STUDY AREA

The sampling sites are located in Dahuang valley (DHG, 38.11°N, 101.40°E), which is a long and narrow valley in the middle reaches of the Xida River in the eastern Qilian Mountains (Fig. 1). The climate in this region is temperate continental arid. Instrumental records from the nearest meteorological station in Yongchang (38.14°N, 101.58°E, 1976.9 m a.s.l., 1958–2007) show a mean annual precipitation of 200.7 mm and a mean annual temperature of 5.1 °C. The peaks of temperature and precipitation occur in July (Fig. 2). The forest belt ranges from 2700 to 3500 m a.s.l. in the study area. It is dominated by Qinghai spruce (Picea crassifolia Kom.) on the wetter, shaded slopes, and Qilian juniper (Juniperus przewalskii Kom.) on the drier, sunnier slopes at higher elevations. The dominant shrubs are Salix oritrepha C. K. Schneid., Caragana jubata (Pall.) Poir., and Potentilla fruticosa, while the dominant herbs are Stipa bungeana Trin., Stipa breviflora Griseb., Achnatherum splendens (Trin.) Nevski, Artemisia spp., and Iris lactea var. chinensis (Cheng et al., 2004).

FIGURE 2.

Average monthly precipitation and mean temperature at the Yongchang station for 1958–2007.

TABLE 1

Tree-ring sampling sites and climate data used in this study.

TREE-RING DATA

Tree-ring samples used in this study were collected from Qinghai spruce (Picea crassifolia). Five sites were sampled at approximately 100-m vertical intervals between 2850 and 3230 m (Table 1). The mountain slopes for the five sites are all greater than 30°, and the slope in DH3 is particularly steep (45–55°). The forest canopy density is about 0.3 to 0.5. There are small shrubs under the trees, while the herbs are gramineous. Two increment cores per tree were taken from healthy living trees at breast height (approximately 1.3 m). In total, 334 radii from 185 trees were collected. All the tree cores were taken into the laboratory, and standard techniques of dendrochronology were used to process them (Stokes and Smiley, 1968). After the cores were air dried, mounted, and sanded to a high polish, all the samples were carefully cross-dated using skeleton plots. Tree-ring widths were measured to 0.001 mm precision with the Velmex measuring system, and the cross-dated measurements were checked using both the TSAP and COFECHA programs (Holmes, 1983; Rinn, 2003). These methods ensured exact dating of each ring-width series.

FIGURE 3.

The five standardized ring-width chronologies and sample size (number of cores). The vertical dashed lines indicate the starting year of subsample signal strength (SSS) > 0.85.

In order to preserve low-frequency climate signals while removing the age-related growth trends, we used negative exponential functions or linear functions of negative slope to detrend 67–77% of all 118 series at each site. The remaining ∼30% of our tree-ring data were detrended using cubic splines with a 50% frequency response cut-off equal to 67% of the length of series. Finally, individual indices were combined into a standard chronology (STD) by the bi-weight robust mean using the program ARSTAN (Cook, 1985; Cook et al., 2012) (Fig. 3). The subsample signal strength (SSS) statistic was employed with a threshold value of 0.85 to determine the most reliable period of the chronologies (Wigley et al., 1984). Standard statistical parameters of the five ring-width chronologies were evaluated, and these parameters are shown in Figure 4.

FIGURE 4.

Descriptive statistics of the five chronologies along an altitudinal transect. The statistics are: maximum age (Age_max), mean ring width (MW), mean series-length (ML), standard deviation (SD), the percentage of absent rings (PA), interseries correlation (IC), mean sensitivity (MS), and first-order autocorrelation (AC1).

CLIMATE DATA

Climate data used in our study were from the Yongchang meteorological station and data from the Self-calibrated Palmer Drought Severity Index (scPDSI) grid point (Dai et al., 2004; Dai, 2011) closest to the sampling sites (Fig. 1). Monthly total precipitation and mean temperature data obtained from Yongchang station were available only for the period 1958–2007. The scPDSI metric was first proposed by Wells et al. (2004). The large differences between PDSI and scPDSI, on regional scales, relates to the inappropriateness of parameters in the “standard” PDSI calculation; scPDSI has been shown to be much more appropriate for characterizing drought conditions than the original PDSI (Van der Schrier et al., 2006a, 2006b). The record of the closest scPDSI grid point to our sampling sites (38.75°N, 101.25°E; Fig. 1) begins in 1935; however, we only used the data during the reliable period of 1958–2010 in order to make comparisons with the nearest meteorological station (Yongchang), which begins in 1958.

DIFFERENCES BETWEEN THE CHRONOLOGIES' CHARACTERISTICS

The statistical parameters of the five tree-ring chronologies (Fig. 4) show that the trees growing at the intermediate elevation site (DH3) have the maximum values of mean series-length, tree age, mean sensitivity, mean interseries correlation, and percentage of absent rings, while the mean ring width is lowest. These values may be related to the steep slopes and low soil water storage capacity at this site (Fig. 4, Table 1). At the other four sites (DH1, DH2, DH4, and DH5), the values of percentage of absent rings, mean sensitivity, and maximum age of trees are very similar. The values of mean ring width (1.26 mm) and standard deviation (0.572) for the chronology at the lowest elevation site (DH5) are clearly higher than the other chronologies, while the mean series length at site DH5 is the shortest. All five chronologies showed similar values in their first-order autocorrelation coefficients and mean interseries correlations. The mean interseries correlations are high for all sites (0.574–0.646), indicating that our crossdating is reliable, and the tree growth is synchronous across the stand.

CLIMATE-GROWTH RELATIONSHIPS

Considering tree growth in the previous year could affect the ring-width in the current year, we investigated the correlation relationships between standard ring-width chronologies and climate factors from May of the previous year to the current October (1958–2007). Results show that the growth-temperature and precipitation correlation patterns are relatively consistent among the five chronologies (Fig. 5). The four lowest sites show a significant negative correlation (p < 0.05) with July and August temperatures in the previous year and March temperature in the current growth year. Sites DH3, DH4, and DH5 display a common positive correlation with precipitation in the previous August (though the correlation coefficient is only significant [p < 0.05] for DH4). The ring-width chronology from the upper treeline site (DH1) is also significantly correlated with previous August temperature and December precipitation; however, the significant negative correlation with temperatures in the prior July and current March, as reported for the other four chronologies, is not observed. Negative correlations with current year temperatures and positive correlations with precipitation in the previous July–August indicate the radial growth of trees in this area is limited by moisture conditions. This result is consistent with earlier studies conducted in semi-arid and arid regions (Cai and Liu, 2007; Fang et al., 2010; Tian et al., 2007). Studies by Fritts (1962) in the American Southwest found that the ring width of American beech is inversely related to the severity of drought during August of the previous year, and the latewood width of white oak is inversely related to the severity of drought represented by high temperature in June and high evapotranspiration deficits in July. These results are consistent with the present study, indicating that the tree growth in semi-arid and arid regions is related to the moisture supply during August and temperature of the preceding year. High temperatures in July and August would increase evaporation and reduce soil moisture, which might limit the growth of trees and reduce storage of assimilates for growth in the next year (Liang et al., 2010a, 2010b; Fritts, 1976). Hence, ring width in the current growth year shows a negative correlation with July–August temperature and positive correlation with precipitation in the previous year. Fritts (1962) also reported that bud formation and food accumulation in the previous year could produce a lagged growth response. The negative correlation with March temperature in the current year might be related to the respiration of trees. The annual ring width of Qinghai spruce is mainly determined by the width of earlywood, and abundant material storage from the previous year would be beneficial to the formation of earlywood (formed in the second half of the current May) (Venugopal and Liangkuwang, 2007; Liang et al., 2009b). Conversely, higher temperatures in March would result in strong respiration in trees and more material consumption, which might be adverse to the formation of earlywood. It is worth noting that the number of statistically significant correlations between precipitation and these chronologies is relatively small, which may be related to differences between precipitation at the sampling sites and the meteorological station. The elevation for Yongchang station is 1976.9 m a.s.l., whereas the elevation of the sampling sites ranges from 2850 to 3230 m a.s.l. The lack of strong correlations between chronologies and precipitation records may also reflect differences in elevation and vegetation between the forested sample sites and desert location for the climate station. However, due to the lack of weather stations in the high-elevation mountain areas of the arid inland regions of northwest China, we are prevented from using better climate records in this study.

FIGURE 5.

The correlation coefficients between the five standardized tree-ring width chronologies and mean monthly temperature (left) and precipitation (right) for 1958–2007. The dashed lines indicate 95% confidence level. (P6 = June of the previous year, C6 = June of the current year, etc.).

Correlation analysis between the standard chronologies and scPDSI data further confirmed that tree growth in this region is predominantly constrained by drought. As shown in Figure 6, the correlation coefficients between chronologies and the scPDSI data are generally higher and the significantly correlated months are more contiguous than the correlations with temperature and precipitation. This is especially true for the chronology from the lowest elevation site (DH5), which shows a significant (p < 0.05) correlation with scPDSI from the prior May to current October (except for July in the current growth year). The strong correlations with scPDSI found in all the chronologies suggest that drought affects tree growth in this area.

While moisture conditions are shown to be the limiting factor for radial growth of trees in the region, our study also found that altitudinal factors have rather weak effects, as similar characteristics and climate-growth relationships are found in the chronologies at different elevations. Consistent and significant correlations between tree rings and precipitation/soil moisture along an altitudinal gradient have been reported in other studies of arid regions of northwestern China (Wang et al., 2005), suggesting that tree growth in arid regions is constrained predominantly by moisture conditions. By investigating the growth variation in Abies georgei var. smithii (Smith fir) along altitudinal gradients (3550–4390 m) in the Sygera Mountains, southeastern Tibetan Plateau, Liang et al. (2010b) showed that the radial growth of Smith fir trees was markedly similar in response to common climatic signals regardless of difference in stand elevation, topographical aspect, and tree age. These results together indicate that climate factors have a more important influence on tree growth than altitude in some regions.

FIGURE 6.

Correlations of the five tree-ring chronologies with monthly scPDSI data during 1958– 2010. The dashed lines indicate 95% confidence level.

TABLE 2

Cross-correlation matrix of the five chronologies over the 1793–2010 period. All correlation coefficients are significant at the 99% confidence level.

DROUGHT RECORDS BY THE FIVE CHRONOLOGIES

Despite slight differences in the statistical parameters of the chronologies and chronologies' climate-growth relationships, all show similar variation patterns across this altitudinal gradient. Cross correlation between individual chronologies shows that all pairings are significant (p < 0.05; Table 2), especially for the chronologies with similar elevations. The strong correlations between the chronologies confirm that trees in the study area are influenced by common climate factors, and these tree-ring chronologies could be used to investigate past drought variation. We defined wet spells as periods with ring widths above average growth continuously for more than 10 years, and drought spells as intervals with ring widths below average growth continuously for more than 10 years. The common drought periods revealed by the five chronologies are identified as A.D. 1860–1875, the 1880s, and 1905–1935. Common wet intervals occurred during A.D. 1800–1810 and 1975–1990 (some intervals have an exception for one year). Additionally, A.D. 1928 is the narrowest year in all five chronologies, and all chronologies show a trend to wetter conditions since 2000.

FIGURE 7.

The comparisons between (a) the first principal component (PC1) derived from the five chronologies in this study and (b) the Palmer Drought Severity Index (PDSI) reconstructions derived from Deng et al. (2013) and (c) Cook et al. (2010) in the eastern Qilian Mountains. The thick lines represent their 5-year moving average. Consistent dry periods are shaded by bars with diagonal lines.

When comparing the first principal component (PC1) of the five chronologies to the PDSI reconstructions in the eastern Qilian Mountains derived from Cook et al. (2010) and Deng et al. (2013), we found that the prominent drought periods A.D. 1880–1885, 1925–1933, 1947–1955, and 1997–2003 are consistent in these series (Fig. 7). The drought conditions in our tree-ring records during A.D. 1905–1920 are also consistent with Deng et al.'s (2013) reconstruction as are the drought years A.D. 1793–1798 identified by Cook et al. (2010). However, there also are differences between these tree-ring records, especially during A.D. 1860–1875 and 1980–1990. These discrepancies might be caused by the spatial heterogeneity of drought/pluvial conditions and particularly the use of many remote tree-ring chronologies in Cook's reconstruction (Cook et al., 2010).