Climate of the Past Reconstruction of southeast Tibetan Plateau summer climate using tree ring δ 18 O : moisture variability over the past two centuries

A tree-ring δ18O chronology of Linzhi spruce, spanning from AD 1781 to 2005, was developed in Bomi, Southeast Tibetan Plateau (TP). During the period with instrumental data (AD 1961–2005), this record is strongly correlated with regional CRU (Climate Research Unit) summer cloud data, which is supported by a precipitation δ18O simulation conducted with the isotope-enabled atmospheric general circulation model LMDZiso. A reconstruction of a regional summer cloud index, based upon the empirical relationship between cloud and diurnal temperature range, was therefore achieved. This index reflects regional moisture variability in the past 225 yr. The climate appears drier and more stable in the 20th century than previously. The drying trend in late 19th century of our reconstruction is consistent with a decrease in the TP glacier accumulation recorded in ice cores. An exceptional dry decade is documented in the 1810s, possibly related to the impact of repeated volcanic eruptions on monsoon flow.


Introduction
Clouds formed by condensation of atmospheric moisture release latent heat in the atmosphere and can scatter, absorb and reflect radiative fluxes.Through their impacts on the atmosphere and surface energy budgets, they play a crucial role in the climate system (Yu et al., 2004;Stephens, 2005).Cloud feedbacks are particularly important in the response of the climate system to radiative perturbations (Dufresne and Bony, 2008).Regional cloud cover is also sensitive to atmospheric aerosol load, which is strongly affected by volcanic eruptions (Robock, 2000) and human activities (Khain et al., 2005;You et al., 2010), with potential positive feedbacks (Jacobson, 2001;Ramanathan et al., 2005).
On the Tibetan Plateau (TP), a significant increase in low clouds has been shown on the eastern TP since the 1980s (You et al., 2010), as well as a decreased cloud variability (Zhang et al., 2008).An increase in night cloud cover (Duan and Wu, 2006), which has amplified the fast regional warming on the TP (Duan and Wu, 2006;Wang et al., 2008), is also reported.Because of the important role of cloud cover in climate dynamics, it is crucial to place the recent changes in the context of a longer time scale to understand the full spectrum of cloud cover variability beyond the short instrumental records and to investigate whether these changes exceed the natural cloud cover variability.
Cloud cover is negatively correlated with diurnal temperature range and sunshine duration and positively correlated with precipitation and summer relative humidity at both regional and global scales (Karl et al., 1993;Dai et al., 1997;Leathers et al., 1998;Trigo et al., 2002;Auer et al., 2007; in the Qinghai-Tibet plateau: Zhang et al., 2008) and at individual weather stations (Wang et al., 1993;Ruschy et al., 1991).Correlations between tree-ring cellulose δ 18 O and both cloud cover and sunshine duration have been reported.They were attributed to the close relationships between cloudiness and key factors controlling cellulose δ 18 O, such as precipitation and humidity (Hilasvuori and Berninger, 2010;Kress et al., 2010).
We have recently measured tree-ring cellulose δ 18 O in Bomi (southeast TP, 95.55 • E, 29.87 • N) and examined its correlations with local and regional climatic parameters during the instrumental period (Shi et al., 2011).The Bomi treering δ 18 O is significantly correlated with local and regional CRU cloud cover data (New et al., 1999(New et al., , 2000)).These gridded data were constructed using a combination of observed data and empirical relationships with TP diurnal temperature range (Mitchell and Jones, 2005).The strongest squared correlation coefficient (r 2 = 0.40, p < 0.01, AD 1956AD -2005) ) is observed between Bomi tree-ring δ 18 O and June-August regional CRU cloud data, upstream of Bomi (28-31 • N 90-95 • E).This shows the potential of longer tree-ring δ 18 O records to reconstruct past variations of the regional cloudiness and moisture variability.
In this manuscript, we extend the Bomi tree-ring cellulose δ 18 O record back to AD 1780 and reconstruct a summer regional cloud index for the past 225 yr.The variability of this reconstruction is analysed in terms of trends, regime shifts and frequency of extreme years.Our reconstruction is also compared with past snow accumulation and δ 18 O derived from ice cores.

Sampling site and strategy
Our tree-ring sampling was conducted in Gangcun Natural Forest Reserve (95.55 • E, 29.87 • N, ca.2760 m), which is located 20 km west of Bomi city and 60 km from the turning point of Yaluzangbo-Brahmaputra River (Fig. 1).The forest is well protected and does not exhibit any sign of anthropogenic influence.The main species of this forest are Linzhi spruce (Picea likiangensis var.lintziensis) and East Himalayan Fir (Abies spectabilis).Linzhi spruce is a shallow rooted and shade tolerant species, mainly distributed within moist areas of southeast TP.We selected 27 Linzhi spruce trees that have no distorted stem or internal rot for sampling.One core per tree was collected at breast height, using an increment borer of 5.14 mm inner diameter.The length of  tree-ring cellulose δ 18 O and the regional CRU cloud data (25°N-35°N 85°E-100°E, 518 rectangular in the map); contours are displayed for increments of the correlation 519 coefficients (p<0.05,n=46).The cloud data were obtained from CRU version 2.0 520 (New et al., 1999;2000).(New et al., 1999(New et al., , 2000)).
cores ranges from 30 cm to 55 cm and most cores reach the pith of the tree.

Sample preparation and crossdating
Cores were air dried and polished with progressively finer sandpaper until rings were clearly visible.Tree-ring widths (TRW) were measured using the videoment ® system and cross-dated by visual inspection of the ring-width pattern.A total of 26 cores were successfully cross-dated and the dating quality was checked using the COFECHA software (Holmes, 1998).

Tree-ring δ 18 O measurement
Among the successfully cross-dated cores, 11 old ones (labelled O1 to O11 following age ascendant order) having regular ring boundaries and no missing rings, were chosen for tree-ring δ 18 O analyses.Each ring of each core was cut using a scalpel blade and stored in an individual plastic tube.For the time period 1956-2005, individual tree-ring δ 18 O measurements were conducted on five trees and compared to the value obtained on the pooled six other trees.The average of the individual measurements and the pooled result are highly correlated (r = 0.90, p < 0.01) with no difference in mean level (Shi et al., 2011).We had previously shown that four trees with one core per tree were sufficient to represent a population signal, and that no age effect was found for this species (Shi et al., 2011).We therefore pooled all the tree-ring samples prior to AD 1956 before conducting the cellulose δ 18 O measurements.
Each wood sample was milled, and alpha cellulose was extracted by chemical extraction, following the procedure described in Raffalli-Delerce et al. (2004).Samples of 0.1-0.2mg of cellulose were packed in silver capsules and δ 18 O was measured with a high temperature analyzer (TC/EA) coupled with a Finnigan MAT252 mass spectrometer at LSCE (Gif/Yvette of France) (Shi et al., 2011).The δ 18 O measurements were conducted with reference to CC31 cellulose standards (31.85 ‰), and the standard deviations of 10 consecutive CC31 (0 to 2 values being rejected) was controlled to be lower than 0.20 ‰.Each sample was measured at least twice, and up to four times if the cellulose amount was sufficient.The analytical uncertainty is 0.2 ‰ (maximum accepted standard deviation on the mean).

Tree-ring δ 18 O chronology
The tree-ring δ 18 O and analytical uncertainties are displayed in Fig. 2a, together with the number of trees pooled for δ 18 O analysis.

Calibration and verification of the tree-ring δ 18 O in relation to CRU cloud data
Our previous calibration study (Shi et al., 2011) revealed a significant correlation with regional June-August CRU cloud data, which is based upon the empirical relationship between diurnal temperature range and cloud cover (r 2 = 0.40, n = 46, p < 0.01).This correlation can be explained by several mechanisms.First, cloud cover is correlated with other surface climate factors such as relative humidity and precipitation (r = 0.46 p < 0.01; r = 0.35 p < 0.05 for humidity and precipitation respectively), directly affecting tree physiology.Second, cloud cover is likely related to inter-annual variations in monsoon flow and precipitation amount, themselves related to surface vapour and precipitation δ 18 O, through the amount effect demonstrated at Bomi at the event base (Jing et al., 2011).We will investigate this last point in Sect.4.1 using atmospheric model results.Two outlier years, AD 1978 and AD 1991, are detected (Fig. 3, blue arrows).The δ 18 O measurements of AD 1978 did not yield reproducible results, as six replicates failed to produce a reliable mean value (SD = 0.5 ‰, which extends beyond our accepted threshold of 0.2 ‰).When ignoring this outlier year, 50 % variance of tree-ring δ 18 O is explained by the variability of June-August CRU cloud data (n = 45, p < 0.01) (Fig. 3).The linear calibration equation relating June-August CRU cloud cover data (CC JJA , expressed in %) with tree ring cellulose δ 18 O (in ‰) is: Leave-one-out cross verification was conducted.RE (reduction of error statistic) and CE (coefficient of efficiency) are both equal to 0.44, which means that the linear calibration is valid.We therefore propose to estimate, prior to the instrumental period, a reconstructed cloud cover index, from this calibration with CRU cloud cover data.We chose to use the term "cloud index" to highlight the fact that (i) our calibration is not performed with direct cloud observations, but with the CRU product based on meteorological data scaled to a short cloud observational record; and (ii) our tree ring data are sensitive to a variety of moisture indicators (precipitation, relative humidity, etc.) in addition to the strongest linear relationship with CRU cloud data (Shi et al., 2011).
The uncertainty on the reconstructed cloud index arises from the uncertainty associated with the linear model and from the uncertainty on the δ 18 O measurements (on average ±0.20 ‰).The uncertainty associated with the linear model was estimated using a bootstrap method.Two thirds of the data (calibration samples) were randomly sampled with replacement; the best multiple linear regression was calculated on these data and the quality of the reconstruction was assessed on the last third of the data (verification samples).Further uncertainties on the reconstructed CC JJA may arise from the uncertainty on the proxy measurements.Our analytical protocol warrants an uncertainty on each annual δ 18 O measurement within ±0.2 ‰; we have used the bootstrap method to test the quality of the linear regression model, taking into account these uncertainties on the proxies by randomly modifying the proxy data within their uncertainty range.In this case, the standard deviation of the residuals obtained over verification sub-datasets was ±1.48 % (1000 iterations).Therefore, in order to have a conservative estimate of the quality of the reconstruction, we consider the error on the reconstructed CC JJA to be ±2.2 (1.5 σ ).

Reconstruction of past cloud cover variability
The past variability of the regional cloud index (28-31 • N, 90-95 • E) was reconstructed by applying Eq. ( 1) to the 225yr long chronology of tree-ring δ 18 O (Fig. 2b).The longterm variability of the cloud index reconstruction was analysed using 30 yr Gaussian low-pass filter.To test the mean value shift of our cloud index reconstruction, a regime shift analysis was conducted with software Regime Shift Detection v3.2 (Rodionov, 2004).The long-term trend is shown in Fig. 2b, and three major shifts in years AD 1807, AD 1818 and AD 1887 are found.From the 1780s to the 1880s, the reconstructed CC JJA shows a plateau of relatively high values, interrupted by an 11-yr sharp minimum (from AD 1807 to 1817).A downward trend is observed from the 1870s to the 1890s, followed by small fluctuations around stable mean   values.The recent increasing trend is not unprecedented and appears small in comparison to past regime shifts.
To depict the inter-annual variability, we investigated extreme years with high/low cloud index, as well as the moving standard deviation (STD) of the residual record (yearly cloud index minus the mean regime).The extreme years are defined as those with a reconstructed residual cloud index exceeding 1.5 standard deviation from the base level (dashed blue lines in Fig. 2b), and represented with red and blue vertical bars, respectively (Fig. 2b).From the 1920s onwards, few extreme years are evidenced and the variability of the 31-yr moving STD is low (Fig. 2c).The inter-annual to decadal variability of the reconstructed cloud index appears to have been maximum in the years AD 1900-1930, and about 25 % higher than during the last 50 yr (Fig. 2c).

Analysis of atmospheric simulations
The oxygen isotopic compositions of regional precipitation and vapour were simulated using the isotope-enabled general circulation model LMDZ (Hourdin et al., 2006), nudged by atmospheric reanalyses over the AD 1981-2007 period after 3 yr of spin-up (Gao et al., 2010;Risi et al., 2010).The simulated summer precipitation and vapour δ 18 O at Bomi are significantly anti-correlated with the simulated summer regional (25 • N-35 • N, 85 • E-100 • E) cloud water content and low cloud cover (r = −0.41,p < 0.05 in both cases).Spatially, correlation is maximum with cloud water content upstream the air mass trajectories.In the model, this anticorrelation mostly results from the amount effect, with an anti-correlation between precipitation δ 18 O and regional precipitation amount, itself closely linked with the regional low cloud cover and cloud water content.Cellulose δ 18 O therefore appears related to cloud cover through the link with the precipitation δ 18 O, and possibly (albeit not quantitatively investigated) through leaf water enrichment effects (Shi et al., 2010).
We therefore conclude that the most likely explanation for the Bomi cellulose δ 18 O variability is a simultaneous variability in regional cloud cover associated with changes in upstream monsoon flow, which also correlates with changes in Bomi surface air relative humidity.This is consistent with several tree-ring ecophysiological studies showing that relative humidity is the major driving force of cellulose δ 18 O (Roden and Ehleringer, 2000;Helliker and Richter, 2008;Sternberg, 2009).

The 1807-1817 anomaly
In this section, we focus on the regime shift observed from AD 1807 to 1817.This period is characterized by the lowest cloud index.According to large-scale tree-ring based temperature reconstructions, the 1810s is the coldest decade of the past 250 yr to 600 yr in the Northern Hemisphere (Briffa et al., 1998(Briffa et al., , 2001;;Briffa and Osborn, 1999;Esper et al., 2002).This cold event was reported in many Asian areas, including Sichuan (Bräuning and Mantwill, 2004;Song et al., 2007) and Northwest Yunan (Fan et al., 2008), two Chinese provinces adjacent to the TP.A cold anomaly was also evidenced in southeast TP (150 km southwest of our site) in a TRW-based summer temperature reconstruction, which showed the 1810s-1820s as the coldest period in the last 242 yr (ca.0.5 • C below the mean) (Liang et al., 2009).The climate anomaly of the 1810s was attributed to two consecutive volcanic eruptions: a large tropical eruption of unknown origin, which occurred near the beginning of AD 1809 (Dai et al., 1991;Cole-Dai et al., 2009;Dai, 2010); followed by the AD 1815 eruption of the Indonesian Tambora Volcano (Briffa et al., 1998).Several General Circulation Models (GCMs) predict that large volcanic eruptions should also result in anomalous dry conditions throughout much of monsoon Asia (Oman et al., 2005;Schneider et al., 2009;Fan et al., 2009).Indeed, in the Yunnan province during the AD 1815-1817 interval, serious deficiencies in the monsoon rainfall led to crop failures and to the worst starvation recorded in documented history (Yang et al., 2005).In contrast with modelling and historical sources, at the grid points closest to Bomi, the Palmer Drough Severity Indices reconstructions of the Monsoon Asia Drought Atlas show a moderately dry, almost normal 1810s decade (Cook et al., 2010;Anchukaitis et al., 2010).
If Bomi cellulose δ 18 O was temperature driven, colder conditions would induce a δ 18 O drop, in contrast with our data.Today, the cellulose δ 18 O is mainly controlled by amount effects (via the δ 18 O of the precipitation) and evaporative enrichment (in direct connection with local humidity).A weakening of the monsoon, which would lead to a decrease of cloud cover, precipitation and relative humidity, would be consistent with the observed increase of cellulose δ 18 O (decreased rainfall resulting in an increase of precipitation δ 18 O, combined with decreased evaporative enrichment).
In order to assess if the Bomi signal is associated with a widespread precipitation isotopic composition anomaly, we have compared our record with published ice-core δ 18 O from Dunde, Dasuopu and Guliya drilling sites, albeit not located upstream or downstream of the monsoon flow to Bomi (only decadal averaged data are available) (Thompson et al., 1989(Thompson et al., , 1997(Thompson et al., , 2000) ) (Fig. 4).There is no ice core evidence for a positive isotopic anomaly in precipitation in the 1810s.
We conclude that the sharp tree ring cellulose anomaly detected in Bomi tree-ring cellulose δ 18 O in the 1810s is likely caused by a reduced local monsoon flow, and drier conditions of exceptional decadal duration, possibly related with the regional impacts of repeated volcanic eruptions.

Southeast Tibetan Plateau drying in the late 19th century
The Bomi data show a significant regime shift in the late 19th century towards drier conditions.Other moisture records are available to assess if this is a local or a more widespread signal.Yao et al. (2008) reconstructed past annual precipitation accumulation using four TP ice cores (Dasuopu, Dunde, Guliya, Puruogangri).Dunde, Guliya and Puruogangri Glacier all depict a dramatic decreased accumulation at the end of the 19th century (Yao et al., 2008), consistent with a reduced low-level jet since the late 19th century in the Indian Ocean (Gong and Luterbacher, 2008).These results support our interpretation of the 1890s regime shift detected in the Bomi tree-ring δ 18 O in terms of TP moisture availability.The cause for this shift remains to be investigated.(Thompson et al., 1989;Thompson et al., 1997;Thompson et al., 2000).(Thompson et al., 1989(Thompson et al., , 1997(Thompson et al., , 2000)).
These drier conditions since the late 19th century contradict another tree-ring δ 18 O reconstruction of annual precipitation in North Pakistan, showing that post-1850s is the wettest period of the last millennium (Treydte et al., 2006).This mismatch could be due to the different climate contexts, since the precipitation of North Pakistan is concentrated in winter and controlled by westerly flow, while Bomi precipitation mainly falls in summer and is dominated by Indian monsoon flow (Gao et al., 2010).

Conclusions and perspectives
A 225-yr record of tree-ring cellulose δ 18 O is obtained for southeast TP.Based on the relationship between the tree-ring δ 18 O and CRU cloud data derived from diurnal temperature range in the period AD 1956-2005, a summer cloud index is reconstructed, which reflects moisture variability in past 225 yr.
Our reconstruction suggests drier conditions in the 20th century than during the 19th century.The occurrence of years of extreme low or high cloud index appears to have strongly decreased since the 1920s, suggesting a relatively stable summer moisture condition in southeast TP in spite of the increasing impact of human activities on climate.The sudden increased tree-ring δ 18 O in AD 1807-1817 reflects an exceptional decade marked by persistent dry conditions, possibly related to the impact of repeated volcanic eruptions on regional monsoon flow.
More regional tree-ring δ 18 O records are needed to investigate the spatial changes associated with the regime shifts recorded at Bomi.Process-based understanding of the links between cloud cover and tree-ring cellulose δ 18 O must be further investigated using atmospheric/land surface isotopic models and ecophysiological models.The tree-ring δ 18 O data offer the potential for multi-centennial comparisons with climate model results at the regional scale.

Figure 1 :
Figure 1: Map of the study area and location of sites: Tree ring sampling site (black 514 517 521

Fig. 1 .
Fig. 1.Map of the study area and location of sites: tree ring sampling site (black circles), Bomi and Lhasa meteorological stations (blue circles), glacier ice core drilling sites (black triangles).The shading depicts the local topography.The subplot in the top-left corner shows the inter-annual correlation coefficients between Bomi tree-ring cellulose δ 18 O and the regional CRU cloud data (25 • N-35 • N 85 • E-100 • E, rectangular in the map); contours are displayed for increments of the correlation coefficients (p < 0.05, n = 46).The cloud data were obtained from CRU version 2.0(New et al., 1999(New et al., , 2000)).

Fig. 2 .Figure 3 :Fig. 3 .
Fig. 2. Panel (a): tree-ring δ 18 O with measurement uncertainty (grey shading) and number of pooled trees.Panel (b): reconstructed cloud cover (CC) and regime shift.The black line is the reconstructed CC JJA , with its uncertainty depicted in grey shading (1.5 σ ), the solid blue line is the regime shifts (window length =15 yr, confident at 95 %); the thick red line is 30 yr Gaussian low-pass filter of the reconstructed CC JJA ; the red and blue bars correspond to extreme high/low cloud cover years (exceeding 1.5 standard deviations at the base of regime).Panel (c): 31 yr moving standard deviation (STD) of the CC residual (calculated as the difference between the reconstructed CC JJA and its mean baseline, shown by the solid blue line in panel b).

Figure 4 .
Figure 4. Comparison of tree-ring δ 18 O and ice-core δ 18 O records on Tibetan Plateau