Moisture variations during the first millennium CE and their linkage with social developments along the Silk Road in northwestern China

Moisture conditions, especially those that occur as multi-decadal anomalies, have profound impacts on society, especially in arid and semi-arid regions. However, the lack of high-resolution climate data for the first millennium CE greatly limits our understanding of how moisture variations have influenced history. Here, we present an 1882-year (134–2015 CE) tree-ring chronology developed from Qilian juniper (Juniperus przewalskii Kom.) growing in the western Qilian Mountains, northwest China. The tree-ring index correlates significantly with the May–June self-calibrating Palmer Drought Severity Index (sc-PDSI) and can be used to reconstruct May–June moisture variations since 241 CE. The reconstruction reflects moisture conditions at the annual to multi-decadal time scales over the past two millennia. During the period from the third to eighth centuries, there were prominent interdecadal fluctuations, with the third century and the late fifth century being the wettest and driest periods during the reconstruction, respectively. The transition from the wet third century to the dry fifth century corresponded with key events in Chinese history, namely the demise of the Western Jin Dynasty and the chaotic Southern and Northern Dynasties. Thus, our reconstruction provides new evidence for the potential linkage between abnormal climate conditions and social changes in ancient times.


Introduction
Moisture conditions, especially multi-decadal anomalies, have significant impacts on human societies. For example, the persistent wet conditions in the Mongol Empire during the thirteenth century (Pederson et al. 2014), the pronounced drought and pluvial spells in central Europe over the past 2110 years (Büntgen et al. 2021), and the megadrought during the late Ming Dynasty in China (Zhang et al. 2008;Zhao et al. 2021) all had profound impacts on society. Understanding the characteristics of moisture variations at the multi-decadal time scale is very important for assessing the impacts of climatic change on economic and societal development (Meehl et al. 2009). Because instrumental records are limited to the last century, it is necessary to use proxy records such as tree rings to study multi-decadal events prior to the twentieth century. Fortunately, tree rings provide widely distributed, relatively accessible, and annually resolved data. For this reason, they have been used to reconstruct interannual-decadal climate variations during the past 2000 years in many parts of the world (Esper et al. 2012;Jones and Mann 2004;Mann and Jones 2003). However, because the availability of tree-ring-based reconstructions decreases with time, there is a lack of reliable reconstructions during the first millennium CE, especially prior to approximately 800 CE (Breitenmoser et al. 2012;Esper and Büntgen 2021;Ljungqvist et al. 2016).
The period prior to 800 CE is extremely important because several known climate anomalies had significant impacts on human history. These include the Roman Warm Period (first century-third century) and the Late Antique Little Ice Age (sixth century-seventh century) in Europe (Büntgen et al. 2011;Büntgen et al. 2016), as well as the warm periods during the Sui and Tang Dynasties (541-810 CE) and the cold periods during the Wei, Jin, and North-South Dynasties (181-540 CE) in ancient China (Ge et al. 2013). Additionally, the demise of the western Roman Empire (265-316 CE), the turmoil of the Migration Period in Europe (fourth-sixth centuries), and frequent dynasty successions with severe social/political upheavals in China (304-581 CE) all happened in the first 8 centuries of the Common Era (Büntgen et al. 2011;Büntgen et al. 2016;Ebrey 1996).
To help fill this gap in climate history, we developed a new, nearly 2000-year-long treering-width chronology for the Qilian Mountains that are located along the north-eastern margin of the Tibetan Plateau. This region serves as a major water source for the famous Hexi Corridor in arid northwestern China. This area has been an important channel for economic and cultural exchanges between China and Central Asian and European countries since ancient times. For this reason, the characteristics and impacts of climate change in the Qilian Mountains throughout history are of profound interest. We collected tree-ring samples from the semi-arid western Qilian Mountains, which is an area that is particularly sensitive to climate change and human activities (Huang et al. 2016;Reynolds et al. 2007). Working in the same area, Tian et al. (2007) and Liang et al. (2009) reconstructed soil moisture and precipitation chronologies, respectively, but only for approximately the past 200 years. Yang et al. (2011) extended the drought record to 620 years in this region. In 2015, an 850-year reconstruction of May-July Palmer Drought Severity Index (PDSI) was established in this area and used to analyze severe multi-decadal drought events and the impact of solar activity during the Little Ice Age (Gou et al. 2015a). Recently, Yang et al. (2019) reported a 1556-year-long ringwidth chronology for the Hexi Corridor and compared the hydroclimatic changes between the Hexi Corridor and Qaidam Basin on different time-scales. They looked in particular at the Medieval Climate Anomaly (MCA, ~ 800-1200 CE) and the drying trend during the twentieth century. With this study, we extend the ring-width chronology to 134 CE. We examine the connection between climate and society by analyzing multi-decadal variations in moisture conditions during the past 1775 years and their possible linkages with societal activities in and around the Qilian Mountains. We focus in particular on the period from the third century to the eighth century.

Tree-ring and climatic data
The study area is in the Jiuquan area, which is located in the western Hexi Corridor, north of the western Qilian Mountains (Fig. 1). It has a typical continental arid climate with an annual mean temperature of 7.5 °C and annual total precipitation of 87.2 mm according to the data recorded at the Jiuquan meteorological station over the 1951-2015 period (data available from http:// data. cma. cn). The highest temperature and most precipitation both occur in July. Qilian junipers grow sparsely on south-and semi-south-facing slopes at elevations of 3100-3700 m above sea level (m a.s.l.) . Years of investigation in this region have revealed that Qilian junipers in this area are seldom older than approx. 300 years. However, a few small patches of old-growth stands exist in some of the high and mostly inaccessible reaches of the mountains, where they have survived for more than 500 years. Some standing old dead trees that are much older than the living trees can also be found in these areas.
Tree-ring samples were obtained in 2016 from the Yaoquan Valley (E97.88, N39.61), 50 km west of Jiuquan City, Gansu Province, China. The Yaoquan Valley runs from north to south; the tree-ring samples were taken from four sites located on the sunny, south-facing slopes on both sides of the valley. The distances between the sites are less than 10 km and the sampling elevations range from 3100 to 3500 m a.s.l. The standing old dead trees are distributed at various elevations but are mainly found at elevations between 3250 and 3450 m a.s.l.). There were no signs of insect damage on the bark side of the standing dead trees, but fire scars could be seen on the exposed roots of a few trees. Two or more increment cores were extracted from each living tree and each dead tree. In total, 107 cores were collected from 47 living trees and 171 cores were collected from 82 dead trees.

Chronology development
All of the sample cores were prepared in the laboratory using standard dendrochronological techniques (Stokes and Smiley 1996). We cross-dated the rings visually (Fritts 1976;Stokes and Smiley 1996) and measured the ring widths to the nearest 0.01 mm using a Lintab 5 measuring system. We then statistically verified the cross-dating accuracy using the COFECHA program (Holmes 1983). Because the four sampling sites are proximate to one another and share similar environments, tree growth at these sites should be governed by the same climatic conditions. Indeed, all ring-width series cross-dated well between the sites, with the average segment correlation coefficients of individual series with the master series ranging from 0.57 to 0.83. The mean of correlation coefficients between each series and the master series was 0.77. For this reason, all of the raw measurements were combined into a single chronology (hereafter referred to as YQ; Fig. 2). Because the mean segment length of all measurements is 282 years, with the longest measurement being 1481 years, we could not obtain a reliable regional tree growth curve for the Regional Curve Standardization (RCS) detrending technique, which is an effective method Fig. 1 Locations of the tree-ring sampling sites, Jiuquan station, four nearby sc-PDSI grids points, and sites of other climate proxies used in this study (JQ, Gou et al. 2015a;QF, Yang et al. 2019;HYG, Zhang et al., 2011b;Tian'E lake, Zhang et al. 2018;Qadam basin, Yin et al. 2016b;northeastern TP, Yang et al. 2014;northern China, Zheng et al. 2006;Huangye Cave, Tan et al. 2010). The 200-mm isohyet of annual mean precipitation  was calculated based on the CRU precipitation dataset (red line) for retaining low-frequency variations Melvin and Briffa 2008). The ring-width series were instead detrended using the Signal-Free (SF) procedure, which can reduce or remove the distortion introduced during the traditional standardization process . In this procedure, individual series were detrended using cubic smoothing splines with a 50% frequency response cut-off at approximately 67% of the series length for the original chronology . To account for changes in the variance associated with changes in sample depth through time, the variance in the chronology was stabilized using a method introduced by Osborn et al. (1997). The stabilized signal-free chronology was then used in the subsequent reconstruction. We used the RCSigFree program (Cook et al. 2014; https:// www. ldeo. colum bia. edu/ tree-ringlabor atory/ resou rces/ softw are) to develop the SF chronology. Then we used the ARSTAN program with the same spline function to fit the growth trend (Cook 1985) and obtained the basic statistical information for the common period 1230-1440 CE, including average sensitivity, signal-to-noise ratio, inter-series correlation (Rbar), and expressed population signal (EPS). The length of the reliable chronology is determined using the EPS statistic: an EPS greater than 0.85 is generally considered to be an acceptable threshold for a reliable chronology for climatic reconstruction (Cook and Kairiukstis 1990). In addition, standing dead tree (YQD) and living tree (YQL) measurements were combined into two different chronologies to allow for the comparison of living and dead trees.

Climate-growth relationship and reconstruction
Correlation analysis between the original and first differenced YQ chronology and climate variables from previous September to current September over the period 1951-2015 CE was used to investigate the climatic information contained in the YQ chronology and to determine the main factors limiting tree growth at the study site (Blasing et al. 1984;Fritts et al. 1971). Regression analysis was used to establish the transfer function of reconstruction using the data for the full calibration period. Both the leave-one-out correlation and the split-period verification methods were applied to examine the robustness of the reconstruction model (Meko and Graybill 1995). The statistics of the calibration and verification include the sign test (ST), the first difference sign test (FST), the product-mean T test (PMT), reduction of error (RE), and coefficient of efficiency (CE) (Fritts 1976). RE and CE values above zero indicate that the model has robust estimation skills (Cook et al. 1999).
To assess the regional significance of the reconstruction, the CRU gridded precipitation data (TS 4.01) (Mitchell and Jones 2005) and CRU self-calibrating PDSI data 0.5° × 0.5°) were correlated with the actual and reconstructed series, respectively, using the KNMI Climate Explorer (http:// clime xp. knmi. nl). We also correlated the reconstruction series with the gridded June-August PDSI reconstruction from the Monsoon Asia Drought Atlas (MADA) (Cook et al. 2010) for the longer time scale during 1300-2005 CE. Due to the lack of long chronologies in the Xinjiang area and eastern China, the interval of correlation analysis was set from 1700 to 2005 CE. Table 1 shows the statistics of the YQ standard chronology. The YQ chronology exhibits a relatively high mean sensitivity value (0.385) and a low first-order autocorrelation (0.195), indicating its potential for reflecting past interannual environmental changes, most likely moisture variations. Values for the average correlation among series, the signal-to-noise ratio (S/N), the variance explained by the first principal component, and the representativeness of the sample population are all relatively high, indicating that variation in ring widths across the sampling sites is consistent. The total length of the chronology is 1882 years (134-2015 CE); the starting year of the robust chronology (when EPS reaches 0.85) is 241 CE (Fig. 2).

Correlation analysis
As shown in Fig. 3a, the correlations between the tree-ring index and precipitation in the early growing season (May and June) are statistically significant and positive. At the same time, the correlations between the index and temperature in March, May, and June are negative (statistically significant in March and June). Correlations between the treering index and the monthly sc-PDSI are consistently positive and statistically significant from September of the previous year to August of the current year. The results of the first order differenced data are similar to those of the original data (Fig.3b), but the correlation coefficients are slightly higher. For instance, correlations with March temperature and precipitation in May and June remained statistically significant. The same is true for the positive correlations with May and June sc-PDSI. Further analysis of various monthly/seasonal combinations revealed that the highest correlation (0.68; p < 0.001) occurred between the tree-ring index and the average May-June sc-PDSI at the four grid points. For this reason, May-June sc-PDSI was identified as the target of the reconstruction.

Calibration and verification
The scatter plot shows that the tree-ring index has a non-linear relationship with the May-June sc-PDSI (Fig. 4a). This means that while dry years (less precipitation) generally cause narrow ring widths, wet years produce rings with a reduced range of variation for the same increment of the sc-PDSI values, from medium to wide ring widths. In fact, there seems to be a power relationship between the tree-ring index and May-June sc-PDSI. To test this, we logarithmically transformed the tree-ring index and sc-PDSI using a base-10 logarithm with a constant of 5 added to the sc-PDSI to make all values positive (Feng et al. 2013). The resulting scatter plot shows a more linear relationship with a correlation coefficient of 0.71 (Fig. 4b). We therefore estimated the transfer function with linear regression using the logarithmically transformed data: where sc-PDSI is the average May-June sc-PDSI at the four grid points closest to the sampling area and SSF is the ring-width index of the YQ stabilized signal-free chronology. The final reconstruction transfer function is: The model explains 50.5% of the variance in the May-June sc-PDSI for the entire calibration period. The results of the split-period verification and leave-one-out analyses indicate good model prediction skills, as represented by positive RE and CE statistics in all cases ( Table 2) (2) − = (10 ) 0.71 − 5.

Variations in interannual and interdecadal moisture patterns
The reconstructed May-June sc-PDSI series reveals prominent interannual and interdecadal moisture variations over the past 1775 years (Fig. 4d), with sc-PDSI values ranging from -4.6 to 3.4. Based on the 11-year running means and the ± 1.0 standard deviation of the reconstructed series, 19 distinct pluvial and 15 dry periods were identified (Fig. 4d).
During the past 1775 years, the driest and wettest periods were the late fifth century and the middle-late third century, respectively. For the period spanning the middle of the third century to the end of the eighth century, decadal moisture variation is characterized by large fluctuations. In particular, the third century was significantly wetter than normal but quickly became drier. By the end of the fifth century, moisture conditions had become the driest on record. Although it became wetter again in the sixth century, the period from the fourth to the eighth centuries was generally dry.
To evaluate the spatial representativeness of our reconstruction, we correlated the observed and reconstructed May-June sc-PDSI with the CRU gridded sc-PDSI field for the period 1952-2015 (Fig. 5). The results of the correlation analysis show that the actual sc-PDSI is significantly correlated with observed moisture variation in the surrounding area with a spatial extent of approx. 35-42°N and 75-102°E (Fig. 5a). The spatial pattern of the correlation coefficients between the reconstructed May-June sc-PDSI and observed moisture variation in the surrounding area (Fig. 5b) is very similar to that of the actual sc-PDSI (Fig. 5a). The regions with significant correlations are mainly distributed in the Qilian Mountains and the area to the west, i.e., the northern part of the Tibetan Plateau (Fig. 5b). These spatial patterns remain essentially unchanged after detrending the data (Fig. 5c), suggesting robust relationships on both interannual and interdecadal time scales. The results of the correlation between our reconstruction and the PDSI series from the MADA data (Cook et al. 2010) reveal a belt of significant correlations running NW-SE from the north-eastern Tibetan Plateau (including the Qilian Mountains) to central China (Fig. 5d).

Climatic implications of the YQ chronology and the validity of the regression model
Moisture is a common limiting factor for tree growth in arid and semi-arid regions (Fritts 1976). Our sampling sites are located in an arid area that receives less than 100 mm of precipitation annually, as indicated by the records from the nearby Jiuquan weather station, although the mountain slopes with the Qilian juniper stands receive more precipitation (Shao et al. 2005). The radial growth of Qilian junipers is mostly dependent on soil moisture recharged by precipitation and thus responds positively to precipitation and sc-PDSI variability (Gou et al. 2015b). As shown in Fig. 3, the positive (negative) correlations with precipitation (temperature, especially maximum temperature) in the early growing season indicate the typical moisture stress on Qilian juniper growth (Shao et al. 2010;Yang et al. 2019;Yin et al. 2008). Such relationships are widely reported for other studies in the Qilian Mountains and surrounding areas (Gou et al. 2015a;Yang et al. 2014;Zhang et al. 2015bZhang et al. , 2013. The significant positive correlations between the tree-ring index and precipitation and sc-PDSI in May and June indicate that moisture conditions in the early growing season of the current year influence ring width. A related study showed that moisture availability from May to June affects the radial growth rate and that this period accounts for more than 60% of the total annual growth (Zhang et al. 2016). Therefore, moisture availability in May and June can significantly influence the ring width of any given year. The results of the correlation analysis with the gridded sc-PDSI field (Fig. 5) confirm that drought variations along the northern margin of the Tibetan Plateau are well represented. This is especially true for the north-eastern part of the Tibetan Plateau, which further validates our decision to use the May-June sc-PDSI as the reconstruction target.
It should be noted that our chronology combines both living trees and dead trees. A previous study by Zhang et al. (2015a) has found that the Qilian junipers growing below the upper 20% of the forest belt on the northeastern Tibetan Plateau are precipitationsensitive. At YQ, the forest belt of Qilian junipers ranges from approx. 2950 to 3650 m a.s.l, all of our samples were taken below 3500 m a.s.l. and are moisture-sensitive as described above. Most of dead tree measurements (169 cores out of 171 cores) correlate significantly with the YQL master series over their respective periods, with an average correlation coefficient of 0.585. As shown in Fig. 2b, the variability of both the YQL and YQD chronologies are very consistent at high and low frequencies; during the reliable common period (541-1790 CE), the correlation coefficient is 0.78 (p < 0.001). This indicates that trees from both series responded to climatic factors in a similar manner. In addition, many of the standing dead trees died hundreds of years ago, and 7 standing dead trees died over one thousand years ago. During the sampling process, we observed that the bark of many of the dead trees have decayed and fallen off, as did the wood immediately beneath the bark. Due to subsequent weathering, it was very difficult to retrieve a complete core. As a result, we speculate that the rings representing the years when the trees were weakened before their eventual death were lost to decay or during the sampling process. However, the dead tree cores all show good growth conditions in their interior portions. The impacts of the abnormal factors leading to the death of these trees have therefore most likely been eliminated and are not reflected in the final chronology.
The ST results did not pass the 0.95 confidence level for the verification periods of 1952-1983 and 1984-2015 (Table 2), which can partly be explained by the disagreements between the observed and predicted values in pluvial years (Fig. 4c). In dry regions, narrower rings provide more precise climatic information than do wider rings (Fritts 1976). In pluvial years, precipitation is abundant and the additional water is unavailable to the tree due to runoff or percolation below the root zone, and is therefore of no benefit to growth (Fritts 1976;Sun et al. 2017). Accordingly, the sensitivity of Qilian juniper ring-width increments to available precipitation decreases. Furthermore, because the sc-PDSI calculation considers the cumulative effect of water deficits in the previous 9-12 months (Wells and Goddard 2004), the observed sc-PDSI values probably do not reflect the true moisture conditions in some years, such as the dry years of 1957 and 1986. In addition, the disagreement between the observed and predicted values mostly occurs after 2001, which has also been observed in studies of different areas and different tree species (Fang et al. 2012;Gou et al. 2015b;Yang et al. 2019;Chen et al 2015;Chen et al. 2014;Shi et al. 2015). Given that the automatic weather station system in China was completed around 2003 (Li et al. 2013), the difference between manual observations and automatic observations may be a reason for such disagreement. A recent study by Liu et al. (2021) indicates that temperature signals in tree-ring-width data can profoundly affect the low-frequency trends of precipitation reconstructions for the northeastern Tibetan Plateau. For this reason, temperature should also be considered as a potential explanation.

Comparisons with regional hydroclimate reconstructions
As shown in Fig. 5d, significant correlations between our reconstruction and the PDSI series from the MADA data (Cook et al. 2010) were found for the north-eastern Tibetan Plateau and its surroundings, which also extend toward the southeast (Fig. 5d). We further compared our reconstruction to several reconstructions of moisture conditions in the vicinity, including the ratios of Artemisia and Amaranthaceae (A/C) in Tian'E Lake   (Fig. 6a) and two tree-ring-based moisture reconstructions (JQ and QF) (Gou et al. 2015a;Yang et al. 2019) (Fig. 6b) located in the Jiuquan area. The wet and dry periods in our series correspond well with those found in the Tian'E Lake series. Notable dry periods include the mid-nineteenth century, the mid-late seventeenth century, the middle and late fifteenth century, and the mid-late thirteenth century; and notable wet periods include the 1220s, the 1610s, the 18th and the late nineteenth centuries. However, there are also differences between the two series, especially before 1200 CE. For instance, there was an abnormal wet period at Tian'E Lake around the 720s, whereas the YQ series shows a relatively dry period at this time. Tian'E Lake is located at 3012 m a.s.l., which is 400 m lower than the location of our sampling site. This precipitation difference may be explained by differences in elevation and microclimate, as differences in precipitation between mountain and lowland have been observed in the eastern Qaidam Basin of the Tibetan Plateau (Dong et al. 2021). Similarly, both the JQ and QF series are in good agreement with our reconstruction on interannual to interdecadal timescales, with correlation coefficients of 0.86 (JQ, 1161-2010 CE) and 0.54 (QF, 455-2011 CE), respectively. Additionally, 46 drought events are recorded in historical documents from Jiuquan and Dunhuang during the period 241-1960 CE (Dunhuang City Annals Compilation Committee 1994; Jiuquan History Office 1998). These include 9 spring droughts, 7 summer droughts, 4 autumn droughts, and 26 droughts with unspecified seasons. A total of 24 of these drought events correspond with the dry years in our reconstruction, including 6 spring droughts, 4 summer droughts, 2 autumn droughts, and 12 droughts with unspecified seasons. This indicates that our reconstruction is especially adept at capturing spring/summer drought events.
To test the spatial representation of our reconstruction, we further compared it with several tree-ring-based reconstructions farther from our study area, including an annual previous August-current July precipitation reconstruction for the Zhangye area (HYG) (Zhang et al. 2011b), a January-June moisture balance reconstruction for the eastern Qaidam Basin (Yin et al. 2016b), and an annual (previous July-current June) precipitation reconstruction for the north-eastern Tibetan Plateau . We also included the May-September historical document-based dry-wet index series for northern China (Zheng et al. 2006) and the stalagmite oxygen stable isotope record from the Huangye Cave in central China (Tan et al. 2010) (Fig. 6). The correlation coefficients between the YQ series and the three tree-ring-based reconstructions listed above are 0.55, 0.39, and 0.40, respectively, all of which are statistically significant at the 0.01 level. Decadal-scale dry and wet fluctuations in our reconstruction are consistent with those of the earlier tree-ring-based reconstructions, as are most of the multi-decadal drought and pluvial events. In addition, some of these events, such as the droughts in the 460s-500s and 1450s-1510s and the pluvial events in the 790s-820s and 1860s-1910s, are also found in the historical document series for northern China and the stalagmite series for central China. Overall, the major wet and Fig. 6 Comparison between the reconstruction presented in this paper (c) and other moisture-related series from the Jiuquan area (a. Zhang et al. 2018;(b. Gou et al. 2015a (JQ, red line) and Yang et al. 2019 (QF, black line)), the Zhangye area (d. HYG, Zhang et al. 2011b), the Qadam basin (e, Yin et al. 2016b), the northeastern Tibetan Plateau (f. Yang et al. 2014), northern China (g. Zheng et al. 2006), and Huangye Cave in central China (h. Tan et al. 2010). Blue columns indicate the prosperous periods of the Silk Road; red columns indicate spatially extensive drought periods; green columns indicate spatially extensive pluvial periods dry events found in our reconstruction are also found in other reconstructions. This suggests that our reconstruction captures signals of regional inter-annual to decadal moisture variability over the northeastern Tibetan Plateau and beyond. Due to regional differences in precipitation or moisture changes, some differences in the long-term dry and wet conditions among different reconstructions can also be observed, such as the dry and wet conditions around the third century, the eighth century, and the twentieth century. Moreover, the duration of large-scale droughts tends to be longer in western China, as was the case for the droughts around the fifteenth and seventeenth centuries. This phenomenon has also been found in other studies; for example, multiple studies have shown that the drought event in the 1920s lasted longer in arid areas than in other areas (Gou et al. 2014;Liang et al. 2006;Zhang et al. 2011a).

The link between interdecadal moisture variations and human activities during the third through the eighth centuries
Extended periods of wet or dry conditions in arid regions often have great impacts on human society; this was especially true in ancient times. For instance, Zhang et al. (2018) found that persistent droughts were responsible for the abandonment of several settlements and cities in the Qilian Mountains/Hexi Corridor region, including the ancient city of Dunhuang. As mentioned above, the tree-ring reconstructions for the northeastern TP are in good agreement with one another, whereas some differences exist between our reconstruction from a high-mountain area and the lake sediment-based reconstruction from the lowlands. However, a recent study indicates that precipitation in the surrounding mountains instead of in the lowlands facilitated the prosperity of ancient civilizations in mountainbasin systems (Dong et al. 2021). Precipitation falling in the high mountains provided abundant meltwater runoff to nourish the oasis in the lowland and facilitated agro-pastoral production and human settlement, which were more conducive to social stability and economic development. For this reason, we used our reconstruction and others from the region to further investigate the possible linkage between regional hydroclimate and social activities.
As shown in Fig. 6, the period from 240 to 310 CE was one of the most significant wet periods of the past 1775 years. Historical archives and the stalagmite record ( Fig. 6g and h) indicate that the climate during this period was wet throughout northern and central China. This period corresponded to the Western Jin Dynasty (266-316 CE), the dynasty that unified the warring states of the Three Kingdoms and controlled our study area at the time (Ebrey 1996). The climate then rapidly became drier starting around 310 CE, which was when the Western Jin Dynasty collapsed and nomads from the north and northwestern parts of China invaded central China (Fig. 7a). Although the civil unrest of ethnic minorities at the time was related to the emergence of a minority population in northern inland China and the escalation of national oppression (Cui 2012;Ebrey 1996;Yoshiaki 2019), the dry conditions during this time would have inevitably led to the degradation of pastures in northern and northwestern China and hence to increased livestock mortality (Begzsuren et al. 2004;Sternberg 2008). When drought occurred, nomads were pushed to migrate southward and invade agriculturalist polities, namely the agricultural settlements of the Han Chinese (Pei and Zhang 2014;Pei et al. 2019).
The increasing dryness in this region eventually peaked around the end of the fifth century. Nomadic groups conquered and occupied the northern part of central China during this period (Su et al. 2016), while frequent regime changes occurred in our study area (Ebrey 1996;Yoshiaki 2019). When moisture conditions improved in the latter half of the sixth century, the region that included our study area was governed successively by two unified dynasties, the Sui Dynasty (581-617 CE) and the Tang Dynasty (618-907 CE) (Twitchett and Fairbank 1979). Similarly, the ancient Silk Road region entered a period of prosperity around the year 609 CE (Jiuquan History Office 1998). In fact, moisture conditions in and around our study region were either normal or wetter than normal during two other Silk Road boom periods, the 240s-250s and the 710s-750s CE (Dunhuang City Annals Compilation Committee 1994; Gao 2011).
As shown in Fig. 7, periods in which wars were frequently initiated between nomadic and farming groups often correspond with dry periods between the third and eighth centuries. However, we would add that the lack of warfare in northwestern China around the late fifth century, a period of notable dryness, was largely the result of a southward shift in the warring zones associated with the invasion of the nomads. Similarly, the frequent wars during the relatively humid period around 600 AD were mainly related to social unrest before and after dynastic change. The reduction of resources caused by climate deterioration was more likely to lead to conflicts between nomadic and farming groups. On the time scales of 10-30 years, precipitation has a more significant effect on social vicissitudes (Yin et al. 2016a). For example, in the late third century, persistent drought in northwestern China (He et al. 2013;Liu et al. 2009;Wang et al. 2019;Yang et al. 2014;Yin et al. 2016b) resulted in the migration of more than 100,000 people from Gansu Province to Sichuan Province due to years of famine. This in turn led to the outbreak of war and the subsequent establishment of the Cheng Han regime (Compile Group of Chinese Military History 2002). Similarly, too little precipitation can trigger pasture deterioration and nomadic migrations, especially for people in northern and northwestern China (Pei and Zhang 2014). It should be emphasized that many factors played a role in determining the political and economic conditions that led to instances of warfare between nomads and the agricultural Han Chinese dynasties. However, our results indicate that unusual moisture variations seemed to have played indirect but important roles in influencing human activities during ancient times.

Conclusion
In this paper, we present a 1775-year May-June sc-PDSI reconstruction for the western Qilian Mountains based on a new tree-ring-width chronology developed from Qilian juniper by combining 107 cores from healthy living trees and 171 cores from standing dead trees. This study provides new climate information for the period prior to 800 CE, a period for which high-resolution climate data are especially scarce. The reconstruction reveals several distinct wet and dry periods during the past 1775 years, including an especially wet period during the third century (240s-250s CE /280s-310s CE) and an especially dry period in the late fifth century (450s-490s CE). It also provides new evidence that significant events in human history were likely related to climate change and variations on the interannual to interdecadal timescales. Future work is needed to provide more information about the impacts of interdecadal droughts on human history.