The response of annual minimum temperature on the eastern central Tibetan Plateau to large volcanic eruptions for the period 1380–2014 AD

Volcanic eruptions have a significant impact on global temperature; their consequences are of particular interest in regions that are especially sensitive to climate change, like the Tibetan Plateau. In this study, we develop a temperature-sensitive tree-ring width standard chronology covering the period 1348–2014 AD using Qilian juniper (Sabina przewalskii Kom.) samples collected from Animaqin Mountain on the Tibetan Plateau. We 15 reconstruct the annual (prior August to current July) mean minimum temperature (Tmin) since 1380 AD and show that our reconstruction explains 58% of the variance during the 1960-2014 calibration period. Our results demonstrate that in 77.8% of cases in which a volcanic eruption with a volcanic explosivity index of 5 or greater occurs, temperature decreases in the year of or the year following the eruption. The results of the Superposed Epoch Analysis also indicate that there is a high probability that the Tmin decreases within 2 years of a large volcanic 20 eruption, especially when such eruptions occur in the Northern Hemisphere.


Introduction
Large volcanic eruptions can affect the climate of the Earth (Robock, 2000) and have played a major role in past global temperature changes (Salzer and Hughes, 2007). Eruptions emit large amounts of ash particles and gases 25 into the atmosphere, much of which are carried to other regions by atmospheric movement. These materials efficiently reflect incident solar radiation, resulting in the cooling of the Earth's surface. However, volcanic eruptions of similar magnitudes do not necessarily result in cooling across all areas of the world. For example, the 1991 eruption of Mount Pinatubo, Philippines, caused summer cooling over much of the globe in 1992, but the temperature in some areas was above average (Robock and Mao, 1992). Thus, it is not necessarily clear to what 30 extent or in what manner a strong volcanic eruption will influence temperature in a particular region.
Often referred to as the "third pole", the Tibetan Plateau (TP) is especially sensitive to climate change (Yao et al., 2012) and may therefore be more profoundly influenced by volcanic eruptions. As early as 1985, studies of the relationship between large volcanic eruptions (dust veil index > 1000m 3 ) and temperature variations from 1951 to 1980 in China (Zhang and Zhang 1985) demonstrated that temperature on the northeastern TP decreased 8, 15, and 35 18 months after eruptions. However, temperatures in the 6 months immediately following an eruption were found to be relatively high. Jia and Shi (2001) studied climate signals following volcanic eruptions from 1950 to 1997 and found that temperature on the TP dropped within 2 years after eruptions during this period. However, new research focusing on regional differences over China of winter temperature response to large volcanic eruptions with different latitudes and seasons from 1956 to 2005 found that winter volcanic eruptions led to extensive 40 warming of winter temperature over Tibet Plateau .These studies show that temperature on the TP is affected by volcanic activities, but it is important to note that they are based on instrumental data, which covers a relatively short time span. Temperature changes caused by strong volcanic eruptions can affect tree growth (Tognetti et al., 2012), an influence that can be seen in the rings of certain trees (D'Arrigo et al., 2013;Filion et al., 1986;Lamarche and Hirschboeck, 1984) and used to identify past volcanic activity (D'Arrigo et al., 2013;Filion 45 et al., 1986;Lamarche and Hirschboeck, 1984). Especially when long-lived trees are available, tree-rings can serve as temperature-sensitive proxies for investigating climate responses to volcanic eruptions that occurred prior to the instrumental record (D'Arrigo et al., 2013;Salzer and Hughes, 2007).
Tree rings from the TP can potentially be used to study the climate response to volcanic activities. Previous treering based studies showed that some cold years can be closely correlated with large volcanic events (Liang et al., 50 2008;Liang et al., 2016;Zhang et al., 2014). Li et al (2017a) quantitatively assessed the correlation between temperature changes in the southeastern TP and volcanic eruptions and showed that most of the years of extreme cold in the past 304 years occurred 1-2 years after major volcanic eruptions. However, the influence of volcanic eruptions on temperatures on the TP over the long-term is less well understood due to the paucity of data in this region. 55 The Animaqin Mountains are located on the northeastern TP, many long tree-ring series, with hundreds of years or even thousands of years (Chen et al., 2016;Gou et al., 2007;Gou et al., 2008;Gou et al., 2010), were developed here, and provided a significant proxy for the tree growth response to volcanic dynamics, but the response studying of tree rings to volcanic activity is still rare across this area even the TP. Using tree-ring samples of Qilian juniper (Sabina przewalskii Kom.) collected from a new sample site in southeast part of the Animaqin Mountains, a 667-year tree-ring width chronology is developed, and then is used to reconstruct annual mean minimum temperatures (Tmin) across the east-central TP. Finally, this study explores the response of Tmin to strong volcanic eruptions (Volcanic Explosivity Index (VEI) ˃ 4) over the past six centuries. Qilian juniper samples were taken from a natural woodland 35km west of Ningmute town, Henan County, Qinghai Province, China (E100.96,N34.62,3806 m a.s.l). This area is located in the sub-frigid zone and has a semi-humid climate (Zheng et al., 2013). According to climate data of the Henan meteorological station ; Table 1), mean annual temperature is 0.31℃ and mean annual precipitation is 582 mm. Precipitation is mainly concentrated between May and September. Because the site is close to the Yellow River and the soil layer 70 in the forest area is thick, the moisture conditions of the forest are good. The regional vegetation zone is coniferous forest and the main tree species include Sabina przewalskii Kom., Picea crassifolia Kom., Betula spp., and Salix cheilophila. The study area is located on the southern slope of the Animaqin Mountains and ranges in elevation from 3523 to 3900 m a.s.l. (Figure 1). The gradient of the slope is 30°-40°. A total of 110 cores from 55 trees were drilled at breast height with an increment borer in 2015. 75 (figure 1 near here) The cores were air-dried, fixed to wooden mounts, polished, and cross-dated. The cores were then measured using a Lintab 6 tree-ring width measuring instrument with a resolution of 0.01 mm. The COFECHA program (Holmes, 1983) was used to check the quality of the cross-dating and the accuracy of the measurements. The signalfree (SF) standardization method (Melvin and Briffa, 2008) was adopted to standardize the tree-ring data in order 80 to minimize trend distortion in the chronologies produced by the straight-line detrending method. The RCSigFree program (https://www.ldeo.columbia.edu/tree-ring-laboratory/resources/software) was used to establish the treering width chronology. In this procedure, the age trend curve was fitted to cubic smoothing splines with a 50% cutoff at approximately 67% of the mean segment length. The validity of the SF chronology was assessed using the mean correlation coefficient for the tree-ring series (Rbar) and the value of the expressed population signal 85 (EPS).
Thirteen meteorological stations were identified in the region of the sampling site: Zhongxinzhan (ZXZ), Dari  Because the observation intervals of the stations shown in Table 1 differ, it was necessary to select the longest 95 record to ensure the stability of the correlation function. Seven stations with instrumental data spanning <30 years were excluded: GL, GN, and TR (all of which were established in the 1990s), and ZXZ, TD, ZK, and LMS (all of which ceased monitoring in the 1980s or 90s). Climate data from the other six stations (DR, XH, HN, JZ, MQ, and HZ) were used for the following analysis (Fig. 1). During the quality control of the data from the six stations, we found that there are some problems in the instrumental data from the HN, JZ and XH station. representative of the regional climate. The monthly climate data from the previous July to the current September were used to analyze the response of tree growth to climate change.

Methods
The correlation function was used to analyze the relationship between the tree-ring width index and climatic factors.
A fitted equation was then established using a simple linear regression and verified by the cross-validation method (Michaelsen, 1987) and split-period calibration/verification analysis (Meko and Graybill, 1995). Several statistical 120 tests (sign test, product mean value, reduction error, and coefficient of efficiency) were utilized. The split periods for calibration were 1960-1991 and 1984-2014. The correlations between the observed and reconstructed series and the gridded dataset (TS3.22; Mitchell and Jones, 2005) from the University of East Anglia Climatic Research Unit (CRU) were analyzed using the KNMI Climate Explorer research tool (http://climexp.knmi.nl). The Superposed Epoch Analysis (SEA) method (Haurwitz and Brier, 1981) was also used to analyze the influence of 125 volcanic activity on regional temperature. SEA is a statistical method used to resolve significant signal-to-noise problems and is often used to study the link between climate and discrete events such as solar activity, fire events, volcanic activity etc. (Adams et al., 2003;Singh and Badruddin, 2006;Swetnam 1993;Esper et al., 2013;). In this study, the year of a volcanic eruption is regarded as year 0. The years before the volcanic eruption are denoted as -1, -2, -3 and so forth, whereas the years after the eruption regarded as 1, 2, 3, etc.. The impact of volcanic eruption 130 on temperature in the east-central TP was analyzed by comparing differences in temperature in the years leading up to and following an eruption. The significance of responses is determined by a Monte Carlo resampling procedure (10,000 iterations) (Adams et al., 2003). February, and has a positive correlation with Tmean, Tmax, and Tmin, most notably with Tmean and Tmin. For the 15month period from the previous July to the current September, the tree-ring width index correlates significantly and positively with the monthly Tmin except for the previous August Tmin. Correlation with annual Tmin (from 145 previous August through current July; hereafter referred to as Tmin87) is at the 0.01 significance level.
The first-differenced correlations between the tree-ring width index and climate data are weak; in some months, the first-order correlations are even negative (Figure 3b). The first-differenced chronology correlates significantly and positively with precipitation of the previous September and the current May, but negatively with precipitation of the previous December. For temperature, the tree-ring width index shows clear correlations with Tmean and Tmin 150 and correlates significantly and positively with Tmin of the previous September and November and current February, June, and July. The positive correlation with the annual Tmin is at the 0.01 significance level.
The tree-ring width index is most strongly correlated with Tmin87 (r = 0.767, p<0.001). This correlation remains strong and positive after first-differencing (r = 0.583, p<0.001), which indicates that the tree-ring width index is suitable for reconstructing the Tmin87 for the given period.

Reconstruction development and verification
The tree-ring width index for the current year (SFt) was selected as the predictor to reconstruct the annual mean minimum temperature departure (Tmin87): (1) 160 The reconstruction accounts for 58% of the variance in the instrumental series during the calibration period . The F value is 74.187, and exceeds the confidence level of 0.001. The transfer function is therefore highly significant (Figure 4a and 4b).

Spatial representation 185
Correlations between the reconstructed series, instrumental data, and the CRU gridded annual Tmin are high over most of China and even across most of Asia for the period 1960-2014 (Figure 5a and 5b). However, the firstdifferenced instrumental data only correlate significantly with the CRU gridded data over the TP (Figure 5c (Figure 7a). In the year immediately following the eruption, the relationship is at the 0.05 significance level. Eruptions in the Northern and Southern Hemispheres (Figure 7b and 7c) also coincide with a Tmin decrease in the year of the eruption and for one to two years thereafter. However, eruptions in the Northern Hemisphere have a more obvious influence on Tmin in the study area, as indicated by the strong drop in temperature in the year following an eruption (Figure 7b). Southern Hemisphere mid-latitude eruptions occur at greater 205 distances from the TP and therefore have a correspondingly weaker influence on Tmin.
Tmin in the study area decreases significantly in the years following volcanic activity at low latitudes ( Figure   7d). Eruptions in the mid-latitudes of the Northern Hemisphere clearly coincide with drops in temperature, but the decreases are not statistically significant (Figure 7e). Eruptions in the mid-latitudes of the Southern Hemisphere and in the high latitudes of the Northern Hemisphere also coincide with reduced Tmin in the year of the eruption and 210 in the following year, but the decreases are not statistically significant (Figure 7f and 7g).

Reliability of the Tmin reconstruction
The correlation between the tree-ring width index and climate factors shows that the relationship between tree radial growth and precipitation is not statistically significant except in February. With the thick topsoil and humid 215 climate in spring and summer, the study area meets the needs of trees for radial growth. However, according to instrumental data from the GL weather station (Table 1), the elevation of which is close to that of the sampling site, the annual Tmean and Tmin are 2.25℃ and -6.76℃, respectively. These temperatures are quite low for tree growth, and the statistically significant positive correlation between growth and temperature shows that tree radial growth in this area is restricted by temperature, especially Tmin. Tmin before and during the growing season may affect tree 220 growth in several ways. In winter and early spring, warmer Tmin protects roots from cold damage and triggers earlier snowmelt. Warmer Tmin may therefore result in a longer growing season, and trees may experience increased growth in the subsequent growing season (Pederson et al., 2004;Fu et al. 2012;Hollesen et al. 2015;Williams et al. 2015).
In the other hand, Tmin is known to affect conifer tracheid division and enlargement by controlling the onset and conclusion of xylogenesis during the growing season (Deslauriers et al., 2003;Rossi et al., 2008;Li et al., 2017b;225 Hosoo et al. 2002;Steppe et al., 2015).
It should be mentioned that the CE value of the validation period remained negative for the period from 1960 to 1983 even though we tested different meteorological stations and different observation intervals in the analysis process. One reason for this could be a distortion of the meteorological data due to the poor management and/or the relocation of meteorological stations during the 1950s and 1960s on the TP. However, the cross-validation results indicate that the equation is otherwise reliable, the negative value of CE should not influence our analysis about the high frequency climate response to volcanism.
The results of correlations of the reconstructed series with CRU Tmin reflect the regional significance of the reconstruction in general, however, the consistent warming trend of Tmin over most of Asia (Dong et al., 2017) may result in the large area of significant positive correlations, the results of first-order correlation analysis are 235 therefore more referential, i.e. our reconstruction can reflect the temperature variations in the Animaqin Mountains and the area to the west.
The reconstructed Tmin87 in this study was further compared with minimum temperature reconstructions from other regions of the TP, including HY (northeastern TP; Zhang et al., 2014), QML(ZHD) (central TP; He et al., 2014), HBL (eastern TP; Gou et al., 2007), YS (south-central TP;Liang et al., 2008), and LIT(TAN) (southeastern 240 TP; Li and Li, 2017). As a result of differences in reconstruction factors, study regions, and methods of chronology establishment, there are some differences between the chronologies. For example, there are significant regional differences in Tmin reconstructions over the past 100 years. However, there is also notable consistency at the interannual-multidecadal scale. The correlation coefficients between our reconstruction and each sequence are 0.227 (HY; p < 0.01), 0.328 (QML(ZHD); p < 0.01), 0.499 (HBL; p < 0.000), 0.235 (YS; p < 0.01), and 0.317 245 (LIT(TAN); p < 0.01). The closer the sequence is to the study site, the more similar the changes in Tmin are. For example, the Tmin87 of this study is most strongly correlated with the HBL reconstruction (Figure 8d), which is nearest to our study area. Both reconstructions show low-temperature periods at the end of the 16th century and from the 1670s to the 1720s. Although it is located further from the study area, the HY reconstruction also shows cold periods in the late 15th century, at the end of the 16th century, at the beginning of the 18th century, and in the 250 middle of the 19th century (Figure 8a). Similarly, the low-temperature periods in the late 15th and late 16th centuries, the 1670s to 1720s, and the 1960s to 1980s are in agreement with those in the QML(ZHD) reconstruction ( Figure 8b). The low-temperature interval in the 1960s-80s coincides with the cold interval in the YS reconstruction ( Figure 8e). Finally, although the LIT(TAN) reconstruction is also located in another region of the TP, its lowtemperature intervals are consistent with those in our reconstruction (e.g., at the end of the 15th century, 1670s-255 1720s, 1900s-1920s, and 1960s-80s) (Figure 8f). It is interesting that higher Rbar values appear in some special intervals, i.e., the cold intervals of the 1460s-1500s and 1800s-1820s and the warmest period of 1980-2014. These higher Rbar values indicate a good consistency among tree-ring series during these periods; in fact, these three intervals are evident in tree-ring-based studies from elsewhere on the Tibetan Plateau (Huang et al. 2019;Liang et al. 2016;Shi et al. 2019). The two cold periods identified in our series correspond to period of weaker solar activity 260 (the Spörer Minimum and the Dalton Minimum), and to a few very strong tropical eruptions (e.g., the Tambora eruption in 1815 and another stratospheric eruption in 1809) (Cole-Dai et al., 2009). Similarly, the warming in 1980-2014 is closely related to the influence of human activities. These responses are indicative of the consistent response of tree growth to strong external forcing factors and of the reliability of our reconstruction. As shown in Figure 6, the cooling probability in the year of or the year following a large volcanic eruption is very high. The effects of the Tambora volcano in 1815 (VEI = 7) were recorded in many parts of the world. Our reconstruction indicates that temperatures on the TP dropped by about 0.5°C in 1816, the year following the eruption. On the southeastern TP, the cold period from 1816 to 1822 may have also been related to the Tambora 270 eruption (Liang et al., 2008). Other research in the northeastern TP has shown that cold years can be matched with known tropical volcanic events in 15 of 21 cases (Zhang et al., 2014). We compared years of cooling we identified in this study with those identified by Zhang et al. (2014), the cooling years identified by the two studies are either the same or within a year of each other. On the southeastern TP, Liang et al. (2016) showed that the 15 coldest years of the past 304 years occurred mostly within 1-2 years of a major volcanic eruption, nine of these 15 cooling 275 years are also seen in our study. The results of SEA analysis further confirm that the temperature of the TP is affected by strong volcanic eruptions, especially those occurring at the Northern Hemisphere and low latitudes, and that cooling occurs in the study area within a year or two of a major eruption.
Studies have shown that some cold intervals in the eastern and southern TP may be influenced by large volcanic eruption in low-latitude regions (Bi et al., 2020;Duan et al., 2019a;Krusic et al., 2015), and the surface air 280 temperature in the TP were cooling in the first winter based on the ensemble simulation of the climate response to high-latitude volcanic eruption (Oman et al., 2005). Using the fully coupled NCAR Community Climate System Model (CCSM3), Schneider et al. (2009)  hemisphere seem to be more pronounced. The mechanisms related to how volcanic eruptions influence local temperature are undoubtedly complex, and temperature variability is further driven by other factors, such as circulation patterns like ENSO (Breitenmoser et al., 2012;D'Arrigo et al., 2011;Duan et al., 2019b). Cooling on 290 the TP as a result of large volcanic eruptions may be weakened or masked by other influencing factors (Duan et al., 2018), which is likely why a one-to-one correspondence between large eruptions and cooling is not observed.
Establishing a deeper understanding of the relationship between eruptions and cooling will depend on expanding the spatial network of long-term, temperature-sensitive chronologies from the TP.

Conclusions
This study establishes a 667-year-long tree-ring width chronology for the east-central TP. The anomaly sequence of the annual Tmin (Tmin87) was reconstructed for the period 1380-2014 AD. The lowest temperatures with the longest duration occurred in 1586-1602; the longest and warmest interval was 1991-2009. The Tmin on the TP has been increasing, most notably since the 1980s. The SEA analysis shows that Tmin decreases in the year following 300 strong volcanic activity in the Northern Hemisphere. Thus, the climate across the east-central TP is sensitive to large-scale background climatic factors such as volcanic activity. Further investigations are needed to fully understand the connection between volcanic eruptions and temperature on the TP, and to incorporate this information into regional annual and decadal predictions of temperature.

Author contribution 305
Yong Zhang and Xuemei Shao took the tree ring samples and designed this study; Yajun Wang and Yong Zhang analysed the data and prepared the manuscript. Mingqi Li prepared the volcanic data and took the tree ring samples.       (Gou et al., 2007); (e) YS, June-August Tmin on the southcentral TP (Liang et al., 2008); (f) LIT(TAN), April-March Tmin on the southeastern TP (Li and Li, 2017). slash columns indicate warm periods in our reconstruction; grey columns indicate cold periods.