1 Introduction

The Tibetan Plateau (TP), with an area of about 2,300,000 km2, influences the large-scale circulation system, such as the Asian Monsoon, through its thermal effect (He et al. 1987; Li and Yanai 1996; Wu and Zhang 1998; Zhou et al. 2009). It is important to understand the climate variability over the TP. However, most meteorological records on the TP are of short length (less than 60 years) and distributed sparsely, limiting the study of the climate variability on long-term timescale. An enhanced understanding of past climatic variability on the TP must rely on climatic proxies.

Tree rings, as an annually resolved proxy record, have been increasingly used to reconstruct past climate changes on the TP (e.g., Bräuning and Mantwill 2004; Gou et al. 2008; Liang and Eckstein 2009; Liang et al. 2008; Liu et al. 2007; Liu et al. 2006; Shao et al. 2005; Shi et al. 2010; Wang et al. 2009; Yang et al. 2009; Yin et al. 2008; Zhang et al. 2003; Zhu et al. 2008). For example, Liu et al. (2007) reconstructed winter temperature variations since ad 1000 for the middle Qilian Mountains using Sabina przewalskii. Based on tree-ring samples from the same species, Gou et al. (2008) presented a maximum temperature reconstruction for the past 700 years in the Anymaqen Mountains area; On the southeastern TP, a summer temperature reconstruction since ad 1765 was established using tree-ring samples from Abies georgei var. smithii timberlines (Liang et al. 2009). These climate reconstructions described climate history of several areas of the TP. However, the number of available tree-ring chronologies from the TP is still sparse.

Sabina tibetica, as one of the dominant species, has a wide distribution on the eastern TP, such as in the area of Yushu, Qamdo, and Linzhi. Tree-ring width of S. tibetica was found to be significantly correlated with the early summer climate (Bräuning 2001; Qin et al. 2003; Shi et al. 2010; Wang et al. 2008). In this paper, we present a temperature reconstruction based on a ring-width chronology of S. tibetica from Leiwuqi. The reconstruction should expand our knowledge of climate variability for this area.

2 Materials and methods

2.1 Study area

Our study area is located in Leiwuqi, Qamdo area of the eastern TP (Fig. 1). It is the source region of the Lancangjiang River. The climate of this area is controlled by the South Asian Monsoon in summer and by the southern branch of the westliers in winter. According to the meteorological record in Qamdo (97° 10′, 31° 9′, 3,307 m above sea level (a.s.l.); Fig. 2), July is the warmest month with a mean temperature of 16.4°C, and January the coldest (−2.19°C). Summer (June–August) precipitation (296 mm) accounts for 61.4% of the annual precipitation (482 mm). S. tibetica and Picea likiangensis var. balfouriana are the two dominant species in this area. In general, S. tibetica grows on the south-facing slope, while Balfour spruce on the north-facing slope.

Fig. 1
figure 1

Map showing tree-ring sampling sites, meteorological station, and compared sites in the text. Leiwuqi is the tree-ring sampling site in this study. Qamdo is the meteorological station. Zaduo (Shi et al. 2010) and Anymaqen (Gou et al. 2008) are the two compared tree-ring records

Full size image
Fig. 2
figure 2

Climate diagram from the meteorological station in Qamdo in the eastern Tibet Plateau from 1954 to 2007. The climate variables include the monthly maximum, mean and minimum temperatures, and monthly total precipitation (bars). The horizontal dashed line represents 0°C at 4,000 m with an adjustment of a lapse rate of −0.6°C/100 m from Qamdo meteorological station

Full size image

2.2 Tree-ring sampling and chronology development

Tree-ring samples were bored from a well-drained, open-canopy, and south-facing S. tibetica forest at an elevation range from 3,967 to 4,042 m a.s.l. There is little evidence of disturbances due to fire or human activities. Twenty-five isolated trees were selected for sampling. Two (seldomly one) cores were taken for each tree at a height of about 0.70 m above ground. The tree-ring samples were glued, smoothed, and crossdated through traditional process of dendrochronology in the lab (Stokes and Smiley 1968). Then, we measured the ring width using Lintab with a resolution of 0.01-mm resolution. The Cofecha program (Holmes 1983) was used to check the quality of the crossdating and measurement.

Tree-ring chronology was developed using program Arstan (Cook 1985). We tried several methods of standardization on the raw ring-width data: 30, 80, 180, 230, 280, 330, 380 yearsr, 67% of series length smoothing spline, and negative exponential function or linear regression. The correlation between trees decreased with increasing first-order autocorrelation, suggesting the chronology signal weakened as more low-frequency variation was preserved in chronology. Accordingly, we finally used a 30-year smoothing spline with a 50% frequency cut-off to detrend each raw ring-width series to enhance growth variations on interannual to decadal timescales. The resulted index series were averaged bi-weightly into a chronology to diminish the influence of outliers. To quantify the signal strength among different indexed series, we conducted common interval analysis during 1800–2000. Several statistics were calculated, including variance explained by the first principal component (PC1), correlation between trees (BTR), correlation within trees (WTR), and expressed population signal (EPS) (Cook and Kairiukstis 1990). We also calculated running EPS every 50 years with a 25-year overlap to evaluate the representation of the chronology for population through time.

2.3 Tree growth-climate relationship and climate reconstruction

We correlated the ring-width chronology with monthly climate records from previous October to current September to investigate the tree growth-climate relationship. The climate variables included monthly mean/maximum/minimum temperatures and monthly total precipitation. The meteorological station in Leiwuqi (96° 36′, 31° 13′, 3,811 m) was the nearest one to the tree-ring sampling site. However, its record only started from 1991. Hence, we used the longer record (since 1954) from the Qamdo meteorological station, which is about 65 km southeast to the tree-ring sampling site. The temperature data of four seasons from Qamdo station had high correlations (the least one is 0.89 in summer) with those from Leiwuqi during 1991–2007. For precipitation, most correlations were higher than 0.84 except for the winter season (0.63).

We established several combinations of climate data for calculating correlation coefficients between the tree-ring chronology and climate variables. The seasonal variable having the highest correlation with the chronology was selected for the final reconstruction. The climate data were regressed against the ring-width chronology. The skill of the regression equation for the reconstruction back to ad 1440 was tested by cross-calibration/verification for the sub-periods 1954–1980 and 1981–2006, and by a leave-one-out cross-validation (LOOCV) (Michaelsen 1987) over the full-period 1954–2006. Evaluative statistics included the variance explained (R 2), the adjusted variance explained \( \left( {R_{\rm{adj}}^2} \right) \), the variance predicted (r 2), the sign test of both raw data and their first difference, the reduction of error (RE; Fritts 1976) and the coefficient of efficiency (CE; Briffa et al. 1988).

3 Results and discussion

3.1 Tree-ring chronology and statistics of the common interval analysis

Figure 3 shows the tree-ring-width chronology of S. tibetica from Leiwuqi. The chronology extended back to ad 1384, and could be considered reliable after ad 1440, when ten cores from six trees are available and EPS exceeds the recommended threshold of 0.85 (Wigley et al. 1984; Fig. 3). Both the variance explained by PC1 and the mean correlation BTR during 1800–2000 (Table 1) indicated high consistency between the different ring-width series.

Fig. 3
figure 3

Tree-ring width chronology of Sabina tibetica from Leiwuqi. a The 50-year running expressed population signal (EPS); b ring-width index of the chronology; c the sample depth of the chronology

Full size image
Table 1 Site information of the tree-ring sampling site and the statistics in common interval analysis (1800–2000)
Full size table

3.2 Relationships between chronology and climate data

The chronology showed negative correlations with temperatures from March to July and positive ones with precipitation from February to July (Fig. 4). The highest correlation between tree growth and temperature (p < 0.001) were found for the seasonal window May–June. Seasonal precipitation also correlated significantly (p < 0.05) with the chronology; however with lower correlation coefficients.

Fig. 4
figure 4

Correlations between the ring-width chronology of Sabina tibetica and the mean monthly/seasonal climate variables from October of the pre-growth year to August of the current-growth year during 1954–2006. Horizontal dotted lines denote a significance level of p = 0.05

Full size image

The negative correlations between tree growth and climate (Fig. 4) suggest that high temperature, particularly in May and June, limits the growth of S. tibetica in the eastern TP. This is in agreement with earlier studies (Bräuning 2001; Shi et al. 2010; Wang et al. 2008) on the same species from the eastern, central, and southwestern TP. Negative influences of early summer warmth on tree growth were reported for S. przewalskii on the northeastern TP (Gou et al. 2008; Liang et al. 2010; Liu et al. 2006; Shao et al. 2010; Sheppard et al. 2004; Zhang et al. 2003). In the western Himalayas, warm summers also limit the growth of Juniperus polycarpos (Yadav et al. 2010), Cedrus deodara (Yadav et al. 1997, 1999; Yadav et al. 2004) and Abies pindrow (Hughes 1992). In addition, Cai et al. (2008) found the limiting effect of May–Jun temperature on tree growth of Pinus tabulaeformis from the southeastern Chinese Loess Plateau. High temperature in early summer without sufficient precipitation increases transpiration and soil evaporation, and thus leads to strong moisture stress on tree growth (Cai et al. 2010; Gou et al. 2008; Yadav et al. 2010).

3.3 Calibration/verification statistics and reconstructed temperatures

According to the correlations between tree-ring width and climate variables, we selected the May–June mean temperature for final reconstruction. The regression equations explained over 60% of the variance in the instrumental May–June temperature data, both in the calibration and verification periods (Table 2). Sign test showed that the estimated data were quite similar with the instrumental records. Both RE and CE indicated that the equations had good predictive skills for the mean May–June temperature. In addition, the regression coefficients were essentially the same for the two subperiods and the whole period. Finally, the equation during the whole period (1954–2006) was selected to reconstruct the mean May–June temperature variation back to ad 1440.

Table 2 The calibration and verification statistics for the May–June temperatures
Full size table

As shown in Fig. 5a, the reconstructed temperatures closely matched the instrumental record. However, our reconstruction might not capture low-frequency variations of May–June temperatures due to the application of the rigid 30-year spline in the standardization process of chronology construction. To ascertain the problem, we examined the 11-year smoothing curve of the instrumental and the estimated temperature data (Fig. 5a). The two curves show very similar decadal variations during 1954–2006. There are no significant autocorrelation (r = −0.16, Durbin–Watson = 2.00) in the residuals between the instrumental and the estimated May–June temperature (Fig. 5b). In addition, the scatter plot clearly presented the linear relationship between the instrumental and estimated data (Fig. 5c). These analyses suggested that there should be no obvious loss of low-frequency variation in our reconstruction of May–June temperature at least during the instrumental period. Nevertheless, there may have occurred low-frequency temperature variations over the whole period (1440–2006) that are not captured by our reconstruction due to the data treatment.

Fig. 5
figure 5

Comparisons between the instrumental and estimated May–June temperature of Qamdo. a The instrumental and estimated May–June temperatures with their 11-year moving averages (the thick solid lines); b the residuals of the instrumental and estimated May–June temperatures; c the scatter plot of instrumental and estimated May–June temperatures

Full size image

Figure 6 shows the reconstructed May–June temperature back to ad 1440 for Qamdo area. According to the 11-year moving average curve and their long-term (1440–2006) mean, warm periods could be identified during 1501–1514, 1528–1538, 1598–1609, 1624–1636, 1650–1668, 1695–1705, 1752–1762, 1794–1804, 1878–1890, 1909–1921, 1938–1949, and 1979–1991. Cool early summers occurred during 1440–1454, 1482–1500, 1515–1527, 1576–1597, 1610–1621, 1669–1679, 1706–1716, 1782–1793, 1863–1873, 1894–1908, and 1922–1937. The top five of the 28 high-temperature extremes (anomaly ≥ 2 STD) were indicated in 1944, 1883, 1777, 1799, and 1460. The reconstruction showed only four cold extremes in 1904, 1957, 1763, and 1769. The lower number of cold extremes captured in the reconstruction may result from the finding that tree growth is more limited by higher temperatures.

Fig. 6
figure 6

Comparison between our temperature records with other reconstructions along the eastern margin of the Tibetan Plateau over the 1440–2001 period. a April–September maximum temperature reconstruction for the Anymaqen area (Gou et al. 2008); b the reconstructed Zaduo May–June maximum temperature (Shi et al. 2010); c our May–June temperature reconstruction for Qamdo. The thick lines represent their 11-year moving average. The solid circles are their high/low extremes (≥2 STD)

Full size image

We examined the spectral characteristics of our reconstruction using Multi-taper Method (Mann and Lees 1996). The reconstructed May–June temperature showed significant cycles at 23.8, 19.7, 13.0, 7.7, 5.1, 3.5, 2.6, 2.2 years (Fig. 7). Liang et al. (2008) also found a significant peak at 2.7 and 4.7 years in a summer temperature reconstruction for the Yushu area. The cycles of 23.8, 19.7, and 13.0 years are near the sunspot cycle of 11 years and its double.

Fig. 7
figure 7

Power spectra of the reconstructed May–June temperature for Qamdo during 1440–2006

Full size image

3.4 Comparison with nearby temperature reconstructions

Our temperature reconstruction shows several warm/cold periods and extremes with other nearby temperature reconstructions. For example, the warm periods 1650–1668 and 1878–1890, and cool periods 1482–1500, 1610–1621, 1669–1679, 1706–716, 1762–1772, 1706–1716, 1894–1908, and 1922–1937 in our reconstruction are consistent with the Zaduo May–June maximum temperature reconstruction (Shi et al. 2010; Fig. 6). The warm periods 1624–1636, 1650–1668, and 1794–1804 are in good agreement with the March–May temperature reconstruction for the western Himalaya (Yadav et al. 1999). The western Himalayan record also shows cool conditions identified in our reconstruction during 1440–1454, 1515–1527, and 1922–1937 periods. A regional tree-ring width chronology of S. tibetica for the Qamdo area (Bräuning 2001), which had negative correlations with temperature from March to June of the growth year, indicated relatively higher growth rates during 1490–1500, 1570–1580, and 1780–1790 periods. These periods correspond to the cooler episodes in our reconstruction. In addition, our reconstruction shows common warm extremes in 1799 and 1979, cold one in 1769 with the Zaduo reconstruction (Shi et al. 2010). The cold extreme in 1904 identified in our reconstruction is also indicated to be extreme in the Anymaqen April–September maximum temperature reconstruction (Gou et al. 2008; Fig. 6). The agreement between our reconstruction and these nearby proxy records indicates that our reconstruction should be of good reliability.

In general, there is little agreement between our reconstruction and the Anymaqen April–September maximum temperature record (Gou et al. 2008; Fig. 6). They had opposite warm/cold variations during some periods, such as 1482–1550, 1794–1804, and 1922–1937. Spatial correlations between the Qamdo reconstruction and the Anymaqen record (Gou et al. 2008) with the corresponding CRU-gridded temperature dataset (Mitchell and Jones 2005) reveal their different geographical representation (Fig. 8a and c). The Qamdo reconstruction is associated with the temperature field south of approximately 34° N with a north–south extension along the moisture passage from the Bay of the Bengal (Fig. 8a). However, the temperature field of the Anymaqen reconstruction (Gou et al. 2008) is north to about 33–34oN with the core region in the Anymaqen Mountains area (Fig. 8c). The temperature fields of the proxy data are basically validated by the meteorological data, despite their relatively larger spatial representation (Fig. 8b, d). Further investigation of the synoptic circulation features using the 500 hpa geopotential height from the NCEP reanalysis dataset (Kalnay et al. 1996) indicated that the climate of the Qamdo area and the Anymaqen area are influenced by different synoptic regimes. The Qamdo area is located at the eastern front of India–Burma trough (Wu and Zhang 1998) (Fig. 8e) during the May–June period, leading to dominant southwesterly flow (Fig. 8f). On the contrary, relatively fewer southwest flows can reach the Anymaqen area in May to June. The area is more influenced by northwesterlies.

Fig. 8
figure 8

Spatial correlations of temperature records with corresponding CRU-gridded temperature dataset (Mitchell and Jones 2005) over 1959–2001 and the synoptic circulation features. a and b are the correlation fields (with p < 0.05 filtered out) of the reconstructed and instrumental Qamdo May–June mean temperature, respectively (plotted using the Climate Explorer: http://climexp.knmi.nl); c and d are the correlation fields of the reconstructed and instrumental April–September maximum temperature for the Anymaqen area (Gou et al. 2008); e and f are the composite mean of NCEP (Kalnay et al. 1996) May–June geopotential height and vector wind at the 500 hpa during 1954–2006 (plotted using the PSD Interactive Plotting and Analysis Pages: http://www.esrl.noaa.gov)

Full size image

4 Conclusion

In this paper, we presented a 567-year May–June temperature reconstruction for the Qamdo area based on a new tree-ring width chronology of S. tibetica. The reconstruction indicates that warm conditions occurred during 1501–1514, 1528–1538, 1598–1609, 1624–1636, 1650–1668, 1695–1705, 1752–1762, 1794–1804, 1878–1890, 1909–1921, 1938–1949, and 1979–1991. The Qamdo area experienced cool early summers during periods of 1440–1454, 1482–1500, 1515–1527, 1576–1597, 1610–1621, 1669–1679, 1706–1716, 1782–1793, 1863–1873, 1894–1908, and 1922–1937. Comparisons with other proxy records and meteorological record validate our reconstruction on both spatial and temporal scales. However, there is obvious spatial variability in the May–June temperature variations along the eastern margin of the TP. Since tree growth of Sabina trees on the TP shares coherently inverse relationship with early summer temperatures, there should be great potential to develop a tree-ring network for reconstruct spatial and temporal variability of early summer temperatures over the past few hundred years for this vast area.