The YD event is captured in select stalagmite records, including those from China11, India12, Brazil13 and the United States14, but few stalagmite records exist from the deep tropics. This paucity of rainfall records is especially true in caves on islands in the Western Pacific Warm Pool15,16. The Western Pacific Warm Pool represents a major source of heat and water vapour in the global climate system, and its changes affect climate variability across the globe on a multitude of timescales. Paleoclimate records from in and around the western tropical Pacific show minimal temperature responses during the YD17,18,19,20,21,22,23,24,25,26,27,28, with slight cooling confined to the areas around mainland Asia29,30,31,32,33,34 (Fig. 1a, Supplementary Table 1).Figure 1: Synthesis of published paleoclimate data during the Younger Dryas from the western Pacific.
Seasonal differences (JJA minus DJF) in (a) SST (C; 19982012 (ref. 65)) and (b) precipitation (mmh1; 19982012 (ref. 66)) in the western tropical Pacific. (a) The symbols indicate the temperature signal in proxy data during the YD (Supplementary Table 1). The blue arrows denote cooling of 1C or less. The darker line is the coast at 12 kyr ago, that is, the isobath at 60m below modern sea level. b The symbols indicate the hydroclimate signal in proxy data during the YD (Supplementary Table 1). Letters denote the locations of sites discussed: C, China, P, Palawan, B, Borneo and F, Flores. Both temperature and precipitation proxy data indicate that cooling and drying were restricted to the areas of the Asian summer monsoon.
The hydroclimate response to the YD in the Western tropical Pacific and adjacent regions was spatially and seasonally variable (Fig. 1b): the northern summer monsoon weakened, as suggested by 18Osw marine sediment records (a proxy for salinity) from the South China Sea29,30,33, Sulu Sea17 and eastern Indian Ocean20,23 and stalagmite records from southeastern China35 (with a hint of a concomitant increase in the boreal winter and austral summer monsoons in China36 and Flores16,37, respectively). In contrast, most proxy records from around Borneo15,38, the Banda Sea19, the Timor Sea24 and east of the Philippines19,31, where the primary rainy season occurs during boreal winter, show no hydroclimate response to the YD, with a few exceptions25,26,27,28 that are closely tied to the Indonesian throughflow. These records suggest that the YD primarily weakened the boreal summer monsoon in the western tropical Pacific and may have slightly strengthened the austral summer monsoon. A recent study39, however, argues that the stalagmite records from southeastern China may reflect isotopic changes in precipitation over India via atmospheric moisture transport and not local hydroclimate, but the records would still indicate a reduction in the boreal summer monsoon. To our knowledge, there is no direct evidence of boreal summer monsoon changes in the Western tropical Pacific to date.
Here we present a stalagmite 18O record of hydroclimate variability from the Puerto Princesa Subterranean River National Park in Palawan, Philippines (10.2 N and 118.9 E), to better understand the precipitation response of the Western tropical Pacific during the YD. The rainy season today in Palawan (MayNovember) contributes 80% of the total annual rainfall of 2,000mm, and the seasonal variation in average temperature is small (2729C). The limited seasonal variation in temperature and the high humidity of tropical cave environments suggest that changes in rainfall 18O values dominate the stalagmite 18O signal in Palawan40. The stalagmite 18O signal most likely records changes in wet season rainfall amount via the amount effect, which is an empirical relationship between rainfall amount and rainfall 18O values in the tropics41. Studies at similar sites in the Western tropical Pacific40,42 show an inverse relationship between rainfall amount and rainfall 18O composition: a change of 1,000mm of rainfall per year results in a 1 change in rainfall (that is, stalagmite) 18O values on suborbital timescales (decades to centuries) when ocean basin water transfer is minimal43.
The bottom of the stalagmite is UTh dated to 13.7 kyr ago, and a hiatus in growth constrains the youngest part of the record to 10.8 kyr ago (Supplementary Fig. 1). The age model for the stalagmite 18O record consists of 12 UTh dates with an average precision of 78 years (2; Methods; Supplementary Table 2; Supplementary Fig. 2). Subsamples for 18O analysis were drilled every 1m per subsample for the upper 2cm and 0.5m for the lower 16.5cm of the stalagmite, resulting in an average temporal resolution of 7 years per 18O data point. We assessed whether the stalagmite formed in isotopic equilibrium with cave dripwater and conclude that non-equilibrium effects on the stalagmite isotopic composition are limited (Supplementary Figs 3 and 4).
The Palawan stalagmite 18O record (Fig. 2, red line) shows an increase during the YD (12.911.7 kyr ago, Greenland YD chronology8; Fig. 2, blue shading), reflecting drier conditions. The initiation of the YD event, represented by a decrease in stalagmite 18O values of 0.5 (decadal-scale shifts of 0.5m per year or 1.4mm per day), occurs at 12.920.08 kyr ago (Supplementary Fig. 5), which is not significantly different from the cooling in Greenland at 12.790.08 kyr ago (Fig. 3). In addition, the timing of the initiation of the recovery in Palawan at 11.790.08 kyr ago matches the warming in Greenland at 11.670.07 kyr ago. However, those events represent the extent of the similarities in timing, as the initiation and recovery of the YD in Greenland are more abrupt than in Palawan. Indeed, the final recovery of the termination in Palawan occurs at 11.360.08 kyr ago (Supplementary Fig. 5), which is significantly later than the timing of the full recovery in Greenland at 11.630.08 kyr ago.Figure 2: Proxy data for temperature, hydroclimate and ocean circulation highlighting the timing and magnitude of the YD.
The blue shaded bars indicate the timing of the canonical 12.811.6 kyr ago YD derived from Greenland temperatures. The yellow shading indicates a later, more gradual end to the YD 300 years later based on proxy time series of tropical hydroclimate. The red data points with error bars are the UTh dates for the new stalagmite record from Palawan, Western Philippines. The timing of the blue bar is more coeval with Greenland temperature8 and Cariaco Basin SST47, which respond to the AMOC7 and sea ice46, whereas the yellow bar timing is seen in proxy records outside the northern North Atlantic35,48,49,50 and is more concurrent with the timing of increases in NH summer insolation67 and CO2 (ref. 68).Figure 3: The timing of the onset and termination of the YD in well-dated paleoclimate proxy time series.
The initiation (green circles) and completion (red triangles) of both the onset and the termination are depicted. Filled symbols indicate that the timing is calculated using a Bayesian change point method69. When the method did not detect a change point, visual determinations were used (open symbols). Error bars on the timing come from the age uncertainty reported in the published record. The exception to this is the Palawan record, for which the error bars are derived from the width of the distributions in Supplementary Fig. 5. All time series show a contemporaneous initiation of the onset and the termination, as has been established in previous studies. However, the completion of both parts of the YD in hydroclimate and temperature proxies outside of the North Atlantic occurs more gradually and does not fully recover until at least 300 years after Greenland temperatures.
The timing of the initiation of the onset and termination of the YD are contemporaneous among the various proxy records in the Atlantic and Pacific sectors, suggesting a common driver: the AMOC (Fig. 2). At the onset of the YD, the North Atlantic sea surface temperature (SST)44,45 and Greenland temperature1 decreased as the AMOC weakened7 and sea ice extent increased rapidly46. SSTs in the Cariaco Basin off the coast of Venezuela in the Caribbean Sea also decreased abruptly47. Cooling in the North Atlantic reduced the meridional SST gradient in the tropical Atlantic and eventually shifted the intertropical convergence zone (ITCZ) southward48, resulting in decreased rainfall over the Cariaco Basin48 and western Africa49. In the North Pacific, an increase in sea ice accompanied lower SSTs50. During the period of reduced AMOC, the northern summer monsoons in China11 and the western tropical Pacific (this study) weakened. Sea surface salinity in select locations around the western Pacific also exhibited a change towards more saline conditions17,20,25,26,27,28,33.
Although proxy data agree on how the YD affected climate, data on the temporal evolution of the YD fall into two categories: gradual (duration of >100 years to reach and recover from full YD conditions) or abrupt (duration of 10100 years to reach and recover from full YD conditions). These timings are based on the adaptation time of human and natural systems to drastic changes in climate2. The rapid resumption of the AMOC (100 years), coupled with the retreat of insulating sea ice at 11.6 kyr ago46, likely led to the abrupt warming (40 years to recover from full YD conditions) in Greenland surface air temperature51,52. SSTs from a well-dated marine sediment core from the Cariaco Basin in the Caribbean Sea show the same rapid onset (48 years to reach full YD conditions) and termination (147 years to recover from full YD conditions) of the YD47. However, a proxy record of hydroclimate from the same basin48 shows a much more gradual onset (261 years) and termination (327 years), such that the completion of the termination occurred at 11.300.15 kyr ago (Fig. 2). Well-dated hydroclimate records from several tropical locations also show a gradual YD response. In addition, rainfall in western Africa49 displays a gradual onset (331 years) and termination (547 years), with completion of the termination occurring at 11.120.15 kyr ago. A gradual onset and termination also occurred in the northern summer monsoons in China11 (onset: 437 years; termination: 343 years; completion of termination: 11.180.08 kyr ago), in the Western tropical Pacific (this study; onset: 565 years; termination: 440 years), and in the sea surface hydroclimate conditions in the eastern Indian Ocean20 (onset: 210 years; termination: 400 years; completion of termination: 11.400.15 kyr ago). A gradual onset (240 years) and termination (280 years) are also observed in a well-dated SST record from outside of the tropics in the North Pacific (completion of termination: 11.250.15 kyr ago, at which time the local sea ice also disappeared) (Fig. 2). The time required to reach and recover from full YD conditions in the tropical hydroclimate records and temperature in the North Pacific (gradual; Fig. 3) clearly differs, beyond the range of error, from that of temperature in the Atlantic, sea ice in the North Atlantic, and the AMOC (abrupt; Fig. 3).
We compare the proxy records with the deglacial climate transition simulated by two climate models of different complexities (CCSM3 (ref. 53) and LOVECLIM54; see Methods). These climate models are forced with temporally varying greenhouse gas concentrations, insolation, ice-sheet topography and meltwater fluxes from the last glacial maximum (1921 kyr ago) through the deglaciation. The temporal evolution of climate simulated by these models qualitatively agrees with the proxy records, indicating that cooling in the North Atlantic led to cooling and drying in the northern tropical Atlantic and Indo-Pacific (Fig. 4). The timing of the simulated YD closely follows the evolution of prescribed meltwater influx into the North Atlantic, which supports the hypothesis that this climate event was mediated by changes in the AMOC. The magnitude of SST change in the tropical Indo-Pacific is <0.5C and agrees with the uncertainty of paleoclimate reconstructions of SST25,27. The magnitude of precipitation change in the tropical Indo-Pacific is, however, substantially smaller than the amount suggested by the new stalagmite proxy data from Palawan (Fig. 2; Supplementary Fig. 7). The underestimated hydroclimate response in the tropical Indo-Pacific may be attributed to several factors, although we are not able to pinpoint the exact cause: the comparison of grid-scale model output to proxy data at a specific grid point, the relatively low horizontal resolutions of the atmospheric component models (3.75 in CCSM3 and 5.63 in LOVECLIM), uncertainty in the meltwater forcing, and/or uncertainty in the conversion from stalagmite 18O values to rainfall amount.Figure 4: Climate model output highlighting the timing and magnitude of the YD.
The blue shaded bars indicate the timing of the canonical YD from 12.8 to 11.6 kyr ago. Climate model outputs from the CCSM3 TraCE-21ka53 (black curves) and LOVECLIM DGNS54 (grey curves; secondary y-axis has the same units as the primary y-axis) experiments are averaged over Greenland (annual mean; 6575 N, 5030 W), the tropical North Atlantic (JJA; 515 N, 7020 W), and the tropical Indo-Pacific (JJA; 515 N, 80140 E) to approximate the locations of the proxy records in Fig. 2. The model data are smoothed with a 21-year running mean filter. NH freshwater forcing in the two model simulations is shown in the bottom panel.
The abrupt recovery of Greenland temperature from the YD is reproduced in LOVECLIM but not in CCSM3 due to the different behaviour of sea ice in these models (Fig. 4). In LOVECLIM, sea ice expands to the south of Greenland during the YD and then retreats somewhat rapidly (110 years for the onset and 270 years for the termination) to the north on the sudden termination of the meltwater flux at 12.2 kyr ago (Fig. 5). In CCSM3, sea ice continues to cover the oceans adjacent to Greenland throughout the YD recovery (Fig. 5), whereas surface temperature in Greenland increases much more slowly (900 years for both the onset and termination) than in LOVECLIM (Fig. 4). In both models, the temperature and precipitation in the northern tropical Atlantic (temperature: 480 years onset, 680 years termination in CCSM and 150 years onset, 600 termination in LOVECLIM; precipitation: 620 years onset, 800 years termination in CCSM and 180 years onset, 420 termination in LOVECLIM) and Indo-Pacific (temperature: no change points detected in CCSM or LOVECLIM; precipitation: 930 years onset, 950 years termination in CCSM and 260 years onset, 450 termination in LOVECLIM) increase more gradually, in agreement with the proxy records. These results suggest that the regional sea ice extent controls the surface air temperature in Greenland51. The temperature and precipitation in the northern tropical Atlantic and Indo-Pacific increase more gradually, in agreement with the proxy records. Tropical hydroclimate is likely primarily influenced by hemispheric-scale temperature changes, which require a longer adjustment time than regional sea ice changes52. In addition to an increase in the AMOC during the recovery, increasing summer insolation and greenhouse gas concentrations (Fig. 2) also contribute to the slow recovery from the YD in the northern tropics. Decomposition of the climatic responses to these different forcings, based on empirical orthogonal function analysis, suggests that insolation and greenhouse gases may account for approximately two-thirds of the precipitation increase in the tropical Indo-Pacific during the YD recovery (Supplementary Fig. 6), and AMOC changes account for the remaining one-third of the change in Indo-Pacific precipitation.Figure 5: Sea ice coverage in the North Atlantic during the YD from two climate models.
Output is from CCSM3 (a) and LOVECLIM (b). Sea ice data are based on March values and averaged over a 100-year period centred at the year indicated at the top left of each panel. Sea ice in LOVECLIM is at near-modern extents from 12.4 to 11.6 kyr ago, coincident with the abrupt warming in Greenland temperature in this model (Fig. 4). In CCSM3, however, sea ice continues to extend to the south of Greenland even at the end of the YD (11.2 kyr ago), and Greenland temperatures do not show an abrupt warming. This pattern indicates that sea ice acts as a switch in the models and can abruptly influence temperatures in Greenland.
In CCSM3, the tropical Atlantic acts as a link that connects AMOC reduction during the YD in the North Atlantic to northern summer monsoon rainfall changes in the Indo-Pacific. During boreal summer, the region of decreased precipitation extends from the northern tropical Atlantic into the eastern tropical Pacific, with alternating signs of precipitation changes in the central (increase) and western Pacific (decrease; Fig. 6). The precipitation decrease over the western Atlantic warm pool in the model, which is also clearly captured in proxy data from the Cariaco Basin48, directly stems from the reduction of the AMOC as the ITCZ shifts southward. Reduced precipitation in the western tropical Atlantic intensifies the northeasterly trade winds across Central America, reducing SST and precipitation in the northeastern tropical Pacific55. Attendant oceanatmosphere interactions reorganize the Walker and Hadley circulations56,57, resulting in decreased precipitation over the western Pacific during boreal summer. Proxy data from sites in the western tropical Pacific that have a pronounced wet season during boreal summer, such as China, the South China Sea and the eastern Philippines (Fig. 1), generally exhibit a decrease in precipitation during the YD (Fig. 7). During boreal winter, tropical precipitation changes simulated by models exhibit a more zonally symmetric pattern relative to that in boreal summer (Fig. 6; Supplementary Fig. 8). Rainfall associated with the ITCZ increases across the Southern Hemisphere in association with the strong cooling in the Northern Hemisphere resulting from the reduced AMOC. The intensified winter westerly jet effectively spreads the cooling in the North Atlantic across the Northern Hemisphere, resulting in a more zonally symmetric response in the tropical hydroclimate58, with signs of increased rainfall in the southern tropical Indo-Pacific. These results agree with a similar study of proxy data and climate models studying the response of the western Pacific to changes in the AMOC27.Figure 6: Seasonal distribution of surface temperatures in the CCSM3 model.
Temperature (C, colour shading) and rainfall changes (mm per day, contours) are from before (13.413.0 kyr ago) and during (12.412.0 kyr ago) the YD for (a) JJA and (b) DJF as simulated in the CCSM3 TraCE-21ka experiment31. Brown and green contours represent decreased and increased rainfall, respectively, during the YD time period (contour interval: 0.4mm per day). A general decrease in rainfall is observed over the tropical Indo-Pacific during JJA, although the magnitude of annual rainfall change is only 3% (Supplementary Fig. 6), much smaller than the estimate based on the Palawan record (25%).Figure 7: Proxy data for hydroclimate in the WPWP.
The record from Palawan, which has a boreal summer bias, shows strong evidence for reduced rainfall, whereas the record from Flores16,70, which has a boreal winter bias, shows evidence for increased rainfall during the YD. The compilation of marine sediment core seawater 18Osw25, Borneo stalagmite 18O15 and Sulawesi Dleafwax38 data shows no evidence of a YD signal because these data integrate across the seasons. One of the notable exceptions among marine core 18Osw records is from the Sulu Sea17, which formed a more enclosed basin during the YD (Fig. 1, bold coastline) and likely integrated the seasonal precipitation and runoff changes.
Seasonally dependent changes in precipitation during the YD explain most of the proxy data from the Indo-Pacific (Figs 1 and 7). Rainfall records from sites that are biased towards boreal summer tend to show a decrease in precipitation during the event, whereas rainfall records from sites that are biased towards boreal winter show an increase in precipitation. Sites that represent annual conditions, that is, no seasonal bias, tend to show no response to the YD, which includes both precipitation and surface salinity records. Offsetting changes, that is, a decrease in one season and an increase in the other, during the YD might lead to no change in mean annual conditions, which may be the case in the precipitation records from Borneo15 and Sulawesi38 and in the marine sediment inferred seawater 18O (18Osw) records (Figs 1 and 7). Many marine sediment 18Osw records19,20,23,24,25,31, a proxy for salinity, integrate the offsetting precipitation changes throughout the year and, hence, do not record a YD signal (Fig. 1b). The marine sediment records in the WPWP that do reveal a hydroclimate response to the YD are in the South China Sea29,30,33, the Sulu Sea17 and the eastern Indian Ocean just off of the coast of Sumatra20. The hydroclimate signal in the Sulu Sea most likely stems from the fact that this sea, located entirely in the Northern Hemisphere, was more of an enclosed basin when sea level was lower at 12 kyr ago (thick black line in Fig. 1). Therefore, the changes recorded in this sea are likely due to changes in local precipitation and runoff. In addition to local freshwater input, advection by ocean currents likely contributed to the YD salinity signal in the South China Sea and the eastern Indian Ocean marine sediment 18Osw records. Advection of less saline water by ocean currents also explains the marine sediment cores from south of the Equator that indicate drier conditions during the YD25,26,27,28. These areas likely did not experience drier conditions locally; rather, the sites recorded the signal of the weaker boreal summer monsoon changes north of the Equator that were then transported southward by the strong currents of the Indonesian throughflow.