South China (22.5°–30°N, 100°–120°E) is becoming vulnerable to the increased frequency of summer drought in recent decades (Liu et al., 2015;Ha et al., 2016). For example, the severe summer drought in 2013 impacted 10 provinces, 85.9 million people, and 8.0 million hectare of crops in South China, and resulted in a direct economic loss of 59.0 billion Yuan (Hou et al., 2014). To minimize the damages, it is necessary to accurately forecast summer dry events, which should be based on a full knowledge of their long-term variation and the underlying driving mechanism. El Niño–Southern Oscillation (ENSO) is an important driver of summer hydroclimate over East Asia (Huang and Wu, 1989;Zhang et al., 1999;Zhai et al., 2015). However, the relationship between preceding ENSO and summer hydroclimate has been found to be unstable in instrumental records.Chang et al. (2000) indicated that the relationship changed around 1977/78 in the southeastern coast area of China.Gao et al. (2006) pointed out that the relationship became weak from the early 1980s in eastern China.Zong et al. (2010) concluded that there were three obvious changes in the last 50 years—in the early 1960s, late 1970s, and late 1980s to early 1990s, respectively. The relationship prior to instrumental records, however, has been seldom discussed. A recent study based on a reconstructed precipi-tation series revealed an insignificant relationship from 1920 to 1940 in the lower reaches of the Yangtze River (Xu et al., 2016). How the relationship changes with time, though, is still unclear in South China.

Both historical documents and physical proxies can record past climate variations. Documentary data have been applied in the quantitative reconstruction of hydroclimate variations in South China (Chinese Academy of Meteorological Sciences, 1981;Zheng et al., 2006). However, documents from different sources often make it difficult to keep the reconstruction objective and uniform throughout the whole series, and the reconstructed hydroclimate series from historical documents needs to be verified by other physical proxies, such as tree rings (Zheng et al., 2014). With the characteristics of annual resolution, high continuity, and wide spatial distribution, tree rings have been widely applied in paleoclimate reconstruction in subtropical China in recent years. Most of the tree-ring width chronologies reveal temperature variations (Shi et al., 2010,2013,2015a;Chen et al., 2012;Duan et al., 2012,2013;Zheng et al., 2012;Cai and Liu, 2013,2016). There are only a few cases of tree-ring width based hydroclimatic reconstructions in Southeast China (Shi et al., 2015b;Chen et al., 2016a).

Several studies have indicated strong summer hydroclimate signals in tree-ring latewood width (LWW) chronologies in North America (Stahle et al., 2009;Griffin et al., 2013) and North China (Liu et al., 2004). However, LWW-based hydroclimatic reconstructions have not been reported in South China until now. Latewood is the later part of the annual growth ring, with dark color and heavy density, which forms in the mid–late growing season, during which it may record climatic information (Larson, 1969). Because LWW can be preconditioned by the adjacent earlywood width (EWW),Meko and Baisan (2001) succeeded in enhancing its climate signals in the North American monsoon region by applying an adjusted LWW (LWW_adj) to remove the systematic dependence of LWW on EWW.

Old forests of Tsuga longibracteata are distributed at subalpine sites in subtropical China (Chen et al., 2015). The sharp anatomical change from earlywood to latewood (Ho et al., 1984) makes it possible to create ring-width chronologies for both latewood and earlywood, separately. In the present study, for the first time, we produced a summer hydroclimate reconstruction based on the LWW_adj of T. longibracteata in South China, and explored its linkage with ENSO, over the past 127 years.

The sampling site, Qishanzhai (QSZ) (26°14′N, 110°14′E), is located at the boundary between Hunan and Guangxi, with an altitude of 1030 m above sea level (a.s.l.). The sampled T. longibracteata trees grow along a southeast facing hillside, which is well-drained due to the steep slope of nearly 60° and thin soil layer (Fig. 1). Annual mean temperature and annual total precipitation are 16.8°C and 1371.0 mm, respectively, as calculated from instrumental data of Wugang meteorological station over the period 1953–2014. July and August are the hottest two months, with only 155.0-mm precipitation in total, and become the relatively dry period of the growing season (Fig. 2).

Fig. 1 Overview of the sampling site. (a) Location of the sampling site, QSZ, and Wugang meteorological station; (b) geomorphology; and (c)T. longibracteata trees, indicated by arrows.

Figure options

Fig. 2 Monthly mean temperature, monthly total precipitation, and monthly relative humidity at Wugang meteorological station (1953–2014).

Figure options

Two increment cores were taken from each tree with an increment borer of 5 mm in diameter, and 28 cores from 14 trees were obtained in total. The cores were dried, mounted, polished, and cross-dated according to standard dendrochronological procedures (Stokes and Smiley, 1996). Given the well-defined boundary between latewood and earlywood (Fig. 3), we measured LWW and EWW separately with a tree-ring width measurement system LINTAB 5 at a resolution of 0.001 mm, following the protocol outlined by Stahle et al. (2009). The quality of the cross-dating and measured data were checked with the software COFECHA (Holmes, 1983), and two cores were eliminated because of their low correlation with the master chronology. The software ARSTAN (Cook, 1985) was used to create earlywood and latewood chronologies (Fig. 4a). All width series were detrended by fitted cubic smoothing spline curves with a 50% frequency response cutoff equal to 67% of the series length to remove age-related growth trends. Residuals between raw width and fitted curves were calculated as the ring-width indices. All the detrended series were averaged to form a standard chronology by using a bi-weight robust mean to minimize the influence of outliers. Variances of the chronology were stabilized to reduce the effects caused by sample size (Osborn et al., 1997). An SSS (subsample signal strength) of larger than 0.8 was employed to determine the most reliable period (Wigley et al., 1984) and, as such, the reliable periods for the LWW and EWW chronologies were established to be 1888–2014 and 1887–2014, respectively. Finally, the LWW_adj chronology was constructed based on the common reliable period (1888–2014) of the two chronologies (Fig. 4c). Specifically, the LWW was regressed on the EWW by using a simple linear regression, and the residuals plus a constant 1.0 were defined as the LWW_adj chronology (Meko and Baisan, 2001).

Fig. 3 Photo of T. longibracteata tree rings from QSZ. EW and LW denote earlywood and latewood, respectively.

Figure options

Fig. 4 (a) Standard chronologies of latewood width (black line) and earlywood width (gray line), along with the (b) sample depth and (c) LWWadj (adjusted latewood width) chronology.

Figure options

The closest station to the sampling site is Wugang meteorological station (26°44′N, 110°38′E; 340.1 m a.s.l.; 1953–2014). Monthly maximum, minimum, and mean temperatures and monthly total precipitation were obtained from the National Meteorological Information Center ( http://data.cma.cn). Monthly maximum and mini-mum temperatures and monthly total precipitation were used to calculate the standardized precipitation–evapotranspiration index (SPEI) on one- and two-month scales by using the SPEI R package (Vicente-Serrano et al., 2010). Potential evapotranspiration was computed by using the method of Hargreaves (1994). Furthermore, the SPEI of the grid (26°15′N, 110°15′E; 1901–2014) from SPEIbase version2.4 ( http://sac.csic.es/spei/database.html) was selected for comparison with the reconstructed SPEI_Jul–Aug.

The ENSO intensity was represented by the averaged anomalies of Niño3.4 sea surface temperature (SST) from the extended reconstructed sea surface temperature (ERSST) datasets version 4 ( http://www.ncdc.noaa.gov/data-access/marineocean-data/extended-reconstructed-sea-surface-temperature-ersst-v4). The spatial range of the western Pacific subtropical high (WPSH)—an important weather system that directly influences the summer hydroclimate in China—was defined by the 5860- and 5880-gpm contours at the geopotential height of 500 hPa using the NCEP–NCAR reanalysis dataset ( http://www.esrl.noaa.gov/psd/data/gridded/data.ncep.reana-lysis.pressure.html;Tao and Wei, 2006).

The relationship between LWW_adj and climate factors was explored by Pearson's correlation. A linear regression model was employed to reconstruct the dominant climate factor, and a split calibration–verification procedure was used to examine the model fidelity (Meko and Graybill, 1995). The spatial representativeness of the reconstructed SPEI was investigated by the Royal Netherlands Meteorological Institute Climate Explorer ( http://climexp.knmi.nl). Significant periodicities of the reconstruction were calculated by power-spectrum analysis (Mann and Lees, 1996). The time-dependent impact of ENSO on the reconstructed SPEI was explored by using 21-yr moving correlation.

As shown in Fig. 5, LWW_adj is dominated by July–August climate factors of the growing year. LWW_adj is significantly negatively correlated with temperature, yet positively with precipitation, during these two months. The most significant relationship is between LWW_adj and July–August SPEI (SPEI_Jul–Aug), with a correlation coefficient of 0.63 (p < 0.001). In the study area, July and August are the relatively dry months of the growing season, which may cause severe water stress on latewood growth of T. longibracteata, and a cool and wet condition is beneficial to the incremental growth of latewood. Such a relationship has also been found in other studies (Liu et al., 2004;Griffin et al., 2013).

Fig. 5 Correlation between LWWadj and (a) monthly mean tempera-ture, (b) monthly total precipitation, and (c) monthly SPEI. The dashed lines denote the 95% confidence level. SPEIJul–Aug is calculated on a two-month scale.

Figure options

Considering that SPEI_Jul–Aug is the strongest limiting factor on LWW_adj, a linear regression model was used to estimate SPEI_Jul–Aug. The model is as follows:

Generally, the reconstructed series exhibits a similar pattern to the actual values (Fig. 6a). The extreme year 1994 is not well reconstructed. If 1994 is excluded from the reconstruction equation, the explained variance of the model will increase to 0.47. However, we still reserve it because the calibration with only moderate values may underestimate the chronology variability (Fang et al., 2012). The split calibration–verification results are shown in Table 1. The correlation coefficients between actual and reconstructed data (r) and the results of sign-tests all reach the significance level of 0.05, and the reduction of error (RE) and coefficient of efficiency (CE) are positive, indicating the validity of the regression model (Fritts, 1976).

Fig. 6 (a) Comparison between the reconstructed (black line) and actual (gray line) SPEIJul–Aug. (b) The reconstructed SPEIJul–Aug (black line) with its 10-yr low-pass filter (gray line). The dashed line represents the mean value of the reconstructed SPEIJul–Aug. The dotted lines represent mean ± 1.5 standard deviations. (c) SPEIJul–Aug of the grid (26°15′N, 111°15′E) from SPEIbase version 2.4 (black line) with its 10-yr low-pass filter (gray line).

Figure options

Defined by the mean ± 1.5 standard deviations of the reconstructed SPEI_Jul–Aug, extremely wet summers were found in 1915, 1923, 1931, 1941, 1948, 1955, 1970, and 1973, while extremely dry summers occurred in 1900, 1911, 1926, 1928, 1934, 1935, 1957, 1972, and 2013. Most of the extremely dry and wet summer events are consistent with descriptions from historical documents (Table 2). According to the rank of extreme years, the severe drought in summer 2013 (Hou et al., 2014) is not unusual because more extreme drought events occurred in the first half of the 20th century, such as in 1926, 1934, and 1900.

The reconstructed series is quite similar to the adjacent grid SPEI (Fig. 6c), with a correlation coefficient of 0.48 (p < 0.01) over the period 1901–2014, validating the reliability of our reconstruction. Spatial correlation analysis shows that the reconstructed SPEI_Jul–Aug represents a large region of South China, including northern Guangxi, Hunan, and Guizhou (Fig. 7).

Fig. 7 Spatial correlation between (a) actual and (b) reconstructed SPEIJul–Aug and SPEIbase grids from 1953 to 2013. The tree symbol repre-sents the sampling site.

Figure options

Power spectrum analysis shows 2.0–3.6-yr periodici-ties in our reconstruction that fall within the range of ENSO cycles (Fig. 8;D'Arrigo et al., 2005), indicating that ENSO might have some impacts on the SPEI_Jul–Aug variations in the study area. ENSO affects the summer rainfall over East Asia by changing the strength of WPSH (Huang and Wu, 1989;Zou and Ni, 1998;Zhang et al., 1999;Chang et al., 2000;Gong et al., 2002;Cai et al., 2003;He et al., 2015). During the mature phase of El Niño events, the warm SST anomalies (SSTAs) in the central–eastern Pacific strengthen the WPSH via the overturning Walker circulation (Ying and Sun, 2000). Meanwhile, the cold SSTAs in the West Pacific enhance the WPSH by forcing an anomalous anticyclone (Wang et al., 2000). Even in the decaying summer of El Niño, the WPSH can still be indirectly enhanced through the so called “capacitor effect” of the tropical Indian Ocean (TIO), which prolongs the impact of ENSO on the WPSH (Xie et al., 2008;Chen et al., 2016b). The anomalous WPSH would shift its northwest rain belt over East Asia, changing the spatiotemporal distribution of precipitation in China (Tao and Wei, 2006). Therefore, autumn–winter ENSO indices have important implications for the WPSH's strength and the hydroclimate of the next summer in South China (Cai et al., 2003;Gao et al., 2006).

Fig. 8 Periodicities of the reconstructed SPEIJul–Aug calculated by power spectrum analysis.

Figure options

Whether the relationship between SPEI_Jul–Aug and ENSO is time-dependent is evaluated by 21-yr moving correlation analysis (Fig. 9). It was found that the two indices were in weak association before the 1950s. This phenomenon could be due partly to the small variance of ENSO in 1920–1960 (Torrence and Webster, 1999;Li et al., 2013;Xu et al., 2016). In addition, the “capacitor effect” of the TIO has been reported to be weak during the period 1910–1970 (Chowdary et al., 2012), which may also contribute to the insensitivity of the SPEI_Jul–Aug response to ENSO. From the 1950s to early 1990s, the correlations became significantly positive, indicating that there was a strong relationship between SPEI_Jul–Aug and ENSO. After that, the relationship dropped to be weak and even out of phase.

Fig. 9 (a) Comparison of the reconstructed SPEIJul–Aug (black) and Niño3.4 SSTAs (SST anomalies) averaged from the previous September to current February (gray). (b) 21-yr moving correlations between SPEIIJul–Aug and Niño3.4 SSTAs.

Figure options

The weakened relationship in the early 1990s may have been directly caused by a significant westward extension of the July–August WPSH (WPSH_Jul–Aug) on interdecadal timescales. Obvious differences were found between the distributions of WPSH_Jul–Aug before and after 1992. From 1969 to 1991, the main body of the WPSH_Jul–Aug, indicated by 5860-gpm contours, was lo-cated in the east of the study region, whether in dry or wet years (Figs. 10a and 10b). When preceding SSTs became warm in ENSO-related regions, the WPSH_Jul–Aug would be enhanced and its northwest edge extended to the study area (Fig. 10b). The northwest edge of the WPSH is characterized by a low-level jet carrying large amounts of water vapor (Zhou and Yu, 2005), which brings the study region more precipitation. Therefore, the SPEI_Jul–Aug and preceding Niño3.4 SSTs were in phase before 1992. However, after 1992, the northwest edge of the WPSH_Jul–Aug passed completely through the study area when preceding SSTs were warmer (Fig. 10c), which could have created continuous stable downdraft over the study region and made the regional hydroclimate dry. Only at the stage of cooler preceding SSTs could the WPSH_Jul–Aug draw its northwest edge back to the study region (Fig. 10d), and the regional climate tended to be wet. Thus, variations of SPEI_Jul–Aug and prece-ding Niño3.4 SSTs became out of phase after 1992. The interdecadal westward extension of the WPSH has been reported to be caused by the ocean warming in the early 1990s—particularly the increasd SSTs in the western Pacific, which induced the expansion of the atmospheric column and an increase in geopotential height (Gao et al., 2014).

Fig. 10 Composite July–August 500-hPa geopotential height and preceding SSTs for (a) dry and (b) wet years during 1969–1991, and (c) dry and (d) wet years during 1992–2014. Wet and dry years are derived from the reconstructed SPEIJul–Aug arithmetic mean over the period 1969–2014. Preceding SSTs are from ERSST version 4, averaged from the previous September to current Feburary. The solid line, dashed line, rectangle, and tree symbol represent 5860 gpm, 5880 gpm, Niño3.4 region, and the QSZ sampling site, respectively.

Figure options

We reconstructed the SPEI_Jul–Aug series from 1888 to 2014, which, to the best of the authors' knowledge, is the first paleoclimate reconstruction by using an adjusted latewood width chronology in South China. The reconstructed series is representative of large-scale hydroclimate change in South China. From the perspective of the past 127 years, the extreme summer drought in 2013 was not unprecedented because more extreme events occurred in the first half of the 20th century. The 2.0–3.6-yr periodicities of the reconstruction indicated connections between SPEI_Jul–Aug and ENSO. However, the relationship was unstable during the whole reconstruction period. It was weak before the 1950s, became significant from the 1950s to early 1990s, and then dropped to be weak again and even out of phase from the early 1990s, which may be attributable to the significant westward extension of the WPSH. This study indicates that summer hydroclimate in South China can be reconstructed based on adjusted latewood width, and will be better understood when more and longer adjusted latewood width chronologies are obtained in the near future.

Acknowledgments . We thank Yanwu Shi, Zhou Yu, and Xinyuan Hou for their assistance in the field; Dr. Jinbao Li and Prof. Jan Bloemendal for improving the language; and the two anonymous reviewers for their constructive comments.