Tree ring record of annual runoff for Issyk Lake, Central Asia

Water provides the basics of food supply and productive environment for organisms. There are more than 6,000 lakes with a total area of 12,300 km² in Central Asia (Savvaitova & Petr 1992). Lakes are believed to be not only important water resources in this region but also sensitive indicators of climate change and regional environment variations (Mason et al. 1994). Under the combined influences of climate change and human activity, inland lakes in Central Asia have dramatically changed, leading to a series of ecological disasters in recent decades (Stanev et al. 2004; Cretaux et al. 2009; Bai et al. 2011). Thus, understanding the dynamic distribution and evolution of a given lake in that arid and semi-arid area is meaningful to evaluate the influences of climate change and human activity and to reduce losses caused by hydrologic variations.

Issyk Lake is an alpine closed lake in Central Asia, and has the greatest depth and volume in the world among alpine lakes. In recent years, many studies have focused on variations of runoff and water level of this lake based on instrumental records (Wang et al. 2006; Salamat et al. 2015) and remote sensing data (Li et al. 2011; Cheng et al. 2015). However, no more than 140 years of data hardly describe water variations over centuries. Thus, proxy data are needed to extend our knowledge of hydrologic variability over centuries on seasonal or annual timescales. As one of the best of proxy data sources, tree-ring data are frequently used in evaluating the relationship between tree growth and streamflow change, extending hydrologic records worldwide. This is because they allow precise dating, annual resolution, and comparability with recorded hydrometeorological data (Pederson et al. 2001; Woodhouse 2001; Polacek et al. 2006; Lara et al. 2008; Gou et al. 2010; Maxwell et al. 2011; Cook et al. 2013; Devineni et al. 2013; Shah et al. 2014).

The widespread long-lived coniferous forest in the mountainous area of Central Asia provides a good opportunity for dendrohydrologic studies to evaluate water bodies over a long period. Wooden cores from Central Asia have been systematically collected since the 1990s (Esper et al. 2007). Many dendrochronological studies related to ring-width variations (Esper 2000), tree growth climate response (Esper et al. 2003; Winter et al. 2009), and climatic reconstruction (Chen et al. 2012; Zhang et al. 2015a, 2017) have been carried out. Compared with tree ring-based climatic series, dendrohydrologic reconstructions are relatively rare (Chen et al. 2017), and there is limited knowledge of the relationship between the growth of coniferous trees in a radial pattern and surface runoff in Central Asia.

Given the above, the purposes of this study were to: (1) develop a tree ring width-based regional chronology for the mountainous area of Issyk Lake; (2) reconstruct runoff variations for the lake beyond the period covered by observed data; (3) explore variation of the runoff reconstruction; (4) compare with other reconstructions to assess hydroclimatic signals inherent in the runoff reconstruction.

Issyk Lake is between the Terskei and Kungei Mountains of Central Asia, and its center is at 77.33°E and 42.42°N (Figure 1). The lake is 178 km long and 60.1 km wide, and its area is approximately 6,236 km². Its average depth is 278 m and its maximum depth is 702 m. The lake water volume is 1,735 km³. The area of modern glaciers originating from the Issyk Lake watershed is approximately 650.4 km², and their storage is 48 km³. The lake has a temperate continental climate. Mean temperature in January is −6 °C, and the mean temperature in July varies from 15 to 25 °C. Sources of water supply for the lake are runoff, precipitation and underground water (Yang & Shao 1993).

A kind of long-lived spruce (Picea schrenkiana) was selected as our study target. These spruce trees often grow to 40 m in height and frequently live more than 200 years. More than 90% of the forest from 1,200 to 2,600 m a.s.l. in the Tien Shan Mountains comprise Picea schrenkiana, and it is the dominant tree species in the forest of the mountain zone. Forest stands are moderately open and canopy densities are low. Four tree-ring sites were sampled in 2012 and 2014. We tended to select healthy pine trees with little evidence of bushfire, landslide or human/animal disturbance, to avoid sampling of non-climatic effects on radial growth. For most sample trees, we extracted two cores from different directions. A total of 142 cores from 79 trees were collected from sites Tubeshu (coded as TBX), Kezersun (KZE), Kezersun (KZM), and Calaker (CAK) with increment borers. Additional tree-ring width data from five sites around Issyk Lake were obtained from the National Centers for Environmental Information, NOAA (National Oceanic and Atmospheric Administration) (http://www.ncdc.noaa.gov/), i.e., a total of 108 cores from 60 trees at sites Sarekungey (RUSS150 and RUSS151), Sarejmek (RUSS152), and Karabatkak (RUSS156) by Fritz Schweingruber, and KAZ001 (Ulken) by Olga Solomina. Information regarding sampling sites is provided in Table 1, such as location, elevation, and maximum tree age. The oldest tree at the KAZ001 site was nearly 432 years old.

According to normal dendrochronological techniques (Speer 2010), the sampled tree-ring cores were dried naturally and mounted on a wooden plank with grooves. Then, each core was sanded with abrasive papers and marked with needles under a microscope. Every ring on the sanded cores was measured by a Velmex Measuring System at resolution 0.001 mm. All tree-ring width data from the four newly sampled (TBX, KZE, KZM, and CAK) and five other (RUSS150, RUSS151, RUSS152, RUSS156, and KAZ001) sites were combined to develop a regional chronology (coded as RCC) in the further analyses. Cross-dating quality control in the COFECHA program was used after obtaining each tree-ring width (Grissino-Mayer 2001). This program was written in ANSI standard Fortran-77 by Holmes (1983) at the Laboratory of Tree-Ring Research of the University of Arizona, Tucson, Arizona, USA. It identifies portions of tree-ring series that may have major dating errors in measurement. Thus, each individual tree-ring was accurately associated with a calendar year based on comparison of ring width. Development of the tree-ring chronology was via the ARSTAN program (Cook & Krusic 2005). The concept and method of the ARSTAN program were developed by Cook (1985) at the Tree-Ring Laboratory, Lamont-Doherty Earth Observatory of Columbia University in the Palisades, New York, USA. Undesirable growth trends related to age and stand dynamics but unrelated to climatic variations were removed from each tree-ring width series in this program. A negative exponential function was executed for each tree-ring width series detrending to conserve as much common signal in the low-frequency domain as possible. Then, all individual detrended ring-width series were combined into a single chronology by computing a bi-weight robust mean. Eventually, standardized, residual and autoregressive standardized tree-ring chronologies were obtained. Reliability of the tree-ring chronology was evaluated by an expressed population signal (EPS) and mean series inter-correlation (Rbar) (Wigley et al. 1984). 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. Rbar is a measure of common variance between single series, independent of the number of measurement series (Cook & Kairiukstis 1990). The values of EPS and Rbar are listed in the output of the ARSTAN program. Generally, 50-year intervals with an overlap of 25 years for statistical analysis are used. Considering a smaller interval and overlap for short tree-ring chronologies may help to confirm relatively precise and reliable chronology lengths. Thus, the statistical analyses were done in 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 (Esper et al. 2003). Thus, the reliable length of chronology RCC was 345 (1670–2014) years. The RCC chronology with its sample depths, EPS, and Rbar is shown in Figure 2, and general statistics of this chronology for a common period (1890 to 1990) are listed in Table 2.

For further analyses, we selected monthly precipitation and mean temperature of the gridded 2.5° × 2.5° Climatic Research Unit Time-Series (CRU TS) 4.00 dataset (40°–45°N, 75°–80°E; 1935–2014; Mitchell & Jones 2005) to describe the climatic conditions around Issyk Lake. The above climatic data were acquired from the Royal Netherlands Meteorological Institute (KNMI) Climate Explorer (http://climexp.knmi.nl). The hydrologic data for Issyk Lake were from a published paper (1935–2000; Wang et al. 2006).

Linear testing was used to evaluate trends of hydrometeorological data. Pearson correlation was used to explore strengths of hydrologic and climatic signals inherent in the tree-ring width chronology from pine trees in the study area. The significance level of correlation coefficients was evaluated by the two-tailed test. After confirming the strongest relationship between tree-ring width and observed data, a linear regression model was used for reconstruction. We used Bootstrap (Young 1994), Leave-one-out cross-validation (Michaelsen 1987) and split-sample calibration-verification test (Meko & Graybill 1995) methods to evaluate statistical reliability of the reconstruction model. During the split-sample calibration-verification tests, the period of climatic data was split into two parts, for calibration and verification. Several statistics including a coefficient of efficiency, product mean test, and sign test were calculated to evaluate the observed and estimated data (Cook et al. 1999). Power spectrum analysis was used to examine reasonable periodicities in our reconstruction (Fritts 1976). This analysis was performed over the full range of the reconstruction. Wavelet analysis was used with a Morlet wavelet to investigate the periodicity of the reconstructed series and to examine how this periodicity changed over time (http://climexp.knmi.nl). We evaluated spatial correlation between our reconstruction and the gridded 0.5° × 0.5° CRU self-calibrating Palmer Drought Severity Index (PDSI) 3.21 dataset (Wells et al. 2004), using the KNMI Climate Explorer. The Climate Explorer is a web-based collection of climate data analysis tools maintained by the KNMI to assess the regional significance of reconstructed climatic series. A 13-year reciprocal filter was used to decompose the newly developed runoff series and other tree ring-based hydroclimatic reconstructions into high- and low-frequency domains (Yuan et al. 2013). Relationships between our reconstruction and others in the original, high- and low-frequency domains were assessed by correlation analysis.

Figure 3 shows that the highest temperature periods in the study area were in summer (June–August), with peaks in July (17.2 °C). A major proportion of the total annual precipitation fell during April–June (120.3 mm). There were two precipitation peaks in May (45.5 mm) and October (22.9 mm), respectively constituting 15.4% and 7.8% of the total annual precipitation. Climate data recorded since 1935 show a significant increasing trend of annual mean temperature in the study area (Y= 0.0263X − 48.71, R² = 0.476, p < 0.001) and a non-significant increasing trend of annual precipitation (Y= 0.5037X − 699.24, R² = 0.063, p < 0.1).

Figure 3 also shows that annual runoff peaked in 1942 (732.3 mm), and the smallest value was in 1972 (486.3 mm). The difference between the maximum and minimum is 246.0 mm. Annual runoff had a significant rising trend in the period 1940–1999 (Y= 0.9398X − 1246.5, R² = 0.111, p < 0.01), at a rate of 9.6 mm/10a. Correlation coefficients of annual runoff between annual total precipitation and mean temperature are 0.421 (p < 0.001, n = 66) and 0.251 (p < 0.05, n = 66), respectively.

Monthly and annual precipitation, temperature, and runoff of the previous July–December and current January–September during 1935–2000 were selected to evaluate how climatic and hydrologic elements influenced the radial growth of spruce trees around Issyk Lake.

Correlation coefficients of tree-ring chronology and hydrometeorological data are listed in Table 3. Figure 4 indicates that the relationship between ring width and precipitation was generally positive, and significant correlation coefficients were found for July (r= 0.31), August (r= 0.34), November (r= 0.26) and December (r= 0.27) of the previous year and current March (r= 0.25). The response of radial growth of spruce trees to temperature was negative, and the RCC chronology was negatively correlated with mean temperature of the previous July (r= − 0.37) and current April (r= − 0.40), at the 95% confidence level. After testing annual precipitation, temperature and runoff, their respective correlation coefficients were 0.20, 0.07 and 0.54. The RCC chronology and annual runoff had a strong relationship. Generally, precipitation was the main climatic limitation on the development of tree rings in the arid and semi-arid regions, but temperature always affects the radial growth of trees by modulating the amount of soil moisture (Zhang et al. 2014). More precipitation may enhance the potential for accumulating water reserves in the soil, resulting in wide rings, whereas higher temperature may increase evapotranspiration rates and moisture stress, leading to narrow rings (Zhang et al. 2015a).

The significant correlation coefficient between annual runoff and annual total precipitation (r= 0.421, p < 0.001, n = 66) reveals coherence among their variations. More rainfall may cause not only more runoff for a given lake, but also wider ring widths of sampled trees. The effect of runoff on tree growth is similar to that found for precipitation in the study area. Abundant rainfall is beneficial to the radial growth of sampled trees (Bao et al. 2012; Sun et al. 2013). Furthermore, similar relationships between ring width and hydrometeorological elements in previous studies are considered to improve the reasonability of tree growth-hydrologic response. Positive correlations between tree-ring width chronologies of Picea schrenkiana and water-year streamflow for the Manasi and Akesu rivers, originating from the Tien Shan Mountains were demonstrated by Yuan et al. (2007) and Zhang et al. (2016). Chen et al. (2016a) indicated that the influence of precipitation on tree-ring growth of Turkestan juniper was positive, while the influence of temperature was negative. Annual streamflow for the Upper Kurshab River in the Pamir-Alai Mountains of Kyrgyzstan was reconstructed using tree-ring chronology, owing to their strong correlation. Furthermore, similar influences of streamflow or runoff on tree-ring growth have been widely found in northern China, such as Sabina przewalskii Kom. in the Heihe River watershed (Liu et al. 2010; Yang et al. 2011), Juniperus przewalski in the Yellow River watershed (Gou et al. 2007), and Pinus sylvestris var. mongolica in the Yimin River watershed (Bao et al. 2012).

The results of Bootstrap (100 iterations in the recomputation) and Leave-one-out tests show that all values of r, R², R²_adj, standard error (SE), F-value (F) and Durbin-Watson (D/W) were very similar to those of the original regression model (1) (Table 4). Table 5 shows statistics resulting from the split-sample calibration/verification tests for the reconstructed annual runoff series. The values of coefficient of efficiency (CE) exceed or equal 0. Values of the product mean test (t) are positive, indicating significant accuracy in the tree-ring estimates. Values of the low-frequency (S₁) and high-frequency sign test (S₂), which describe how well the predicted value tracks the direction of the actual data, are 28 + /8– (p < 0.01) and 27 + /8– (p < 0.01) for the period 1965–2000, and 26 + /10– (p < 0.05) and 23 + /12– (p < 0.10) for 1935–1970, respectively. These test results, which indicate significant skill in the tree-ring estimates (Fritts 1976), all demonstrate that model (1) is stable and reliable and that it may be used for the annual runoff reconstruction. Consequently, the annual runoff for Issyk Lake during 1670–2014 was reconstructed based on model (1), yielding a mean value of 599.87 mm and standard deviation of σ= 28.24 mm (Figure 5(c)).

For the runoff reconstruction, we defined a wet year using values above the mean + σ (628.11 mm), and a dry year as those below the mean – σ (599.87 mm), according to the method of division used by Liu et al. (2013). The newly reconstructed runoff series revealed that 59 years could be categorized as wet (accounting for 17.1% of the total), and 54 years as dry (15.7% of the total). The remaining 232 years were categorized as normal (67.2% of the total). The extreme values of years and decades are listed in Table 6. It is seen that the difference between the wettest (1994) and driest (1917) years is 189.7 mm, and that between the wettest and driest decades (1950s and 1910s, respectively) is 131.7 mm.

Table 6 also shows the long-term means and variation coefficients (Bai et al. 2014) of the annual runoff reconstruction. The results demonstrate that the water resource for Issyk Lake was comparatively abundant in the 18th century, but an obvious decrease of runoff appeared in the 19th century. Thereafter, runoff began to increase in the 20th century. The variation coefficient of that century was slightly larger than other centuries. The number of anomalies throughout the 20th century, including the three wettest and three driest years, and four wettest and three driest decades (Table 6), certifies the greater hydrologic fluctuation in the century.

Decadal variability was highlighted using a 21-year moving average in the reconstruction (Figure 5(c)), from which seven wet and six dry periods can be distinguished. The wet periods (above the mean value of the reconstruction) are 1680–1693 (average 601.9 mm), 1717–1760 (603.4 mm), 1782–1789 (602.1 mm), 1828–1836 (601.1 mm), 1872–1903 (602.4 mm), 1926–1964 (603.8 mm), and 1983–2004 (605.5 mm). The dry periods (below the mean value) are 1694–1716 (598.3 mm), 1761–1781 (596.9 mm), 1790–1827 (595.4 mm), 1837–1871 (596.6 mm), 1904–1925 (595.7 mm), and 1965–1982 (596.1 mm). Comparison between seven mean values of wet periods reveals that under the background of the recent warming-wetting trend, there was more annual runoff for Issyk Lake in the recent 350 years during 1983–2004.

The results of power spectral analysis over the entire reconstruction period (1670–2014) further indicated significant periodicity at frequencies of 9.6 (90%), 5.4 (95%), 2.5 (90%), and 2.1 (90%) years (Figure 6(a)). Temporal characteristics of the different cycles were also evaluated using wavelet analysis and are illustrated in Figure 6(b). The significant cycles in our reconstruction detected by power spectral and wavelet analyses are in relative agreement. The wavelet analysis indicates a robust approximate 10-year cycle from 1780 to 1820 and 1850 to 1870. A significant (although less robust) power with approximate 5-year cycle was detected in the periods 1680–1690, 1730–1740, 1910–1920, and 1990–2000. Solar activity has alternated between active and quiet phases with a period of more or less 11 years (Nagovitsyn 1997). The 10-year cycle in our reconstructed runoff series suggests a possibly strong connection with the 11-year Schwabe cycle of solar activity. In addition, the 5.4-, 2.5-, and 2.1-year cycles in the runoff reconstruction fall within the range of the El Niño-Southern Oscillation (ENSO) (Allan et al. 1996). Chen et al. (2016a) found a 11.5-year cycle in a tree ring width-based streamflow reconstruction of the upper Kurshab River in the Pamir-Alai Mountains of Kyrgyzstan. These interannual and interdecadal cycles have also been widely inferred from other tree ring-based hydrometeorological reconstructions for the western (Zhang et al. 2015a, 2016), central (Li et al. 2006; Zhang et al. 2013), and eastern (Wang et al. 2007; Zhang et al. 2015b) Tien Shan Mountains.

Dai et al. (2004) showed that basin-average annual PDSI covaried closely with runoff for seven of the world's largest rivers and several smaller rivers. Thus, the PDSI is regarded as a good proxy of both surface moisture conditions and runoff. We used spatial correlation to evaluate the regional significance of the tree ring width-based annual runoff reconstruction. The runoff reconstruction correlates (>0.3) with gridded August–July PDSI data for 1935–2014 in a large area between approximately 40°–48°N and 70°–90°E, with the strongest correlations (>0.4) in the western Tien Shan Mountains and Junggar Basin (Figure 7).

Three hydroclimatic reconstructions based on tree-ring data, i.e., annual streamflow for the Upper Kurshab River (code SKR, 1720–2013; Chen et al. 2016a), October–September streamflow for the Aksu River (SAR, 1692–2005; Zhang et al. 2016), and August–January standardized precipitation evapotranspiration index for the southeastern Kazakhstan (SPEI, 1785–2014; Chen et al. 2016b), were compared with the newly reconstructed annual runoff series of Issyk Lake (RIL). The locations of these reconstructions are marked in Figure 7. These reconstructions were standardized and decomposed by 13-year reciprocal filters to assess their coherence in a larger spatial context. Table 7 shows that although the correlations between RIL and SKR in the original, high-frequency, and low-frequency domains exceed the 0.05 significance level in the common period 1785–2005, their correlation coefficients are relatively small. The correlations between RIL and SAR in the original (r= 0.358, p < 0.05) and high-frequency (r= 0.530, p < 0.05) domains are relatively strong, whereas the correlation in the low-frequency domain is not significant (r= 0.109, p= 0.116). The strongest correlations are between RIL and SPEI, and the three correlation coefficients from the original, high-frequency, and low-frequency domains all exceed 0.65.

Zhang et al. (2015b) showed that the correlations between tree ring-based precipitation reconstructions from the eastern to western Tien Shan Mountains weakened with increasing distance in the low-frequency domain. Table 7 reveals that the relatively weak correlations in the low-frequency domain coincide with the above finding and confirm the local variations of precipitation in the Tien Shan Mountains. The relatively strong correlations in the high-frequency domain indicate coherence among the extreme values of these reconstructions. The 10 wettest and 10 driest years of our reconstruction (Table 6) were compared with historical documents (Wen et al. 2006) and three hydroclimatic reconstructions (SKR, SAR, and SPEI). The results show good agreement between these years and the related sources (Table 8). These records allow four of the 10 wettest years (1804, 1924, 1994, and 1999) and three of the 10 driest years (1917, 1961, and 1997) in the RIL to be compared with the historical records. Furthermore, the two wettest years (1804 and 1886) and the three driest years (1808, 1829, and 1917) were found in the SKR, SAR, and SPEI. Thus, our reconstructed hydrologic series captured the signals of flood or drought disasters in the western Tien Shan Mountains well.

A 345-year long regional tree-ring width chronology was developed using 250 cores from 139 spruce trees in the mountainous area of Issyk Lake. Because there was a large correlation coefficient between hydrologic data and the radial growth of sampled trees, it was possible to develop an annual runoff reconstruction for Issyk Lake using this regional chronology. The characteristics of annual runoff reconstruction showed that the 20th century had the three wettest and three driest years, and the four wettest and three driest decades. It also had the largest hydrologic fluctuations for the lake in the last three centuries. After applying a 21-year moving average, seven wet periods were identified, i.e., 1680–1693, 1717–1760, 1782–1789, 1828–1836, 1872–1903, 1926–1964, and 1983–2004. Six periods, 1694–1716, 1761–1781, 1790–1827, 1837–1871, 1904–1925, and 1965–1982 were relatively dry in the low-frequency domain. There were frequent 10-, and 2.1–5.4-year cycles in the reconstructed series in some tree-ring based hydrometeorological reconstructions from the western to eastern Tien Shan Mountains, suggesting that annual runoff variability of Issyk Lake may be influenced by solar activity and the atmosphere–ocean system. Spatial correlations between the annual runoff reconstruction and a gridded August–July PDSI dataset revealed significant positive correlation in the western Tien Shan Mountains and Junggar Basin. Comparison between the annual runoff reconstruction and three hydroclimatic reconstructions based on tree-ring data of the Upper Kurshab River, Aksu River, and southeastern Kazakhstan revealed that the coherence of these reconstructions in the high-frequency domain is stronger than in the low-frequency domain. Thus, the newly reconstructed runoff series could accurately capture some flood (1804, 1886, 1924, 1994, and 1999) and drought events (1808, 1829, 1917, 1961, and 1997) noted in the meteorological records and other tree ring-based hydroclimatic reconstructions.

We thank Ali Mamtimin PhD and Zhao Yong PhD for their great help in the process of collecting samples. Particular thanks are extended to the anonymous reviewers and editors whose comments and suggestions greatly benefited this manuscript. This research was supported by Shanghai Cooperation Organisation Science and Technology Partnership (2017E01032), National Natural Science Foundation of China (41605047), Climate Change Special Project of China Meteorological Administration (CCSF201601), and Autonomous Region Youth Science and Technology Innovation Talents Training Project (qn2015bs025).