The study area is located at Laobai Mountain (128∘03′ E, 44∘06′ N), the boundary zone between Jilin and Heilongjiang provinces, and is also an ecotone between Changbai and Xiaoxing'an Mountain. Laobai Mountain is the third highest peak in northeast China and rises to 1650 m above sea level (a.s.l.). Almost no inhabitants live on or near the mountain, so the forest ecosystem is very well preserved and the native vegetation remains predominantly intact (Fig. 1). Five forest vegetation types from temperate to frigid change with the altitude increase, which is the broad-leaved Quercus mongolica forest below 800 m a.s.l., the mixed broadleaved Korean pine forest from 800 to 1050 m, dark conifer forest with Picea jezoensis from 1050 to 1350 m, Betula ermanii forest between 1350 and 1640 m, and Pinus pumila forest and subalpine meadow above 1640 m. Plant flora transitions from Changbai Mountain to Xiaoxing'an Mountain. Five tree species were cored in this area, but only Korean pine (Pinus koraiensis) cores were used in this study. Korean pine is a sun-loving plant (shade-tolerant when it is young) and has shallow roots, widely distributed on well-drained wet mountain slopes close to the subalpine timberline where the brown forest soil is covered. The forest vegetation in sampling area is the mixed broadleaved Korean pine forest dominated by Pinus koraiensis, Picea jezoensis, and Abies nephrolepis as well as broadleaf tree species, such as Juglans mandshurica, Fraxinus mandshurica, and Acer mono (Bu et al., 2003).

This region belongs to a temperate continental monsoon climate. Climate data are collected from the nearest meteorological station in Dunhua. The mean annual temperature from 1956 to 2013 is 3.3 ∘C, with July (20.1 ∘C) and January (-16.8 ∘C) being the warmest and the coldest month, respectively. The mean monthly minimum and maximum temperatures are -2.5 and 9.8 ∘C, respectively. The mean annual total precipitation is 627 mm, the majority (63.1 %) of which falls during June–August. The annual frost-free period is approximately 90–110 days (Fig. 2).

Korean pine tree-ring samples were obtained from the south slope of Laobai Mountain along an elevational gradient from 950 to 1050 m from an almost pristine area containing well-preserved old forests largely uninfluenced by human activity. One or two cores per undamaged tree (71 cores from 41 trees) were extracted from cross-slope sides of the trunks at breast height using an increment borer. Cores were air dried, glued firmly to grooved wooden mounts and sanded with progressively finer grade abrasive paper up to 800 grit. Then the samples were cross-dated using a skeleton plot method (Stokes and Smiley, 1968); each tree-ring width was measured with a precision of 0.001 mm using the Velmex tree-ring width measurement system (Velmex, Inc., Bloomfield, NY, USA). Data were checked for missing or false rings and dating errors using the quality control program COFECHA (Holmes, 1983).

The ARSTAN program was used to detrend and standardize cross-dated tree-ring width series into a tree-ring chronology (Cook, 1985). During this detrending process, to remove biological factors (such as age-related trends) and non-climatic variations and preserve as much low-frequency signal as possible, each ring-width series was fitted with a straight line or negative exponential function. A 67 % cubic smoothing spline with a 50 % cutoff frequency was further used in a few cases when anomalous growth trends occurred. The detrended data from individual tree cores were then averaged using a bi-weight robust mean to develop the standard (std) and residual (RES) chronologies (Cook and Kairiukstis, 1990).

Statistical characteristics for the std and RES chronologies of Pinus koraiensis on Laobai Mountain are shown in Table 1. These statistic characteristics of tree-ring chronologies contained strong climate signals, common growth-limiting signals, and the amount of different frequency information. As shown in Fig. 3, the amplitude of std chronology (Fig. 3a) in low-frequency variability was larger than that in RES chronology (Fig. 3c). This indicated that std chronology preserved more low-frequency signals, while RES chronology reflected high-frequency signals. The mean sensitivity of RES chronology was larger than std chronology, which quantitatively illustrated that RES chronology exhibited more high-frequency climate information than std chronology did (Table 1). The full length of tree-ring series spanned from 1600 to 2015. The expressed population signal (EPS) was used to assess the quality of std chronology (Wigley et al., 1984). A generally acceptable threshold of the EPS was consistently greater than 0.85 from AD 1660 to 2015 (11 trees; Fig. 3b), which affirmed that this is a reliable period. However, although the EPS value from AD 1600 to 1659 was less than 0.85, it matches a minimum sample depth of six trees in this segment. It is very important to extend the reconstruction tree-ring chronology as possible as we could because of a few long climate reconstructions in this area. Therefore, we kept this part from 1600 to 1659 to be used in the reconstruction. In addition, the std chronology was used in the subsequent analyses to obtain more low-frequency signals.

Meteorological data were obtained from the National Meteorological Information Center (http://data.cma.cn/). Considering the proximity to sampling sites and climate record length, climate data from Dunhua meteorological station (43∘22′ N, 128∘12′ E; elevation 524.9 m a.s.l.; 1956–2013) were selected to identify climate signals in the tree-ring series. The climate variable included monthly total precipitation, mean maximum temperature (Tmax), mean temperature (Tmean), and mean minimum temperature (Tmin). Months from the previous July to current August were selected for the analysis of the relationship between climate variables and Korean pine growth.

To identify climate–growth relationships of Korean pine on Laobai Mountain, a Pearson's correlation was performed between climate variables and tree growth. The stability and reliability of the reconstruction equation was assessed by the split-period calibration and verification analyses (Fritts, 1976; Cook and Kairiukstis, 1990) for the two periods 1956–1984 and 1985–2013. The Pearson's correlation coefficient (r), R square (R2), sign test (ST), the reduction of the error (RE), the coefficient of efficiency (CE), and the product means test (PMT) are the tools used to verify the results. All statistical analyses presented in this paper were performed using a commercial software, SPSS12.0 (SPSS, Inc., Chicago, IL, USA).

Relationships between the std and RES chronologies and monthly climate data in Dunhua were shown in Fig. 4. Temperatures were more crucial to Korean pine growth compared with precipitation. In contrast, the correlation coefficients between Korean pine chronologies and mean minimum temperature were positive and higher than those for maximum and mean temperature. The significant correlation months between std chronology (Fig. 4a) and mean minimum temperature were not found in the RES chronology (Fig. 4b). This indicated that the std chronology recorded the minimum temperature signals in low frequency but not at high frequencies. Different combinations of months were also considered (Table 2). The best-correlated 3-month season, April–July (r= 0.757, p < 0.0001), was then selected for temperature reconstruction of the mean minimum temperature (MMT; Table 2).

It was generally accepted that extreme temperatures limited tree growth at the tree line or at high-latitude forests, especially spring or early summer minimum temperature (Wilson and Luckman, 2002; Körner and Paulsen, 2004; Porter et al., 2013; Yin et at., 2015). Moreover, Tmax, Tmean, and Tmin during the observed period of 1956–2013 illustrated similar interannual variations (Fig. 5), while the increasing trend in Tmin was much higher than Tmean and Tmax, especially since 1976. This phenomenon was consistent with the results in Karl et al. (1993), Ren et al. (1998), and Tang et al. (2005). They indicated that climate warming over past decades was mostly owing to the faster rise of night or minimum temperatures. This seemed to be the case in northeast China as well. Based on the relationship between std chronology and climate data, we found that the minimum April–July temperature played more important roles in limiting Korean pine radial growth on Laobai Mountain compared with the maximum and mean temperatures. This also meant that less warm and wet conditions in this area were suitable for Korean pine growth.

This may have two reasons. One, the sampling site was located at high elevations close to the upper limit of the Korean pine distribution, which may cause tree growth to be more sensitive to minimum temperature (Szeicz and MacDonald, 1995; D'Arrigo et al., 2009; Li et al., 2011; Yu et al., 2011; Flower and Smith, 2012). High minimum temperatures in the early growing season can inhibit frost damage and thus allow the formation of a wider ring (Wu, 1990; Akkemik, 2000; Mäkinen et al., 2003). High nighttime temperatures can also promote tree respiration and enhance physiological activities, thereby producing more auxin, promoting cell enlargement and forming a wider ring in growing season (Fritts et al., 1976). Increasing temperature may allow trees to conduct photosynthesis at the early stage of the growing season, which might produce more auxin. A crucial growth period of Korean pine in every year was from April to July. During this period, temperature could have direct effects on the photosynthesis rate, cambium activity, and respiration efficiency, etc., which affect the formation of ring width (Li et al., 2000; Yu et al., 2011). Therefore, Korean pine radial growth was positively correlated with the average minimum temperature from April to July.

Based on the above analysis, a linear regression equation was established to reconstruct the April–July MMT. The transfer function was as follows: Y=2.987Xt+4.829(N=58,R=0.757R2=0.573,Radj2=0.565,F=75.161,p< 0.0001), where Y is the April–July MMT and X is the tree-ring index of the Korean pine std chronology at year t. As shown in Fig. 6a, the reconstructed values closely tracked the observed temperature. The calibration and verification statistics were shown in Table 3. Parameters of the split-sample validation periods indicated that the reconstruction equation was stable over the whole period. Positive RE and CE values revealed a useful paleoclimatic information (Cook et al., 1999). Significant results of std and PMT indicated a good agreement between the actual and reconstructed data. However, the first difference correlation (not shown) between the std chronology and temperature did not exceed the 95 % confidence level. This confirmed that the std chronology might capture the low-frequency variability better than high-frequency variability. In addition, the correlation coefficient between the first-order difference series of the actual and reconstructed values was not significant at the 0.05 level (r= 0.12, p > 0.05); hence, this reconstructed minimum temperature series was more consistent with the observed series at low-frequency variability. The final calibration equation accounted for 57.3 % of the total variance (p < 0.0001) and passed all calibration and verification statistical requirements. Hence, this equation was reliable and allowed for the accurate reconstruction of the April–July MMT on Laobai Mountain.

The reconstructed average April–July MMT variations since AD 1600 and its 11-year moving average were shown in Fig. 6b. The 11-year moving average of the reconstructed series was used to obtain low-frequency information and analyze temperature variability in this region. The mean value of the 414-year reconstructed temperature was 7.66 ∘C, with a standard deviation of ±0.53 ∘C. The warm and cold periods were defined when temperatures exceeded the mean value plus and minus 0.5 times standard deviation, respectively (Fig. 6b). The reconstructed April–July MMT series exhibited six cold and seven warm periods. The longest cold period lasted from AD 1645 to 1677 (33 years), with an average temperature of 0.5 ∘C below the mean value. The longest warm period, however, lasted from AD 1767 to 1785 (19 years), and the average temperature was 0.69 ∘C above the mean value (Table 4). Four cold (1605–1616, 1645–1677, 1911–1924, and 1951–1969) and warm (1795–1807, 1838–1848, 1856–1873, and 1991–2008) periods were consistent with other results of tree-ring reconstructions in northeast China (Shao and Wu, 1997; Yin et al., 2009; Wang et al., 2012; Zhu et al., 2015). In addition, two cold periods (1645–1677 and 1684–1691) were consistent with the Maunder Minimum (1645–1715), an interval of decreased solar irradiance (Bard et al., 2000). The cold period 1645–1677 also appeared in other proxy records of reconstructed temperatures, which coincided with the Little Ice Age (LIA) in the Northern Hemisphere (Lin et al., 2004; Wang et al., 2006; Hong et al., 2009). Cold conditions during the 17th century and the rapid warming during the mid-19th and late 20th century in northeast China was present in this series, suggesting it could be a good proxy for regional temperature variations in northeast China.

To further evaluate the reliability of this reconstruction, we compared our reconstruction series with two nearby tree-ring-based reconstruction temperature series from Dunhua (Li and Wang, 2013; Fig. 7a) and Changbai Mountain (Zhu et al., 2009; Fig. 7b) and the Northern Hemisphere temperature reconstruction (D'Arrigo et al., 2006; Fig. 7d). Interestingly, a significant negative correlation (r= -0.18, p < 0.01) between our reconstruction (Fig. 7c) and the Northern Hemisphere temperature reconstruction (D'Arrigo et al., 2006) was found (Fig. 7d), while our reconstruction of April–July MMT had similar variations in the April–September temperature reconstruction in Dunhua (r= 0.50, p < 0.01; Fig. 7a) and the February–April temperature reconstruction on Changbai Mountain (r= 0.45, p < 0.01; Fig. 7b). The three temperature series exhibited significantly low-temperature periods during the 1950s–1970s, which coincided with a slight decrease in solar activity from AD 1940 to 1970 (Beer et al., 2000; Fig. 7).

It was widely believed that the LIA in China exhibited three cold periods in the 15th, 17th, and 19th centuries (Wang et al., 2003), and this was confirmed by our reconstruction series (Fig. 7c and Table 4). The first cold period in our series was less obvious, while the second one was the most obvious of all. A different beginning and ending year of the second cold period in our reconstruction was found (Fig. 7c and Table 4). In addition, there existed a regional difference for the third cold period, that is, it was obvious in south China, while had the opposite phase in northeast China (Wang et al., 1998; Wang et al., 2003). The third cold period in 19th century was not obvious in our reconstruction, which was consistent with Wu (2013) and Wang et al. (1998). This also led to a bad match with the Northern Hemisphere temperature (D'Arrigo et al., 2006). Another notable feature in Fig. 7 was a sharp temperature increase since the 1980s, and temperature rose to a peak in the early 2000s. The temperature increase in this area was consistent with the report from the Intergovernmental Panel on Climate Change (IPCC, 2007). This series displayed similar patterns of low-frequency variations suggesting that the reconstructed temperature in northeast China was significantly correlated with large-scale variations (Fig. 7).

Unfortunately, three compared temperature series also showed dissimilar variations in some cold and warm years (Fig. 7). This might be due to differences in reconstructed temperature months, parameters (such as mean, minimum, and maximum temperature), and habitat conditions in different sampling areas. Recent studies suggested that the mean, minimum, and maximum temperature variations were often asymmetric (Karl et al., 1993; Xie and Cao, 1996; Wilson and Luckman, 2002, 2003; Gou et al., 2008). Global warming over the past decades was mostly owing to the faster rise of night or minimum temperatures but not maximum temperature. The unsynchronized variability among mean, minimum, and maximum temperatures was found at Dunhua meteorological station (Fig. 5). The sampled site was located at a border zone between Jilin and Heilongjiang provinces, further north than Changbai Mountain. Meanwhile, some differences in the reconstructed temperature series were explained reasonably well from the comparison with analogous regions. Consequently, these findings could reveal more characteristics of regional climate variations and provide reliable data for larger-scale climate reconstructions in northeast China.

As the minimum temperature approached or fell below the freezing point, it may have limited biological activity and growth. Therefore, years with low temperatures were often accompanied by cold damage or severe frost events. The evidence from historical documents (Sun et al., 2007) showed that cold damage or frost disaster events have occurred in Heilongjiang Province since 1675 (Table 5). Extremely cold damage or frost disaster events were in good agreement with eight low-temperature years (1675, 1682, 1689, 1699, 1730, 1748, 1812, and 1885 in Fig. 6b and Table 4) in reconstructed April–July MMT series during the 1600s–1800s. At the beginning of 20th century, three severe frost periods occurred in the periods 1909–1918, 1934–1945, and 1956–1972 in Heilongjiang Province (Sun et al., 2007) and were represented in our reconstruction (Fig. 6b and Table 4). In addition, other low-temperature years in our reconstruction corresponded to extreme frost disaster events occurred in the periods 1902–1903, 1912–1914, 1920, 1932, 1934–1936, 1940, 1947, 1956–1961, 1964–1965, 1967, and 1969 (Fig. 6b and Table 4). The results revealed that 27 of the 30 cold damage or frost disaster events corresponded to the April–July MMT lower than the 27-year moving average, while the remaining three events corresponded to higher than April–July MMT values. In contrast, we found a decreasing trend in the annual extreme low-temperature frequency and cold damage or severe frost events with the warming since the 1980s. In summary, the reconstructed April–July MMT on Laobai Mountain strongly revealed the cold damage or frost disaster events in the past 414 years.