Climate change and ecological engineering jointly induced vegetation greening in global karst regions from 2001 to 2020

Vegetation dynamics play a dominant role in the global carbon cycle and climate, especially in vulnerable karst ecosystem. Many studies have examined the past several decades changes in vegetation greenness and the associated with climate drivers. Yet, few studies have analyzed the vegetation change in global karst regions particularly in the last decades when climate change and anthropogenic disturbance widely occurred. In this study, we investigated the spatio-temporal variations in vegetation dynamic using the Seasonally Integrated Normalized Difference Vegetation Index (SINDVI) and examined their relationship with climate changes using correlation analysis, the ordinary least squares method investigate the variation trends and the Mann-Kendal test to detect the turning points from 2001 to 2020. As expected, there were greening trends in global karst SINDVI from 2001 to 2020, with significant increasing trends in China (range = 0.836, P < 0.05), Europe (range = 0.456, P < 0.05) and many other regions. According to correlation analyses, SINDVI was water-limited in arid and semi-arid regions, such as Middle East and central Asia, and temperature-limited in northern high-latitude. Our results suggest that anthropogenic activities were mainly responsible for the increasing vegetation greenness in tailoring management measures (e.g., Ecological Engineering, the Grain to Green Project) in China and Europe, and intensive farm in Middle East. Coupling warming temperature and increasing precipitation, southeastern Asia and Russia showed increasing trends in SINDVI. In general, climate factors were the dominant drivers for the variation in vegetation greenness in globally karst regions during research period.


Introduction
Karst physiognomy is created by mechanical erosion and chemical solubility (Liu et al. 2012), and there are substantive movement and large reciprocity of water and rock in karst areas. The global area of karst regions is approximately 2.2 × 10 7 km 2 , around 15% of global land areas (Falkowski et al. 2000).
Karst dynamic system is characterized by the epikarst vegetation interaction, providing more than 25% of groundwater for global population (Cao et al. 2017). The karst ecosystem is particularly vulnerable mainly due to the alternation of seasonal drought and flood, soil erosion, surface subsidence, and rocky desertification (Huang et al. 2000;Yuan 1993). Under intensive anthropogenic disturbances and extreme climate changes, fragile karst terrains have been facing the threat of serious ecosystem degradation. This type of degradation often manifests as a reduction in vegetation cover and the consequence of soil erosion, is one of the most serious environmental problems in the world (Yuan 2000;Yue et al. 2010). Therefore, the dynamics of vegetation cover have been considered as an essential indicator of the ecosystem changes in karst regions (Kelly et al. 2011;Piao et al. 2015).
Vegetation is the main component of the terrestrial ecosystem and it plays an important role in global carbon cycle, water and energy balance. Under global climate change, plant's photosynthesis responds to warming temperature and extreme climate events. Analysing the relationship between vegetation greenness with temperature and precipitation is essential for designing policies to mitigate the impacts of extreme climate and anthropogenic activities on ecosystem and human society (Begue et al. 2011;Peters et al. 2012). Chen et al. (2019) and De Jong et al. (2013) have attempted to examined the factors responsible for the vegetation greenness by investigating the response of vegetation to climate changes, as represented by variables such as temperature and precipitation. However, different geological backgrounds may result in significant differences in soil and moisture regimes for the weathered crust, which restricts the spatial allocation of regional water and soil, as well as constrains the geochemical cycling processes of nutrients, leading to impacts on vegetation greenness (Florinsky and Kuryakova 1996;Seddon et al. 2016). As for the driving factors for spatiotemporal changes in vegetation, natural elements such as elevation, precipitation and air temperature are usually consider to explore the spatial distribution of greening and browning trends (Chen and Pan 2010;Jiang et al. 2017;Ndayisaba et al. 2016). Furthermore, anthropogenic effects have a significant impact on vegetation dynamics and distribution (Martínez and Gilabert 2009). Many studies have investigated the relationship between human practices and vegetation variations at different spatial scales Sun et al. 2015;Tong et al. 2017). The vegetation in karst regions is an essential part of the global ecosystem and has strong spatial heterogeneity in different regions (Cao et al. 2017;Vermote et al. 2002). Understanding the characteristics of temporal and spatial variation in vegetation cover and quantitatively analyzing the correlation between its changes with geographical and climate factors are critical to study the relationship of vegetation, climate change and human activity to unveil the process of global environmental evolution and to predict future development trends (Aguilar et al. 2012;Pouliot et al. 2009).
The remote sensing (RS) technology have been widely applied in ecological research (Cabello et al. 2012;Feng et al. 2010;Pettorelli et al. 2005;Xie et al. 2008). Moderate Resolution Imaging Spectroradiometer (MODIS) data have been widely used since the launch of Terra platform in December 1999. The Normalized Difference Vegetation Index (NDVI) products from MODIS have been widely applied for vegetation research (Eisfelder et al. 2012;Fensholt et al. 2012). In recent decades, many studies have used NDVI data to monitor and evaluate variations in vegetation over different spatial and temporal scales, including research into vegetation phenology and 1 3 terrestrial carbon cycle (Azzali and Menenti 2000), the characteristics of different crop species (Jakubauskas et al. 2002), forest fires (Jia et al. 2004;Leblon et al. 2001), the impacts of human activities on vegetation dynamics and distribution (Possingham et al. 2016), and vegetation cover characteristics and its response to climate change (Fensholt et al. 2009;Ma and Frank 2006;Pelkey et al. 2003;Wen et al. 2017). However, these impacts were rarely considered in the greenness trend of vegetation in global karst area, so as the influences of natural process and human activities on vegetation change. The study could provide foundations for the predicting of vegetation growth trends, ecosystem evolution and environmental change in the global karst regions. What's more, due to the seasonal variations in solar radiation and vegetation growth, Seasonally Integrated Normalized Difference Vegetation Index (SINDVI) works better than NDVI in reflecting inter-annual mutability and integrative vegetation cover trends (Sun et al. 2010). In addition, SINDVI, defined as the sum of NDVI values (NDVI > 0.1) in all time phases for each pixel during the growing season, can effectively eliminate the impact of regions with bare and sparse vegetation Stow et al. 2003). SINDVI has been increasingly applied in exploring the land use and vegetation change, such as in Ejin Oasis (Hu et al. 2015), and the effect of vegetation change on albedo , and land surface temperature (Song et al. 2018). Based on previous research, we aimed to provide the different vegetation in the global karst regions from the beginning of the 21 st century. There were two main research objectives of the current study: (1) to investigate the overall trends of vegetation greenness and climate variables, applied Mann-Kendall to detect potential turning points of SINDVI time series data, and (2) to analyzed changes in SINDVI and their relationships with climate factors and land cover change.

Study area
Global karst regions, with a total area of 2.2 × 10 7 km 2 were chooses as the study area in this research (Fig. 1). The World Karst Map V3.0 was obtained from Geography and Environmental Science, University of Auckland (https:// www. fos. auckl and. ac. nz/ our_ resea rch/ karst/ index. html). To improve the accuracy of boundary information for the karst areas, the World Karst Map V3.0 data were integrated with detailed karst data in each continent by Karst Scientific Data Center (http:// www. karst data. cn). There is a prominent karst zone around the world from western China through the Middle East to the Mediterranean and along the coast of the western Atlantic. There are   Soil (2022) 475:193-212 195 three major karst regions in the world: East Asia, the Mediterranean coast and eastern America. In specific, these include southwestern China, northern Vietnam, central and southern Europe, the central plateau of France, Ural in Russia, northeastern America, Cuba and Jamaica. According to climatic conditions, karst areas are divided into four categories: the glacial karst area, Eurasian plate karst area, North American plate karst area, and Gondwana continental karst area. Most karst regions have suffered from a series of ecosystem damages, including rocky desertification, soil erosion, vegetation degeneration, and productivity decline. In particular, Mexico, Middle East, Southeast Asia (Ford and Williams 2007), southwest China (Jiang et al. 2014) and Mediterranean basin (Yassoglou 2000) are the most ecologically fragile areas in the world (Kelly et al. 2011).

Vegetation index data
This study selected the MODIS/Terra NDVI (collection 6) dataset, which was based on cloud-free composites images from 16 days with a spatial resolution of 1 km (Wen et al. 2017). The NDVI data were supplied as a Level 3 product projected onto a 0.05-degree (5600 m) Climate Modeling Grid (CMG) with monthly temporal resolution, provided by Land Processes Distributed Active Archive Center (LP DAAC), the NASA Earth Science Data and Information System project (https:// doi. org/ 10. 5067/ MODIS/ MOD13 C2. 006). SINDVI was defined as the sum of NDVI values greater than certain threshold (0.1 was selected in this study) in all time phases throughout the year of each pixel Ma and Frank 2006;Stow et al. 2003), and it was calculated by summing monthly NDVI values when NDVI > 0.1 for each pixel as bellow: where i represents the ith month. (1)

Meteorological data
Meteorological parameters including air temperature and precipitation were obtained from GLDAS_ NOAH025_M data, Goddard Earth Sciences Data and Information Services Center (https:// disc. gsfc. nasa. gov/ datas ets/ GLDAS_ NOAH0 25_M_ 2.1/ summa ry? keywo rds= GLDAS), which were processed in monthly 0.25-degree resolution. The GLDAS air temperature was instantaneous observation per 3 h, which firstly generated into monthly temperature by averaging, and then generated to the scale of annual temperature. While the GLDAS precipitation was the sum of rain and snow amount per 3 h, which was firstly calculated to monthly precipitation and then accumulated to annual values. GLDAS-2.1 data were archived in machine independent and self-describing NetCDF format. MATLAB R2016b was used to convert data format and synthesize meteorological data to a yearly scale.

Land cover data
To monitor and quantify the land cover change over the full time period in global karst regions, the Terra and Aqua combined MODIS (MCD12C1, V 6.0) data product (https:// doi. org/ 10. 5067/ MODIS/ MCD12 Q1. 006) were used. Maps of the International Geosphere-Biosphere Programme (IGBP), University of Maryland (UMD), and Leaf Area Index (LAI) classifications schemes are provided at annual scale at 0.05 degree (5,600 m) spatial resolution for the entire globe, and the dataset is available from 2001 to 2019.

Reconstruct NDVI data
The Mean Value Iteration filter (MVI), a simple and effective method was used to reconstruct NDVI and reduce the noise caused by the atmosphere water vapor and clouds. The reconstruction process can also eliminate some NDVI outliers in the original image, improving the homogeneity of each land cover type and spatial uniformity of NDVI image (Ma and Veroustraete 2006). A high quality of NDVI time series were calculated as follows:

3
where i means the i th monthly NDVI. 10% of the multiyear NDVI average of each pixel was set at a threshold (ΔT), when Δi is greater than ΔT, NDVI will be replaced by NDVI i -(NDVI i-1 + NDVI i+1 ) /2. Finally, when all Δi are less than ΔT, the iteration ended.

Time series analyses
The monthly SINDVI, air temperature, and precipitation from 2001 to 2020 were processed to synthesize the annual data, therefore time series in each pixel had a length of 20 years. In order to estimate the inter-annual trends of SINDVI, meteorological parameters, and area for different land use types, we used an ordinary least squares method to calculate their slopes and ranges in different karst regions (Hu et al. 2015;Peng et al. 2018). The slopes, representing the change rates of each pixel from 2001 to 2020, were calculated as follows: where n is the cumulative number of years in the monitoring period; i is the serial number of years and i = 1, 2, … 20; and X i is the SINDVI, air temperature, or precipitation in the ith year. Slopes > 0 indicates that variables are increasing in n years, while slopes < 0 means a downward trend. The more the slope values are near to 0, the much less changes occurr in the trends (Jafary et al. 2018). To detect statistical significance of the variables trends in karst regions from 2001 to 2020, a Pearson correlation analysis was used and significance level was set to 0.05 (P < 0.05) in this study.
The range was used to assess the magnitude of the changes. The value calculated for each pixel indicates the change between 2001 and 2020.
where n is the length of study periods. (

Mann-Kendall test
Mann-Kendall analysis is a trend estimation method of time series based on nonparametric rank (Kendall 1975;Mann 1945), which is suitable for elastic processing of outliers (Lanzante 1996). The sequential version of the Mann-Kendal rank statistics can be performed using the following steps: a. Replace original values of the series x i by their ranks y i , arrange in ascending order. b. The magnitudes of y i (i = 2, …, n) were compared with y j (j = 1,…,i-1). At each comparison, the number of cases y i > y j was marked by r i . c. For the time series X, which containing n samples, a rank sequence (S m ) was first constructed as follows: where m is length of study periods; S m is the sum of all individual S statistics (r i ) for all years. d. r i is the cumulative number of the i th sample when y i > y j and it was defined as follows: e. Considering that the time series are independent and random, the statistical UFm is defined as follows: f. E (Sm) is the mathematical probabilities and Var (Sm) is the variances. E (Sm) and Var (Sm) were calculated as follows: where n is the number of observations. A positive value of UF m indicates an increasing trend, which is a series of positive statistics calculated by range x 1 , x 2 , x 3 , …, x n . Subsequently, calculating the inverted sequence of time series X (x n , x n-1 , …, x 1 ) by repeating the above process, to obtain the reverse statistical sequence UB m . If UB m and UF m intersect, their intersection may be a breakpoint (P < 0.05).

Correlation analysis
In order to explore the relationships between SINDVI and meteorological parameters (air temperature and precipitation), Pearson's correlation coefficients (r) were calculated as follows: where r xy is the correlation coefficient between x and y, and the values range from -1 to 1; x i and y i are the values of the i th year; x and y are their averages for 20 years.
Partial correlations were conducted to assess the dominant factor of each pixel. When estimating the partial correlation between temperature and SINDVI, the influence of precipitation will be considered as a constant. For example, if the partial correlation coefficient between SINDVI and temperature was greater than the correlation coefficient between SINDVI and precipitation, it indicates that temperature had larger influence on SINDVI. The partial correlation coefficients were calculated as follows: where r yz·x is the partial correlation coefficient between y and z without consideration the influence of x; and r xy and r xz are the simple correlation coefficient between x and y. For each pixel, the dominant by comparing the partial coefficient between NDVI and temperature or precipitation. Different significance levels were also set for difference confidence interval (Extremely significant, P < 0.01; very significant, P < 0.05; significant, P < 0.1; not significant, P >0.1).

Spatiotemporal variation in SINDVI
Temporally, SINDVI showed an obvious increase trend during the period of 2001 -2020 in karst regions, with an average of 3.99. While, there was large spatial heterogeneity (Fig. 2a), with the largest average SINDVI values appearring in China and Europe.
There were remarkable and widespread increment of SINDVI in the global karst areas during the research period (Fig. 2). In general, approximately 72.35% of the pixels showed upward trends. Meanwhile, there were also spatial differences between (11) r yz⋅x = r yz − r yx r zx √ 1 − r 2 yx 1 − r 2 zx various regions. Especial the obvious disparity of the increasing trends appeared in tropical regions, subtropical humid regions and temperate regions. Notably, the increasing trend of the SINDVI in China was the most obvious, which increased 0.836 from 2001 to 2020 (Fig. 2b). In contrast, the decreasing trends of SINDVI mainly distributed in arid areas, such asCentral Asia (-0.020) and South America (-0.006).
The positive trends of SINDVI at low latitudes in the Northern Hemisphere were generally weaker than in middle and higher latitudes (Fig. 2b). In additional, SINDVI in the Northern Hemisphere grew 0.28, while there was no significant trend in the Southern Hemisphere.
From 2001 to 2020, SINDVI in the global karst regions showed an increasing trend with large volatility ( Fig. 3b) (slope = 0.011/a, P < 0.01). The sequential version of the Mann-Kendall tests were conducted to detect the breakpoints (Fig. 3a). The forward (UF) and backward (UB) statistics sequence curves intersected near the year of 2012, indicating a significant change in SINDVI around 2012. Before the breakpoint, SINDVI exhibited a stronger positive trend with a rate of 0.011/a (P < 0.05). However, after the breakpoint, the SINDVI exhibited a relatively weaker positive trend with a rate of 0.009/a (P < 0.05). In order to determine the patterns of inter-annual trends, we calculated the trends of both before the breakpoint (BBP) and after the breakpoint (ABP) for all regions and classified them into four categories (Fig. 4): increased in BBP but decreased in ABP, decreased in BBP but increase in ABP, upward trends in both BBP and ABP, and downward trends in both BBP and ABP. The results showed that the NINDVI in Australia, Middle East, Russia and South America showed climbed in BBP but decreased trends in ABP. China, North America, Europe, Africa and Southeast Asia showed sustained increasing trends from 2001 to 2020 at the rate of 0.034/a, 0.028/a, 0.011/a, 0.006/a and 0.005/a in BBP, respectively. After that, with the exception of North America, the growth trends of the other three regions were stronger after the turning points. Central Asia showed a continuous downward trend, with the breakpoints in 2009, and the decreasing trends in ABP were weaker than those in BBP (Table 1). Figure 4 illustrated the spatial patterns of the trends in precipitation and temperature in global karst areas from 2001 to 2020. The annual average temperature showed an increasing trend of 0.06 °C/a (P < 0.05) (Fig. 4a). Spatially, all the karst regions experienced air temperature warming trends, particularly in Africa and Middle East where the magnitude of the temperature trend exceeds 2 ℃ during research period. Conversely, precipitation experienced slightly decreased at the rate of -0.36 mm/a (P > 0.05) (Fig. 4b), only China and Southeast Asia had increasing trends. The maximum of temperature and precipitation appeared in 2010, in a good agreement with the long-term trends in precipitation in previous study of Adler et al. (2017).

Relationship between SINDVI and meteorological factors
In order to determine the spatial correlation between SINDVI with precipitation and air temperature, correlation coefficients were calculated for each pixels (Fig. 5). About 21.12% of SINDVI pixels was affected by precipitation (P < 0.05) in global karst regions. (Fig. 5a). Correlation coefficients greater than 0.5 mainly occurred in Middle East peninsula, Mexico, Southwest China, Southern Australia and Southern Africa (Fig. 5c). Approximately15.22% of all pixels indicated that temperature had significant influenced on the variation in SINDVI (Fig. 5b), with r values larger than 0.5 mainlyin Central Eurasia, Eastern Mediterranean, Eastern and Northwest Canada, Southwestern China and Eastern United States (Fig. 5d).
The partial correlation coefficient was calculated for each pixel to determine the main factor of SINDVI change (Fig. 6), and only 21.06% of the areas are significance (P < 0.05). The karst regions of several different regions were analyzed below.
In the arid regions in the Southern Hemisphere, including Australia and South Africa, SINDVI dynamics were synergistic with precipitation change (Fig. 6). Near the equator, where the precipitation is abundant throughout the year, temperature seems to play a powerful role in the fluctuation of SINDVI.
In the high latitude areas of the Northern Hemisphere, temperature acted as the dominant factor affecting the SINDVI. All of the rising temperature, stronger photosynthesis and longer growth season, facilitated the plant growth vigorously and then promoted SINDVI.

Trends of land cover change in karst regions
This study focused on five land use types in the karst regions: cropland, forest, shrub land, grassland and bare land, and the dominant types in global karst regions were grassland and forest. Between 2001 and 2019, grassland area has increased from 438.986 × 10 4 km 2 to 443.810 × 10 4 km 2 with a rate of 0.2868 × 10 4 km 2 /a, cropland area has a slightly increasing at the rate of 0.0936 × 10 4 km 2 /a. In contrast, the shrub land, forest and bare land exhibited decreasing trends with rates of -0.1103 × 10 4 km 2 /a, -0.1598 × 10 4 km 2 /a and -0.0771 × 10 4 km 2 /a, respectively (Fig. 7). At the regional scale, the various land cover types underwent different changes at different regions, and a detailed statistical analysis of the five main land cover types were performed for ten karst regions (Fig. 8). The cropland area reduced in Europe, China, Africa and Southeast Asia. In terms of forest, with exception of Russia, Africa, North America, South America and Middle East, all other regions showed increasing trends. In additional, most regions experienced the shrinking of shrub land, except three regions (Central Asia, Australia and Southeast Asia). For grasslands, besides five regions (Central Asia, Australia, Europe, Middle East and Southeast Asia), all other regions showed upward trends. Regarding bare land, China, Africa and North America remarkably dropped, conversely; Middle East had positive trend; other regions had no visible changes.

Discussions
In this study, the spatio-temporal variations of SINDVI over the global karst regions show a greening trend during 2001-2020 (slope = 0.011/a, P < 0.01), which is consistent with previous study , Piao et al. 2020). There are two distinct periods with growing trends in SINDVI (Fig. 3): a significant increase of SINDVI from 2001 to 2012 (slope = 0.011/a, P < 0.01) and a relatively weaker grown from 2012 to 2020 (slope = 0.009/a, P < 0.01). Changes in the latter period was largely contributed by the browning vegetationin Australia and Middle East after 2012, and Russia after 2011, respectively. This offset the greening trend before 2012 in other karst regions, and this finding is also consistent with previous reports (Long et al. 2018;Piao et al. 2020).
Our results indicate that climate change is the most important factor influencing vegetation dynamics in karst regions (Fig. 5), which is similar with previous findings (Piao et al. 2014).During the research period, two meteorological factors (air temperature and precipitation) had significantly positive relationships with the inter-annual SINDVI variation in the global karst regions, indicating that the increase of temperature and precipitation could promote the growth of vegetation.
Despite the positive correlation between temperature and SINDVI in most karst areas of the Northern Hemisphere, including China, Europe and North America, the significant negative correlation also appeared in most dry areas and the middle-latitude areas, such as South Africa, Southern Australia, Middle East (Fig. 5d). In the dry areas, the persistent warming and decreasing of precipitation can seriously aggravate the water deficit and evapotranspiration loss, bringing heavy damage to plant growth (Zhou et al. 2015). Because of obvious spatiotemporal differences, this study chosen several regions with obvious variations in SINDVI to explore the changes in global karst areas.
Increasing trends in the SINDVI Vegetation in the karst areas of China has increased in the last decade (Fig. 2b). The relationship between SINDVI and temperature was closer than that between SINDVI and precipitation (Fig. 6). This might be related to the location of Southwest China, which is the subtropical monsoon climate zone, with adequate precipitation. Therefore, small changes of precipitation may not cause seriously change on vegetation growth (Lai et al. 2020;Hou et al. 2015;Zhao et al. 2017). In addition, the dual hydrological structures of surface and underground in karst region in Southwest China lead to drainage of precipitation through leakage and runoff (Tong et al. 2016), indicating the precipitation has limited effect of vegetation growth. As for the reasons for the greening in this area, on the one hand, governments have taken some effective measures (e.g., relocation and ecological migration) to reduce land abandonment and land degradation caused by the outflow of rural population ). On the other hand, ecosystem restoration projects (for example, the Grain to Green Project, the Karst Rocky Desertification Restoration Project, and the Natural Forest Protection Project) have been implemented (Mueller et al. 2014;Tong et al. 2018;Zhang 2008) and afforestation is an important factor in turning green, especially since 2008.
The similar political decisions appeared in Europe, the afforestation and forest management have accelerated the growth of vegetation (Naudts et al. 2016), causing the dramatic increase in the forest area of 0.3485 × 10 3 km 2 /a (Fig. 8) and contributing to the upward trend in the SINDVI. Furthermore, the rising temperature (Range = 1.46 ℃) might have promoted the growth of vegetation, which may explain the greening trend. Previous study indicated that the positive effect of meteorological parameters change in the northern high latitudes were attributed to enhanced photosynthesis and prolonged growing seasons (LeBauer and Treseder 2008;Liu et al. 2015;Menzel et al. 2006;Xu et al. 2013). In central and northeastern America, the increasing temperature was the dominant factor affecting the SINDVI (Fig. 5d), and this upward trend was contributed by vegetation propagation caused by global warming (Zhu et al. 2016). Moreover, agricultural fertilization and irrigation would also have a significant impacts on the trend of SINDVI dynamics (Neigh et al. 2008). In contrast, the effect of precipitation was stronger than that of temperature in southern North America (Fig. 6). Figure 3 shows that the BP of the inter-annual variation in SINDVI occurred in 2012, the SINDVI showed an increasing trend with a slope of 0.028/a before breaking point (BBP); while, the upward trend was relatively slight (slope = 0.011/a) after the breakpoint In arid and semi-arid areas, the land use types are mainly savanna and desert. Although the trend in vegetation cover is similar to that of global level, their extents of changes are differently. Precipitation is characterized by regular seasonal drought and large inter-annual fluctuation. Due to the non-irrigated crops and natural ecosystems mainly depend on soil moisture produced by snow melt in spring and seasonal rainfall (Baldocchi et al. 2004;Zaitchik et al. 2007), there were large inter-annual fluctuations in the vegetation coverage (Fig. 4). The drought episodes that struck the Middle Eastern countries in 2007-2009 were the worst drought events with large impacts on the region in more than 60 years (Villa et al. 2014). What's more, the impacts of the lower precipitation and the increasing temperature (Table 2) provide the sufficient conditions of extreme drought in this area. Nonetheless,the greening trends were due to farming in the North Iraq and Turkeywhere groundwater and desalinated seawater were extensively used for agricultural activities (Rousta et al. 2020). In the low latitudes, except for temperature and precipitation, solar radiation is also a limiting factor (Nemani et al. 2003). In Southeast Asia, the NDVI product may include some inaccuracies. This research applied the Maximum Value Composite (MVC) method to reduce the outliers of data, but the selected maximum values were not always the nadir value that tended to overestimate NDVI values (Huete   et al. 2002). Moreover, the NDVI value saturates easily under high vegetation coverage and cannot sensitively monitor the changes in vegetation growth (Wang et al. 2003). These issues associated with NDVI also influence the accuracy of the correlations. This study also evaluated the temporal evolution of low vegetation areas when NDVI < 0.1 (Fig. 9), and the results show that Middle East, Arctic tundra and the Sahara had greening trends, while central Asia experienced a browning trend. These results, at least in part, are in agreement with previous studies (Berner et al. 2020;Piao et al. 2020;Rousta et al. 2020).

Decreasing trends in SINDVI
Located in the hinterlands of the Eurasian continent, Central Asia is one of the driest areas in the world (Cihlar et al. 1997) with low vegetation coverage. Most area is desert in this region besides Tajikistan and Kyrgyzstan which are predominated by mountainous. The variation in SINDVI indicates that vegetation degradation mainly occurred in northern Kazakhstan, southern Tajikistan and southwestern Kyrgyzstan. The spatial distribution of temperature from 2001 to 2020 showed a significance warming trend (Fig. 4), although the slightly increasing trend in precipitation, sufficient sunlight and solar energy resources in these domains are abundant than other regions, resulting in annual precipitation less than annual evapotranspiration; thus, excessive evapotranspiration is the hinge process of soil moisture loss and it leads to the shallow roots of desert plants to withered Yuan et al. 2017). In the drought periods, moisture insufficiency inhibited the growth of vegetation and reduced photosynthesis of plants (Stocker et al. 2019;Zhang et al. 2012).
This study found the sharply decreasing trend of the forest cover in Russia, Africa, North America, South America and Middle East (Fig. 8), mainly caused by fire, diseases or even over-logging (for example, large areas of tree mortality in North America owning to plant disease and insect pests ; Middle East and South America experienced a widespread deforestation and fire (Phalke et al. 2020;Qin et al. 2017)). This study also discovered that the increment of grassland and cropland areas in these regions made contribution to the positive trend of SINDVI, which is consistent with the previous findings that cropland land increase in some regions and the increase in vegetation greenness occurred in regions of intensive agricultural activities (Hansen et al. 2014;Lamchin et al. 2020).
In the Northern Hemisphere, vegetation growth was mainly affected by temperature, while precipitation was the dominant factor to the change of vegetation greenness in the Southern Hemisphere. The karst areas of South America are mainly in the east of  Soil (2022) 475:193-212 206 Brazil. According to Table 2, SINDVI here showed a downward trend, and the vegetation coverage in this region did not couple with temperature and precipitation. This lack of correlation is attributed to heavy human intervention. Under pressures from human activities, land use change has been increasingly threatening tropical rainforest area (Dirzo and Raven 2003). The quick transformations of forests into agricultural, timber production and other uses have produced a vast and human dominated landscape, which will bring immeasurable destruction to the rainforest (Gardner et al. 2009;Hansen et al. 2014). However, some studies have shown that the occurrence of extreme weather in recent years has aggravated the natural disasters, such as the severe droughts occurred in Brazil in 2007 (Cunha et al. 2019).
This study aimed to explore the impacts of climate change and human activities on vegetation dynamics from 2001 to 2020 in global karst areas. Our results indicate that the karst ecosystem is particularly vulnerable to the alternation of seasonal drought and flood. Especially in the 21 th century, under intensive anthropogenic activities and climate changes (extreme weather events), fragile karst terrains have been facing the threat of serious ecosystem degradation. This study assessesed the important issues including anthropogenic disturbance and climate change in karst area on global or regional scales. In particular, we have elaborated the impacts of climate change in Southwest China. Human activities were the dominate factor of vegetation greening, while in 2009/2010 there was a downward trend in SINDVI due to a period of extreme drought (Li et al. 2019;Zhao et al. 2017). Many previous studies have also confirmed that abnormal cyclone in the western North Pacific during an El Niño period resulted in a severe reduction in precipitation in Southwest China in the autumn of 2009 (Zhang et al. 2013). Climate change has caused large variations in seasonal and annual precipitation and temperature, both locally and globally (Fernández et al. 2014). Especially in the recent decades, drought occurred with longer durations, greater intensities and more frequently, with the serious impact on karst ecosystem (Trenberth et al. 2014;Williams et al. 2015). Deep rooting plant access to bedrock water storage and groundwater is an importance feature for plant survival in drought environments (Ding et al. 2021. For example, the effects of droughts have been widely reported in the karst areas in the United States (Ganguli and Ganguly 2016), Europe (Spinoni et al. 2017) and central equatorial Africa (Hua et al. 2016). A better knowledge of the dynamics of vegetation greenness and it responses to extremely climate is an essential process in understanding current situations and future changes. The current study is mainly about the global karst vegetation change, with focuses on several fragile karst regions. The results provide scientific evidence for the protection of fragile karst ecosystem, while detailed studies in each karst area are still needed.

Conclusions
In summary, this study assessed spatiotemporal variations in vegetation cover by estimating global SINDVI trends for karst region. On a global karst scale, inter-annual SINDVI of the karst regions from 2001 to 2020 exhibited an upward trend at 0.011/a. Temperature and precipitation had prominent effect on inter-annual changes in SINDVI with large spatiotemporal heterogeneities. In dry areas such as Central Asia, persistent warming coupled with decreasing precipitation caused serious water deficit and great evapotranspiration loss, thus affected the growth of vegetation. However, in some middle and high latitude areas, warming temperature was considered to be the main reason for greening trends, such as in northern Eurasia and northern North America. In addition to studying climate factors (temperature and precipitation), this study found anthropogenic activities (e.g., Ecological Engineering, the Grain to Green Project) also had significant impact on the variations in SINDVI, particularly in Southwest China.