Introduction

Over the 21st century, continuing relative sea level (RSL) rise poses an existential threat to low-lying islands, coastal deltas and lowlands and their respective environments and populations1,2,3,4,5. The implications of rising sea level, however, will be spatially disproportionate with equatorial and tropical latitudes of Asia facing the greatest impacts where substantial populations live below projected sea- and flood-levels1,6,7. Singapore, a small (~730 km2) equatorial island sustaining ~5.9 million people in Southeast Asia (Fig. 1), is highly exposed to sea-level rise with centres of population, industry, urban, and transport infrastructure within 2 m elevation of present sea level8,9. Accurate sea-level projections are required to implement appropriate adaptation, mitigation, and engineering strategies8,10.

Fig. 1: Southeast Asia study region and location of relative sea level (RSL) datasets.
figure 1

a RSL predictions at the last glacial maximum 26 thousand years before present from glacial isostatic adjustment model consisting of the ICE-6G_C ice model37,38 and HetM-LHL140 3D Earth model124,125. Location of paleo river systems113,115 and geological RSL reconstructions from the Sunda Shelf26,30 and Mapur Island, Indonesia32. b Singapore study area showing location of geological RSL reconstructions31 and tide gauge stations15 shown in Fig. 2 and detailed in Table 1. Present-day simplified bathymetry depth and approximate distribution of coastal habitats adapted after refs. 92 and 102, respectively. Black dashed line in panel a indicates the approximate boundary of the Sundaland block and the dark blue barbed line indicates the location of major subduction zones (after ref. 35). Location of active and potentially active volcanoes after ref. 120. Numbering of tide gauge stations in panel b and associated names are provided in Table 1.

Full size image

A prerequisite for sea-level projections is understanding the relationship between changes in the climate and changes in sea level11,12,13. Most instrumental sea-level records, however, are temporally limited to the mid-to-late twentieth and early twenty-first centuries14,15 and capture a sea-level/climate relationship during which anthropogenic forcing dominates4,16,17. This is especially true in Singapore, where near-complete instrumental records only began in the early 1970s (Fig. 2; Table 1). Reconstructions of RSL change based on geological proxies can provide complementary archives, demonstrating the longer-term response of sea level over centuries to millennia to a wider range of climate forcing mechanisms, unforced climate variability, and boundary conditions12,18,19,20,21. Both geological reconstructions and instrumental records, however, share fundamental challenges in understanding spatial and temporal variability driven by processes that cause regional sea level to deviate from the global mean11,13,22,23.

Fig. 2: Singapore tide gauge records.
figure 2

Annual relative sea level (RSL) change recorded by Singapore tide gauge stations15 shown in Fig. 1b and detailed in Table 1. Thicker lines (stations 1–4) represent tide gauge stations used to construct an averaged instrumental RSL record discussed in text.

Full size image
Table 1 Summary of Singapore tide gauge stations.
Full size table

Geological reconstructions from equatorial and tropical latitudes have provided detailed RSL histories since the last glacial maximum (LGM) ~ 26 thousand years (kyr) to ~19 kyr before present (BP)24,25,26,27. These reconstructions have been central to our understanding of global mean sea level (GMSL), determine ice-volume changes and constrain geophysical models of the glacial isostatic adjustment (GIA) process19,28. Understanding the timing, magnitude and driving processes of past RSL changes provides a useful background to contextualise sea-level projections12 that is important in the context of effective coastal risk management strategies for emerging contingencies5,29.

Here we present a synthesis of the evolution of RSL in equatorial Southeast Asia combining geological reconstructions and instrumental records of RSL change from the Sunda Shelf and Singapore. This region is important because of the availability of detailed RSL datasets spanning the LGM (~21.5 kyr BP) through the last deglacial transition (LDT, ~18–11 kyr BP)26,30, Holocene (~11 kyr BP)31 and present instrumental period (20th and 21st century)15,32 that are not unduly affected by vertical land motion processes33,34,35. We apply an Error-In-Variables Integrated Gaussian Process (EIV-IGP) model36 to quantify magnitudes and rates of RSL change through time and assess and discuss paleogeographic changes across the Sunda Shelf in response to RSL change using a GIA model37,38. We conclude with the latest Intergovernmental Panel on Climate Change (IPCC) sea-level rise projections to 2150 for Singapore4,39 and use the geological past to quantitatively provide probability perspectives when projected rates of sea level were last exceeded. We show that projected rates of RSL rise under a moderate emission scenario were only last exceeded (at least 99% probability) during rapid ice mass loss events of the last deglacial period.

Results and discussion

Last glacial maximum to present relative sea-level changes

The geological RSL reconstructions from the former North Sunda River on the central Sunda Shelf (Fig. 1a) provide a sea-level index point (SLIP) dataset (n = 33) that constrain RSL between ~21.5 kyr and ~13.3 kyr BP with an average vertical and age uncertainty of ±5 m (2σ) and ±393 years (2σ), respectively (Fig. 3a). The vertical uncertainty incorporates spatial differences in GIA between the Sunda Shelf and Singapore. Application of the EIV-IGP model shows RSL rose from a lowstand of −121.1 m at 20.7 kyr BP to −112.3 m at ~19 kyr BP at rates of RSL rise up to 7 ± 5.8 mm/yr (Fig. 3d, e). The rate of RSL subsequently slowed to an average ~3 mm/yr as RSL continued rising to −105.2 m at ~16 kyr BP. Between ~16 kyr and ~13 kyr BP, RSL rose to −~70 m with a cluster of SLIPs (n = 17) between ~15 kyr and ~14 kyr BP associated with the meltwater pulse 1A (MWP1A) event (Fig. 3d). The average (250-time interval) rate of RSL rise between 16 kyr and 13 kyr BP reached 15.4 ± 8.2 mm/yr (Fig. 3e). An absence of SLIP data constraining the LDT into the early Holocene is reflected by the increased uncertainty of RSL changes during this period.

Fig. 3: Geological reconstructions and instrumental records of relative sea level (RSL) change from the Sunda Shelf and Singapore.
figure 3

Sea level index points (SLIPs) showing RSL change since (a) the last glacial maximum and deglacial transition26,30, b Holocene31, and c 20th and 21st century15,32 (note axis change among graphs). Application of the Error-in-Variables Integrated Gaussian Process (EIV-IGP) model36 to a combined sea-level dataset showing (mean and 95% credible interval) d magnitudes and e rates of RSL change. Inset panels shows magnitudes and rates of RSL change for the combined geological and averaged instrumental dataset for the 20th and 21st century. EIV-IGP model mean dashed line in (d) represents time periods with an absence of SLIPs and increasing RSL change uncertainty.

Full size image

The geological RSL reconstructions from mainland Singapore and neighbouring islands (Fig. 1b) provide a SLIP dataset (n = 64) that constrain RSL between ~9.4 kyr and ~1.2 kyr BP with an average vertical and age uncertainty of ±1.1 m and ±487 years, respectively (Fig. 3b). Most SLIPs (n = 59) are concentrated towards the early- to mid-Holocene and shows a continuous rise in RSL. The EIV-IGP model shows RSL rose from −20.6 m at 9.4 kyr BP to −0.25 m at ~7 kyr BP (Fig. 3d) at a maximum rate of 15.2 ± 3.5 mm/yr (Fig. 3e). The rate of RSL rise subsequently slowed as RSL continued to rise and reached a mid-Holocene highstand of ~4.6 m at 5.2 kyr BP. SLIPs constraining the mid- to late-Holocene transition are sparse (n = 4) but suggest RSL fell below present level to −2.2 m between ~2.5 kyr and ~0.25 kyr BP at a rate of −1 mm/yr.

The geological RSL reconstructions from Mapur Island, Indonesia (Fig. 1a) provide a SLIP dataset (n = 16) that constrain RSL between 1915 and 2012 CE with an average vertical and age uncertainty of ±0.1 m and ±1.4 years, respectively (Fig. 3c). The averaged tide gauge record for Singapore provides annual instrumental measurements of RSL change between 1972 and 2020 CE (Fig. 3c). The EIV-IGP model applied to the combined dataset shows RSL rose 0.15 m between 1915 and 2020 CE (Fig. 3d) at rates of RSL rise increasing from 1.7 ± 1 mm/yr to 2.2 ± 0.6 mm/yr (Fig. 3e).

Paleogeographic changes

Rising RSL dramatically altered the paleogeographic landscape of Southeast Asia (Fig. 4a). At the LGM, the area of the land exposed on the Sunda Shelf was greater (relative to present-day topography) by ~2.3 million km2 (Fig. 4b). During the LDT, RSL rise flooded the Sunda Shelf at an average lateral shoreline migration rate of 57 m/yr and by 16 kyr BP, exposed land area had reduced to ~2.1 million km2. The rapid increase in RSL rise during MWP1A reduced land area to ~1.4 million km2 and lateral shoreline migration rates increased to ~335 m/yr. By the early Holocene (~10 kyr BP), land bridges between continental and insular islands of Southeast Asia such as Sumatra and Borneo had been severed and exposed land area was further reduced to ~336 ka km2 at a maximum rate of ~15 m/yr. During the mid-Holocene (~6 kyr BP), rising RSL reaching above present level decreased exposed land area to a minimum of −88 ka km2 before subsequently increasing again thereafter as RSL fell towards present reaching a near present-day coastline configuration by ~2–1 kyr BP.

Fig. 4: Land area change since the last glacial maximum (LGM).
figure 4

a Paleotopographic maps showing land area change as rising sea level flooded the Sunda Shelf after the LGM 26 thousand years (kyr) before present (BP) to 20 kyr BP and then at 2 kyr intervals thereafter. b Calculated land area difference relative with present-day topography132 and ice equivalent sea level from 26 kyr BP based on the ICE-6G_C global ice history model37,38. Paleogeographic changes continued as ice equivalent sea level ceased during the mid-Holocene because of spatially variable GIA processes as discussed in text. Red transect line A-A’ in 26 kyr BP panel used to calculate lateral transgression rates discussed in text.

Full size image

Last glacial maximum and deglacial transition

Geological reconstructions from the Sunda Shelf and Singapore reveal the response of sea level to climate forcing as the Earth transitioned from glacial to interglacial conditions. Between ~26 kyr and ~19 kyr BP, global ice sheet complexes grew to their maximal extent19,27,40 and GMSL reached a lowstand of ~120 to ~130 m below present level (Fig. 5a, b). Atmospheric CO2 concentrations were between ~188 and ~194 ppm and global mean surface temperature (GMST) cooler (relative to 1850–1900 CE) by 5°–7°C41,42.

Fig. 5: Geological relative sea level (RSL) reconstructions since the last glacial maximum (LGM) from equatorial and tropical latitudes.
figure 5

a Global map showing RSL predictions at 26 thousand years (kyr) before present (BP) from glacial isostatic adjustment model consisting of the ICE-6G_C ice model37,38 and HetM-LHL140 3D Earth model124,125. Location of geological reconstructions constraining RSL changes since the LGM within equatorial and tropical latitudes (white dashed lines). b Geological reconstructions of RSL change from sites in (a) since 22 kyr BP including the Sunda Shelf and Singapore (green circles, this study), Barbados (yellow triangles28,54), Tahiti (dark blue cross-hairs25,49), Bonapate Gulf (purple diamonds27), Great Barrier Reef (red crosses, Hydrographer’s Passage134) and Papua New Guinea (light blue squares135,136). Each geological reconstruction are derived using sea-level indicators (e.g., coral reefs and mangroves) that are chronologically constrained using a variety of dating techniques (e.g., U-Th and C14 dating). The reader should see original publications for discussions on their associated vertical and temporal uncertainties, respectively. No additional corrections for local vertical land motion have been applied to the geological reconstructions and the reader should see original publications for discussions on their regional setting. Bright green outline with white shading in panel a represents spatial extent of ice sheets at the LGM. Dark grey vertical shading in (b) represents approximate global timing and name of late Quaternary rapid increases in global mean sea level discussed in text47,70. MWP meltwater Pulse, EHSLR early Holocene sea level rise.

Full size image

Increasing northern summer insolation began the onset of the LDT towards the early Holocene12,43 as atmospheric CO2 concentrations increased to ~270 ppm and GMST rose 1°–1.5 °C kyr42. In response, GMSL rose as ~50 million km3 land-based ice was transferred to the global ocean primarily from deglaciating Northern Hemisphere ice masses27,40,44,45. This ice melt caused RSL on the Sunda Shelf to rise ~93 m over ~12,000 years from 21.5 kyr to 9.5 kyr BP at an average rate of ~7.6 mm/yr.

Superimposed on long-term secular rising GMSL have been several periods of short-term rapid increases46,47. An initial rapid increase in GMSL between 19.5 kyr and 18.8 kyr BP (MWP1Ao, Fig. 5b) caused RSL on the Sunda Shelf to rise ~8 m at rates of RSL rise up to 7 mm/yr (Fig. 3d, e). Similar rapid increases in RSL following termination of the LGM have been reported from geological reconstructions in both low27 (Fig. 5a, b) and high48 latitude settings.

A second rapid increase in GMSL between 14.8 kyr and 13 kyr BP (MWP1A, Fig. 5b) caused RSL on the Sunda Shelf to rise ~32 m at rates of RSL rise up to ~15 mm/yr (Fig. 3d, e). This averaged rate is conservative compared to annual rates of RSL rise of ~40 mm/yr26,47,49. Rising GMSL during MWP1A was the most rapid of the last deglacial period and was driven by the sudden influx of meltwater as deglaciation of Northern Hemisphere ice sheets continued and ice-dammed and subglacial lakes drained increasing ocean volume by ~470,000 km3 19,47,50,51. Geological reconstructions to support MWP1A are global in scale with widespread evidence within equatorial and tropical latitudes (Fig. 5 and references therein) that are also corroborated at higher latitudes52,53.

An additional rapid increase in GMSL during the LDT is postulated between ~11.5 kyr and ~11 kyr BP (MWP1B, Fig. 5b). Evidence from coral reconstructions in Barbados suggests RSL rose ~13 m at rates of RSL rise increasing ~20–40 mm/yr24,25,54 while Gargani55 reported rates of RSL rise may have been even faster at ~65 mm/yr. The global importance of MWP1B, however, remains contested19 and is currently unresolved in other equatorial and tropical reconstructions with an equivalent change in RSL not recorded on the Sunda Shelf and Tahiti26,49. Indeed ref. 19 suggests elevated rates of RSL increasing at ~16.5 mm/yr occurred just prior to MWP1B between 12.5 ka and 11.5 ka BP after the Younger Dryas. While mangrove and intertidal data (n = 5) from the Southern Vietnam Shelf provide evidence of RSL change between ~-65 m and ~-50 m between ~13 kyr and ~11 kyr BP26 (Supplementary Fig. 1), they were not included here due to their spatial distance from North Sunda River and Singapore SLIP datasets and their proximal location to the Mekong River Delta and its associated influence of subsidence19,56.

The rise in GMSL after the LGM had profound impacts on shelf margins57 including the Sunda Shelf and transformed the palaeogeography of the region58,59,60. The low shelf gradient allowed sea level to rapdily transgress laterally increasing from an average rate of ~57 m/yr to ~335 m/yr during MWP1A. Rising RSL submerged coastal landscapes60,61 and segregated insular islands of Southeast Asia from the continental mainland dislocating flora and fauna migration routes59,62. Indeed, evidence from whole-genome sequencing datasets suggests rising RSL, particularly during MWP1A, played an important role in the spatial distrbution of modern human demography in Asia as local populations became segregated63. In the Singapore Strait, the land bridge that existed during low GMSL was severed as rising seas flooded across sills to the east from the South China Sea and to the west from the Malacca Strait when RSL reached ~−30 m toward the end of the LDT64 (Fig. 1b). The flooding of the Sunda Shelf also altered the interchange of water between the Indian Ocean and South China Sea distrupting regional oceanographic and atmospheric climate systems65,66,67.

Holocene

Climate forcing during the Holocene up until the pre-industrial Common Era (~1850 CE) was relatively mild compared to the preceding LDT. Atmospheric CO2 concentrations were between ~260 and ~285 ppm42 and variability in GMST was reduced68 showing a slight but steadily warming trend of ~0.25–0.5 °C41,69. Despite relative climate stability, GMSL continued to rise tens of metres during the early Holocene as Northern Hemisphere ice sheets entered their final stages of disintegration and coastal ice streams broke19,70,71,72.

In Singapore, RSL rose ~21 m between 9.5 kyr and 7 kyr BP at rates of RSL rise up to ~15 mm/yr (Fig. 3d, e). Bird et al.73,74 suggested rapid RSL rise during the early Holocene in Singapore was temporally punctuated by a near cessation in RSL rate between 7.8 kyr and 7.4 kyr BP before continuing to rapidly increase again thereafter. Lambeck et al.19, however, noted that the near-zero RSL rise during this period possibly reflects local processes because of an absence of similar trends at the global scale. Furthermore, Chua et al.31 concluded accurate verification of oscillating RSL in Singapore is precluded by large vertical and temporal scatter of SLIP data following their standardisation75.

Rising RSL reached near present level by ~7 kyr BP and continued rising to a mid-Holocene highstand that is characteristic of far-field regions distal from ice sheets and driven by regional hydro-isostatic processes when meltwater input decreased21,22. The magnitude and timing of the highstand varies around the Sunda Shelf 31,76,77 and in Singapore, RSL reached ~4.6 m at ~5 kyr BP. Falling RSL from the mid-Holocene highstand was driven by both hydro- and glacio-isostatic loading of the Earth’s surface (equatorial ocean syphoning and continental levering)78 and rotational feedback79. Late-Holocene SLIPs show RSL below present at −~2 m between ~2.5 kyr and ~0.5 kyr BP. Evidence to support RSL below present during the late Holocene is corroborated by mangrove SLIPs from Peninsula Malaysia which show RSL at ~−0.7 m ~ 0.8 kyr BP33,80,81,82.

The rise and fall in RSL during the Holocene continued to alter the palaeogeography of the region. Rapid RSL rise during the early Holocene flooded lowlands surrounding Singapore and in the Johor Strait segregating Singapore from Peninsula Malaysia64,83. As the rate of RSL rise declined below ~7 mm/yr after ~8.5 kyr BP, widespread development of mangrove forests commenced as mangroves began to maintain their vertical position through sediment accretion84. The rise in RSL above present during the mid-Holocene continued to encroach and submerge low elevations of Singapore decreasing the terrestrial land area. The marine sediments that were deposited were subsequently weathered and partially eroded by sub-aerial processes as RSL fell during the late Holocene increasing land area as previous shorelines became progressively exposed85.

Twentieth and twenty-first century

Instrumental records cover changes in Earth’s climate driven by anthropogenic forcing86. Atmospheric CO2 concentrations have increased to ~424 ppm in 2023 CE87, which is unprecedented in at least the last two million years, and GMST rose ~1.1 °C between 1850–1900 and 2011–2020, which is warmer than any multi-centennial interval during the LDT42. The response of GMSL to this forcing has so far been an increase in rate from ~1.4 mm/yr4,88,89,90 between 1901 and 1990 to 3.3 mm/yr between 1993 and 2018 that was driven by accelerating land-ice losses and thermal expansion that vary in relative contribution depending on time period analysed4. For example, sea-level budgets for the 20th and 21st century reveal thermal expansion accounted for 32% of GMSL rise between 1901 and 1990 compared to 46% between 1993 and 20184. Long-term (i.e. >50 years) instrumental records coupled with overlapping geological reconstructions have also demonstrated rising GMSL during the 20th century was faster (P ≥ 0.999) than the proceeding 3 kyr17, with a timing of emergence above background variability centred on 1863 CE91. In Southeast Asia, geological reconstructions using mangrove sediments from Peninsula Malaysia have suggested an increase in RSL rate from 1.26 mm/yr to 3.2 mm/yr after 1900 CE80.

The addition of geological reconstructions from Mapur Island, Indonesia provides an important methodological step toward further extending instrumental records of RSL change in equatorial and tropical latitudes such as Singapore32. The combined dataset shows RSL rose 0.15 m between 1915 and 2020 CE increasing from a rate of ~1.7 mm/yr to ~2.2 mm/yr. Singapore’s temporally limited instrumental records provide only a brief history of RSL changes with variability on different timescales92,93,94. On daily to seasonal timescales, sea-level variability is influenced by meteorological forcing across the region95 and prevailing wind directions during Northern Hemisphere and Southern Hemisphere monsoon systems can cause sea level to vary up to ±30 cm92. Indeed, storm surges driven by winds over the South China sea can result in localised coastal flooding when coinciding with high spring tides96. On interannual timescales, sea-level variability is dominated by the El Niño–Southern Oscillation and during El Niño and La Niña events, changes in sea-surface height can vary by up to ±5 cm, with lower sea-surface height observed during El Niño events92. On decadal and inter-decadal timescales, sea-level variability in Singapore and wider Southeast Asia is influenced by basin-scale climate modes including the Pacific Decadal Oscillation and Interdecadal Pacific Oscillation97,98.

Coastal landscape changes following British colonial establishment in 1819 and during the 20th and 21st century in Singapore largely reflect anthropogenic modifications to accommodate industrial and urban development99. Land reclamation projects expanded land area ~25% from 581.5 km2 in 1960 to 733.2 km2 in 202299 that significantly reduced coastal habitat extent100,101,102. Between 1922 and 2011, tidal flats and coral reefs reduced in area from 33 km2 to 5 km2 and 32 km2 to 9.5 km2, respectively. Furthermore, the damming of mangrove-fringed estuaries to create freshwater reservoirs resulted in a 91% decrease in mangrove forest extent, reducing in area from 75 km2 to 6.4 km2 101,102.

Perspectives of future sea level

The magnitude RSL rise since the end of LGM through the LDT, Holocene and towards the present demonstrates the long-term commitment and sensitivity of sea levels to climate forcing on timescales of centuries to millennia12. While the extent and grounding line of ice sheets were spatially different at the LGM compared to present conditions4,103 (Fig. 5a), it is virtually certain (at least 99% probability) Greenland and likely (at least 66% probability) Antarctic ice sheets will continue losing mass throughout the 21st century4. Indeed, rising GMSL initiated during the 20th century is virtually certain to continue even if CO2 emissions are drastically reduced due to the lagging integrated response time of deep ocean heat uptake and ice sheets12,104,105,106,107. Over the next 2000 years, GMSL is committed to rise 2–3 m if GMST is limited to 1.5 °C warming, 2–6 m if limited to 2 °C warming and 19–22 m with 5°C of warming4. Projections of GMSL rise are consistent with geological reconstructions during past warm climate periods when GMST were higher108. For example, during the last interglacial ~126 kyr to 116 kyr BP, GMSL was ~1–5 m higher than present when GMST were just 0.5–1 °C warmer than today4,108,109,110.

Future sea-level rise in Singapore will primarily be caused by global increases in ocean mass and volume associated with meltwater input from land-based ice sheets and glaciers and thermal expansion to warming temperatures4. Faster-than-projected disintegration of marine ice shelves may also exacerbate sea-level rise through marine ice cliff instability processes111,112 for which there is low confidence4. In contrast, the impact of vertical land motion processes will be negligible due to the tectonic stability and overall low subsidence rates throughout the central Sunda Shelf including Singapore33,34,64.

Under the very low emissions shared socioeconomic pathway (SSP)1–1.9 scenario, future RSL rise (relative to a 1995–2014 baseline) in Singapore will increase 0.58 m (likely range of 0.31–0.93 m) at a rate of 3.4 mm/yr (1.2–6.2 mm/yr) by 2150 (Fig. 6a, b, Supplementary Table 1). Magnitude 0.5 m and 1.0 m thresholds under SSP1–1.9 are expected to be surpassed by 2127 (likely range of 2084–>2300) and 2279 (2162–>2300), respectively. The geological past provides probability perspectives to when equivalent rates of RSL rise were last exceeded (Fig. 7). Rates of RSL rise increasing at greater than 3.4 mm/yr were very likely (at least 90% probability) between ~20 kyr and ~19.5 kyr BP and about as likely as not (between 33 and 66% probability) between 18.75 kyr and 16 kyr BP (Fig. 7a). Rates of RSL rise exceeding 3.4 mm/yr were virtually certain to have occurred between 15.25 kyr and 13.5 kyr BP and between 9.5 kyr and 8.25 kyr BP. During the last ~5 kyr, it is unlikely (less than 33% probability) rates of RSL rise exceeded 3.4 mm/yr.

Fig. 6: Future projections of relative sea-level (RSL) rise to 2150 for Singapore.
figure 6

a Magnitudes and b rates of RSL rise to 2150 for Singapore showing 50th percentile (solid line) and likely (17th–83rd percentile; shading) ranges for Shared Socioeconomic Pathways (SSP) 1-1.9, SSP2-4.5 and SSP5-8.5 scenarios. Red dashed line represents the 83rd percentile projection for SSP5-8.5 low confidence (LC) scenario. See Supplementary Table 1 for full decadal values to 2150. Projections are relative to 1995–2014 baseline period for the Sultan Shoal tide gauge station4,39 (Fig. 1b). Grey horizontal dashed lines in panel a used to reflect timing of exceedance of 0.5 m and 1.0 m magnitude thresholds discussed in text.

Full size image
Fig. 7: Geological probability perspectives of future relative sea-level (RSL) rise.
figure 7

Probability (%) of rates of RSL rise during the geological past based on 250-year average bins exceeding a 3.4 mm/yr, b 7.3 mm/yr and c 11 mm/yr that reflect projected (50th percentile) rates of RSL rise by 2150 under SSP1-1.9, SSP2-4.5 and SSP5-8.5 scenarios, respectively. Grey horizontal dashed lines indicate likelihood ranges discussed in text.

Full size image

Under the moderate emissions SSP2-4.5 scenario, future RSL in Singapore will increase 0.95 m (0.62–1.4 m) at a rate of 7.3 mm/yr (4.4–11.6 mm/yr) by 2150 (Fig. 6a, b, Supplementary Table 1). Magnitude 0.5 m and 1.0 m thresholds under SSP2-4.5 are expected to be surpassed by 2092 (2072–2125) and 2157 (2114–2264), respectively. Rates of RSL rise exceeding 7.3 mm/yr were about as likely as not between ~20 kyr and ~19.5 kyr BP and unlikely up to 15.75 kyr BP (Fig. 7b). Rates of RSL rise exceeding 7.3 mm/yr were virtually certain between 15 kyr and 14 kyr BP and between 9.25 kyr and 8.75 kyr BP. During the last ~8 kyr, it is unlikely rates of RSL rise exceeded 7.3 mm/yr.

Under the very high emissions SSP5-8.5 scenario, future RSL rise in Singapore will increase 1.37 m (0.94–2.02 m) at a rate of 11 mm/yr (6.6–17.6 mm/yr) by 2150 (Fig. 6a, b, Supplementary Table 1). Magnitude 0.5 m and 1.0 m thresholds under SSP5-8.5 are expected to be surpassed by 2080 (2066–2097) and 2119 (2096–2159), respectively. Rates of RSL rise exceeding 11 mm/yr were very unlikely (less than 10% probability) between 21.5 kyr and 15.75 kyr BP while it very likely occurred between 15.25 kyr and 14.75 kyr BP and virtually certain at ~9 kyr BP (Fig. 7c). During the last ~8.25 kyr, it is extremely unlikely (less than 5% probability) rates of RSL rise exceeded 11 mm/yr.

Considering ice-sheet processes in which there is currently low confidence in the scientific ability to model raises the potential sea-level rise contributions, particularly under high emissions scenarios. The 83rd percentile projection for SSP5-8.5 including low confidence marine ice cliff instability processes reaches 5.3 m and 111 mm/yr by 2150 (Supplementary Table 1) and moves the crossing of a 1.0 m magnitude threshold as soon as 2076 (Fig. 6a). Such high rates of RSL rise have no precedent in the last ~21.5 kyr.

The geological past also provides useful perspectives to future topographic changes. While present and future coastal configurations are anthropogenically modified and future RSL rise will be spatially variable across Southeast Asia, a projected rise of 0.95 m by 2150 under SSP2-4.5 (Supplementary Table 1), for example, caused an equivalent reduction in the paleogeographic landscape of −19.7 ka km2 during the mid-Holocene. A projected RSL rise of ~5 m by ~2150 under the high emissions low confidence scenario (Supplementary Table 1) caused the paleogeographic landscape to reduce −87.6 ka km2 during the mid-Holocene (Fig. 4b) and illustrates the vulnerability of equatorial and tropical coastal regions and population densities to future RSL rise1.

Methods

Regional setting and sea-level datasets

Singapore is located at the centre of the Sunda Shelf, the largest tropical shelf margin in the world, which extends ~800 km laterally over low geomorphic gradients towards the deep-ocean South China Sea (Fig. 1). During glacial periods of low sea level, present-day islands of Borneo, Java, Sumatra, and Singapore were connected with continental Asia59,60. This sub-aerially exposed land mass, known as Sundaland113,114, was dissected by several major river systems that drained the surrounding exposed shelf and its hinterland113,115 (Fig. 1a) and have preserved deposits pertaining to the paleoenvironmental evolution of the region58,59,60,115,116. Plate tectonic movements and the northward moving Indo-Australian Plate colliding with the Eurasian Plate117,118 causes major seismicity from tectonics and volcanism to be concentrated towards plate exterior subduction and collision zones35,119,120 (Fig. 1a). The interior of Sundaland, however, is virtually free of seismicity and volcanic activity35,119,120 and vertical deformation is considered minimal during the Quaternary33,121. Decade long GPS velocities support this inference showing a very low rate of shallow seismicity35 driven by transient dynamic topography due to underlying mantle flow34. Although differential surface loading of sediment volumes likely results in localised spatially variable subsidence rates60, geomorphological evidence from the central Sunda Shelf at Belitung Island and in the Singapore Strait (Fig. 1) indicates long-wavelength subsidence at rates of 0.2–0.3 mm/yr34,64 that are representative for the central Sunda Shelf during the Quaternary34.

The Quaternary stratigraphy of Singapore provides evidence of paleoenvironmental change from intertidal sediments and sea-level indicators preserved in paleo channels, coastal deposits and deep-drill boreholes64,73,74,83,85. The natural predeveloped coastline of Singapore was encompassed by extensive coastal to shallow marine ecosystems including coral reefs, intertidal flats, and mangrove forests100,101,102,122. Rapid industrial and urban development during the 20th century, however, has since reduced habitat extent to relatively small remnant patches mainly positioned along northern coastlines and offshore islands (Fig. 1b). The coastal waters surrounding Singapore and in the Johor Strait are relatively shallow (<20 m) but reach depths of ~30 m to ~120 m in the Singapore Strait64,92 (Fig. 1b). Today, water flow through the region connects to the Pacific Ocean via the South China Sea to the northeast, the Indian Ocean through the Malacca Strait to the west and the Java Sea to the southeast (Fig. 1a). The convergence of tidal waters within the Singapore Strait creates a diurnal and semi-diurnal tidal regime with a mean amplitude of 2.4 m during spring tides and 1 m during neap tides8.

We compiled existing RSL data from the central Sunda Shelf, Singapore and Mapur Island, Indonesia123 (Fig. 1a, b). The geological reconstructions and instrumental records for the Holocene, including the 20th and 21st century, use RSL data from Singapore31 and Mapur Island, Indonesia32, respectively. Mapur Island has minimal differences in present-day rate of GIA (<0.1 mm/yr) and sea-surface height (<2 cm) with Singapore32. To extend the temporal scale of the RSL reconstruction to the LGM, we included RSL data from the central Sunda Shelf 26,30. To account for the GIA difference between the central Sunda Shelf and Singapore, we included a vertical uncertainty for the central Sunda Shelf RSL data using predictions from the ICE-6G_C global ice history model37,38 and HetM-LHL140 3D Earth model124,125.

The geological RSL reconstructions are based on mangrove root remnants and intertidal deposits26,30,31 and coral microatolls32 that are used as proxy sea-level indicators to develop SLIPs. A SLIP defines the past position of RSL in time and space with an associated vertical and temporal uncertainty13,126. Standardised protocols in the collation and validation of SLIP data75 require four key attributes including: (1) elevation of the sample relative to a modern tidal datum; (2) vertical relationship of the sample to contemporaneous sea level, termed the indicative meaning; (3) age of formation or growth (e.g., through radiometric methods); and (4) geographic location.

The elevation of SLIPs on the central Sunda Shelf was measured using depths below modern water surface26,30 and in Singapore and Mapur Island using total station surveys31,32. The indicative meaning of SLIPs incorporates the central tendency (reference water level) and indicative (vertical) range of the sample’s distribution relative to tidal levels126. Indicative meanings for mangrove and coral microatoll SLIPs were established using the indicative range of modern analogues detailed in original publications31,32. For the Sunda Shelf SLIPs, we applied the same indicative range as those applied to SLIPs in Singapore. The age of mangrove and intertidal SLIPs were radiocarbon dated and (re)calibrated using IntCal20127. The age of coral microatoll SLIPs were constrained through annual growth-band counting from the living edge of the coral structure32. Geographic locations were extracted from original publications, respectively.

The instrumental RSL records are based on Singapore’s network of tide gauge stations (Fig. 1b) that provide water level measurements during the 20th and 21st centuries of varying time length and completeness (Table 1). From these we constructed an averaged instrumental record using annual data from the four longest, most complete tide gauges including Sultan Shoal, Raffles Lighthouse, Tanjong Pagar and Sembawang (Table 1). While Sembawang was operational since 1954 CE, we excluded measurements prior to 1972 CE because of anomalously high data points that precedes a long data gap between 1961 and 1971 CE after which the station was relocated92. Data for Sultan Shoal are also not available prior to 1972 CE (Table 1).

Statistical analyses

We combined the geological reconstructions and averaged instrumental record and quantified magnitudes and rates of RSL change using an Error-In-Variables Integrated Gaussian Process (EIV-IGP) model36. The EIV-IGP model takes an unevenly distributed RSL time series, prone to vertical and temporal uncertainties, as input and produces estimates of RSL and rates of RSL with 95% credible intervals. The EIV-IGP model models rates of RSL change using a Gaussian process128 (GP) and models RSL as the integral of the GP (IGP) plus (measured and estimated) vertical uncertainty. Temporal uncertainties are accounted for through setting the IGP model in an errors-in-variables (EIV) framework129. The geological reconstructions spanning the end of the LGM through the LDT and Holocene were analysed at 250-year time intervals. The combined geological and averaged instrumental record for the 20th and 21st century were analysed at annual time intervals. We included a vertical uncertainty for the averaged tide gauge record calculated from the standard deviation of RSL values following ref. 36.

We make use of the probabilistic nature of the model-based estimates from the EIV-IGP model to provide perspective to future scenario-based sea-level projections demonstrating the probability of when equivalent rates of RSL rise were last exceeded. Specifically, we use the posterior samples of RSL rise obtained from the EIV-IGP model to estimate the probability that RSL rise in a given year t exceeded sea level projections under various climate scenarios. To estimate the probabilities, we let \({\omega }_{x}^{(s)}\) be posterior sample s of RSL rise in year x and let \({p}_{x}\) be the probability that RSL rise in year \(x\) was greater than a chosen rate of change (denoted \(\delta\)), such that:

$${p}_{x}=(1/{{{{{\rm{M}}}}}})\mathop{\sum }\limits_{s=1}^{M}{\mathbb{I}}({\omega }_{x}^{\left(s\right)} > \delta )$$
(1)

where \({\mathbb{I}}\) is an indicator function such that \({\mathbb{I}}\)(condition) = 1 if the condition in the brackets holds true and 0 otherwise. For the purposes of this study, we included values of \(\delta\) reflecting projected rates of RSL rise by 2150 under different climate scenarios outlined below.

Sea-level projections

We utilise the latest, most up to date regional sea-level projection data currently available for Singapore developed by the IPCC AR6 report4,39. The sea-level projections adopt differing future SSP scenarios, which reflect a range of possible changes in socioeconomic conditions and the geophysical driving mechanisms of climate change, including radiative forcing and greenhouse gas emission and concentration futures4,130. From these SSPs, we selected sea-level rise projections to 2150 (relative to a 1995–2014 baseline period) for Singapore under SSP1-1.9, SSP2-4.5, and SSP5-8.5 emissions scenarios that broadly encompass the lowest and highest emission scenarios, respectively4,39.

Each SSP scenario encompasses an associated GMST and greenhouse gas emission range. Under a very strong mitigation scenario (SSP1-1.9), the GMST (relative to a 1850–1900 baseline period) increase is limited to 1.0–1.8 °C warming by 2100, with net zero CO2 emissions achieved by 2050. Under the moderate emissions scenario SSP2-4.5, GMST is limited to 2.1–3.5 °C warming by 2100, with CO2 emissions falling by 2050 but not reaching net-zero until after 2100. Conversely, under a very high emissions scenario (SSP5-8.5) that reverses current climate policy, GMST increases 3.3–5.7 °C by 2100, with CO2 emissions double current levels by 2050 and continuing to increase thereafter. AR6 also assessed the potential contributions to sea-level rise under SSP1-2.6 and SSP5-8.5 of ice-sheet processes, such as MICI, currently characterised by low confidence. We also include low confidence SSP5-8.5 projections as an indicator of potential high-end sea level projections. Under each SSP, we also report the projected timing of exceedance of 0.5 m and 1.0 m RSL rise magnitude thresholds.

Paleotopographic maps

We produced paleotopographic maps of the Sunda Shelf region at 500-year time intervals to demonstrate spatial and temporal land area changes and lateral shoreline migration in response to RSL change following the LGM. The maps were generated following ref. 131 and use the ICE-6G_C global ice history model37,38 and HetM-LHL140 3D Earth model124,125:

$$T\left(\theta ,\,\lambda ,\,t\right)=S\,\left(\theta ,\,\lambda ,\,t\right)+\left[{T}_{p}\left(\theta ,\,\lambda \right)-S\left(\theta ,\,\lambda ,\,{t}_{p}\right)\right]$$
(2)

Where, θ, λ and t represent latitude, longitude and time, respectively; \(T\left(\theta ,\,\lambda ,{t}\right)\) is the paleotopography at time t; \({T}_{p}\left(\theta ,\,\lambda \right)\) is the present topography from ETOPO1132, \(S(\theta ,\,\lambda ,\,{t}_{p})\) and \(S\left(\theta ,\,\lambda ,{t}\right)\) are the present-day sea level and sea level at time t respectively, which are predicted by a GIA model with the ICE-6G_C ice history model37,38 and the HetM-LHL140 3D earth model124,125. The HetM-LHL140 3D Earth model includes lateral variations both in the lithospheric thickness and mantle viscosity.

We then used the paleotopographic maps to calculate land area difference compared to the present-day topography and rates of landward lateral shoreline migration across a hypothetical transect extending from the South China Sea toward Singapore (Fig. 4a). While lateral shoreline migration rates may respond to variety of regional and local sedimentary processes, for example, delta progradation, sedimentation and erosion133, our modelled results provide an estimated response of the paleogeographic landscape to rising RSL since the LGM.

We also compare and discuss output from the EIV-IGP model results and RSL datasets from the Sunda Shelf, Vietnam Shelf and Singapore with ICE-6G_C model predictions of RSL change (Supplementary discussion).