Ha and Thuy2012; Tran et al., Reference Tran, Hung, Thu, Shakirov, Syrbu, Toan, Khien, Huong, Lee, Thanh, Nhan, Trang, Thao, Dung and Dien2022). Hanebuth et al. (Reference Hanebuth, Saito, Tanabe, Vu and Ngo2006) identified Marine Isotope Stage (MIS)-3 sea-level fluctuations from a core taken at Nam Ðịnh, ~50 km east of Tràng An and 7 km inland from the present coastline. Organic contents from a sequence from the lower 30 m of a 60 m core were dated to around 45 cal. ka BP and attributed to a transgressive event. Further support for this comes from stalagmite analysis from central Vietnam (Patterson et al., Reference Patterson, Johnson, Griffiths, Kinsley, McGee, Du, Pico, Wolf, Ersek, Mortlock and Yamoah2023), which reports wet conditions and likely regional inundation during the period 45–35 ka.
Geomorphological markers in the form of marine terraces at 10–15 m asl have also been attributed by association to the Vinh Phuc transgression (UNESCO, 2014a), but the proposed age of these terraces is brought into question by the presence of oyster beds of similar age in erosional notches at 2.9 m and 4.5 m asl. Since tropical oysters settle in the intertidal zone (Angell, Reference Angell1986) and tidal notches form at sea level within the tidal minimum and maximum (Pirazzoli, Reference Pirazzoli and Ovd1986), wave-cut terraces are unlikely to have formed ~10 m above these features contemporaneously. It is more likely that the terraces formed significantly earlier, or that other processes produced these geomorphological features. The absence of direct dating evidence precludes a conclusive interpretation.
If we accept the proposed adjustment for isostasy, the Tam Coc and Co Vien Lau notches were uplifted between 1.4 and 2.2 m since the Vinh Phuc transgression. This suggests that the Vinh Phuc transgression reached levels of at least 0.9–2.6 m above present sea level. The presence of oysters of Holocene date is unlikely to be synonymous with the formation of the notches. By the time the oysters settled, the notches had already eroded deep into the cliffs. It is therefore more likely that the notches themselves date to an earlier high-stand, such as the MIS 5e transgression. Boyd and Lam (Reference Boyd and Lam2004) recorded oyster shells of > 40 kBP at elevations of 7–10 m asl. While no specific dates were provided by them, these may correspond to the same transgression as our dates. The difference in elevation of up to 8.5 m between the two sample sites may offer some insight into differential VLM in the RRD during the Pleistocene.
Holocene sequence
All new samples dated to the Mid-Late Holocene were collected at Thay Vi 1 and the Than Boat Wharf notches (Figure 1b,c), which belong to an extensive notch system that developed along the cliff faces of the Thai Vi and Tam Coc valleys. The three Tam Coc samples collected by Boyd and Lam (Reference Boyd and Lam2004) came from the same notch system, albeit further upriver. Posterior dates and adjusted elevations are presented in Table 1.
Observed discrepancies in VLM across the area of the RRD have been principally confined to Holocene or present-day analysis of subsidence. For example, in relation to coastal erosion in the vicinity of Nam Dinh (Nguyen Hao and Takewaka, Reference Nguyen Hao and Takewaka2023) and groundwater extraction in the southern districts of Hanoi city (Dang et al., Reference Dang, Doubre, Weber, Gourmelen and Masson2014). However, a connection to the active RRFZ is presumed. Indications are that uplift on the SW side of the fault has been limited between the Pliocene and Quaternary (Zuchiewicz et al., Reference Zuchiewicz, Cu’ò’ng, Zasadni and Yêm2013). The RRFZ and its continuation under the Gulf of Tonkin exhibit tectonic structures predominantly associated with horizontal strike-slip displacement; however, differential VLM is reported to affect the central area of the South China Sea (SCS) (Dung et al., Reference Dung, Kulinich, Minh, Van Sang, Duong, Thanh, Dai, Lap, Dung, Tung, Cham and Son2021), including along the northern coast of Hainan Island (e.g., Yao et al., Reference Yao, Harff, Meyer and Zhan2009). Terrestrial uplift along the northern margin of the delta, for example, the Yen Tu massif (Phan et al., Reference Phan, Ngo Van, Nguyen Van, Hoang Quang, Bui Van, Bui Thi, Mai Thanh and Nguyen2012), may lend further support to the suggestion that the NE and SW parts of the RRD are subject to differing vertical displacement, which may factor in the discrepancies reported herein for Pleistocene records. Even so, further research is needed to confirm this.
The close geographic proximity and chronological coverage of our sample set allow reconstruction of a detailed sea-level chronology for the area that was most affected by marine inundations and place it in a broader temporal context of the RRD (Boyd and Lam, Reference Boyd and Lam2004; UNESCO, 2014a; Kahlert et al., Reference Kahlert, O’Donnell, Stimpson, Mai Hương, Hill, Utting and Rabett2021). The resulting sea-level curve (Figure 3) illustrates a plateau of comparative stability and tidal notch development 6–5.4 cal. kBP at c. 4.8 m asl. This was followed by a peak 5.4–4.5 cal. kBP, c. + 5.4 m cal. asl., representative of the Dong Da (Flandrian) transgression high-stand, and provides an update to the reference literature pertaining to the RRD. Following the high-stand, there is a regression of 4.5–3.9 cal. kBP, where the sea level drops to + 3–4 m cal. asl., after which there is another plateau and shallow regression of 3.9–1.7 cal. kBP.
Sea-level reconstruction for the RRD during the Mid–Late Holocene. Ostreidae and Meretrix sp. (blue points, 95% ranges as error bars) and the single charcoal (red, 95% range) radiocarbon ages are plotted against measured elevations. The black line represents the loess-smoothed median SL curve with 95% uncertainty (green shaded ribbon), based on Marine20 calibration with ΔR = −190 ± 23 years. Orange bars represent the regional onset and duration of the 4.2 and 2.8 cal. kBP Cold Events.
Figure 3. Long description
A line graph plots elevation in meters above sea level (a s l) on the Y-axis from 3.0 to 6.0 against time in cal B P on the X-axis from 6000 to 2000.
* The primary data is a black loess-smoothed median curve surrounded by a green shaded ribbon representing 95 percent uncertainty.
* The curve begins at 6000 cal B P at approximately 4.8 meters, rises to a peak of 5.2 meters around 4600 cal B P, then descends sharply to 3.6 meters by 3200 cal B P, remaining relatively stable until 2000 cal B P.
* Blue violin-shaped data points with horizontal error bars represent Ostreidae and Meretrix species samples. These are labeled with codes such as W K-8268, W K-8267, U B A-45068, and W K-8260.
* A single red violin-shaped point labeled U B A-44782 represents charcoal radiocarbon age near 2900 cal B P.
* Two orange rectangular bars at the bottom of the plot indicate cold events. The first spans from approximately 4800 to 3800 cal B P. The second spans from approximately 3000 to 2400 cal B P.
* The background is filled with a dense scatter of small grey dots representing individual data iterations.
Sea-level conditions in the northern SCS during the Mid-Late Holocene
Proxy records and simulations indicate that Global Mean Sea Level fluctuations during the Early- to Mid-Holocene were tied to episodic meltwater pulses overprinting a steady rise due to the continuing retreat of high-latitude icesheets (Fox-Kemper et al., Reference Fox-Kemper, Hewitt, Xiao, Adalgeirsdottir, Drijfhout, Edwards, Golledge, Hemer, Kopp, Krinner, Mix, Notz, Nowicki, Nurhati, Ruiz, Sallée, Slangen, Yu, Masson-Delmotte, Zhai, Pirani, Connors, Péan, Berger, Caud, Chen, Goldfarb, Gomis, Huang, Leitzell, Lonnoy, JBR, Maycock, Waterfield, Yelekçi, Yu and Zhou2021). These forcing mechanisms, though, are considered to have had a limited effect on RSL change in Southeast Asia where, by the Late Holocene, sea-level variability is thought to have been principally driven by sterodynamic mechanisms (ocean density and circulation changes) (Swapna et al., Reference Swapna, Gan, Lau and Fung2009; Lambeck et al., Reference Lambeck, Rouby, Purcell, Sun and Sambridge2014; Stammer et al., Reference Stammer, Le Cozannet, Slangen, Carson, Cazenave, Dangendorf, Hill, Yang and Zorita2025). Detailed modelling of these drivers in relation to the two RRD stabilisation intervals identified here is outside the scope of the current article; however, we present the following hypothesis for testing.
Across the transition from the Mid-to Late Holocene (5–3 cal. kBP), southwestern China and mainland Southeast Asia were subject to a rapid reduction in (and centennial-scale internal fluctuations of) the Asian Summer Monsoon (ASM) (Wang et al., Reference Wang, Cheng, Edwards, Kong, An, Wu, Kelly, Dykoski and Li2005; Chu et al., Reference Chu, Zhu, Sun, Su, Xie, Zaw and Sein2020; Wang et al., Reference Wang, Li, Zhang, Xu and Wu2022; Zhang et al., Reference Zhang, Zhang, Shang, Chen, Liang, Shao, Zhao, Wu and Wang2026). Coeval climate disruptions have been reported globally and are linked to cultural collapse in several regions (e.g., Li et al., Reference Li, Li, Zheng, Yu, Tang, Li and Cui2018; Griffiths et al., Reference Griffiths, Johnson, Pausata, White, Henderson, Wood, Yang, Ersek, Conrad and Sekhon2020; Groucutt et al., Reference Groucutt, Carleton, Fenech, Gauci, Grima, Scerri, Stewart and Vella2022; Younes and Bakry, Reference Younes and Bakry2022; Helama, Reference Helama2024; Jiang et al., Reference Jiang, Luo, Zhou, Li, Luo, Chen, Liu, Zhao, Zhang and Zhang2025). The associated 4.2 ka Cold ‘megadrought’ Event (c. 4.5–3.88 ka) now formally marks the onset of the Late Holocene (or Meghalayan Stage) and is generally recognised as a significant climatic perturbation (Walker et al. Reference Walker, Head, Berkelhammer, Björck, Cheng, Cwynar, Fisher, Gkinis, Long, Lowe, Newnham, Rasmussen and Weiss2018). The precise cause of the 4.2 ka Cold Event remains elusive. Linkage to an increase in El Niño-La Niña activity across the Pacific is widely considered to be a likely driver (Li et al., Reference Li, Li, Zheng, Yu, Tang, Li and Cui2018; Toth and Aronson, Reference Toth and Aronson2019; Dang et al., Reference Dang, Yu, Tao, Han, Zhang and Jiang2020; Helama, Reference Helama2024; Huang et al., Reference Huang, Yu, Jiang and Jiang2026), though Dang et al. (Reference Dang, Liu, Yu, Tao, Zhang, Wang and Lan2024) suggest that it may be a response rather than a direct triggering mechanism.
Under modern conditions, the El Niño-La Niña system strongly influences RSL within the SCS, with cold-phase ENSO events playing a significant role (Rong et al., Reference Rong, Liu, Zong and Cheng2007; Swapna et al., Reference Swapna, Gan, Lau and Fung2009; Han et al., Reference Han, Stammer, Thompson, Ezer, Palanisamy, Zhang, Domingues, Zhang and Yuan2019; Huang et al., Reference Huang, Yu, Jiang and Jiang2026). Across the Mid-Late Holocene transition, studies point to multiple cold sea surface temperature (SST) intervals indicative of a La Niña-like signal, including a 1 °C drop 4,420–4,312 cal. BP and a 1.6 °C drop 4,189–4,112 cal. BP relative to AD 1887–2007 (Zhou et al., Reference Zhou, Yan, Han, Liu, Yang, Luo, Liu, Wang, Shen, Shi, Dodson and Zhou2022; You et al., Reference You, Li, Zhang, Ge, Zhang, Zhang, Wang, Fan, Wang, Cao and Li2024; Huang et al., Reference Huang, Yu, Jiang and Jiang2026; Zhang et al., Reference Zhang, Zhang, Shang, Chen, Liang, Shao, Zhao, Wu and Wang2026), and a mean SST 25.6 ± 0.4 °C (1σ), c. 0.7 °C cooler compared to a modern baseline of 26.3 ± 0.4 °C (1982–2023) (Huang et al., Reference Huang, Yu, Jiang and Jiang2026). High-resolution datasets also indicate warmer SST excursions in the SCS equated with periods of strengthened ASM and weaker ENSO – including at the onset of the 4.2 ka Event (Dang et al., Reference Dang, Liu, Yu, Tao, Zhang, Wang and Lan2024), which may have had a tripartite ‘arid/non-arid/arid’ structure (You et al., Reference You, Li, Zhang, Ge, Zhang, Zhang, Wang, Fan, Wang, Cao and Li2024; Zeng et al., Reference Zeng, Li, Qing, Ye, Yang, Zhu, Amorosi, Yu and Ouyang2026), highlighting signal complexity in an overall trend towards strengthening ENSO activity and progressive weakening of the ASM after c. 7.5 ka (Yu et al. Reference Yu, Hua, Zhao, Hodge, Fink and Barbetti2010; Zhou et al., Reference Zhou, Yan, Han, Liu, Yang, Luo, Liu, Wang, Shen, Shi, Dodson and Zhou2022).
We acknowledge that correspondence between the Mid-Late Holocene ASM El Niño-La Niña system signal and the intervals illustrated in our model can be considered no more than tentative at this point and recognise the likely contribution of other drivers, including wider changes in the intensity of the Pacific Walker circulation (Yan et al., Reference Yan, Yu, Jiang, Li and Zhao2024), as well as migration of the mid-latitude westerlies and the Intertropical Convergence Zone (ITCZ) (Gao et al., Reference Gao, Jiang, He, Hu, Shen and Zhang2023) and regional climate complexity (e.g., Yang et al., Reference Yang, Zhao, Wei, Liu, Zhang, Long, Li and Xu2023; Hao et al., Reference Hao, Duan, Chen and Cao2025). With these caveats in mind, we make the following observations:
-
1. The period 6–5.4 cal. kBP sees a plateau c. +4.8 m or slight increase in sea level based on dated tidal notches from both sides of the RRD (see Table 2) that finds correspondence to 14C-dated marine terrace formation on Vietnam’s coastal islands at 5 m (Korotky et al., Reference Korotky, Razjigaeva, Ganzey, Volkov, Grebennikova, Bazarova and Kovalukh1995) and associated evidence (c. + 5 m) from sea notch analysis in the Ha Tien–Kien Luong area of the MRD (Tran and Ha, Reference Tran and Ha2020). Globally, this interval falls at the end of the Holocene Thermal Maximum (HTM) (c. 11–5 ka BP) (Renssen et al., Reference Renssen, Seppä, Crosta, Goosse and Roche2012). Its expression in the tropics is reported to have been more muted (Cartapanis et al., Reference Cartapanis, Jaccard, Jonkers and de Vernal2022), though a comparatively warm SST (28.2 ± 0.5 °C) has been reported 6,500–5,550 cal. BP from the southern coast of the SCS (Woodson et al., Reference Woodson, Leorri, Culver, Mallinson, Parham, Thunell, Vijayan and Curtis2017).
-
2. From c. 5.4–4.8 cal. kBP, there is a notable rise in sea level. We take this to be the Mid-Holocene high-stand, though we acknowledge it may represent a late surge within a longer high-stand, as described by Stattegger et al. (Reference Stattegger, Tjallingii, Saito, Michelli, Nguyen and Wetzel2013) for the MRD, where the Mid-Holocene high-stand is considered to have spanned 6.7–5 cal. kBP. Our maximum transgression lies within the range reported in the SCS Mid-Holocene high-stand overview (Zhang et al., Reference Zhang, Lü, Xue, Mo, Wang, Zhao and Hu2025) but is later than the average age of 5,660 cal. BP presented therein. Accounting for VLM, the degree of sea-level rise in our curve also exceeds the average value within SCS reported by those authors (av. + 2–3 m asl.). A strong intrusion of warm waters from the SCS into the Beibu Gulf between Hainan Island and the RRD is reported via fossil diatom analysis (Zhang et al., Reference Zhang, Tomczak, Witkowski, Zhena and Li2023), which may point to the influence of thermal expansion of seawater as a factor influencing this period of sea-level rise, though the window of chronological resolution is quite coarse (5.8–3.8 cal. kBP). A more resolved window finds anomalous SSTs 4,870–4,860 cal. BP from high-resolution Tridacna Sr/Ca records in the northern SCS (Zhou et al., Reference Zhou, Yan, Han, Liu, Yang, Luo, Liu, Wang, Shen, Shi, Dodson and Zhou2022), linked to conditions of El Niño-dominated extreme ENSO variability.
-
3. The comparatively steep sea-level regression phase of c. 1.0 m (4.5–3.9 cal. kBP) finds local support from coral evidence on the east coast of Hainan Island, where a c. 0.73 m drop in sea level is registered c. 5.4–3.4 kBP (U-Th date) (Yue et al., Reference Yue, Tang, Yu and Huang2024). Climatically, this interval coincides with an abrupt weakening of the ASM c. 4.4 kBP (U-Th date) (Wang et al., Reference Wang, Cheng, Edwards, Kong, An, Wu, Kelly, Dykoski and Li2005) and follows the U-series chronology for the ‘early’, ‘top’ and ‘end’ stages of the 4.2 Cold Event, 4,478–3,909 ka, as defined by Dang et al. (Reference Dang, Yu, Tao, Han, Zhang and Jiang2020). Periodic incidence of Porites sp. coral bleaching episodes from the Wenchang fringing reef, Hainan Island (Wang et al., Reference Wang, Yu, Tao, Xu, Yu, Shen and Wang2021), indicates that this cooling trend was not uniform. However, La Niña-dominated ENSO variability during this period (Zhou et al., Reference Zhou, Yan, Han, Liu, Yang, Luo, Liu, Wang, Shen, Shi, Dodson and Zhou2022) raises the possibility that Late Holocene regression (as well as transgression) was linked to temperature-induced fluctuations in the volume of seawater leading to its thermal contraction (or expansion) (Zhang et al., Reference Zhang, Lü, Xue, Mo, Wang, Zhao and Hu2025).
-
4. Subsequent to 3.9 cal. kBP, we see a period of sea-level quiescence centred around a modelled elevation of +3.6 m until c. 3 cal. kBP that also accords with local 14C-dated marine terrace formation +3–4 m (Korotky et al., Reference Korotky, Razjigaeva, Ganzey, Volkov, Grebennikova, Bazarova and Kovalukh1995) and the Ha Tien–Kien Luong sea notch data of c. + 3 m (Tran and Ha, Reference Tran and Ha2020). The Thai Vi lower sequence may also equate to the Quang Xuong sea transgression, proposed by Nguyen Duc (Reference Nguyen Duc2011), though this is still to be determined.
-
5. A further shallow regression phase follows, ending c. 1.7 cal. kBP. Climatically, this corresponds with a period of weakened ASM conditions, 100–500 years in length, centred at 2.7 kBP (U-Th date) (Wang et al., Reference Wang, Cheng, Edwards, Kong, An, Wu, Kelly, Dykoski and Li2005; also see Li et al., Reference Li, Zhang, Li and Zhao2010; Huang et al. Reference Huang, Zeng, Ye and Wei2019): the 2.8 kBP Cold Event. Employing a high-resolution speleothem record from SE China, Zhang et al. (Reference Zhang, Zhang, Shang, Chen, Liang, Shao, Zhao, Wu and Wang2026) identify a two-phase event commencing 2,948 ± 38 BP and ending 2,686 ± 40 BP (U-Th dates) and noting the impact this had on regional cultures (also see Jia et al. Reference Jia, Zhao, Storozum, Shi, Bai, Liu, Hu, Sun, Wang and Li2022). SCS evidence for this event is still nascent but has been identified recently through dinoflagellate cyst assemblages in the northern SCS as lasting c. 2,700–2,400 cal. BP, during which time summer SSTs reportedly decreased by 1.5 °C (Li et al., Reference Li, Pospelova, Liu, Francois, Wu, Mertens, Saito, Zhou, Song and Xie2021; also see Kong et al., Reference Kong, Zong, Jia, Wei, Chen and Liu2014). This latter study did not, however, observe the 4.2 ka Cold Event; and as such, we cannot rule out the possibility of different causal drivers for the two RSL regression intervals reported here.
Table 2.Calibration results (R – rice) of Holocene dates and elevations adjusted to isostasy
Table 2. Long description
The table contains 10 columns: Lab I D, Site, Material, 14 C age, S D, 95% range (cal. B P), 95% hpd ranges (cal. B P), q 50, m asl, and m cal asl.
Key data rows include:
* U B A 44780: Thay Vi 1 lower, Ostreidae, 2,317 14 C age, 3.5 m asl.
* U B A 44781: Thay Vi 1 lower, Meretrix, 2,536 14 C age, 3.5 m asl.
* U B A 44782: Thay Vi 1 lower, Charcoal, 2,827 14 C age, 3.5 m asl.
* U B A 45068: Thay Vi 1 upper, Ostreidae, 4,275 14 C age, 5.6 m asl.
* U B A 54213: Than Boat Wharf, Ostreidae, 2,931 14 C age, 3.7 m asl.
* U B A 54214: Thay Vi 1 lower, Ostreidae, 3,326 14 C age, 3.3 m asl.
* W K 8255: Halong Bay (Loc. 1), Ostreidae, 4,990 14 C age, 4.9 m asl.
* W K 8256: Halong Bay (Loc. 2), Ostreidae, 4,100 14 C age, 4.6 m asl.
* W K 8257: Halong Bay (Loc. 2), Ostreidae, 3,820 14 C age, 4.1 m asl.
* W K 8258: Halong Bay (Loc. 3), Ostreidae, 4,050 14 C age, 4.9 m asl.
* W K 8259: Halong Bay (Loc. 3), Ostreidae, 3,280 14 C age, 4.3 m asl.
* W K 8260: Tam Coc (Loc. 1), Ostreidae, 2,280 14 C age, 3.5 m asl.
* W K 8262: Tam Coc (Loc. 2), Ostreidae, 4,770 14 C age, 4.8 m asl.
* W K 8265: Tam Coc (Loc. 3), Ostreidae, 4,420 14 C age, 5.4 m asl.
* W K 8267: Halong Bay (Loc. 1), Ostreidae, 5,040 14 C age, 5.4 m asl.
* W K 8268: Halong Bay (Loc. 1), Ostreidae, 5,300 14 C age, 5.1 m asl.
* W K 8269: Halong Bay (Loc. 2), Ostreidae, 3,000 14 C age, 4.0 m asl.
Elevations (m cal asl) range from 3.1 to 5.4, with the highest values associated with older 14 C ages at Thay Vi 1 upper and Tam Coc (Loc. 3).
Limitations
Chronological uncertainty and measured vertical uncertainty were propagated simultaneously through the Monte Carlo ensemble reconstruction. However, the confidence envelope shown in Figure 3 does not incorporate uncertainty associated with the indicative meaning of the sea-level indicators. Consequently, the uncertainty ribbon should be interpreted as representing the combined effects of chronological and measurement uncertainty, while the total uncertainty associated with the relationship between notch elevation and former sea level is likely to be greater.
A remaining limitation is the absence of a quantitatively constrained indicative range for oyster-bearing tidal notches, as advocated by Rovere et al. (Reference Rovere, Stocchi and Vacchi2016a, Reference Rovere, Raymo, Vacchi, Lorscheid, Stocchi, Gómez-Pujol, Harris, Casella, O’Leary and Hearty2016b). While this does not affect the propagation of chronological and measurement uncertainty within the model, it limits the precision with which notch elevations can be related to former sea level. Future surveys of modern oyster-bearing tidal notches in Ha Long Bay would provide the data needed to establish a locally constrained indicative range and further refine relative sea-level reconstructions for northern Vietnam.
Conclusion
The findings presented here provide the first absolute dating of Pleistocene tidal sea notches associated with the poorly resolved Vinh Phuc transgression and help refine the chronological record of sea-level change in the southern RRD during the Mid–Late Holocene. We observe potential correspondence between Mid–Late Holocene sea-level change, SST trends and extreme ENSO-related 4.2 and 2.8 ka Cold Events. Detailed testing and modelling to investigate this systematically is outside the scope of this article. However, under a future scenario of heightened ENSO frequency and severity (e.g., Cai et al. Reference Cai, Santoso, Collins, Dewitte, Karamperidou, Kug, Lengaigne, McPhaden, Stuecker, Taschetto, Timmermann, Wu, Yeh, Wang, Ng, Jia, Yang, Ying, Zheng, Bayr, Brown, Capotondi, Cobb, Gan, Geng, Ham, Jin, Jo, Li, Lin, McGregor, Park, Stein, Yang, Zhang and Zhong2021; Coquereau et al. Reference Coquereau, Sévellec, Huck and Fedorov2025) and growing emphasis on the value of the palaeo-ENSO system to better understand those conditions (e.g., Jirapinyakul et al., Reference Jirapinyakul, Nudnara, Punwong, Nohal, Englong, Phujareanchaiwon, Yamoah and Choowong2023; Huang et al., Reference Huang, Yu, Jiang and Jiang2026; Zeng et al., Reference Zeng, Li, Qing, Ye, Yang, Zhu, Amorosi, Yu and Ouyang2026), we echo the argument of Zhang et al. (Reference Zhang, Lü, Xue, Mo, Wang, Zhao and Hu2025), that further investigation of the potential relationship between hydroclimate conditions and RSL is warranted. On the basis of data presented herein, we argue such work may have particular value to isolate the likelihood of future ENSO-linked cold events and their capacity to instigate conditions associated with sea-level regression. The potential that such conditions might result in intervals of slower coastal inundation or even quiescence would offer opportunity for positive exploitation by policy-makers and coastal managers to allow stabilisation initiatives that are sensitive to RSL, such as mangrove or saltmarsh rehabilitation (e.g., Gilman et al., Reference Gilman, Ellison, Duke and Field2008; Kirwan et al., Reference Kirwan, Guntenspergen, D’Alpaos, Morris, Mudd and Temmerman2010; Krauss et al., Reference Krauss, MeKee, Lovelock, Cahoon, Saintilan, Reef and Chen2013; Dang et al., Reference Dang, Reid and Kumar2022; Ouyang et al., Reference Ouyang, Connolly and Lee2022; Tiggeloven et al., Reference Tiggeloven, van Zelst, Mortensen, van Wesenbeeck, Worthington, Spalding, de Moel and Ward2026) to take hold, finding advantage rather than disadvantage in climate instability towards building effective protection for the region’s coastal communities and especially its heavily populated mega-deltas.
Open peer review
To view the open peer review materials for this article, please visit http://doi.org/10.1017/cft.2026.10050.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/cft.2026.10050.
Acknowledgements
The authors also thank the two anonymous reviewers for their constructive comments that have appreciably enhanced the article.
Author contribution
The authors made the following contributions. Conceptualization: T.K., R.R.; Funding acquisition: R.R.; Data curation: T.K.; Formal Analysis: T.K., M.B.; Investigation: T.K., R.R., K.S.P., S.O., E.H.; Methodology: T.K., R.R., M.B.; Project administration: T.K., R.R.; Software: T.K.; Visualisation: T.K.; Writing - original draft preparation: T.K., R.R., S.O.; Writing - review and editing: T.K., R.R., M.B.
Competing interests
The authors declare no competing interests.
Data availability statement
Principal data sets have been made available at https://zenodo.org/records/21279507 (DOI:10.5281/zenodo.21279506):
-
1. R-script for the SL curve, including input data used in its construction (Sea-level analysis adj asl.R)
-
2. Notch survey data (Notch metrics.xlsx)
-
3. Illustrations showing where measurements were taken at two representative notches can be accessed (Annex 1.docx)
Funding statement
The authors would like to express their gratitude to the Xuan Truong Corporation Enterprise, Ninh Binh and the Tràng An Landscape Complex Management Board, Ninh Binh, for their financial and logistical support. This work forms part of the SUNDASIA Project - Human Adaptation to Coastal Evolution: Late Quaternary evidence from Southeast Asia, the principal UK funding for which came from a Global Challenges Research Fund (GCRF) AHRC grant number AH/N005902/1 and a UKRI Covid-19 Grant Extension Allocation Fund, both to RR.
Open access
Comments
No accompanying comment.