The arid regions of northwestern China account for 30% of the total Chinese territory. Droughts are one of the most devastating climate-related natural harassment faced by the people living in these ecologically vulnerable regions. During the 20th century, several severe droughts had large impacts on economy, society and the environment. For example, the severe drought during the 1920s lead to a low-flow period in the upper and middle reaches of the Yellow River and a low level of the Daihai Lake during 1927–1929. As a consequence, drought-induced famines and diseases led to the death of about four million residents in Gansu, Shanxi, Inner Mongolia, Ningxia and Qinghai (Xu et al., 1997).
Because the instrumental meteorological records only cover the past 50 years and thus contain a limited subset of drought events, they are too short to investigate multi-decadal or longer climate oscillations. Hence, it is difficult to evaluate the magnitude and frequency of modern droughts in a historic context. Precise knowledge of natural long-term climate variability, however, is of great value for local administrations to anticipate and plan reasonable preventive measures for possible future droughts. Therefore, regional climate reconstructions are of great relevance to detect the temporal and spatial extent of the ongoing climate changes and to test whether these changes are still within the range of the natural climate variability during historic times. Among other natural archives, tree ring chronologies from drought-sensitive sites provide annually resolved palaeoclimate proxy records that allow the reconstruction of moisture supply with accurate dating for the past centuries or even millennia (Fritts, 1976; Hughes et al., 1994; Cook et al., 2004; Treydte et al., 2006). Recent reconstructions of precipitation, drought, and river streamflow from tree rings (Hughes et al., 1994; Kang et al., 2002; Yuan et al., 2003; Zhang et al., 2003; Sheppard et al., 2004; Liu et al., 2004a, 2004b; Shao et al., 2005a, 2005b; Li et al., 2006; Liang et al., 2006; Gou et al., 2007; Huang and Zhang, 2007; Holmes et al., 2007; Yin et al., 2008) indicate that there is great potential in developing century-to-millennia long reconstructions of hydrologic variability over arid and semi-arid northern China. To date, however, almost all of these reconstructions are derived from semi-arid northern China including the westernmost Xinjiang Province and the Qilian Mountains located in the northeastern part of the Tibetan Plateau (TP). Few studies concentrated on the northern side of the Qilian Mountains-middle Hexi Corridor in Gansu Province of northern China (Tian et al., 2007; Liang et al., 2009). These drought reconstructions, however, only span the past 150 and 200 years and are barely sufficient to study the patterns of climate variability at multi-decadal and centennial timescales.
Here, we report a new drought reconstruction from the arid Hexi Corridor in interior northwest China derived from a 620-year ring width chronology of Qilian juniper (Sabina przewalskii) growing at elevations around 3000–3500 m. This study site lies at the western distribution limit of the species in arid interior northwest China. The tree ring-based drought record contributes to fill the information gap about climate history in the arid Hexi Corridor during the past six centuries, where long high-resolution histories of climatic variability are scarce.
2. Study area
The study area is situated at the north slope of the middle Qilian Mountains in the northeastern part of the Tibetan Plateau, Gansu Province, northern China. The locations of the sampling site, Qifeng town [39°33′N, 98°05′E, 3000 m above sea level (asl)] and the meteorological station nearby, are shown in Figure 1. The region is presently characterized by continental arid climate and primarily influenced by the Westerlies. Because the region is located very close to the present northwestern boundary of the Asian summer monsoon (ASM), it should be sensitive to variations in the strength of different atmospheric circulation patterns. According to the meteorological station Jiuquan (39°46′N, 98°29′E, 1477 m asl) close to the sampling site, the mean January and July temperatures are − 9.4 and 21.9 °C, respectively. Mean annual precipitation amounts 87 mm for the period 1951–2007, which is certainly a too low estimate for the tree ring site. Eighty percent of the annual precipitation is falling during the summer period from May to September (Figure 2).
This region is of great concern because it serves as a base of agriculture development in arid northwest China, where availability of water resources associated with precipitation variability plays an important role in economic development. It is also well known for its geographical importance to be the most important passage from North China to Xinjiang and Central Asia in historical times.
3. Tree ring material and methods
Qilian juniper (Sabina przewalskii Kom.) grows in open stands on south-facing slopes between 3000 and 3500 m asl. Thus, the trees have no root contact to groundwater and their water supply depends solely on precipitation. In total, 56 cores from 31 trees were extracted from living trees with increment borers. Samples were air-dried, fine-sanded and cross-dated using standard dendrochronological techniques (Stokes and Smiley, 1968). The ring width of each tree core was measured to the nearest 0.01 mm. The cross-dated tree ring series were quality checked with the COFECHA program (Holmes, 1983). The removal of the biological age trend for the individual ring width series was accomplished by the calculation of differences between the raw data and an exponential or linear growth trend curve with the ARSTAN program. To reduce the potential influence of outliers, a data adaptive power transformation was applied to the data before removing biological trends (Cook and Peters, 1997). The detrended tree ring index values were averaged to a standard site chronology by a biweight robust estimate of the mean to minimize the influence of outliers (Cook et al., 1990). Before calculating the final chronologies, we stabilized the variance to minimize the effect of changing sampling size and mean inter-series correlation through time by the methods described by Osborn et al. (1997), Frank et al. (2007) and Büntgen et al. (2008). All calculations were carried out with the ARSTAN program. Two versions of the chronology were produced: the standard (STD) and residual (RES) chronologies. The STD retains low-order persistence, whereas RES is a pre-whitened chronology where significant persistence has been removed (Cook, 1985).
Because sample depth decreases back in time, theoretical or true population signal inherent in the chronologies weakens. Expressed population signal (EPS) and the mean inter-series correlation (Rbar) (Wigley et al., 1984) were calculated to determine the statistically reliable time periods of the STD and RES chronologies using a 30-year moving window with 15-year overlaps. A cutoff value of 0.85 of EPS is generally considered to be a reasonable threshold for acceptance of chronology quality (Wigley et al., 1984).
Monthly temperature and precipitation data used for the calibration of the tree ring climate relationships were obtained from the nearest station Jiuquan and are available since 1951. The Palmer drought severity index (PDSI; Palmer, 1965) monthly 2.5° × 2.5° grid dataset (Dai et al., 2004) was also applied to investigate the incorporating effect of temperature, precipitation and soil moisture on tree growth. The grid point (38.75°N, 98.75°E) closest to our sampling site was used in this study. The PDSI grid point has a time span of 1935–2005, but we only use the PDSI data starting at 1951 because the earliest instrumental observations in the region do not begin until 1951.
Correlation coefficients and response functions (Fritts, 1976) between the ring width chronology and monthly values of temperature, precipitation and the PDSI were calculated for the respective periods 1951–2007 and 1951–2005. The response function was carried out by using the software DENDROCLIM 2002 (Biondi and Waikul, 2004). To test the validity of our climate growth model, we used split sample calibration verification tests (Meko and Graybill, 1995) using the Pearson correlation coefficient (r), reduction of error (RE), sign test and coefficient of efficiency (CE) statistic. Both RE and CE are measures of shared variance between the actual and modelled series. Because the common period of climate and tree ring data is rather short, we also performed a leave-one-out verification (Michaelsen, 1987) for the derived climate–growth relationship. Because this regression-based reconstruction did not capture the full variance in the instrumental data, we also applied the scaling approach (Esper et al., 2005; Büntgen et al., 2010) by adjusting the STD chronology to the mean and variance of the 1952–2007 instrumental record.
To illustrate that our reconstruction and the local instrumental observations reflect large-scale climate variability, we correlated these series with the monthly gridded climate dataset of CRU TS3 (New et al., 2000). The analyses were conducted using the KNMI climate explorer (http://www.knmi.nl).
4. Results and discussion
The mean length of all tree ring series is 307 years, indicating that decadal, inter-decadal and century-scale variability could be recovered (Cook et al., 1995). Both the RES and the STD chronologies have a high mean sensitivity (MS; Fritts, 1976) and inter-series correlation (R). The values of signal-to-noise ratio (SNR), EPS and variance in the first eigenvector (VFE) (Table I) indicate that the chronologies contain a strong common signal probably related to climatic forcing of tree growth. It is noteworthy that the values of R, MS, SNR, EPS and VFE in this study are comparable with those reported in dendroclimatological studies carried out in the southern Qilian Mountains (Shao et al., 2005a), but they are considerably higher than values found in the northeastern part of the Qilian Mountains (Kang et al., 2002). This eastwest trend in the statistical characteristics of tree ring sites is paralleled by a trend of decreasing precipitation and reflects increasing climate sensitivity with decreasing moisture availability.
Table I. Statistics of the standard (STD) and residual (RES) chronologies
EPS > 0.85
ML, mean segment length; MS, mean sensitivity; SD, standard deviation; AC1, first-order autocorrelation; Rbar, mean inter-series correlation of all series; EPS, expressed population signal; SNRC, VFEC and EPSC are signal-to-noise ratio, variance in the first eigenvector and mean EPS over the common interval 1700–2000, respectively.
EPS > 0.85, year and number of trees with signal strength up to 0.85.
The threshold of 0.85 for EPS was reached at a sample depth of four trees after AD 1360 for the RES chronology. In comparison, the STD chronology has weaker signal strength before AD 1390 and around AD 1480 (EPS = 0.717) and 1510 (EPS = 0.82). Running Rbar values are above 0.3 for most of the time for the STD chronology but usually above 0.4 for the RES chronology. Thus, the RES chronology shows higher statistical fidelity in comparison with the STD chronology. However, the STD chronology contains a considerable portion of the climatologically relevant low-frequency information as indicated by its first-order autocorrelation (Table I). Therefore, both chronologies are regarded in the following analysis of climate–growth relationships.
The STD chronology exhibits prolonged periods of low index values at ca. 1420–1530, 1650–1790 and since 1920, and periods of high index values at ca. 1600–1650 and 1800–1920. The period 1420–1530 is characterized by the lowest index values over the past 600 years. Generally, the RES chronology captures similar low-frequency variations as the STD series but with smaller amplitude.
4.2. Dendroclimatic modelling
For both chronologies, the relationship between monthly means of precipitation and temperature and tree growth was examined for the period from June of the year before growth to September of the growth year over the common period 1951–2007. The results of the correlation and response function analysis are shown in Figure 3. As expected, the STD and RES chronologies show similar correlation features. Both chronologies correlate positively with precipitation in most months, whereas they are mostly negatively correlated with mean temperature. Response function analysis shows significant response with precipitation in July–August of the previous year and current June for the STD chronology, and in July–August of the prior year and current May–June for the RES chronology. This finding is consistent with earlier dendroclimatological studies in the Qilian Mountains in Qinghai Province (Tarasov et al., 2003; Zhang et al., 2003; Sheppard et al., 2004; Shao et al., 2005a, 2005b; Huang and Zhang, 2007), and in the Anemaqin Shan (Li et al., 2008) in the northeastern TP (Qin et al., 2003). The chronologies correlate negatively with mean temperature in July–August and November–December of the previous year, as well as January–March and May–June of the current growth year, although only June of the current year is significant for both the STD and RES chronologies. Negative effects of high temperatures on tree ring growth in the early growing season were also reported in other tree ring studies on the northeastern TP (Qin et al., 2003; Zhang et al., 2003; Sheppard et al., 2004; Shao et al., 2005a, 2005b; Liu et al., 2006). The negative correlations with mean temperatures are probably a consequence of the negative correlation between temperature and precipitation as reflected in the instrumental data (Liang et al., 2009). It can be derived that warm and dry conditions during the growing season hamper juniper growth at this arid site, whereas humid conditions during the summer preceding the growth year and during the vegetation period favour growth. This argument is corroborated by the fact that the correlations of tree ring width with the PDSI are positive in all months for both chronologies although significant correlations only occur in June. Thus, we conclude that temperature does not play an important role for tree growth at Qifeng except for very warm conditions during summer that increase drought stress for the trees and confirm the general nature of the Qifeng tree ring record as a proxy for moisture availability.
We also investigated the correlation coefficients between ring width and various seasonal assemblages of average temperatures to determine the most suitable seasonal window for a climate reconstruction. Highest correlations with seasonal averages of temperature are found during March to June with r = − 0.20 with the STD chronology and r = − 0.16 with the RES chronology. In contrast, the seasonal average of PDSI during October–September (annual) and April–August (summer) seasons shows correlation coefficients of 0.52 and 0.56 for the STD chronology (p < 0.01) and 0.43 and 0.47 for the RES chronology (p < 0.05), respectively. However, the highest correlation found is with the annual precipitation averaged from July of the year preceding growth until June of the growth season, with correlation coefficients of 0.675 and 0.59 for the RES and the STD chronology, respectively (p < 0.01 in both cases). The positive response of tree ring width to annual precipitation is identical with results of Sheppard et al. (2004), Shao et al. (2005a, 2005b) and Liu et al. (2006) for juniper stands on the northeastern TP. Because our aim is to provide information on long-term (decadal to centennial timescales) precipitation variations, we selected the STD chronology to reconstruct annual precipitation for the period from prior July to June of the growth year.
We used a linear regression model to reconstruct drought history during the past several centuries. The regression model accounts for 34.9% of the observed instrumental precipitation variance during the period 1952 to 2007 (Table II). However, it is obvious that the reconstruction systematically underestimates precipitation maxima, which is a typical weakness of tree ring-derived climate reconstruction (Fritts, 1976). One reason might be that a significant amount of rainfall during strong rainfall events results in surface runoff on the strongly inclined slopes and can thus not be used by the trees for biomass production. On the other hand, water saturation of the soil after strong rainfall events might lead to a saturation effect so that additional available moisture does not lead to more growth. Although measured soil moisture data are not available, we favour the first explanation based on our knowledge of the study area and due to the structure of the mountain soils that are coarse grained and do not show any sign of stagnant water. In this paper, however, we focus on long-term precipitation trends rather than on precipitation maxima. As Figure 5(a) shows, long-term trends in the precipitation data are adequately captured by our reconstruction.
Table II. Results of regression model statistics for different calibration and verification periods
rcal, correlation coefficient for the calibration period; R2, explained variance of the regression model for the calibration period; R2adj, R2 adjusted for the loss of degrees of freedom; rver, correlation coefficient for the verification period; r2, explained variance of the regression model for the verification period; RE, reduction of error; CE, coefficient of efficiency.
Sign test is the number of sign agreements/disagreements of paired actual and estimated departures from their respective mean.
To test the validity of our model, we used split sample calibration verification tests (Meko and Graybill, 1995). The reduction of error and coefficient of efficiency statistics have positive values in all cases (Table II), thus confirming the reliability of the derived reconstructions (Fritts, 1976). The leave-one-out verification for the full period of available temperature data produced a RE value of 0.305 (Table II). The results of sign test between actual and estimated values in the verification period were significant at the 99% confidence level. These results confirm the robustness of the reconstruction model. For the final annual temperature reconstruction, we used the full period 1952–2007 of the observational climate data for the calibration of the regression model. Besides, we show the scaled STD chronology adjusted to the mean and variance of the instrumental data (Figures 4 and 5). In comparison with the regressed climate reconstruction, the scaled version shows a higher overall variability as the regression-based reconstruction tends to underestimate climatic extremes (Esper et al., 2005).
4.3. Annual rainfall reconstruction
The reconstructed annual (July–June) and low-pass filtered precipitation for Jiuquan region in the Hexi Corridor is presented in Figure 5(b). The reconstruction contains considerable low-frequency climate variations during the past 620 years. The long-term means of the precipitation reconstruction based on regression and on scaling are 99.3 and 104.8 mm for the period AD 1390–2007, respectively. According to regression-based reconstruction, dry periods with below-average precipitation occurred in AD 1414–1424, 1451–1529, 1650–1791, 1921–1936, 1950–1979 and 1986–2007. Among them, the interval AD 1451–1529 witnessed the most severe and extended dry episode in the Hexi Corridor over the past 620 years. The 20th century was dominated by dry conditions interrupted only by several short humid periods, but it was not abnormally dry in the context of the past six centuries. Periods of relatively wet years are identified for AD 1390–1413, 1425–1450, 1530–1649, 1792–1920, 1937–1949 and 1980–1985. In general, our precipitation reconstruction shows similar variations of humidity during the past 200 years than the reconstructions made by Tian et al. (2007) and Liang et al. (2009). In comparison with the regression-based reconstruction, the scaled reconstruction indicates considerably wetter conditions during 1390–1413, 1425–1450, 1570–1630 and 1790–1920. The dry period 1460–1520 appears much more arid than in the regression-based reconstruction, whereas for other drier periods do not differ strongly between both reconstruction methods.
To demonstrate that our reconstruction represents regional precipitation variations, we conducted spatial correlations between our precipitation reconstruction, instrumental July–June precipitation series and the updated 0.5 × 0.5 gridded precipitation of CRU TS3.0 (Climate Research Unit, New et al., 2000) for the period 1951–2007 by the use of the KNMI climate explorer (http://climexp.knmi.nl). It is found that the Qifeng ring width record (Figure 6(a)) and actual Jiuquan precipitation series (Figure 6(b)) correlate significantly with annual (July–June) gridded surface precipitation and show very similar spatial correlation fields. Significant positive correlations are found with the southern and northern sides of Qilian Mountains in the northeastern TP, with highest correlations occurring in the middle and western Qilian Mountains, especially in the arid Hexi Corridor. The results confirm that our annual precipitation reconstruction captures broad-scale regional climatic variations.
4.4. Comparison with other proxy precipitation records
Several moisture-sensitive tree ring width series have been developed from the southern and northern sides of the Qilian Mountains (Figure 1). Shao et al. (2005a, 2005b) developed a reconstruction of annual precipitation (for the period from previous July to current June) for the last 1000 years for the eastern Qaidam Basin (Delingha and Wulan region) south of the Qilian Mountains (Figure 1), capturing 65% of the precipitation variance in the calibration period 1955–2002. Kang et al. (1997) developed a chronology from Dulan region, southeast of Delingha. Later studies (Zhang et al., 2003; Sheppard et al., 2004; Liu et al., 2006) showed that tree ring width was sensitive to annual precipitation variations. Kang et al. (2002) reconstructed annual streamflow for the Heihe River and Zhamashike River to the east of our sampling site, both originating from the southern Qilian Mountains. The Delingha and Dulan chronologies have adequate replication after AD 1390, comparable with our Qifeng RES chronology. To meet the replication criterion, we truncated the Heihe and Zhamashike tree ring series before AD 1400 and 1500, respectively. For better visual comparison, all reconstructions of moisture variability were scaled to a mean of zero and smoothed with an 11-year Fast Fourier Transform (FFT) filter to highlight low-frequency climate signals. Our new reconstruction shows similar wet/dry periods to the other proxy records. Regional dry conditions during AD 1450–1525, 1650–1730, 1921–1935 and 1955–1975 and wet conditions during AD 1390–1449, 1565–1645 and 1835–1902 found in this study synchronously occurred in other regions of the southern and northern Qilian Mountains. All reconstructions indicate that the amplitude of precipitation variation during the 20th century lies within the variability range of the past 620 years (Figure 7). The common large-scale climate signals found in our reconstruction and the compared records suggest that our reconstruction represents broad-scale regional climatic variations. It is noteworthy that four prolonged periods of high microparticle concentrations (MPC) during AD 1440–1580, 1670–1770, 1810–1830 and 1900–1950 discovered in nearby Dunde ice core (38°06′N, 96°24′E, 5325 m asl) (Mosley-Thompson et al., 1993) broadly correspond to the five regional drought events mentioned above. Because the ice core MPC is an indicator of dust storm frequency and indirectly reflects wet/dry climate conditions, the good correspondence provides further evidence that our reconstruction is capable of reflecting low-frequency climate information.
Discrepancies in regional moisture variations are found during the 20th century, when the southern Qilian Mountains experienced a humid interval, whereas alternating wet and dry conditions occurred in Jiuquan, Heihe and Zhamashike in the northern Qilian Mountains. Furthermore, the Jiuquan precipitation curve witnesses a dry interval during 1985–2007. The 20th century humidification found in the southern Qilian Mountains is consistent with a dendroclimatological study in the Anemaqin Shan by Li et al. (2008), who reported that a persistent trend of moisture increase occurred in that region since the 1880s. The generally dry conditions occurring at Jiuquan in the northern Qilian Mountains in the 20th century are still evident when we used a cubic spline with various frequency response cutoff to remove the biological age trend in our raw ring width data. Whether these differences reflect regional variations of climate variability or if human impact or other environmental factors are involved awaits further investigations.
In the southern Qilian Mountains, the driest period during the 20th century occurred during 1920–1939 and has also been found in tree ring chronologies from Inner Mongolia (Hughes et al., 1994; Liang et al., 2003, 2006; Liu et al., 2004a). At Jiuquan of the Hexi Corridor, however, it was only a moderately dry interval compared with two following more severe drought events during 1955–1975 and 1986–2007. Additional discrepancy is found around AD 1745–1790/AD 1800, when our reconstruction shows a dry pattern, whereas the tree ring chronologies from the southern Qilian Mountains display wet conditions.
Of particular interest is the regional drought event during AD 1450–1525, which characterizes the most intense and extended dry spell in each of the compared records during the past several hundred years (Figure 8(a)). The period AD 1450–1500 is also documented by tree ring records from the Anemaqin Shan in the eastern TP as the most severe dry epoch in that area during the last 700 years (Gou et al., 2007; Li et al., 2008). The dry period 1450–1525 was not only recorded in the northeastern TP but also reported from eastern and northern China (region C in Figure 1). Several decadal-scale drought-frequency series derived from historical documents witness an extended drought in the semi-arid zones of northern China (Yan et al., 1993; Xu et al., 1997; Wang et al., 2004) and eastern China (Figure 8(c); Zheng et al., 2006) during this period. In addition, the composite record of δ18O in stalagmites of Longquan Cave (Qin, 2008) and Heshang Cave (Hu et al., 2008) in the Asian monsoon realm (courtesy from Tan Ming, personal communication) show the most δ18O-depleted period during the past millennium in AD 1450–1520 (Figure 8(d)). The stalagmite δ18O composite is considered as an indicator of intensity variations in the Asian summer monsoon. Analyses of the δ18O record from Dongge Cave (Wang et al., 2005) confirm this view. Therefore, we come to the same conclusion as Li et al. (2008) that this large-scale drought episode may have been caused by a substantial monsoon failure.
General Circulation Model (GCM) modelling results indicate that changes in solar activity combined with volcanic forcing were mainly responsible for the weakening of the Asian summer monsoon during the mid-15th century drought and for the drought event found in our precipitation reconstruction record around 1690 (Figure 8(b)) (Liu et al., 2009). This finding encouraged us to compare the Jiuquan precipitation reconstruction with a PDSI reconstruction derived from tree rings for the westerly dominated central Tian Shan region (region D in Figure 1). As shown in Figure 8, our precipitation reconstruction is out of phase with the precipitation reconstruction from the central Tian Shan (Figure 8(a)) during the past 300 years. In contrast, our reconstruction shares many common features with the Asian summer monsoon record indicated by the stalagmite δ18O over the last several centuries. It seems that Jiuquan in the Hexi Corridor was under the control of the Asian monsoon circulation at inter-decadal to centennial timescales in the past centuries although modern meteorological observations show that this region is presently predominantly controlled by the Westerlies. In particular, our precipitation reconstruction exhibits a downward trend during 1985–2007, implying that a declining intensity of the Asian summer monsoon contrasts with a consistent moisture increase in the central Tian Shan brought by the Westerlies. Likewise, the Tian Shan PDSI was below the long-term mean during the 19th century, when wet conditions occurred in the Hexi Corridor and the Asian monsoon area. The contrasting trend in moisture variation between the monsoon-dominated and westerly dominated areas was also observed in the context of the whole Holocene period (Chen et al., 2008). Seen together, these results suggest an out-of-phase relationship between the Asian summer monsoon and the Westerlies. However, the mechanism how these two circulation systems interact and how they control regional moisture variability at various timescales awaits further investigation.
The study was jointly funded by the 973 project (2009CB421306), the Chinese Academy of Sciences (CAS) 100 Talents Project (29082762) and the NSFC (grant no. 40671196 and 40811120028). It is also supported by the CAREERI of CAS (grant no. 51). Bao Yang appreciates the support by the Alexander von Humboldt Foundation. We also thank two anonymous reviewers for useful comments to improve the manuscript.