Introduction

There is a general agreement that global warming and related changes in the water cycle are likely to enhance the frequency and severity of extreme climate events (Holmgren et al. 2006; Wallace et al. 2014; Tao et al. 2014), which have already caused more severe floods and droughts (Kerr 2003; Schuster et al. 2012; Woodward et al. 2014; Kim et al. 2014), and subsequent significant impacts on natural ecosystems and human societies (Hasson et al. 2009; Rocklov and Forsberg 2009; Frank et al. 2015). A current view is that extreme climate events trigger vegetation shifts, as these events can induce generalized mortality disrupting a situation in which vegetation would either remain stable or follow successional replacement in equilibrium under stable climate conditions. There are important challenges imposed by the need of long-term studies, but a research agenda focused on the condition and inertia of vegetation under extreme events is well suited to understand the variety of responses (Lloret et al. 2012). Recently, long-term remote sensing data on spatial scales of sub-meter (LiDAR) to kilometers (AVHRR) provide a consistent and repeatable measurement for capturing the effects of many processes causing vegetation changes (Hui et al. 2008; Verbesselt et al. 2010). The Normalized Difference Vegetation Index (NDVI) is a widely used index characterizing vegetation growth conditions due to its close correlation with the fraction of photosynthetically active radiation absorbed by plant canopies, the Leaf Area Index (LAI), the potential photosynthesis and the biomass amount. Vegetation can be differentiated from water, cloud, snow and other objects by the reflectivity differences of the near-infrared and red wave bands (Liu et al. 2012). It can also reduce most irradiance changes relative to instrument calibration, sun angle, terrain, cloud shadow and atmospheric conditions (Tucker, 1979; Crippen 1990). Thus, the NDVI is a good proxy for evaluating the potential links of the variability in regional ecosystems and local ecotones with the spatio-temporal variability to climate conditions (e.g., Tourre et al. 2008; Koepke et al. 2010; Zhang et al. 2013; Frost et al. 2014).

Accordingly, increasing concern on extreme climatic events has triggered extensive research on vegetation dynamics, especially the relationship of the NDVI to climatic indices (e.g., Wang et al. 2010; Van Asch and Hoek 2012; Shevyrnogov et al. 2013). Actually, the impacts of climate changes on vegetation growth are diverse and not constant. Numerous studies have validated the responses of NDVI to climatic changes, and concluded that they differed among vegetation types (Breda et al. 2006; John et al. 2013). For example, Li and Tao (2000) demonstrated that correlation between annual NDVI and annual precipitation stronger over grasslands than over areas covered with shrubs and forests. Reichstein et al. (2013) also proposed that there were specific anticipated effects of climate extremes on forests, peatlands, grasslands, and croplands depending on the type of ecosystem. Liu et al. (2013) analyzed vegetation extremes of eight biomes in Amazonia, and found that extreme low precipitation shows the strongest correlation with declines in temperate broadleaf forest and temperate grassland.

Previous studies found that the vegetation type is an important factor in the strength of the correlations between the NDVI and climate indices. The sensitivity of different biomes to climatic changes has been proven to have different response times which might be the consequence of a more lasting or even profound impact after a certain time lag. Valladares et al. (2007) claimed that physiological and developmental responses to temperature shift of the vegetation (organism) from normal acclimation required a few days to weeks. Cavender-Bares and Bazzaz (2000) proposed that adult trees had a suit of morphological and physiological traits that could attenuate the impact of severe conditions. Miriti et al. (2007) believed facilitative interactions varied throughout the different life-history stages, which in turn might exhibit different sensitivity to extreme climatic events. Knowledge of dominant time lag between vegetation and climate extremes might be helpful to assess the resistance and resilience of vegetation and improve our understanding of vegetation vulnerability to climate change. However, the delayed responses of vegetation to climate extremes variation and the mechanisms are poorly understood.

Despite several efforts to evaluate changes in the climate indices and their impacts on NDVI, climate conditions of regions unobserved are usually predicted using limited meteorological datasets in previous studies. Interpolation methods are always used based on the hypothesis of spatial homogeneity. But climate conditions generally bear high uncertainty, because they are jointly affected by complicated topography, land use, human activity and other factors. Moreover, most former studies concentrated on several common climate indicators. How the changes in climate extremes affect NDVI remains unknown and require further investigation.

The Poyang Lake Watershed is one of the most frequently affected areas by a variety of flood and drought events in China, and is well known for its ecological and economic importance (Guo et al. 2008; Zhang et al. 2012). The Poyang Lake Basin is located in Jiangxi Province, which is the center of China’s rice growing region. During severe flood events, around 10,000 km2 of farmland and rural areas are still inadequately protected by levees (Shankman et al. 2006). Some 10 million people are at risk of being affected by floods and droughts. Historical severe drought and flood events have caused huge economic losses, numerous fatalities, and serious damages to rural and urban areas surrounding Poyang Lake. The Poyang Lake Basin is famous for abundant biomes, which, for example, provide sufficient bait and natural habitat for hundreds of thousands of migrating birds. The dramatic changes in precipitation and temperature over the Poyang Lake Basin play an important role in shaping the vegetation ecosystem (Kanai et al. 2002). The knowledge of the severity and spatial patterns of such ecological impacts becomes crucial (Seneviratne et al. 2012). The Poyang Lake Basin has experienced historical changes in climate conditions and vegetation cover, and thus provides an excellent study site.

At present, the temporal and spatial patterns of the responses of the NDVI to climate extremes have not been investigated for the Poyang Lake Basin. Therefore, the three objectives of this research are to: (1) investigate the dominant effects of monthly climate extremes on the NDVI for the period 1982–2006; (2) estimate the temporal and spatial heterogeneity of the relationship between typical climate extremes and the NDVI; and (3) analyze if “lag effects” between climate extremes and the NDVI exist.

Datasets and Methodology

Study Area

The Poyang Lake is located in the northern parts of Jiangxi province, China, and connected to the south bank of the middle and lower reaches of the Yangtze River (Wang et al. 2012). Poyang Lake is fed by the inflows of five tributaries (Ganjiang, Fuhe, Xinjiang, Raohe and Xiushui) and discharges into the Yangtze River through a narrow outlet located in Hukou (Fig. 1). The lake has a drainage basin area of 162, 225 km2, about 97 % of Jiangxi province (Shankman and Liang 2003). The wide alluvial plain is surrounded by hills and mountains with a maximum elevation of around 2200 m above mean sea level. In recent decades, the lake and the surrounding catchments have suffered from frequent occurrences of droughts and floods, especially in the 1990s and 2000s (Wang et al. 2008; Min et al. 2011).

Fig. 1
figure 1

Topography, location of the 14 meteorological stations, and the river network (containing five main tributaries: Ganjiang, Fuhe, Xinjiang, Raohe and Xiushui) of the Poyang Lake Basin

Full size image

Located in the core area of the Poyang Lake Basin is the Poyang Lake Wetland, which has been registered as a Ramsar wetland site of international importance, in 1992 (Zhang et al. 2012). There exists abundant wetland vegetation in and around the area, while the hills and mountains in the east, west, and south are covered by subtropical forest. The central area of the Poyang Lake Basin is relatively flat and mainly used for farming (mainly crops are rice and oilseed rape)or for urban dwellings. Hence, the vegetation in this area is often affected by human activities. Usually, shrubs and perennial or annual grasses (e.g. Carex cinerascens, Triarrhena lutarioriparia and Phragmites australis) cover the floodplains around the lake. The amount of biomass in the flood prone areas is often affected by the water level fluctuation of the lake (Liao et al. 2013). According to the generalized NDVI, the plants start growing rapidly in May, and the biomass peaks in August and September (Fig. 2).

Fig. 2
figure 2

Inner-annual cycle of the NDVI, mean temperature and precipitation in the Poyang Lake Basin, 1982–2006

Full size image

The major growth period of the vegetation in the Poyang Lake Basin takes place in spring and summer with influences from the subtropical monsoon. The monthly average temperature ranges from 5.8 °C in January to 28.7 °C in August (Fig. 2). The annual average precipitation is about 1712 mm. There is a significant transition of monthly precipitation in June, when it decreases dramatically after that. With respect to the annual cycle, about 54.2 % of the total precipitation falls in March to June, but only 15.9 % occur from October to the next January.

Datasets

Monthly characteristics of the plants are closely related to characteristics of the annual cycle of weather patterns, therefore changes in plant phenological events may signal important year-to-year climatic variations. The land surface conditions (1982–2006) of the Poyang Lake Basin are derived using the remote sensing data from the Advanced Very High Resolution Radiometer (AVHRR), provided by the Global Inventory Modeling and Mapping Study (GIMMS) group of NASA (Tucker et al. 2005). The final output have been screened for large-view angles and clouds, and partially corrected for atmospheric effects (Gurgel and Ferreira 2003). It is worthy noting that GIMMS NDVI provides biweekly composite NDVI, by applying a maximum value compositing (MVC) algorithm. It is believed that MVC algorithm can largely reduce effects of atmosphere and land anisotropy.

The monthly NDVI has been generated from the processed biweekly NDVI composites. The MVC procedure was used to minimize the effects of cloud contamination and to reconstruct the NDVI to a monthly time series. In this study, the monthly NDVI dataset with a spatial resolution of 8 km for a time period from January 1982 to December 2006 is used to investigate the vegetation characteristics. There are usually two ways to match remote sensing data and site observations. The first way is to extract remote sensing data according to site coordinates. The second way is to interpolate site observations with resolution of remote sensing data. Since interpolation may introduce extra uncertainty, site-centered mean value of 3 × 3 NDVI pixels was calculated to represent site NDVI. Mean NDVI values around meteorological sites were presented in Table 1. Applying this averaging method seems more reliable, as it reduces (i.e. smoothers) the spatial error and keeps the condition of the underlying surface vegetation at each weather station as uniform as possible.

Table 1 Geographic information of 14 meteorological stations and their mean monthly precipitation, average annual temperature, and average annual NDVI for 1982–2006. The mean monthly precipitation and average annual temperature are calculated from daily meteorological data. The average annual NDVI is calculated using biweekly remote sensing data
Full size table

Considering the continuity and validity of the datasets, 14 meteorological stations of Jiangxi province were chosen for research, each of which has a long-term data from 1982 to 2006. Detailed information of 14 stations as well as statistical information of precipitation and temperature is summarized in Table 1. For each station, the China Meteorological Data Sharing Service System (http://cdc.cma.gov.cn/) has provided daily maximum air temperature (Tmax), daily minimum air temperature (Tmin) and 24 h precipitation (PRCP) for the time period of 1982–2006.

In the present study, crucial indices (2 precipitation indices and 8 temperature indices) of climate extremes are selected and calculated using the software RClimDex (Version 1.0) as developed and maintained by Zhang Xuebin and Yang Feng at the Climate Research Branch of the Meteorological Service of Canada (Hyndman and Fan, 1996; Zhang et al. 2005). All indices have been recommended by the CCl/CLIVAR Expert Team for Climate Change Detection Monitoring and Indices (ETCCDMI). The precipitation and temperature data have been quality controlled before the selected indices have been computed in advance. The details of the ten selected indices are shown in Table 2.

Table 2 Descriptions of the selected ten extreme climate indices
Full size table

Methodology

In this study, correlations between climate indices and NDVI were estimated by the Pearson correlation coefficient. Pearson’s R measures the strength and direction (decreasing or increasing, depending on the sign) of a linear relationship between two variables. It ranged from −1.0 to 1.0, with higher values indicating better agreement, and is given by

$$ \mathrm{R}=\frac{{\displaystyle \sum_{i=1}^N\left({O}_i-\overline{O}\right)\left({P}_i-\overline{P}\right)}}{{\left[{\displaystyle \sum_{i=1}^N{\left({O}_i-\overline{O}\right)}^2}\right]}^{0.5}{\left[{\displaystyle \sum_{i=1}^N{\left({P}_i-\overline{P}\right)}^2}\right]}^{0.5}} $$
(1)

where O i and P i denote the monthly value of the two variables, and \( \overset{-}{O} \) and \( \overset{-}{P} \) denote the means of the two variables, respectively. In general, this paper addresses the comparison of climate indices (P) with the NDVI (O) based on the same set of conditions (i.e., a pairwise comparison) for a 25-year period with N time increments for an arbitrary duration (here: at seasonal and monthly time scale).

Results

Relations Between NDVI and Extreme Indices on a Monthly Basis

According to the annual cycle of climate extremes, TXn (Minimum Tmax), TNn (Minimum Tmin) and TXx (Maximum Tmax) all increased by about 0.05 °C·year−1, less than the increase of TX90p (Warm days, 0.38 °C·year−1) and TN90p (Warm nights, 0.36 °C·year−1). Precipitation indices of RX1day (Maximum 1-day precipitation amount) and RX5day (Maximum 5-day precipitation amount) increased by about 0.21 mm·year−1and 0.44 mm·year−1, respectively. However, the increase of annual temperature and precipitation did not lead to a significant increase or decrease in NDVI (rate of change was smaller than 0.01 %).

Correlation analysis using deviation from the value of normal year (monthly average of the 25 years data) for every climatic variable and NDVI will give the more exact information on direct effects of climate extremes on vegetation. Accordingly, relations between NDVI and extreme indices on a monthly basis were investigated using deviation of long term dataset (Fig. 3a). Figure 3a shows a decreasing strength of the monthly NDVI response to extreme indices in order of TXn, TNn, TNx (Maximum Tmin), TXx, RX5day, RX1day, TN10p (Cool nights), TX10p (Cool days), TX90p and TN90p. Correlations between NDVI and all the former seven indices are significant (at the 95 % confidence level). Namely, NDVI in the study region was mainly dominated by the extreme minimum night-time and day-time temperatures indices. Monthly NDVI generally responds positively to temperature indices and negatively to precipitation indices.

Fig. 3
figure 3

Correlation coefficient between the selected ten climate indices and the NDVI on monthly time scale, 1982–2006: a whole year (N = 300, r0.05 = 0.11, r0.01 = 0.15), the horizontal dashed lines (green) denote the 95 % confidence interval, solid line (red) denote the 99 % confidence interval; b four seasons (N = 75, r0.05 = 0.23, r0.01 = 0.30), blank denotes no significant correlation between climate indices and NDVI (at the 95 % confidence level)

Full size image

Meanwhile, relations between monthly NDVI and extremes indices in different seasons were evaluated and presented in a color image (Fig. 3b). The former six climate indices (TXn, TNn, TNx, TXx, RX1day, and RX5day) show significant positive correlations with NDVI in spring and autumn (at the 95 % confidence level). However, correlations between NDVI and the former four temperature indices (TXn, TNn, TNx and TXx) are not significant in summer and winter. On the other hand, NDVI presents very significant negative response to RX1day (Rmean = −0.37, p < 0.01) and RX5day (Rmean = −0.47, p < 0.01) in summer and winter.

Because TN10p does not impact NDVI significantly in either season (Fig. 3b), the former six indices (TXn, TNn, TNx, TXx, RX1day and RX5day) were chosen according to the value of correlation coefficients in Fig. 3a to evaluate the spatial heterogeneity of relations between NDVI and climate extremes.

Spatial Distribution of the Correlation Coefficients of NDVI and Extreme Indices

For simplicity, temperature indices of TXn, TNn, TNx and TXx, precipitation indices of RX1day and RX5day were selected here as typical extreme climate indices. Spatial patterns of relations between NDVI and typical indices were investigated among different seasons on a monthly time scale (Fig. 4).

Fig. 4
figure 4figure 4

Spatial patterns of correlation coefficients between monthly mean NDVI and extreme climate indices (a: TXn (°C); b: TNn (°C); c: TNx (°C); d: TXx (°C); e: RX1day (mm); f: RX5day (mm)) in different seasons. The height of each bar presents the value of correlation coefficient relative to the reference value in each legend, respectively. Bars above the bottom denote positive values, bars below the bottom denote negative values. N = 75, r0.05 = 0.22, r0.01 = 0.30

Full size image

The authors find that in spring and autumn, NDVI variation responses to TXn, TNn, TNx and TXx positively and significantly (Fig. 4a, b, c and d). But relations are much weaker in summer and winter, they also vary in sign.

Moreover, the impacts of RX1day and RX5day on NDVI differed among seasons (Fig. 4e, f). Positive correlations were detected between NDVI and the two extreme precipitation indices in spring and autumn for almost all meteorological stations, while most of the correlations were negative in summer and winter.

According to the spatial heterogeneity of relations between NDVI and climate extremes in specific season, we can conclude that, NDVI in the middle basin is less correlated to climate variation than NDVI in other regions of the study area. These variances are much more apparent for RX1day and RX5day (Fig. 4e, f). For instance, the mean value of absolute correlation coefficients between spring NDVI in meteorological stations of the middle basin (Boyang, Nanchang and Zhangshu) and RX1day is 0.08. The significance level is much lower than that of other stations for the same period (Rmean = 0.31, P < 0.01). Distribution of relations between NDVI and other indices also presents the similar patterns.

Delayed Responses of NDVI to Extreme Climate Indices

In order to recognize lag effects of climate extremes, the authors calculated the correlation coefficients between monthly NDVI and climate extremes with different time lags (Table 3). It is easy to conclude that TXn without time lags (or with a lag less than 1 month) have the most significant effects on monthly NDVI. Strongest relations between NDVI and other temperature indices of TNn, TXn and TXx were found with a lag of 1 month. On the other hand, NDVI exhibited the strongest responses to precipitation indices of RX1day and RX5day with a lag of 2 months.

Table 3 Cartogram of multiple correlation coefficients between the basin averaged NDVI and monthly climate extreme indices with different time lags
Full size table

Discussion and Conclusion

Monthly analysis of relations between NDVI and climate extremes proposed that extreme temperature indices (TXn, TNn, TNx and TXx) have stronger impacts on NDVI than precipitation indices (RX1day and RX5day) in Poyang Lake Basin. This conclusion is consistent with the hypothesis that for southern humid vegetation, where water is not a limiting factor for vegetation growth, higher temperature rather than increased precipitation is associated with the increase of NDVI (Zhang et al. 2010; Piao et al. 2014). This weak correlation between NDVI and precipitation indices may be attributed to the coincident reduction of incoming solar radiation during wet period (Beer et al. 2010; Piao et al. 2003).

Actually, relations between NDVI and climate extremes vary in different seasons. The correlations of NDVI with temperature indices are weaker in summer and winter than in spring and autumn. High temperature increases the efficiency of plant photosynthesis and water use efficiency, and promotes the growth of vegetation in spring. By contrast, in autumn, metabolism of vegetation weakens with the decrease of available energy. On the other hand, high temperature may lead to increase of evaporation and decline of soil moisture in summer. The vegetation fraction of annual herbaceous plants, as well as deciduous forests decreased due to the phenological feature of vegetation rather than climate change in winter (Chen et al. 2006).

Compared to extreme temperature indices, significant negative correlations between monthly NDVI and extreme precipitation indices (RX1day and RX5day) in summer and winter implied that increased precipitation may lead to a decline of NDVI during these two seasons. As discussed above, there are intensive river network and robust irrigation system in southern plains of China. Meanwhile, precipitation in this region concentrated mainly in summer, more rainfall might cause soil supersaturation and decrease bioactivity in the wet season. Landslide caused by big flood and excessive precipitation is one of the biggest threatens of mountain vegetation. Inundation of vegetation in lowland areas, especially in the areas around the Poyang Lake by flood can also lead to the decrease of NDVI. In addition, much cloud in rainy day also reduced the solar radiation and inhibited photosynthesis of plants. In winter, much precipitation might result in a decrease in soil temperature, and decreased soil temperature would probably become the key factor for restraining vegetation growth, according to Liebig’s law of the minimum (Zhang et al. 2013; Von Liebig 1842).

Furthermore, relations between NDVI and typical extreme climate indices exhibited spatial heterogeneity among different meteorological stations. Vegetation in middle basin was less sensitive to climate change than other regions (Fig. 4). To the best of the authors’ knowledge, the flat middle basin covered with more arable lands is more likely to be affected by artificial factors. Farming processes such as seed, weeding, fertilization and harvest, and so on would possibly change the greenness of plants. These changes were not controlled by climate conditions exactly. Moreover, concentrated constructing lands (urban land, factory and mine, road, oil field and airport, and so on) in middle area would not respond to the climate changes either.

It is noteworthy that vegetation ecosystems adapt to climate change attributes to physiological, anatomical, and functional strategies. These strategies are mainly reducing water loss, respiration costs, photosynthetic activity, and growth rate. But it takes a certain time lag (Potter and Brooks 1998; Vicente-Serrano et al. 2013). In the present study, a notable discovery was that extreme temperature indices with a lag of 1 month (or less than 1 month) had the strongest impacts on NDVI. Meanwhile, most significant response of NDVI to extreme precipitation indices was observed with a lag of 2 months. Delayed responses of monthly NDVI in Poyang Lake Basin to climate extremes were detected in this study, but the specific reasons and the differences of response patterns need further investigation and discussion.

The authors’ investigation of relationship between climate indices and vegetation, as well as the lag effects of temperature and precipitation on NDVI are consistent with other studies (Gu et al. 2009; Zhang et al. 2011). Different scholars diverged in their climate change impact assessments because multiple processes at different temporal, spatial, and organizational scale could influence ecosystem dynamics in conflicting directions. Biotic and abiotic factors not explicitly dealt with in the authors’ study include the competition and adaptation among species, soil properties, soil moisture content, topography, fire occurrence, grazing management and urbanization, all of which could be important in affecting vegetation activity at the site-specific level (Yang et al. 1998). Estimating change from remotely sensed data however was not straightforward, since time series contained a combination of seasonal, gradual and abrupt ecosystem changes occurring in parallel (Verbesselt et al. 2012).

Our present works together with previous studies highlighted the role of climate extremes to the NDVI and confirmed the high vulnerability of ecosystem in the Poyang Lake Basin. Understanding the characteristics of climate extremes and the mechanisms how they affect vegetation is crucial for improving the knowledge of vegetation vulnerability to climate fluctuations and climate change.