Int J Biometeorol DOI 10.1007/s00484-013-0762-8
ORIGINAL PAPER
Spatiotemporal analysis of ground-based woody plant leafing in response to temperature in temperate eastern China Guohua Liu & Qiuhong Tang & Xingcai Liu & Junhu Dai & Xuezhen Zhang & Quansheng Ge & Yin Tang
Received: 25 May 2013 / Revised: 15 October 2013 / Accepted: 29 October 2013 # ISB 2013
Abstract The analysis of woody plant leafing in response to regional-scale temperature variation using ground-based phenology is usually limited by the sparse coverage and missing data of ground observation. In this study, a station-based multispecies method was proposed to generate spatiotemporal variation of woody plant leafing date using ground observations from the Chinese Phenological Observation Network during 1974–1996. The results show that the leafing date had slightly insignificant advance (−0.56 day decade−1), and the Arctic Oscillation (AO) index could explain 36 % variance of the spring leafing date anomaly. The leafing date had been substantially delayed (4 days) when AO shifted from an extreme high index state (2) in 1989–1990 to a relatively low state (0.1) in 1991–1996. The canonical correlation analysis (CCA) was used to demonstrate the temporal evolutions and spatial structures of interannual variations of the spring temperature and leafing date anomalies. The three CCA spatial patterns of leafing date anomaly are similar to those of spring temperature anomaly. The first spatial pattern shows ubiquitous warming, which is consistent with the ubiquitous advance in leafing date across the study area. The second and third spatial patterns present the regional differences featured by advanced (delayed) leafing associated with high (low) temperature. The results suggest that the spring leafing date anomaly is spatiotemporally coherent with the regional-scale temperature variations. Although we focus here on woody Electronic supplementary material The online version of this article (doi:10.1007/s00484-013-0762-8) contains supplementary material, which is available to authorized users. G. Liu : Q. Tang (*) : X. Liu : J. Dai : X. Zhang : Q. Ge : Y. Tang Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, China e-mail: [email protected] G. Liu University of Chinese Academy of Sciences, Beijing, China
plant leafing in a historical period in temperate eastern China, our station-based multispecies method may be applicable to analysis of the ground-based phenology in response to regional-scale climatic variation in other regions. Keywords Leafing . Woody plant . Spring temperature . Canonical correlation analysis . Temperate eastern China
Introduction Study on plant phenology has received increasing attention in the light of global warming (Cleland et al. 2007; Morin et al. 2009; Richardson et al. 2013). Plant leafing (leaf unfolding), one of the most obvious phenological events in spring, is sensitive to climatic factors such as air temperature (Menzel et al. 2005; Schwartz et al. 2006; Chen and Xu 2012a; Polgar and Primack 2013), photoperiod (Partanen et al. 1998; Körner and Basler 2010; Basler and Körner 2012; Vitasse and Basler 2013), and precipitation (Jolly and Running 2004; Zhang et al. 2005; Piao et al. 2006; Crimmins et al. 2010; Bradley et al. 2011; Grogan and Schulze 2012; Zhang et al. 2013). Therefore, plant leafing date, as an observable biological indicator, may be used to study climatic change and variability (Schwartz 1998; Menzel et al. 2006; Piao et al. 2006). In addition, shifts in the timing of plant leafing could alter the length and strength of plant activities and further affect regional and global carbon cycles (Barr et al. 2007; Piao et al. 2007; Cong et al. 2012). Hence a thorough understanding of the plant leafing and its dynamic responses to large-scale climatic changes is important for climatic and ecological research (Menzel 2002; Walther et al. 2002; Schwartz et al. 2006; Morisette et al. 2008; Barnosky et al. 2012). Ground observation and remote sensing are the two widely used but completely different approaches to measuring plant phenology (Badeck et al. 2004; Studer et al. 2005). Ground
Int J Biometeorol
observation comes with a high temporal resolution and detailed information on individual species but poor spatial coverage, whereas satellite data provide wide spatial coverage but low temporal resolution and less detailed species information (Tang et al. 2012). Satellite observations have been used to investigate spatiotemporal variation of vegetation phenology and climate change impacts in temperate China (Piao et al. 2006; Tao et al. 2008; Peng et al. 2010; Piao et al. 2011; Cong et al. 2013; Wu and Liu 2013). The satellite-based studies have shown an advanced trend of spring green-up in response to a warming trend in China (Piao et al. 2006; Ma and Zhou 2012; Cong et al. 2013). Although satellite observation is a powerful tool in monitoring large-scale vegetation phenology, it should be noted that satellite observations do not obtain the same traits as ground observations (Badeck et al. 2004). Ground observations can describe the onset of leafing phenophase at the level of the individual plant. The groundbased leafing dates have been linked with regional climatic changes, particularly in Europe (Peñuelas and Filella 2001; Menzel et al. 2006, 2011; Reyer et al. 2013), North America (Worrall 1999; Schwartz et al. 2006; Wolkovich et al. 2012), and China (Zheng et al. 2006; Dai et al. 2013a; Ma and Zhou 2012; Chen and Xu 2012a). Using ground observations and satellite observations from 1982 to 1993, Chen et al. (2005) showed that spatial patterns of growing season beginning dates correlate significantly with those of spring temperatures in temperate eastern China. Zheng et al. (2006) found that the leafing dates of woody plants were closely correlated with spring temperature and had advanced 1–5 days per decade in 1963–1996 based on ground observations in the region north of 33°N over the eastern China. Ground observations in 1986–2005 have also been used to investigate the phenological responses of specific species to climate change in the temperate zone of China (Chen and Xu 2012a, b). Using the ground-based leafing dates and a process-based model, Wang et al. (2012) demonstrated the spatiotemporal dynamics of spring phenophase changes of Fraxinus chinensis from 1952 to 2007 in China. These analyses are generally siteand species-based and mainly focus on the developmental shifts of individual species. Hence, it is difficult to obtain large-scale phenology information from site-based data due to the sparse coverage and missing data of ground observation (Badeck et al. 2004; Liang et al. 2011). A few studies have been conducted to derive the phenological patterns of some specific species using the ground-based phenology (Rötzer and Chmielewski 2001; Estrella et al. 2009; Dai et al. 2013b; Xu and Chen 2013). These studies require phenological observation of a certain number of specific species at all (or most) observation stations. However, the phenology observation network in temperate eastern China is lacking of many specific species at most stations, and a method which requires specific species across most stations is therefore not suitable.
Moreover, the use of the specific species may reflect only the phenological pattern of these species but not necessarily the pattern of the majority of species. In this study, we proposed a station-based multispecies method which combined leafing anomalies from all available species at each station to generate spatiotemporal variation of woody plant leafing date over temperate eastern China. This method could minimize the data limitation problem of ground-based observations while retaining as much as possible information of the regional-scale phenological variation. Furthermore, for better understanding, the dynamic (spatial and temporal) responses of leafing date to regional-scale temperature variation, the canonical correlation analysis (CCA) was used to extract the canonical spatial structures and temporal evolutions between pairs of variables (i.e., leafing date anomaly and temperature anomaly). The main objective of this study is to investigate the spatiotemporal variations of the ground-based woody plant leafing and to demonstrate the association between the ground-based phenology and regional-scale temperature variations in temperate eastern China from 1974 to 1996. First, we proposed a station-based multispecies method to generate spatiotemporal variations of woody plant leafing date from ground observations. Then the spatiotemporal variations of woody plant leafing in relation to spring air temperature changes are demonstrated.
Materials and methods Study area We chose the temperate eastern China (north to 35°N and east to 110°E in China but excluding Inner Mongolia, Fig. 1) as our study area. Most of the Chinese Phenological Observation Network (CPON) stations with long-term phenological observations during the period of 1974–1996 are located in the study area (Zheng et al. 2002). The three provinces (grey areas in Fig. 1) of the study area lie in Northeast China, and the south of the study area (dark grey areas in Fig. 1) is in North China. The climate over the study area is dominated by the East Asian monsoon with the remarkable seasonality and spatial variability in temperature. The main biomes comprise temperate hemi-boreal, mixed, and deciduous forests (Wu and Liu 2013). The biome dynamics in the study area are strongly influenced by the monsoon climate with a distinct seasonality (Chen et al. 2005). Phenological data The leafing date data of woody plants were taken from CPON, a network of phenological monitoring station administered through the Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences. We selected the species of deciduous woody plant with at least 10-
Int J Biometeorol
monthly temperature data came with a spatial resolution of 0.5°, matching the gridded field of leafing date anomaly. The spring (March, April, and May) temperature anomalies were calculated based on the gridded monthly data. The Arctic Oscillation (AO) index was obtained from the National Oceanographic and Atmospheric Administration (NOAA) Climate Prediction Center (http://www.cpc.noaa.gov/). The mean AO in later winter and spring (from January to May) was used. The linear least-squares regression model was applied to calculate the trend of study area average spring temperature, leafing date, mean AO, and the sensitivity (i.e., the regression slope) of leafing date anomaly to spring temperature change at each grid in the study period. The statistical significance of the trends, correlation, and sensitivity was calculated according to the two-tailed Student’s t test. Fig. 1 The study area and the CPON stations. The light grey area is Northeast China and the dark grey area is North China
year records at each station. We removed the stations where the number of selected species was less than five. We excluded the years when the number of stations was less than eight over the whole study areas and identified the study period of 1974–1996. A total of 5,654 leafing records from 176 species at 19 stations (see Appendix S1 for details) were used during the study period. The leafing date anomalies from the species’ mean leafing date during the study period were calculated for each species at each station. The median and interquartile range of the leafing date anomalies of all the available deciduous woody plant species were computed for each station. The median leafing date anomaly represents the majority of the leafing date anomaly of the deciduous woody plant species at the station, offering an estimate of leafing phenological responses to climatic change at regional scale. It should be noted that the median leafing date anomaly index does not give the difference of mean leafing date between different locations. Rather, it reflects the advance or delay of the leafing date at each location. If one species shows the same sign as the median leafing date anomaly in more than 70 % of the data available years during the study period, we define the specific species as a responsive species which has the same response (in change direction) as the leafing date anomaly of the majority of species at the station. We calculated the number of the responsive species at each station. Using the inverse distance weighted method, we interpolated the median leafing date anomalies at the stations to a 0.5°×0.5° grid over the study area for each year in the study period (Chmielewski and Rötzer 2002).
Climatic data Surface air temperature data were obtained from China Meteorological Administration (Xu et al. 2009). The gridded
Canonical correlation analysis The canonical correlation analysis (CCA) (Hotelling 1936; Barnett and Preisendorfer 1987) is a statistical technique to study the relationship between two sets of variable. It could identify pairs of optimally correlated patterns by analyzing the linear combinations of original variables. The first CCA pair gives the maximum correlation between two sets of variable, followed by the second CCA and so on. We adopted the approach proposed by Barnett and Preisendorfer (1987) to demonstrate the temporal evolutions and spatial structures of interannual variation of the spring temperature and leafing date anomalies from 1974 to 1996 over the study area. Prior to the CCA, the spring temperature and leafing anomaly fields were first approximated by two sets of leading empirical orthogonal functions (EOFs). The use of EOF not only reduces the dimension of original variable but also prefilters the data to eliminate noise and render the resulting modes more stable (Xoplaki et al. 2000). We computed the EOFs for the spring temperature and leafing date anomaly fields. Then we selected the first three EOFs of each data set and used CCA to search for pairs of weight sets such that the correlations between the linear combinations of the temperature EOFs and the leafing EOFs were maximized. A sequence of paired canonical patterns (i.e., the linear combinations of the EOFs) was identified with the highest possible correlation for the first pair and decreased highest correlations for the latter pairs. We used the first three pairs, which had included the primary interrelated information of the spring temperature and leafing anomaly fields (Chmielewski and Rötzer 2002). The temporal coefficient (canonical vectors) series associated with the most important three CCA pairs were analyzed. According to the time series of the three canonical patterns (CCA1, CCA2, and CCA3 modes) for leafing date anomaly, we classified the years in the study period to the canonical patterns. For an individual year, if the maximum absolute value of the three sets of
Int J Biometeorol
Results Observed leafing date
Fig. 2 The numbers of the observed deciduous woody species and responsive species at the 19 selected CPON stations
normalized canonical vector is over 1.0, the associated canonical mode (CCA1, CCA2, or CCA3) with the maximum absolute value was identified. The individual year was classified into the associated canonical mode. The year which was classified into the CCA1 (or CCA2, CCA3) mode is called CCA1 (or CCA2, CCA3) year. The leafing date anomaly in the positive CCA1 year has a spatial pattern similar to the CCA1 mode. The leafing date anomaly in the reversed CCA1 year has a spatial pattern similar to the CCA1 mode but with opposite sign. If the maximum absolute value was less than 1.0, the individual year was taken as an unclassified year. The spatial pattern of the spring temperature and observed leafing date anomalies were shown for the three classifications corresponding to the three canonical patterns. Table 1 The number of the data available years (NY), the mean leafing date (LD), the standard deviation (SD) of the median leafing date anomaly, and the half mean interquartile range (IQR) of leafing date anomaly in the data available years at the CPON stations
Figure 2 shows the number of the observed deciduous species and responsive species at each selected CPON station. The number of the observed deciduous woody species at each station varies in the range of 6–64. The mean leafing date of the observed deciduous woody species ranges from March to May, with the earlier date in the south and later date in the north (Table 1). Most (more than 60 %) species is responsive species (Fig. 2), indicating most species have the same leafing response (in the change direction) as that of the majority of species. The median and the interquartile range of the leafing date anomalies of all the available deciduous woody plant species at each selected CPON station are shown in Fig. 3. Although there are considerable differences between the leafing date anomaly of single species and the median value at some particular stations and in some particular years (e.g., at station Shenyang2 in 1991–1995 and Haerbin1 in 1989), the interquartile range is generally small. The half of mean interquartile range in the data available years at each station ranges from 1.5 to 3.2 days (Table 1). The standard deviation of the median value in the data available years at each station is about 5 days. The half mean interquartile range is generally less than the standard deviation of the median value. There are two nearby station pairs
Station
Code
Latitude
Longitude
NY
LD
SD (day)
IQR/2 (day)
Nenjiang Deduxian Yichun Jiamusi
1 2 3 4
49.17 48.50 47.72 46.82
125.22 126.20 128.88 130.40
23 22 10 21
May.12 May.20 May.16 May.8
4.8 4.4 5.4 4.7
2.6 1.5 2.5 1.5
Haerbin1 Haerbin2 Mudanjiang Shenyang1 Shenyang2 Chengde Gaixian Beijing1 Beijing2 Qinhuangdao Tianjin Dezhou Xingtai Liaocheng Taian
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
45.75 45.74 44.58 41.94 41.80 40.95 40.42 40.00 40.00 39.93 39.06 37.44 37.06 36.45 36.18
126.65 126.65 129.60 123.52 123.43 117.85 122.37 116.27 116.20 119.59 117.27 116.28 114.50 115.97 117.10
15 13 19 11 11 21 18 23 12 14 13 14 15 19 13
May.11 Apr.28 May.5 Apr.29 Apr.26 Apr.25 Apr.22 Apr.18 Apr.9 Apr.21 Apr.13 Apr.13 Apr.7 Apr.9 Apr.9
4.4 5.7 4.6 3.9 4.9 5.7 5.2 4.7 4.2 5.2 4.2 4.5 4.2 4.1 4.7
2.5 2.0 1.9 3.2 2.4 2.4 2.6 2.4 2.4 2.1 2.1 1.8 2.1 2.1 2.4
Int J Biometeorol Fig. 3 The median and the interquartile range of deciduous woody plant leafing date anomaly at the CPON stations in 1974– 1996. The gap shows where the leafing date data is missing
(Beijing1 and Beijing2, and Haerbin1 and Haerbin2) with a distance of around 8 km in between the paired stations and overlapped period of 1975–1986. The time series of the median values at the paired stations are strongly correlated in the overlapped period with a correlation coefficient of 0.85 (p < 0.01) for Beijing and 0.69 (p
ORIGINAL PAPER
Spatiotemporal analysis of ground-based woody plant leafing in response to temperature in temperate eastern China Guohua Liu & Qiuhong Tang & Xingcai Liu & Junhu Dai & Xuezhen Zhang & Quansheng Ge & Yin Tang
Received: 25 May 2013 / Revised: 15 October 2013 / Accepted: 29 October 2013 # ISB 2013
Abstract The analysis of woody plant leafing in response to regional-scale temperature variation using ground-based phenology is usually limited by the sparse coverage and missing data of ground observation. In this study, a station-based multispecies method was proposed to generate spatiotemporal variation of woody plant leafing date using ground observations from the Chinese Phenological Observation Network during 1974–1996. The results show that the leafing date had slightly insignificant advance (−0.56 day decade−1), and the Arctic Oscillation (AO) index could explain 36 % variance of the spring leafing date anomaly. The leafing date had been substantially delayed (4 days) when AO shifted from an extreme high index state (2) in 1989–1990 to a relatively low state (0.1) in 1991–1996. The canonical correlation analysis (CCA) was used to demonstrate the temporal evolutions and spatial structures of interannual variations of the spring temperature and leafing date anomalies. The three CCA spatial patterns of leafing date anomaly are similar to those of spring temperature anomaly. The first spatial pattern shows ubiquitous warming, which is consistent with the ubiquitous advance in leafing date across the study area. The second and third spatial patterns present the regional differences featured by advanced (delayed) leafing associated with high (low) temperature. The results suggest that the spring leafing date anomaly is spatiotemporally coherent with the regional-scale temperature variations. Although we focus here on woody Electronic supplementary material The online version of this article (doi:10.1007/s00484-013-0762-8) contains supplementary material, which is available to authorized users. G. Liu : Q. Tang (*) : X. Liu : J. Dai : X. Zhang : Q. Ge : Y. Tang Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, China e-mail: [email protected] G. Liu University of Chinese Academy of Sciences, Beijing, China
plant leafing in a historical period in temperate eastern China, our station-based multispecies method may be applicable to analysis of the ground-based phenology in response to regional-scale climatic variation in other regions. Keywords Leafing . Woody plant . Spring temperature . Canonical correlation analysis . Temperate eastern China
Introduction Study on plant phenology has received increasing attention in the light of global warming (Cleland et al. 2007; Morin et al. 2009; Richardson et al. 2013). Plant leafing (leaf unfolding), one of the most obvious phenological events in spring, is sensitive to climatic factors such as air temperature (Menzel et al. 2005; Schwartz et al. 2006; Chen and Xu 2012a; Polgar and Primack 2013), photoperiod (Partanen et al. 1998; Körner and Basler 2010; Basler and Körner 2012; Vitasse and Basler 2013), and precipitation (Jolly and Running 2004; Zhang et al. 2005; Piao et al. 2006; Crimmins et al. 2010; Bradley et al. 2011; Grogan and Schulze 2012; Zhang et al. 2013). Therefore, plant leafing date, as an observable biological indicator, may be used to study climatic change and variability (Schwartz 1998; Menzel et al. 2006; Piao et al. 2006). In addition, shifts in the timing of plant leafing could alter the length and strength of plant activities and further affect regional and global carbon cycles (Barr et al. 2007; Piao et al. 2007; Cong et al. 2012). Hence a thorough understanding of the plant leafing and its dynamic responses to large-scale climatic changes is important for climatic and ecological research (Menzel 2002; Walther et al. 2002; Schwartz et al. 2006; Morisette et al. 2008; Barnosky et al. 2012). Ground observation and remote sensing are the two widely used but completely different approaches to measuring plant phenology (Badeck et al. 2004; Studer et al. 2005). Ground
Int J Biometeorol
observation comes with a high temporal resolution and detailed information on individual species but poor spatial coverage, whereas satellite data provide wide spatial coverage but low temporal resolution and less detailed species information (Tang et al. 2012). Satellite observations have been used to investigate spatiotemporal variation of vegetation phenology and climate change impacts in temperate China (Piao et al. 2006; Tao et al. 2008; Peng et al. 2010; Piao et al. 2011; Cong et al. 2013; Wu and Liu 2013). The satellite-based studies have shown an advanced trend of spring green-up in response to a warming trend in China (Piao et al. 2006; Ma and Zhou 2012; Cong et al. 2013). Although satellite observation is a powerful tool in monitoring large-scale vegetation phenology, it should be noted that satellite observations do not obtain the same traits as ground observations (Badeck et al. 2004). Ground observations can describe the onset of leafing phenophase at the level of the individual plant. The groundbased leafing dates have been linked with regional climatic changes, particularly in Europe (Peñuelas and Filella 2001; Menzel et al. 2006, 2011; Reyer et al. 2013), North America (Worrall 1999; Schwartz et al. 2006; Wolkovich et al. 2012), and China (Zheng et al. 2006; Dai et al. 2013a; Ma and Zhou 2012; Chen and Xu 2012a). Using ground observations and satellite observations from 1982 to 1993, Chen et al. (2005) showed that spatial patterns of growing season beginning dates correlate significantly with those of spring temperatures in temperate eastern China. Zheng et al. (2006) found that the leafing dates of woody plants were closely correlated with spring temperature and had advanced 1–5 days per decade in 1963–1996 based on ground observations in the region north of 33°N over the eastern China. Ground observations in 1986–2005 have also been used to investigate the phenological responses of specific species to climate change in the temperate zone of China (Chen and Xu 2012a, b). Using the ground-based leafing dates and a process-based model, Wang et al. (2012) demonstrated the spatiotemporal dynamics of spring phenophase changes of Fraxinus chinensis from 1952 to 2007 in China. These analyses are generally siteand species-based and mainly focus on the developmental shifts of individual species. Hence, it is difficult to obtain large-scale phenology information from site-based data due to the sparse coverage and missing data of ground observation (Badeck et al. 2004; Liang et al. 2011). A few studies have been conducted to derive the phenological patterns of some specific species using the ground-based phenology (Rötzer and Chmielewski 2001; Estrella et al. 2009; Dai et al. 2013b; Xu and Chen 2013). These studies require phenological observation of a certain number of specific species at all (or most) observation stations. However, the phenology observation network in temperate eastern China is lacking of many specific species at most stations, and a method which requires specific species across most stations is therefore not suitable.
Moreover, the use of the specific species may reflect only the phenological pattern of these species but not necessarily the pattern of the majority of species. In this study, we proposed a station-based multispecies method which combined leafing anomalies from all available species at each station to generate spatiotemporal variation of woody plant leafing date over temperate eastern China. This method could minimize the data limitation problem of ground-based observations while retaining as much as possible information of the regional-scale phenological variation. Furthermore, for better understanding, the dynamic (spatial and temporal) responses of leafing date to regional-scale temperature variation, the canonical correlation analysis (CCA) was used to extract the canonical spatial structures and temporal evolutions between pairs of variables (i.e., leafing date anomaly and temperature anomaly). The main objective of this study is to investigate the spatiotemporal variations of the ground-based woody plant leafing and to demonstrate the association between the ground-based phenology and regional-scale temperature variations in temperate eastern China from 1974 to 1996. First, we proposed a station-based multispecies method to generate spatiotemporal variations of woody plant leafing date from ground observations. Then the spatiotemporal variations of woody plant leafing in relation to spring air temperature changes are demonstrated.
Materials and methods Study area We chose the temperate eastern China (north to 35°N and east to 110°E in China but excluding Inner Mongolia, Fig. 1) as our study area. Most of the Chinese Phenological Observation Network (CPON) stations with long-term phenological observations during the period of 1974–1996 are located in the study area (Zheng et al. 2002). The three provinces (grey areas in Fig. 1) of the study area lie in Northeast China, and the south of the study area (dark grey areas in Fig. 1) is in North China. The climate over the study area is dominated by the East Asian monsoon with the remarkable seasonality and spatial variability in temperature. The main biomes comprise temperate hemi-boreal, mixed, and deciduous forests (Wu and Liu 2013). The biome dynamics in the study area are strongly influenced by the monsoon climate with a distinct seasonality (Chen et al. 2005). Phenological data The leafing date data of woody plants were taken from CPON, a network of phenological monitoring station administered through the Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences. We selected the species of deciduous woody plant with at least 10-
Int J Biometeorol
monthly temperature data came with a spatial resolution of 0.5°, matching the gridded field of leafing date anomaly. The spring (March, April, and May) temperature anomalies were calculated based on the gridded monthly data. The Arctic Oscillation (AO) index was obtained from the National Oceanographic and Atmospheric Administration (NOAA) Climate Prediction Center (http://www.cpc.noaa.gov/). The mean AO in later winter and spring (from January to May) was used. The linear least-squares regression model was applied to calculate the trend of study area average spring temperature, leafing date, mean AO, and the sensitivity (i.e., the regression slope) of leafing date anomaly to spring temperature change at each grid in the study period. The statistical significance of the trends, correlation, and sensitivity was calculated according to the two-tailed Student’s t test. Fig. 1 The study area and the CPON stations. The light grey area is Northeast China and the dark grey area is North China
year records at each station. We removed the stations where the number of selected species was less than five. We excluded the years when the number of stations was less than eight over the whole study areas and identified the study period of 1974–1996. A total of 5,654 leafing records from 176 species at 19 stations (see Appendix S1 for details) were used during the study period. The leafing date anomalies from the species’ mean leafing date during the study period were calculated for each species at each station. The median and interquartile range of the leafing date anomalies of all the available deciduous woody plant species were computed for each station. The median leafing date anomaly represents the majority of the leafing date anomaly of the deciduous woody plant species at the station, offering an estimate of leafing phenological responses to climatic change at regional scale. It should be noted that the median leafing date anomaly index does not give the difference of mean leafing date between different locations. Rather, it reflects the advance or delay of the leafing date at each location. If one species shows the same sign as the median leafing date anomaly in more than 70 % of the data available years during the study period, we define the specific species as a responsive species which has the same response (in change direction) as the leafing date anomaly of the majority of species at the station. We calculated the number of the responsive species at each station. Using the inverse distance weighted method, we interpolated the median leafing date anomalies at the stations to a 0.5°×0.5° grid over the study area for each year in the study period (Chmielewski and Rötzer 2002).
Climatic data Surface air temperature data were obtained from China Meteorological Administration (Xu et al. 2009). The gridded
Canonical correlation analysis The canonical correlation analysis (CCA) (Hotelling 1936; Barnett and Preisendorfer 1987) is a statistical technique to study the relationship between two sets of variable. It could identify pairs of optimally correlated patterns by analyzing the linear combinations of original variables. The first CCA pair gives the maximum correlation between two sets of variable, followed by the second CCA and so on. We adopted the approach proposed by Barnett and Preisendorfer (1987) to demonstrate the temporal evolutions and spatial structures of interannual variation of the spring temperature and leafing date anomalies from 1974 to 1996 over the study area. Prior to the CCA, the spring temperature and leafing anomaly fields were first approximated by two sets of leading empirical orthogonal functions (EOFs). The use of EOF not only reduces the dimension of original variable but also prefilters the data to eliminate noise and render the resulting modes more stable (Xoplaki et al. 2000). We computed the EOFs for the spring temperature and leafing date anomaly fields. Then we selected the first three EOFs of each data set and used CCA to search for pairs of weight sets such that the correlations between the linear combinations of the temperature EOFs and the leafing EOFs were maximized. A sequence of paired canonical patterns (i.e., the linear combinations of the EOFs) was identified with the highest possible correlation for the first pair and decreased highest correlations for the latter pairs. We used the first three pairs, which had included the primary interrelated information of the spring temperature and leafing anomaly fields (Chmielewski and Rötzer 2002). The temporal coefficient (canonical vectors) series associated with the most important three CCA pairs were analyzed. According to the time series of the three canonical patterns (CCA1, CCA2, and CCA3 modes) for leafing date anomaly, we classified the years in the study period to the canonical patterns. For an individual year, if the maximum absolute value of the three sets of
Int J Biometeorol
Results Observed leafing date
Fig. 2 The numbers of the observed deciduous woody species and responsive species at the 19 selected CPON stations
normalized canonical vector is over 1.0, the associated canonical mode (CCA1, CCA2, or CCA3) with the maximum absolute value was identified. The individual year was classified into the associated canonical mode. The year which was classified into the CCA1 (or CCA2, CCA3) mode is called CCA1 (or CCA2, CCA3) year. The leafing date anomaly in the positive CCA1 year has a spatial pattern similar to the CCA1 mode. The leafing date anomaly in the reversed CCA1 year has a spatial pattern similar to the CCA1 mode but with opposite sign. If the maximum absolute value was less than 1.0, the individual year was taken as an unclassified year. The spatial pattern of the spring temperature and observed leafing date anomalies were shown for the three classifications corresponding to the three canonical patterns. Table 1 The number of the data available years (NY), the mean leafing date (LD), the standard deviation (SD) of the median leafing date anomaly, and the half mean interquartile range (IQR) of leafing date anomaly in the data available years at the CPON stations
Figure 2 shows the number of the observed deciduous species and responsive species at each selected CPON station. The number of the observed deciduous woody species at each station varies in the range of 6–64. The mean leafing date of the observed deciduous woody species ranges from March to May, with the earlier date in the south and later date in the north (Table 1). Most (more than 60 %) species is responsive species (Fig. 2), indicating most species have the same leafing response (in the change direction) as that of the majority of species. The median and the interquartile range of the leafing date anomalies of all the available deciduous woody plant species at each selected CPON station are shown in Fig. 3. Although there are considerable differences between the leafing date anomaly of single species and the median value at some particular stations and in some particular years (e.g., at station Shenyang2 in 1991–1995 and Haerbin1 in 1989), the interquartile range is generally small. The half of mean interquartile range in the data available years at each station ranges from 1.5 to 3.2 days (Table 1). The standard deviation of the median value in the data available years at each station is about 5 days. The half mean interquartile range is generally less than the standard deviation of the median value. There are two nearby station pairs
Station
Code
Latitude
Longitude
NY
LD
SD (day)
IQR/2 (day)
Nenjiang Deduxian Yichun Jiamusi
1 2 3 4
49.17 48.50 47.72 46.82
125.22 126.20 128.88 130.40
23 22 10 21
May.12 May.20 May.16 May.8
4.8 4.4 5.4 4.7
2.6 1.5 2.5 1.5
Haerbin1 Haerbin2 Mudanjiang Shenyang1 Shenyang2 Chengde Gaixian Beijing1 Beijing2 Qinhuangdao Tianjin Dezhou Xingtai Liaocheng Taian
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
45.75 45.74 44.58 41.94 41.80 40.95 40.42 40.00 40.00 39.93 39.06 37.44 37.06 36.45 36.18
126.65 126.65 129.60 123.52 123.43 117.85 122.37 116.27 116.20 119.59 117.27 116.28 114.50 115.97 117.10
15 13 19 11 11 21 18 23 12 14 13 14 15 19 13
May.11 Apr.28 May.5 Apr.29 Apr.26 Apr.25 Apr.22 Apr.18 Apr.9 Apr.21 Apr.13 Apr.13 Apr.7 Apr.9 Apr.9
4.4 5.7 4.6 3.9 4.9 5.7 5.2 4.7 4.2 5.2 4.2 4.5 4.2 4.1 4.7
2.5 2.0 1.9 3.2 2.4 2.4 2.6 2.4 2.4 2.1 2.1 1.8 2.1 2.1 2.4
Int J Biometeorol Fig. 3 The median and the interquartile range of deciduous woody plant leafing date anomaly at the CPON stations in 1974– 1996. The gap shows where the leafing date data is missing
(Beijing1 and Beijing2, and Haerbin1 and Haerbin2) with a distance of around 8 km in between the paired stations and overlapped period of 1975–1986. The time series of the median values at the paired stations are strongly correlated in the overlapped period with a correlation coefficient of 0.85 (p < 0.01) for Beijing and 0.69 (p
