Introduction

Temperature seasonality, defined as the annual cycle of surface temperature (summer minus winter temperature), is a key climate variable in the mid- to high-latitude land areas, accounting for over 90% of total temperature variance1,2. Changes in temperature seasonality have attracted increasing attention in recent years due to their significant ecological and societal impacts, such as alterations in plant biological cycles3, shifts in animal migration patterns4, and effects on seasonal influenza outbreaks and disease transmission5.

Observational data reveal a marked reduction in temperature seasonality across East Asia (EA) since the mid-19th century6, surpassing the historical variability measured since the 14th century7. Climate model simulations from the Coupled Model Intercomparison Project Phase 5/6 (CMIP5/6) have mainly attributed this weakening trend in temperature seasonality to anthropogenic forcing (i.e., greenhouse gas and aerosol concentrations)8,9,10,11. Specifically, increased greenhouse gases reduce outgoing long-wave radiation from the surface and mainly prevent the winter surface temperature from decreasing, while increased aerosol concentrations act to reflect or scatter incoming solar radiation and mainly prevent the summer surface temperature from rising, ultimately weakening the temperature seasonality8. However, modeling results may substantially underestimate the observed reduction in temperature seasonality7. This discrepancy likely arises from challenges in detecting the effects of natural external forcing, given the limited temporal availability of instrumental data.

The last deglaciation (19–11 ka) was a period of intense global warming. Investigating changes in temperature seasonality along with the underlying mechanisms during this period provides a context for understanding the roles of natural forcing. In EA, numerous studies have documented increasing trends in annual mean or warm season temperatures during the last deglaciation12,13,14,15,16, while only a few reconstructions revealed the winter temperature changes. Previous reconstructions based on pollen records demonstrated a more pronounced temperature seasonality in the North Hemisphere at the Last Glacial Maximum (LGM; 21 ka) compared to present-day conditions17,18. Time series of pollen-based seasonal temperature in Yunnan Province in southwest China showed a greater winter warming than summer during the last deglaciation, i.e., a decreasing trend of temperature seasonality19. Overall, the evolution of EA temperature seasonality during the last deglaciation was very limited in previous reconstruction studies.

Similar research gaps also exist in previous modeling studies. For example, an ensemble mean of 13 simulations from the Paleoclimate Modeling Intercomparison Project Phase 4 (PMIP4) supported the reconstructions in that simulated temperature seasonality over the extratropical Asia was more pronounced during the LGM than present-day conditions20, but also lacking information on its temporal features during the last deglaciation. In Greenland, transient simulations of TraCE-21ka have demonstrated the effects of different external forcing, which indicated that changes in orbital insolation (ORB) contributed to enhancing the temperature seasonality during the last deglaciation by causing significant anomalous summer warming, while rising greenhouse gas (GHG) concentrations contributed to stronger winter than summer warming21,22. However, the impacts of different external forcing on EA temperature seasonality have not been quantified.

During the last deglaciation, the global mean sea level rose significantly, by approximately 70 m23, leading to the inundation of the EA continental shelf area, an area spanning more than 800,000 km2, accounting for more than 40% of total changes in subtropics (20°–45° N and 20°–45° S)24 (Fig. 1c). Recent modeling studies have highlighted the influence of sea-level changes on global atmospheric and oceanic circulation patterns25. For example, during the LGM, the exposure of continental shelves increased surface albedo, ultimately leading to a reduction in global annual mean surface temperature26,27,28,29. In EA, it can be assumed that changes in the land‒sea configuration could affect thermal heat capacity, which might modulate land‒sea thermal contrasts in summer and winter30,31, thus influencing local temperature seasonality. However, this hypothesis has not yet been quantified under the last deglaciation conditions, and the relative climatic impacts of surface albedo and thermal heat capacity of the expanded EA ocean remain unknown.

Fig. 1: Changes in East Asia (EA) surface temperature during the last deglaciation and roles of each external forcing.
figure 1

a Simulated June-July-August minus December-January-February (JJA–DJF) EA (20°‒45° N, 100°‒130° E) surface temperature in ICE (red), ICE + ORB (orange), ICE + ORB + GHG (black), ICE + ORB + GHG + MWF (iTraCE) (blue) experiments. The green line represents reconstructed temperature differences between the warmest and coldest months based on pollen data from Tengchongqinghai Lake in Yunnan Province19. b Contributions of different forcing to EA temperature seasonality from 14 to 11 ka. Bars from left to right represent total changes (iTraCE) and contributions of EA land‒sea configuration (EALSC), ICE minus EALSC, ORB, GHG, and MWF, respectively. ICE minus EALSC denotes the combined effect of ice sheet changes and land‒sea configuration changes in other regions (especially the Maritime Continent and Bering Strait). c Global land‒sea configurations from 21 ka to 11 ka. The shadings represent the topography at 11 ka. The red areas denote the regions inundated due to sea level rise from 21 ka to 11 ka. The time series attached on the figure show the EA land area changes in the blue box revealed by ICE-6G. The blue circles indicate the land area set in EA_OCN and EA_LND experiment, respectively, the difference of whom represents the EALSC forcing.

Full size image

Therefore, in this study, we use isotope-enabled transient climate evolution (iTraCE) simulations32,33 and conduct a series of land‒sea configuration sensitivity experiments to investigate the characteristics and mechanisms behind temperature seasonality changes in EA during the last deglaciation. Our research focuses on three main objectives: (i) identifying the specific characteristics of temperature seasonality changes in EA during this period; (ii) assessing the contributions of different external forcing, with a particular focus on the climatic impact of the inundation of the EA continental shelf area; and (iii) exploring the underlying mechanisms.

Results

Changes in EA temperature seasonality during the last deglaciation

Figure 1a shows the simulated changes in summer-minus-winter temperatures over EA (20°‒45° N, 100°‒130° E) during the last deglaciation, derived from iTraCE simulations. EA temperature seasonality generally exhibited an increasing trend during the period 18–14 ka, followed by a strong decrease from 14 to 11 ka, which is dominated by prominent winter warming during this period. This trend is mainly characterized by two abrupt decreases at 14 ka and 12 ka, accompanied by multi-millennial timescale fluctuations induced by meltwater flux (MWF) (Fig. 1a). At the two points of 14 ka and 12 ka, global land–sea configurations were modified substantially in the ICE forcing in iTraCE simulation to describe the significant sea level rise during the last deglaciation, which mainly occurred during 14.5–11 ka revealed by the ICE-6G reconstruction24. Spatial patterns of temperature seasonality from 14 ka to 11 ka revealed a general decrease in the EA landmass, and a notable decrease of approximately 9.6 °C over the EA shelf area (Fig. 2a).

Fig. 2: Contributions of East Asia (EA) continental shelf area inundation to seasonal surface temperature (TS) variations.
figure 2

a June-July-August (JJA) minus December-January-February (DJF) temperature (°C) changes from 14 to 11 ka in the iTraCE simulation (ICE + ORB + GHG + MWF). b JJA minus DJF, c JJA, and d DJF temperature (°C) responses to EA continental shelf area inundation. Red lines denote the inundated shelf area where the topography was changed. Temperature labels indicate the regional average temperature anomalies within the region enclosed by red lines. Dotted areas exceed the 90% confidence level based on the two-tailed Student’s t test.

Full size image

A model-data comparison was conducted to evaluate the simulated temperature seasonality in EA during the last deglaciation. A reconstructed time series of temperature seasonality based on pollen records at Lake Tengchongqinghai (25.1°N, 98.6°E) in Yunnan Province in southwest China19 well supports the simulated results. It shows that the temperature seasonality decreased during the last deglaciation, especially after 15 ka, when the global land–sea configurations changed significantly due to sea level rise (Fig. 1a). The record suggests a decrease in temperature seasonality of about 0.7 °C from 14 ka to 11 ka, comparable to the simulated 0.8 °C decrease in the same site. Additionally, we also examined the reconstructions of EA temperature seasonality anomalies in the time slice of LGM compared to present-day conditions (Fig. S1a–c)17,18. Although the time span of these reconstructions is longer than the simulated results, they contain more data in East Asia, which helps to bring more information on winter temperature changes. The temperature seasonality reconstructions showed a general decrease in China between the LGM and present (Fig. S1c), similar to the seasonal temperature changes simulated by iTraCE.

Contributions of EA continental shelf area inundation

The effects of single external forcing on temperature changes in EA are depicted in Figure S2. The impacts of ORB and GHG were consistent with those observed in previous studies21,34. Orbital-induced insolation changes increased summer temperature but decreased winter temperature in EA during the last deglaciation, increasing EA temperature seasonality by 2.3 °C (Fig. S2c). Conversely, increasing GHG concentrations caused stronger warming in winter compared with that in summer (Fig. S2d), weakening the seasonality by 0.8 °C. They primarily exerted their influence from 18 to 14 ka, whereas their impact was negligible during 14–11 ka.

Notably, ICE forcing greatly weakened temperature seasonality (−2.5 °C) by causing sharp winter temperature increases at 14 and 12 ka (Fig. S2b), consistent with the times at which land‒sea configurations were modified. Our land‒sea configuration sensitivity experiment conducted in EA (EA_OCN minus EA_LND) further separated the impact of EA land‒sea transition (EALSC) from that of ICE forcing (Fig. 1c). The results revealed that the decrease in temperature seasonality was primarily driven by the EA land‒sea transition (−1.3 °C) (Fig. 1b). The secondary contributor was ICE minus EALSC, reflecting the combined effect of ice-sheet changes and land‒sea configuration changes in other regions (Fig. 1c). The ice sheet configurations were modified every 1000 years in the iTraCE. However, shown by the effects of ICE forcing, the temperature seasonality kept almost unchanged except at 14 ka and 12 ka, when the global land–sea configurations were modified (Fig. S2b). It suggests that the role of land‒sea configuration changes is more important than that of ice-sheet melting.

Given the dominant role of EA land‒sea transition in decreasing temperature seasonality, we examined the spatial patterns of temperature changes in the EA_OCN minus EA_LND sensitivity experiment. The results revealed similar patterns to those observed in the iTraCE. Temperature seasonality exhibited a general decrease across EA, concentrated over the continental shelf area (Fig. 2b). Temperature seasonality over the shelf area decreased by 5.9 °C, accounting for 61% of the changes observed in the iTraCE. The land‒sea transition caused anomalous cooling in summer (−1.6 °C; Fig. 2c) but warming in winter (4.3 °C; Fig. 2d) over the shelf area. The EA landmass also experienced cooling anomalies in summer (Fig. 2c), implying that shelf area inundation suppressed summer warming across the EA landmass, and explaining the weaker summer warming relative to winter warming observed in the iTraCE (Fig. S2a). Winter anomalies across the EA landmass were less pronounced. Some coastal regions warmed, while there was a cooling anomaly over central China (Fig. 2d). Consequently, temperature seasonality decreased in northeast and southern China (Fig. 2b). In central China, temperature seasonality responses to the EA land‒sea transition were negligible (Fig. 2b), indicating that the seasonality decreases observed in the iTraCE (Fig. 2a) and ICE (Fig. S8a) simulations were likely driven by ice-sheet changes or land‒sea configuration changes in other regions.

According to ICE-6G reconstructions24, the EA shelf area continued to be inundated during the period 11–7.5 ka, shrinking by approximately 300,000 km2. To further explore the effects of EA land‒sea transition during this period, we conducted an additional group of sensitivity experiments. The results showed a similar temperature seasonality response to that of the previous period (14.5–11 ka), indicating a temporally robust reduction in temperature seasonality induced by the EA land‒sea transition (Fig. S3).

Mechanisms

Previous studies have suggested that the primary mechanisms by which land‒sea changes drive climate change involve alterations in surface albedo and heat capacity28. To separate these effects, we conducted a sensitivity experiment (EA_OCN_LndAlb) in which the surface albedo of the EA shelf area was increased from 5% to 30%, with other boundary conditions remaining the same as the EA_OCN experiment (Fig. 3b).

Fig. 3: Contributions of surface albedo and thermal heat capacity to temperature change over the East Asia (EA) shelf area.
figure 3

a Topography (m) and b surface albedo (%) changes in EA_OCN minus EA_LND experiments. In EA_OCN minus EA_OCN_LndAlb experiments, only the surface albedo of the shelf area decreased. c Annual mean (ANN), June-July-August (JJA) minus December-January-February (DJF), JJA, and DJF temperature responses (°C) over the inundated shelf area. Gold, red, and blue bars represent the effects of land‒sea transition (EA_OCN minus EA_LND), decrease in surface albedo (EA_OCN minus EA_OCN_LndAlb), and increase in thermal heat capacity (EA_OCN_LndAlb minus EA_LND), respectively. d Same as (c), but for responses over EA (20°‒45° N, 100°‒130° E).

Full size image

The lower surface albedo of the ocean (EA_OCN minus EA_OCN_LndAlb) resulted in the shelf area absorbing more solar radiation, increasing the annual mean surface temperature by 0.5 °C (Fig. 3c), consistent with a previous study on the Maritime Continent28. This phenomenon was more pronounced in summer because the shelf area received more solar insolation in summer compared with winter. Consequently, the local temperature responses to the decrease in albedo were seasonally asymmetric, with surface temperature increasing by 0.7 °C in summer but only 0.1 °C in winter, slightly increasing the temperature seasonality (by 0.6 °C).

The increase in thermal heat capacity (EA_OCN_LndAlb minus EA_LND) caused local cooling anomalies (−2.3 °C) in summer and warming anomalies (4.2 °C) in winter, much stronger than responses to the decrease in albedo. Temperature seasonality was markedly reduced by 6.5 °C. This differed from the results of a previous study on the Maritime Continent28. Compared with tropical regions, subtropical regions have larger seasonal variations in solar insolation, leading to a greater variation in land‒sea temperature contrast. Hence, temperature seasonality responses to land‒sea transition in EA were much stronger than those in the Maritime Continent. This seasonality-damping effect of the ocean compared to land has been proposed previously in North Africa during the Eocene–Oligocene transition35. But the effect has not been quantified by isolating it from surface albedo changes, and how the resulting processes provide feedbacks on summer and winter temperature respectively have not been analyzed.

As shown by our quantified results of temperature changes over the inundated shelf area, there was an asymmetry in the intensity of seasonal temperature responses, with the absolute values of temperature anomalies in winter being substantially larger than those in summer (Fig. 3c). This asymmetry persisted even in the absence of surface albedo changes. This implies an important role of atmospheric feedback processes triggered by increased heat capacity (EA_OCN_LndAlb minus EA_LND).

In winter, the water content and the surface humidity over the EA shelf area were increased after land–sea transition, increasing the local evaporation (Fig. S4f) and thermal heat capacity. Increased heat capacity has a damping effect on the low temperature in winter, thus causing a warming anomaly. The initial warming could further enhanced surface evaporation. The resulting water vapor rose and formed increased amount of cloud over the continental shelf area (Fig. 4b). These clouds reduced outgoing longwave radiation and increased the cloud‒longwave radiation effect (Fig. 4d), intensifying surface warming through a positive cloud‒radiation feedback (Fig. 6b). It also had opposing effects by reducing shortwave radiation, but the cloud‒longwave radiation effect was more pronounced in winter. Additionally, this local warming generated anomalous southeasterly winds over the shelf area, which brought warm air into the region and induced positive warm advection (Fig. 4h), further amplifying the local warming effect.

Fig. 4: Climatic impacts of increased thermal heat capacity of the East Asia shelf area.
figure 4

Responses of (a) June-July-August (JJA) and b December-January-February (DJF) vertically integrated total cloud cover (CLD; %). Responses of the (c) JJA shortwave cloud radiation effect (SWCRE; W m−2) and d DJF longwave cloud radiation effect (LWCRE; W m−2). Responses of (e) JJA and (f) DJF precipitation (Pr; mm day−1, shading) and winds at 850 hPa (UV850; m s−1, vectors). g JJA and h DJF 850‒500 hPa averaged advection of mean temperature by anomalous winds (\(-{{\bf{V}}}^{{\prime} }\cdot \nabla {\rm{T}}\); 10−5 °C s−1). Dotted areas exceed the 90% confidence level based on the two-tailed Student’s t test. Only vectors exceeding the 90% confidence level are shown.

Full size image

In summer, the initial cooling caused by greater heat capacity reduced local evaporation (Fig. S4e), leading to a decrease in cloud cover over the continental shelf area (Fig. 4a). This reduction in cloud cover in summer triggered a warming effect by increasing downward shortwave radiation penetrating cloud layers (Fig. 4c), which limited cooling anomalies over the EA shelf area. These different feedback processes in summer compared with those in winter contributed to the asymmetry in seasonal temperature responses (Fig. 6a).

Changes of temperature seasonality over EA continental shelf area expanded to the EA landmass due to the resulting atmospheric responses (Fig. 3d). The cooling in summer triggered a strong anticyclonic circulation anomaly (Fig. 4e), with anomalous easterly winds bringing cooler air from the Pacific Ocean onto the EA landmass. This cold advection further decreased temperatures over the region (Fig. 4g). Additionally, the easterlies carrying moisture underwent orographic uplift in the east of the Yunnan-Guizhou Plateau, initiating strong convective activity (Fig. S5). This convergence of moisture supply and enhanced anomalous ascending motion increased cloud coverage and precipitation, consequently weakening the downward shortwave radiation in the EA (Fig. 4a, c, e). The two processes of cold advection and cloud radiation effect induced remote summer cooling over the EA continent, ultimately weakening temperature seasonality over the EA continent.

Furthermore, EA land‒sea transition also triggered a circum-global teleconnection (CGT)-like pattern across the Northern Hemisphere (Fig. S6). Temperature anomalies stimulated a Rossby wave train that propagated downstream along the westerly jet; this phenomenon was particularly pronounced in winter when the westerly jet was stronger (Fig. S6c, f), and resulted in a marked decrease in temperature seasonality across mid-to-high latitudes of the Eurasian continent (Fig. S6a). These teleconnections may have further amplified the weakening of temperature seasonality over the EA continent.

Inundation of the EA continental shelf area also contributed to the decreasing trend in the oxygen isotope signature of precipitation (δ18Op) over the EA landmass during the last deglaciation (Fig. 5a), consistent with findings from cave stalagmite reconstructions (Fig. 5b)36,37,38,39. This finding challenges the conventional “land bridge” hypothesis, which has proposed that EA continental shelf area inundation would increase δ18Op over the EA landmass by shortening the moisture transport distance from the Pacific Ocean40,41. Increased precipitation in the winter decreased the δ¹⁸Op over the inundated shelf area and EA coastal regions (Fig. S7f). Additionally, the isotopic composition effect in summer also played a significant role in decreasing the δ¹⁸Op over the EA landmass (Fig. S7b). It potentially reflects changes in water vapor transported from the Pacific and Indian Ocean and changes in upstream depletion (Fig. 4). Quantified contributions of each process are not allowed based on our current experiments. Further tagging experiments can be conducted in our future study focusing on δ¹⁸Op responses in Asian Monsoon regions to global land-sea configuration changes.

Fig. 5: Responses of precipitation δ18O (δ18Op) to East Asia (EA) shelf area inundation and comparison with reconstructions.
figure 5

a Simulated δ18Op responses (vSMOW, Vienna Standard Mean Ocean Water, 1‰) to East Asia (EA) continental shelf area inundation. Green triangles denote the locations of proxies. Dotted areas exceed the 90% confidence level based on the two-tailed Student’s t test. b Reconstructed δ18O from cave records (δ18Oc, vPDB, Vienna PeeDee Belemnite, 1‰). The red, orange, green, blue lines denote records from Hulu34, Sanbao35, Haozhu36, Dongge cave37, respectively. Gray solid line is the ensemble mean of four cave records, while the gray dashed line represents its trend. Red stars represent the regional averaged simulated δ18Oc (vPDB, 1‰) in the EA (20°‒45° N, 100°‒130° E) responses to EA continental shelf area inundation.

Full size image

Discussion

In this study, we used iTraCE simulations and conducted a series of land‒sea configuration sensitivity experiments to investigate changes in EA temperature seasonality during the last deglaciation. The simulations revealed a significant decrease in EA temperature seasonality from 14 to 11 ka, consistent with paleoclimatic reconstructions. Single-forcing simulations and sensitivity analyses suggest that this decreasing trend was primarily caused by land‒sea transition in EA due to sea-level rise, rather than the traditional views of orbital forcing, greenhouse gases, and ice sheet. The effects of changes in albedo and heat capacity were quantified; reduced albedo led to shelf area warming, especially during summer, while the ocean’s higher heat capacity compared to land led to anomalous cooling in summer and warming in winter, contributing to the overall reduction in temperature seasonality. Notably, there was an asymmetry in the magnitude of local anomalous temperature responses, with the absolute values of winter temperature anomalies markedly exceeding those in summer. This asymmetry was influenced by cloud‒radiation feedback, temperature advection, and a CGT-like pattern, which also expanded the local impacts across the Eurasia continent (Fig. 6). Although some previous studies cared about the temperature seasonality changes during the LGM compared to present-day conditions, most of them focused on the period of Holocene, and few studies demonstrated the feature and mechanism of seasonality evolution during the period of last deglaciation.

Fig. 6: Diagrams of the mechanisms by which East Asia (EA) continental shelf area inundation weakens temperature seasonality over EA.
figure 6

a Local annual temperature cycle responses to EA continental shelf area inundation. The increase in thermal heat capacity weakens the annual temperature cycle (ATC). The decrease in surface albedo along with cloud‒radiation feedback processes contribute to warming in both winter and summer, increasing the annual mean temperature (MAT). Consequently, winter warming is stronger than summer cooling over the shelf area. b Diagram illustrating negative cloud‒radiation feedback in summer. Cooling in summer weakens local evaporation, thus reducing the amount of cloud over the shelf. Reduced cloud cover increases the shortwave cloud radiation effect, suppressing the cooling of the shelf area. Local cooling then triggers an anomalous anticyclone, further cooling the EA landmass. c Diagram illustrating positive cloud‒radiation feedback in winter. Warming enhances evaporation from the surface. Meanwhile, the ocean has a higher evaporative capacity compared with that of land owing to its higher surface humidity. Water vapor moves upwards, forming more clouds over the shelf area; this reduces outgoing longwave radiation and increases the longwave cloud radiation effect, further enhancing warming. Anomalous southerly winds also strengthen local warming by inducing warm advection.

Full size image

Changes in ice sheets and land‒sea configurations in other regions also markedly affected EA temperature seasonality (Fig. 1b). The abrupt changes at 14 and 12 ka, contrasted with negligible variations during other periods in the ICE experiment (Fig. S2b) suggest that the role of land‒sea configuration changes is more important than that of ice-sheet melting. Between 14 and 11 ka, cooling anomalies in summer and warming anomalies in winter were observed along coastlines from EA to the Bering Strait (Fig. S8). This implies that our result—namely, that shelf area inundation reduces local temperature seasonality—could also be applicable to global higher-latitude regions.

Today, global mean sea level is rising at an accelerating rate of about 4.5 mm/year, compared to the rate of 2.1 mm/year in 199242. The current sea level is about 20 cm higher than it was at the beginning of the 20th century43. CMIP6 Projections indicate that by 2100, sea levels could rise by 0.38 m (under Shared Socio-economics Pathways SSP1‒1.9) to 0.77 m (under SSP5‒8.5)44. EA is also one of the most sensitive regions in the context of future sea-level rise. Based on a high-resolution global elevation model, a previous study suggests that 81,900 km2 of coastal land area in China will be very vulnerable to sea level by 2100—an area almost comparable to the size of Jiangsu Province45. Although it could not cause such significant continental shelf area inundation as during the last deglaciation, we highlight the sensitivity of temperature seasonality over coastal regions to heat capacity changes due to land‒sea transition in mid- to high-latitudes.

Data and methods

iTraCE simulations during 20–11 ka

We used the isotope-enabled Transient Climate Experiment (iTraCE) which starts from the LGM (20 ka) and ended at early Holocene (11 ka)32,33. These simulations were conducted using the isotope-enabled Community Earth System Model, version 1.3 (iCESM 1.3), which can simulate the transport and transformation of water isotopes (e.g., H218O) across different components of the Earth system46. The atmosphere model used has a nominal 2° solution (1.9° in latitude and 2.5° in longitude) in the horizontal and 30 hybrid coordinate levels in the vertical. The ocean model has a nominal 1° horizontal resolution (gx1v6) and 60 vertical levels.

The baseline simulation was forced by changing ice-sheet configurations and land–sea configurations (ICE), based on the ICE-6G reconstruction24. Ice-sheet configurations were modified every 1000 years. The land–sea configurations were changed twice at 14 ka and 12 ka based on the sea level rise implied by the ICE-6G, applied to the present-day sea-floor topography. These changes in land–sea configurations mainly occurred in EA, the Maritime Continent (i.e., islands, peninsulas, and shallow seas of Southeast Asia), and Bering Strait. Subsequently, orbital insolation forcing was added (ICE + ORB), followed by greenhouse gas concentrations (ICE + ORB + GHG), and lastly meltwater fluxes (ICE + ORB + GHG + MWF) to generate the full forcing simulation (iTraCE). The effects of each single forcing were approximately isolated as follows: the ice sheet and land–sea configuration effect (ICE), orbital effect (ICE + ORB – ICE), GHG effect (ICE + ORB + GHG – ICE + ORB) and meltwater effect (iTraCE - ICE + ORB + GHG)32,33.

Sensitivity experiments of changes in the EA shelf area

To investigate the contribution of EA shelf area inundation, we conducted a group of sensitivity experiments using iCESM 1.3. First, we conducted an equilibrium experiment which was branched from the iTraCE ICE + ORB + GHG experiment at 11 ka, named EA_OCN (Table S1). The transient experiment of ICE + ORB + GHG has been integrated for 9000 years and we further integrated it for 100 years based on the restart files at 11 ka, with all boundary conditions consistent with the iTraCE simulation at 11 ka. We checked its top of the atmosphere (TOA) and ocean temperature conditions, which can be balance and similar to climatic conditions at 11 ka in the ICE + ORB + GHG experiment. Second, given that the major phase of EA continental shelf area inundation occurred between 14.5 and 11 ka, we conducted a sensitivity experiment (EA_LND) following the EA_OCN, in which the local ocean bathymetry of the EA continental shelf was lifted by about 60 m to expose it, based on the ICE-6G topography at 14.5 ka (Fig. 3a). The topography and ocean bathymetry in other regions were kept as 11 ka. All other boundary conditions were also set at 11 ka, consistent with the EA_OCN experiment. Surface properties for the newly exposed land areas in EA_LND were determined through nearest-neighbor extrapolation (mainly covered by C3 grass). This sensitivity experiment was integrated for 150 years to reach equilibrium (no significant oceanic circulation responses). The effect of EA shelf area inundation (EALSC) was approximated as the result of EA_OCN minus EA_LND.

To isolate the climatic impacts of changes in surface albedo and heat capacity owing to continental shelf area changes on EA temperature, we conducted a further sensitivity experiment (EA_OCN_LndAlb) following the EA_OCN. In this experiment, we maintained the same land‒sea configuration as that in the EA_OCN experiment but increased the surface albedo over the inundated shelf area from 5 to 30%, making it similar to the land albedo (Table S1). The climatic impact of the albedo change associated with the transition from land to ocean was then isolated using the results of EA_OCN minus EA_OCN_LndAlb, while the approximate contribution of heat capacity was isolated using the results of EA_OCN_LndAlb minus EA_LND.

Additionally, to help testify the climate impacts of EA continental shelf area inundation, we further extend the research period to 11–7.5 ka when the EA shelf area continued to be inundated. Similarly, following the EA_OCN, we lowered local topography by about 25 m according to the ICE-6G to make the land further inundated and conducted a sensitivity experiment which is integrated for 150 years. The effect of EA shelf area inundation during 11–7.5 ka was approximated as the result of this experiment minus EA_OCN. The results from the last 30 years in these experiment in this section were used for calculations.

Comparison with pollen-based temperature reconstruction

Temperature reconstructions were employed for comparison with the simulations. We used a pollen-based temperature reconstruction spanning the last 18.5 ka from Lake Tengchongqinghai (25.1°N, 98.6°E)19. In this study, 5 reconstruction models were used to perform quantitative reconstruction of mean temperature of the warmest month (MTWM) and of the coldest month (MTCM). The differences between MTWM and MTCM were calculated for comparison with simulated temperature seasonality changes. We also referred to reconstructions of MTWM and MTCM in approximately 50 sites in China during LGM based on an improved inverse vegetation model, which incorporated physiological processes combined with an augmented China Quaternary Pollen Database18. Reconstructions of the same variables during the LGM from a site in Mongolia17 were also employed.

Comparison with cave δ¹⁸Oc

We collected four δ18Oc records in Hulu36, Sanbao37, Haozhu38, Dongge cave39, respectively. They were linearly interpolated to a 500-year resolution and averaged to be compared with simulated regional averaged δ18Oc.

The simulated δ18Op should be converted to δ18Oc to be compared with cave records. First, the unit of δ18Op-SMOW is converted to δ18Op-PDB as47:

$${\delta }^{18}{O}_{p-{PDB}}=0.97002\times {\delta }^{18}{O}_{p-{SMOW}}-29.98.$$

Second, we use the temperature-dependent equilibrium equation of the inorganic carbonate oxygen isotope to derive δ18Oc48:

$${\delta }^{18}{O}_{c}={\delta }^{18}{O}_{p-{PDB}}+2.7\times {10}^{6}/{T}^{2}-3.29,$$

where T (°K) is the model annual mean surface temperature on the cave site.

The δ¹⁸Op changes can be decomposed to separate the isotopic composition effect and precipitation seasonality effect as follows:

$$\Delta {\delta }^{18}{O}_{p}=\sum \left(\frac{{P}_{i}}{P}\right)\Delta {\delta }^{18}{O}_{i}+\sum {\delta }^{18}{O}_{i}\Delta \left(\frac{{P}_{i}}{P}\right)$$

where \(i\) is calendar month, \(P\) is annual precipitation accumulation, the sum of \({P}_{i}\). The total \(\Delta {\delta }^{18}{O}_{p}\) is composed by the change in the value of the isotope \(\Delta {\delta }^{18}{O}_{i}\) in precipitation (isotopic composition effect) and the change associated with the precipitation weight \(\Delta (\frac{{P}_{i}}{P})\) (precipitation seasonality effect).

Temperature advection diagnosis

To help understand the mechanisms of temperature anomalies in EA, we used a temperature advection diagnosis method, dividing this variable into three terms49 as follows:

$$\frac{\partial {T}^{{\prime} }}{\partial t}=-{\boldsymbol{V}}{\boldsymbol{\cdot }}\nabla T=-{u}^{{\prime} }\frac{\partial \bar{T}}{\partial x}-{v}^{{\prime} }\frac{\partial \bar{T}}{\partial y}-\bar{u}\frac{\partial {T}^{{\prime} }}{\partial x}-\bar{v}\frac{\partial {T}^{{\prime} }}{\partial y}-{u}^{{\prime} }\frac{\partial {T}^{{\prime} }}{\partial x}-{v}^{{\prime} }\frac{\partial {T}^{{\prime} }}{\partial y}\,,$$

where, T represents temperature; u and v represent zonal wind and meridional wind, respectively; and overbars and primes represent climatological (EA_OCN) and anomalous variables, respectively. The first two terms on the right-hand side of the equation represent the advection of mean temperature by anomalous wind; the next two terms denote the advection of anomalous temperature by mean wind; and the last two terms are nonlinear eddy terms. 850–500 hPa averaged temperature and winds were used for this calculation.