Volume 40, Issue 3 p. 1477-1491
RESEARCH ARTICLE
Open Access

A 256-year-long precipitation reconstruction for northern Kyrgyzstan based on tree-ring width

Tongwen Zhang

Corresponding Author

Tongwen Zhang

Institute of Desert Meteorology, China Meteorological Administration, Urumqi, China

Key Laboratory of Tree-Ring Physical and Chemical Research of China Meteorological Administration, Urumqi, China

Key Laboratory of Tree-Ring Ecology of Uigur Autonomous Region, Urumqi, China

Correspondence

Tongwen Zhang, Institute of Desert Meteorology, China Meteorological Administration, 46 Jianguo Road, Urumqi 830002, China.

Email: [email protected]

Search for more papers by this author
Bo Lu

Bo Lu

Laboratory for Climate Studies, National Climate Center, China Meteorological Administration, Beijing, China

Search for more papers by this author
Ruibo Zhang

Ruibo Zhang

Institute of Desert Meteorology, China Meteorological Administration, Urumqi, China

Key Laboratory of Tree-Ring Physical and Chemical Research of China Meteorological Administration, Urumqi, China

Key Laboratory of Tree-Ring Ecology of Uigur Autonomous Region, Urumqi, China

Search for more papers by this author
Mamatkanov Diushen

Mamatkanov Diushen

Institute of Water Problem and Hydropower of National Academy of Sciences of Kyrgyz Republic, Bishkek, Kyrgyzstan

Search for more papers by this author
Satylkanov Rysbek

Satylkanov Rysbek

Institute of Water Problem and Hydropower of National Academy of Sciences of Kyrgyz Republic, Bishkek, Kyrgyzstan

Search for more papers by this author
Ermenbaev Bakytbek

Ermenbaev Bakytbek

Institute of Water Problem and Hydropower of National Academy of Sciences of Kyrgyz Republic, Bishkek, Kyrgyzstan

Search for more papers by this author
Feng Chen

Feng Chen

Institute of Desert Meteorology, China Meteorological Administration, Urumqi, China

Key Laboratory of Tree-Ring Physical and Chemical Research of China Meteorological Administration, Urumqi, China

Key Laboratory of Tree-Ring Ecology of Uigur Autonomous Region, Urumqi, China

Search for more papers by this author
Shulong Yu

Shulong Yu

Institute of Desert Meteorology, China Meteorological Administration, Urumqi, China

Key Laboratory of Tree-Ring Physical and Chemical Research of China Meteorological Administration, Urumqi, China

Key Laboratory of Tree-Ring Ecology of Uigur Autonomous Region, Urumqi, China

Search for more papers by this author
Shengxia Jiang

Shengxia Jiang

Institute of Desert Meteorology, China Meteorological Administration, Urumqi, China

Key Laboratory of Tree-Ring Physical and Chemical Research of China Meteorological Administration, Urumqi, China

Key Laboratory of Tree-Ring Ecology of Uigur Autonomous Region, Urumqi, China

Search for more papers by this author
Heli Zhang

Heli Zhang

Institute of Desert Meteorology, China Meteorological Administration, Urumqi, China

Key Laboratory of Tree-Ring Physical and Chemical Research of China Meteorological Administration, Urumqi, China

Key Laboratory of Tree-Ring Ecology of Uigur Autonomous Region, Urumqi, China

Search for more papers by this author
First published: 13 August 2019
Citations: 7

Funding information: Key Laboratory Opening Subject of Xinjiang Uigur Autonomous Region, Grant/Award Numbers: 2016D03005, 2018D04028; National Natural Science Foundation of China, Grant/Award Number: 41605047; Shanghai Cooperation Organisation of Science and Technology Partnership, Grant/Award Number: 2017E01032; Tianshan Cedar Project of Xinjiang Uigur Autonomous Region, Grant/Award Number: 2017XS18

Abstract

In this study, 85 tree cores from 45 Schrenk spruces (Picea schrenkiana) were used to establish a regional tree-ring chronology. A 256-year pJune–May precipitation (where p denotes a month from the previous year) for northern Kyrgyzstan in central Asia was reconstructed using this newly developed chronology. The tree-ring-based precipitation reconstruction tracks the observed data well (r = .622, p < .0001, n = 105). Spatial correlation proved that the reconstructed precipitation series contains climatic signals representative of a larger area, including eastern Kyrgyzstan and parts of Kazakhstan. A comparison between the newly reconstructed precipitation series and four tree-ring-based precipitation reconstructions for the surrounding areas reveals similar variations, particularly in the high-frequency domain. Furthermore, this reconstructed precipitation series matches nine flood (1785, 1786, 1792, 1952, 1953, 1970, 1971, 1973, and 2000) and four drought (1917, 1919, 1927, and 1944) events noted in the historical documents and captures a dry decade that occurred in the 1910s in central Asia.

1 INTRODUCTION

Central Asia, located in the heart of Eurasia, not only sensitively responses to climate change but also has a fragile environmental (Becker and Li, 1990; Chen et al., 2009). Global warming has caused significant snow rate reduction and serious glacial recession in the mountainous areas (Farinotti et al., 2015) and has transformed the hydrological cycle and water system of central Asia—the source area of numerous international rivers (Chen et al., 2017a). The increased unreliability of the water resources could provoke new conflicts among central Asian countries and severely affect regional stability and national security (Stone, 2012).

Climatic variations in central Asia show both integrity and independence in spatial distribution and temporal evolution (Song et al., 2003; Wang et al., 2010). An increasing trend (12 mm/10 a) during 1940–1991 in the annual precipitation of the Tianshan Mountains has been identified based on climatic data from 110 sites in the period (Aizen et al., 1997). Wang et al. (2008) found that there were no specific precipitation responses to the first warming period from the 1920s to the 1940s in arid central Asia, while the whole area had an overall precipitation response to the second warming period after the 1970s. Spatiotemporal analyses of precipitation variations in arid central Asia, using monthly gridded precipitation from the Climatic Research Unit (CRU), have shown that annual precipitation and winter precipitation significantly increased in the last 80 years (Chen et al., 2011). Overall, although reanalysis data can remedy the missing observation in the 1990s and the sparse distribution of meteorological stations in central Asia to some extent, the relatively short length of data does not satisfactorily describe climate variations over centuries.

The widespread coniferous forests in central Asia, comprising mostly local tree species (Juniperus turkestanica Kom., Larix sibirica Ledeb., Pinus wallichiana, and Picea schrenkiana), provide a good opportunity for dendrochronological studies (Esper, 2000; Chen et al., 2012; Zhang et al., 2014; Seim et al., 2016; Opała et al., 2017). Wooden cores derived from central Asia have been systematically collected since the 1990s (Bräuning, 1994; Esper et al., 2007). Many dendrochronological studies have been carried out. These studies involved ring-width variations (Esper, 2000), tree growth climate response (Esper et al., 2003; Winter et al., 2009), and hydro-climate reconstruction (Zhang et al., 2015a; Chen et al., 2017b). However, when compared with dendrochronological studies carried out in surrounding areas, there have been few studies on tree growth response to climate and tree-ring-based climatic reconstruction for central Asia. More studies are needed to enhance and broaden the understanding of climatic spatiotemporal dynamics in this area.

Therefore, to address this knowledge gap, the aims of this study are to (a) reconstruct historical climate series for the study area based on our newly developed tree-ring width chronologies and (b) explore variations in the reconstruction and assess climatic signals inhering in it.

2 MATERIALS AND METHODS

2.1 Sample collection

The tree-ring sampling sites are located in the Kyrgyz Range (Figure 1), which is an active basement-cored range where the northern margin of the Tianshan Orogen overthrusts the Kazakh Platform (Oskin and Burbank, 2007). The Kyrgyz Range extends approximately 375 km from east (Chu River) to west (Talas River), and the maximum elevation is 4,875 m a.s.l. (West Alamedin Peak). The Kyrgyz Range experiences a temperate continental climate. Annual precipitation generally increases with elevation, but winter cyclonic storm systems dominate at low elevations, while summer convective storms are more important sources of rain and snow at high elevations (Aizen et al., 1995). The widespread, long-living, coniferous forest growing in this range provides a good opportunity for carrying out dendrochronological studies to assess tree growth–climate response and evaluate climate variations over a long-term period.

Details are in the caption following the image
Location of the study area and sampling sites (KGS and KGX). The selected gridded climatic data (42°–43°N, 75°–76°E) are represented by the red dash-block diagram

A long-lived (more than 200 years) Schrenk spruce (P. schrenkiana) (Yang et al., 1992) was collected for our study. These spruces often grow up to 40 m in height and 70–100 cm in radius. More than 90% of the forest (from 1,200 to 3,000 m a.s.l. in the Tianshan Mountains) contains P. schrenkiana, which is the dominant tree species in the forest of the mountain zone. Tree-ring cores from two sites were sampled in September 2015. Forest stands are moderately open and the soil thickness of the forest is relatively thin. To avoid sampling non-climatic effects on radial growth, we selected healthy spruces with little evidence of damage from bushfire, landslides, earthquakes, human destruction, or animal invasion. For most sampled trees, we extracted two cores in parallel at breast height from one tree. The two cores were collected from different directions at two flanks of a given tree. To collect tree cores that contained consistent climate signals, the altitude difference between the highest and lowest locations of one sampling site was kept to less than 50 m. In total, we obtained 92 cores from 47 trees using increment borers (bore diameter: 10 mm) at two sites: KGS (location: 42.49°N–75.08°E; elevation: ~2,600 m a.s.l.) and KGX (42.56°N–75.11°E; ~1,990 m a.s.l.) in the Kegeti region.

2.2 Tree-ring core treatments and ring-width measurements

Following standard dendrochronological techniques (Speer, 2010), the sampled tree-ring cores were dried naturally and mounted on a wooden plank with grooves. Each core was sanded using abrasive paper with the mesh number from low to high until the tree-ring cells were clearly distinguished under a microscope. Then, decade, fifty years, and century of tree-rings in each core were marked with different signs using needles under a microscope. At last, every ring width was measured using a binocular microscope (LINTAB Measuring Systems: the standard of North America's Dendrological Research Community) at a resolution of 0.001 mm.

2.3 Tree-ring width chronology development

Two programs, COFECHA (Grissino-Mayer, 2001) and ARSTAN (Cook and Krusic, 2005), were used for cross-dating quality control and developing chronologies. The detrended data from individual tree cores were processed to produce a standard chronology using a bi-weight robust mean to minimize the influence of outliers, extreme values, or biases (e.g., from spurious trends) in the tree-ring indices. For each tree-ring width series, temporal autocorrelation was removed using autoregressive moving average time series models to produce a residual chronology. The pooled model of autoregression was reincorporated into a residual chronology to develop an Arstan chronology containing the persistent, common, and synchronous signals among a large proportion of the series from the sampled tree cores. Finally, standardized, residual, and autoregressive standardized tree-ring chronologies were obtained (Cook and Kairiukstis, 1990). The expressed population signal (EPS) can be used to assess the reliability of the chronology. EPS is an absolute measure of chronology error that determines how well a chronology, based on a finite number of trees, estimates the theoretical population chronology from which it was drawn. The statistical analyses were performed at 20-year intervals with an overlap of 10 years across the chronology. An EPS of ≥0.85 was used to ensure a reliable chronology length (Wigley et al., 1984).

2.4 Climatic data

For further analyses, we selected to use monthly values of precipitation and temperature of the CRU Time-Series (CRU TS 4.01; Harris et al., 2014) of the gridded 0.5 × 0.5° data set (42°–43°N, 75°–76°E), available from 1910 to 2015, to describe the climatic conditions in the study area. These climatic data were acquired from the Royal Netherlands Meteorological Institute (KNMI) Climate Explorer (http://climexp.knmi.nl). A substantial body of dendroclimatological studies have frequently stressed that tree-ring formation is not only influenced by climatic conditions during the growing season but also by climatic conditions in the period prior to the growing season (D'Arrigo et al., 2005; Liu et al., 2011; Opała and Mendecki, 2014). Thus, different combinations of monthly climatic data from the previous year to the current year were used to reveal the relationship between tree growth and climatic conditions.

2.5 Statistical analysis

To evaluate the patterns of variation in various frequency ranges, we used 13-year reciprocal filters to decompose the developed tree-ring chronologies and the reconstructed climatic series into high- and low-pass components (Yuan et al., 2013). The models used for high- and low-pass filter transmits are given by the following equations:
urn:x-wiley:08998418:media:joc6280:joc6280-math-0001(1)
urn:x-wiley:08998418:media:joc6280:joc6280-math-0002(2)

Variation patterns in the original and decomposed chronologies were assessed using Pearson correlations for the original, high-frequency, and low-frequency domains. Correlation analyses were also used to quantify the strengths of climatic signals inherent in chronologies from spruces at the study site. The strongest seasonal relationship between tree-ring width and climatic data was identified, and then a linear regression model was employed for reconstruction. Bootstrap (Young, 1994), leave-one-out cross-validation (Michaelsen, 1987), and split-sample calibration–verification test (Meko and Graybill, 1995) methods were used to evaluate the statistical reliability of the reconstruction model. During the split-sample calibration–verification tests, the climate data period was split into two parts for calibration and verification. Several statistical parameters, including reduction in error, coefficient of efficiency, product mean test, and sign test, were calculated to evaluate the observed and estimated data (Cook et al., 1999). The above analyses were carried out in SPSS and Microsoft Office Excel. Spatial correlation was used to identify coherence between our reconstruction and the gridded 0.5 × 0.5° CRU self-calibrating precipitation 4.01 data set over a large region. To assess their coherence in a larger spatial context, some tree-ring-based precipitation reconstructions and the newly developed precipitation series were standardized using zero-mean normalization and decomposed using 13-year reciprocal filters. The NCEP Twentieth Century Reanalysis (V2) data set was adopted to reveal the historical circulation (Compo et al., 2011). The V2 horizontal resolution is 2 × 2° and covers the period from 1871 to 2012. We used composite analysis to obtain the common features of the atmospheric circulation during wet and dry years.

3 RESULTS

3.1 Variation characteristics of the meteorological data

Figure 2a shows that the highest temperature periods were in summer (June–August) in the study area, and the maximum temperature was in July. A major part of the annual precipitation falls during spring (134.4 mm), which represents 39.6% of the total annual precipitation, whereas much less precipitation falls in winter (50.8 mm). There are two precipitation maxima: in May (56.3 mm) and October (25.2 mm), which represent 16.6 and 7.4%, respectively, of the total annual rainfall. Analysis of climate data for the period since 1910 shows significant increasing trends in total annual precipitation (Y = 0.7978X – 1226.5, R2 = 0.164, p < .001; Figure 2b) and mean annual temperature (Y = 0.0180X – 33.848, R2 = 0.371, p < .001; Figure 2c).

Details are in the caption following the image
Gridded climatic data for the study area during 1910–2015: (a) monthly and (b) annual mean temperature; (c) monthly and (d) annual precipitation. Dashed lines represent trends in climatic data

3.2 Statistical characteristics of chronologies and development of a regional chronology

After cross-dating quality control, three cores from the KGS site and four cores from the KGX site were rejected because of the low correlations between the subseries and the master series. Ultimately, 47 cores from 25 spruces at the KGS site and 38 cores from 20 spruces at the KGX site were used to develop ring-width chronologies. To proceed with the analysis, we utilized the standard chronologies that contained the common variations for series of tree samples and retained a low- to high-frequency common variance. This variance is hypothesized as being dependent on climate (Cook, 1985). Two chronologies, the depths of their samples, and EPS are demonstrated in Figure 3. General statistics for these chronologies for a common period of analysis (from 1910 to 2009) are listed in Table 1. The reliable lengths of KGS and KGX chronologies were 256 (1760–2015) and 106 (1910–2015) years, respectively.

Details are in the caption following the image
Chronologies (thin lines) with their sample depth (dashed lines). Blue lines represent EPS data: (a) KGS, (b) KGX, and (c) REG
Table 1. Statistical characteristics of chronologies for the 1910–2009 common period
Statistic KGX KGS REG
Rate of absent rings 0.273% 0.214% 0.231%
Mean index (MI) 1.005 1.075 1.015
Standard deviation (SD) 0.269 0.308 0.273
Mean sensitivity (MS) 0.236 0.212 0.203
First-order autocorrelation (AC1) 0.481 0.634 0.607
Interseries correlation (trees) 0.298 0.318 0.279
Interseries correlation (all series) 0.327 0.331 0.289
Mean within-tree correlation 0.935 0.834 0.851
Signal-to-noise ratio (SNR) 5.818 18.818 20.307
Expressed population signal (EPS) 0.853 0.950 0.953
The first principal component (PC#1) 0.426 0.367 0.326
Number of trees/cores 20/38 25/47 45/85
First year EPS >0.85 1910 1760 1760

In addition to the reciprocal filters, we also used Pearson correlation coefficients to analyse the three sets of data. These data comprised original unfiltered data, high-pass filtered data, and low-pass filtered data (Table S1). The correlations between the two chronologies in the original, high-frequency, and low-frequency domains over the common period of 1910–2015 are 0.683, 0.650, and 0.717, respectively. The 10 highest-value years and 10 lowest-value years of the KGS and KGX chronologies within the 1910–2015 common period are shown in Table 2. After comparing these extreme-value years, five highest-value years (1971, 1973, 2000, 2002, and 2005) and five lowest-value years (1917, 1927, 1943, 1944, and 2008) were observed in the two chronologies. The results indicated a good coherence of the extreme values between the two chronologies. Therefore, we combined all the ring-width data from the KGS and KGX sites to establish a regional chronology (REG).

Table 2. Highest- and lowest-value years of tree-ring chronologies for the 1910–2015 common period
KGS KGX
10 highest-value years 10 lowest-value years 10 highest-value years 10 lowest-value years
Year Value Year Value Year Value Year Value
1953 2.021 1917 0.044 2000 1.601 1917 0.175
1952 1.757 1984 0.385 1999 1.579 1957 0.420
2002 1.615 1919 0.425 1971 1.521 1918 0.492
2005 1.558 1985 0.471 1970 1.445 1944 0.504
2000 1.541 1945 0.490 1973 1.440 1927 0.509
1973 1.468 1944 0.528 1924 1.417 2008 0.538
2004 1.416 2008 0.555 2003 1.416 1911 0.565
1968 1.403 1979 0.584 1967 1.405 1940 0.566
1969 1.386 1927 0.589 2002 1.385 1943 0.602
1971 1.386 1943 0.609 2005 1.385 1995 0.634

The REG chronology and the depths of the samples are shown in Figure 3c, and general descriptive statistics are listed in Table 1. Values of interseries correlation (trees) and interseries correlation (all series) for the REG chronology are slightly lower than for the KGS and KGX chronologies because of the combination of samples. First-order autocorrelation was used to assess the relationships between tree growth in a current year and previous growth. The autocorrelation values for the KGS, KGX, and REG chronologies ranged from 0.481 to 0.634. This indicates that tree-ring widths possessed low-frequency variance, which was affected by the lag effects of climate and tree physiology. The REG chronology had higher signal-to-noise ratios (SNR) and EPS, and exhibited more climatic signals in the regional chronology. The reliable length of the REG chronology was 256 years (1760–2015) based on the initial year (EPS > 0.85).

3.3 Correlation between tree-ring width and climatic data

A strong biological lag effect was indicated by the high values of first-order autocorrelation for these tree-ring chronologies (Table 1). Therefore, the climatic data per month (1910–2015) from June in the previous year to October in the current year (over a 17-month period) were applied to assess the influence of climatic factors on the radial growth of spruces (Figure 4). The results of the correlation analysis revealed a generally positive relationship between the radial growth of spruces and precipitation. The REG chronology was significantly positively correlated with precipitation in June (r = .326, p < .001), July (r = .355, p < .001), and August (r = .394, p < .0001) of the previous year, and April (r = .250, p < .01) and June (r = .305, p < .01) of the current year. In contrast, the correlation between tree-ring width and mean temperature was relatively low. The REG chronology was negatively correlated with mean temperature in July (r = −.402) and August (r = −.285) of the previous year at the 0.01 significance level. After testing different combinations of months, the highest correlation coefficient was found for the REG chronology and the precipitation from June of the previous year to May of the current year (r = .622, n = 105), while the correlation between regional chronology and mean temperature from July to August of the previous year was distinctly lower (r = −.429, n = 105).

Details are in the caption following the image
Pearson correlations between the REG chronology and climatic data (precipitation and mean temperature)
urn:x-wiley:08998418:media:joc6280:joc6280-math-0003(3)

3.4 Precipitation reconstruction and stability tests

We computed the correlation coefficients between spruces' growth in a radial pattern within our study area and various assemblages of months for the precipitation for the period 1911–2015, so that the most suitable season for reconstruction could be selected. To do this, we reconstructed the pJune–May precipitation (where p denotes a month from the previous year) for the study area using the regional chronology. We also used a linear regression model to portray the relationship between the REG chronology and the precipitation. The model is as follows:where PpJune–May is the pJune–May precipitation for the study area, and REG refers to the regional chronology. The model accounts for 38.7% of the precipitation variance in the REG chronology during the 1911–2015 calibration period. Figure 5a shows that the reconstructed precipitation tracks the observations well and that the first difference (year-to-year changes) between the reconstruction and observation had a correlation coefficient of 0.542 (p < .0001, n = 104) (Figure 5b). The results of the leave-one-out and bootstrap tests (100 iterations in the recomputation process) revealed that values of r, R2, R2adj, standard error (SE), F value (F), and Durbin–Watson (D/W) are nearly equal to those of the original regression model in Equation (3) (Table 3). Table 4 shows statistics from the split-sample calibration–verification tests for the reconstructed precipitation series. The values for reduction in error (RE) and coefficient of efficiency (CE) clearly exceed 0.3 and 0, respectively. Values of the product mean test (t) are positive, which indicates significant accuracy in the tree-ring estimates. Values of the sign test (S), which describe how well the predicted value tracks the direction of the actual data, are 35+/17– (p < .05) for the period 1964–2015, and 38+/14– (p < .01) for 1911–1962, respectively. The above results, which indicate significant accuracy in the tree-ring estimation, all demonstrate that the model (Equation (3)) is stable and reliable. Therefore, we used Equation (3) to reconstruct the precipitation for northern Kyrgyzstan from June of the previous year to May of the current year during the period of 1760–2015 (Table S2); an average value of 333.2 mm and a standard deviation of σ = 32.2 mm (Figure 5c) were obtained.

Details are in the caption following the image
(a) Comparison between the reconstructed (red line) and observed (blue line) pJune–May precipitation for northern Kyrgyzstan. (b) Comparison between first differences (year-to-year changes) of reconstructed (red line) and recorded (blue line) precipitation series. (c) pJune–May precipitation reconstruction for northern Kyrgyzstan since 1760 (solid line). Dashed horizontal lines depict the long-term mean for 1760–2015 and the mean value ±σ. The 10 red and 10 blue dots represent the driest and wettest years, respectively
Table 3. Verification results from the leave-one-out and bootstrap tests for the precipitation reconstruction
Statistic Calibration Leave-one-out mean (range) Bootstrap (100 iterations) mean (range)
r 0.622 0.622 (0.590–0.646) 0.624 (0.502–0.767)
Squared multiple correlation (R2) 0.387 0.387 (0.348–0.418) 0.393 (0.252–0.589)
Adjusted squared multiple correlation (R2adj) 0.381 0.381 (0.342–0.412) 0.389 (0.245–0.585)
Standard error (SE) 45.003 45.002 (43.164–45.223) 0.228 (0.195–0.263)
F-value 65.119 64.510 (54.407–73.153) 68.865 (34.736–147.437)
Durbin–Watson (D/W) 1.382 1.384 (1.318–1.461) 0.865 (0.506–1.192)
Table 4. Statistics of split-sample calibration–verification tests for the precipitation reconstruction
Statistic Calibration (1911–1963) Verification (1964–2015) Calibration (1963–2015) Verification (1911–1962) Full calibration (1911–2015)
r 0.638 0.550 0.549 0.637 0.622
R2 0.407 0.303 0.302 0.406 0.387
R2adj 0.395 0.288 0.381
RE 0.417 0.504
CE 0.108 0.173
t 6.789 4.053
S 35+/17– (p < .05) 38+/14– (p < .01)

3.5 Characteristics of the reconstruction

For the precipitation reconstruction, we defined a wet year as having values above the mean + σ (365.4 mm), and a dry year as having values below the mean –σ (301.1 mm), according to the method of division as used by Liu et al. (2016). The newly reconstructed precipitation series revealed that 43 years could be categorized as wet (accounting for 16.8% of the total), and 36 years as dry (14.1% of the total). The remaining 177 years were categorized as normal (69.1% of the total). The extreme values of years/decades and the mean centenary values are listed in Table 5. It can be seen that the difference between the wettest (1785) and driest (1917) years is 201.6 mm and that the difference between the wettest and driest decades (1780s and 1910s, respectively) is 72.8 mm.

Table 5. Summary of precipitation reconstruction characteristics
10 wettest years 10 driest years 10 wettest decades 10 driest decades Long term
Year Value Year Value Decade Mean Decade Mean Years Mean
1785 416.2 1917 214.6 1780s 372.8 1910s 300.0 1760–1799 345.5
1952 415.5 1835 265.6 1950s 363.6 1830s 300.6 1800–1899 321.7
1953 413.7 1858 266.7 2000s 361.7 1870s 303.7 1900–1999 337.2
1971 403.5 1919 267.5 1960s 357.2 1860s 313.1 2000–2014 350.0
2000 403.4 1895 272.8 1970s 355.5 1770s 314.2 1714–2014 333.2
1970 401.5 1879 273.1 1790s 355.4 1850s 316.2
1792 401.1 1880 274.2 1810s 348.5 1940s 318.6
1973 399.6 1830 275.2 1990s 342.3 1820s 319.5
1786 395.6 1927 276.1 1930s 340.3 1880s 320.1
1783 395.1 1944 277.5 1760s 339.7 1920s 324.3

4 DISCUSSION

Moisture stress is regarded as one of the main climatic stresses involved in the process of tree-ring formation. The stress has been confirmed in dendroclimatological studies using Schrenk spruce tree-ring samples for other regions in central Asia (Chen et al., 2017b; Huo et al., 2017; Zhang et al., 2017a). Our correlation analyses indicated that the positive relationship between the radial growth of spruces and precipitation is strong, while the correlation between the regional tree-ring width chronology and temperature is relatively low. Based on the high correlation between the REG chronology and pJune–May precipitation, the total precipitation over 12 months is regarded as the main climatic limitation on spruce tree-ring formation in the study area.

Wang (2000) showed that the optimum photosynthetic temperature for evergreen conifers ranges from 10 to 25°C and that photosynthesis may cease at temperatures below −5 to −3°C or above 35 to 42°C. Figure 2a shows that the mean temperature in April and October was 4 and 3°C, respectively, while the mean temperature dropped to approximately −5°C in March and November. Thus, the period from April to October was regarded as the growth season for spruce trees in the study area. When the mean results were combined with the results of the correlation analyses, two periods of precipitation affecting the radial growth of Schrenk spruce were identified: June–August of the previous year and April–June of the current year (Figure 4a). July–August of the previous year was identified as the period when temperature affected the radial growth of the spruces (Figure 4b).

June–August of the previous year covers the period from the fast growth season to the end of growth season for the spruce trees in the study area. More rainfall may enhance photosynthesis resulting in an abundance of larger leaves, buds, and roots and abundant nutrient storage (Liang et al., 2001; Liu et al., 2010; Gou et al., 2015). This increases the total photosynthetic and absorptive areas of Schrenk spruces, which are able to absorb moisture or produce growth substances when favourable climatic conditions occur in the following year (Fritts, 1976; Liu et al., 2011). Two negative correlation coefficients at the 0.01 significance level, between the REG chronology and mean temperature in July and August of the previous year, suggested that high temperatures in these months may aggravate the moisture stress in the fast-growing season of the previous year (Figure 4b). April–June of the current year is the beginning of the growth season. Additional rainfall resulting in higher soil moisture may reduce water stress and benefit cambial cell division in the rapid growth season (Liu et al., 2004). In addition, high solar radiation in this arid area, coupled with a thin soil thickness and an open canopy, may result in high evaporative water loss (Yu et al., 2007). Similar tree growth responses to climate have been frequently verified for Schrenk spruce and other species in previous studies, including P. schrenkiana in the Tianshan Mountains (Li et al., 2006; Jiao et al., 2017; Zhang et al., 2017a), Picea obovata in the Altay Mountains (Chen et al., 2014), L. sibirica in Mongolia (Davi et al., 2006), Picea crassifolia Kom. in the Xinglong Mountains (Liu et al., 2013), and P. crassifolia and Sabina przewalskii in the northern Tibetan Plateau (Liang et al., 2009).

The results of the spatial correlation revealed the regional significance of the gridded precipitation data and the ring-width-based precipitation reconstruction (Figure 6). The relatively high correlations between the gridded precipitation used for reconstruction and the gridded precipitation data set for the period 1911–2015 applies to the precipitation field that lies in a large area between approximately 40°–46°N and 70°–89°E. This area includes most of Kyrgyzstan, the northern parts of Tajikistan, southern Kazakhstan, and northwestern Xinjiang, China (Figure 6a). Although the correlation coefficients are relatively low, the results of the spatial correlation analysis show that the correlation area of our reconstruction is similar to the correlation area of the gridded precipitation. The area of precipitation with correlation coefficients >0.40 for the reconstructed precipitation series and the gridded precipitation data set lies in eastern Kyrgyzstan and parts of Kazakhstan, with the highest correlations occurring in the mountainous area surrounding the sampling sites (Figure 6b).

Details are in the caption following the image
Spatial correlation of the (a) recorded and (b) reconstructed pJune–May precipitation for the study area with the gridded meteorological data set for 1911–2014. The solid circle represents the precipitation reconstruction for the study area (PNK), and the dashed circles represent the precipitation reconstructions (PSK, PIL, PIR, and PAK) used in the comparison analyses

Although some previous studies based on the observed and reconstructed data revealed that the coherence of precipitation variation among the different areas of the Tianshan Mountains is relatively weak (Fang et al., 2010; Zhang et al., 2015b), the newly reconstructed precipitation series for northern Kyrgyzstan (PNK) were compared with four tree-ring-based precipitation reconstructions: (a) pJune–May precipitation for southern Kazakhstan (PSK, 1770–2015; Zhang et al., 2017b); (b) pJuly–June precipitation for the Issyk Lake (PIL, 1756–2012; Zhang et al., 2015a); (c) pJune–May precipitation for the Ili region (PIR, 1682–1995; Yuan et al., 2000); and (d) pAugust–April precipitation for the Aksu region (PAK, 1396–2005; Zhang et al., 2009). The locations of these reconstructions are indicated in Figure 6b. Table 6 shows that the correlation coefficients for the standardized precipitation reconstruction for northern Kyrgyzstan and the other precipitation reconstructions for the surrounding areas all exceed the 0.05 significance level in the original and high-frequency domains in the common period of 1770–1995, while the correlations are relatively weak in the low-frequency domain. The good coherence in the original and high-frequency domains indicates that our newly developed precipitation reconstruction may contain relatively precise high-frequency climatic signals in the larger area. Furthermore, the driest decade (the 1910s) was identified in our precipitation reconstruction and also in the four other precipitation reconstructions in the low-frequency domain (Figure 7). This dry decade has been frequently captured in other tree-ring-based hydroclimatic reconstructions for the western (Zhang et al., 2016), middle (Zhang et al., 2013), and eastern (Chen et al., 2016a) Tianshan Mountains, and in the Hexi corridor to the east (Chen et al., 2016b). This indicates that there was a widespread and persistent drought in central Asia and that this drought event predated the 1920s great drought in the semi-arid and arid areas of northern China (Liang et al., 2006). Furthermore, spatial reconstructions of drought for central High Asia based on tree-ring data showed that the wettest period for both the western and eastern mode reconstructions occurred from the 1940s to the 1950s, while the driest period of the western central High Asia was from the 1640s to the 1650s (Fang et al., 2010). Although our precipitation reconstruction does not show the dry period of the 1640s–1650s because of the limit of the length of reconstruction, the 1950s, the second wettest decade (Table 5), matched well with the wettest period for central High Asia. The above comparative analysis of the driest and wettest periods indicated a large area influenced by extreme climatic conditions.

Table 6. Coherence between the precipitation reconstruction for northern Kyrgyzstan and other precipitation reconstructions for the surrounding areas for the 1770–1995 common period
Original domain (n = 226) High-frequency domain (n = 214) Low-frequency domain (n = 214)
PSK PIL PIR PAK PSK PIL PIR PAK PSK PIL PIR PAK
PNK 0.519** 0.398** 0.258** 0.142* 0.566** 0.415** 0.311** 0.140* 0.445** 0.267** 0.110 0.110
  • Note: The table shows correlation coefficients. Results for original, high- and low-pass-filtered reconstructions are shown.
  • Significant at *p < .05; **p < .01.
Details are in the caption following the image
Graphical comparison of the precipitation reconstruction for northern Kyrgyzstan (PNK: black line) and four precipitation reconstructions for the surrounding areas (PSK: red line; PIL: blue line; PIR: grey line; PAK: green line). The red bar indicates the dry decade (1910s)

Unlike the detailed historical documents from southern China, the records of meteorological disasters in central Asia are sporadic and isolated because of the historically low levels of human presence. The limited historical records cannot describe the past climatic changes adequately, but they can verify the reliability of our reconstruction. The 10 wettest and 10 driest years of the precipitation reconstruction for northern Kyrgyzstan were compared with historical documents from Xinjiang Province, China (Wen et al., 2006). Table 7 reveals that the nine wettest and four driest years have been recorded in the historical documents. This shows that our reconstructed precipitation series can capture the signals of flood and drought events for the surrounding area.

Table 7. Wettest and driest years in the precipitation reconstruction for northern Kyrgyzstan in comparison with meteorological records
Wettest year Short description of flood or drought disaster for the study area and surrounding areas
1785–1786 Frequently intense rainfall occurred in Dihua (Urumqi now) on the northern slope of central Tienshan Mountains and inundation occurred in Urumqi River in summer of 1785 and 1786
1792 Great snowfall in Dihua (Urumqi now) on the northern slope of central Tienshan Mountains in winter and spring, 1792
1952–1953 Frequently intense rainfall occurred in the Ili and Akesu regions west Tienshan Mountains in spring and summer of 1951–1952
1970–1971 Intense rainfall occurred in the Akesu, Kezilesu, and Boertala regions, west Tienshan Mountains in summer of 1969 and 1970. Continuous and intense snowfall occurred in the Akesu region on the southern slope of west Tienshan Mountains in July 1970
1973 Frequently intense rainfall occurred in the Akesu region on the southern slope of west Tienshan Mountains from June to July in 1972
2000 Frequently intense rainfall occurred in the Ili region, west Tienshan Mountains from June to August, in 1999. Heavy snowfall (snow depth: 100–200 cm) occurred in the Ili region, west Tienshan Mountains in January 1999
Driest year Short description of flood or drought disaster for the study area and surrounding areas
1917 Great drought occurred in the Ili region, west Tienshan Mountains in 1917. People flee from home, and 90% of rooms are empty
1919 Great drought occurred in the southern Xinjiang province in 1918
1927 Severe drought occurred in the Ili region, west Tienshan Mountains in spring and summer, 1926
1944 Less snowfall and rainfall resulting in great drought occurred in the northern Xinjiang province in 1943

To further study the large-scale climate anomalies associated with the newly developed precipitation series, the circulation during wet years (1924, 1935, 1952, 1953, 1954, 1955, 1966, 1967, 1968, 1969, 1970, 1971, 1973, 1983, 1993, 1994, 1999, 2000, 2002, 2003, 2004, 2005, and 2006) and dry years (1911, 1914, 1917, 1918, 1919, 1927, 1928, 1943, 1944, 1945, 1957, 1979, 1984, 1995, and 2008) were investigated. The large-scale precipitation anomalies can be clearly seen to match the precipitation series for northern Kyrgyzstan (Figure 8). During the wet years, abundant precipitation occurred in central Asia and southeastern China, while below normal rainfall was observed in northern India. In contrast, during the dry years, the water vapour transport from the North Atlantic was displaced northwards and the shift in climate patterns caused a large-scale drought from the Mediterranean Sea to central Asia. In addition, a drought in southeastern China and a flood in northern India can be clearly associated with the dry years in the precipitation series for the study area. This in-phase relationship between precipitation in northern Kyrgyzstan and southeastern China and out-of-phase relationship between precipitation in northern Kyrgyzstan and northern India suggests the potential value of applying our newly developed precipitation series in remote regions.

Details are in the caption following the image
Composite patterns of the precipitation anomalies (shading; mm/month) and horizontal wind at 850 hPa (vector; m/s) during the (a) wet years (1924, 1935, 1952, 1953, 1954, 1955, 1966, 1967, 1968, 1969, 1970, 1971, 1973, 1983, 1993, 1994, 1999, 2000, 2002, 2003, 2004, 2005, and 2006) and (b) dry years (1911, 1914, 1917, 1918, 1919, 1927, 1928, 1943, 1944, 1945, 1957, 1979, 1984, 1995, and 2008) in the newly developed precipitation series for northern Kyrgyzstan

5 CONCLUSIONS

Tree cores were sampled from Schrenk spruces on the northern slopes of the Kyrgyz Range. After eliminating several series because of low correlations with the master series, a regional tree-ring chronology was developed using 85 ring-width series from 45 living and healthy trees. Based on the higher correlation coefficient between ring width and precipitation, a pJune–May precipitation reconstruction for northern Kyrgyzstan in central Asia was developed using the regional chronology. Spatial correlations between the precipitation reconstruction and a gridded precipitation data set were positive for eastern Kyrgyzstan and parts of Kazakhstan. A comparison between our precipitation reconstruction and tree-ring-based precipitation reconstructions for southern Kazakhstan, Issyk Lake, the Ili region, and the Aksu region revealed that the coherence of these reconstructions in the high-frequency domain is stronger than in the low-frequency domain. The precipitation reconstruction matched the observed data well and precisely captured flood (1785, 1786, 1792, 1952, 1953, 1970, 1971, 1973, and 2000) and drought (1917, 1919, 1927, and 1944) events mentioned in historical documents for the surrounding areas. A dry decade occurred in the 1910s according to the newly developed precipitation series; this was corroborated for different parts of the Tianshan Mountains and for the Hexi corridor, China. During dry years, the wet air flow from the North Atlantic shifts to higher latitudes, which results in drought conditions in central Asia. A significant correlation is evident between the precipitation in northern Kyrgyzstan and the rainfall over southeastern China and northern India, implying teleconnection among various sites.

ACKNOWLEDGEMENTS

The research was supported by the National Natural Science Foundation of China (41605047), the Shanghai Cooperation Organization of Science and Technology Partnership (2017E01032), the Key Laboratory Opening Subject of Xinjiang Uigur Autonomous Region (2018D04028 and 2016D03005), and the Tianshan Cedar Project of Xinjiang Uigur Autonomous Region (2017XS18). We thank Paul Seward, PhD, from Liwen Bianji, Edanz Group China (www.liwenbianji.cn/ac), for editing the English text of a draft of this manuscript.