Summer temperature and summer monsoon history on the Tibetan plateau during the last 400 years recorded by tree rings

2.1. Tree Ring Samples and Study Area

[3] Increment cores from 271 spruce (Picea balfouriana, P. purpurea), fir (Abies delavayi var. motouensis, A. squamata, A. fabri) and larch (Larix griffithiana, L. potaninii) trees from 22 temperature sensitive sites (Figure 1) from north facing slopes [Briffa et al., 2001] between 3700 m and 4500 m asl. were analyzed by x-ray densitometry [Schweingruber, 1988]. Cross dating was accomplished by standard dendrochronological procedures [Cook and Kairiukstis, 1990]. Since the biological age trend of the MLD curves was weak and mostly linear, non-climatic variance in the raw data was minimized by calculating residuals from a linear trend line that was computed by least square fitting to the original data. The resulting tree ring index curves preserve much of the low-frequency signal [Cook and Kairiukstis, 1990; Jacoby and D'Arrigo, 1995], although some multicentury variance may be lost depending on the tree longevity [Cook et al., 1995; Briffa et al., 2001; Collins et al., 2002]. To test the homogeneity of the climatic signal recorded by the MLD chronologies and to derive a spatial division of the study area, Empirical Orthogonal Functions (EOFs) of the total set of chronologies were calculated. Hierarchical Cluster Analysis (HCA) was applied on a set of five Principal Components with an eigenvalue >1 that were derived by a Varimax Rotated Principal Component Analysis.

2.2. Relationships Between Tree Ring Chronologies and Regional Climate

[4] Time series of regional variations of temperature and precipitation were calculated as deviations of monthly means of the common period 1961–1990 for each climate station and averaged for each growth region. Correlation functions were calculated between the regional MLD chronologies and 14 series of monthly temperature and precipitation data from August of the year prior to growth until September of the growth year. Although some individual climate records already started in 1951, the common period of climate data only covers the years 1961–1990. For calibration and verification statistics this period was divided into the subperiods 1961–1980 and 1981–1990. To validate the stability of the derived models, a second set of periods 1971–1990 (calibration) and 1961–1970 (verification) was used. In addition, a leave-one-out procedure [Michaelsen, 1987] was applied for the 1961–1990 period (Table 1). Although the climate-growth relationships remained stable, significance levels were not reached due to the shortness of the verification period. However, the high altitude of the studied sites, the stringent dependency of MLD upon one dominant forcing climatic factor (summer temperature) and the results of the leave-one-out validation justify confidence in the temporal stability of the derived interrelation. To derive the climate-growth models for the final temperature reconstructions, regressions coefficients calculated for the whole period 1961–1990 were used (Table 1).

3.1. Spatial Division of the Chronology Network and Regional Temperature Variability

[5] The first EOF of all MLD chronologies explains 49.6% of the common signal and has positive loadings at all sites (Figure 3a) which points to a strong common forcing among all MLD chronologies. The correlation coefficient for the period 1961–1990 between the principal component (PC) of EOF#1 and late summer temperature (August to September mean) of Qamdo is 0.722 and explains 52.2% of the variance. The temperature pattern of Qamdo is representative for the whole Tibetan plateau (r = 0.74, n = 36) [Liu and Chen, 2000]. As a result of HCA, the study area can be divided into two growth provinces (I and II in Figure 1) separating the humid margins of the Tibetan plateau from the more continental interior parts. Four growth regions (A to D) are outlined from northeast to southwest, region B being identical with province I. Due to local disturbance, two chronologies derived from larch can not be clearly assigned to one region and were excluded from further analyses. Regional master chronologies of the four growth regions were recalculated as means of all trees included in each respective region. From region B, 90 trees older than 150 years were selected from 147 trees available. Regions A, C and D comprise 24, 65 and 35 trees, respectively. As an indicator of chronology reliability, the expressed population signal EPS [Wigley et al., 1984] was calculated. Due to decreasing sample size in the older parts of the chronologies, EPS drops to 0.65 which means that the first decades of the chronologies still represent two thirds of the population signal.

[6] In all cases, significant positive correlations between the regional MLD chronologies and summer temperature of the growth year were found (Table 1) whereas the influence of climate of the year preceding growth was negligible. Ring width of high elevation conifers is often reduced by low winter temperatures as a consequence of bud damage, frost desiccation and reduced root activity due to low soil temperatures [Körner, 1998]. In contrast, MLD chronologies are insensitive to carryover effects of growing conditions during years preceding the growth year due to the use of stored carbohydrates for earlywood formation. However, the ‘Reduction of Error’ statistic RE (a measure of common variance) is negative for the temperature reconstructions of regions A and D which indicates limited reliability of the derived reconstructions [Fritts, 1976]. Very probably these limitations are to some extent caused by the short calibration periods rather than by constraints of the climate-growth model.

[7] The regional reconstructions of summer temperature (Figures 2a–2d) reveal some periods of cool summers, like around 1700, 1730–1750, 1810–1820 and in the 1970s. They occurred synchronously in all growth regions and reflect broad-scale climate anomalies. Most of these periods are also found in other North hemispheric MLD chronologies [Briffa et al., 1998a] and other temperature reconstructions from the Tibetan – Eastern Himalayan area [Cook et al., 2003; Wu, 1992; Wu and Shao, 1995]. However, individual deviations in the Tibetan chronologies point to regional differences from the general trend which probably reflect the influence of different monsoonal air masses affecting temperature at high elevations.

[8] Interestingly, pronounced negative temperature trends from 1970–1990 can be detected in growth regions A, C and D which represent the marginal areas of the Tibetan plateau that receive abundant summer monsoon rainfall. However, in region A where only relatively weak correlations to the meteorological data are found (Table 1), an influence of human impact on some trees can not be completely excluded. Additional tree ring material from this area is needed to verify the fidelity of this regional chronology. Weak responses of MLD to recent warming trends in annual mean temperature are also known from the subarctic and boreal zones [Jacoby and D'Arrigo, 1995; Collins et al., 2002; Briffa et al., 1998b] and the central Himalaya, where summer temperatures declined after 1960 [Cook et al., 2003].

3.2. Spatial Pattern and History of Summer Precipitation

[9] Chronologies B, C and D also show significant negative correlations to summer rainfall (Table 1). At high elevation sites, abundant rainfall is generally combined with enhanced cloudiness and reduced radiation input and temperature. Thus, the mean temperature and the precipitation sum of the August–September season at Qamdo (3241 m asl.) are negatively correlated (r = −0.715 for the 1951–2000 period, p < 0.001). To examine the spatial pattern of rainfall variability across the study region, EOFs of summer (August to September) precipitation were extracted from the set of meteorological stations (Figures 3b–3d). The three first EOFs explain 80.9% of the summer rainfall variance. The leading EOF#1 explains 51.6% of the total variance and has weak or negative loadings at the eastern margin of the Tibetan plateau (Figure 3b). The loading pattern of EOF#2 that accounts for 21.1% of the total variance has a reverse spatial distribution (Figure 3c) which implies the influence of different branches of the summer monsoon system [Wang et al., 2001]. The northeastern part of Tibet receives rainfall from the East Asian branch of the summer monsoon circulation whereas the southern part of eastern Tibet receives rainfall from the Indian summer monsoon. This finding is corroborated by the spatial distribution of the stable isotope composition of present day precipitation water [Aggarwal et al., 2004]. EOF#3 accounts for 8.2% of the total variance of summer precipitation. Its loading pattern (Figure 3d) separates the humid margins from the continental interior parts of the Tibetan plateau.

[10] The PC time series of EOF#1 of August and September rainfall was reconstructed by multiple regression of the regional MLD chronologies B and C (Figure 2e). The multiple correlation coefficient is −0.733, the explained variance adjusted for the loss of degrees of freedom is 50.4% (F = 15.715). Correlation coefficients between the original and the reconstructed PC in the calibration (1961–1980) and verification (1981–1990) periods are 0.68 and 0.88, respectively. The RE value is 0.53, indicating the reliability of the derived reconstruction [Fritts, 1976]. Periods of reduced summer monsoon activity are reported by historical records of droughts in India during 1790–1796 und 1876–1877 and by maxima of dust concentration, chloride and δ¹⁸O in the Dasuopu ice core from southern Tibet [Thompson et al., 2000]. Besides the periods of humid summers between the first half of the 17th century and 1900–1920, the last decades since 1980 show an increase in summer rainfall unprecedented in the past 350 years (Figure 2e).