Tree rings reveal changes in the temperature pattern in eastern China before and during the Anthropocene

A comprehensive understanding of the spatial characteristics of warming trends and temperature variability is important given global warming. Paleoclimate reconstruction has played an important role in evaluating warming and extreme climactic events in recent decades. Using the ring width of Pinus tabulaeformis, we reconstructed the ground surface temperature changes at Mengshan Mountain, in the central part of eastern China (EC), from 1667 to 2019. There were 3 extremely low-temperature years, 42 low-temperature years, 256 normal years, 37 high-temperature years and 15 extremely high-temperature years over the past 353 years. High and extremely high temperatures mainly occurred in the first half of the 19th century, the end of the 20th century and the beginning of the 21st century; low and extremely low temperatures occurred in almost all periods except for the last 50 years. By combining the past temperature reconstructions from the Great Xing’an Mountains in northern EC and Taiwan in southern EC, we also found that after the start of the Anthropocene, there were strong positive correlations among the temperatures of the northern, central and southern parts of EC. However, before the Anthropocene, these correlations were weak, and there was even a significant negative correlation during some periods. Additionally, the temperature in EC exhibited different change patterns before and during the Anthropocene. Before the Anthropocene, the temperature in southern EC showed an upward trend; in central EC, the temperature first rose and then began to decline in the 1820s; there was no obvious trend in the temperature changes in northern EC. The spatial consistency of temperature changes during the Anthropocene might be related to the fact that greenhouse gases emitted by human activities diffuse evenly withatmospheric circulation and absorb longwave radiation to directly heat the atmosphere.


Introduction
Human activity has induced global warming, in which extremely high-temperature events occur frequently (IPCC 2021). However, there are spatial differences in warming trends and temperature variability (NCCCMA 2021); thus, it is highly important to scientifically evaluate the temperature rise and frequency of extreme climatic events in individual areas. To evaluate the rate and range of temperature rise in the modern warming period, the spatio-temporal patterns of past temperature changes have been examined by governments and climatologists. For example, the PAGES 2k Consortium (2013) pointed out that past temperatures for seven continental-scale regions showed generally cold conditions between 1580 and 1880, but the worldwide medieval climate anomaly (MCA) and little ice age (LIA) had no synchronous multi-decadal warm or cold intervals at global scale. At present, research on temperature changes in many regions of China is mainly based on observed data. However, most observations began in approximately 1950, a period too short to comprehensively understand the characteristics and causes of temperature change. It is therefore necessary to reconstruct past temperatures, paying particular attention to interannual variations in temperature. Understanding the past variability and change of temperature provides a longer-term context for recent increasing trends and improves characterization of the magnitude and spectra of modes of natural low-frequency variability (Cook et al 2013).
In the field of paleoclimatology, tree rings have been widely used in climate reconstruction studies due to their advantages of long sequences, accurate cross-dating, continuity, high resolution and high sensitivity to climate change (George 2014, Ljungqvist et al 2016, He et al 2019, Liu et al 2019a. These tree rings provide temperature reconstructions with annual resolution (Esper et al 2016, Christiansen and Ljungqvist 2017, Büntgen et al 2021. However, due to a shortage of old trees in eastern China (EC), there have been fewer temperature reconstructions, especially ground surface temperature (GST) reconstructions, based on tree rings (Liu et al 2019a). GST is defined as the soil temperature at 0 cm depth and is affected by surface air temperature (SAT) but is not entirely consistent with SAT variations in terrestrial regions (Mann and Schmidt 2003). At present, there are only two series of GST reconstruction in EC. Zhang et al (2018) reconstructed the summer GST in the Great Xing'an Mountains (GXAMs) in Northeast China over the past 400 years, and Cai et al (2020) reconstructed the dekad-6 to dekad-27 minimum GST of Jinggangshan in southeastern China from 1791 to 2014. Therefore, to fully understand GST changes in EC, it is necessary to conduct new GST reconstruction series. In addition, temperature reconstructions based on tree rings in EC have been mainly performed in the northern and southern areas and less in the central region of EC (Liu et al 2019a). Moreover, the timespan covered by existing temperature reconstruction is relatively small (He et al 2019).
Since the Anthropocene (Subramanian 2019), the temperature in some regions has shown the same warming trend (Qian and Lin 2004, Jimenez-Munoz et al 2013, Zang et al 2019 or a strong correlation in recent decades (Chooprateep and McNeil 2016). Because these studies were based on observed and reanalyzed meteorological data, it remains unclear whether the spatial patterns of temperature changes in recent decades illustrate a similar temperature relationship in the past few centuries. In addition, in the study of dendroclimatology, the spatial correlation during the observed time period is often used to indicate the past spatial representation of reconstructed climate series (Li et al 2017, Mei et al 2019, Liu et al 2019b, Cui et al 2021. Is this representation based on spatial correlation suitable for all reconstruction? In the study, tree ring research was carried out in the Yimeng mountains in central EC ( figure 1(a)). Our main scientific objectives were to: (a) explore the relationship between tree rings of Pinus tabulaeformis on Mengshan Mountain (MSM) and climate factors; (b) reconstruct the GST of MSM over the past 353 years and analyze its change characteristics and (c) reveal the temporal patterns of temperature change in different regions of EC.

Sampling site
MSM is a branch of the Tai-Yi Mountain Range and is located in southern Mengyin County, Shandong Province. This region is dominated by a warm, temperate monsoon climate with distinct seasons. The annual average temperature is 13.2 • C. The hottest month is July, with an average temperature of 26.2 • C, and the coldest month is January, with an average temperature of −2.0 • C ( figure 1(b)). The annual precipitation is 759 mm; the maximum precipitation in July is 233 mm, and the minimum precipitation in January is 8 mm ( figure 1(b)). In July 2008, 56 cores were collected from 28 trees at one site (MS1), located at 35.562 • N, 117.848 • E, with an altitude of 1100 m. In November 2020, 52 cores were collected from 26 trees at another site (MS2) located at 35.559 • N, 117.846 • E, with an altitude of 1048 m.

Tree-ring width chorology
According to dendrochronological methods, all cores were visually dated and then the tree rings were measured using the LINTAB measurement system with an accuracy of up to 0.01 mm. The quality of crossdating was controlled by the COFECHA program (Holmes 1983); this program was also used to exclude possible false or missing rings. After excluding the fractured cores and cores with an uncertain calendar year after cross-dating, 53 cores extracted from 27 trees of MS1 and 50 cores from 25 trees of MS2 were finally used to establish tree-ring chronology. As the correlation coefficient between chronologies of the two sites was significant (r = 0.503, p < 0.001) during their common time span of 1668-2007 and the sampling trees were all P. tabulaeformis, we combined these cores to create the MS group (table S1). The average correlation coefficient between these 103 cores was 0.54, and the mean sensitivity was 0.40. To eliminate age-related trends and retain as many climate signals as possible, the chronology was established using the signal-free method with regional curve standardization, which retained more middleand low-frequency climate signals, in the CRUST program (Melvin and Briffa 2014). This method can minimize the impact of the common forcing signal on the variance over a long time scale (Briffa and Melvin 2011). In the CRUST program, there are three types of chronology: standard (STD) chronology, residual (RES) chronology, and autoregressive (ARS) chronology. Because the STD chronology retained all frequencies and the RES and ARS chronologies were less sensitive to the climate signal of interest, the STD chronology was used for further analysis in this study (figure S1).
In addition, Rbar and the expressed population signal (EPS) were calculated (Cook and Kairuikstis 1990) to evaluate the quality of the STD chronology. Rbar is the correlation coefficient of cores on the detrended ring-width indexes used to develop the chronology, and the EPS is related to the number of replicate series and the value of Rbar. An EPS value greater than 0.85 is considered to indicate a reliable chronology (Wigley et al 1984, Cook andKairiukstis 1990). This EPS threshold was reached during the period from 1668 to 2020, with at least 10 cores contributing to the STD chronology (figure S1).

Meteorological data
The nearest meteorological station to the sampling site is Mengyin Station (35.750 • N, 117.983 • E; 202.9 m a.s.l). Monthly precipitation, SAT and GST were the climatic variables used in this study. To verify the reliability of the observed data at Mengyin from the National Meteorological Science Data Center (http://data.cma.cn/), we compared it with that of the nearby Pingyi Station (35.467 • N, 117.617 • E; 182.8 m a.s.l). From 1959 to 2015, the correlation coefficients between the values of the variables at the two stations were as follows: precipitation was 0.855, SAT was 0.989, and GST was 0.914. In addition, the 1.88 • × 1.88 • grid GST data from the National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) (Kalnay et al 1996) and the 0.5 • × 0.5 • grid meteorological data from the Climatic Research Unit (CRU) (Harris et al 2020) were also used to verify the observed data at Mengyin. Because the GST grid (118.12 • E, 35.24 • N) covers a wider area than the sampling sites and Mengyin, the GST correlation between the NCEP/NCAR and Mengyin was only 0.738 during 1959-2015. However, the precipitation correlation between the CRU grid (117.75 • E, 35.75 • N) and Mengyin was 0.808 and the SAT correlation between them was 0.952. These strong correlations indicated that the meteorological data from Mengyin were reliable and could be used for further analysis. The precipitation change was relatively stable and no significant increasing or decreasing trend was observed ( figure 1(c)). The GST values were higher than the SAT values, and both showed an upward trend ( figure 1(d)). The correlation between GST and SAT was 0.897, indicating that GST was related to SAT but also differed from it. The correlation between GST and precipitation was -0.316, indicating that GST decreased when there was more precipitation.

Methods
Pearson correlation analysis was used to explore the relationship between chronology and climate, and linear regressions based on the least squares (LSQs) method, robust method and scaling method (Esper et al 2005) were used to reconstruct past GST. The reliability and stability of the reconstruction models was assessed with the split calibration-verification procedure (Cook and Kairiukstis 1990). The statistical parameters provided for the calibration period were the Pearson correlation coefficient (r) and the explained variance (R 2 ); the fidelity of the calibrations during the verification period was assessed using r, R 2 , the reduction of error (RE), the coefficient of efficiency (CE) and the sign test (ST). For the model to be considered acceptable, the values of RE and CE must be greater than zero (Cook et al 1999). The multitaper method (MTM) of spectral analysis (Mann and Lees 1996) was conducted to detect the periodicity of the reconstructed GST series. The ensemble empirical mode decomposition (EEMD) method (Wu et al 2009), an adaptive and temporally local time-series analysis method designed for analyzing nonlinear and nonstationary climate data, was used to decompose the reconstructed GST series. Spatial correlation analyses were adopted to represent the regional-scale variability in the climate signal by using the KNMI Climate Explorer (https://climexp.knmi.nl/start.cgi).

Tree-ring climate response
Since the growth of trees is not only related to the climate of the current year but also affected by the climate of the previous year (Fritts 1976), the season in which the climate response of tree rings were analyzed lasted from March of the previous year to October of the current year, that is, from the previous spring to the current autumn (figure 2(a)). The correlation between tree-ring chronology and precipitation was significant in only April of the previous year (r = −0.377, p < 0.01). This result indicated that precipitation had little negative impact (table S2) on the radial growth of P. tabulaeformis on MSM where precipitation is abundant and the annual amount is approximately 750 mm. P. tabulaeformis is a photophilous species. High precipitation is usually accompanied by increased cloud cover which reduces the solar radiation flux (Fritts 1976) and the solar radiation is favorable for leaf photosynthesis. In addition, high precipitation may also saturate the soil, contributing to low aeration and low root growth. The correlation pattern between chronology and SAT was similar to that between chronology and GST (table S2; figure 2(a)). These correlations were the most significant (coefficients greater than 0.400, p < 0.01) mainly in March-May and July of the previous year and February-April of the current year.
After combining the monthly meteorological data for all seasons, the strongest correlation between treering chronology and SAT was from March to May of the previous year (r = 0.567, p < 0.001), while the strongest correlation between tree-ring chronology and GST was from March to July of the previous year (r = 0.641, p < 0.001). These results are similar to the seasonal signal contained by the P. taiwanensis in the Dabie Mountain (Cai et al 2018), which mainly reflected the local temperature from April to June of the previous year. Both monthly and seasonal correlations showed that the temperature in the pre-and early growing season of the previous year and the pregrowing season of the current year had a substantial influence on tree growth. The temperature rise in the pregrowing season leads to the advance of growing season, and higher temperatures result in snow melt, thus providing water for trees and promoting growth. The significant relationship between tree-ring width and the temperature of previous growing season may represent a legacy effect caused by carbohydrate accumulation in warm (cold) years that led to growth increase (decrease) in the subsequent year (Campioli et al 2011). High temperatures during the previous growing season can help trees produce and accumulate more photosynthate before dormancy in winter; this photosynthate can be partly utilized to form the earlywood in the following growing seasons (McCarroll andLoader 2004, Ols et al 2016).

Transfer function and GST reconstruction
According to climate response results, the tree-ring width mainly reflected the change in GST from May to July (GST MJ ) of the previous year and their linear relationship ( figure 2(b)). Therefore, three transfer functions based on the LSQs, robust and scaling methods were constructed during 1959-2015. By comparison and split test (figure S2; tables 1, S3 and S4), the LSQs method was suitable and could be used. The transfer function based on LSQ was: GST MJ = 1.57 × W t+1 + 20.36 (n = 57, r = 0.641, R 2 = 0.411, R 2 adj = 0.400, F = 38.40, p < 0.001). In this equation, GST MJ is the temperature of the current year and W t+1 is the treering width index of the next year. For example, the GST MJ in 2000 corresponds to the tree-ring width index in 2001. The variance explained by reconstruction of the observed temperatures was 41.1%, and the root mean square error (RMSE) was 0.882. The RMSE was calculated using the formula: where the obsGST and recGST were the observed and reconstructed GST MJ during 1959-2015. The results of the split calibration-verification test showed that our reconstruction was stable and reliable (table 1). The reconstruction closely tracked the observed GST ( figure 2(c)). In addition, the correlation between the first-order differences of the reconstructed and observed GST was significant (r = 0.414, p < 0.005). Based on this transfer function and the length of the tree-ring chronology, we reconstructed the GST MJ change at MSM from 1667 to 2019 (figure 2(d)).   (Kalnay et al 1996), also indicates our GST reconstruction was reliable and could reflect past temperature change.

GST changes over the past 353 years
During the period of 1667-2019, the mean value of GST MJ was 27.77 • C and the standard deviation was 0.64 • C. We defined years when the GST values were greater (less) than the mean value plus two standard deviations as extremely high-(low-) temperature years, years when the GST values were greater (less)  (Kalnay et al 1996). than the mean value plus one standard deviation as high-(low-) temperature years, and the rest of the years as normal years. We found 3 extremely lowtemperature years, 15 extremely high-temperature years, 42 low-temperature years, 37 high-temperature years and 256 normal years. High-and extremely high-temperature years mainly occurred in the first half of the 19th century, the end of the 20th century and the beginning of the 21st century ( figure 2(d)). Specifically, 73.3% of the extremely high-temperature years occurred at the end of the 20th century and the beginning of the 21st century (table S5) The MTM showed that the reconstructed GST was characterized by 2.1-, 3.4-and 7.9-year interannual periods, 12-and 26 year interdecadal periods and 60 year multidecadal period (figure S3). To better understand the GST changes, we conducted EEMD on the reconstructed series and obtained change characteristics at interannual, interdecadal, multi-decadal and centennial time scales and the trend ( figure S4). The proportions of total variance in these five components were 15.5%, 27.5%, 27.3%, 16.9% and 12.8%, respectively. The main periods of GST changes at the four different time scales were 3.4 years, 12 years, 51 years and 145 years, respectively. From the perspective of the interannual and interdecadal scales, the temperature in this region fluctuated greatly from the middle of the 18th century to the end of the 19th century, while the temperature fluctuation was small in other periods. On the centennial scale, warming began during 1940-1950, which corresponds to the beginning of the Anthropocene (Subramanian 2019).

Representativeness of the reconstructed GST series
The spatial correlation pattern (figures 4(a) and (b)) between the observed GST MJ and gridded GST MJ from NCEP/NCAR was similar to that between the reconstructed GST MJ and gridded GST MJ (figures 4(c) and (d)). The little differences between the two spatial correlation patterns were because that the reconstructed GST MJ explained 41.1% of the variance of the observed GST MJ . This result indicates that the reconstruction represented the GST change in central EC and reflected the temperature changes in northern and southern EC to a certain extent. Moreover, at a longer time period , the spatial correlation between our reconstruction and the annual GST also demonstrated that the reconstructed GST was spatially representative (figures 4(e) and (f)). Due to the different climate, latitude and land cover of different regions, there were some differences in spatial correlation coefficients in EC. Based on these results, we assumed that the reconstruction generally represented the past temperature change in EC.

Possible driving mechanism for GST changes
According to the results of MTM and EEMD, the GST reconstruction over the past 353 years mainly contained the interannual, interdecadal and multidecadal periods. The 2.1, 3.4 and 7.9 year interannual periods fell within the range of variability of El Niño-Southern Oscillation (ENSO) (Diaz and Markgraf 2000), indicating that GST in central EC was likely affected by ENSO. The 26 year interdecadal and the 51 and 60 year multidecadal periods might be related to the Pacific Decadal Oscillation (PDO), which has a period of 20-30 years (Mantua et al 1997) and a main pentadecadal oscillation period of 50-70 years (Minobe 1997, MacDonald andCase 2005). The 12 year period corresponded to the 11 year period of solar activity (Lean 2000). These results might suggest ENSO, PDO and solar activity have certain influences on GST changes in central EC. Of course, the specific relationships among them need further study and analysis.

Temporal and spatial differences in temperature changes in EC
To test the assumption that the GST reconstruction generally represented the past temperature change in EC, where long temperature reconstructions covering the past 350 years were relatively rare (He et al 2019), we compared the reconstructed GST from the GXAM in northern EC (Zhang et al 2018), our reconstructed GST at MSM in central EC and the reconstructed SAT from Taiwan (TW) in southern EC  over the past 350 years (figure 5). The temperature in the three regions began to rise in approximately 1950; the warmest period was at the end of the 20th century and the beginning of the 21st century. The relationships among the temperatures in the three regions before and after 1950 were quite different. Before 1950, the temperatures were not correlated, but after 1950, they were significantly correlated (table S6). In addition, the gridded series of the annual temperature among the three regions were significantly correlated during 1948-2019. Specifically, the correlation between the GXAM GST and MSM GST was 0.553, the correlation between the TW SAT and GXAM GST was 0.506, and the correlation between the MSM GST and the TW SAT was 0.716.
The 51-year running correlation also showed that the temperatures in the three regions of EC exhibited significant correlations after entering the Anthropocene ( figure 5(d)). However, before the Anthropocene, these correlations were nonsignificant. In addition, there were significant negative correlations among these three EC regions at the end of the 17th century, the middle of the 18th century and the beginning of the 20th century. These results imply that the temperatures in different EC regions changed synchronously in the Anthropocene and that prior to the Anthropocene, there were significant regional differences in temperature fluctuations at the multidecadal-centennial time scales. Before the Anthropocene, the temperature in southern EC showed an upward trend; the temperature in central EC first rose and then began to decline in the 1820s, and the temperature in northern EC showed no trend change. These results indicate that our reconstructed GST did not reflect the temperature changes in EC before the Anthropocene. These results also indirectly demonstrated that significant spatial correlations are necessary but not sufficient to establish the spatial representativeness of a reconstruction sequence.
In addition, gridded summer temperature reconstructions (Cook et al 2013) from tree rings were used to analyze the spatial distribution of temperature changes every half century. For each gridded temperature dataset, anomalies were determined according to the average value of the reconstruction during 1981-2009. The results showed that the temperature changes in EC were spatially consistent in the Anthropocene; however, before the Anthropocene, the temperature changes exhibited large spatial differences (figure 6).

Forces of changes in temperature pattern before and during the Anthropocene
Moreover, this phenomenon (synchronous temperature rise in the Anthropocene; large temperature differences before the Anthropocene) appears not only in EC but also in Eurasia . For example, the warm period of the 20th century occurred synchronously worldwide, but cold and warm periods, such as the LIA and MCA, occurred asynchronously in different regions (Neukom et al 2019). These dissimilarities were not only due to regional differences in temperature changes, but also different temporal and spatial sensitivities to temperature changes.
Warming in the Anthropocene was mainly due to greenhouse gas emissions from human activities (Wang et al 2021). In the atmosphere, greenhouse gases uniformly diffuse with atmospheric circulation and absorb longwave radiation that directly heats the atmosphere; thus, their warming is highly consistent worldwide (Ge et al 2015). In contrast, the temperature changes prior to the Anthropocene were mainly driven by natural factors, such as external forcing anomalies (solar radiation and volcanic eruptions), and the corresponding internal variability in the climate system, resulting in asynchronous cold and warm conditions across regions.

Conclusions
In this study, the GST change at MSM in central EC during 1667-2019 was reconstructed using the tree-ring width chronology of P. tabulaeformis. The reconstructed GST explained 41.1% of the total variance in the observed GST during 1959GST during -2015 Over the past 353 years, there were 3 extremely low-temperature years, 42 low-temperature years, 37 high-temperature years and 15 extremely hightemperature years. The largest temperature fluctuations were observed from the middle of the 18th century to the end of the 19th century; temperature fluctuations were relatively small in other periods.
By comparing the temperature reconstructions from GXAM in northern EC and TW in southern EC, we found that temperature changes in EC exhibited spatial consistency during the Anthropocene, while before the Anthropocene, these temperature changes exhibited large spatial differences. Specifically, the temperature in southern EC showed an upward trend, the temperature in central EC first rose and then began to decline in the 1820s, and the temperature in northern EC showed no trend change. In addition, our study also indicates that significant spatial correlation is necessary but not sufficient to establish the spatial representativeness of a reconstruction sequence.