Assessing the Impacts of Extreme Climate Events on Vegetation Activity in the North South Transect of Eastern China (NSTEC)

: Extreme climate events frequently exert serious e ﬀ ects on terrestrial vegetation activity. However, these e ﬀ ects are still uncertain in widely distributed areas with di ﬀ erent climate zones. Transect analysis is important to understand how terrestrial vegetation responds to climate change, especially extreme climate events, by substituting space for time. In this paper, seven extreme climate indices and the Normalized Di ﬀ erence Vegetation Index (NDVI) are employed to examine changes in the extreme climate events and vegetation activity. To reduce the uncertainty of the NDVI, two satellite-derived NDVI datasets, including the third generation Global Inventory Monitoring and Modeling System (GIMMS-3g) NDVI dataset and the NDVI from the National Oceanic and Atmospheric Administration (NOAA) satellites on Star Web Servers (SWS), were employed to capture changes in vegetation activity. The impacts of climate extremes on vegetation activity were then assessed over the period of 1982–2012 using the North–South Transect of Eastern China (NSTEC) as a case. The results show that vegetation activity was overall strengthened from 1982 to 2012 in the NSTEC. In addition, extreme high temperature events revealed an increased trend of approximately 5.15 days per decade, while a weakened trend (not signiﬁcant) was found in extreme cold temperature events. The strengthened vegetation activities could be associated with enhanced extreme high temperature events and weakened extreme cold temperature events over the past decades in most of the NSTEC. Despite this, inversed changes were also found locally between vegetation activity and extreme climate events (e.g., in the Northeast Plain). These phenomena could be associated with di ﬀ erences in vegetation type, human activity, as well as the combined e ﬀ ects of the frequency and intensity of extreme climate events. This study highlights the importance of accounting for the vital roles of extreme climate e ﬀ ects on vegetation activity.


Introduction
Past studies showed that the global mean temperature rose by approximately 0.85 • C from 1880 to 2012 [1,2]. However, changes in extreme climate events are more uncertain than average climate conditions. With increasing global climate warming, extreme climate events, including temperature extremes and precipitation extremes, occur more frequently than ever before. This phenomenon exerts serious effects on humans due to probable flooding, droughts, and so on [3][4][5][6], and has become a research focus for the scientific community.
The NDVI (Normalized Difference Vegetation Index) provides a critical historical perspective on vegetation activities. The consistency of the NDVI derived from various sensors in satellites has been evaluated with ground-based sensors [7,8]. On these bases, the NDVI is widely employed to detect vegetation dynamics, as well as responses to climate change at regional and global scales [9][10][11][12][13][14][15]. Climate extremes could alter physiological and ecological processes and even ecosystem functions in the terrestrial ecosystem [16,17]. For instance, Barichivich et al. analyzed the relationships between multiple climate and vegetation indicators of the growing season of northern ecosystems and their connection with the carbon cycle using NDVI as the indicator [14]. Tan et al. explored the relationships between vegetation growth and climate extremes in the Poyang Lake River Basin [18]. They found that variations in the NDVI are generally determined by temperature but not precipitation extremes. Sangeeta et al. studied the characterization of vegetation dynamics over South Asia using satellite time series of different NDVI sources and found that the spatiotemporal vegetation trends derived from the NDVI-3g are analogous to both the NDVI and EVI from the Moderate-Resolution Imaging Spectroradiometer (MODIS) for the Earth Observing System (EOS), thereby indicating greening over semi-arid regions [19]. Zhao et al. studied the change of vegetation growth and its correlation with extreme climate events in the Loess Plateau. They found that vegetation coverage responded positively to precipitation and temperature extremes in the region [20]. By using Xinjiang in China as a case study, Yao et al. showed that climate extremes could have negative effects on vegetation growth in the arid region of Central Asia [21]. Li et al. estimated the trends of several extreme climate indices and the NDVI in the Mongolia Plateau. They found positive correlations between most of the extreme climate indices and the NDVI at monthly and seasonal scales [22]. Despite being widely employed in past studies, remote sensed data are affected by several disturbances when monitoring vegetation activities, including systematic errors, accidental errors, and problematic processing methods [23,24]. In addition, past studies were frequently performed by taking topographic units such as plateaus or basins as the case. Instead, transect analysis is an important and effective approach to understand how terrestrial ecosystems would respond to global climate change by substituting space for time [25,26].
By using the North-South Transect of Eastern China (NSTEC) as a case, this paper analyzes the impacts of climate extremes (i.e., temperature extremes and precipitation extremes) on terrestrial vegetation activity over the period of 1982-2012. The NSTEC is one of the 15 global climate change transections in the International Geosphere-Biosphere Programme (IGBP) [25]. It encompasses a large climate gradient spanning from temperate to tropical climates. Specifically, this paper aims to (1) examine the spatiotemporal variation of vegetation activity in the NSTEC using two different satellite-derived NDVI datasets; (2) analyze the changes in extreme climate events using seven extreme climate indices from the Expert Team on Climate Change Detection and Indices (ETCCDI); (3) explore the relationships between vegetation activity and extreme climate events.

Data and Preprocessing
The data sources in this study include satellite-based NDVI data, Leaf Area Index (LAI) data, daily meteorological data, and some other auxiliary data (e.g., climate zones, vegetation types, and elevation data). In detail, weekly NDVI data, which cover the period of 1982-2012, were obtained from National Oceanic and Atmospheric Administration (NOAA) satellites on Star Web Servers (NOAA-SWS) at a resolution of 4 km. In addition, the GIMMS-3g NDVI dataset, at a resolution of 8 km, was obtained from the third generation Global Inventory Monitoring and Modeling System (GIMMS) from the Advanced Very High Resolution Radiometer (AVHRR) sensors. The dataset has been corrected for sensor degradation, inter-sensor differences, cloud cover, solar zenith angle, viewing angle effects due to satellite drift, and volcanic aerosols [27]. The NDVI data were then synthesized to monthly and yearly scales, respectively. In addition, the annual maximum LAI data were obtained from the long-term ground-based observations of China FLUX and published studies [28] (Figure 1).
For the meteorological data, the daily maximum temperature, daily minimum temperature, and daily precipitation data were collected over the period of 1982-2012 from the National Meteorological Information Center in China. Quality control of the data was first conducted via a manual inspection of time consistency and extremum validation [29]. We further validated the data by screening and eliminating suspicious and missing records to ensure their continuity and consistency. In detail, the stations that were newly-built or abandoned later were removed to ensure the data's consistency and accuracy. Finally, 355 meteorological stations were selected for further analysis. To extract the corresponding NDVI value around the meteorological stations, the grid value of the site's location

Data and Preprocessing
The data sources in this study include satellite-based NDVI data, Leaf Area Index (LAI) data, daily meteorological data, and some other auxiliary data (e.g., climate zones, vegetation types, and elevation data). In detail, weekly NDVI data, which cover the period of 1982-2012, were obtained from National Oceanic and Atmospheric Administration (NOAA) satellites on Star Web Servers (NOAA-SWS) at a resolution of 4 km. In addition, the GIMMS-3g NDVI dataset, at a resolution of 8 km, was obtained from the third generation Global Inventory Monitoring and Modeling System (GIMMS) from the Advanced Very High Resolution Radiometer (AVHRR) sensors. The dataset has been corrected for sensor degradation, inter-sensor differences, cloud cover, solar zenith angle, viewing angle effects due to satellite drift, and volcanic aerosols [27]. The NDVI data were then synthesized to monthly and yearly scales, respectively. In addition, the annual maximum LAI data were obtained from the long-term ground-based observations of China FLUX and published studies [28] (Figure 1).
For the meteorological data, the daily maximum temperature, daily minimum temperature, and daily precipitation data were collected over the period of 1982-2012 from the National Meteorological Information Center in China. Quality control of the data was first conducted via a manual inspection of time consistency and extremum validation [29]. We further validated the data by screening and eliminating suspicious and missing records to ensure their continuity and consistency. In detail, the stations that were newly-built or abandoned later were removed to ensure the data's consistency and accuracy. Finally, 355 meteorological stations were selected for further analysis. To extract the corresponding NDVI value around the meteorological stations, the grid value of the site's location was extracted to participate in subsequent calculations using the Geographic Information System (GIS).

Monitoring Vegetation Activities by Using the NDVI as an Indicator
At present, more than 30 vegetation indices, including the NDVI, the Enhanced Vegetation Index (EVI), and the Difference Vegetation Index (DVI), were developed to reflect changes in vegetation activities [30][31][32][33]. Although many new vegetation indices were proposed to account for many factors, such as soil and atmospheric effects [11], the NDVI is still a good indicator when dealing with large-scale vegetation coverage and vegetation productivity [34][35][36]. For instance, Barichivich et al. found that the GIMMS-3g is highly consistent with MODIS NDVI [14]. In this paper, the NDVI was employed as an indicator to monitor changes in vegetation activities in the NSTEC. To reduce the uncertainty of the NDVI datasets at different resolutions, two satellite-based NDVI datasets from GIMMS-3g and NOAA-SWS were employed due to the dramatic improvement of the GIMMS-3g and NOAA-SWS datasets [37][38][39]. In detail, monthly and yearly NDVIs were synthesized from the original weekly data. A simple linear regression was then conducted to examine the linear trend of the NDVI (Y) [40,41] where t is the time period(year), and a and b are regression coefficients. If a is larger (or less) than zero at a 95% significance level, then there is statistical evidence to support the presence of increased (or decreased) trends of the NDVI (i.e., vegetation activity). In this paper, linear regression was performed using the R Software [42] and the trend package [43].

Characterizing Extreme Climate Events Using Extreme Climate Indices
The climate extreme index (CEI) was first proposed in the United States in 1996. However, the CEI was found to have some limitations, such as its lack of suitability and effectiveness when applied to other countries and regions [44]. The European Climate Assessment (ECA) indices were then developed by accounting for temperature, precipitation, frost, and growth season [45]. Many climate extreme indices, including most of the ECA indices, were also proposed by the Expert Team on Climate Change Detection and Indices (ETCCDI) [46]. The ETCCDI has been used in the fourth and fifth Intergovernmental Panel on Climate Change (IPCC) [47,48], which makes the indices widely used in extreme climate events [49][50][51].
In this paper, seven indices from the ETCCDI were selected to reflect the intensity and frequency of extreme climate events to examine the changes in temperature and precipitation extremes in the NSTEC (Table 1). In detail, this study includes three extreme cold temperature indices (CSDI, TNN, and TN10p), three extreme high temperature indices (WSDI, TXX, and TX90p), and one extreme precipitation index (R95p). TXX and TNN reflect the intensity of extreme temperature events. In addition, the R95p is employed to characterize the extreme precipitation intensity. The remaining four indices, TN10p, TX90p, CSDI, and WSDI, represent the frequency of extreme temperature events. To examine changes in the intensity of extreme climate events, linear regression was also performed using TXX, TNN, and R95p as indicators. If the slope (a in Equation (1)) is larger (or smaller) than zero, there is an increased (or decreased) trend in the extreme climate events.
Given the discrete nature of the frequency of extreme climate events, a Poisson regression model was employed to examine their monotonic trends. If the probability of a random variable N equals k, then we get the Poisson distribution density function as Equation (2).
where k represents the frequency of the observed samples at a certain time, and λ i is the parameter of incidence. To evaluate whether the extreme precipitation frequency at each station has an increasing or decreasing statistical trend, a linear relationship between occurrence rate parameters (λ) and time (t) is established: If the regression coefficient β is larger (or smaller) than zero at 5%, statistical evidence is found to support temporal trends in the occurrence of extreme climate events. In this paper, Poisson regression was applied in four indices, including TN10p, TX90p, CSDI, and WSDI, respectively. Similarly, all these algorithms were performed using the R Software [42] and the trend package [43].

Assessing the Impacts of Extreme Climate Events on Vegetation Activities
The LAI, which is defined as one half of the total green leaf area per unit ground surface area [52], has been widely used to characterize the structure and function of vegetation [53]. In this paper, the LAI data from ground observation were used to verify the reliability of the aforementioned two NDVI datasets. On this basis, we examined the spatiotemporal trends of the vegetation activity over the period 1982-2012 in the NSTEC using the NDVI as an indicator. The changes in climate extremes, including extreme temperature events and extreme precipitation events, were analyzed using the aforementioned seven extreme climate indices from ETCCDI as indicators. Pearson correlation analyses were then conducted between extreme climate indices, and the corresponding NDVI values were sampled according to the meteorological stations. The combined effects of the frequency and intensity of extreme climate events on vegetation activities were also analyzed based on these numerical values. The effect of extreme climate events on vegetation activity was further analyzed by climatic ecozones in the NSTEC. Possible causes of vegetation activity changes to extreme climate events were then discussed from the aspects of vegetation type, human activity, and the combined effects of the frequency and intensity of extreme climate events.

Validation of the NDVI Datasets
In past studies, many NDVI datasets were widely evaluated through sample-based validation or comparison with existing NDVI products by third-party researchers. For instance, Fensholt and Proud [54] compared the performances of a time series of GIMMS and MODIS NDVI data to evaluate long-term vegetation trends at a global scale. In this paper, the accuracy of the aforementioned NDVI datasets was further evaluated by comparing it to ground-based LAI observations. As seen in Figure 2, a significant correlation was found between the NOAA-SWS-derived NDVI and the LAI (R = 0.768, N = 22, P = 0.000). Similarly, a significant correlation was also found between the GIMMS-3g-derived NDVI (R = 0.769, N = 22, P = 0.000) and the ground-based LAI. datasets was further evaluated by comparing it to ground-based LAI observations. As seen in Figure  2, a significant correlation was found between the NOAA-SWS-derived NDVI and the LAI (R = 0.768, N = 22, P = 0.000). Similarly, a significant correlation was also found between the GIMMS-3g-derived NDVI (R = 0.769, N = 22, P = 0.000) and the ground-based LAI.

Variations of Vegetation Activity
As shown in Figure 3, the two NDVI datasets linearly fit close to each other. Specifically, both the satellite-derived observations from the GIMMS-3g and NOAA-SWS NDVI datasets revealed increasing trends from 1982 to 2012 at a rate of 0.0113 ± 0.0026 per decade (R 2 = 0.386) from the SWS and 0.0111 ± 0.0026 per decade (R 2 = 0.526) from the GIMMS-3g, respectively. These results indicate the overall enhanced vegetation activity in the past few decades (taking the NSTEC as a whole).

Variations of Vegetation Activity
As shown in Figure 3, the two NDVI datasets linearly fit close to each other. Specifically, both the satellite-derived observations from the GIMMS-3g and NOAA-SWS NDVI datasets revealed increasing trends from 1982 to 2012 at a rate of 0.0113 ± 0.0026 per decade (R 2 = 0.386) from the SWS and 0.0111 ± 0.0026 per decade (R 2 = 0.526) from the GIMMS-3g, respectively. These results indicate the overall enhanced vegetation activity in the past few decades (taking the NSTEC as a whole).  The spatial heterogeneity in the vegetation activity was further analyzed by examining the annual mean values and trends of the NDVI from 1982 to 2012. We found that the mean NDVI is relatively high in broad-leaved forests and coniferous forests in the regions including the Greater Khingan Range (G1) and the Changbai Mountains (G4) in the northeastern NSTEC, the Qinling Mountains in the Central parts of the NSTEC, the Wuyi Mountains (G9), and the Hainan Island in the southern NSTEC. On the contrary, the lowest NDVI was found in grasslands and human landuse areas. Concretely, the areas with relatively small NDVIs were mainly distributed in the Northeast Plain (G3), the Loess Plateau (G5), the Yangtze River Basin (G8), and the Pearl River Delta (G10). Figure 4 shows the spatial patterns of the trend in NDVI derived from the GIMMS-3g and NOAA-SWS datasets during the period from 1982 to 2012. Both datasets revealed ascending trends in most parts of the NSTEC over the past decades. Novertheless, decreased NDVI could also be found in the Inner Mongolia Plateau (G2), the eastern Great Hinggan Mountains (G1), and the Pearl River Basin (G10) in Figures 1 and 4. The NDVI trend was further analyzed by vegetation types. The results showed that all types of vegetation had mainly increased trends in the NSTEC over the past few The spatial heterogeneity in the vegetation activity was further analyzed by examining the annual mean values and trends of the NDVI from 1982 to 2012. We found that the mean NDVI is relatively high in broad-leaved forests and coniferous forests in the regions including the Greater Khingan Range (G1) and the Changbai Mountains (G4) in the northeastern NSTEC, the Qinling Mountains in the Central parts of the NSTEC, the Wuyi Mountains (G9), and the Hainan Island in the southern NSTEC. On the contrary, the lowest NDVI was found in grasslands and human land-use areas. Concretely, the areas with relatively small NDVIs were mainly distributed in the Northeast Plain (G3), the Loess Plateau (G5), the Yangtze River Basin (G8), and the Pearl River Delta (G10). Figure 4 shows the spatial patterns of the trend in NDVI derived from the GIMMS-3g and NOAA-SWS datasets during the period from 1982 to 2012. Both datasets revealed ascending trends in most parts of the NSTEC over the past decades. Novertheless, decreased NDVI could also be found in the Inner Mongolia Plateau (G2), the eastern Great Hinggan Mountains (G1), and the Pearl River Basin (G10) in Figures 1 and 4. The NDVI trend was further analyzed by vegetation types. The results showed that all types of vegetation had mainly increased trends in the NSTEC over the past few decades. Specifically, 73% of tropical rainforests showed an increasing trend, reaching 0.00185·year −1 in Hainan Island. The slowest growth was found in temperate coniferous and broad-leaved mixed forests at approximately 0.0084 per decade.

Frequency of Extreme Temperature Events
As shown in Figure 5, extreme cold temperature indices, including CSDI and TN10p, revealed decreased trends at a rate of −0.066 ± 0.112 days•year −1 (R 2 = 0.012, P = 0.230) and −0.230 ± 0.166 days•year −1 (R 2 = 0.062, P = 0.177). On the contrary, extreme high temperature indices showed obvious upward trends, reaching 0.136 ± 0.041 days•year −1 for the WSDI (R 2 = 0.278, P = 0.002) and 0.515 ± 0.088 days•year −1 for the TX90p (R 2 = 0.544, P = 0.000). We first analyzed the spatial patterns of the mean annual precipitation from 1982 to 2012 in the NSTEC (Figure 6a). We found that it decreased from southeast to northwest, owing to the combining influences of southeast monsoons and southwest monsoons [55]. Furthermore, we analyzed the spatial heterogeneity of the trends in extreme temperature frequencies using CSDI, TN10p, WSDI, and TX90p as indicators (Figure 6b-e). For extreme cold temperature events, an overall decreasing trend could be found in the NSTEC based on an analysis of the two extreme cold temperature indices (i.e., CSDI and TN10p). Concretely, the CSDI showed an ascent in the Greater Hinggan Mountains

Frequency of Extreme Temperature Events
As shown in Figure 5, extreme cold temperature indices, including CSDI and TN10p, revealed decreased trends at a rate of −0.066 ± 0.112 days•year −1 (R 2 = 0.012, P = 0.230) and −0.230 ± 0.166 days•year −1 (R 2 = 0.062, P = 0.177). On the contrary, extreme high temperature indices showed obvious upward trends, reaching 0.136 ± 0.041 days•year −1 for the WSDI (R 2 = 0.278, P = 0.002) and 0.515 ± 0.088 days•year −1 for the TX90p (R 2 = 0.544, P = 0.000). We first analyzed the spatial patterns of the mean annual precipitation from 1982 to 2012 in the NSTEC (Figure 6a). We found that it decreased from southeast to northwest, owing to the combining influences of southeast monsoons and southwest monsoons [55]. Furthermore, we analyzed the spatial heterogeneity of the trends in extreme temperature frequencies using CSDI, TN10p, WSDI, We first analyzed the spatial patterns of the mean annual precipitation from 1982 to 2012 in the NSTEC (Figure 6a). We found that it decreased from southeast to northwest, owing to the combining influences of southeast monsoons and southwest monsoons [55]. Furthermore, we analyzed the spatial heterogeneity of the trends in extreme temperature frequencies using CSDI, TN10p, WSDI, and TX90p as indicators (Figure 6b-e). For extreme cold temperature events, an overall decreasing trend could be found in the NSTEC based on an analysis of the two extreme cold temperature indices (i.e., CSDI and TN10p). Concretely, the CSDI showed an ascent in the Greater Hinggan Mountains (G1), the Northeast Plain (G3), the Changbai Mountains (G4), the southern part of the Loess Plateau (G5), and parts of the Yangtze River Basin (G8). However, these trends were reversed in some other areas, including the Hainan Island, the Wuyi Mountain (G9), the East China Plain, and the Yangtze River Basin (G8). This result is consistent with Du (2015) [56]. The TN10p showed generally weak trends in the southern parts of the NSTEC. In addition, the trend of TN10p is consistent with that of the CSDI but different in magnitude. In terms of extreme high temperature indices (i.e., WSDI and TX90p), similar changes could be found between them (Figure 6d,e). In detail, the WSDI and TX90p showed increasing trends in most areas of the NSTEC. However, an obvious decrease in the WSDI could be found in the Northeast Plain.  [56]. The TN10p showed generally weak trends in the southern parts of the NSTEC. In addition, the trend of TN10p is consistent with that of the CSDI but different in magnitude. In terms of extreme high temperature indices (i.e., WSDI and TX90p), similar changes could be found between them (Figure 6d,e). In detail, the WSDI and TX90p showed increasing trends in most areas of the NSTEC. However, an obvious decrease in the WSDI could be found in the Northeast Plain.    An overall fluctuation was found for the extreme precipitation index (R95p) in the NSTEC from 1982 to 2012 (4.3 ± 8.4 mm·year −1 , R 2 = 0.009, P = 0.611). The spatial distribution mainly showed a decreasing trend in the northern NSTEC. Both increasing and decreasing trends could be found in the southern NSTEC [57]. Precipitation may affect vegetation types deeply and even the NDVI because the main potential climatic constraints for plant growth are sunlight in the south, water in the middle, and temperature in the north of the NSTEC, according to Nemani et al. [58]. Figures 7 and 8 show the changes of extreme temperature intensity in the NSTEC from 1982 to 2012. In detail, the intensity of extreme cold temperature events (TNN) showed a fluctuation from 1982 to 2012 (0.013 ± 0.024 °C•year −1 , R 2 = 0.001, P = 0.966). However, an enhanced warming trend could be found in Changbai Mountains and the northern parts of the North China Plain. For extreme high temperature events (TXX), small increasing trends could be found in the NSTEC over the past decades (0.036 ± 0.013 °C•year −1 , R 2 = 0.220, P = 0.634). An overall fluctuation was found for the extreme precipitation index (R95p) in the NSTEC from 1982 to 2012 (4.3 ± 8.4 mm•year −1 , R 2 = 0.009, P = 0.611). The spatial distribution mainly showed a decreasing trend in the northern NSTEC. Both increasing and decreasing trends could be found in the southern NSTEC [57]. Precipitation may affect vegetation types deeply and even the NDVI because the main potential climatic constraints for plant growth are sunlight in the south, water in the middle, and temperature in the north of the NSTEC, according to Nemani et al. [58].

Correlation between the Frequency Indices of Extreme Climate Events and Vegetation Activity
The correlations between frequency indices of extreme temperature events (i.e., CSDI, TN10p, WSDI, and TX90p) and NDVI were first analyzed over the period 1982-2012 in the NSTEC. Negative correlations could be found between the CSDI and the SWS-derived NDVI in 78% of the meteorological stations (R < −0.1; P < 0.05). Similar correlations could also be found in 74% of the meteorological stations when using GIMMS-3g-derived NDVI as variables (R < −0.1; P < 0.05). The TN10p showed negative correlations with the SWS-derived NDVI and GIMMS-3g NDVI in 81% and 73% of the stations, respectively. These results indicate that the frequency of extreme low temperature events, which are represented by the CSDI and the TN10p, were most negatively correlated with the NDVI during 1982-2012 in the NSTEC.
In addition, the frequency of extreme high temperature indices was positively correlated with the NDVI in most stations. According to the analysis of the SWS-derived NDVI and GIMMS-3g NDVI, 65% and 57% of the stations showed positive correlations between the WSDI and the two NDVI datasets at a 95% confidence level. Moreover, 70% and 88% of the TX90p stations showed significant positive correlations between the index and the two NDVI datasets (R > 0.1; P < 0.05). Figure 9 shows the spatial distributions of the correlation between the frequency indices of extreme climate events and the NDVI datasets from the NOAA-SWS and GIMMS-3g. Similar patterns were found on the correlations between the two NDVI datasets and the frequency indices of extreme

Correlation between the Frequency Indices of Extreme Climate Events and Vegetation Activity
The correlations between frequency indices of extreme temperature events (i.e., CSDI, TN10p, WSDI, and TX90p) and NDVI were first analyzed over the period 1982-2012 in the NSTEC. Negative correlations could be found between the CSDI and the SWS-derived NDVI in 78% of the meteorological stations (R < −0.1; P < 0.05). Similar correlations could also be found in 74% of the meteorological stations when using GIMMS-3g-derived NDVI as variables (R < −0.1; P < 0.05). The TN10p showed negative correlations with the SWS-derived NDVI and GIMMS-3g NDVI in 81% and 73% of the stations, respectively. These results indicate that the frequency of extreme low temperature events, which are represented by the CSDI and the TN10p, were most negatively correlated with the NDVI during 1982-2012 in the NSTEC.
In addition, the frequency of extreme high temperature indices was positively correlated with the NDVI in most stations. According to the analysis of the SWS-derived NDVI and GIMMS-3g NDVI, 65% and 57% of the stations showed positive correlations between the WSDI and the two NDVI datasets at a 95% confidence level. Moreover, 70% and 88% of the TX90p stations showed significant positive correlations between the index and the two NDVI datasets (R > 0.1; P < 0.05). Figure 9 shows the spatial distributions of the correlation between the frequency indices of extreme climate events and the NDVI datasets from the NOAA-SWS and GIMMS-3g. Similar patterns were found on the correlations between the two NDVI datasets and the frequency indices of extreme climate events. Concretely, the CSDI and the TN10p were negatively correlated with the NDVI in most stations in the Northeast Plain (G3) and the Yangtze River Basin (G8) in Figures 8e and 9a. As one of the main crop production areas, the NDVI for crop plants in the Northeast Plain was higher (or lower) when the number of consecutive days of cold temperature decreased (or increased) in past decades. These results could be due in part to the fact that crop plants are subject to suffering cold stress [59,60]. Further, crop growth may also be affected by irrigation [61] and soil erosion [62].
Water 2019, 11, x FOR PEER REVIEW 11 of 20 also be found in the northernmost part of the NSTEC, including the regions in the Changbai Mountain (G4) and the northern of Great Hinggan Mountains (G1) [63]. In addition, negative correlations could be found in the semi-arid areas where water is insufficient, including the Inner Mongolia Plateau (G2) and the Northeast Plain (G3) [63]. Negative correlations could be due to the fact that the vegetation growth of grasslands and croplands is inhibited, as extremely high temperatures can accelerate transpiration and evapotranspiration and thus threaten vegetation growth [64]. Similar patterns could also be found between the TX90p and the NDVI (Figure 9d,h).

Correlation between the Intensity Indices of Extreme Climate Events and Vegetation Activity
For extreme high temperature intensities, most stations passed a significance test according to the correlation analysis between the TXX and the NDVI. Among them, 63% and 50% of the stations showed positive correlations (R > 0.1; P < 0.05) according to the analysis of the SWS-derived NDVI and GIMMS-3g NDVI. For the extreme cold temperature intensity (TNN), we found that 53% and 63% of the stations showed a positive correlation between the TNN and the two NDVI datasets, respectively.
As shown in Figure 10, consistent correlations could be found between the two satellite-derived NDVI observations and the intensity indices of extreme climate events (i.e., TNN, TXX, and R95p). That is, the correlation between extreme high temperature and NDVI is mainly positive in the As shown in Figure 9c,g, positive correlations between the WSDI and the NDVI are mainly distributed in the eastern and southern parts of the NSTEC. This result indicates the possible enhancement of vegetation activity owing to the frequent occurrence of extreme high temperatures. In the southern NSTEC with a deciduous broad-leaved forest, concurrences of rain and heat energy provide favorable conditions for vegetation growth in monsoon climates. Similar phenomena could also be found in the northernmost part of the NSTEC, including the regions in the Changbai Mountain (G4) and the northern of Great Hinggan Mountains (G1) [63]. In addition, negative correlations could be found in the semi-arid areas where water is insufficient, including the Inner Mongolia Plateau (G2) and the Northeast Plain (G3) [63]. Negative correlations could be due to the fact that the vegetation growth of grasslands and croplands is inhibited, as extremely high temperatures can accelerate transpiration and evapotranspiration and thus threaten vegetation growth [64]. Similar patterns could also be found between the TX90p and the NDVI (Figure 9d,h).

Correlation between the Intensity Indices of Extreme Climate Events and Vegetation Activity
For extreme high temperature intensities, most stations passed a significance test according to the correlation analysis between the TXX and the NDVI. Among them, 63% and 50% of the stations showed positive correlations (R > 0.1; P < 0.05) according to the analysis of the SWS-derived NDVI and GIMMS-3g NDVI. For the extreme cold temperature intensity (TNN), we found that 53% and 63% of the stations showed a positive correlation between the TNN and the two NDVI datasets, respectively.
As shown in Figure 10, consistent correlations could be found between the two satellite-derived NDVI observations and the intensity indices of extreme climate events (i.e., TNN, TXX, and R95p). That is, the correlation between extreme high temperature and NDVI is mainly positive in the northeast and south parts of the NSTEC. In addition, the correlation between the NDVI and the TXX is highly consistent with that of the WSDI and the TX90p (Figure 9c,d,g,h and Figure 10b). However, for extreme cold temperature intensity, the spatial distribution of the correlation between the NDVI and the TNN was different from that of the frequency of extreme cold temperature events. In detail, the TNN was positively correlated with the NDVI in most of the NSTEC. northeast and south parts of the NSTEC. In addition, the correlation between the NDVI and the TXX is highly consistent with that of the WSDI and the TX90p (Figures 10b and 9c,d,g,h). However, for extreme cold temperature intensity, the spatial distribution of the correlation between the NDVI and the TNN was different from that of the frequency of extreme cold temperature events. In detail, the TNN was positively correlated with the NDVI in most of the NSTEC. For extreme precipitation intensity, negative correlations between the R95p and the NDVI were widely found in the Yangtze Basin, the Pearl River Basin, and the northernmost of the NSTEC. On the other hand, the positive correlations between R95p and NDVI were mainly concentrated in the central region of the NSTEC (Figure 10c,f), such as the Loess Plateau (G5), the Inner Mongolia Plateau (G2), and the Northeast Plain (G3). This phenomenon could be due to the fact that rainstorms can promote vegetation growth in the water-limited conditions in semi-arid regions [65].
The effect of extreme climate events on the NDVI were also analyzed by combining the For extreme precipitation intensity, negative correlations between the R95p and the NDVI were widely found in the Yangtze Basin, the Pearl River Basin, and the northernmost of the NSTEC. On the other hand, the positive correlations between R95p and NDVI were mainly concentrated in the central region of the NSTEC (Figure 10c,f), such as the Loess Plateau (G5), the Inner Mongolia Plateau (G2), and the Northeast Plain (G3). This phenomenon could be due to the fact that rainstorms can promote vegetation growth in the water-limited conditions in semi-arid regions [65].
The effect of extreme climate events on the NDVI were also analyzed by combining the frequency and intensity of extreme climate events as a whole. For instance, both the frequency and intensity of extreme high temperature indices (i.e., WSDI, TX90p, and TXX) showed overall increasing trends in the past decades in the NSTEC (Figures 5 and 7). Accordingly, the NDVI revealed an enhanced trend during this period. Furthermore, an obvious negative effect of extreme climate events on vegetation activity could be found in 1984, 1993, 2000, 2005, and 2007 (Table 2). For instance, the frequency and intensity of extreme cold temperatures (i.e., CSDI, TN10p, and TNN) were relatively severe in 1984. That is, extreme cold temperature events occurred more frequently and lasted for a longer time than during any other periods. Consequently, the NDVI showed a lower level than in any other years. Similar phenomena could also be found in 2005. Table 2. Changes of the NDVI and the extreme climate indices when severe extreme climate events occurred (where +/− represents the fact that extreme climate indices are higher/lower than usual).

Year
Higher than Usual Lower than Usual NDVI-SWS  NDVI-GIMMS   1984  CSDI, TN10p, TNN  --1993  CSDI, TN10p  TNN, WSDI, TX90p, TXX  --2000  TX90p, TXX  R95p  --2005  CSDI, TN10p, TNN  --2007 TNN, TX90p CSDI, TN10p + + In 1993, the frequency of extreme cold temperature events (e.g., CSDI and TN10p) was increased, while the intensity of the extreme cold temperature events was strengthened ( Table 2). As a result, the NDVI was relatively small due to the frequent occurrence of severe extreme cold temperatures. In 2000, both the frequency and intensity of extreme high temperatures were found to be enhanced (i.e., a large TX90p and large TXX). Meanwhile, a weak intensity for extreme precipitation events was found in the NSTEC ( Figure 5). Consequently, the NDVI was relatively small in comparison with that during other periods. This NDVI reduction could be attributed to the large-scale droughts in the region [66]. In 2007, the frequency and intensity of extreme cold temperature events were relatively weak (i.e., a small CSDI, small TN10p, and high TNN). Thus, the NDVI was relatively high in that period.
We further explored the combined spatial effects of the frequency and intensity of extreme climate events on vegetation in the NSTEC. Three typical regions were then analyzed: A1, a water-limited area; A2, a monsoon climate zone; and A3, an urbanized area ( Figure 11). In detail, In the region A1, extreme precipitation events showed a weakening trend (Figure 11e). Meanwhile, we found concurrent occurrences of increased frequency, as well as enhanced intensity of extreme high/cold temperatures (Figure 11b-d,f-h). Consequently, the NDVI showed mainly weakening trends over the past decades. Nevertheless, the NDVI showed an increasing trend in the northern Greater Hinggan Mountains (G1). This increase could be associated with the obvious increase in the days of the biological growth season [67], as well as the wide distribution of forest lands, including coniferous forests and coniferous mixed forests. In region A2, the NDVI mainly showed an upward trend (e.g., Jianghan-Dongting-Lake Plain; G7 in Figure 11). This increased NDVI could be associated with the abundant moisture and heat in a typical monsoon climate. Concretely, the intensity and frequency of extreme high temperatures increased significantly, while extreme precipitation revealed overall increasing trends. In addition, the intensity and frequency of extreme cold temperatures showed decreasing trends. Consequently, the concurrences of rain and heat energy provided a suitable environment for plant growth. A3 is mainly located in the Pearl River Delta (G10). In this region, the NDVI decreases significantly. This decrease could be mainly associated with rapid urban expansion in past decades. Since the 1980s, urbanization in the Pearl River Delta has increased by more than 60% [68]. Although the frequency and intensity of extreme high temperatures increased significantly in A3, the increase of urban land use will lead to a significant reduction in vegetation activity [69].

Analysis of the Correlation between Different Climatic Ecozones
As shown in Table 3, consistent correlations could be found in most cases from both the NOAA-SWS and GIMMS-3g datasets. For instances, extreme temperature events mainly showed positive effects on cold temperate coniferous forests. In addition, temperate grasslands are affected by extreme cold temperature events. That is, the frequency of extreme cold temperature mainly has negative relationships with NDVI. That is, more extreme cold temperature events could reduce the NDVI in temperate grasslands. Extreme high temperature indices (i.e., TXX, WSDI and TX90p) mainly have a negative effect on grasslands. In addition, extreme precipitation events have a positive influence on the grassland according to both satellite-derived observations (likely induced by the watering effects in semi-arid areas).
For forest vegetation, including temperate mixed forests and subtropical evergreen broadleaved forests, relatively consistent correlations could be found from both the NOAA-SWS and GIMMS-3g observations. For temperate mixed forests, subtropical evergreen broad-leaved forests, and tropical monsoon rainforests, both the frequency and intensity of extreme high temperature have positive effects on vegetation. In addition, extreme precipitation events had negative effects on the vegetation.

Analysis of the Correlation between Different Climatic Ecozones
As shown in Table 3, consistent correlations could be found in most cases from both the NOAA-SWS and GIMMS-3g datasets. For instances, extreme temperature events mainly showed positive effects on cold temperate coniferous forests. In addition, temperate grasslands are affected by extreme cold temperature events. That is, the frequency of extreme cold temperature mainly has negative relationships with NDVI. That is, more extreme cold temperature events could reduce the NDVI in temperate grasslands. Extreme high temperature indices (i.e., TXX, WSDI and TX90p) mainly have a negative effect on grasslands. In addition, extreme precipitation events have a positive influence on the grassland according to both satellite-derived observations (likely induced by the watering effects in semi-arid areas). For forest vegetation, including temperate mixed forests and subtropical evergreen broad-leaved forests, relatively consistent correlations could be found from both the NOAA-SWS and GIMMS-3g observations. For temperate mixed forests, subtropical evergreen broad-leaved forests, and tropical monsoon rainforests, both the frequency and intensity of extreme high temperature have positive effects on vegetation. In addition, extreme precipitation events had negative effects on the vegetation.
On the other hand, the NDVIs are positively correlated with the extreme precipitation index in warm-temperate deciduous broad-leaved forests. Furthermore, both the intensity and frequency of extreme high temperature have a negative effect on vegetation growth.

Possible Causes of the Vegetation Activity Changes to Extreme Climate Events
According to the aforementioned results, the vegetation NDVI reveals an overall increasing trend in the NSTEC from 1982 to 2012, even though spatial heterogeneity exists. This increase could be associated with extreme climate characteristics, vegetation types, and human activities. For instance, the vegetation growth of grasslands and croplands could be inhibited by changes in extreme high temperature events and extreme precipitation events in the Inner Mongolia Plateau and the Northeast Plain [32]. Similar results were also found in the case study of European in Baumbach et al. [70].
In addition, human activities, including urban land development, could dramatically affect local vegetation activity. One of the typical examples is in the Pearl River Delta. Despite favorable climate conditions, vegetation activity in this region still decreased in past decades. The main reason for this phenomenon could be attributed to large urban land development.
Furthermore, climate change, especially the combined effects of extreme precipitation events, as well as the frequency and intensity of extreme temperature events, could be attributed to changes in vegetation activity. For instance, vegetation activity revealed increasing trends in the Jianghan-Dongting-Lake Plain, the Yangtze River Basin, and the Wuyishan District due to the concurrent occurrences of enhanced high temperature extremes, precipitation extremes, and weakened cold temperature extremes.

Conclusions
Transect analysis has obvious advantages on understanding the effects of climate change on terrestrial vegetation activity by substituting space for time [25,26]. In this paper, the impacts of extreme climate events on vegetation activity were assessed over the period of 1982-2012 using the North South Transect of Eastern China (NSTEC) as a case study. Seven extreme climate indices from ETCCDI were selected to reflect changes of extreme climate events from aspects of frequency and intensity. We found that the frequency and the intensity of extreme high temperatures events showed an overall enhanced trend. However, we also found an enhanced frequency and weakened intensity of extreme cold temperature events, although they were not significant. Under such conditions, vegetation activity revealed an overall strengthened trend in the NSTEC over the past decades. Despite this, this trend is locally uncertain. The possible causes of the vegetation activity changes to extreme climate events were explored from the perspectives of vegetation type and human activity, as well as the combined effects of precipitation extremes and the frequency and intensity of temperature extremes.
This paper highlights the necessity to clarify the vital role of climate extremes on vegetation activity. However, there are also limitations in this study. For instances, vegetation activity was captured using NDVI as an indicator. Nevertheless, the NDVI still has some limitations. For example, it is easily saturated in rainforest areas and is susceptible to water vapor, dust aerosol pollution, and so on [30]. In addition, uncertainties still exist when sampling the NDVI based on meteorological stations from satellite-derived datasets at a coarse resolution. In addition, extreme climate events are frequently instantaneous and often exert severe influences on vegetation growth during a certain period. However, vegetation responses to extreme climate events are often lagged [71][72][73]. The lag time period is still uncertain due to the differences in vegetation type, as well as the magnitude, duration, and frequency of extreme climate events. Consequently, the impacts of extreme climate events on vegetation activity deserve further investigation by tacking them with different lag time periods in the future.