A 422-Year Reconstruction of the Kaiken River Streamﬂow, Xinjiang, Northwest China

: Our understanding of Central Asian historical streamﬂow variability is still limited because of short instrumental hydrologcial records. Based on tree-ring cores collected from three sampling sites in Kaiken River basin near Tien Shan, a regional tree-ring width chronology were developed. The correlation analysis showed that the runo ﬀ of Kaiken River from previous August to current June was signiﬁcantly correlated with the regional chronology, and the high correlation coe ﬃ cient was 0.661 ( p < 0.01). Based on the regional chronology, the August-June runo ﬀ of Kaiken River has been reconstructed over the past 422 year, and it accounted for 43.7% of actual runo ﬀ variance during the common period 1983–2013. The reconstruction model is reliable, and the trend of observed and reconstructed data is relatively consistent. The results of multi-taper spectral analysis for the runo ﬀ reconstruction indicated some remarkable cycles for the past 422 years; the 11.5-year cycles correspond to the solar cycle and is found widely in runo ﬀ reconstructions in Central Asia. This may imply a solar inﬂuence on the hydroclimate variations of Tien Shan. The runo ﬀ reconstruction of Kaiken River compares well with runo ﬀ reconstructions the Urumqi River and Manas River, and implies that there is a common driving factor for the runo ﬀ in central Tien Shan, China. The analysis of linkages between climate variation and the runo ﬀ reconstruction of Kaiken River shows that there is a relationship between extremes in runo ﬀ variation and abnormal atmospheric circulations. Our 422-year steamﬂow reconstruction provides long-term perspective on current and 20th century hydrological events in central Tien Shan, is useful for aids sustainable water management and addresses regional climate change challenges.


Introduction
Economic development, as well as urbanization and the increasing population, raises doubts over the supply of water resources to meet the growing needs of society in semiarid and arid areas around the world [1][2][3][4]. Understanding of temporal-spatial hydrological variations is thus vital to improve our knowledge about water resources in drylands [5][6][7][8]. However, instrumental hydrological records over Central Asia are short and sparse in spatial distribution, especially in Tien Shan, China. High-resolution hydrology proxy records, such as history records and tree rings, are needed to interpret the Central Asian hydrological changes. The Kaiken River basin is located in the northern slope of Tien Shan, Central Asia. It originates from the Bogda peak in central Tien Shan. It is the largest

Study Area Description
The Kaiken River is located at the southern margin of Zhungeer Basin in Qitai, Xinjiang, and originated from the Bogda peak in central Tien Shan, China. With a drainage area of 280 km 2 and a length of 64 km. The tree-ring width data that formed the base for the runoff reconstruction came from central Tien Shan which are situated between 43 • 30 to 44 • N, 88 • 30 to 90 • 50 E (Figure 1). The sampling sites information is shown in Table 1. The dominant tree species in central Tien Shan is Schrenk spruce (Picea schrenkiana). At sampling sites, spruce trees grew in open stands with thin and rocky soil in the understory of the forests.

Tree-Ring Chronology
All tree-ring samples were selected from isolated living trees with relatively homogeneous environment. Tree-ring samples were chosen from three different sites in central Tien Shan in August 2008 and July 2014. Two cores per tree were obtained from the relatively old spruce trees at breast height. Tree-ring sampling activities were mainly concentrated in the spruce forests to arrange 1500 to 2500 m a.l.s. Three different sampling sites are hardly affected by human activities and provided 144 cores from 72 trees in total. After dried, mounted on wooden holders, and polished with 400 grit sandpaper to enhance tree-ring bound, annual ring widths of all cores were measured by a manual Henson micrometer (LINTAB™, Rinntech, Heidelberg, Germany) with a 0.01 mm precision. The computer program COFECHA [25,26] was applied to check cross-dated results of tree-ring series. The regional tree-ring width chronology (STD) were developed by the software ARSTAN. Due to the similar growth environment and average high correlation with the master series (r = 0.59), detrended data from all cores were combined into a regional chronology after cross-dated again. Each individual ring-width series was detrended by fitting a negative exponential curve. More than the 0.85 threshold value of the express population signal (EPS) is considered acceptable of chronology quality [27].

Hydroclimate Data
Several observed data were used in this research, including runoff data from the Kaiken River (89°50′ E, 43°36′ N, 1520 m a.s.l.) and regional climate data (Qitai, 89°27′ E, 44°03′ N, 1061.2 m a.s.l.) ( Figure 1). The monthly total precipitation and mean temperature from 1983 to 2013 at the Qitai meteorological station were showed in Figure 2A. The annual total precipitation and mean temperature are 296.8 mm and 5.4 °C. July is the hottest and wettest month (mean temperature 22.8°C and total precipitation 36 mm) while January is the coldest month (mean temperature −17.4°C). The mean annual runoff of the Kaiken River is 5.1 m 3 /s during the period 1983-2013. Figure 2B shows monthly mean runoff of the Kaiken River from 1983 to 2013. The seasonal distributions of runoff and precipitation and runoff is very different ( Figure 2B). There is only one peak (in July) in the monthly precipitation distribution, while there are one peak with high runoff months (May-July). The runoff peak is directly linked to the meltwater input from snowpack of Tien Shan during the warm season.

Statistical Analysis
To understand historical runoff changes, correlation analysis was applied to identify the response of tree-ring widths to hydroclimate. Monthly total precipitation, monthly mean runoff and monthly mean temperature were used for correlation analysis, over a span of 14 months from prior August to current September. A linear regression model was computed between the predictors (treering width series) and the predict runoff for the calibration period 1983-2013. Because the actual runoff record is relatively short, the 'leave-one-out' method was used to assess the reliability of the mode [28]. Verification statistical parameters used included the reduction of error (RE), the sign test (ST), sign test of the first difference (FST) and the product means test (PMT) [29].

Hydroclimate Data
Several observed data were used in this research, including runoff data from the Kaiken River (89 • 50 E, 43 • 36 N, 1520 m a.s.l.) and regional climate data (Qitai, 89 • 27 E, 44 • 03 N, 1061.2 m a.s.l.) ( Figure 1). The monthly total precipitation and mean temperature from 1983 to 2013 at the Qitai meteorological station were showed in Figure 2A. The annual total precipitation and mean temperature are 296.8 mm and 5.4 • C. July is the hottest and wettest month (mean temperature 22.8 • C and total precipitation 36 mm) while January is the coldest month (mean temperature −17.4 • C). The mean annual runoff of the Kaiken River is 5.1 m 3 /s during the period 1983-2013. Figure 2B shows monthly mean runoff of the Kaiken River from 1983 to 2013. The seasonal distributions of runoff and precipitation and runoff is very different ( Figure 2B). There is only one peak (in July) in the monthly precipitation distribution, while there are one peak with high runoff months (May-July). The runoff peak is directly linked to the meltwater input from snowpack of Tien Shan during the warm season.

Statistical Analysis
To understand historical runoff changes, correlation analysis was applied to identify the response of tree-ring widths to hydroclimate. Monthly total precipitation, monthly mean runoff and monthly mean temperature were used for correlation analysis, over a span of 14 months from prior August to current September. A linear regression model was computed between the predictors (tree-ring width series) and the predict runoff for the calibration period 1983-2013. Because the actual runoff record is relatively short, the 'leave-one-out' method was used to assess the reliability of the mode [28]. Verification statistical parameters used included the reduction of error (RE), the sign test (ST), sign test of the first difference (FST) and the product means test (PMT) [29].  Figure 3 shows the correlations between tree-ring width series and monthly total precipitation from prior August to current September, and indicates significant positive correlations (p < 0.05) in prior September and current April. Based on various multi-month combination, the highest correlation (r = 0.53, p < 0.01) was found between tree-ring width series and total August-June precipitation. Significant positive correlations (p < 0.05) between monthly runoff and tree-ring width series were occur at prior September, October and December, and current January, May and June. The highest correlation between tree-ring width series and seasonal runoff combinations were 0.661 (p < 0.01) for prior August to current June. Correlation between August-June runoff and precipitation was 0.55 (p < 0.01) for 1983 to 2013. Thus, precipitation has significantly influences on the growth of spruce tree and the runoff variations in the Kaiken River basin. In addition, correlation coefficients between tree-ring width series and monthly mean temperature are relatively weak, and only June temperature has significantly influence on tree growth.   Figure 3 shows the correlations between tree-ring width series and monthly total precipitation from prior August to current September, and indicates significant positive correlations (p < 0.05) in prior September and current April. Based on various multi-month combination, the highest correlation (r = 0.53, p < 0.01) was found between tree-ring width series and total August-June precipitation. Significant positive correlations (p < 0.05) between monthly runoff and tree-ring width series were occur at prior September, October and December, and current January, May and June. The highest correlation between tree-ring width series and seasonal runoff combinations were 0.661 (p < 0.01) for prior August to current June. Correlation between August-June runoff and precipitation was 0.55 (p < 0.01) for 1983 to 2013. Thus, precipitation has significantly influences on the growth of spruce tree and the runoff variations in the Kaiken River basin. In addition, correlation coefficients between tree-ring width series and monthly mean temperature are relatively weak, and only June temperature has significantly influence on tree growth.  Figure 3 shows the correlations between tree-ring width series and monthly total precipitation from prior August to current September, and indicates significant positive correlations (p < 0.05) in prior September and current April. Based on various multi-month combination, the highest correlation (r = 0.53, p < 0.01) was found between tree-ring width series and total August-June precipitation. Significant positive correlations (p < 0.05) between monthly runoff and tree-ring width series were occur at prior September, October and December, and current January, May and June. The highest correlation between tree-ring width series and seasonal runoff combinations were 0.661 (p < 0.01) for prior August to current June. Correlation between August-June runoff and precipitation was 0.55 (p < 0.01) for 1983 to 2013. Thus, precipitation has significantly influences on the growth of spruce tree and the runoff variations in the Kaiken River basin. In addition, correlation coefficients between tree-ring width series and monthly mean temperature are relatively weak, and only June temperature has significantly influence on tree growth.  Tree-ring width series and runoff correlate to variations of August-June precipitation and June temperature, revealing that corresponding signals from climate factors were contained within tree-ring records and runoff. The strong link between runoff and tree-ring width series supports this conclusion. Recognizing this interlinkage, it was assumed that the physiological linkage between climate factors and tree-ring growth provides some useful explanations for revealing the linkage between the climate-integrated variations of tree-ring growth and the runoff [30]. Our research has showed the physiological linkage between tree-ring growth and fluctuations in August-June precipitation and June temperatures. Seasonal distribution of runoff (with May-July peak) and the significant positive correlation detected between August-June precipitation and tree-ring growth/runoff in this study likely reflect high precipitation, especially for the winter snowpack synchronized with above mean runoff years. Good moisture conditions at the end of the previous growth season and heavy snow in winter can provide enough water for the next year's tree growth, and more water for high mountains and spruce forests is converted into runoff [31][32][33][34]. Thus, high August-June precipitation is indicative of both increased runoff and luxuriant spruce tree growth. High June temperatures enhance evapotranspiration and decrease soil moisture content, and lead to reduced earlywood growth and runoff [35,36].

Runoff Reconstruction of the Kaiken River
Based on the above correlation analysis results, a linear regression model was developed to indicate the August-June runoff variations of the Kaiken River. The transfer model was designed as follows: The dependent variable Y is August-June runoff of the Kaiken River, and the independent variable X is the regional tree-ring chronology. During the calibration period 1983-2013, the regional tree-ring chronology accounts for 43.7% of total variance of the actual runoff data (41.7% after adjustment for loss of degrees of freedom). Figure 4 shows a comparison of actual and reconstructed August-June runoff of the Kaiken River, and reveals that the reconstructed runoff is consistent with the actual runoff during the calibration period 1983-2013. The results of reduction of error (RE) the sign test (ST), sign test of the first difference (FST) and product mean test (PMT) are showed in Table 2.
Atmosphere 2020, 11, x FOR PEER REVIEW 5 of 11 Tree-ring width series and runoff correlate to variations of August-June precipitation and June temperature, revealing that corresponding signals from climate factors were contained within treering records and runoff. The strong link between runoff and tree-ring width series supports this conclusion. Recognizing this interlinkage, it was assumed that the physiological linkage between climate factors and tree-ring growth provides some useful explanations for revealing the linkage between the climate-integrated variations of tree-ring growth and the runoff [30]. Our research has showed the physiological linkage between tree-ring growth and fluctuations in August-June precipitation and June temperatures. Seasonal distribution of runoff (with May-July peak) and the significant positive correlation detected between August-June precipitation and tree-ring growth/runoff in this study likely reflect high precipitation, especially for the winter snowpack synchronized with above mean runoff years. Good moisture conditions at the end of the previous growth season and heavy snow in winter can provide enough water for the next year's tree growth, and more water for high mountains and spruce forests is converted into runoff [31][32][33][34]. Thus, high August-June precipitation is indicative of both increased runoff and luxuriant spruce tree growth. High June temperatures enhance evapotranspiration and decrease soil moisture content, and lead to reduced earlywood growth and runoff [35,36].

Runoff Reconstruction of the Kaiken River
Based on the above correlation analysis results, a linear regression model was developed to indicate the August-June runoff variations of the Kaiken River. The transfer model was designed as follows: The dependent variable Y is August-June runoff of the Kaiken River, and the independent variable X is the regional tree-ring chronology. During the calibration period 1983-2013, the regional tree-ring chronology accounts for 43.7% of total variance of the actual runoff data (41.7% after adjustment for loss of degrees of freedom). Figure 4 shows a comparison of actual and reconstructed August-June runoff of the Kaiken River, and reveals that the reconstructed runoff is consistent with the actual runoff during the calibration period 1983-2013. The results of reduction of error (RE) the sign test (ST), sign test of the first difference (FST) and product mean test (PMT) are showed in Table  2. All evaluative statistics exceed the 0.01 confidence level, and demonstrate that the credibility of regression model. We further compared the runoff reconstruction with drought reconstruction of All evaluative statistics exceed the 0.01 confidence level, and demonstrate that the credibility of regression model. We further compared the runoff reconstruction with drought reconstruction of eastern Tien Shan [8], and high correlation (r = 0.76, p < 0.01) as found between the series during the  Atmosphere 2020, 11, x FOR PEER REVIEW 6 of 11 eastern Tien Shan [8], and high correlation (r = 0.76, p < 0.01) as found between the series during the common period 1725-2013. Therefore, based on this model and the 0.85 EPS threshold value, the runoff variations of the Kaiken River have been reconstructed for the period 1592-2013 CE ( Figure 5).

The Characteristics of the Runoff Reconstruction
The August-June runoff reconstruction of the Kaiken River spanned 422 years; the average runoff  Figure 5B).
Several tree-ring based runoff reconstructions for surrounding river basin have recently been developed. Yuan et al. [16] developed an October-September runoff reconstruction since 1629 CE for the Manasi River, capturing 51% of the total variance in the calibration period . Yuan et al. [28] developed 15 Schrenk spruce tree-ring width chronologies and reconstructed annual (January-December) runoff of the Urumqi River for the period 1633-1989. The calibration model explained 69.4% of the observed runoff variances for the period from 1958 to 1989. Correlations between the Kaiken reconstruction and other runoff reconstructions by Yuan et al. [16,37], computed over the common periods, are 0.25 and 0.33, respectively, and increase to 0.38 and 0.43 after 21-year low-pass filtering. Our runoff reconstruction reflects similar high/low runoff intervals in the neighboring areas as runoff reconstructions ( Figure 6). As showed by first principal component (PC1), regional low runoff conditions during the 1641-1644, 1704-1727, 1753-1777, 1804-1832, 1853-1869 and 1905-1932 found in this study occurred synchronously in the Urumqi and Manasi River basin. The common runoff signals found in the three runoff reconstructions suggest that our reconstruction  Figure 5B).
Several tree-ring based runoff reconstructions for surrounding river basin have recently been developed. Yuan et al. [16] developed an October-September runoff reconstruction since 1629 CE for the Manasi River, capturing 51% of the total variance in the calibration period . Yuan et al. [28] developed 15 Schrenk spruce tree-ring width chronologies and reconstructed annual (January-December) runoff of the Urumqi River for the period 1633-1989. The calibration model explained 69.4% of the observed runoff variances for the period from 1958 to 1989. Correlations between the Kaiken reconstruction and other runoff reconstructions by Yuan et al. [16,37], computed over the common periods, are 0.25 and 0.33, respectively, and increase to 0.38 and 0.43 after 21-year low-pass filtering. Our runoff reconstruction reflects similar high/low runoff intervals in the neighboring areas as runoff reconstructions ( Figure 6). As showed by first principal component (PC1), regional low runoff conditions during the 1641-1644, 1704-1727, 1753-1777, 1804-1832, 1853-1869 and 1905-1932 found in this study occurred synchronously in the Urumqi and Manasi River basin. The common runoff signals found in the three runoff reconstructions suggest that our reconstruction represents broad-scale regional runoff variations. The low runoff period 1641-1644 and 1753-1777 is also two well-documented historical droughts the Ming Dynasty drought [38] and the Strange Parallels drought [39], and demonstrated that large-scale drought extreme events in Central and eastern Asia can effectively affect the runoff variation of arid areas in Asian inland. Many studies also reported that the 1917-1919 drought was most severe [40][41][42]. Of particular interest is the downward trend during the 1950s-2010s. Although the regional climate is still warm and wet [43], evaporation increases with the rise in temperatures lead to the decline of runoff, especially in the eastern Xinjiang and Mongolia [18,44,45].
Multi-taper method spectral analysis [46] was employed to reveal the characteristics of the runoff variation of the Kaiken in the frequency domain. The analysis over the range of our runoff reconstruction indicated low-and high-frequency periodicities (Figure 7). The low-frequency (60.2 years) and high-frequency peaks (5.6, 3.2, and 2.2 years) exceed the 0.01 confidence level based on a red noise null continuum. Other significant high-frequency periodicities were found at 11.5 year, 9.5 year, 3.7 year, and 2.0 year (95%) (Figure 7).
Atmosphere 2020, 11, x FOR PEER REVIEW 7 of 11 represents broad-scale regional runoff variations. The low runoff period 1641-1644 and 1753-1777 is also two well-documented historical droughts the Ming Dynasty drought [38] and the Strange Parallels drought [39], and demonstrated that large-scale drought extreme events in Central and eastern Asia can effectively affect the runoff variation of arid areas in Asian inland. Many studies also reported that the 1917-1919 drought was most severe [40][41][42]. Of particular interest is the downward trend during the 1950s-2010s. Although the regional climate is still warm and wet [43], evaporation increases with the rise in temperatures lead to the decline of runoff, especially in the eastern Xinjiang and Mongolia [18,44,45]. Multi-taper method spectral analysis [46] was employed to reveal the characteristics of the runoff variation of the Kaiken in the frequency domain. The analysis over the range of our runoff reconstruction indicated low-and high-frequency periodicities (Figure 7). The low-frequency (60.2 years) and high-frequency peaks (5.6, 3.2, and 2.2 years) exceed the 0.01 confidence level based on a red noise null continuum. Other significant high-frequency periodicities were found at 11.5 year, 9.5 year, 3.7 year, and 2.0 year (95%) (Figure 7). The August-June runoff reconstruction of the Kaiken River presented in this article are characterized by cyclical fluctuations of above and below mean runoff. The 11.5-year cycle resembles other findings in neighboring areas and implies the impact of solar activity on Central Asian runoff there [17,18,21]. The appearance of these low and high frequency cycles (60.2, 5.6, 3.7, 3.2, 2.2, and 2.0 years) suggests that the oscillations of the ocean-atmosphere system have impacted on the hydrologic cycle of the Kaiken River basin for the past 422 years, such as North Atlantic Oscillation (NAO) and Arctic Oscillation (AO) [44,47,48]. Over the historical period, runoff reconstruction for the Kaiken River show significant correlation with the January-June NAO index (r = −0.23, p < 0.01) and the January-February AO index (r = −0.21, p < 0.01) [49] during the common period 1850-2013. Some studies have revealed that the positive NAO and AO phase could lead to the precipitation decrease in Central Asia, include Xinjiang, owing to decrease in eastward water-vapor transport from Europe to Central Asia, which causes an increase in column water vapor content in Central Asia, and vice versa [48][49][50]. The August-June runoff reconstruction of the Kaiken River presented in this article are characterized by cyclical fluctuations of above and below mean runoff. The 11.5-year cycle resembles other findings in neighboring areas and implies the impact of solar activity on Central Asian runoff there [17,18,21]. The appearance of these low and high frequency cycles (60.2, 5.6, 3.7, 3.2, 2.2, and 2.0 years) suggests that the oscillations of the ocean-atmosphere system have impacted on the hydrologic cycle of the Kaiken River basin for the past 422 years, such as North Atlantic Oscillation (NAO) and Arctic Oscillation (AO) [44,47,48]. Over the historical period, runoff reconstruction for the Kaiken River show significant correlation with the January-June NAO index (r = −0.23, p < 0.01) and the January-February AO index (r = −0.21, p < 0.01) [49] during the common period 1850-2013. Some studies have revealed that the positive NAO and AO phase could lead to the precipitation decrease in Central Asia, include Xinjiang, owing to decrease in eastward water-vapor transport from Atmosphere 2020, 11, 1100 8 of 11 Europe to Central Asia, which causes an increase in column water vapor content in Central Asia, and vice versa [48][49][50]. The 500-hPa vector wind composite anomalies of the highest and lowest reconstructed runoff (n = 10) in the period 1948-2013 support the connection with AO and NAO (Figure 8). High reconstructed runoff year mean 500 hPa winds exhibit strong westerly and northwesterly flow over Central Asia (Figure 8A), indicating a strong phase of the westerlies. This is consistent with the enhanced westerly flow of cold air from the high latitude areas, which should lead to increased snowfall in Central Asia, and providing sufficient water tree growth and runoff formation in the following year. During the lowest runoff years, the opposite pattern occurs. As discussed above, runoff variation in our study area is related to the intensification of the mid-latitude westerly circulation.

Conclusions
The correlation analyses indicated that the tree-ring width of spruce trees have significant correlation with the runoff of the Kaiken River, and the highest correlation between the seasonalized runoff and tree rings was found in August-June. The results of the research reveal the feasibility of combining actual runoff data and tree-ring width series for runoff reconstruction for the Kaiken River. The runoff reconstruction has revealed the runoff variations of the Kaiken River during 1592-2013 CE. Comparison with runoff reconstructions for the Urimqi and Manasi rivers indicates high The 500-hPa vector wind composite anomalies of the highest and lowest reconstructed runoff (n = 10) in the period 1948-2013 support the connection with AO and NAO (Figure 8). High reconstructed runoff year mean 500 hPa winds exhibit strong westerly and northwesterly flow over Central Asia (Figure 8A), indicating a strong phase of the westerlies. This is consistent with the enhanced westerly flow of cold air from the high latitude areas, which should lead to increased snowfall in Central Asia, and providing sufficient water tree growth and runoff formation in the following year. During the lowest runoff years, the opposite pattern occurs. As discussed above, runoff variation in our study area is related to the intensification of the mid-latitude westerly circulation. The 500-hPa vector wind composite anomalies of the highest and lowest reconstructed runoff (n = 10) in the period 1948-2013 support the connection with AO and NAO (Figure 8). High reconstructed runoff year mean 500 hPa winds exhibit strong westerly and northwesterly flow over Central Asia (Figure 8A), indicating a strong phase of the westerlies. This is consistent with the enhanced westerly flow of cold air from the high latitude areas, which should lead to increased snowfall in Central Asia, and providing sufficient water tree growth and runoff formation in the following year. During the lowest runoff years, the opposite pattern occurs. As discussed above, runoff variation in our study area is related to the intensification of the mid-latitude westerly circulation.

Conclusions
The correlation analyses indicated that the tree-ring width of spruce trees have significant correlation with the runoff of the Kaiken River, and the highest correlation between the seasonalized runoff and tree rings was found in August-June. The results of the research reveal the feasibility of combining actual runoff data and tree-ring width series for runoff reconstruction for the Kaiken River. The runoff reconstruction has revealed the runoff variations of the Kaiken River during 1592-2013 CE. Comparison with runoff reconstructions for the Urimqi and Manasi rivers indicates high

Conclusions
The correlation analyses indicated that the tree-ring width of spruce trees have significant correlation with the runoff of the Kaiken River, and the highest correlation between the seasonalized