1 Introduction

Global land-surface air temperature has risen rapidly since the late nineteenth century, and this warming has been particularly marked since the 1970s (Hartmann et al. 2013). The Tibetan Plateau (TP) is one of the most sensitive areas to global climate change (Liu and Chen 2000; Yao et al. 2019), and meanwhile it plays an important role in modulating large-scale atmospheric circulations and climate over Asia (Broccoli and Manabe 1992; Li and Yanai 1996; Wu et al. 2012). For example, the thermal conditions on the TP have a strong influence on the strength of the Asian summer monsoon that produces much of the urgently needed precipitation for the Asian continent (Fu et al. 2020; Xu et al. 2019). Although it is well acknowledged that the temperature increasing rate in the TP is higher than that in the Northern Hemisphere and the same latitudinal zone over the second half of the twentieth century (Liu and Chen 2000), its long-term variations, especially on decadal to multi-decadal scales, are still not well understood mainly due to two reasons. One is that the meteorological data are too short because most of the stations were built after 1958 (Wei et al. 2003). The other is that high quality temperature reconstructions based on paleoclimate proxy records are still limited for this extensive plateau.

Tree rings have characteristics of annual resolution and precise dating (Hughes 2002), and have been widely used for temperature reconstructions worldwide (Cook et al. 2000; Esper et al. 2002; Shi et al. 2017a). Many tree-ring based temperature reconstructions have been conducted in the southeastern TP (Bi et al. 2020; Bräuning and Mantwill 2004; Duan et al. 2019; Huang et al. 2019; Shao and Fan 1999; Shi et al. 2015, 2017b; Wang et al. 2015), and fewer in the northeastern TP (Gou et al. 2007, 2008; Liu et al. 2009). The southeastern TP is influenced by the Indian summer monsoon that brings a lot of moisture from the Indian ocean. It also features a low temperature due to high elevations. High precipitation and low temperature make trees growing there sensitive to temperature. On the other hand, the northeastern TP is located in inner continent with less precipitation, and as a result trees there are sensitive to precipitation other than temperature (Liu et al. 2006; Shao et al. 2010; Yang et al. 2021; Zhang et al. 2015). Understanding climate change on the eastern TP is complicated due to its large spatial extent and complex climate and orography. The transitional zone between the northeastern and southeastern TP, i.e., central eastern TP, is critical to understand the climate on the eastern TP. Previous studies suggest that tree growth in the central eastern TP is generally stressed by warm-season temperatures (Liang et al. 2008; Xiao et al. 2015; Zhu et al. 2016), which shows the potential for warm-season temperature reconstructions.

Our aims in this study are to (1) build a long, well-replicated tree-ring width chronology using the samples from two independent sites on the central eastern TP, (2) identify and reconstruct the most critical climate limiting factor on tree growth, (3) explore its spatial representativeness and validate its reliability by comparing with other nearby temperature reconstructions, and (4) explore its major features and teleconnections with large-scale oceanic and atmospheric circulation modes.

2 Data and methods

2.1 Meteorological data

The study area is located in the eastern border of the TP, where the average elevation is higher than 3,000 m above sea level (a.s.l.). Influenced by alternations of the westerlies in winter and the Indian summer monsoon in summer, distinct dry and wet seasons can be found in this area. Meteorological data from three meteorological stations (Songpan, Hongyuan and Ruo’ergai) close to the sampling sites were used in this study (Fig. 1; Table 1). Climate parameters include monthly mean, maximum, and minimum temperatures, and monthly total precipitation. The common period of climate data from the three stations is from 1961 to 2015. Songpan station was relocated to a new site in 2013 that is a few kilometers away from the original one. Therefore, we adjusted Songpan temperature data from 2013 to 2015 by comparing the monthly differences between Songpan and the average of Ruo’ergai and Hongyuan over the periods 2013–2015 and 1961–2012. We did not adjust precipitation because it varies locally and is not the focus of this study. Meteorological data from the three stations were averaged to represent the regional situation (Fig. S1). The hottest month is July with a temperature of 12.3 oC, and the coldest is -7.8 oC in January. April-October is the rainy season, with a precipitation of 653 mm, accounting for 92.2% of the annual total precipitation.

Fig. 1
figure 1

Map showing the two tree-ring sampling sites and the three meteorological stations. The red line of the inset delineates the boundary of the TP

Full size image
Table 1 Characteristics of the meteorological stations
Full size table

2.2 Tree-ring data

Tree cores were collected from two sites, SP01 (32.87°N, 103.36°E, 3680 m a.s.l.) and SP02 (32.85°N, 103.36°E, 3650–3666 m a.s.l.). The two sites are 1.5 km away from each other in a straight distance. The dominant tree species is Picea purpurea, with a very small percentage of Abies faxoniana. Most of the trees from which we took cores are P. purpurea, with a few A. faxoniana trees collected as well. Only one core per tree was collected at SP01, and two cores per tree at SP02. All samples were mounted and sanded following standard procedures of Dendrochronology (Stokes and Smiley 1996). Each ring width was measured to a precision of 0.001 mm using a LINTAB 5.0 system after being visually cross-dated under a microscope. The quality of cross-dating was checked with the COFECHA program (Holmes 1983). Eventually, 53 cores from 53 trees at SP01 and 58 cores from 34 trees at SP02 were used in the following analyses.

The raw series of ring-width measurements were detrended using the program RCSigFree (http://www.ldeo.columbia.edu/tree-ring-laboratory/resources/software). Age-dependent spline was applied to detrend each raw series. Ring-width index was calculated as the ratio of actual ring-width and the estimated value, and the robust biweight mean was used to assemble the chronology (Cook 1985). The “signal-free” approach was applied to reduce potential trend distortion in tree-ring chronologies (Melvin and Briffa 2008). The mean inter-series correlation (Rbar) and the expressed population signal (EPS) were calculated for 51-year moving window (Wigley et al. 1984). The EPS threshold value of 0.85 was used to assess the most reliable period of the chronology.

2.3 Methods

Correlation and response functions (Fritts et al. 1970) were used to analyze the relationship between tree growth and climate. The response functions were calculated using the program DENDROCLIM2002 (Biondi and Waikul 2004). A linear regression function was applied to reconstruct June-July mean temperature, the main limiting factor on tree growth. A split sample calibration and verification method was used to assess the reliability of the reconstruction model. Spatial representativeness was explored using the KNMI (http://climexp.knmi.nl). Multi-taper method (MTM) spectral analysis (Mann and Lees 1996) was used to examine the frequency characteristics of the reconstruction.

3 Results

3.1 Characteristics of the tree-ring chronologies

Two site chronologies were built using the signal-free method (Fig. 2). The correlation coefficient between them is 0.75 over their common period 1701–2015, and 0.84 over their common reliable period 1748–2015 based on the EPS value higher than 0.85, both significant at 0.001 level. The correlation coefficient is 0.88 between their first differenced series over the period 1749–2015, significant at 0.001 level. Further considering that the straight distance between the two sites is only 1.5 km and that they have a similar elevation, we pooled all the samples from these two sites together to build a robust chronology, named SP0102 (Fig. 3). COFECHA results showed that for all raw measurements in SP0102, the mean segment length is 240 years, with a series intercorrelation of 0.62 and a mean average sensitivity of 0.14. The absent rings account for 0.04% of the total rings. The same method as used for site chronologies was used to develop the robust regional chronology, with a reliable period of 1730–2015 based on EPS higher than 0.85 (Fig. 3c). Only the robust chronology was used in the following analyses.

Fig. 2
figure 2

a The tree-ring width chronology from SP01 (red) and SP02 (blue) sites. b Low-frequency ring width series with a 10-year fast Fourier transform (FFT) filter. c The sample size

Full size image
Fig. 3
figure 3

a The tree-ring width chronology (thin black line) covering the past 318 years (AD 1698–2015) with a 10-year FFT filter (thick red line). b Rbar statistics calculated over 51-year window. c EPS statistics calculated over 51-year window. The dotted line denotes the 0.85 threshold value. d The sample size

Full size image

3.2 Relationships between tree growth and climate

Correlation analyses were conducted between the SP0102 chronology and monthly mean, maximum, minimum temperatures and monthly total precipitation in Songpan, Hongyuan, Ruo’ergai and the regional average from previous September to current October over the period 1961–2015. As shown in Fig. 4, there are quite similar patterns for Songpan, Ruo’ergai, Hongyuan, and their average. Therefore, we focus on the results of their average. Generally, tree growth correlates positively with monthly mean, maximum, and minimum temperatures. The correlations are the highest for monthly mean temperature (Fig. 4a, b, and c). The two highest correlations occur in June and July. On the other hand, precipitation shows weak correlations with radial growth, with only a marginally significant correlation in current May (Fig. 4d). When seasons are considered, the highest correlation was found between the chronology and June-July mean temperature (r = 0.81, p < 0.001). Furthermore, response functions were calculated between the chronology and regional monthly mean temperature and monthly total precipitation. The results show that June and July temperatures and June precipitation are significant. Since June precipitation is non-significantly correlated with tree growth in terms of correlation analysis, we choose June-July mean temperature as the reconstruction target.

Fig. 4
figure 4

Correlation coefficients (bar) and response functions (circle) between the chronology and a monthly mean temperature, b maximum temperature, c minimum temperature, and d precipitation based on Songpan, Hongyuan, Ruo’ergai meteorological data and the regional average over the period 1962–2015. The horizontal dashed lines denotes 99% confidence level for correlation coefficient. Response functions were only calculated for a regional monthly mean temperature and d monthly total precipitation. The black solid circle denotes significant value at 0.05 level. “p” means previous year

Full size image

3.3 Reconstruction of June-July mean temperature

We reconstructed June-July mean temperature for the period 1730–2015 using a simple linear regression model (Fig. 5a). The reconstruction accounted for 65.6% of the actual temperature variance during the calibration period 1961–2015. In order to evaluate the reliability of the relationship between tree growth and June-July temperature, we applied the split sample calibration and verification method. As shown in Table 2, the correlation coefficients are all significant at the 99% confidence level, and the reduction of error (RE) and the coefficient efficiency (CE) are all positive, even for the first-order difference series. The final reconstruction is from 1730 to 2015 (Fig. 5b). The mean and standard deviation (SD) of the reconstruction are adjusted to match those of the instrumental data over their common period.

The mean value of the reconstructed June-July temperature series over 1730–2015 is 10.7 °C, and the SD is 0.8 °C. Warm intervals are 1735–1809, 1932–1967 and 1984–2015, while relatively cold conditions prevail in 1810–1931. Extremely warm early summers (> mean + 1.5SD) are 1739, 1776, 1785, 1805, 1997, 1998, 2000, 2001, 2005–2007 and 2009–2015, while extremely cold early summers (< mean − 1.5SD) include 1779, 1837, 1841–1844, 1852, 1864, 1869–1877, 1916 and 1918. A general warming trend can be observed since the 1870s. The most recent warming period starting from the 1980s reached an unprecedented level since the late 1990s. The last two decades (1996–2015) are the warmest in the context of the past 286 years, with 14 out of the 20 years being extremely warm years.

Fig. 5
figure 5

a Comparison of the observed (solid line) and reconstructed (dotted line) June-July mean temperature during 1961–2015. b Reconstructed June-July mean temperature over the period 1730–2015 (thin line), its 10-year FFT smoothing (thick line), the mean value (solid horizontal line) and ± 1.5 standard deviation (dotted horizontal line) of the reconstructed temperature series

Full size image
Table 2 Calibration and verification statistics for the common period 1961–2015. The values before and after the slash (“/”) denote results for raw and first-differenced data, respectively
Full size table

4 Discussion

4.1 Spatial representativeness and characteristics of the reconstructed June-July mean temperature

The reconstructed temperature series can represent the whole eastern Tibetan Plateau and a large part of central western China in terms of both raw and first-differenced data (Fig. 6). The first-differenced series excludes long-term trends and low-frequency signals and represents high-frequency signals. Using the first-differenced series to make a spatial correlation analysis could avoid information inflation which may be caused by low-frequency signals and trends. In terms of the first-difference analysis results, the reconstructed temperature can represent at least the TP to a relatively reliable level (Fig. 6d). Another feature is that there is a distinct temperature difference between the western and eastern China, which is also evidently shown in April-July mean temperature from the southeastern China (Shi et al. 2017a). The topography and temperature differences between the western and eastern China of the same latitude may make atmospheric circulation (i.e., the East Asian monsoon and South Asian monsoon) more complex.

We compared our reconstructed series with other warm-season temperature reconstructions (Fig. 7), including June-August minimum temperature of Liang et al. (2008), June-July minimum temperature of Yu et al. (2012), July-September mean temperature of Xiao et al. (2015), and July-August maximum temperature of Zhu et al. (2016). These reconstructed temperature series have close seasonal windows and similar climate characteristics with ours and represent the best series that could be used for comparison. As shown, they matched well in most periods, such as the relatively cold periods of 1813–1825, 1840–1850 1864–1880, 1910–1930 and 1965–1985, and the warm periods of 1880–1900, 1930–1965, and 1985-present. The reconstruction of this study is significantly correlated with all other reconstructions at 0.01 level, and the correlations between first-differenced series are considerably higher than those between raw series (Fig. 7), indicating that these tree-ring based reconstructions are more consistent on high-frequency variability. Since the reconstructions are based on tree rings from different locations, climatic homogeneity must be responsible for their consistency.

As for the unstable consistency at low frequency, there are several possible reasons, such as ecological disturbance, local microenvironment, different limiting climatic factors on tree growth, and even human disturbance. In order to evaluate whether there are ecological and/or human disturbances on tree growth, one feasible way is to develop two or more independent site chronologies that have the same limiting climatic factors, and then the consistency of these chronologies can be carefully checked to assess the quality of the chronology in reflecting climatic change. Two site chronologies that meet the above criteria were developed in our study (Fig. 2). The correlations between them are very high, 0.84 for the original series and 0.88 for the first differenced series over the period 1748–2015. Furthermore, the series matched very well visually. The high consistency indicates that it is unlikely that ecological or human disturbances affect the quality of the SP0102 chronology used for our June-July temperature reconstruction.

The reconstructed June-July temperature series can be divided into three major periods, i.e., 1730-1800 s, 1810-1930 s, and 1940s-present (Fig. 5b). The first period features warm temperature with major interannual variability. The second period is cold with large decadal variability. The third period features warm temperature with large decadal variability and a recent rapid warming that reached an unprecedented level in the past two decades. Cloud-radiation feedback may plan an important role in modulating the recent warming (Duan and Xiao 2015) which is also likely a response to the increasing greenhouse gases emissions suggested by two coupled global climate models (Duan et al. 2006). The unprecedented warming caused some mixed environmental changes, such as accelerated glacier retreat, early snowmelt, permafrost degradation, increased net primary production, extended agricultural land area, and increased river runoff contributed by mountain glacier melt (Chen et al. 2015; Kang et al. 2010; Yao et al. 2012).

Fig. 6
figure 6

Spatial correlations of instrumental a, b and reconstructed c, d June-July mean temperature series with the CRU TS4.03 land temperature over 1961–2015 for raw a, c and first differenced b, d data. Non-significant correlations (p ≥ 0.1) are masked out. The green squares denote the sampling sites of this study. The yellow, blue, white and orange triangles denote the sampling sites of Liang et al. (2008), Yu et al. (2012), Xiao et al. (2015), and Zhu et al. (2016), respectively

Full size image
Fig. 7
figure 7

Comparison of the reconstructed June-July mean temperature (red line) with tree-ring based warm-season temperature reconstructions in the nearby regions (black line). a June-August minimum temperature of Liang et al. (2008), b June-July minimum temperature of Yu et al. (2012), c July-September mean temperature of Xiao et al. (2015), and d July-August maximum temperature of Zhu et al. (2016). All series were standardized over their common period 1787–2002, and the bold lines denote the 10-year FFT smoothing. Gray bars denote cold periods in the reconstructions

Full size image

4.2 Periodicity and teleconnections of June-July mean temperature

MTM analysis revealed that the reconstructed temperature series has periodicities of 63.7–51.0, 26.2, 4.5, 3.0, 2.6 and 2.2-2.0 years at 95% confidence level (Fig. 8). The multi-decadal scale cycles fall in the typical periodical range of the Atlantic Multidecadal Oscillation (AMO), and the interannual scale cycles correspond to those of El Niño-Southern Oscillation (ENSO). Therefore, the reconstructed temperature series was compared with the AMO and ENSO series to explore their relationship. Mann et al. (2009) employed a diverse multiproxy network comprising more than a thousand tree-ring, ice core, coral, sediment, and other assorted records to reconstruct global patterns of surface temperature changes over the past 1,500 years, with a decadal-resolution. The decadal-scale AMO series was produced by averaging the spatial temperatures of the corresponding region. Wang et al. (2017) used 46 annually resolved terrestrial proxy records including tree-ring, ice core, and document from the circum-North Atlantic-Arctic region to reconstruct the extended summer (May-September) Atlantic Multidecadal Variability index since 800 CE. Li et al. (2013) used 2,222 tree-ring chronologies from both the tropics and mid-latitudes in both hemispheres to reconstruct a seven-century-long ENSO series. The reconstructed temperature series is significantly correlated with both AMO series at 0.01 level, and the correlation coefficients are 0.37 and 0.42 respectively (Fig. 9). However, the correlation between the temperature reconstruction and ENSO series is not significant.

The relationship between the reconstructed temperature and the Pacific Decadal Oscillation (PDO) was also investigated in this study. The PDO is a long-lived El Niño-like pattern of Pacific climate variability. The PDO series is based on observations (Mantua et al. 1997; Zhang et al. 1997), and is defined as the leading principal component of North Pacific sea surface temperature variability. The correlation coefficient is only − 0.12, not significant at the 0.05 level (Fig. S2). Therefore, the impact of the PDO on June-July temperature in the eastern TP is very low, which was also found using a spatial correlation analysis by Wang et al. (2014).

AMO has modulated temperature variations of the Tibetan Plateau and its surroundings over the past several centuries, i.e., February-July mean temperature in southwest China (Fang et al. 2019), winter, summer, and annual mean temperatures in the southeastern TP (Shi et al. 2017b; Wang et al. 2014, 2015). AMO is also a factor in modulating early summer temperature on the central eastern TP over the past three centuries. Its influences are realized by warming Eurasian middle and upper troposphere, which may be related to the wave guidance mechanism associated with the Asian upper jet (Wang et al. 2009). Both observations and simulations with an atmospheric circulation model indicate that the AMO warm phase can induce a zonal dipole response in sea level pressure across the Atlantic-Eurasia region, with negative surface air pressure anomalies over the North Atlantic (Knight et al. 2006), and high surface air pressure anomalies and corresponding downward atmospheric motion over the TP, causing negative rainfall and positive temperature anomalies over the TP (Shi et al. 2019). The AMO was in a warm phase over the past two decades, which may have contributed to the unprecedented warming to some extent. The warming was also caused by the increasing greenhouse gases and modulated by cloud-radiation feedback. However, the relative contributions of different driving factors are still uncertain, which is worth of further studies.

Fig. 8
figure 8

Results of multi-taper method spectrum analysis of the reconstructed temperature. The solid, dashed and dotted lines indicate the 99%, 95% and 90% confidence levels, respectively

Full size image
Fig. 9
figure 9

Comparison between the temperature reconstruction and two AMO series. The black curves are the reconstructed temperature series. Thin line is original with annual resolution. Thick line is the 10-year FFT smoothing of the original series. a The annual-resolution AMO series (thin red line) and its 10-year FFT smoothing (thick red line) from Wang et al. (2017). b The decadal-resolution AMO series (thick red line) from Mann et al. (2009)

Full size image

5 Conclusion

A robust 286-year chronology was developed from tree-rings collected from two sites on the central eastern Tibetan Plateau. Based on the chronology, June-July mean temperature was reconstructed using a linear regression function that passes all commonly used test parameters, with an explained variance of 65.6%. The recent two decades are the warmest in the past 286 years, with 14 out of 20 years being extremely warm, which clearly shows the unprecedented warming conditions of the last two decades on the central eastern TP. The unprecedented warming will inevitably exert its influence on both mountain glaciers and ecosystems, and further affect the Asian monsoon system, to which enough attention should be paid by the policy-makers. Although the contribution of the AMO to June-July temperature changes was emphasized, the driving forces of temperature changes on the TP are undetermined partly because of its complex topography, spatial heterogeneities and data quality. Therefore, more high-quality and highly replicable temperature reconstructions, like the one in this study, should be produced in different climatic regions of the TP, in order to have an in-depth understanding of the driving forces.