Spatio-Temporal Characteristics and Trend Analysis of Grassland Ecosystem Drought in Asia From 2010 to 2018

Under the background of global climate change, the impact from drought on the ecosystem exhibits the characteristics of complexity and multi-process, especially for the main component, which is the grassland ecosystem of the overall ecosystem. Identifying past droughts and predicting future ones is vital in limiting their effects. However, the random and non-linear nature of drought variables makes accurate drought prediction still a challenging scientific problem. In this study, the boundaries, Land Surface Temperature (LST) and Enhanced Vegetation Index (EVI) of Asian Grassland Ecosystem (AGE) were obtained by Google Earth Engine (GEE), which were used to construct LST-EVI feature spaces to calculate the dry-wet edge fitting equations and Temperature Vegetation Drought Index (TVDI). Mann–Kendall test and Sen trend degrees were further used to analyze the drought trend of AGE. The results showed that there were obvious spatial differences in the wet and dry conditions of AGE, which showed that the TVDI increased from east to west and from north to south, with humid areas mainly concentrated in northern Asia and severe drought areas concentrated in southern Asia. From 2010 to 2018, the area of humid areas and severe drought areas of AGE decreased, and some humid areas changed to normal areas or even drought areas, while the drought in severe drought areas was alleviated. The results of the Sen trend test further show that the aggravating trend of drought in severe drought areas of South Asia is relatively low, and some areas show a trend of changing to humidity. However, there is an obvious aggravating trend of drought in humid areas or low drought areas of South Asia, these areas should also be the focus areas for drought prevention in the future. This study identified the spatio-temporal distribution characteristics and evaluated the evolution trend of the drought of AGE, which is of great significance to the management and prevention of drought of AGE.


INTRODUCTION
Under the influence of global warming, global extreme weather events are frequent. Drought, as one of the extreme weather events, is common in various historical periods of the world. It is characterized by high frequency, long cycles, and a wide range of impacts. With the passage of time, meteorological drought has gradually evolved into agricultural and hydrological drought, which has had profound effects on agricultural production (Zhang et al., 2015;Gan et al., 2016), water resources (Piao et al., 2010;Craft et al., 2015), ecosystem function (Zhong et al., 2019;Bastos et al., 2020) and environmental sustainability (Sun and Li, 2013;Sharafi et al., 2020). During periods of drought, reduced water flows into wetlands and groundwater aquifers, coupled with increased human water demand. This further reduces the availability of water to ecosystems, thus altering hydrogeological processes and affecting water bodies and terrestrial ecosystems in various ways. The end results are ecological droughts, with multiple knock-on effects such as ecological deterioration, ecosystem vulnerability and loss of ecological services Wang et al., 2021).
A large number of scholars have devoted themselves to the research of ecological drought monitoring, the driving mechanism of meteorological factors to ecological aridity, the driving mechanism of groundwater to ecological drought, and the evaluation of the ecological arid vulnerability. For example, Kim et al. (2019) conducted a quantitative assessment of water quality risks due to ecological drought by applying non-parametric kernel density estimates and hypothesizing the probability of rivers exceeding water-quality targets after extreme droughts. Studies have also been carried out to establish ecological drought indicators based on ecological flows. The study of Chu et al. (2019) showed that the Normalized Differential Vegetation Index (NDVI) in the growing season in the northeast of Heilongjiang was mainly regulated by precipitation, and the temperature was the dominant factor affecting the growth of spring vegetation, while there was a negative correlation between NDVI and precipitation in autumn. The work of Zhang G. et al. (2020) established the non-linear relationship between NDVI and groundwater depth based on information entropy. In the study of Crausbay et al. (2017), they established an ecological drought vulnerability research framework to reflect the ability of humans and natural systems to mitigate or adapt to drought, as well as the potential impact of drought on ecosystems. During this period, a series of drought indices, such as the Standardized Drought Vulnerability Index (SDVI) (Oikonomou et al., 2019), the Standard Precipitation Indices (SPI) (Mckee et al., 1993), the Palmer Drought Severity Indicator (PDSI) (Palmer, 1965), and the Standardized Precipitation-Evapotranspiration Index (SPEI) (Vicente-Serrano et al., 2010) were also established to represent the characteristics of drought in different regions and types.
With the development of remote sensing technology, many research methods have been widely used, such as using sensors to obtain spectral information of large-area surface vegetation, establishing different models of remote-sensing inversion and drought monitoring indexes, and then judging the degree and distribution range of the surface drought. For example, the Temperature Vegetation Dryness Index (TVDI) based on remote sensing data of vegetation and surface temperature calculation can indirectly reflect the effects of drought on vegetation and water consumption of vegetation. It is often used to describe the drought condition of terrestrial ecosystems. Several studies have also proved the reliability of this model (Sandholt et al., 2002;Du et al., 2017;Li C. et al., 2020;Kang et al., 2021).
Asia has a vast territory and great differences in topography. Under the combined effects of solar radiation, atmospheric circulation, and other factors, it has formed a prominent climate feature with a complete climate belt and complex climatic types. At the same time, large areas of Asia are affected by droughts, such as Thailand in Southeast Asia and the "no river countries" in West Asia (the "no river countries" refers to a country with no rivers which mainly includes Saudi Arabia, Kuwait, Oman, the United Arab Emirates, Qatar, Yemen, Bahrain, etc.) (Zong et al., 2017;Dai and Wang, 2020). The Taklamakan Desert, the second largest desert in the world, is also located in north-western China, East Asia. Therefore, the drought index measurement and drought trend analysis are very important for the terrestrial ecosystem in Asia, especially the grassland ecosystem, which is the main part of the whole ecosystem. Based on this, the boundaries, Land Surface Temperature (LST) and Enhanced Vegetation Index (EVI) of Asian Grassland Ecosystem (AGE) were obtained by Google Earth Engine (GEE), and TVDI and Sen trend degrees were used to evaluate the spatio-temporal distribution characteristics and evolution trend of AGE. Furthermore, the impact of changes in climate and environmental factors on ecosystems under the background of global climate change was explored. This was for the following aims: (1) provide data for drought monitoring of AGE; (2) provide a scientific basis for decision-makers to formulate targeted prevention and control measures to limit the development of drought; (3) provide technical support for the region to deal with terrestrial ecological drought caused by climate change.

Study Area
As the main body of the Eurasian continent, Asia has the highest plateaus, mountains, and peaks in the world, with a great difference in topography (Figure 1). At the same time, Asia has the widest north-south latitude span and the longest East-West distance (77 • 43 N∼11 • 7 S, 169 • 40W∼26 • 3 E), including all kinds of climatic zones from the equator to the Arctic, and the complicated topography and changeable climate caused the uneven spatio-temporal distribution of temperature and rainfall in Asia (Ge et al., 2015). For example, West Asia, which is deeply inland, is controlled by the continental air mass, making water vapor difficult to reach, so it is relatively arid. Although East Asia and South Asia are offshore, they are seriously affected by unstable monsoons. During the weak summer monsoon years, there is lower rainfall and drought disasters are more likely to occur. In addition, the spatial distribution of surface water in Asia is asymmetrically radiating under the constraints of geomorphological structure, mainly including desert-centered inland water systems, outflow systems on the outskirts of inland watersheds, and a few small rivers that flow into the Black Sea and the Mediterranean Sea, respectively. The extremely uneven distribution of surface water systems even led to the existence of "no river countries" such as Saudi Arabia, Kuwait, Oman, and other countries in West Asia (Gao, 2010;Li, 2020).
The vast spatial distribution has created regional differences between the coastal and inland regions of Asia, while at the same time creating geographical and landscape diversity, extremes, and typical. It provides the basis for the rich grassland ecosystems in Asia. The area of grassland ecosystems in Asia is about 2.897 billion hectares, accounting for 65.84% of the total terrestrial ecosystems of Asia, distributed mainly in Russia, China, Kazakhstan, Mongolia, Turkey, India, Indonesia, Iran, Afghanistan, Uzbekistan, and other countries ( Table 1; Hu, 1999). With the intensification of global climate change, complex climate types in Asia are more affected, with droughts and intense evaporation in most areas, making the ecological environment in the region more fragile and the problem of grassland ecosystem drought more acute.

Data Acquisition
The administrative boundary data used in this study to describe the geographical position in Asia was derived from Global Map data archives, in vector data format, developed in cooperation with the National Geospatial-Intelligence Agency (NGIAs) and geospatial intelligence agency in other countries. The location and administrative boundaries information covering 48 countries or regions of Asia had been obtained in this study.
The data used to calculate TVDI in this study was derived from the MODelation-resolving Image Spectrometer (MODIS), which was processed by the GEE platform due to the large area and long time span of the calculated data (Gao, 2010;Didan, 2015;Wan et al., 2015;Friedl and Sulla-Menashe, 2019;Sun et al., 2021). After data acquisition, quality check, projection,  resampling, clipping, and other pre-processing are carried out and the data at different time scales are calculated as annual data. Asian Grassland Ecosystem was derived from global land cover data of a grid of 500 m 2 provided by the MCD12Q1 V6 product (Friedl and Sulla-Menashe, 2019), which divided global land cover into 17 types. The Woody Savannas (tree cover 30-60%), Savannah (tree cover 10-30%), and Grasslands (dominated by herb bags annuals) from the Land Cover_Type1 (LC_Type1) classification of MCD12Q1 V6 product were extracted to the boundaries of AGE. Finally, the grassland in GlobeLand30 2020 has also been further adapted to modify the boundaries of AGE in this study (Jiang et al., 2006;Liu et al., 2021).
Land Surface Temperature was land surface temperature data of a grid of 1,200 km 2 for an average of 8 days from MOD11A2 V6 (Wan et al., 2015). The value of each pixel in MOD11A2 V6 was a simple mean of all corresponding MODO11A1 LSTs collected during the 8 days and included data on day-time LDST, nighttime LST, 31 and 32 band channel emissivity, and quality control. Twice as long as the 8-day cycle was the precise ground tracking repeat cycle of Terra and Aqua platforms, which to some extent ensured the accuracy of the LST data.
Enhanced Vegetation Index data minimized canopy background changes and maintained sensitivity under dense vegetation conditions. The EVI used in this study selected the best available pixel values from MOD13A2 V6 product collected over a 16-day period, which was calculated based on the atmospheric correction of the two-way surface albedo (Didan, 2015). Water, clouds, heavy aerosols, and cloud shadows were shielded. The EVI also used a blue band to remove residual atmospheric pollution from smog and subpixel thin clouds, using the criteria of low cloud cover, low perspective, and maximum EVI values. The data are presented in Table 2.

Research Framework
The following steps had been taken to develop a framework for drought monitoring and spatio-temporal characteristics of drought trends for AGE (Figure 2): Step 1: The LST and EVI data were pre-processed for quality inspection, projection, resampling, clipping, and so on. The areas with vegetation indices < 0.1 or > 0.85 were removed and the overestimation of sparse cover and underestimation of high-density cover in linear trend fitting were ruled out.
Step 2: At intervals of 0.02, EVI each year was divided into 50 categories. The minimum and maximum surface temperatures corresponding to each of them were determined.
Then, LSTmin-EVI and LSTmax-EVI spatial fitting were carried out to solve the corresponding dry-wet edge fitting equations.
Step 3: A TVDI model based on LST and EVI was established according to the obtained dry-wet edge fitting equations to analyze the spatial distribution of TVDI of AGE from 2010 to 2018.
Step 4: The Mann-Kendall test and Sen trend degrees were used to calculate the Sen trend degrees of annual TVDI on a grid point, and then Mann-Kendall trend test was carried out to calculate the spatial drought trend of AGE.

Temperature Vegetation Drought Index
The inversion of soil drought using TVDI was based on the effect of water evaporation and vegetation transpiration which brought the surface temperature down. The value of TVDI depended on the spatial conditions of LST-EVI. The effect of surface evaporation and transpiration on TVDI resulted from the effect on LST-EVI space (Goward et al., 2002). In recent years, a series of studies had shown that vegetation cover, soil moisture, surface thermal characteristics, and other factors had an influence on the relationship between LST and EVI. However, the remote sensing image sensor selected in the study and resolution has little influence on it (Yan and Wang, 2009;Song et al., 2011). Among them, soil moisture was the main influencing factor of LST-EVI space, and EVI could reflect vegetation cover status. Therefore, TVDI obtained from LST-EVI space could largely reflect the status of soil humidity. The study of Price, Carlson, Sandholt, and others analyzed the acquired remote sensing data and concluded that the characteristic spatial relationship between LST and EVI was triangular (Price, 1990;Carlson et al., 1994;Sandholt et al., 2002). The work of Gillies et al. (1997) further used remote sensing data and the LST-EVI triangle to measure soil moisture content. The study of Goward et al. (2002) proposed a dry-wet edge fitting equation. Based on these studies, Sandholt et al. (2002) proposed a TVDI, which is expressed as follows: Where LST is the land surface temperature of any pixel. LST data of MODIS were used in this study. a 1 +b 1 × EVI is the maximum surface temperature corresponding to a value of EVI. It was the dry edge of characteristic space. a 1 and b 1 are the coefficients of the dry edge fitting equation. a 2 +b 2 × EVI is the minimum surface temperature corresponding to an EVI. This corresponded to the wet edge of characteristic spaces. a 2 and b 2 are the coefficients of the wet edge fitting equation. The value range of TVDI is 0∼1, and the higher the value is, the more severe the water shortage and drought in the area. Conversely, the smaller the value is, the wetter the area.

Mann-Kendall Test and Sen Trend Degrees
Mann-Kendall Tendency Test was the main method for judging trends in the field of meteorological hydrology. It judged the trend salience of the sequence according to the test statistic Z. At the significance level α = 0.05, when the test statistic |Z| > 1.96, it meant significant trend; when the |Z| < 1.96, it indicated no significant trend (Tao and Zhang, 2020).
The Sen trend degree is a method proposed by Sen to study the variation of long time series of variables (Sen, 1968). For the time series x 1 , x 2 , . . ., x n , trend degree β is defined as: where x j and x i are time series values at different times respectively, Median is the median function. When β > 0, it indicated the time series of the variables to be analyzed is increasing; when β < 0, it indicated the time series of the variables to be analyzed is decreasing.

Analysis of Temperature Vegetation Drought Index in Asian Grassland Ecosystem
Based on the data pre-processing, the EVI data of each year were reclassified with the step size of 0.02, the maximum and minimum values of the LST values with the same EVI were extracted. By establishing the two-dimensional feature spaces fitting diagram of LST-EVI, the corresponding equation of drywet edge fitting (Figures 3, 4) was obtained. As can be seen from Figures 3, 4, the two ends of the dry-wet t edge fitting curve were relatively discrete, which was due to the deviation of the actual condition of vegetation reflected by the vegetation index when the vegetation coverage was very low or even bare soil, or when the vegetation coverage was very high. Thus, the points with EVI values lower than 0.1 and higher than 0.85 did not participate in linear regression fitting, so that the dry and wet edge fitting could meet the TVDI extraction precision of most pixels. Based on the obtained dry-wet edge fitting equation, the TVDI of AGE was calculated by using the formula of the TVDI model. The results of TVDI calculated from low to high were reclassified into five grades, 0 < TVDI ≤ 0.3 was humidification, 0.3 < TVDI ≤ 0.5 was normal, 0.5 < TVDI ≤ 0.65 was light drought, 0.65 < TVDI ≤ 0.77 was medium drought, and 0.77 < TVDI ≤ 1 was severe drought (Sun et al., 2021). The result is shown in Figure 5.
The correlation analysis was obtained from soil moisture data and TVDI data at the corresponding locations of AGE. There was a significant correlation between TVDI and measured water content of soil surface layer. The linear fitting results of TVDI and soil moisture both passed the t-test of confidence coefficient α = 0.05, and the correlation was good. This indicated that the temperature vegetation drought index could reflect the changing trend of soil moisture condition, and it was reasonable as an index of drought evaluation.  According to the drought of AGE, as reflected in TVDI (Figure 5), there were marked spatial differences in dry and wet conditions of AGE, specifically the degree of soil aridity increases from east to west and from north to south in general. Under this general trend, combined with the local topography and climate, the degree of drought varied greatly from country to country. The humid and normal areas were mainly concentrated in Russia, Mongolia, Kyrgyzstan, Tajikistan, Afghanistan, Georgia, north of Kazakhstan, and northeast and southwest China. The light drought areas were mainly concentrated in Iran, Afghanistan, Turkey, the center of Kazakhstan, the center of China, and other places in western Asia. Most of the southern and south-western parts of Asia were characterized by medium and severe drought conditions, particularly in Iraq, Pakistan, India, Indonesia, and other places in South Asia, which had reached severe drought.
After further screening and attribute surface area extraction, a histogram of the different levels of drought for AGE from 2010 to 2018 was obtained (Figure 6). Figure 6 shows a decreasing  In 2015, the area of severe drought was significantly less. In 2017, the area of humid areas, normal areas, and light drought areas all decreased significantly. The area of medium and severe drought areas increased greatly. In 2010, humid areas of AGE covered an area of 8.1776 million km 2 , and severe drought areas had covered an area of 4.4141 million km 2 . By 2018, humid areas of grassland ecosystems in Asia had been reduced to 8.0785 million km 2 and severe drought areas to 0.0418 million km 2 . In 2018, the area of humid areas decreased by 1.21% compared with 2010, and the severe drought areas were reduced by 5.27%. Overall, there was an improvement in severe drought for AGE from 2010 to 2018. At the same time, drought was occurring in some humid areas. Regional TVDI changed from humid to normal and even to drought.

Analysis of Mann-Kendall Test and Sen Trend Degrees in Asian Grassland Ecosystem
The Sen trend of the inter-annual scale TVDI was calculated grid point by grid point, and the Mann-Kendall trend test was performed on it (Figure 7). Figure 7 shows that there was an overall increasing trend of year-scale meteorological drought in AGE, mainly concentrated in the regions with humid and normal areas, such as Russia, Mongolia, north of Kazakhstan, and northeast China. Conversely, in the severe drought areas of the south of South Asia, the drought intensity in 2010-2018 was still very severe. It had a relatively low trend toward increased aridity, even moderating in places such as Laos, Thailand, Cambodia, and south of China. The Sen trend of TVDI shifting in direction (toward humid).
In summary, this study used TVDI and Sen trend degrees to evaluate the spatio-temporal distribution characteristics and evolution trend of AGE. It was found that there were obvious spatial differentiation characteristics of drought degree of AGE under the influence of temperature, environment, and other factors. The drought degree has increased or decreased in varying degrees from 2010 to 2018, mainly in South Asia and East Asia with wetter dry conditions had become drier and drier, changing to drier conditions, while severe dry conditions in certain areas had changed to wetter conditions. Other relevant studies have also confirmed the accuracy of this result (Ummenhofer et al., 2013;Manolache et al., 2020;Chandrasekara et al., 2021;Hina et al., 2021). In addition, the study of Manolache et al. (2020) explored the obstacles of different grassland management measures and the uniqueness to be considered in the selection, which was of great significance to provide different grassland management schemes for grassland ecosystems with different drought degrees. Combined with the results of this study, they could help decision-makers better FIGURE 7 | Sen trend degrees of TVDI for AGE.
Frontiers in Ecology and Evolution | www.frontiersin.org formulate targeted prevention and control measures to deal with terrestrial ecological drought caused by climate change.

Limitations
There are limitations in this study starting from the MODIS data which provided data support for the assessment of drought conditions in AGE, with long time series and high temporal resolution and with high spatial reality. However, uncertainties in global data observations and the conversion of the data used in the study to annual data at higher temporal resolutions had, to some extent, eliminated differences in data extremes, especially the differences in regions with different seasons and different degrees of drought (Sun et al., 2021). For example, the seasonal features of LST that were higher in summer and autumn but lower in winter and spring were ignored in the correlation analysis. Therefore, the assessment results of trends in drought conditions by the annual data used in this study require further calibration. In addition, several studies showed that drought is caused by a combination of factors such as precipitation, water resources in rivers and lakes, soil moisture content, and soil potential evapotranspiration (Zhang X. Y. et al., 2020;Hina et al., 2021;Su et al., 2021). The TVDI calculated from LST and EVI data in this study could represent the drought of AGE to some extent. However, further studies on drought in AGE affected by multiple factors are still needed.

CONCLUSION
In this study, the spatio-temporal distribution characteristics and evolution trend of drought in the AGE were analyzed by using TVDI and Sen trend degrees. The results showed that the humid areas of the AGE are mainly concentrated in the north of Asia, and the severe drought areas were mainly concentrated in the south of Asia. From 2010 to 2018, the transition from humid areas to arid areas was obvious, and the results of the Sen trend also showed that the drought degree had an aggravating trend. These findings provided a foundation to explore the impact of climate and environmental factors on ecosystems. Further study should concentrate on the analysis of the seasonal characteristics of drought evolution. Furthermore, the method used in this study could also be used to evaluate the evolution trend of ecosystem drought in other regions, so that the decision-makers could make targeted prevention and control measures to limit the development of drought, which is worth promoting.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author/s.