Introduction

The Global Last Glacial Maximum (LGMG; 26.5–19 ka)1 and the last deglaciation (19–11 ka)2 denote the most recent global climate shift from full glacial to interglacial. To examine the cryosphere’s role in climate system during this period, detailed mapping and dating of moraine sequences have been conducted for the Laurentide Ice Sheet (LIS), Fennoscandian ice sheet, Alaska, and South Alps3,4,5,6. The Tibetan Plateau and its surrounding mountains (TPSM) were recognized as one of the tipping elements in the Earth’s climate system7 and contained the most extensive and largest number of glaciers outside the polar regions8. During the LGMG, summer temperature was 4–8 °C lower than preindustrial period9, the extended glacier and snow cover greatly influenced Asia climate10,11, including weakened Asian summer monsoon, along with strengthened Westerlies and Siberian High12. Reduced interior precipitation on the TPSM prevented the formation of a unified ice sheet during the LGMG, while ice caps and valley glaciers expanded, covering an area approximately four times larger than today13. The last deglaciation has experienced temperature recovery and a substantial atmospheric circulation reorganization over the TPSM, the main axis of the Westerlies remained south of the Tibetan Plateau during the LGMG12, by 16.5 ka, Indian Summer Monsoon (ISM) has replaced the domination of Westerly over Central Tibet14. These varied hydrothermal conditions potentially led to regional differences in glacier activities15. Terrestrial cosmogenic nuclide (TCN) exposure dating has been widely applied to constrain glacier activities on the TPSM since the twenty-first century16,17,18, however, due to the lack of high-resolution preserved moraine sequences since the LGMG, the obtained moraine ages from a single site could only be correlated with millennial-scale cold climate events such as Heinrich Stadial 1 (HS1) or the Younger Dryas19,20, a detailed history of glacier recession comparable with that from the high latitudes is still absent.

Nevertheless, the accumulation of 10Be age data across the TPSM provides the potential to identify spatiotemporal glacier activities and further explore their correlation with climate change. For instance, Owen et al. (2008)21 and Chevalier et al. (2011)16 partitioned the Himalaya-Tibetan orogen into different regions and produced probability density functions (PDFs) of 10Be ages from boulders and polished bedrocks or moraines respectively to compare the timing of long-term glaciations. Dortch et al. (2013)17 and Murari et al. (2014)18 used the Gaussian separation method to determine the age of individual location or moraine from bedrock and boulder 10Be ages, then applied the student’s t-test to identify independent regional glacial events. This new approach revealed five millennial-scale glacier events in the semi-arid regions17 and six in the monsoon-influenced regions18 of the TPSM between 26.5 and 10 ka. Using the similar method described in Dortch et al. (2013)17 and radial plot analysis, Saha et al. (2018.2019)22,23 identified seven Holocene glacial stages in the northwestern end of the Himalayan–Tibetan orogen22 and five Holocene glacial stages over Himalayan-Tibetan23.

In the past decade, there was a large increase in number of 10Be ages from glacial moraine boulders on the TPSM. For the moraine ages spanning 26.5–10 ka, the number of boulder ages increased from 590 in 2014 to 994 by 2024 (Supplementary, Table S1). All the data were recalculated using the default global average 10Be production rate24 and the LSDn scaling model25, and were registered in the Informal Cosmogenic Nuclide Exposure Age Database (ICE-D)26, last updated in March 2023. Here, we utilized the probabilistic cosmogenic age analysis tool (P-CAAT) developed by Dortch et al.27 to derive the moraine ages spanning 26.5–10 ka in this dataset from 994 10Be ages. The exposure age of a single moraine indicates the cessation of moraine construction due to the onset of glacier retreat rather than the beginning of its building28,29. We divided TPSM into seven regions and used the P-CAAT regional analysis function to generate the separated component Gaussians for each region moraines. Then, we compared the peak ages of Gaussians, along with their uncertainties, across the seven regions to reconstruct the spatiotemporalglacier recession process on the TPSM. By comparing the glacier activities with other paleoclimate records, we further explored the coevolution history of glaciers and atmospheric circulation adjustment on the TPSM since LGMG.

Study area

The TPSM, covering 5 million km2 (26–43°N, 67–103°E), is surrounded by the Himalayas, Hengduan Mountains (Hengduan Mts.), Qilian Mountains (Hengduan Mts.), Tianshan, and Pamir Plateau (Fig. 1). Its modern glacier coverage spans 1.2 × 105 km2, primarily over the Himalayas and Karakoram30. Due to its location and broad extension, the TPSM encompasses various climatic zones, including the East Asian summer monsoon (EASM) and ISM in the east and south, the Westerlies in the west, and the Siberian High-driven winter monsoon impacting the north (Fig. 1). Glacier types shift from temperate maritime in the southeast to cold-dry continental in the northwest (Fig. 1), influenced by regional variations in precipitation and temperature31. Since the 1980s, glaciers in monsoon-influenced regions like the Himalayas have receded significantly due to reduced precipitation and warming, while those in Westerlies-influenced regions like the eastern Pamir have remained stable or expanded due to increased precipitation32. In contrast, the Tianshan region, despite Westerly influence and increased rainfall, has experienced substantial glacier mass loss over the past 50 years due to rising temperatures33.

Fig. 1
figure 1

Topographic map of the Tibetan Plateau and surrounding mountains (based on ArcMap 10.4, Esri, 2015, https://desktop.arcgis.com/en/arcmap/10.4/), showing its major atmospheric circulation systems34. Locations of seven regions are enclosed by black dashed lines. Within these dashed lines, bold black font indicates the names of the seven regions, while regular black font denotes specific mountain ranges. The glacial deposit sites we compiled are shown by dots in different colors for these regions. The different glacier types at present are divided by solid black lines: (A) maritime glacier (temperate glacier)35; (B) polythermal glacier (sub-continental glacier); (C) cold-based glacier (continental glacier). Retreat phases across regions are shown as colored fan-shaped sectors: Phase I (22–20 ka), Phase II (19–18 ka), Phase III (17–16 ka), Phase IV (14.5–13 ka), and Phase V (11.6–10 ka). The black rectangles, triangles, and pentagons indicate the locations of lake, loess, and stalagmite records, respectively, used for correlation in this study. 1. Lake Karakul36; 2. Lake Tso Moriri37; 3. Mapam Yum Co38; 4. Tangra Yum Co39,40; 5. Nam Co14; 6. Mabu Co41; 7. Pumoyum Co42; 8. Mawmluh Cave43; 9. Lake Tiancai44,45; 10. Lake Naleng46; 11. Lake Qinghai47,48,49; 12. Lake Balikun50; 13. Loess section in the eastern Yili Basin51; 14. Loess section in the Western Tianshan, Kyrgyzstan52.

Full size image

Given the regional differences in glacier responses to climate, glaciers across the TPSM likely showed similar behavior during the last deglaciation, a period of significant climate change. To explore the relationship between glacier activity and atmospheric circulation during the LGMG and the last deglaciation, we divided the TPSM into seven regions: Tianshan, Pamir, W Himalaya, E Himalaya, Central Tibet, SE Tibet, and NE Tibet. This classification, based on Heyman (2014)53 with minor adjustments, combined E Pamir and W Pamir into a single Pamir region due to their shared Westerlies influence. It also merged W Himalayan and W Tibet into W Himalaya, considering their proximity and limited data for W Tibet between 26.5–10 ka. The area near the Yarlung Tsangpo Grand Canyon, one part of SE Tibet in Heyman (2014)53 and combined with the Central Himalaya, was now redefined as E Himalaya due to its unified ISM pathway through the Grand Canyon into the Yarlung Tsangpo valley (Fig. 1).

Methods

10Be exposure age compilation

Our compilation of 10Be exposure ages from the TPSM is entirely sourced from ICE-D26, with a total of 994 boulder ages from 196 moraines (Supplementary, Table S1). The dataset includes 112 boulders from 28 moraines in Tianshan, 134 boulders from 21 moraines in SE Tibet, 111 boulders from 20 moraines in NE Tibet, 109 boulders from 19 moraines in C Tibet, 222 boulders from 46 moraines in E Himalaya, 64 boulders from 14 moraines in W Himalaya, and 242 boulders from 48 moraines in Pamir (Supplementary, Table S1). ICE-D provides recalculated exposure ages based on the default global average 10Be production rate24 and the LSDn scaling model25, utilizing the CRONUS V3 platform. Bedrock exposure dating is more susceptible to nuclide inheritance, as it is often unclear whether prior exposure has been fully reset by subglacial erosion54. Owen et al. (2008)21observed that ice-polished bedrock ages in their dataset of 777 10Be ages might not accurately reflect the timing of glacial advances due to such geological uncertainties. Therefore, we selected moraines with ages between 26.5 and 10 ka, excluding those from bedrock; the remaining samples include cobbles, pebbles, and boulders. Considering the significant variation in the number of boulders on individual moraines on the TPSM, with most having fewer than five55, and since we aimed to apply a consistent method to calculate moraine ages, our dataset includes only moraines with three or more boulder ages, in accordance with the P-CAAT requirement of at least three boulders per site or moraine.

Recalculating individual moraine ages

Dortch et al. (2022)27 compared the performance of the P-CAAT approach with other common outlier statistical approaches in identifying outliers based on synthetic datasets developed by Applegate et al. (2012)56. The performance of these common outlier detection methods varies widely depending on the dataset size and the type of end-member scenarios (e.g. pre-depositional and post-depositional processes), while the P-CAAT approach consistently performs better than alternative approaches. Therefore, several recent articles on glacial exposure dating began to utilize this approach extensively20,57. P-CAAT method selects a set of internally consistent boulder ages that are most likely to represent the deglaciation age of the moraine, rather than using all ages or remaining ages after outliers are excluded, greatly reducing the error in the calculated moraine ages and making it possible to establish connections between different moraines within the specific region. P-CAAT provides three selectable bandwidth estimators (Mean, MADD, and STD/IQR) and generates a composite probability density estimate (PDE) of all boulder ages on a single moraine and separates the composite PDE into a series of component Gaussians based on the group clustering of boulder ages. With the premise of the bandwidth estimator already confirmed, once a component Gaussian is selected as the age of the moraine, the quantification of uncertainty for the moraine is obtained, including the peak age and its corresponding internal or external error.

In selecting the separated Gaussians to determine moraine ages, we referred to uncertainties in the geological context by referencing the moraine ages reported in the original publications. For moraines with unreported ages or limited geological context in the publications, we selected the highest-probability component Gaussian to represent the moraine age, following the Gaussian selection criteria proposed by Dortch et al. (2013)17 for LGMG and younger moraines. Ultimately, our dataset included 196 moraine ages (Supplementary, Table S1) and their corresponding PDEs (Supplementary, Fig. S1–S25).

P-CAAT regional analysis

After obtaining all the ages and age errors of moraines on the TPSM, we calculated regional moraine ages based on the ages of individual moraines included within each region. In calculating the regional moraine age, we used the P-CAAT regional analysis function, inputting the moraine ages and corresponding uncertainties with the LN mode and the MADD bandwidth selector, which generated multiple component Gaussians for each region. We adopted the peak age of the component Gaussian and its corresponding 1-sigma internal uncertainty to represent the regional moraine deglaciation age. This method was utilized for identifying regional glaciation events27.

Results and interpretation

Using the P-CAAT regional analysis function, we identified eight component Gaussians for the entire TPSM using the total 196 moraine ages (Supplementary, Table S1), indicating continuous but uneven glacier retreat from the LGMG to the Holocene on the TPSM (Fig. 2a). This pattern was consistent with that illustrated by the high-resolution geologic records of glacial retreat from the Southern Alps in New Zealand4 and Alaska5 during this period (Fig. 3d, e). It was evident this method is effective to reveal more complete regional glacier retreat events than that from geologic record of a single site19,20. The component Gaussians of the seven regions, i.e. the Pamir, Tianshan, SE Tibet, NE Tibet, C Tibet, E Himalaya, and W Himalaya were shown in Fig. 2b–h, respectively. Compared to the Gaussian distributions derived from all moraine ages across the entire TPSM, some subregional Gaussians were flatter, indicating larger errors of the Gaussian peak ages. As a result, independent Gaussians could not be distinguished in certain time periods, such as the Gaussian age of 19.9 ± 2.5 ka in the E Himalaya (Fig. 2g). Furthermore, some regions lacked Gaussian ages, such as Tianshan, which lacked a Gaussian age during 19–18 ka (Fig. 2c), and Pamir and NE Tibet, which lacked a Gaussian age during 17–16 ka (Fig. 2b, e).

Fig. 2
figure 2

Comparison of the component Gaussians of 10Be moraine ages from the TPSM and its seven subregions (10Be moraine ages from Supplement, Table S1). The black curve represents the summed Probability Density Estimate (PDE), while the differently colored shadings indicate the component Gaussians separated from the PDE. From top to bottom (a-h) are that for the TPSM, Pamir, Tianshan, SE Tibet, NE Tibet, Central Tibet, E Himalaya, and W Himalaya, respectively. The font colors corresponding to the Gaussians indicate their peak ages and 1-sigma internal uncertainties. The black box contains the bandwidth estimation pattern (MAAD) and their corresponding values, as well as the model fit quality (R2). The numbers in parentheses represent the probability of each component Gaussian. The five gray rectangles represent the five peak retreat stages we defined.

Full size image
Fig. 3
figure 3

Plots showing the peak ages of component Gaussians on the TPSM, compared with regional and global paleoclimate records. (a) The component Gaussians of 10Be moraine ages from the TPSM; (b) Regional glacial stages (semi-arid western Himalaya-Tibetan Stage: SWHTS)17; (c) Regional glacial stages (monsoon Himalayan–Tibetan stages: MOHITS)18; (d) Time-distance diagram of glaciers in the Revelation Mountains, Alaska5. Boulder ages were recalculated using the default global average 10Be production rate and the LSDn scaling model, with moraine ages corrected using the Bayesian age model; (e) Glacier-inferred temperatures reconstructed from Southern Alps glacier chronologies4. Boulder ages were recalculated using the default global average 10Be production rate and the LSDn scaling model; (f) Reconstructed global sea level58; (g) Atmospheric CO2 concentrations from the West Antarctic Ice Sheet (WAIS) Divide59; (h) The δ18O record of NGRIP ice core from Greenland60; (i) Summer July insolation curve for 30°N61; (j) Speleothem δ18O records from Mawmluh Cave in the northern Indian, negative value represents strong Indian summer monsoon and vice versa43; (k) Lake Qinghai Asian summer monsoon index (SMI)47; (l) Lake Qinghai Westerlies climate index (WI, flux of > 0.25 mm fraction)47; (m) Pollen discrimination index (PDI) record from Nam Co, central Tibetan Plateau14.

Full size image

We divided the glacier retreat phases based on independent Gaussian distributions on the TPSM. Since the Gaussian around 24 ka was observed only in the Pamir, Tianshan, and E Himalaya, while the 21 ka Gaussian was more widely distributed (Fig. 2b–h), we defined the first glacier retreat phase as 22–20 ka, based on the peak around 21 ka. The second phase was 19–18 ka, and the third phase was17–16 ka. The Gaussians at 14.5 ± 0.6 ka and 13.0 ± 0.7 ka were independent but both fall within the Bølling-Allerød period (B-A), so we combined them into the fourth phase (14.5–12.9 ka). Since the last two Gaussians during the early Holocene, the fifth phase was defined as 11.6–10 ka.

In the next section, we will explore in detail the synchronicity and differences in glacial retreat across the five stages on the TPSM, comparing them with the SWHTS17and MOHITS18 regional glacial stages (Fig. 3b–c).

Discussion

The first phase of glacier retreat (22–20 ka)

In various regions of the TPSM, widespread signals of glacier recession first appeared around 22–20 ka, although the 19.9 ± 2.5 ka age in the E Himalaya carries considerable uncertainty (Figs. 1, 2). An earlier peak around 24 ka was also observed in the Pamir, Tianshan, and E Himalaya (Fig. 2b, c, g). The 24 ka peak ages in our compiled moraine dataset primarily resulted from a single moraine with relatively small error (Supplementary, Figs. S22d, S3c, S12e). Glacier recession around 21 ka was widespread across TPSM (Figs. 2, 3a) and had previously been identified as the regional glacial stage MOHITS 2E in the monsoon Himalaya and SWHTS 2E in the semi-arid western Himalaya-Tibetan regions (Fig. 3b, c). The SWHTS 2E was based on data from 10 moraines in the W Himalaya and Pamir, while MOHITS 2E was derived from 9 moraines in SE Tibet, C Tibet, E Himalaya, W Himalaya, and NE Tibet (Supplementary, Table S2). This study also identified glacier retreat during this phase in the Tianshan, with a Gaussian age of 20.1 ± 1.0 ka (Fig. 2c). High-resolution moraine records from Alaska and the Southern Alps of New Zealand indicated that glaciers began retreating around 21 ka (Fig. 3d, e).

Low ISM intensity during this period was evident from δ18O records in Mawmluh Cave (Fig. 3j). Records from Lake Qinghai in NE Tibet indicated a weak Asian summer monsoon (SMI, Fig. 3k) and strong Westerlies (WI index, Fig. 3l). The high PDI values from Nam Co in central Tibet remained high from 24 to 16.5 ka, reflecting a higher proportion of airborne pollen carried by the Westerlies, indicating their strength (Fig. 3m). Additionally, high MAR in the Yili Basin within the Tianshan indicated the Siberian High’s strong dust-carrying capacity51.

In summary, the Siberian High influenced the Tianshan, while the Westerlies dominated the rest of the TPSM during this phase14,62. Despite the glacial conditions during this phase, characterized by low atmospheric CO₂ levels (Fig. 3g) and depleted δ18O in Greenland ice cores (Fig. 3h), global sea levels began rising from their low stand, suggesting a decrease in ice volume within the cryosphere (Fig. 3f). We suggest that increased local summer insolation likely triggered consistent early glacier retreat on the TPSM (Fig. 3i), resembling patterns observed in other high-latitude regions of the Northern Hemisphere63.

The second phase of glacier retreat (19–18 ka)

The Tianshan lacked Gaussian age for this phase, and the 19.9 ± 2.5 ka Gaussian age from the E Himalaya had large uncertainties, making it difficult to confirm glacier retreat (Figs. 1, 2c, g). The 17.6 ± 1.0 ka Gaussian age from SE Tibet and the 17.8 ± 1.1 ka Gaussian age from the W Himalaya (Figs. 1, 2d, 2h) fall between the 19–18 ka and 17–16 ka phases, potentially reflecting the limitations of P-CAAT in distinguishing independent Gaussian ages, as the age errors overlap with these intervals. Therefore, we interpreted that glacier retreat occurred in SE Tibet and W Himalaya during both the 19–18 ka and 17–16 ka phases. Pamir, NE Tibet, and C Tibet showed clear signals of glacier retreat (Figs. 1, 2b, e, f) during this phase. The MOHITS 2D report for this stage was based on data from 7 moraines in the W Himalaya, NE Tibet, SE Tibet, Tianshan, and C Tibet, excluding those from the E Himalaya (Supplementary, Table S2). The only moraine age from the Tianshan included in the MOHITS 2D report was 20 ± 4 ka, while we recalculated its Gaussian age as 15.6 ± 0.9 ka (Fig. S1c). Due to the large error in the SWHTS 2E age (Fig. 3b), Dortch et al. (2013)17 did not identify a corresponding regional glacial stage. However, our study detected a glacier retreat signal in the Pamir region, which is part of the semi-arid western Himalaya-Tibetan regions they defined, with a Gaussian age of 18.0 ± 1.2 ka (Fig. 2b). Moraine records from Alaska and the Southern Alps of New Zealand also showed glacier retreat around 18 ka (Fig. 3d, e).

In SE Tibet, summer temperatures reconstructed from sub-fossil chironomid head capsules in Lake Tiancai sediments showed warming around 18.5 ka (Fig. 4f), triggering rapid forest expansion around the lake45, supporting that glacier retreat occurred in this region. Although we could not definitively confirm glacier retreat in the E Himalaya during this period, a surge in sand content in Pumoyum Co sediments starting around 18.5 ka indicated glacial meltwater input (Fig. 4g). In the Tianshan, despite the absence of glacier retreat signals for this phase (Fig. 2c), rising summer lake temperatures at Lake Balikun indicated local warming (Fig. 5a). At Mabu Co in the E Himalaya, high PCA1 values suggested a dry climate controlled by the Westerlies (Fig. 4h). High PDI values at Nam Co in Central Tibet (Fig. 3m) and high WI values in NE Tibet (Fig. 3l) both indicated stronger Westerlies. In the Tianshan, the persistence of the Siberian High was evidenced by high MAR values in the Yili Basin (Fig. 5b). Overall, the atmospheric pattern across the TPSM during this phase remained consistent with the previous phase, with temperature records from multiple subregions showing warming.

Fig. 4
figure 4

Comparison of paleoclimate records for four regions of the TPSM (Central Tibet, W Himalaya, SE Tibet, E Himalaya). (a) Lake volume reconstruction of Tangra Yum Co, Central Tibet39; (b) Principal Component Analysis (PCA) scores of the first axis for the pollen samples from Tangra Yum Co, Central Tibet, and the negative values related to humid conditions and positive values related to arid conditions40; (c) Lake level reconstruction of the Lake Tso Moriri, W Himalaya37; (d) Total organic carbon concentration in the sediment of Mapam Yum Co, W Himalaya38; (e) Cyperaceae percentages indicative for moisture regime from Lake Naleng, SE Tibet46; (f) The chironomid-inferred mean July temperature of the Lake Tiancai in SE Tibet44; (g) Sand content change in PY104PC of Pumoyum Co, E Himalaya42; (h) Principal Component Analysis (PCA) scores of the first axis for the pollen samples from Mabu Co, E Himalaya, and the negative values related to humid conditions and positive values related to arid condition41.

Full size image
Fig. 5
figure 5

Comparison of paleoclimate records for four regions of the TPSM (Tianshan, NE Tibet, SE Tibet, Pamir). (a) The reconstructed summer temperature of Lake Balikun in the Tianshan50; (b) The stacked dust mass accumulation rate (MAR) in the eastern Yili Basin51; (c) The moisture proxy ΔL* from Bishkek (BSK) loess section in the Western Tianshan, Kyrgyzstan52; (d) The δ2Hwax record, after removing the influences of temperature and global ice volume (hereafter, δ2Hwax-TIV), from Lake Qinghai49; (e) Alkenones-inferred summer temperature record of Lake Qinghai48; (f) Inferred salinity changes based on alkenone abundance from Lake Karakul36; (g) Inferred temperature changes based on Uk’37 proxy from Lake Karakul36.

Full size image

In summary, although we did not identify a glacier retreat signal in the Tianshan during this stage, local summer temperature records from the region indicated that the climate was relatively warm (Fig. 5a). Considering that glacier retreat in the other six regions was broadly consistent in timing with the rise in atmospheric CO2 around 18.1 ka (Fig. 3g), we attributed the TPSM glacier retreat during this phase to the onset of CO2 rise and its impact on global climate warming, like other researchers4,5.

The third phase of glacier retreat (17–16 ka)

In the third phase, Gaussian ages were absent for the Pamir and NE Tibet (Figs. 1, 2b, e). The Gaussian ages of 19.9 ± 2.5 ka and 15.3 ± 1.2 ka from the E Himalaya had large errors, making it difficult to determine if glacier retreat occurred during this phase (Figs. 1, 2g). Based on the interpretation in section "The second phase of glacier retreat (19–18 ka)" of the 17.6 ± 1.0 ka Gaussian age from SE Tibet and the 17.8 ± 1.1 ka Gaussian age from the W Himalaya, we interpreted that Tianshan, SE Tibet, C Tibet, and W Himalay all experienced glacier retreat during this phase (Figs. 1, 2c, d, f, h). The MOHITS 2B report for this stage was based on data from 14 moraines in the W Himalaya, E Himalaya, NE Tibet, SE Tibet, and C Tibet (Supplementary, Table S2). Notably, the MOHITS 2B report listed the only moraine age from NE Tibet as 16.6 ± 4.5 ka, while our recalculated Gaussian age is 18.5 ± 3.3 ka (Fig. S7h). The SWHTS 2D report for this stage relied on data from two moraines in the Tianshan and Pamir (Supplementary, Table S2). The moraine age from the Pamir was 16.6 ± 0.3 k, calculated using the St scaling model64 and Balco et al.65 calibration data, which may have resulted in a nearly 2 ka difference compared to our recalculated moraine age of 18.3 ± 0.6 ka (Fig. S23b). Moraine records from Alaska and the Southern Alps of New Zealand also showed glacier retreat during this phase (Fig. 3d, e).

In Central Tibet, glacier meltwater influx during this phase raised the volume of Tangra Yum Co (Fig. 4a) and increased water levels in Lake Tso Moriri in the W Himalaya after 16.4 ka (Fig. 4c). Although we could not confirm glacier retreat in the E Himalaya during this phase, the Mabu Co lake level record showed glacial meltwater input beginning at 16.8 ka66. Depleted stalagmite δ18O values from Mawmluh Cave indicated a strengthening of the ISM (Fig. 3j). Pollen records from Mabu Co in the E Himalaya showed a shift to Cyperaceae-dominated alpine meadows after 17 ka, driven by the strengthened ISM (Fig. 4h). In Central Tibet, the low PDI values from Nam Co indicated a transition of the local atmospheric circulation from Westerlies to ISM dominance (Fig. 3m). Meanwhile, Lake Tiancai in SE Tibet recorded a rapid summer temperature rise between 16.4 and 16.1 ka, reflecting a warmer climate (Fig. 4f). In summary, glacier retreat in the E Himalaya, W Himalaya, SE Tibet, and C Tibet was synchronous with the strengthening of the local ISM. This retreat phase occurred during the early part of Heinrich Stadial 1 (HS1, 17.5–14.7 ka)67. Denton et al. (2005)68 proposed that the intense seasonality of HS1, with warm summers driving glacier retreat, was supported by recent exposure dating studies69,70. Given that chironomid records from SE Tibet confirmed summer warming during this period (Fig. 4f), we support Denton et al. (2005)68 hypothesis, which explained the glacier retreat on the TPSM.

In NE Tibet, the high WI values and low SMI values from Lake Qinghai during this phase suggested a weakened ISM and strengthened Westerlies (Fig. 3k, i). The δ2H wax-TIV proxy, reflecting the source water of Qinghai Lake’s aquatic plants, showed depleted values, indicating that local precipitation primarily originated from snowmelt of winter snowfall brought by the Westerlies (Fig. 5d). In Pamir, Aichner et al. (2019)36 linked the decrease in salinity (Fig. 5f) and temperature decline (Fig. 5g) in Lake Karakul to a colder, wetter climate influenced by Westerlies, likely driven by increased water vapor from the retreat of the European ice sheet and the expansion of the Black and Caspian Seas71. In Tianshan, summer temperatures at Lake Balikun (Fig. 5a) rose transiently between 17 and 16 ka, while high MAR values in the Yili Basin (Fig. 5b) and dry conditions in western Tianshan (Fig. 5c) suggested a continued strong Siberian High. Based on the analysis of local climate in NE Tibet and Pamir, no glacier retreat signal was observed in these regions (Fig. 2b, e), likely due to the control of the colder and wetter Westerlies in this phase. In contrast, the glacier retreat in Tianshan (Fig. 2c) was attributed to dry Siberian air masses and rising summer temperatures (Fig. 5a, b). During this phase, the ISM reached the southern TPSM but had not yet penetrated NE Tibet and Pamir, while the Siberian High continued to influence Tianshan.

The fourth phase of glacier retreat (14.5–12.9 ka)

During the fourth phase, Gaussian ages were present in all seven subregions (Figs. 1, 2). Pamir, Tianshan, and SE Tibet each had one Gaussian age (Fig. 2b–d), while C Tibet and W Himalaya showed two independent Gaussians (Fig. 2f, h). NE Tibet and E Himalaya may have had two potential Gaussians (Fig. 2e, g), but the 12.3 ± 1.3 ka age from NE Tibet and 15.3 ± 1.2 ka age from E Himalaya introduced uncertainty in classifying the retreat phase. The MOHITS 2A report for this stage was based on data from 10 moraines in the W Himalaya, E Himalaya, NE Tibet and C Tibet, but not SE Tibet, while the SWHTS 2C and 2B reports used data from 4 moraines in the Pamir, Tianshan, W Himalaya (Supplementary, Table S2). In contrast, this study also identified a Gaussian age of 14.3 ± 0.6 ka for SE Tibet (Fig. 2d). Moraine records from Alaska and the Southern Alps of New Zealand both showed retreat during this phase, with a faster rate in Alaska and a slower rate in the Southern Alps (Fig. 3d, e).

During this interval, significantly depleted stalagmite δ18O values from Mawmluh Cave indicated an enhanced ISM (Fig. 3j). The low PCA1 values at Lake Tangra Yumco in Central Tibet (Fig. 4b) and the increased Cyperaceae ratio in pollen records from Lake Naleng in SE Tibet (Fig. 4e) suggested a shift to more humid vegetation, reflecting increased precipitation from the ISM. Chironomid records from Lake Tiancai in SE Tibet indicated relatively high local summer temperatures (Fig. 4f). In the W Himalaya, the high total organic carbon (TOC) percentages in Mapam Yum Co (Fig. 4d) signaled warmer, wetter conditions due to the strengthened ISM. In Pamir, the increased evaporation and salinity (Fig. 5f) was attributed by Aichner et al. (2019)36 to rising temperature (Fig. 5g), and the more depleted leaf-wax δD values were linked to a strengthened ISM bringing precipitation36. In NE Tibet, alkenone records from Lake Qinghai indicated higher temperatures (Fig. 5e), while δ2H wax-TIV values showed rapid depletion, suggesting that ISM moisture reached the region by 13.6 ka (Fig. 5d). In Tianshan, higher summer temperatures at Lake Balikun (Fig. 5a) and Zhao et al.(2015)72 suggested a warm, wet climate, consistent with the disappearance of the Grigoriev ice cap and the emergence of organic-rich soil beneath it73. After 15 ka, the weakening of the Siberian High, indicated by lower MAR values in the Yili Basin (Fig. 5b) and increased humidity in western Tianshan (Fig. 5c), suggested that the Westerlies had become the dominant circulation in Tianshan52.

In summary, the ISM had advanced north into the TPSM, affecting regions beyond Tianshan. Additionally, whether in regions influenced by the ISM or the Westerlies, all paleoclimate proxies indicated an increase in precipitation. Therefore, the glacier retreat signals observed in the seven regions (Fig. 2) likely reflected that the high temperatures during this phase (Figs. 4f, 5a, d, f) exceeded the increased precipitation’s supply to the glaciers, leading to their retreat. This hypothesis was supported by records of glacial meltwater input into Lake Qinghai, in the NE Tibet74, and into lakes on the northwestern Tibetan Plateau75. Meanwhile, we suggest that the warming in the TPSM was driven by a revived Atlantic Meridional Overturning Circulation (AMOC)76, reflecting the Northern Hemisphere’s B-A warming.

The fifth phase of glacier retreat (11.6–10 ka)

Pamir, Tianshan, SE Tibet, E Himalaya, and W Himalaya all presented Gaussian ages during this phase (Fig. 2), with E Himalaya showing two distinct ages (Figs. 1, 2g), while the others each had one (Figs. 1, 2b–d, h). C Tibet lacked Gaussian ages (Figs. 1, 2f), and the 12.3 ± 1.3 ka Gaussian age in NE Tibet could not be clearly associated with this stage (Figs. 1, 2e). The MOHITS 1 k and 1j reports for this stage was based on data from 13 moraines in the W Himalaya, E Himalaya, NE Tibet, SE Tibet, and C Tibet (Supplementary, Table S2). In Table 3 of Murari et al. (2014)18, the MOHITS 1 k category listed two moraine ages of 10.6 ± 0.6 ka. Upon verification, we found that one belonged to E Himalaya, while the other, the sole moraine from C Tibet in this category, should have been 19.9 ± 1.4 ka, likely due to a presentation error (Supplementary, Table S2). Therefore, Murari et al. (2014)18 also found no C Tibet moraines for this stage. The SWHTS 2A report for this stage was based on data from five moraines in the Pamir and W Himalaya (Supplementary, Table S2), while we also identified a Gaussian age of 11.3 ± 0.4 ka from the Tianshan (Fig. 2c). During this phase, glaciers in Alaska and the Southern Alps of New Zealand had retreated close to their cirques (Fig. 3d, e). To test whether the absence of glacier retreat records in Central Tibet was due to limited exposure-age data from early Holocene moraines located at higher, less accessible areas, we surveyed glacial valleys where fourth-phase moraines had been reported using satellite imagery. For instance, in Pagele Valley, Xu et al. (2020)77 sampled the PM2 moraine, yielding a P-CAAT age of 14.1 ± 1.9 ka (Supplementary, Table S1), and we found several additional terminal moraines were found about 1 km upstream.

During this phase, summer temperatures at Lake Tiancai in SE Tibet were elevated (Fig. 4f). The low PCA1 values from Mabu Co in the E Himalaya (Fig. 4h) and the higher TOC values from Mapam Yum Co in the W Himalaya (Fig. 4d) both indicated humid conditions influenced by the ISM. In Central Tibet, despite the lack of reliable temperature records78, the low PCA1 values from Lake Tangra Yumco (Fig. 4b) implied a warm, humid climate. Lake Karakul in the Pamir experienced higher temperatures (Fig. 5g), and Heinecke et al. (2016)79 noted that the intensified ISM strengthened local physical weathering. In NE Tibet, summer temperatures at Lake Qinghai were elevated (Fig. 5e), while depleted δ2H wax-TIV values (Fig. 5d) indicated increased ISM precipitation. In Tianshan, summer temperatures at Lake Balikun were higher (Fig. 5a) and while increased humidity influenced by the Westerlies was observed in western Tianshan (Fig. 5c).

In summary, although we did not directly observe glacial retreat signals in C Tibet, paleoclimate analysis and a survey of undated moraines using satellite imagery suggested that the entire TPSM was in a retreating state during this phase, with atmospheric circulation resembling the B-A period. The retreat likely reflected a high-temperature climate, driven by the combined forcings of maximum Northern Hemisphere summer insolation (Fig. 3i) and interglacial CO2 concentrations (Fig. 3g).

Conclusions

We divided the TPSM into seven regions and compiled recalculated 10Be exposure ages from 196 glacial moraines (994 individual samples) using the LSDn model, spanning 26.5–10 ka, to investigate glacier retreat history. Using P-CAAT regional analysis, we calculated component Gaussians of moraine ages and compared peaks across regions to identify synchronous or asynchronous glacier activity phases. The results reveal continuous but uneven glacier retreat from the LGMG to the Holocene, concentrated in five phases: 22–20 ka, 19–18 ka, 17–16 ka, 14.5–12.9 ka, and 11.6–10 ka. The 22–20 ka and 14.5–12.9 ka retreat phases occurred throughout the TPSM, while the other phases showed regional differences. During 22–20 ka and 19–18 ka, glaciers retreated in response to increasing local summer insolation and onset of CO2 rising, respectively, under a Westerly-dominated atmospheric circulation on the TPSM. The 17–16 ka phase marked a shift to ISM dominance, with glaciers in NE Tibet and Pamir remaining stable, contrasting with retreat in other regions, possibly due to lingering Westerly influence. The 14.5–12.9 ka phase of glacier retreat responded to the B-A warming and the recovery of the AMOC, with ISM dominance in the TPSM atmospheric circulation. The 11.6–10 ka phase of glacier retreat was driven by maximum summer insolation and interglacial levels of atmospheric CO2, but was absent in Central Tibet, likely due to the lack of dating for remote and perched moraines. This study identifies five distinct retreat phases between 26.5 and 10 ka, aiding simulations of glacier and paleoclimate changes on the TPSM during this period.

In the future, further 10Be exposure dating, particularly of high-elevation early Holocene moraines, is needed. Additionally, high-resolution glacier retreat rate or ice volume reconstructions from precise moraine dating will provide insights into deglaciation history and its link to global climate change.