A High-Resolution Absolute-Dated Late Pleistocene Monsoon Record from Hulu Cave, China

See allHide authors and affiliations

Science  14 Dec 2001:
Vol. 294, Issue 5550, pp. 2345-2348
DOI: 10.1126/science.1064618

This article has a correction. Please see:


Oxygen isotope records of five stalagmites from Hulu Cave near Nanjing bear a remarkable resemblance to oxygen isotope records from Greenland ice cores, suggesting that East Asian Monsoon intensity changed in concert with Greenland temperature between 11,000 and 75,000 years before the present (yr. B.P.). Between 11,000 and 30,000 yr. B.P., the timing of changes in the monsoon, as established with 230Th dates, generally agrees with the timing of temperature changes from the Greenland Ice Sheet Project Two (GISP2) core, which supports GISP2's chronology in this interval. Our record links North Atlantic climate with the meridional transport of heat and moisture from the warmest part of the ocean where the summer East Asian Monsoon originates.

The Asian and Australian Monsoons are important because they transport heat and moisture from the warmest part of the tropical ocean (the West Pacific Warm Pool) across the equator and to higher latitudes. The East Asian Monsoon, a sub-system of the Asian Monsoon, affects an area east of the Bay of Bengal and the Tibetan Plateau (1). Spring heating of Asia initiates the summer monsoon, which transports northward moisture and heat from north of Australia across the Warm Pool, as far as northern China. The winter monsoon is characterized by cold, dry Siberian air flowing southward across eastern China, ultimately contributing to the Australian summer monsoon (1).

Plausible factors affecting the monsoon are orbitally controlled changes in insolation (2, 3), shifts in sea level causing changes in Warm Pool surface area (4), and circulation changes internal to the climate system (5). Loess records (6, 7) show clear evidence for monsoon changes (1) that are possibly linked to global climate (7) and Heinrich Events (5). However, resolution and dating problems limit the loess work. We reconstruct monsoon history with the oxygen isotopic composition of speleothem calcite, which has key advantages over many archives of past conditions. Well-chosen inorganic calcite can be dated precisely (8,9) with mass spectrometric 230Th methods (10). Speleothems may form continuously for tens of thousands of years and can be sampled at high resolution for dating and δ18O analysis.

We collected five stalagmites from 35 m depth in Hulu Cave, 28 km east of Nanjing (32°30′N,119°10′E). We halved samples along growth axes and sub-sampled on cut surfaces for 230Th dating by thermal ionization (10) and inductively coupled plasma mass spectroscopy (11, 12) and δ18O analysis (13). Fifty-nine 230Th dates (Fig. 1) (14), all in stratigraphic order, have analytical errors equivalent to about ±150 years at 10,000 years and ±400 years at 60,000 years. The oldest age is 74,875 ± 1,010 yr. B.P. (relative to 1950 A.D.) and the youngest is 10,933 ±160 yr. B.P., with at least one stalagmite active during all intervening times. Sample YT has visible banding throughout, and three 230Th ages with errors of ±60 to ±90 years. Numbers of bands are equal to differences in age between dated sub-samples, indicating that the banding is annual. We established the time scale for sample H82 with two 230Th ages (Fig. 2) (14), by matching its oxygen isotope record to YT's at 14.6 thousand years ago (ka), and by band counting between 11.0 and 11.8 ka and between 13.0 and 14.6 ka.

Figure 1

δ18O of Hulu Cave stalagmites (purple, green, and red) and Greenland Ice (22) (dark blue) and insolation at 33°N averaged over the months of June, July, and August (20, 21) (black) versus time.230Th ages and errors are color-coded by stalagmite. Numbers indicate GISs and correlated events at Hulu Cave. The YD and Heinrich events are depicted with vertical bars (24). The brown and blue bars indicate two possible correlations to H5. The average number of years per δ18O analysis is 130 for MSD and 140 for MSL. The δ18O scales are reversed for Hulu (increasing down) as compared with Greenland (increasing up).

Figure 2

δ18O of Hulu stalagmites (purple, black, and blue) and Greenland Ice (22) (dark blue shows 20-year averages; gray shows 3-year averages) versus time. Yellow bands indicate the timing and duration of the YD and the transition into the BA (t-BA); the BA is the interval between the yellow bands. 230Th ages and errors are color-coded by stalagmite. The chronology of YT and most of the chronology of H82 are fixed by annual banding. As YT and H82 are more precisely and continuously dated than PD, we adjusted the time scale of PD between 17 and 14 ka to match the major δ18O features. The slight adjustment is well within the errors of the PD time scale. The thin δ18O trace in this interval depicts the PD record based solely on its own 230Th dates. The average number of years per δ18O analysis is 60 for PD, 9 for YT, and 7 for H82.

A key issue is whether calcite δ18O can be interpreted solely in terms of the δ18O of meteoric precipitation and equilibrium fractionation during calcite precipitation. Hendy (15) described additional processes (e.g., kinetic fractionation) that could also affect δ18O. A robust test is the comparison of δ18O for contemporaneous stalagmites from the same cave (16). If the records replicate, the net effect of additional processes on δ18O must have been the same. Consistent offsets are unlikely because each stalagmite-precipitating drip has a unique combination of flow path, CO2 partial pressure, residence time, concentration of solutes, and degassing history. Thus, replicated records strongly suggest that such additional processes are not important. With modern dating and δ18O measurement techniques, the replication test can be made with high resolution and little temporal ambiguity.

Stalagmites MSD and MSL (Fig. 1) (14) grew contemporaneously between 53 and 36 ka. Considering dating errors and resolution differences, the records are virtually identical over this interval. At least two stalagmites (out of PD, YT, and H82) grew contemporaneously for all times between 17 and 10.5 ka (Figs. 1 and 2) (14). Samples YT and H82, both sampled at high resolution, have an overlapping section with δ18O values that replicate. Sample PD was sampled at lower resolution and has δ18O values offset from the others by a small amount (about 0.5‰) compared with the 5‰ amplitude of the record. Nevertheless, the pattern of δ18O variation is similar among overlapping sections of all five stalagmites, suggesting we can treat them as replicated records. We also tested for positive correlation between δ13C and δ18O values, plausibly indicative of kinetic fractionation (15). R2 values are either low or the correlation is negative for each of six data groupings [data for each of five speleothems grouped individually and as one group; see (14)], thus showing no evidence for kinetic fractionation.

Given records that replicate and the lack of a clear positive correlation between δ13C and δ18O values, the issue becomes how to interpret the record in terms of the δ18O of precipitation and temperature. Mean annual rainfall and temperature at Hulu Cave are 1015 mm and 15.4°C. The summer monsoon (June to September) contributes 80% of annual precipitation with δ18OVSMOW of –9‰ to –13‰. The rest comes during the winter monsoon with δ18O about 10‰ higher (–3 to +2‰) (17). Because of the large seasonal difference, a mechanism that may explain large changes in past mean annual δ18O of precipitation is a change in the ratio of the amount of summer to winter precipitation. Temperature effects are likely to be small because changes in the temperature-dependent fractionation between calcite and water are small (on the order of −0.25‰/°C) (18). On the basis of modern data (19), the effects of summer temperature and rainfall amount on mean δ18O of summer precipitation are also small, with similar relations holding for winter (19).

Hulu Cave δ18O values range from –4 to –9‰ between 75 and 10 ka (14). The large range suggests that a primary control is variance in the summer/winter precipitation ratio. If so, a change in the ratio by a factor of 3 (from today's value of 4 to 1.3) is required to explain the 5‰ amplitude. This factor is likely an upper limit because temperature and amount effects may also contribute.

The long-term Hulu trend (Fig. 1) appears to follow summer (integrated over June, July, and August) insolation (20, 21) at Hulu Cave (33°N), at least for a good portion of the record, suggesting that high summer insolation increases the continent-ocean temperature difference, enhancing the summer monsoon (2,3). However, the record is punctuated by numerous millennial-scale events and by shifts in δ18O over centuries or less, much shorter than orbital time scales.

These features resemble the Greenland ice-core δ18O records (22, 23). If analogous features do represent coincident events, Greenland temperature correlates positively with the summer/winter precipitation ratio in eastern China. To the extent that changes in the ratio result from changes in summer precipitation, warmer Greenland temperatures correlate with a more intense summer East Asian Monsoon. Between 10 and 15 ka, an interval for which ice core chronologies are robust, we can test for synchronicity. In this interval, samples PD and H82 (Fig. 2) exhibit features similar to the Younger Dryas (YD) and the Bolling-Allerod (BA). On the basis of independent time scales, these features are synchronous within errors (Fig. 2), demonstrating a link between the East Asian Monsoon and Greenland temperatures.

In detail, there are both similarities and differences in the Hulu and Greenland deglacial sequences (22, 23). The Hulu record has a sharp increase (about 2‰) in δ18O at 16,073 ± 60 yr. B.P., which takes place in <20 years, at about the time of Heinrich Event 1 (H1) (24). A similar feature is not apparent in the ice records. The slopes of the records during the BA differ, which could result, in part, from the decrease in marine δ18O associated with glacial melting, because this would affect the records in opposite senses. The most rapid portion of the transition into the BA appears to be more gradual at Hulu (180 years by band counting centered at 14,645 ± 60 yr. B.P.) as compared with Greenland (about 100 years centered at about 14,680 ± 290 yr. B.P. in GISP2). In contrast, the transition at the beginning of the YD (12,823 ± 60 yr. B.P. at Hulu and 12,880 ± 260 yr. B.P. in GISP2) is of similar short duration (<20 years). The transition ending the YD [11,473 ± 100 yr. B.P. at Hulu, 11,550 ± 70 yr. B.P. in the Greenland Ice Core Project (GRIP) (23), and 11,640 ± 250 yr. B.P. in GISP2 (22)] is also extremely rapid at both localities (<10 years). The duration of the YD as recorded at Hulu (1350 ± 120 years) is the same within error as its duration in Greenland.

Given the apparent synchronicity between Hulu and Greenland for times when the ice core chronologies are robust, we may correlate older events, for which ice core chronologies are less certain (Fig. 1). This correlation appears straightforward as far back as 38 ka. The highest Hulu δ18O values (at 16,032 ± 60 yr. B.P.) correspond to low Greenland temperatures associated with H1. Low Hulu δ18O values (at 23,310 ±100 yr. B.P.) correspond to Greenland Interstadial (GIS) 2, which is immediately preceded by high δ18O values (24,180 ±100 yr. B.P.) corresponding to low Greenland temperatures associated with H2. Low δ18O excursions correspond to GIS 3 through 8 (Fig. 1). With these correlations, we assign times to Greenland events with Hulu ages. For times between GIS 1 and 8, age offsets between GISP2 and Hulu are less than several hundred years, whereas offsets between GRIP and Hulu increase progressively from 0 at 15 ka to 3000 years at 30 ka (Fig. 3). For this interval, GISP2's time scale, determined by band counting (25), appears robust. For times older than ∼15 ka, the GRIP time scale was determined by flow modeling, with primary accumulation rates estimated by ice δ18O values (23). In Greenland, the 15 to 30 ka interval is characterized by low δ18O and low accumulation rates. Of the many factors that contribute to the construction of each time scale, we cannot uniquely identify those that contribute to discrepancies between time scales. However, use of a somewhat lower accumulation rate for GRIP in this cold interval would largely reconcile the three chronologies between 15 and 30 ka.

Figure 3

Difference in age between Hulu and Greenland ice core time scales [GRIP, (23); GISP2, (25)] versus Hulu age. Each point is based on a correlation between GRIP or GISP2 and Hulu (e.g., the correlations between Hulu and the GIS's depicted in Fig. 1). Positive values are times when the ice core age is less than the Hulu age; negative values are times when the ice core age is more than the Hulu age. Typical 230Th dating errors as a function of time are illustrated by the gray error envelope where the difference in age is 0. Error bars are estimates of the error in the ice core chronologies (23,25).

For GIS 9 through 13, we present two possible correlations. Those represented by blue numbers (Fig. 1) appear to follow logically as we correlate peaks back from GIS 8. However, this correlation places the beginning of GIS 12 (the end of H5) at 48 ka. Both GISP2 (25) and GRIP (23) as well two other high-precision stalagmite records place the end of H5 at ∼45 ka (26, 27). The differences among stalagmite records [Hulu, Sorel Cave, Israel (26), and Crevice Cave, Missouri (27)] highlight important regional differences in past climate at this time. If we take the end of H5 at 45 ka as a tie point, we obtain the correlation depicted by the brown numbers. At present, we cannot distinguish between the two. Beyond GIS 13, the correlation appears straightforward and is consistent with the only other high-resolution speleothem correlation that covers this whole time range (27). The oldest part of our record correlates to the end of GIS 21.

The Hulu record identifies a link between the East Asian Monsoon and North Atlantic climate and supports the idea that millennial-scale events first identified in Greenland are hemispheric or wider in extent (28–32). The Greenland events have been explained by changing rates of North Atlantic deep water formation, resulting in changing heat transport to the North Atlantic (28, 33). The millennial-scale changes that we observe may result similarly from massive and rapid changes in oceanic and atmospheric circulation patterns. The temporal relations between the Hulu and Greenland deglacial sequences are consistent with North Atlantic events that trigger large-scale circulation changes (28). Regardless of the trigger, our observations are consistent with the idea that Northern Hemisphere atmospheric circulation patterns are more meridional in character during Greenland interstadials and are more zonal during stadials. Our data support the idea that changes in the East Asian Monsoon are integral to millennial-scale changes in atmospheric/oceanic circulation patterns and are affected by orbitally induced insolation variations; however, our data do not show clear evidence that sea level itself has an observable effect on East Asian Monsoon intensity.

  • * To whom correspondence should be addressed: E-mail: edwar001{at}


View Abstract

Stay Connected to Science

Navigate This Article