Tree-ring based annual precipitation reconstruction for the southern Three-River Headwaters region, China

The Three-River Headwaters (TRH) region (31°39′–36°16′N, 89°24′–102°23′E) is in the hinterland of the Tibetan Plateau (Liu et al. 2005). The TRH region is the source of the Yangtze, Yellow, and Lancang rivers. This important region of the Tibetan Plateau has provided large amounts of water resources for most Asian countries (such as China, Myanmar, Laos, Thailand, Kampuchea, Vietnam, etc.). Thus, climate change relating to runoff has important influences on the social and economic development of this region. In recent decades, the impact of climate change has progressively been causing deterioration of the ecological environment of the region which has resulted in drought or floods, crop failure and famine, and changes in the hydrological regimes of monsoonal Asia (Zhao & Zhou 2005; Harris 2010; Shao et al. 2010a, 2010b; Wu et al. 2011). Understanding long-term precipitation variability in the region is important because precipitation and the availability of water resources in the region are critical for both the present and the future of the TRH region. However, meteorological stations were not installed in this area until the 1950s, which limits the analysis of long-term climate trends. This situation could be greatly improved by using a dendroclimatology method. Due to the accurately dated, continuous, high-resolution, precisely measured ring widths, and sensitive relations to climate, tree rings play an important role in reconstructing past environmental and climatic changes over the past millennium on regional, hemispheric, and even global scales.

Numerous dendroclimatology studies have been conducted in the TRH region (Sheppard et al. 2004; Shao et al. 2005, 2010c; Liu et al. 2006a, 2006b; Gou et al. 2007a, 2007b, 2008; Li et al. 2008; Wang et al. 2008; Fan et al. 2008, 2009; Gou et al. 2010, 2014a, 2014b; Yang et al. 2014), and improved our understanding about climate change over the past century or even longer in the TRH region. However, considering the immense size of the TRH region, the geographical distribution of the dendroclimatic records is insufficient, and only a few of them were related to variations in precipitation or drought in the southern TRH region (Wang et al. 2008; Fang et al. 2010a, 2010b). Previous studies have provided valuable insights on climate in the past. However, there have been few tree-ring based reconstructions of regional precipitation for the southern TRH region, and numbers of tree-ring chronologies have been very small. Also, only one or two meteorological stations are used in these studies, and the inclusion of more stations in the average meteorological records would represent broader regional climatic conditions. Therefore, we used the averaged dataset of the four stations. The Empirical Mode Decomposition (EMD) method (Huang et al. 1998) was employed to extract multi-scale variability of the reconstructed precipitation, and may provide more comprehensive understanding of the differing time scales for intrinsic climate signals in our study. In the present work, we combine four moisture-sensitive tree-ring width chronologies archived in this area to extract the regional climate signals, and reconstruct a 461-year annual precipitation of the southern TRH region. Furthermore, we explore the commonalities in regional precipitation of the past.

The southern TRH region is located in the east-central area of the Tibetan Plateau (Figure 1). This region is arid with a plateau-continental climate (He 2008). Its climate is affected by the South Asian and the East Asian monsoon. The annual mean temperature is −2.42°C–2.87°C and annual rainfall is 490–680 mm (1951–2006) (He 2008). The tree species used in this study is Juniperus przewalskii Kom., a conifer with a long life span (Zhou et al. 1986).

The meteorological data used in this study are monthly total precipitation and monthly mean temperature from Qumalai, Zhiduo, Zaduo, Qingshuihe, and Yushu meteorological stations (Figure 1 and Table 1). We used these stations because they are the nearest meteorological stations. The climate of the southern TRH region is characterized by mild, wet summers and cold, dry winters (Figure 2). The precipitation distribution has a clear summer maximum reflecting the effect of the Asian summer monsoon. To confirm the coherency of the meteorological data we have also checked for the significant correlations among the five stations for each month (Table 2). For the reason of effectively reducing the small-scale noise or stochastic components contained in a single station and represent broader regional climatic conditions (Davi et al. 2006; Bao et al. 2015), we used the arithmetical averaged dataset of the five stations for our analyses.

The four sampling sites are located in the southern TRH region: Angsai village site (AS), located in Zaduo, Qinghai Province, China; Xiarishi village site (XRS), located in Zhiduo, Qinghai Province, China; Yuegai village site (YG), located in Qumalai, Qinghai Province, China; and Bagan village site (BG), located in Qumalai, Qinghai Province, China (Figure 1 and Table 1). Tree-ring samples were collected from healthy juniper trees. Most samples were taken from isolated trees or trees in small groves to reduce the effects of competition on tree growth by neighboring trees. One hundred and eight trees were collected in this study. All tree-ring samples were taken back to the tree-ring lab. After mounting and sanding, the annual widths of cores were measured with a LINTAB tree-ring measurement station at 0.001 mm precision. The quality of cross-dating was examined with the computer software COFECHA (Holmes 1983; Cook & Kairiukstis 1990; Fritts 2001). There are very few missing rings in the samples. Therefore, a missing ring in other samples could be determined by comparing the skeleton-plotting and ring width series plotting with the samples from younger trees. To retain more climate signal, each tree-ring series was detrended conservatively by fitting negative exponential curves. Finally, the detrended tree-ring series were used to develop the standard (STD), residual (RES), and arstan (ARS) chronologies with the computer software ARSTAN (Fritts 1976; Cook 1985; Fang et al. 2010a, 2010b) In this study, we used the STD chronology, which preserves more climate signals (Table 3). Figure 3 shows the chronology and sample depth (number of core/tree) of the four sampling sites. Due to sample size decreases in the early phase of the chronology, the expressed population signal (EPS > 0.85) was used to evaluate reliability of the tree-ring width chronology (Wigley et al. 1984). Figure 3 also demonstrates that the variations of the four chronologies are highly similar to each other. Table 4 shows the correlation coefficients between the chronologies. These high correlations indicate that tree growth at these four sites is limited by a similar set of environmental factors, most likely related to climate. Accordingly, we decided to use a principal components analysis (PCA) to combine the chronologies into a single predictor variable. The PCA of the ARSTAN chronologies resulted in the first principal component (PC1) with an eigenvalue more than one that explained 84.5% of the variance. For subsequent dendroclimatic analyses, we used PC1 as the primary predictor variables. The reliable portion of the four tree-ring chronologies extended from 1553 to 2013 (461 years).

Correlation analysis was applied to identify the relationship between tree-ring growth and climate factors. Linear regression analysis was utilized to reconstruct the precipitation. To understand the low-frequency variability of the reconstruction, the reconstructed series was subjected to the 10-year fast Fourier transformation (FFT). FFT is an algorithm that samples a signal over a period of time (or space) and divides it into its frequency components. These components are single sinusoidal oscillations at distinct frequencies, each with their own amplitude and phase (Van Loan 1992). EMD (Huang et al. 1998) is a useful method for the analysis of non-stationary and non-linear data. It can decompose any complicated data into finite and often a small number of intrinsic mode functions (IMF). The EMD was used to estimate the periods in the precipitation record.

The PC1 of the four chronologies were significantly and negatively correlated with temperature and significantly and positively correlated with precipitation (Figure 4). The PC1 of the four chronologies correlate positively with precipitation of the previous September–October, the current April–May, and the current July–August, but negatively with temperature of the current February, May, and July.

Based on the above climate correlation analysis results, we concluded that total September–August precipitation is the most appropriate seasonal predictand for developing a climate reconstruction from the tree-ring width time series.

During the 1969–2013 calibration period, the reconstruction explained 38.1% of the variance (36.6% after adjustment for loss of degrees of freedom). Split-sample calibration-verification test results (Table 5) show high correlation coefficients in both calibration and verification periods (Meko & Graybill 1995). Results of the split calibration-verification test showed that the values of reduction of error (RE) and coefficient of efficiency (CE) were positive. The results of the sign test (ST) were significant at the 0.01 level. The transfer model therefore successfully reconstructed the annual precipitation over the 1553–2013 period. We compared the reconstructed precipitation with the meteorological measurement data in order to verify the reliability of the reconstruction results. The comparison indicates close agreement between the meteorological measurement and the reconstructed series (Figure 5).

The reconstructed total precipitation from previous September to current August for the southern TRH region is presented in Figure 6. The mean value and standard deviation (SD) of the reconstructed precipitation for the period 1553–2013 were 476.31 mm and 33.14 mm, respectively.

We defined a wet year as more than +1 SD and a dry year as less than −1 SD. Extremely wet years were defined by values above +2 SD and extremely dry years had values below −2 SD (Li et al. 2007; Chen et al. 2015). Under these definitions, 71 dry years and 65 wet years occurred during the entire period. There were 22 extremely dry years and 9 extremely wet years (Table 6). As shown in Table 6, there were 9 extremely dry years in the 17th century, 4 extremely dry years in the 18th century, 2 extremely dry years in the 19th century, and 4 extremely dry years in the 20th century. This indicated that more extreme single-year droughts occurred in the 17th, 18th, and 20th centuries than 19th century in the southern TRH region. The wettest year is 1624 (590.00 mm) and the driest year is 1749 (355.92 mm).

Wet periods (lasted longer than 11 years) were 1615–1630, 1657–1683, 1701–1733, 1756–1786, 1798–1816, 1844–1855, 1864–1875, 1885–1912, 1933–1952, and 1977–1989. Dry periods (lasted longer than 11 years) were 1567–1597, 1604–1614, 1641–1656, 1684–1700, 1734–1755, 1817–1830, 1913–1932, 1953–1971, and 1990–2005. The longest wet period is 1701–1733 (33 years) and the longest dry period is 1567–1597 (31 years).

The statistical parameters of the IMF components are shown in Figure 7 and Table 7. IMF 1 and IMF 2 were significantly correlated with our reconstruction and were the major components. The amplitudes of IMF 1 and IMF 2 were very high. The secondary components were IMF 3, IMF 4, and IMF 5. In contrast, there was no statistically significant correlation with IMF 6. The major periodic oscillations of IMF 1–IMF 5 were about 2–5 years (IMF 1), 6–10 years (IMF 2), 11–18 years (IMF 3), 28–60 years (IMF 4), and 90–120 years (IMF 5). The 2–5 year cycle may relate to the quasi-biennial and Southern Oscillations influenced by alternating east- and west-wind regimes in the equatorial stratosphere lasting 26–30 months (Naujokat 1986). The 6–10 year cycle may be associated with El Niño–Southern Oscillation (ENSO) variability (Allan et al. 1996). Moreover, IMF 4 showed relatively stable oscillation on 28–60 years cycle. Larger fluctuations are seen at the front and back of our reconstruction. The IMF5 component revealed the potential for an unstable oscillation on a centennial scale. However, low correlation with our reconstruction suggests that this centennial signal has only a low explanatory power.

Temperature affects tree growth by influencing soil moisture, plant respiration, and transpiration (Sheppard et al. 2004; Wang et al. 2008; Fang et al. 2010a, 2010b). These effects are exacerbated if precipitation is not adequate during the growing season. This pattern can often be inferred from tree-ring width chronologies that are positively correlated with rainfall and negatively correlated with temperature (Zhang et al. 2003; Qin et al. 2004; Fan et al. 2008). After combining the monthly climate data, the total precipitation from previous September to current August showed high correlations (r = +0.62, p(a) < 0.01, n = 45) with the PC1 of the four chronologies.

Comparison with the dry/wet events recorded in the history of the southern TRH region from the yearly charts of dryness/wetness in northwest China for the last 500-year period (Bai 2010) shows many extreme wet years (such as 1561, 1624, 1659, and 1726) and dry years (such as 1589, 1649, 1693, 1740, 1749, 1872, 1918, and 1995) in the reconstruction. For instance, there was a serious drought in Zaduo, Zhiduo, and Yushu in 1872 (Wen 2006), which exists in our reconstruction. The Zaqu River in this area was also dry in 1918, and there was a multiyear drought centered on 1995 in the area (Wen 2006).

We compared our reconstruction with three tree-ring chronologies of nearby areas: Delingha (Shao et al. 2005), Wulan (Shao et al. 2005), and Shenge (Sheppard et al. 2004) (Figure 1). Correlation analysis indicated that our reconstruction has strong correlation with tree-ring chronologies from the three areas. The correlation coefficients are 0.65 (p(a) < 0.01, n = 440), 0.60 (p(a) < 0.01, n = 440), and 0.63 (p(a) < 0.01, n = 440), respectively. Figure 8 shows that most dry periods, such as the 1570s–1590s, 1740s–1760s, 1770s–1800s, and 1830s–1840s, coincide with low-growth periods of trees in these regions. Our reconstruction significantly correlates with the precipitation reconstruction of the northeastern Tibetan Plateau (Fang et al. 2010a, 2010b) (r = +0.68, p(a) < 0.01, n = 454) and the drought reconstruction of the southeastern Tibetan Plateau (Yang et al. 2014) (r = +0.66, p(a) < 0.01, n = 454). The correlations between smoothed reconstruction (apply an 11-year moving average) and these two smoothed reconstructions (apply an 11-year moving average) become even higher, the correlation coefficient is 0.71 (p(a) < 0.01, n = 40) and 0.69 (p(a) < 0.01, n = 40), respectively. As shown in Figure 9, the peaks and troughs in these reconstructions lines match closely. These three areas share the same drought (1600s–1610s and 1680s–1700s) on an inter-annual scale. Dry periods (1690s–1700s and 1740s–1750s) are also consistent with the drought index reconstruction for Zaduo in Qinghai Province, China (Shi et al. 2009). Droughts during the 1950s–1970s period also occurred in the precipitation reconstruction of the northeastern Tibetan Plateau (Liu et al. 2006a). The 1604–1614 and 1641–1656 droughts were also widespread in northwest China. These 17th century droughts are also reported in northeastern and southeastern Tibetan Plateau (Liu et al. 2006a, 2006b; Gou et al. 2013; Yang et al. 2014). The 1684–1700 drought was also reported in tree-ring width or oxygen isotope reconstructed precipitation in the Qaidam Basin (Zhang et al. 2003; Wang et al. 2013), and streamflow reconstructions for the Kherlen River (Davi et al. 2013) and the Selenge River (Davi et al. 2006) in Mongolia. Moreover, we compared our reconstruction with the MADA (Monsoon Asia Drought Atlas) (Cook et al. 2010) dataset in the public section (1553–2005) of corresponding grid point dataset (33.75°N, 96.25°E) (Figure 10). Correlation analyses between our reconstruction and MADA were performed from June to August. The correlation coefficients are 0.66 (p(a) < 0.01, n = 452); this connection suggests a possible link with the interactions between our reconstruction and the monsoon system.

A precipitation reconstruction of the southern TRH region was performed for 1553–2013 based on the tree-ring width chronology of juniper. We established a model between the PC1 of the four chronologies and total precipitation for the interval from September to August. Our reconstruction revealed nine dry and ten wet periods during the past 461 years. Comparison of our reconstructions with tree-ring chronologies from nearby areas shows that most dry periods coincide with low-growth periods of trees in these regions, and that tree growth is similar throughout this region. Our reconstruction also has wet/dry periods similar to the reconstructions from nearby areas. The EMD analysis suggests the existence of significant periods with intervals of 2–5, 6–10, 11–18, and 28–60 years. Our results are preliminary and require confirmation from ongoing dendroclimatological studies of the southern TRH region. Thus, further work should be undertaken to develop more reliable tree-ring networks, and to explore the possible links between climatic variation reflected in the reconstruction and large-scale climate forcing. Continued work in this direction should enable us to understand better the variation characteristics of rainfall and drought in and around the TRH region.

This research was supported by the National Natural Science Foundation of China (41772173).