Droughts are among the most influential and complex disasters (Liu et al., 2020). Global warming can increase the intensity of droughts (Trenberth et al., 2014). Under the representative concentration pathways (RCPs) scenarios, dryland area is likely to continue to increase by 23% (RCP8.5) and 11% (RCP4.5) by 2100 compared to the baseline period (1961–1990) (Huang et al., 2016a). In China, areas affected by droughts increased from 1980 to 2015 (Shao et al., 2018), and drought frequency and intensity experienced an increasing tendency since the late 1990s (Chen and Sun, 2015). The magnitude and frequency of droughts are expected to increase in the future (Huang et al., 2018; Yuan et al., 2019), leading to severe economic losses in China (Su et al., 2018).

The impacts of droughts on the global terrestrial ecosystem increased in the 20th century. It was found that there will be more severe and frequent droughts in the 21st century and terrestrial ecosystems will need more time to recover (Schwalm et al., 2017). Vegetation ecosystems play a major role in the global water, energy, and carbon cycle (Guo et al., 2017a; Zhao et al., 2018b; Piao et al., 2020). However, droughts can lead to many negative effects on vegetation, such as reduced vegetation cover, decreased productivity, weakened carbon sink capacity, and increased mortality (Doughty et al., 2015; Huang et al., 2016b; Xu et al., 2016), which may threaten the environment and the sustainable development of societies and economies. It is, therefore, of great importance to characterize the response of vegetation to droughts.

A drought index is normally used as a quantitative indicator for monitoring and forecasting droughts. The Standardized Precipitation Evapotranspiration Index (SPEI) is a statistical indicator that takes both the impacts of precipitation and evapotranspiration into account. It was found that SPEI is more appropriate for revealing the magnitude and frequency of drought than the Standardized Precipitation Index (SPI), which only relies on precipitation. SPEI has the advantage of being temporally scalable compared to the Palmer Drought Index (Huang et al., 2016b; Liu et al., 2016). SPEI has therefore been widely applied to characterize the sensitivity of vegetation to droughts. Vicente-Serrano et al. (2013) analyzed the relations between vegetation and droughts globally based on SPEI and Normalized Difference Vegetation Index (NDVI) and found that the time-scales of droughts played important roles in determining the vulnerability of vegetation to droughts. Bunting et al. (2017) showed that different time-scales of climate impacts played important roles in assessing ecological droughts and predicting vulnerable vegetation communities in dryland plant communities of the United States Southwest under climate change scenarios. SPEI was also used to identify the main land cover types affected by drought in different regions (Gouveia et al., 2017; Zhou et al., 2018). Zhang et al. (2017) found that the relations between SPEI and NDVI differed in different areas with diverse precipitation or water balance in China. Thus, the changes in vegetation activity and biomass were highly impacted by the change of water availability. Additionally, arid and semiarid regions were found more vulnerable to droughts in the northern China (Xu et al., 2018; Zhang and Zhang, 2019). As described previously, relations between vegetation and drought are impacted by the time-scales of drought, land cover types, and regional water availability. However, how the relations between vegetation and drought at different time-scales vary depending on the growth stage still need to be studied further.

The Beijing-Tianjin Sandstorm Source Control Project (BTSSCP) was initiated to address the problem of land desertification around the Beijing and Tianjin areas in China. Vegetation not only plays a key role in environmental protection, such as wind prevention, sand fixation, water and soil conservation, land degradation prevention; but is also the basis for the social and economic development in the BTSSR. Li et al. (2016) found that the accumulated NDVI in 59.30% of the BTSSR showed an increasing trend from 2000 to 2010. Shan et al. (2018) found that regional NDVI of BTSSR increased at the rate of 0.002/yr from 2000 to 2014. However, most areas of the BTSSR are characterized by arid and semiarid climate and have frequent drought disasters. Shan et al. (2015) found that both annual precipitation and potential evapotranspiration in the BTSSR showed a downward trend from 1959 to 2011 (–13.4 mm/decade and –3.8 mm/decade, respectively), and a drier climate may limit regional vegetation restoration. Based on the relations of SPI and NDVI, Wu et al. (2014) found that although the ecological construction promoted the increase of vegetation cover from 2000 to 2010 in the BTSSR, drought had negative impacts on vegetation cover from southwest to northeast. A similar result showed that drought has seriously affected vegetation growth in the BTSSR, and drought events increased in the past 15 yr after the implementation of the BTSSCP (Ma et al., 2018). However, most previous studies were conducted in the region that was delimited during the first phase (2000–2010) of the project and covered 75 counties (Wu et al., 2014; Li et al., 2016; Ma et al., 2018; Shan et al., 2018). As the project advances, the extension of research to the whole BTSSR is particularly important. In addition, the Thornthwaite method was used to calculate potential evapotranspiration of SPEI in the BTSSR (Ma et al., 2018). When the potential evapotranspiration was calculated based on the Penman-Monteith method, SPEI could more accurately reflect drought change in the arid and semi-arid regions of China (Zhao et al., 2015). Additionally, existing studies on the relationship between vegetation and drought in the BTSSR were mainly focused on annual scales. There is still a lack of knowledge on the vegetation responses to drought at different time-scales and in different growth stages, subregions, and vegetation types.

This study aims to analyze the relations between vegetation cover and drought conditions in the BTSSR after the implementation of the BTSSCP. The objectives are: 1) to estimate the spatio-temporal changes of vegetation cover (in terms of NDVI) from 2000 to 2017; 2) to characterize possible relationships between NDVI changes and drought at different time-scales and in different growth stages; 3) to obtain insights into which subregions, what vegetation types and when they are more susceptible to drought. This study is not only beneficial to the environmental protection of the BTSSR but is also helpful to understand the relationship between vegetation and drought.

According to the Plan of BTSSCP, the BTSSR was delimited to fight against sandstorms and improve the quality of the ecological environment of the Beijing-Tianjin areas in the north of China. The BTSSCP includes two phases: phase I, from 2000 to 2010 and which covered 75 counties; phase II, from 2013 to 2022, covering 138 counties across six provinces/municipalities/autonomous region (Beijing, Tianjin, Hebei, Shanxi, Inner Mongolia, and Shaanxi), located between 36°48′N–46°46′N and 105°12′E–120°58′E, and divided into five subregions (Fig. 1a). The subregions were classified based on their physical geographical characteristics and the integrity of the administrative boundaries (The Study Team of the Second Stage of the Beijing-Tianjin Sandstorm Source Control Project, 2013).

Figure 1.  Location of the Beijing-Tianjin Sandstorm Source Region (BTSSR) in China. Overview of the available data: (a) subregions (I: Water conservation regions in the Bashang Plateau and hilly regions of the northern China; II: The Hunshandak-Horqin sandy land region; III: Grassland degradation region in the Xilingol Plateau and Wuzhumuqin Basin; IV: Desert steppe degradation region in the Ulanqab Plateau; V: Sandy land region in the Ordos Plateau), elevation, and location of meteorological stations; (b) vegetation types

The BTSSR is dominated by an arid and semiarid continental climate and elevations varying from –5 to 3054 m above sea level. The average annual temperature and precipitation in this region are 6.18℃ and 357 mm, respectively. Climate differs in different seasons, with the average summer (June to August) temperature reaching a high of 20.99℃, while the mean winter (December to February of next year) temperature is –10.32℃. About 65% of the average annual precipitation in this region occurs in summer (234.14 mm). Snow dominates the precipitation in winter and only accounts for 1.91% (6.83 mm) of the annual precipitation. Compared with two subregions (I, V) in the south, three subregions (II, III, IV) in the north of the BTSSR had a lower annual temperature and less precipitation (Table 1). The main vegetation types in the BTSSR are grassland and cultivated vegetation (58.58% and 23.43%, respectively) (Fig. 1b). Vegetation stops growing in winter due to low temperatures and many areas in the BTSSR are covered by snow or ice. Thus, only the growing season (April to October) in the BTSSR was analyzed in this study (Liu and Lei, 2015; Wu et al., 2016).

Daily meteorological data from 75 stations covering the period of 1998–2017 in and around the BTSSR were downloaded from the China Meteorological Data Service Center (http://data.cma.cn/en), including precipitation, mean, maximum and minimum air temperature, relative humidity, sunshine duration, wind speed data (Fig. 1a).

The NDVI data refer to MOD13Q1 in the Moderate Resolution Imaging Spectroradiometer (MODIS) (http://ladsweb.nascom.nasa.gov/) data repositories, covers a time span of 18 yr (2000–2017), with a temporal resolution of 16-d and a spatial resolution of 250 m × 250 m. NDVI was combined at a monthly scale using the Maximum Value Composite (MVC) method (Holben, 1986). Growing-season NDVI in this study refers to the mean value of NDVI from April to October. However, only grid cells with NDVI above 0.1 during the growing season were defined as vegetated areas and were analyzed (Piao et al., 2003; Zhang et al., 2017).

The vegetation map was obtained from the 1∶1 000 000 Vegetation Atlas of China distributed from the Data Center for Resources and Environmental Sciences, Chinese Academy of Sciences (http://www.resdc.cn/). Grassland, forest, shrubland, and cultivated vegetation are the four major vegetation types in the BTSSR and were analyzed.

SPEI characterizes drought, including the roles of precipitation and potential evapotranspiration (PET), objectively reflecting drought changes in the context of significant temperature changes (Vicente-Serrano et al., 2010). In SPEI, first the difference between precipitation and PET is calculated, and then the degree to which the difference deviates from the multi-year average state is used to indicate drought. A detailed description of the calculation procedure of SPEI can be found in the literature (Vicente-Serrano et al., 2010). Negative SPEI values indicate dryness and positive SPEI values indicate wetness. By using meteorological data, PET was calculated based on the FAO-Penman-Monteith method (Allen et al., 1998). The inverse distance weighted (IDW) method was used to perform the SPEI interpolation (Zhao et al., 2018a). The variation of growing-season NDVI was obtained by linear regression (An et al., 2018; Zhou et al., 2018).

The relations between NDVI (pixel-based) and SPEI were explored using Pearson’s correlation coefficients. During the growing season, each pixel had seven NDVI series (from April to October); the correlation coefficients between each series and SPEI (time-scale of 1–24 mon) were calculated. Thus, each pixel had 168 (7 × 24) correlation coefficients. The maximum of correlation coefficients in the growing season was defined as Rmax and the corresponding time-scale of SPEI was defined as Tmax. The maximum of the correlation coefficients between monthly NDVI and SPEI (time-scale of 1–24 mon) was defined as monthly Rmax. Rmax can reflect the highest sensitivity degree of vegetation to the change of drought conditions (Vicente-Serrano et al., 2013).

The values of growing-season NDVI were higher in the southeast of the BTSSR. Higher values of NDVI were mainly distributed in the east of subregion I, middle of subregion II, east of subregion III, and north of subregion V (Fig. 2a). Vegetation cover in most areas of the BTSSR has increased in past decades (2000–2017). In the total vegetated area of the BTSSR, 88.19% of the pixels have experienced increased NDVI in the growing season, 48.3% of the pixels increased significantly (P < 0.05) and mainly took place in subregion I, subregion II, west and northeast of subregion III, west of subregion IV, and subregion V (Table 2, Fig. 2b). However, pixels in 11.81% of the total vegetated areas showed decreased NDVI in the growing season; these pixels were scattered and most of them decreased insignificantly. Approximately only 1.22% of the vegetated areas in the BTSSR experienced a significant decrease in NDVI (P < 0.05). The proportion of pixels with a significantly increased trend was the largest in subregion I, reaching 80.89%, followed by subregion V, II, III, and IV (Table 2). Although the proportion of pixels with increased trend showed little difference in different vegetation types, the proportion of pixels with significantly increased trend was more than 60% in shrubland, forest and cultivated vegetation, but only 38.04% in grassland (Table 3).

Figure 2.  The spatial distribution (a) and trends (b) of NDVI (Normalized Difference Vegetation Index) in the growing season in the BTSSR during 2000–2017

Fig. 3 shows NDVI anomalies and SPEI in the growing season from 2000 to 2017 in the BTSSR. SPEI in the growing season was represented by SPEI-7 in October, which was equivalent to the total water deficit degree from April to October. NDVI anomalies in the growing season in the BTSSR increased significantly (0.003/yr, P < 0.01) from 2000 to 2017 (Fig. 3a). SPEI in the growing season also increased since 2000, with an increasing trend of 0.03/yr (Fig. 3b). The correlation coefficient between NDVI anomalies and SPEI in the growing season was 0.685 (P < 0.01). The NDVI anomalies were relatively higher in 2003, 2008, 2012, and 2016, which was generally consistent with SPEI change dynamics. Meanwhile, the same held true for the lower NDVI anomalies and SPEI in 2001, 2009, and 2015. From 2005 to 2007, NDVI anomalies decreased continuously and corresponded to continuous negative values of SPEI. In general, the negative SPEI anomalies were consistent with the corresponding negative NDVI anomalies during 2000–2017, except for 2015 and 2017 when the NDVI anomalies were positive but also relatively lower than that in adjacent years.

Figure 3.  Inter-annual variation of NDVI anomalies (a) and SPEI (Standardized Precipitation Evapotranspiration Index) (b) in the growing season from 2000 to 2017 in the BTSSR

The average Rmax of the whole vegetated areas was 0.746 and pixels with positive Rmax accounted for 99.99% of the vegetated areas (Fig. 4a). Pixels with significant positive (P < 0.05) Rmax accounted for 97.84% of the vegetated areas, which indicated that vegetation in most areas was sensitive to the change of drought conditions. Pixels with positive but insignificant Rmax were mainly distributed in the northern part of subregion V. Fig. 4b shows the distribution of Tmax (only pixels with significant positive Rmax are shown). Pixels with Tmax of 1–3 mon occupied the largest proportion (33.9%) of the vegetated areas with significant positive Rmax, followed by Tmax of 10–12 mon, which accounted for 21.73% (Fig. 4b). Pixels with Tmax of 13–24 mon were mainly distributed in subregions I, central and western of subregion II, and subregion V, which were more widely distributed in the south of the BTSSR.

Figure 4.  Spatial distribution of Rmax (maximum correlation coefficient in the growing season between NDVI and SPEI) (a) and Tmax (corresponding time-scale of SPEI) (b) in the BTSSR during 2000–2017 (Rmax > 0.4683 indicates significant positive (P < 0.05), Rmax > 0.5897 indicates extremely significant positive (P < 0.01), Rmax > 0.7084 indicates positive correlation at the significance level of 0.001)

Fig. 5 shows the proportions of pixels in vegetated areas that have positive monthly Rmax at different significance. The pixels with positive monthly Rmax in each month accounted for more than 98% of the vegetated areas (Fig. 5). However, the proportion of pixels with significant (P < 0.05) and extremely significant (P < 0.01) positive correlation experienced distinct temporal variation, which increased first from April to July and then decreased gradually. Area-averaged values of monthly Rmax in the area with significant positive monthly Rmax increased first and then decreased after July (Rmax = 0.696) (Fig. 6). In general, the impact of drought on vegetation increased first from April to July, reached the maximum in July, after which it decreased.

Figure 5.  The proportions of pixels with positive monthly Rmax at different significance levels in the BTSSR during 2000–2017

Figure 6.  Mean value of monthly Rmax in the area with significant positive monthly Rmax in the BTSSR during 2000–2017

Monthly Rmax, which was significant positive, demonstrated that vegetation could be significantly affected by drought in that month. Thus, the distribution of significant (P < 0.05) positive monthly Rmax was spatially explored as well (Fig. 7). The significant positive monthly Rmax was distributed in most areas of the studied subregions in July. However, the significant positive monthly Rmax were mainly distributed in the middle of subregion I, west of subregion II, middle of subregion IV, middle and south of subregion V in April. In October, the pixels with significant positive monthly Rmax were mainly distributed in the east of subregion IV and east of subregion V.

Figure 7.  Spatial distribution of monthly Rmax in the BTSSR during 2000–2017

Fig. 8 shows the proportions of areas with significant positive (P < 0.05) correlation coefficients between NDVI and SPEI in different months. The proportions were higher in July than in other months except at the time scale of 22 mon, which demonstrated that vegetation in July could be more strongly affected by drought changes than in other months (Fig. 8). Proportions greater than 50% were mainly distributed from June to August and at the time-scales of 1–15 mon, with the highest proportion in July at 2-month’ time-scale (76%). Additionally, with the increase in time-scale, the changes of the proportions were various in different months. In April and May, the proportions were higher when time-scales were longer than 8–9 mon. The proportions began to be greater than 27% in April when the time-scales were greater than 9 mon. In May, the proportions were higher at the time scales of 9–22 mon, which were above 39%. From June to August, the proportions were lower when time-scales were longer than 13–14 mon. In June and July, the proportions were higher at the time scales of 2–13 mon, which were greater than 49% and 64%, respectively. In August, the proportions were higher at the time scales of 2–14 mon, which were above 47%.

Figure 8.  The proportions of pixels with significant positive (P < 0.05) correlation coefficients between NDVI and SPEI in different months in the BTSSR during 2000–2017

Area-averaged Rmax in all subregions were above 0.72, with the highest area-averaged Rmax located in the subregion III (0.787), followed by the subregions IV, II, V, and I (Fig. 9). Generally, nearly all the pixels in all the subregions existed positive Rmax, except for the subregion V where positive Rmax accounted for 99.95% (Fig. 10). The proportion of areas with significant (P < 0.05) and extremely significant (P < 0.01) positive Rmax was the highest in the subregion III, followed by the subregion IV, II, I, and V. Comparatively, the vegetation of three subregions (II, III, IV) in the north of the BTSSR were more susceptible to the impacts of drought than subregion I and subregion V that in the south of the BTSSR. However, subregion III was the most sensitive area to the impact of drought change.

Figure 9.  Area-averaged Rmax in five subregions of BTSSR during 2000–2017

Figure 10.  The proportions of pixels with positive Rmax at different significance levels in five subregions during 2000–2017

The proportions of areas with significant positive (P < 0.05) correlation coefficients in different months differed with subregions (Fig. 11). When the proportions were greater than 50%, it indicated that more than half of the pixels were sensitive to the change of drought conditions. In subregion I, the proportions greater than 50% were mainly distributed in July and August, with the highest proportion at 3-month’ time-scale in July (70%) (Fig. 11a). In subregion II, the proportions greater than 50% were mainly distributed from June to August, with the highest proportion at 2-month’ time-scale in July (77%) (Fig. 11b). In subregion III, the proportions greater than 50% were mainly distributed from May to September and at the time-scales shorter than 14 mon (Fig. 11c). The higher proportions were distributed in June and greater than 85% at the time-scales of 2–9 mon. In subregion IV, the proportions greater than 50% were mainly distributed from June to September and at the time-scales shorter than 15 mon (Fig. 11d). The higher proportions were distributed in July and at the time-scales of 2–11 mon, which were greater than 80%. In subregion V, the proportions greater than 50% were mainly distributed from April to July (Fig. 11e). From April to June, the proportions greater than 50% were mainly distributed at the time-scales of 10–24 mon. In July, the proportions at all time-scales were greater than 50% (Fig. 11e).

Figure 11.  The proportions of pixels with significant positive (P < 0.05) correlation coefficients between NDVI and SPEI in different months of subregion I (a), subregion II (b), subregion III (c), subregion IV (d), and subregion V (e) during 2000–2017

Higher Rmax and higher proportions of pixels with significant positive Rmax indicate higher sensitivity of vegetation to drought variations. Generally, Rmax in all vegetation types were greater than 0.71; however, the highest Rmax occurred in grassland (0.762), followed by shrubland (0.736), cultivated vegetation (0.723), and forest (0.719) (Fig. 12). Except for cultivated vegetation, all pixels of other vegetation types existed positive Rmax (Fig. 13). The proportions of areas with a significant positive correlation (P < 0.05) were higher than 96% in all the vegetation types. However, the proportion of areas with extremely significant positive correlation (P < 0.01) was the highest in grassland, followed by shrubland, forest, and cultivated vegetation. Generally, the impacts of the variations of drought were significant in all vegetation types, while was strongest in grassland.

Figure 12.  Area-averaged Rmax for different vegetation types in the BTSSR during 2000–2017

Figure 13.  The proportions of pixels with positive Rmax at different significance levels in different vegetation types in the BTSSR during 2000–2017

The proportions of areas with significant positive (P < 0.05) correlation coefficients in different months differed with vegetation types (Fig. 14). For grassland, the proportions greater than 50% were mainly distributed from May to August (Fig. 14a). The proportions greater than 70% were mainly distributed at the time-scales of 2–12 mon in July and with the highest proportion at the 2-month’ time-scale (80%). For shrubland, the proportions greater than 50% were mainly distributed in July and August, and with the highest proportion at the 2-month’ time-scale in July (72%) (Fig. 14b). For forest, the proportions greater than 50% were mainly distributed in July at the time-scales of 1–15 mon, and with the highest proportion at 2-month’ time-scale in July (68%) (Fig. 14c). For cultivated vegetation, the proportions greater than 50% were mainly distributed in July and August and with the highest proportion at 2-month’ time-scale in July (72%) (Fig. 14d).

Figure 14.  The proportions of pixels with significant positive (P < 0.05) correlation coefficients between NDVI and SPEI in different months of grassland (a), shrubland (b), forest (c), and cultivated vegetation (d) in the BTSSR during 2000–2017

Vegetation cover during the growing season increased in the BTSSR from 2000 to 2017. This can be assessed from both the pixels with increased NDVI (which accounted for 88.19% of the vegetated areas) and the increased tendency of NDVI anomalies in the growing season. Previous studies also showed that more than half of the area in the BTSSR experienced increased vegetation (Wu et al., 2014; Li et al., 2016). From 2000 to 2010, pixels with decreased NDVI distributed in the southwest-to-northeast area of the BTSSR (Wu et al., 2014). This phenomenon was not found in our study, which indicates that vegetation cover has improved after 2010. The phase II of the BTSSCP was started in 2013, which was conducive to vegetation restoration. Previous study found that drought increased in the BTSSR from 1960 to 2014, and drought events were frequent from 2000 to 2014 (Ma et al., 2018). We find the wetter trend of climate contributed to the increased vegetation cover in the BTSSR from 2000 to 2017. The differences in the results may be due to the differences in research periods.

Although the overall NDVI increased, years with negative SPEI usually had lower NDVI anomalies, which indicated that drought had negative impacts on vegetation. Pixels with significant positive Rmax accounted for 97.84% of the vegetated areas, which further showed that most vegetated areas were sensitive to the change of drought. Previous researches also reported that the vegetation cover in arid and semiarid areas of China was vulnerable to drought (Hua et al., 2017; Zhang et al., 2017).

The sensitivity of vegetation to climate conditions differs depending on the growth stage (Guo et al., 2017b; Pang et al., 2017; Chen et al., 2018), thus, identifying the sensitive period of vegetation to drought and the changes in sensitivity during different growth stages are important. In the BTSSR, the impact of drought on vegetation was found to rise first and then decrease, and July was the most sensitive month. This phenomenon may be because the vegetation in the early and late growing season in the BTSSR is greatly affected by temperature rather than moisture factors, and the limitation of temperature on vegetation is weakened in summer and moisture factors become to the dominant factors. Piao et al. (2011) showed that spring and autumn NDVI were controlled by temperature in Eurasia, while NDVI in summer was more affected by precipitation. In dry regions of Inner Asia, Mohammat et al. (2013) also found that vegetation activity was constrained by temperature in spring while mainly negatively affected by drought in summer.

Compared with two subregions (I and V) in the south, subregions in the north of the BTSSR (II, III, IV) were more sensitive to the effect of drought, which may be due to the drier climate in the northern region. Besides, there was a wide distribution of grassland in the north of the BTSSR. Relative to other vegetation types, grassland was more sensitive to the impacts of drought; this result is consistent with related research (Zhang et al., 2017). Grass with shallow roots usually absorbs moisture from the surface or middle layer of soil, while shallow soil moisture responds quickly to changes in meteorological drought (Wu et al., 2018). Even in light drought conditions, grass growth can respond quickly. In addition, the xylem system of grass has poor resistance to drought, with weak water and carbon storage capacity, which cannot guarantee normal growth during drought (Craine et al., 2013; Xu et al., 2018). The sensitivity of forest and cultivated vegetation to drought is lower than that of grassland. Deep-rooted forests can absorb water from deep soil to maintain growth during droughts (Wu et al., 2018). Human activities such as irrigation can reduce the impact of drought on cultivated vegetation.

Vegetation growth is affected by cumulative water deficits of the current month and previous months. The multi-temporal scales of SPEI facilitated the analysis of the cumulative impacts (Vicente-Serrano et al., 2013; Ding et al., 2020). Pixels with Tmax of short time-scales (1–3 mon) were found to occupy the largest proportion (33.9%) of the vegetated areas with significant positive Rmax in the BTSSR. A previous study also found that drought at short time-scales had the most serious impact on vegetation in the arid areas of the Loess Plateau (Zhao et al., 2018a). The impacts of drought at a 1-mon time-scale did not have the largest impact on vegetation, which indicated that vegetation was not mostly affected by climate conditions in the current month, and the accumulation of climate impacts was important (Zhao et al., 2020). Therefore, when quantifying the contribution of climate change to vegetation change, we must take into account the cumulative effects of climate change; otherwise, we underestimate the impact of climate. Tmax was helpful to find the most sensitive time scales of vegetation to drought, but some details could be ignored. The time-scale of drought with higher impact was not fixed or unique. In addition to the time-scale with the highest impact, the other time-scales with high impacts should also be noted in drought prevention. For example, for grassland, the areas with significant impacts of drought on vegetation in July were more than 70% at the time-scales of 2–12 mon. In addition, the cumulative effects of drought on vegetation growth varied with growth stages, regions, and vegetation types.

The impact of drought on vegetation is complicated by the many factors involved, e.g., soil water storage, vegetation self-protection mechanism, or root depth (Craine et al., 2013; Wu et al., 2018; Liu et al., 2019). Correlation analysis allowed to analyze the relationship between vegetation and drought on large spatial scales, but it cannot reveal the response mechanism of vegetation under the impact of drought. Future research needs to combine a process-based model with experimental data to further understand the response mechanism of vegetation to drought. Revegetation was found to approach sustainable water resource limits in the Loess Plateau in China (Feng et al., 2016), and the high density of plants in the Mu Us Sandy Land was found to reduce groundwater storage (Zhang and Wu, 2020). The BTSSR is drier than the Loess Plateau, and thus the relation between vegetation increase and local water resources in the BTSSR needs further study. After the completion of the BTSSCP, the sensitivity of vegetation in the region to drought emphasizes the importance of further vegetation protection. Given the increasing risks of droughts in the future (Huang et al., 2016a; Huang et al., 2018; Su et al., 2018; Yuan et al., 2019; Berdugo et al., 2020), we analyzed the relationships between vegetation cover and drought in the BTSSR and identified the time-scales, growth stages, subregions, and vegetation types that sensitive to drought. The findings of this study can provide references for ecological construction and drought prevention in the BTSSR and other ecologically fragile areas with limited water availability.

This study estimated the spatio-temporal changes of vegetation cover and its response to drought in the BTSSR after the implementation of the BTSSCP. The response characteristics of vegetation to drought at different time-scales, in different growth stages, in different subregions and for different vegetation types were analyzed. The main conclusions are as follows:

(1) Vegetation cover increased in most areas of the BTSSR from 2000 to 2017. Pixels with increased NDVI accounted for 88.19% of the total vegetated areas. However, pixels with significant increased NDVI accounted for 48.3% of the total vegetated areas and were mainly distributed in subregions I and V. A wetter climate contributed to the increased vegetation cover in the BTSSR from 2000 to 2017.

(2) During the growing season, NDVI anomalies were closely related to SPEI values. Pixels with significant positive Rmax accounted for 97.84% of the total vegetated areas, which indicated that most areas in the BTSSR were sensitive to changes in drought conditions. Pixels with Tmax at short time-scale (1–3 mon) accounted for the largest proportion (33.9%) in the vegetated areas with significant positive Rmax.

(3) In the BTSSR, the impacts of drought variations on vegetation cover in the growing season rose first and then decreased, with the highest impact in July. Compared with two subregions (I and V) in the south, subregions in the north of the BTSSR (II, III, IV) were more sensitive to the impacts of drought, especially in subregion III. All four major vegetation types in the BTSSR were sensitive to the effects of drought variations, especially for grassland.

(4) The time-scales of drought with higher impacts varied with growth stages, regions, and vegetation types. Vegetation is not only sensitive to drought at a fixed time scale. In addition to the time-scale with the highest impact, the other time-scales with high impacts should also be noted in drought prevention.