China’s deserts greening and response to climate variability and human activities

RESEARCH ARTICLE China’s deserts greening and response to climate variability and human activities Xiaoyu LiuID1,2, Liangjie XinID1* 1 Key Laboratory of Land Surface Pattern and Simulation, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Science, Beijing, China, 2 College of Resources and Environment, University of Chinese Academy of Sciences, Beijing, China * xinlj@igsnrr.ac.cn Abstract Vegetation, which is a good indicator of the impacts of climate variability and human activi- ties, can reflect desert ecosystem dynamics. To reveal the vegetation variations in China’s deserts, trends in the monthly, seasonal, and annual normalized difference vegetation index (NDVI) from 2000 to 2017 were measured both temporally and spatially by the Theil-Sen estimator and Mann-Kendall test. Additionally, correlation coefficients and residual analysis were employed to evaluate the correlations between the NDVI and climatic factors and to distinguish the impacts of climate variability and human activities. The results showed that China’s deserts underwent greening. The annual NDVI showed a significant increasing trend at a rate of 0.0018/yr, with values of 0.094 in 2000 and 0.126 in 2017. Significant increasing trends in NDVI were observed in all four seasons. The NDVI were higher in sum- mer and autumn than in spring and winter. Both the monthly NDVI and its trends showed an inverted U-shaped curve during the year. Spatially, the greening trends were mainly distrib- uted on the southern edge of the Gurbantunggut Desert, in the northwestern part of the

Taklimakan Desert, and in the Kubuqi Desert. The correlations between the NDVI and cli- matic factors at the monthly and seasonal scales were stronger than those at the annual scale. Temperature and precipitation had positive effects on NDVI at the monthly and sea- sonal scales, but only precipitation had a positive effect at the annual scale. Human activi- ties, especially oasis expansion and sand stabilization measures, were two major causes of large increasing areas of desert greening in China indicated by the NDVI. Introduction Deserts are one of the major land-cover types, covering approximately one-third of the terres- trial area of the earth [1]. Due to topography, atmospheric circulation, latitude, ocean currents, etc., deserts, as a product of arid climatic conditions, are mostly located on the western coasts of continents near 30˚ north and south latitude [2,3]. There is still no universally accepted common or technical definition for deserts, only distinguished by characteristics (e.g., climate, weather, and hydrology) [4]. In addition, deserts are dynamic, and their boundaries are grad- ual and not clearly defined. In China, deserts are mainly distributed in the northwest, which is PLOS ONE PLOS ONE | https://doi.org/10.1371/journal.pone.0256462 August 30, 2021 1 / 20 a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Liu X, Xin L (2021) China’s deserts greening and response to climate variability and human activities. PLoS ONE 16(8): e0256462. https://doi.org/10.1371/journal.pone.0256462

Editor: Gerald Forkuor, United Nations University Institute for Natural Resources in Africa, GHANA Received: June 2, 2021 Accepted: August 6, 2021 Published: August 30, 2021 Copyright: © 2021 Liu, Xin. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The temperature and precipitation data used in our manuscript were obtained through a request from the National Tibetan Plateau Data Center, which is not directly produced by us, but directly uses someone else’s data product. Due to data copyright issues, we are unable to upload the data directly. The data is currently open access and readers can access the data by requesting it directly on the platform (https://data.tpdc.ac.cn/en/). Temperature and precipitation data can be obtained at the following websites: temperature: https://data.tpdc.ac.cn/en/ data/71ab4677-b66c-4fd1-a004-b2a541c4d5bf/? q=1-km%20monthly%20mean%20temperature%

affected by both East Asian monsoons and global westerlies, and the climate variability that occurs within a year is critically influenced by the interplay of monsoonal circulation between summer and winter [5]. As a potential result of global warming [6], desert ecosystems as one of the fragile ecosys- tems are changing dynamically [3]. To clarify the changes of desert ecosystems can help to understand their ecological balance, and to take effective measures in time. It is important to strengthen ecological construction and take effective countermeasures in time. Due to the sen- sitivity to climate variability and good performance in arid areas [7–9], vegetation can be an ideal indicator to detect changes in desert ecosystems dynamics [10]. The normalized differ- ence vegetation index (NDVI), which provides information on vegetation conditions, can identify low biomass and an increase in vegetation cover than other indexes [11], so it has become the most widely used index in studies on vegetation dynamics [12,13]. In recent decades, several studies have focused on NDVI variations in arid and semiarid regions [14– 16]. Their results showed that the NDVI showed greening trends at the annual scale in several regions, such as the Gurbantunggut Desert, which showed an increasing trend that fluctuated from 1981–2003 [17]. At the same time, the NDVI for Central Asia significantly increased, with a value of 11.35% [18]. In global semiarid areas, the average NDVI rise from 1981 to 2007 by about 0.015 NDVI units [19]. Seasonal differences were seen in the global NDVI, for exam-

ple, the upward trend of NDVI in spring, summer, and autumn was observed in some mid- and high-latitude regions of the Northern Hemisphere, and the downward trend of NDVI in summer and autumn was observed in some arid and semi-arid regions of the Southern Hemi- sphere [20]. Variations in the NDVI that occur within a year can be observed at the monthly scale. For instance, the monthly mean NDVI increased between January and August and decreased in December in Nigeria [15]. However, the monthly mean NDVI in Central Asia is highest in January and December [21]. Variations in the NDVI are influenced by multiple factors, which can be divided into envi- ronmental and anthropogenic factors [22–24]. Among various environmental factors, varia- tions in temperature and precipitation, which are forms of climate variability, are considered to be the two most influential factors [25–27]. In most cases, the effect of temperature is posi- tive, but sometimes it has a negative effect. For example, the NDVI is positively associated with the temperature when energy is limited to vegetation growth in higher latitudes and higher altitudes in the North American continent, especially at the beginning of the season, but usu- ally in the mid-season the relationship is negative for lower latitudes when water is the limit of vegetation growth [28]. Temperature has a positive correlation with NDVI at both ends of the growing seasons but a weak negative correlation in the middle of the growing season in the

central Great Plains [29]. Precipitation positively influences the NDVI, especially in arid and semiarid regions that lack water [30–32]. The impact of climate variability also has seasonal differences, as temperature is the dominant factor in spring that lowers the NDVI in inner Asia, which is an environment appropriate for vegetation growth and causes an increase in the melting of winter snow and ice, thereby increasing moisture in the spring [33,34]. Sufficient rainfall during summer benefits the growth of vegetation in arid areas more [35]. The effect of climate has a time lag that is distributed differently in different regions. For example, there was a 3-month lag between the NDVI and temperature at the biome scale in China [36], and there was a 2-year lag in temperature at the global scale [37]. Human activities such as cropland and forest expansion in China and India have increased their relative areas by approximately 25% and 6.8%, respectively, and have played a critical role in global greening [22]. However, a dra- matic decline in NDVI was observed in the Yangtze River and Pearl River deltas due to rapid urbanization [36]. PLOS ONE China’s deserts greening and response to climate variability and human activities PLOS ONE | https://doi.org/10.1371/journal.pone.0256462 August 30, 2021 2 / 20 20dataset%20for%20china%20(1901-2017) precipitation: https://data.tpdc.ac.cn/en/data/ faae7605-a0f2-4d18-b28f-5cee413766a2/?q=1-km %20monthly%20precipitation%20dataset%20for %20China%20(1901-2017). Future researchers first need to click "Apply for Download" on the right

side of the above two web pages, register their accounts and submit instructions on how to use the data, etc., and then obtain the permission to download the data after approval. Funding: This work was supported by Strategic Priority Research Program of Chinese Academy of Sciences “Pan-third pole environmental change and the construction of green Silk Road” [grant numbers XDA20040000, XDA20090000]. Competing interests: The authors have declared that no competing interests exist.

The NDVI is an effective indicator of conditions affecting vegetation growth and can reflect the impacts of environmental and anthropogenic factors. Previous studies have paid more attention to NDVI dynamics and influential factors in different regions (e.g., Hungary [38], Central Asia [21], and Canada [39]), as well as different land cover types (e.g., grassland [40], forests [41], and cropland [42]). By contrast, few studies have considered NDVI variations and driving factors at both the interannual and intra-annual scales in regions with low vegetation cover, especially arid zones and deserts, which are more sensitive to outside factors. Combined with existing studies, it has been found that vegetation in most arid zones around the world has shown a gradual recovery trend recently [8,43]. As a country with vast desert areas, China has been under the attack of wind and sand for years, the government attached great importance to it and made great practices in desert control, especially since the 21st century, ecological protection has been highly emphasized in China. However, the detailed studies on the vegetation dynamics of all deserts in China and the extent to which they are affected by climatic factors and human activities are still lacking. This study aims to analyze NDVI variations and the influential factors in China’s deserts from 2000 to 2017. Specifically, this study seeks to (1) use monthly, seasonal, and annual data to explore trends in the NDVI and climatic factors temporally and spatially; (2) analyze the

correlations between the NDVI and temperature and precipitation at the monthly, seasonal, and annual scales to better understand the relationships between the NDVI and climatic fac- tors and to compare their influence at different time scales; and (3) use residual analysis to dis- tinguish greening regions caused by human activities from 2000 to 2017. Materials and methods Study area Deserts in China are widely distributed in the north, especially northwestern China, which is located between 75˚-115˚E and 35˚-47˚N, with an area of approximately 0.98 million km2. Most part of the deserts is distributed in arid areas, and the remaining part is distributed in semi-arid areas. The annual average temperature of China’s deserts ranges from -10˚C to 15˚C, and the annual accumulated precipitation in most of the areas is less than 200 mm and mostly distributed in summer. Eight deserts are located in the study area: the Gurbantunggut Desert, Taklimakan Desert, Kumtag Desert, Qaidam Basin Desert, Badain Jaran Desert, Teng- ger Desert, Ulan Buh Desert, and Kubuqi Desert (Fig 1(a)). Data sources Distribution of deserts. The desert distribution data used in this study were mainly obtained from the 1:100000 desert (sand) distribution dataset in China [44]; the data were interpreted, extracted, and edited from Thematic Mapper (TM) digital images and the China land-use map for 2000. The range of deserts was corrected by using a topographic map of China and the map of desert distribution at 1:2000000 in China [45].

NDVI data. In this study, the NDVI from MOD13A2 was used as an indicator of vegeta- tion activity. The NDVI dataset was sourced from the MODIS (Moderate Resolution Imaging Spectroradiometer) vegetation index product data on the NASA website (https://lpdaac.usgs. gov/products/mod13a2v006/) [46] and had already been corrected for the effect of atmo- spheric gases, thin cirrus clouds, and aerosols, with a spatial resolution of 1 km and a temporal resolution of 16 days. A total of 411 images were used for the period between December 2017 and February 2000. Data processing, such as image merging, map projection transformation, resampling, and masking by desert areas, was performed on the original downloaded NDVI data. The maximum value composite (MVC) method was applied to calculate the monthly and PLOS ONE China’s deserts greening and response to climate variability and human activities PLOS ONE | https://doi.org/10.1371/journal.pone.0256462 August 30, 2021 3 / 20

annual NDVI data for each pixel to reduce the atmospheric effects of clouds and aerosols [47– 49]. The mean values for spring (March-May), summer (June-August), autumn (September- November), and winter (January, February, and December) were calculated separately to truly reflect seasonal differences in the NDVI. Otherwise, the maximum values of the NDVI for spring, summer, and autumn would be very similar, but a large gap would appear for winter. Climate data. Data on the monthly mean temperature and monthly accumulated precipi- tation from 2000 to 2017 at a 1 km spatial resolution were obtained from the National Tibetan Plateau Data Center (http://data.tpdc.ac.cn) [50–52]. The original climate data were merged, projection transformed, masked, and registered to be consistent with the NDVI data. The sea- sonal and annual mean temperature and accumulated precipitation were calculated from monthly values to aid the analysis. Multi-year averages of temperature (Fig 1(b)) and precipita- tion (Fig 1(c)) from 2000 to 2017 were also calculated for understanding the climatic condi- tions of the study area. Fig 1. Spatial distribution of deserts in China(a), and mean annual temperature(b) and precipitation(c) of China’s deserts from 2000 to 2017. https://doi.org/10.1371/journal.pone.0256462.g001 PLOS ONE China’s deserts greening and response to climate variability and human activities PLOS ONE | https://doi.org/10.1371/journal.pone.0256462 August 30, 2021 4 / 20

Both the NDVI raster data and climate raster data were processed and analyzed by using ArcGIS10.5, MATLAB R2016a, and Python 3.7. Methods Theil-Sen estimator. The Theil-Sen estimator calculates the slope estimate as the median of the single-point slope and the intercept as the median of the single-point intercept. The Theil-Sen estimator can effectively eliminate the influence of outliers on the estimation results (Eq 1) [53,54]. In contrast to ordinary least squares used for linear regression, the Theil-Sen estimator is an unbiased estimate of a linear trend and is more robust; therefore, it has been widely used for estimating the slope of the NDVI [55–58]. slope ¼ Median NDVIj NDVIi j i � � ; 8j > i ð1Þ where NDVIi and NDVIj are the NDVI values at times i and j, respectively. The Theil-Sen esti- mator was used to calculate the trend for each pixel; in general, slope >0 indicates a positive trend in the NDVI value of the pixel during the period, and vice versa. Mann-Kendall test. The Mann-Kendall test, which is a nonparametric test, was applied to evaluate the significance of the trend of the Theil-Sen estimator [59,60]. And it has been widely used in analyses of NDVI trends [39,61–63]. The Mann-Kendall test was computed using Eqs (2–4). S ¼ X n 1 i¼1 X n j¼iþ1 sign NDVIj NDVIi � � ð2Þ sign NDVIj NDVIi � � ¼ 1 ðNDVIj NDVIi > 0Þ 0 ðNDVIj NDVIi ¼ 0Þ 1 ðNDVIj NDVIi < 0Þ 8 > < > : ð3Þ Z ¼ S 1 ffiffiffiffiffiffiffiffiffiffiffiffiffiffi Var S ð Þ p ðS > 0Þ 0 S ¼ 0 ð Þ S þ 1 ffiffiffiffiffiffiffiffiffiffiffiffiffiffi Var S ð Þ p ðS < 0Þ 8 > > > > > > <

> > > > > > : ð4Þ where n is the length of the time-series data (in this study, n = 18), S represents the Mann-Ken- dall statistic, and Z is a standardized statistic that follows a standard normal distribution. |Z| > 1.96 indicates that the time-series data change significantly at the 5% level. Combined with the Mann-Kendall statistic and the Theil-Sen estimator, a pixel shows a greening trend when slope > 0 and |Z| > 1.96, the pixel shows a browning trend when slope < 0 and |Z| > 1.96, and the pixel does not show a significant change when |Z| � 1.96. Correlation analysis of the NDVI and climatic factors. The Pearson correlation coeffi- cient (r) was commonly used in previous studies to evaluate the relationships between vegeta- tion dynamics and climatic factors [43,64,65]. In this study, this method was applied to calculate correlations between monthly, seasonal, and annual NDVI and climatic factors (i.e., temperature and precipitation). P-values were used to indicate the significance of the correlations. PLOS ONE China’s deserts greening and response to climate variability and human activities PLOS ONE | https://doi.org/10.1371/journal.pone.0256462 August 30, 2021 5 / 20

Residual analysis. NDVI variations are jointly affected by climate variability and human activities [64,66,67]. However, the impact of human activities on the NDVI of deserts is diffi- cult to assess by using only one indicator or several indicators. The residual analysis method can separate the influence of climate factors and human activities with the help of multiple lin- ear regression (Eqs 5 and 6), so it is widely used to identify the impact of human activities [23,68]. NDVIpre;n ¼ a�TMPn þ b�PREn þ c ð5Þ NDVIhuman ¼ NDVIobs NDVIpre;n ð6Þ where NDVIpre,n is the value of the NDVI predicted by temperature and precipitation in year n. TMPn and PREn are annual mean temperature and annual accumulated precipitation in year n. a and b are the coefficients of temperature and precipitation, respectively, calculated by multiple linear regression to obtain. c represents the intercept of the multiple linear regression. NDVIhuman represents the variations in NDVI affected by human activities. NDVIobs represents observed NDVI. Results Temporal variation of NDVI and climatic factors The mean values of the monthly, seasonal, and annual NDVI, temperature, and precipita- tion of the whole study area were calculated for use in the analysis of the temporal varia- tions, and the trends were identified by using the Theil-Sen estimator and Mann-Kendall test. According to the results (Fig 2a-2c), the annual NDVI showed a significant increasing trend at a rate of 0.0018/yr (|Z| > 1.96), up from 0.094 in 2000 to 0.126 in 2017. However,

temperature and precipitation showed a slight decreasing trend (-0.0010/yr) and an increas- ing trend (0.7976/yr), respectively, and both fluctuated greatly and showed insignificant trends. The seasonal differences in the NDVI were obvious. The mean values of the NDVI from 2000 to 2017 for each season were as follows: summer (0.1005)>autumn (0.0837)>spring (0.0729)>winter (0.0601). The seasonal temperature showed a different pattern and was higher in spring than in autumn. Because of the large fluctuations of seasonal precipitation in the time series, the variation curves for spring and autumn precipitation were staggered. With respect to the trends, the seasonal NDVI (Fig 2(d)) showed a significant increasing trend in all four seasons (|Z| > 1.96), with summer showing the highest rate of increase at 0.0017/yr, fol- lowed by autumn (0.0013/yr) and spring (0.0010/yr), and winter showing the lowest rate of increase (0.0008/yr). Seasonal temperature and precipitation did not show significant fluctua- tion trends. The temperature showed slight fluctuations, with increasing trends in summer (0.0054˚C/yr), and decreasing trends in winter (-0.0233˚C/yr). There were greater fluctuations in seasonal precipitation, especially in summer, with increasing trends in summer (0.4295 mm/yr) and a decreasing trend in winter (-0.0627 mm/yr). The data showed monthly NDVI variations (Fig 3). The mean values of the monthly NDVI from 2000 to 2017 showed an inverted U-shaped curve, which was the lowest in January (0.0539) and the highest in August

(0.0989). The variations in monthly temperature and precipitation were the same as the NDVI variations with one exception: the maximum values of both temperature and precipitation occurred in July. The trend of the NDVI over 12 months was positive, implying that the NDVI increased throughout the year during the study period. PLOS ONE China’s deserts greening and response to climate variability and human activities PLOS ONE | https://doi.org/10.1371/journal.pone.0256462 August 30, 2021 6 / 20

Spatial differences in greening and browning trends Greening and browning trends indicated by the annual NDVI. On the annual scale, the deserts showed greening trends spatially (Fig 4), with 62.8% of the area showing greening trends, mainly on the desert margins, especially on the southern edge of the Gurbantunggut Desert, in the northwestern part of the Taklimakan Desert, and in the Kubuqi Desert. Only 0.78% of the area showed browning trends, distributed in the northeastern Taklimakan Desert and the eastern Qaidam Basin Desert. Fig 3. Mean values (black lines) and trends (red lines) of the monthly NDVI (a), temperature (b), and precipitation (c) in desert areas from 2000 to 2017. https://doi.org/10.1371/journal.pone.0256462.g003 Fig 2. Variations and trends in annual precipitation (a), temperature (b), and NDVI (c), and seasonal NDVI (d), temperature (e), and precipitation (f) in desert areas from 2000 to 2017. https://doi.org/10.1371/journal.pone.0256462.g002 PLOS ONE China’s deserts greening and response to climate variability and human activities PLOS ONE | https://doi.org/10.1371/journal.pone.0256462 August 30, 2021 7 / 20

Fig 4 also shows the relative frequency distributions of the greening and browning trends over longitude and latitude. The area with greening trends was positively correlated with the desert area. Although the area with browning trends was very small, it was distributed over longitudes and latitudes with large desert areas. The NDVI trends differed for different deserts (Table 1). The proportion of greening in the Badain Jaran Desert was the highest among the deserts at 88.26%, which increased signifi- cantly, followed by the Kumtag Desert (75.26%) and Tengger Desert (70.07%). While the pro- portion of greening in the Gurbantunggut Desert was only 31.22%, the lowest of the deserts. Fig 4. Spatial distribution of greening and browning trends and statistical results based on the longitude and latitude of China’s desert areas from 2000 to 2017. Gray areas represent desert regions with no significant trends at the 5% confidence level. The green and brown areas indicate that the pixel has a significant greening and browning trend, respectively. https://doi.org/10.1371/journal.pone.0256462.g004 Table 1. The NDVI trends in different deserts. Desert Significant Increase (%) Increase but not significant (%) Significant decrease (%) Decrease but not significant (%) Gurbantunggut Desert 31.22 62.50 0.17 6.12 Kumtag Desert 75.26 23.30 0.07 1.38 Qaidam Basin Desert 63.60 30.46 1.25 4.68 Kubuqi Desert 60.31 33.67 0.67 5.35 Badain Jaran Desert 88.26 11.14 0.05 0.55 Ulan Buh Desert 40.41 50.40 0.30 8.89 Tengger Desert 70.07 27.67 0.11 2.15

Taklimakan Desert 65.82 29.69 0.84 3.65 https://doi.org/10.1371/journal.pone.0256462.t001 PLOS ONE China’s deserts greening and response to climate variability and human activities PLOS ONE | https://doi.org/10.1371/journal.pone.0256462 August 30, 2021 8 / 20

The Qaidam Basin Desert (1.25%) showed the highest proportion of browning, followed by the Taklimakan Desert (0.84%) and the Kubuqi Desert (0.67%). Greening and browning trends indicated by the seasonal NDVI. The greening and browning trends were calculated by the Theil-Sen estimator and Mann-Kendall test using the seasonal NDVI in desert areas from 2000 to 2017 (Fig 5) and the statistics are shown in Table 2. The spatial distribution of the greening and browning trends had obvious seasonal dif- ferences. In spring and winter, there was little spatial difference among the deserts, and the areas showed approximately the same degree of greening. The proportion of greening areas in spring was 73.05%, the highest among the seasons. However, the degree of the greening of NDVI in autumn was higher than that in spring, and the greening trends increased on the southern edge of the Gurbantunggut Desert, in the northwestern part of the Taklimakan Fig 5. Spatial distribution of greening and browning trends for spring (a), summer (b), autumn (c), and winter (d) in China’s desert areas from 2000 to 2017. Gray areas represent desert regions with no significant trends at the 5% confidence level. The green and brown areas indicate that the pixel has a significant greening and browning trend, respectively. https://doi.org/10.1371/journal.pone.0256462.g005 Table 2. Proportion of the area for seasonal NDVI trends. Season Significant Increase (%) Increase but not significant (%) Significant decrease (%) Decrease but not significant (%) Spring 73.05 22.30 0.76 3.89

Summer 69.37 25.95 0.96 3.72 Autumn 44.93 46.02 1.14 7.91 Winter 56.49 35.34 0.86 7.30 https://doi.org/10.1371/journal.pone.0256462.t002 PLOS ONE China’s deserts greening and response to climate variability and human activities PLOS ONE | https://doi.org/10.1371/journal.pone.0256462 August 30, 2021 9 / 20

Desert, and in the Kubuqi Desert. The spatial distribution of the NDVI trends in summer was the most similar to the spatial distribution of the annual NDVI trends; this may have occurred because the annual NDVI was calculated by the MVC method, while the maximum values of NDVI in most regions occurred in summer. Correlations between the NDVI and climatic factors The Pearson correlation coefficients between monthly, seasonal, and annual NDVI and cli- matic factors were calculated to measure the effect of climate variability on NDVI in deserts and to compare the impact of different time scales on the correlations, as presented in Table 3. Comparing the differences in the effects of climatic factors on NDVI at different scales, it was found that the correlations between climatic factors and NDVI were significant at seasonal and monthly scales, while only precipitation was significantly correlated with NDVI at annual scales. This suggested that climate variability was the main cause of monthly and seasonal vari- ation in NDVI, while only precipitation had some influence on NDVI variation at the interan- nual scale. Lagged effect of climatic factors on NDVI. Considering the possible lag in the effect of temperature and precipitation on the NDVI, the Pearson correlation coefficients for the corre- lations between the current NDVI of each month and the climatic factors of the current month and previous months were calculated (Table 4). The results showed that temperature and precipitation had no lagged effect on the monthly

NDVI because the largest correlation coefficients for the correlations between the NDVI and temperature and precipitation in the same month were 0.796 and 0.869, respectively, and the correlation decreased as the number of lagged months increased. When the lagged months increased to 4 months, NDVI started to be negatively correlated with temperature and precipitation. Table 3. The Pearson correlation coefficients of the climatic factors and NDVI at different time scales. Scale Temperature Precipitation Annual -0.087 0.544� Seasonal 0.865�� 0.866�� Monthly 0.869�� 0.796�� � Significant at level 0.10; �� significant at level 0.05. https://doi.org/10.1371/journal.pone.0256462.t003 Table 4. The Pearson correlation coefficient for the relationship between the monthly NDVI and temperature and precipitation for different time lags. Time lag Temperature Precipitation 0 0.796�� 0.869�� 1 0.764�� 0.864�� 2 0.532�� 0.622�� 3 0.115 0.226�� 4 -0.286�� -0.227�� 0 indicates the correlation coefficients are calculated by the NDVI and climatic factors in the same month, 1 indicates the correlation coefficients are calculated by the NDVI in the current month and climatic factors in the previous month, etc. �� Significant at level 0.05. https://doi.org/10.1371/journal.pone.0256462.t004 PLOS ONE China’s deserts greening and response to climate variability and human activities PLOS ONE | https://doi.org/10.1371/journal.pone.0256462 August 30, 2021 10 / 20

Spatial differences in correlations. To further compare the differences in the impacts of climate variability on NDVI at different scales spatially, the Pearson correlation coefficients for the relationship between monthly and annual NDVI and temperature and precipitation from 2000 to 2017 were calculated for each pixel, and the proportion of the area of the correlation coefficient is shown in Table 5. Fig 6(a) illustrates the spatial distribution of the temperature effect on the NDVI at the monthly scale; 94.15% of the area in which the NDVI had a positive correlation with tempera- ture was distributed in areas other than the eastern part of the Taklimakan Desert and the central part of the Qaidam Basin Desert. The strong positive correlation was principally dis- tributed in the Gurbantunggut Desert, the western edge of the Taklimakan Desert, the western and eastern edges of the Qaidam Basin Desert, the Tengger Desert, and the Kubuqi Desert, indicating the area where the NDVI showed significant greening trends. Fig 6(b) illustrates the spatial distribution of the influence of precipitation on NDVI at the monthly scale, which was similar to that of temperature; however, the correlation in the Gur- bantunggut Desert was not as strong as that of temperature. Overall, 83.28% of the area showed a positive correlation, and 68.19% was significant (p<0.05), which was lower than that of temperature. In addition, 16.72% of the area had a negative correlation; this area was distrib- uted in the eastern part of the Taklimakan Desert and the central part of the Qaidam Basin Desert.

Annual Significant positive (%) 0.50 14.99 Positive but not significant (%) 38.35 69.97 Significant negative (%) 3.98 0.15 Negative but not significant (%) 57.17 14.89 “Significant positive” denotes pixels r>0 and p<0.05, “Positive but not significant” denotes pixels r>0 and p>0.05, “Significant negative” denotes pixels r<0 and p<0.05, and “Negative but not significant” denotes pixels r<0 and p>0.05. https://doi.org/10.1371/journal.pone.0256462.t005 PLOS ONE China’s deserts greening and response to climate variability and human activities PLOS ONE | https://doi.org/10.1371/journal.pone.0256462 August 30, 2021 11 / 20

Impact of human activities The impact of human activities was calculated by residual analysis. Areas where human activi- ties caused NDVI variations were greater than climate variability were classified as human activities impact areas. Fig 7 shows the region where NDVI variations were caused by human activities, where considerable greening trends were observed. This region included the south- ern part of the Gurbantunggut Desert, the edge of the Taklimakan Desert, and the Kubuqi Desert. To further identify which human activities caused the deserts greening, we compared Land- sat images and Google Earth images of the desert greening areas with those of the desert non- greening areas. The images for 2000 and 2017 indicate that land cover changes can be clearly observed. By comparison, there were two main types of human activities that caused deserts greening: oasis expansion and sand stabilization measures. Specifically, the areas showed greening trends in the southern part of the Gurbantunggut Desert, the northwestern edge of the Taklimakan Desert, the eastern Kumtag Desert, the Badain Jaran Desert, and the Tengger Desert were transformed into cropland due to the expansion of oasis. The greening areas in the Kubuqi Desert changed into shrubs and grass grids due to sand stabilization measures. Due to the space limitation, only a few image comparisons before and after the greening of the deserts caused by human activities are shown in Fig 7. More remote sensing image evidence

Fig 6. Spatial distribution of the Pearson correlation coefficient for the relationships between the NDVI and climatic factors. (a) Correlation between the monthly NDVI and temperature. (b) Correlation between the monthly NDVI and precipitation. (c) Correlation between the annual NDVI and temperature. (d) Correlation between the annual NDVI and precipitation. https://doi.org/10.1371/journal.pone.0256462.g006 PLOS ONE China’s deserts greening and response to climate variability and human activities PLOS ONE | https://doi.org/10.1371/journal.pone.0256462 August 30, 2021 12 / 20

about oasis expansion and sand stabilization measures as the two major factors of desert green- ing in China’s deserts are listed in the Supplementary Materials. Discussion Variations in the NDVI and its relationship with the climatic factors China’s deserts showed significant greening trends both temporally and spatially, reflecting the increased vegetation cover of China’s deserts during the study period. Desert greening has far-reaching impacts on climate variability and environmental improvement. Firstly, it increases the land space available for human use, and the increased vegetation cover in deserts increases the carbon sink capacity. Secondly, increased vegetation helps improve the water cycle and increases local air humidity, thus improving the global climate [69,70]. In addition, it also reduces dust storm weather in northern China and improves air quality [71,72]. In a similar period, greening was observed not only in the China’s deserts, but also in other arid and semiarid regions [19]. For example, a slight increase was observed in more than 80% of the arid mountain-oasis river basin in northwestern China [73]. In the Hexi region, approxi- mately 70% of the vegetation area presented a significant increasing trend, and only 2.85% pre- sented a significant decreasing trend [74,75]. And a significant increasing trend of the NDVI was observed in Xinjiang [76]. In general, precipitation has been shown to be the main cause of NDVI rise in arid and semiarid regions, such as the Hexi region, Xinjiang, and Central Asia

[34,73,75,76]. However, from this study, although the correlation between precipitation and NDVI was significant on the annual scale, the area with significant correlation was only 15.14%. The positive effect of precipitation and the negative effect of temperature on NDVI, Fig 7. Greening regions caused by human activities. Green areas denote desert regions where greening trends were influenced by human activities. https://doi.org/10.1371/journal.pone.0256462.g007 PLOS ONE China’s deserts greening and response to climate variability and human activities PLOS ONE | https://doi.org/10.1371/journal.pone.0256462 August 30, 2021 13 / 20

and the greater effect of precipitation compared to temperature were also observed in related studies [34,36,56,77,78], mainly because the vegetation growth in arid regions is more limited due to the lack of moisture. The lagged effects of climatic factors on NDVI were not observed at the monthly scale, which may be because the time scale was not fine enough that the lagged effect might be less than one month. Related studies show that in temperate grassland deserts, the average lag time between NDVI and temperature and precipitation is 16.4 days and 11.8 days, respec- tively [79]. In general, the correlations at the seasonal and monthly scales were significantly strong but weaker at the annual scale. Spatially, the correlations with the NDVI at the monthly scale were stronger, and those at the annual scale were weaker and showed a completely different spatial distribution. These results confirm that climate variability is the main explanation of intra- annual NDVI variations, but could not effectively explain the greening and browning trends of China’s deserts from 2000 to 2017 reflected by NDVI at the annual scale. Human activities The results show that the majority of the desert area showed a significantly increasing trend from 2000 to 2017. By comparing remote sensing images of all desert areas that had turned green with those that had not, we found that there are two principal human activities contrib- uting to the greening of deserts in China: oasis expansion and sand stabilization measures (Fig

7). These two anthropogenic impacted desert areas differ. Oasis expansion, especially in the form of cropland expansion, led to considerable greening areas that were mainly distributed in the southern part of the Gurbantunggut Desert, the northwestern edge of the Taklimakan Des- ert, patches at the edges of the Qaidam Basin Desert, the Badain Jaran Desert, and the Tengger Desert. The expansion of crops in the desert region can be connected to inadequate transfer of rural labor forces and grain subsidies, which can lead to excessive consumption of water and disrupt the natural balance between the oasis and the desert [80]. Policies to control the mod- erate expansion of oasis need to be strengthened like facilitating the transfer of agricultural labor to non-agricultural labor and reasonably controlling the grain subsidies in these areas to effectively control the expansion of oasis. Similarly, agricultural practices are the primary cause of greening trends in most regions of the world [22,36]. Sand stabilization measures such as using straw checkerboards and planting grass, shrubs, and trees contributed to quite a few greening areas in the Kubuqi Desert. This result provides evidence of the effectiveness of desertification control in China [81]. By employing sustainable grazing practices, sustainable agriculture practices, water use quotas, and planting wind-shelter forests, China has paid great attention to desertification management [82]. These efforts in China provide a valuable exam-

ple for sustainable desertified land management throughout the world [83]. Since 2000, desert- ification in China has been implemented through national projects (e.g., the Sandstorm Source Control project and the Three-North Shelterbelt System Construction) [82,84]. Despite great progress in the greening of desert areas in China, there is still a long way to go, indicating the need for increased efforts by the government and funding for effective policies and projects [85]. Conclusions This study analyzed greening and browning trends in China’s deserts from 2000 to 2017 shown in the monthly, seasonal, and annual NDVI both temporally and spatially and mea- sured the effects of climate variability and human activities by correlation analysis and residual analysis. The following conclusions were obtained. PLOS ONE China’s deserts greening and response to climate variability and human activities PLOS ONE | https://doi.org/10.1371/journal.pone.0256462 August 30, 2021 14 / 20

The annual NDVI showed a significant increasing trend at a rate of 0.0018/yr, with values of 0.094 in 2000 and 0.126 in 2017. At the seasonal scale, the mean values of the NDVI were higher in summer and autumn, and lower in spring and winter, which showed significant increases in all four seasons, also higher in summer and autumn, and lower in spring and win- ter. The monthly NDVI showed an inverted U-shaped curve during the year, which was the lowest in January (0.0539) and the highest in August (0.0989). The variation of the monthly NDVI trends was similar. In terms of the climatic factors, the variations in monthly, seasonal, and annual temperature and precipitation did not show significant trends from 2000 to 2017. A total of 62.8% of the desert area showed greening trends, and only 0.78% showed brow- ning trends in spatial. The greening trends were mainly distributed on the southern edge of the Gurbantunggut Desert, in the northwestern part of the Taklimakan Desert, and in the Kubuqi Desert. The browning trends were distributed in the northwestern part of the Takli- makan Desert and the eastern part of the Qaidam Basin Desert. In terms of longitude and lati- tude, the area with greening trends was positively correlated with the desert area. The Badain Jaran Desert showed the highest percentage of greening (88.26%), followed by the Kumtag Desert (75.26%) and the Tengger Desert (70.07%); however, the Qaidam Basin Desert showed the highest percentage of browning (1.25%). In terms of seasons, spring had the largest propor-

tion of greening area, followed by summer, winter, and autumn. The spatial distribution of the greening and browning trends showed seasonal differences, and the spatial pattern in summer was the most similar to that at the annual scale, followed by autumn, winter, and spring. At the annual scale, climatic factors had limited explanatory power for NDVI variations, and only precipitation had a significant effect on NDVI, and the effect of precipitation was stronger than that of temperature. At the seasonal and monthly scales, temperature and precipitation were the main factors affecting NDVI variations. Furthermore, the lagged effects of climate fac- tors on NDVI at monthly scales were not observed because the time scale was not fine enough. Human activities played a critical role in the greening of China’s deserts from 2000 to 2017. Specifically, oasis expansion and sand stabilization measures were the two major reasons that led to desert greening. The former occurred mainly in the southern part of the Gurbantunggut Desert, the northwestern edge of the Taklimakan Desert, and patches at the edges of the Qai- dam Basin Desert, the Badain Jaran Desert, and the Tengger Desert, while the latter occurred mainly in the Kubuqi Desert. Supporting information S1 File. Image evidence of human activities. (PDF) S2 File. Data set (part 1). (ZIP) S3 File. Data set (part 2). (ZIP) S4 File. Data set (part 3). (ZIP) S5 File. Data set (part 4). (ZIP) S6 File. Data set (part 5). (ZIP) PLOS ONE China’s deserts greening and response to climate variability and human activities

PLOS ONE | https://doi.org/10.1371/journal.pone.0256462 August 30, 2021 15 / 20

S7 File. Data set (part 6). (ZIP) S8 File. Data set (part 7). (ZIP) Author Contributions Data curation: Xiaoyu Liu. Formal analysis: Xiaoyu Liu. Funding acquisition: Liangjie Xin. Investigation: Xiaoyu Liu. Methodology: Xiaoyu Liu, Liangjie Xin. Project administration: Liangjie Xin. Resources: Liangjie Xin. Supervision: Liangjie Xin. Validation: Xiaoyu Liu. Visualization: Xiaoyu Liu. Writing – original draft: Xiaoyu Liu. Writing – review & editing: Xiaoyu Liu, Liangjie Xin. References 1. Mabbutt JA. Desert landforms: Australian National University Press; 1977. 2. Cooke RU, Warren A, Goudie AS. Desert geomorphology: CRC Press; 1993. 3. Whitford WG, Duval BD. Ecology of desert systems: Academic Press; 2019. 4. Warner TT. Desert meteorology: Cambridge University Press; 2009. 5. Laity JJ. Deserts and desert environments: John Wiley & Sons; 2009. 6. Stocker TF, Qin D, Plattner G-K, Tignor M, Allen SK, Boschung J, et al. Climate change 2013: The physical science basis. Contribution of working group I to the fifth assessment report of the intergovern- mental panel on climate change. 2013;1535. 7. Kim J, Hogue TS. Evaluation and sensitivity testing of a coupled Landsat-MODIS downscaling method for land surface temperature and vegetation indices in semi-arid regions. Journal of Applied Remote Sensing. 2012; 6(1):063569. 8. Xu Y, Yang J, Chen Y. NDVI-based vegetation responses to climate change in an arid area of China. Theoretical and Applied Climatology. 2016; 126(1–2):213–22.

9. Zhou J, Cai W, Qin Y, Lai L, Guan T, Zhang X, et al. Alpine vegetation phenology dynamic over 16 years and its covariation with climate in a semi-arid region of China. Science of the Total Environment. 2016; 572:119–28. 10. Yin H, Li Z, Wang Y, Cai F. Assessment of desertification using time series analysis of hyper-temporal vegetation indicator in Inner Mongolia. Acta Geographica Sinica. 2011; 66(5):653–61. 11. Leprieur C, Kerr Y, Mastorchio S, Meunier J. Monitoring vegetation cover across semi-arid regions: comparison of remote observations from various scales. International Journal of Remote Sensing. 2000; 21(2):281–300. 12. Goward SN, Tucker CJ, Dye DG. North American vegetation patterns observed with the NOAA-7 advanced very high resolution radiometer. Vegetatio. 1985; 64(1):3–14. 13. Malingreau J, Tucker C, Laporte N. AVHRR for monitoring global tropical deforestation. International Journal of Remote Sensing. 1989; 10(4–5):855–67. PLOS ONE China’s deserts greening and response to climate variability and human activities PLOS ONE | https://doi.org/10.1371/journal.pone.0256462 August 30, 2021 16 / 20

14. Martiny N, Camberlin P, Richard Y, Philippon N. Compared regimes of NDVI and rainfall in semi-arid regions of Africa. International Journal of Remote Sensing. 2006; 27(23):5201–23. 15. Olusegun CF, Adeyewa ZD. Spatial and temporal variation of normalized difference vegetation index (NDVI) and rainfall in the north east arid zone of Nigeria. Atmospheric and Climate Sciences. 2013; 2013. 16. Weiss JL, Gutzler DS, Coonrod JEA, Dahm CN. Long-term vegetation monitoring with NDVI in a diverse semi-arid setting, central New Mexico, USA. Journal of Arid Environments. 2004; 58(2):249–72. 17. Li Y, Liu Y, Zhang P, YIN Y-h. Research on the spatio-temporal change of NDVI in the Gurbantunggut Desert. Arid Zone Research. 2009; 26(5):686–93. 18. Propastin P, Kappas M, Muratova N. Inter-Annual Changes in Vegetation Activities and Their Relation- ship to Temperature and Precipitation in Central Asia from 1982 to 2003. Journal of Environmental Informatics. 2008; 12(2). 19. Fensholt R, Langanke T, Rasmussen K, Reenberg A, Prince SD, Tucker C, et al. Greenness in semi- arid areas across the globe 1981–2007—an Earth Observing Satellite based analysis of trends and driv- ers. Remote sensing of environment. 2012; 121:144–58. 20. Kawabata A, Ichii K, Yamaguchi Y. Global monitoring of interannual changes in vegetation activities using NDVI and its relationships to temperature and precipitation. International journal of remote sens- ing. 2001; 22(7):1377–82. 21. Formica AF, Burnside RJ, Dolman PM. Rainfall validates MODIS-derived NDVI as an index of spatio-

temporal variation in green biomass across non-montane semi-arid and arid Central Asia. Journal of Arid Environments. 2017; 142:11–21. 22. Chen C, Park T, Wang X, Piao S, Xu B, Chaturvedi RK, et al. China and India lead in greening of the world through land-use management. Nature sustainability. 2019; 2(2):122–9. https://doi.org/10.1038/ s41893-019-0220-7 PMID: 30778399 23. Evans J, Geerken R. Discrimination between climate and human-induced dryland degradation. Journal of arid environments. 2004; 57(4):535–54. 24. Wessels KJ, Prince S, Malherbe J, Small J, Frost P, VanZyl D. Can human-induced land degradation be distinguished from the effects of rainfall variability? A case study in South Africa. Journal of arid envi- ronments. 2007; 68(2):271–97. 25. He B, Chen A, Wang H, Wang Q. Dynamic response of satellite-derived vegetation growth to climate change in the Three North Shelter Forest Region in China. Remote Sensing. 2015; 7(8):9998–10016. 26. Ichii K, Kawabata A, Yamaguchi Y. Global correlation analysis for NDVI and climatic variables and NDVI trends: 1982–1990. International journal of remote sensing. 2002; 23(18):3873–8. 27. Zhou L, Tucker CJ, Kaufmann RK, Slayback D, Shabanov NV, Myneni RB. Variations in northern vege- tation activity inferred from satellite data of vegetation index during 1981 to 1999. Journal of Geophys- ical Research: Atmospheres. 2001; 106(D17):20069–83. 28. Karnieli A, Agam N, Pinker RT, Anderson M, Imhoff ML, Gutman GG, et al. Use of NDVI and land sur-