Quantification of Holocene Asian monsoon rainfall from spatially separated cave records

A reconstruction of Holocene rainfall is presented for southwest China — an area prone to drought and flooding due to variability in the East Asian monsoon. The reconstruction is derived by comparing a new high-resolution stalagmite δ 18 O record with an existing record from the same moisture transport pathway. The new record is from Heshang Cave (30°27 ′ N, 110°25 ′ E; 294 m) and shows no sign of kinetic or evaporative effects so can be reliably interpreted as a record of local rainfall composition and temperature. Heshang lies 600 km downwind from Dongge Cave which has a published high-resolution δ 18 O record (Wang, Y.J., Cheng, H., Edwards, R.L., He, Y.Q., Kong, X.G., An, Z.S., Wu, J.Y., Kelly, M.J., Dykoski, C.A., Li, X.D., 2005. The Holocene Asian monsoon: links to solar changes and North Atlantic climate. Science 308, 854 – 857). By differencing co-eval δ 18 O values for the two caves, secondary controls on δ 18 O (e.g. moisture source, moisture transport, non-local rainfall, temperature) are circumvented and the resulting Δ δ 18 O signal is controlled directly by the amount of rain falling between the two sites. This is confirmed by comparison with rainfall data from the instrumental record, which also allows a calibration of the Δ δ 18 O proxy. The calibrated Δ δ 18 O record provides a quantitative history of rainfall in southwest China which demonstrates that rainfall was 8% higher than today during the Holocene climatic optimum ( ≈ 6 ka), but only 3% higher during the early Holocene. Significant multi-centennial variability also occurred, with notable dry periods at 8.2 ka, 4.8 – 4.1 ka, 3.7 – 3.1 ka, 1.4 – 1.0 ka and during the Little Ice Age. This Holocene rainfall record provides a good target with which to test climate models. The approach used here, of combining stalagmite records from more than one location, will also allow quantification of rainfall patterns for past times in other regions.

with present climate models. Quantitative records of past rainfall, which extend the relatively short instrumental record, would offer significant insight into natural rainfall variability and the pattern of rainfall under different climate states. Such records are particularly important for regions impacted by major and potentially variable climate systems such as the Asian monsoon, a region where prediction of future rainfall is difficult (IPCC, 2007).
Stalagmites can provide precisely-dated records of δ 18 Oa geochemical proxy that, in regions dominated by atmospheric convection, is strongly influenced by the amount of rainfall (Dansgaard, 1964). Several highresolution δ 18 O records have recently been published from such regions including China (Wang et al., 2005(Wang et al., , 2001Dykoski et al., 2005), Oman Fleitmann et al., 2003), and Brazil (Wang et al., 2004). Such records from the region impacted by the East Asian monsoon indicate dramatic changes of rainfall associated with cool events in the North Atlantic region (Wang et al., 2001) and the possible influence of solar variation in controlling monsoon strength (Wang et al., 2005).
Converting stalagmite δ 18 O records into a quantitative assessment of past rainfall amount is difficult, however, because δ 18 O in cave carbonates is also controlled by other environmental variables (Bar-Matthews et al., 1999;McDermott, 2004) including cave temperature, local evaporation, the δ 18 O of the source, and the transport distance from the source. In the Asian monsoon region, for instance, although rainfall amount is thought to be the dominant control on stalagmite δ 18 O (Johnson and Ingram, 2004) and high quality δ 18 O records exist e.g. (Wang et al., 2005;Fleitmann et al., 2003), converting these records into a quantitative assessment of past rainfall is uncertain and has not been performed. In addition, a single record of δ 18 O provides no information about the location of the rainfall. Rainfall δ 18 O reflects the total moisture loss along the atmospheric transport path so, in regions where transport distances are long, rainfall variability a considerable distance from the measurement site might control the  (Wang et al., 2005). Arrows show the average total-column moisture transport for June-July-August (Ding et al., 2004) in kg m − 1 s − 1 . These values are based on reanalysis of modern atmospheric measurements and are derived by multiplying humidity (kg/m 3 ) by wind speed (m/s) and summing over the whole air column. The box indicates the specific region for which comparison of Heshang and Dongge allows rainfall reconstruction, and the red stars show the location of six modern rainfall stations (Enshi, Yichang, Youyang, Zhijiang, Rongjiang, and Guilin). measured δ 18 O signal. This inability to quantify rainfall amount or location from speleothem records represents a significant limitation on the use of these otherwise powerful archives of high-resolution paleoclimate.
In this paper we present a new approach to allow quantification of rainfall by comparing stalagmite δ 18 O records from two caves along the same moisture transport path. We use this approach to quantify the history of rainfall in southwest China during the Holocene by comparing a new precisely-dated stalagmite δ 18 O record described below with a published record (Wang et al., 2005). This is a region prone to flooding when monsoon rainfall is strong (part of the motivation for the recent construction of the Three Gorges Dam in the region) so understanding the amplitude of rainfall variation in this area has a particular societal importance.

Samples and location
Heshang Cave is situated in the Qingjiang Valley of the middle reaches of the Yangtze River (30°27′N, 110°25′E; 294 m) (Fig. 1). Annual climate in this region is strongly monsoonal. In the boreal summer, average temperatures of 28°C drive convection, drawing moist air from the south to generate rainfall peaking in July at 270 mm/month. In the boreal winter, temperature drops below 10°C, air movement is from the north, and rainfall is less than 50 mm/month. Heshang Cave is approximately horizontal and extends a distance of ≈ 250 m backwards from its rea-sonably large opening. It is overlain by ≈400 m of Cambrian dolomite with recharge therefore occurring at an altitude of ≈650 m. The cave is well decorated with stalagmites, rimstone pools, and less frequent stalactites. Three samples have been recovered, all from a distance of ≈150 m within the cave: HS-2 ; HS-4 (the main focus of this study); and HS-6 (from which limited results are also presented here). Stalagmite HS-4 is 2.5 m long (Fig. 2) and was actively growing when recovered from Heshang Cave in 2001. It shows clear annual banding throughout its length, generated by the strong seasonal cycle at this site, and associated with geochemical signals which are the focus of ongoing efforts to reconstruct past climate at seasonal resolution (Johnson et al., 2006). Such seasonal variations are averaged out at the resolution of measurement presented in this paper.

Chronology and analytical techniques
The chronology of stalagmite HS-4 has been established independently by U-Th dating and layer counting. For U/Th analysis, 21 samples of ≈ 1 g of stalagmite calcite were dissolved, spiked with a mixed 229 Th-236 U spike, purified by ion-exchange chemistry, and analysed by multi-collector ICP mass spectrometer at Oxford (Nu Instruments). Techniques broadly followed those used for other carbonate samples at Oxford (Robinson et al., 2002). Samples were handled in a class-1000 clean lab and chemical blanks, although corrected for, were insignificant. U/Th ages are corrected for the presence of small amounts of initial Th using a ( 230 Th/ 232 Th) ratio (where round brackets signify activity ratio) measured directly on the drip waters that formed HS-4. A beaker was placed under the drip for some days, during which time the water it collected degassed and precipitated calcite which could readily be recovered for analysis. This collection technique is reasonably analogous to the growth of stalagmite calcite and gives an accurate assessment of the modern day ( 230 Th/ 232 Th) value captured by HS-4. The measured value (1.97) is higher than typical crustal values (≈1.0), indicating the addition of 230 Th from the overlying dolomite and requiring a somewhat larger age correction for a given 232 Th concentration. Age corrections are made using this ratio and the measured 232 Th for each sample. They average 112 yrs and, with an assumed 50% uncertainty, dominate final age uncertainties which average 67 yrs.
Stable isotope samples were drilled continuously from the entire length of HS-4. A total of 1223 samples were measured on a Finnegan MAT 251 instrument at Wuhan, China with a precision of 0.1‰ (based on longterm repeatability of standard measurements). These  Table 1). The blue line represents layer-counting chronology and is in close agreement with the U/Th dates. have a resolution of ≈ 2 yrs in the uppermost 2000 yrs and, on average, 16 yrs in the rest of the record. In addition to these growth-axis samples, samples were taken from HS-4 along growth layers at each of two depths (≈ 173 cm and ≈ 216 cm). These samples were analysed for stable isotopes to assess the presence of possible kinetic effects during stalagmite growth (i.e. a Hendy test, (1971)). A total of 105 samples were also analysed from the top of a second stalagmite (HS-6) to assess whether the HS-4 record is representative of the Heshang environment.

Results
U/Th ages increase systematically with depth to 9.45 ka (all ages are quoted relative to 1950) at the base of HS-4, (Table 1, Fig. 3) and indicate an approximately constant growth rate averaging 270 μm per year. The layer-counting chronology is in broad agreement (Fig. 3), indicating clearly that the lamina in HS-4 are annual. Over short periods the layer counting chronology is likely to be more precise than the U/Th chronology, but cumulative errors and subjectivity in counting make the U/Th chronology more accurate over longer time periods. We therefore use the U/Th chronology for this study except for the uppermost 150 yrs of the stalagmite where layer counting provides better precision. The most significant deviation between the two chronologies occurs at ≈6-7 ka, a period where chronological uncertainty would not cause any change to the interpretation presented here. A smaller deviation between the chronologies at ≈8 ka would make events at this period about 100 yrs older if the layer counting chronology was used in place of the U/Th chronology. The age model beyond 150 yrs is derived by simple linear interpolation between U/Th ages.
Oxygen isotope values range from − 10.6 to − 7.3‰ (relative to the V-PDB standard) with low values during the early Holocene (≈9 to 5 kyr) and higher values more recently (Fig. 4). Some millennial structure is apparent, with high δ 18 O intervals at around 3.3, 4.3 and 8.2 kyr. Carbon isotope values are reasonably constant at − 11‰, with millennial structure most apparent as two excursions to higher values at 7.3 and 8.2 ka (Fig. 4). There is  (Cheng et al., 2000) and is quoted relative to 1950. Uncertainties incorporates mass spectrometric uncertainty, weighing uncertainty, and spike uncertainty. d Raw ages are corrected for initial 230 Th content using the measured 232 Th content and assuming an initial ( 230 Th/ 232 Th) = 1.97 (a value based on a measurement of the present-day drip water chemistry). Uncertainty in this correction is assumed to be 50% of the size of the correction. Fig. 4. δ 18 O and δ 13 C for HS-4 and a cross plot of δ 18 O versus δ 13 C. The latter demonstrates the lack of a relationship between the two isotope systems, suggesting the absence of any kinetic control on isotope fractionation during stalagmite growth. The full dataset is available at the NOAA National Geophysical Data Centre. no correlation between δ 18 O and δ 13 C (Fig. 4). Samples drilled from single layers in HS-4 show significant increases in δ 13 C (by up to 4‰) towards the edge of the sample, but approximately constant δ 18 O values with no indication of a systematic trend with distance from the centre of the stalagmite (Fig. 5). The oxygen isotope composition of samples from the upper section of HS-6 agree very closely with those from HS-4 (Fig. 6).

Confirmation of isotope equilibrium
If calcite precipitation is rapid during stalagmite growth there can be insufficient time for oxygen isotopes to equilibrate between water and the bicarbonate ion (Hendy, 1971). This leads to Rayleigh fractionation processes in the bicarbonate reservoir and generates systematic increases in δ 18 O which have no direct relationship to δ 18 O of dripwater or to cave temperature. Such kinetic effects at best complicate, and in most cases prevent, the use of stalagmites for climate reconstruction. It is therefore important to assess whether stalagmite growth has occurred in equilibrium conditions before interpreting δ 18 O records for paleoclimate. Four lines of evidence indicate that δ 18 O in HS-4 formed in equilibrium with the dripwaters in Heshang Cave: i. Recently formed carbonate has a δ 18 O value identical to that predicted from equilibrium (Kim and O'Neil, 1997), given weighted-average drip-water δ 18 O of − 7.4 and an annual average cave temperature of 17.1°C (Johnson et al., 2006).
ii. There is no correlation between δ 13 C and δ 18 O in the Holocene record (Fig. 3). Kinetic effects leads to such correlation (Hendy, 1971) and, although there are other ways to generate such a correlation in a stalagmite time-series, its absence from the HS-4 record indicates the lack of any kinetic effect. iii. δ 18 O in the upper section of two Heshang stalagmite is identical (Fig. 6) which is difficult to explain unless precipitation is at equilibrium in both records. iv. δ 18 O is uniform within each of two growth band in HS-4. Progressive degassing and calcite precipitation leads to a systematic δ 13 C increase as drip waters flows over the surface of the stalagmite. In the absence of kinetic effects, this δ 13 C increase will not be accompanied by changes in δ 18 O, because an insignificant proportion of the oxygen present in the water is precipitated as calcite. HS-4 passes this "Hendy test" (Hendy, 1971).
The absence of any indication of kinetic effects indicate that the HS-4 δ 18 O record is a function only of dripwater δ 18 O and cave temperature. Modern dripwaters in the cave have a weighted average δ 18 O of −7.4 (Johnson et al., 2006) identical to that expected for average rainfall at the location based on modelling of rainfall data (Bowen and Revenaugh, 2003), indicating that evaporation does not alter the δ 18 O of rainwater before it precipitates carbonate. This is not surprising because recharge into Heshang cave is rapid, and because the cave is wet throughout the year with measured humidity levels always close to 100%. The HS-4 record can therefore be interpreted as a function only of local rainwater δ 18 O and cave temperature.

Isolating rainfall information by comparison of δ 18 O at two sites
In common with other records of the East Asian monsoon (e.g. Wang et al., 2005) δ 18 O in the HS-4 stalagmite is low during the early Holocene and increases during the mid Holocene with superimposed millennial and centennial variability (Fig. 4). Such δ 18 O changes have generally been interpreted as reflecting higher rainfall during the mid Holocene. Most simply, moisture from a single southerly source becomes progressively lighter as rainfall amount increases due to Rayleigh fractionation and evaporative effects in the atmosphere (Dansgaard, 1964). Other controls also influence stalagmite δ 18 O, however, and most be considered before records can be interpreted as rainfall amount. Temperature is likely to co-vary with monsoon Fig. 5. δ 18 O and δ 13 C analysis from two horizons in HS-4 taken from the centre to the edge of the stalagmite. Carbon isotopes become progressively heavier towards the edge due to degassing of isotopically light CO 2 as waters flow over the surface. Oxygen isotopes show no systematic trend indicating that there is sufficient time for isotopic equilibration with oxygen in the water as the carbonate forms. HS-4 therefore passes the "Hendy test" (Hendy, 1971). strength and will influence the size of the isotope fractionation imposed during carbonate growth. Similarly, δ 18 O in the source region may change with time, or the location of this source may move. Additionally, increased rainfall upwind of Heshang Cave, in the South China Sea for instance, could lead to low δ 18 O in southwest China even if local rainfall actually decreased. This is a significant point because the East Asian monsoon is not characterized by a uniform change in conditions, but rather by changes in the distribution of rainfall along the moisture transport pathway from sea to land (Wang, 2002;Ding et al., 2004).
These secondary controls can, however, be removed from the δ 18 O record by comparing δ 18 O from two stalagmites growing along the same atmospheric moisture transport pathway. As long as both stalagmites grew without kinetic or evaporative modification of their δ 18 O signals, then the difference between δ 18 O at the two sites only reflects loss of moisture from the atmosphere by rainfall as the air travels between them. The magnitude of the difference in δ 18 O between the two sites reflects the amount of rain removed. Source-water δ 18 O; the location of the source; and the amount of rainfall since leaving the source may all influence δ 18 O at both locations, but will not influence the difference between the two stations. If the two caves are reasonably close to one another they will also have a similar temperature history so changes in temperature will also not influence the difference in δ 18 O between the two sites. A caveat to this approach is that the altitude of the cave recharge must be similar at the two sites or, if different, must be corrected for (because increasing altitude causes systematic lightening of rainfall δ 18 O) but this can readily be achieved.
In this study we have followed this approach by comparing our new Holocene δ 18 O record with a published record from Dongge Cave (25°17′N, 108°5′E). Dongge Cave lies 600 km SW of Heshang, directly upstream along the atmospheric pathway by which moisture arrives at Heshang during the summer monsoon (Fig. 1). Winter moisture transport is broadly the reverse of this summer transport, but less than 20% of annual rainfall falls in the winter so rainfall (and therefore stalagmite) δ 18 O is dominated by the summer monsoon. There are two published Holocene δ 18 O records from Dongge (Wang et al., 2005;Dykoski et al., 2005) which show some differences in detail but which, overall, have similar structure. Here we focus on the record from stalagmite DA (Wang et al., 2005) which is at higher resolution and features a more uniform growth rate. The same analysis performed with stalagmite D4  supports the interpretation presented below. The δ 18 O record of Heshang and Dongge show some similarity of structure, reflecting their location in the same climate system. A plot of fifty year averages of δ 18 O has an r 2 of 0.78 indicating that much, though not all, of the variability in the two records is common.
As expected for a site further along the moisture transport pathway, speleothem δ 18 O at Heshang is lower than at Dongge throughout the Holocene (Fig. 7). The average difference is 1.0‰, of which less than a third can be explained by the small (1.5°C) temperature difference between the two caves. The altitude of recharge at the two sites is quite similar but Heshang is slightly lower (and hence would be expected to have a higher δ 18 O) so altitude differences also cannot explain the observed difference. This systematic difference in δ 18 O between the two sites therefore appears to be controlled by the amount of rain falling between them, potentially providing a quantitative Holocene rainfall history for SW China.
Quantifying rainfall using δ 18 O at these two sites assumes that the dominant annual moisture transport has remained similar to that seen today. Several arguments suggest this to be the case, at least for the Holocene. The geography of the region (e.g. coast lines and the Himalayan massif) exerts a strong control on the pattern of winds and has not changed during the Holocene. Analysis of inter-annual variation in moisture transport during the instrumental record from 1952 to 2001 also indicate a uniform moisture transport pathway (Fig. 8). Perhaps most convincingly, a constant moisture trajectory is indicated by numerous GCM simulations of the Asian monsoon at 6 ka (Joussaume, 1999;Braconnot et al., 2002;deNoblet et al., 1996;Bush, 2002). The monsoon circulation is primarily driven by the land-sea thermal contrast, which was higher during the mid-Holocene. Joussaume et al. (1999) report that 18 different model simulations for 6 ka all indicate enhanced lowlevel convergence into the monsoon low over Eurasia, with the summer monsoon flow extending further inland at this time. While there are differences in model sensitivity, the general pattern is similar in all models, with most atmospheric variables exhibiting only small anomalies relative to the modern. The most significant differences are an increased meridional component of the low-level circulation at 6 ka manifested as northward shifts in monsoon rainfall (Braconnot et al., 2002;deNoblet et al., 1996). These models therefore suggest that any circulation changes were small and parallel with the modern trajectories, so use of the two-site δ 18 O approach appears appropriate in this setting.  Fig. 9 (lower panel). Note the inverted scale in the upper panel, and that Heshang is always more depleted than Dongge. The solid blue line in the lower panel is Δδ 18 O. This is calculated by differencing a 100 yrs average for both records. With typical age uncertainty of ±67 yrs, this averaging process encompasses the age uncertainty. To further assess the effect of age uncertainty, the Dongge record is shifted 50 yrs young, and 50 yrs old, and Δδ 18 O recalculated, as shown by the two faint blue lines. This analysis demonstrates that the Δδ 18 O record is robust to age uncertainty. Analytical uncertainty in Δδ 18 O is ≈ 0.15‰ based on combination of the errors on analyses in the two caves and is shown by the bar on the top left of the figure. The bar on the top right also encompasses uncertainty due to scatter about the linear relationship shown in Fig. 9. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

A quantitative history of Holocene rainfall for SW China
The difference between δ 18 O at Dongge and Heshang (Δδ 18 O) can be calibrated to rainfall using the instrumental record of the last 54 yrs. Five δ 18 O samples from Dongge with an average resolution of 9 yrs are compared with averages of annual samples from the same period from Heshang to yield approximately decadal values for Δδ 18 O (Fig. 4). These values are compared with the average rainfall between the two caves for each timeperiod based on the six rainfall stations (Fig. 1). As expected if rainfall is the dominant control on Δδ 18 O a strong relationship is observed (Fig. 9). In addition to supporting the use of the two-site approach, this relationship also allows a scaling of Δδ 18 O to average annual rainfall which incorporates uncertainty due to winter rainfall or continental water recycling. Decadal average values of Δδ 18 O span a range of 0.7 and capture most of the variation seen during the Holocene so that little extrapolation beyond this calibrated range is required.
Assessment of uncertainty in this calibration involves both analytical uncertainty in Δδ 18 O, and uncertainty in the rainfall to Δδ 18 O relationship. We assess Δδ 18 O error as the quadratic sum of measurement of a single δ 18 O value at the two sites. This is a conservative estimate of error because Δδ 18 O calculation averages many indivi-dual δ 18 O values. We assume that the uncertainty in conversion of Δδ 18 O to rainfall is represented by the scatter about their co-relation during the period of the instrumental record (Fig. 9). This assumes that the instrumental record is long enough to assess uncertainty, Fig. 9. Comparison of instrumental rainfall records from the last 50 yrs with Δδ 18 O. Age ranges for each datapoint reflect those captured by sampling intervals in the Dongge record (Wang et al., 2005). Rainfall data is the average annual rainfall at the six stations shown in Fig. 1 in mm/yr. As expected, larger differences between δ 18 O at the two caves relates to higher annual rainfall. The observed relationship is not a simple Rayleigh Fractionation due to the effects of local terrestrial recycling of moisture (Brubaker et al., 1993;Ingraham and Taylor, 1991). and that it incorporates any variability due to slight changes in moisture transport pathway or water-recycling. These are reasonable assumptions, although their robustness cannot be further tested with the available data. Future modelling work may help to provide additional quantification of the errors associated with the conversion of Δδ 18 O to rainfall.
We use the relationship with rainfall to convert measured Δδ 18 O into a quantitative reconstruction of rainfall in the region between Dongge and Heshang during the Holocene (Fig. 7). This region, although representing only a portion of the total area influenced by the Asian monsoon, is nevertheless representative of the East Asian monsoon, and has particular significance because it lies in SW China, close to the Yangtze river and regions of significant flooding in years of unusually strong monsoon rainfall.
The record of southwest-China rainfall indicates that rainfall was 3% higher than modern in the period 8.2-6.4 ka, 8% higher than modern during between 6.4 and 5.0 ka, and 4% higher during the period 5.0-3.0 ka. Higher rainfall than present throughout the Holocene agrees with the qualitative interpretation of δ 18 O records from either Dongge or Heshang and is expected based on higher northern hemisphere insolation driving stronger continental heating. But in other ways the Δδ 18 O record differs from individual records. In both of the individual records there is little change in δ 18 O between ≈8 ka and ≈5 ka, suggesting uniform strong rainfall during the early Holocene. But Δδ 18 O indicates lower rainfall during the period 8.2-6.4 ka, with an increase after 6.4 ka. This difference may reflect a change in the δ 18 O and location of the source of moisture which influenced δ 18 O at both locations but not the difference between them. Sea-level was still increasing slowly during the early Holocene following deglaciation, leading to small changes in the δ 18 O of seawater and changes in coastlines in Indonesia and elsewhere. Alternatively, the difference between regional and single-site rainfall reconstructions might reflect significantly higher rainfall upwind of Dongge (decreasing δ 18 O before it reaches Dongge).
At multi-centennial timescales, the Heshang record shows significantly more variability than the Dongge record. Larger variability of the more inland site is well explained by rainfall-induced δ 18 O change between the two sites and argues against a change in overall atmospheric circulation which would be expected to influence both records together. This adds confidence to the interpretation of dry and wet multi-centennial events in the Δδ 18 O record. The most pronounced dry event occurs at 8.25 ka when rainfall is 7% lower than present, and 14% lower than that at ≈6 ka. This event is likely to be correlative with the 8.2 ka event first observed in the North Atlantic region (Alley and Agustsdottir, 2005). Weakening of the monsoon during cooling in the northern hemisphere at this time mirrors the relationship seen during millennial events of MIS 3 at Hulu Cave (Wang et al., 2001) and may be related to changing ITCZ position caused by this Northern Hemisphere cooling e.g. (Chiang and Bitz, 2005). A lower-amplitude millennial cold period around the abrupt 8.2 ka event has been noted in Greenland ice-core records (Alley and Agustsdottir, 2005;Rohling and Palikie, 2005). The slightly early onset and longer duration of the event in the southwest China rainfall record (and in the δ 13 C record from HS-4) may suggest some influence of this longer cold period in the monsoon, although the duration, nature and timing of the abrupt 8.2 ka event in the north Atlantic region is still a matter of some debate (Ellison et al., 2006).
Other millennial to centennial periods of monsoon weakening in the Δδ 18 O record are of lower amplitude, but are clearly resolved, particularly at 4.8-4.1 ka, 3.7-3.1 ka, and 1.4-1.0 ka. The same pattern of events is not seen clearly in Holocene records from other areas e.g. (Mayewski et al., 2004). The first of these events has the longest duration and has been recognized in previous lake and pollen records from China (e.g. Morrill et al., 2003 and refs therein). This event is not clearly correlative with any event in the North Atlantic region, but a similar dry period is seen in other regions impacted by the Asian monsoon. This event correlates with the 4.2 ka event that may have led to collapse of the Akkadian empire in Mesopotamia (Cullen et al., 2000) and the Neolithic culture of China (Wu and Liu, 2004).
Multicentennial scale variations in rainfall continue until the present, including a wet and dry period that correlate with the Medieval Warm Period and the Little Ice Age respectively. Above average rainfall is seen for the period 1000AD to 1250AD (correlating with the Medieval Warm Perioda period of relative warmth in the Northern hemisphere Moberg et al., 2005) and below average rainfall in the period 1250AD to 1750 (correlating with the cool European conditions of the Little Ice Age). The sense of change is the same as that seen for other abrupt changes in monsoon strength (i.e. cold North Atlantic with dry Asia), and is again consistent with an atmospheric teleconnection, such as control of the ITCZ position by Northern Hemisphere temperature (Chiang and Bitz, 2005).

Implications for future work
The recognition and quantification of changing Holocene rainfall in southwest China demonstrates the power of combining spatially-separated stalagmite δ 18 O records to provide information that neither record delivers on its own. The resulting rainfall record, for a continental interior influenced by a major climate system, provides a good target with which to test and refine climate models, and helps to assess the expected magnitude of change in future rainfall in this area.
The use of Dongge and Heshang caves for the Holocene period represents perhaps a best-case scenario for the combination of spatially separated δ 18 O records. Both records are high resolution with precise U/Th-based age models and they are situated on a single moisture transport pathway which is expected to be invariant during the period of study. The simple differencing of two sites might also be applied in other areas but in many settings the situation is likely to be more complicated. In such settings, however, inter-comparison of stalagmite δ 18 O could be performed using GCM models which capture aspects of the processes controlling atmospheric δ 18 O fractionation, and the transport of moisture in the atmosphere e.g. (Schmidt et al., 2005;Sturm et al., 2005). Coupled with such models, the approach described in this paper could be extended to include δ 18 O from more than two sites, and could provide quantitative reconstruction of past rainfall in many regions, allowing better understanding of natural rainfall variability and its response in varied climate states.

Conclusions
The difference in δ 18 O between two stalagmite records from SW China provides a reconstruction of Holocene rainfall in this region. The use of two records removes the effects of secondary controls on δ 18 O so that the dominant control becomes regional rainfall. This approach should be applicable in other settings and periods, and might be improved by appropriate atmospheric modelling. In this study it has been used to demonstrate that rainfall in SW China was 8% higher than present at the Holocene optimum, but only 3% higher in the early Holocene. It also indicates pronounced multi-centennial dry periods at ≈8.2 ka, 4.5 ka, and during the Little Ice Age. The long term and multi-centennial record of rainfall provides a good target with which to test atmospheric models and indicates the teleconnection of monsoon dry periods to cold periods in the north Atlantic.