Areas with High Fractional Vegetation Cover in the Mu Us Desert (China) Are More Susceptible to Drought

Abstract

Largescale vegetation reconstruction projects in the western and northern parts of China, along with climate change and increased humidity, have significantly boosted fractional vegetation cover (FVC) in the Mu Us Desert. However, this increase may impact the area’s vulnerability to drought stress. Here, we assessed the area’s susceptibility to hydrometeorological drought by analyzing the maximum correlation coefficients (MCC) derived from the spatiotemporal relationships between FVC and estimates of standardized precipitation evapotranspiration index (SPEI) for the area. The results of the study were as follows: (1) FVC exhibited an increasing trend throughout the growing seasons from 2003 to 2022. Although the region experienced an overall wetting trend, drought events still occurred in some years. MCC-values were predominantly positive across all timescales, suggesting that vegetation generally responded favorably to drought conditions. (2) The order of response of land covertype to drought, from greatest to lowest, was grassland, cultivated land, forestland, and sand land. Cultivated land and grassland exhibited heightened sensitivity to short-term drought; forestland and sand land showed greater sensitivity to long-term drought. (3) With a high FVC, the response of grassland and sand land to drought was significantly enhanced, whereas the response of cultivated land and forestland was less noticeable. (4) Low FVC grassland and sand land have not yet reached the VCCSW threshold and can support moderate vegetation restoration. In contrast, forestland and cultivated land exhibit drought sensitivity regardless of FVC levels, indicating that increasing vegetation should be approached with caution. This research offers a method to evaluate the impact of drought stress on ecosystem stability, with findings applicable to planning and managing vegetation cover in arid and semiarid regions globally.

1. Introduction

Drought stands as one of the most devastating natural disasters, characterized by its unpredictable nature, widespread consequences, and intense severity [1,2]. In the context of global warming, the frequency and intensity of droughts have increased significantly. Generally, drought events trigger water deficits and impose water stress on vegetation, leading to changes in physiological and ecological characteristics, as well as structural functions. Prolonged and sustained droughts can result in extensive mortality in vegetation cover. These impacts cause significant damage to agriculture, water resources, and ecosystems in affected areas, while also posing serious challenges to socio-economic development and ecosystem sustainability [3]. Therefore, a thorough investigation into the relationship between vegetation and drought is of practical significance and scientific value. It offers a theoretical framework for understanding how vegetation responds to climate change and provides a scientific basis for vegetation planning and management in arid and semiarid regions worldwide.

There are various drought indices currently being used to assess the severity of drought [4]. These include the Palmer Drought Severity Index (PDSI) [5], the Standardized Precipitation Index (SPI) [5], and the SPEI [6]. On the one hand, PDSI, based on the soil moisture balance equation, is limited by its inability to monitor droughts across fixed timescales [5]. SPI, on the other hand, relies solely on precipitation data, neglecting other essential factors, such as temperature and evapotranspiration that influence drought intensity and frequency. In semi-arid regions, high temperatures and low vapor pressure make evapotranspiration a key driver of water deficits and ecological stress. SPEI overcomes the limitations of SPI and PDSI by incorporating both precipitation and monthly mean temperature to calculate potential evapotranspiration (PET), thereby accounting for both water supply (precipitation) and atmospheric water demand (evapotranspiration) in the water balance, and allowing assessment across multiple timescales. This makes SPEI particularly effective at characterizing drought events in arid and semiarid regions under ongoing and future climate change.

Drought affects vegetation development and growth [7]. Several indicators, such as the NDVI [8], Vegetation Condition Index (VCI) [9], and Solar-induced Chlorophyll Fluorescence (SIF) [10], have been employed to study the response of vegetation to drought at regional scales. The correlation between NDVI and SPEI has been shown to vary across different vegetation types and seasonal conditions in China [11]. Additionally, in Qinghai Province, NDVI demonstrated a positive correlation with SPEI. This positive correlation suggests that as drought intensified, indicated by decreasing SPEI, NDVI also decreased, pointing to general decline in the region’s grasslands [12]. An analysis of VCI and SPEI revealed that the impact of drought on vegetation covertypes differed across climatic zones, with grasslands being the most sensitive covertype in northwestern China [13]. SPEI exhibited a stronger correlation with SIF compared to NDVI. Both SIF and NDVI responded more rapidly to drought in cultivated lands and grasslands, but more slowly in evergreen broadleaf and mixed forests of the Yangtze River Basin [14]. Although NDVI, VCI, and SIF effectively capture the responses of different land covertypes to drought across multiple timescales, these indices lack ecological interpretability. In contrast, FVC provides a more direct and ecologically meaningful measure by quantifying the actual proportion of green vegetation on the land surface. Unlike NDVI, VCI, and SIF, FVC better captures structural changes in vegetation and can more effectively reflect drought-induced impacts across different land covertypes and timescales.

In addition to these vegetation indices, FVC has proven useful in assessing land-surface degradation, ecosystem characterization, physiological drought, and patterns of vegetation productivity [15]. The relationship between FVC and SPEI in China’s drought-prone Karst region was predominantly positive. However, in the northeastern part of the country, this correlation during spring and summer was mostly negative [16]. Similarly, an analysis of FVC and SPEI in northern Asir, Saudi Arabia, has revealed that changes in vegetation cover tended to be driven by episodes of drought [17].

In the Mu Us Desert of China, Wang et al. [18] quantitatively assessed the response of FVC to climate change and human activities. They found that the correlations between FVC and precipitation were slightly positive, while the correlation between FVC and temperature exhibited significant regional differences. Sun et al. [19] investigated the relationship between vegetation and climate change, finding that NDVI had significant positive correlation with atmospheric temperature and relative humidity, and a significant negative correlation with wind speed and sunshine hours. Wang et al. [20] examined the response of phenology to climatic factors and found that the onset of the growing season was negatively correlated with the air temperature and precipitation of the previous month. In contrast, the end of the growing season was positively correlated with the air temperature in the previous month and cumulative precipitation over the preceding months of the growing season.

However, most existing studies mainly focus on the response of NDVI to drought under different land covertypes or the linear relationship between FVC and environmental factors, and lack sensitivity analysis of different FVC to drought at different time scales. To address this gap, we stratified FVC for each land covertype and calculated the MCC of different FVC and SPEI at different time scales. By analyzing the MCC of different FVC and SPEI at different time scales, we investigated vegetation response to drought changes with increasing FVC, which is beneficial for a more comprehensive understanding of the interaction between vegetation and drought.

In order to prevent desertification, land degradation, and sandstorms, the Chinese central government has implemented large-scale ecological reconstruction projects, resulting in a significant increase in vegetation cover throughout the country. Regions with increased FVC may be more vulnerable to the effects of drought, potentially undermining ecological reconstruction efforts. Vegetation carrying capacity of soil water (VCCSW) quantifies the maximum vegetation cover or biomass that can be sustained when soil-water consumption equals soil-water supply within the root zone under specific climatic conditions, soil textures, and management practices [21]. When vegetation cover or biomass exceeds the VCCSW, plant growth becomes limited and soil desiccation may occur [22,23], thereby undermining ecological restoration efforts. By systematically analyzing how varied FVC vegetation responds to drought stress across different time scales, we investigated the VCCSW of different land covertypes in the arid and semi-arid regions.

The objectives of this study were to: (i) investigate the spatiotemporal, seasonal and growing season variation in FVC and SPEI at multiple timescales (i.e., 1, 3, 6, 9, and 12 months); (ii) analyze the maximum correlation between FVC and SPEI to identify the dominant drought timescales and the response time between vegetation cover and drought; (iii) determine the MCC for different land covertypes to assess their sensitivity to drought; and (iv) explore how various land covertypes with high to low FVC may respond to drought.

2. Materials and Methods

2.1. Study Area

The Mu Us Desert, situated in northcentral China (37°29′–39°21′ N, 107°22′–110°28′ E), spans approximately 39,507 km² and encompasses 10 administrative units, including various counties, banners, and districts (Figure 1a). The study focused on four primary land covertypes: cultivated land, forestland, grassland, and sand land (Figure 1b). The topography is predominantly a plateau, with an average elevation of 1302 m. The elevation gradually decreases from northwest to southeast of the plateau. The climate is temperate continental semiarid, influenced by the East Asian monsoon. The average annual precipitation and temperature ranges from 295.6 to 467.4 mm from northwest to southeast and 7.6 to 9.7 °C from south to north, respectively (Figure 2a,b). The growing season extends from April to October, with spring, summer, and autumn occurring from April to May, June to August, and September to October, respectively [24].

2.2. Data Sources and Preprocessing

2.2.1. Hydrometeorological Data

Hydrometeorological data from nine weather stations surrounding the Mu Us Desert were downloaded from the daily meteorological observation dataset of the China Meteorological Data Service Center (https://data.cma.cn/) (accessed on 30 April 2023). Estimates of SPEI were calculated for several timescales (i.e., monthly, quarterly, semi-annually, nine-months, and annually) based on daily temperature and precipitation from 2003 to 2022. As there is missing data in only three months, missing data were gap-filled using the average values from the same month for those years with available data [25] and then projected for the entire study area using inverse distance weighted (IDW)-based interpolation. Since the IDW interpolation method does not require the estimation of a variogram, it is well suited for regions with relatively low station density [26]. By employing leave-one-out cross-validation, the mean absolute error (MAE) and root mean square error (RMSE) were 0.69 °C and 0.85 °C for temperature, and 9.9 mm and 12.4 mm for precipitation, respectively. These results indicated that the use of IDW method in the study area is reliable. Detailed information about the nine weather stations is provided in

Table 1. Information pertaining to the nine weather stations adjacent to the study area.

Station Number	Number (Figure 1a)	Station Name	Longitude (°)	Latitude (°)	Elevation (m)
53529	1	Otog Banner	107.58	39.05	1381.4
53543	2	Dongsheng	109.59	39.5	1461.9
53545	3	Ejin Hore Banner	109.43	39.34	1367.0
53646	4	Yulin	109.47	38.16	1157.0
53651	5	Shenmu	110.28	38.49	941.1
53723	6	Yanchi	107.23	37.28	1349.3
53725	7	Dingbian	107.35	37.35	1360.3
53735	8	Jingbian	108.48	37.37	1336.7
53740	9	Hengshan	109.14	37.56	1111.0

.

2.2.2. NDVI

NDVI-values were obtained from the MOD13A1 product for the period from 2003 to 2022, provided by the National Aeronautics and Space Administration (NASA; https://www.nasa.gov/) (accessed on 15 May 2023). The acquired images had a spatial resolution of 500 m × 500 m and a temporal resolution of 16 days. The MODIS reprojection tool was used to preprocess the images, including providing batch conversion of file formats and projection transformation to the WGS 1984 UTM Zone 49N coordinate system, resulting in 360 NDVI images. Monthly NDVI data were generated using the maximum value composite method to reduce the effects of clouds and atmospheric interference [27]. Outliers were identified using the Z-score statistic, and both outliers and missing values were replaced or filled with adjacent values [28].

2.2.3. Land Covertype and DEM

Land covertype data for the years 2000, 2005, 2010, 2015, and 2020 were obtained from the Data Center for Resources and Environmental Sciences of the Chinese Academy of Sciences (https://www.resdc.cn/) (accessed on 20 May 2023), with a spatial resolution of 30 m. Land covertypes of the study area were classified into cultivated land, forestland, grassland, waterbodies, construction land, sand land, and unexploited land. To minimize the influence of land covertype changes, only areas of unchanged vegetation were selected to analyze the response of major land covertypes, namely cultivated land, forestland, grassland, and sand land, to episodes of drought [13]. Digital elevation data were sourced from the Geospatial Data Cloud GDEMV3 30-m dataset (https://www.gscloud.cn/search) (accessed on 22 May 2023).

2.3. Methods

2.3.1. Pixel Dichotomy Model

Fractional vegetation cover (FVC) is the planar fraction of green foliage to the land surface area [29]. Fractional cover is calculated using the image pixel dichotomous model, which assumes that each pixel of a given image consists of vegetated and non-vegetated (bare ground) fractions. The resulting spectral information is a linearly weighted combination of the two fractions [30,31]. The formula for calculating FVC typically involves $NDVI$ and can be expressed as:

$$\mathrm{F}\mathrm{V}\mathrm{C}={\displaystyle \frac{NDVI-{NDVI}_{\mathrm{s}\mathrm{o}\mathrm{i}\mathrm{l}}}{{NDVI}_{\mathrm{v}\mathrm{e}\mathrm{g}}-{NDVI}_{\mathrm{s}\mathrm{o}\mathrm{i}\mathrm{l}}}}$$

(1)

where ${NDVI}_{\mathrm{s}\mathrm{o}\mathrm{i}\mathrm{l}}$ is the non-vegetated fraction of an image pixel, and ${NDVI}_{\mathrm{v}\mathrm{e}\mathrm{g}}$ the vegetated fraction of the same pixel in the 20 years. The upper and lower thresholds of $NDVI$ were set at the 5 and 95% confidence level to approximate ${NDVI}_{\mathrm{s}\mathrm{o}\mathrm{i}\mathrm{l}}$ and ${NDVI}_{\mathrm{v}\mathrm{e}\mathrm{g}}$, respectively. $NDVI$ is the average value of each pixel of in the 20 years. As the objective of this study is to analyze the differences in drought response across varying classes of FVC, FVC-values for all land covertypes of interest were categorized into five equal-interval classes based on the equidistant classification method. As the maximum FVC in the study area is 70%, each interval was 14%. The range of observed FVC values (from minimum to maximum) was divided into equal-sized intervals: (i) very low vegetation cover (0–14%), (ii) low vegetation cover (14–28%), (iii) medium vegetation cover (28–42%), (iv) medium-high vegetation cover (42–56%), and (v) high vegetation cover (56–70%).

2.3.2. SPEI

SPEI is calculated based on the water balance represented by the difference between precipitation and potential evapotranspiration. The calculation of SPEI is based on:

$$D_m=P_m-{PET}_m$$

(2)

where $m$ represents the month, $D_m$ the difference between monthly precipitation and potential evapotranspiration (mm), $P_m$ monthly precipitation (mm), and ${PET}_m$ the monthly potential evapotranspiration (mm) calculated using the Thornthwaite equation [32].

The cumulative water surplus and deficit series at different timescales is calculated using the following equation:

$$D_n^k=\sum _{i=0}^{k-1}\left(P_{n-1}-{PET}_{n-1}\right), n\ge k$$

(3)

where $D_n^k$ is the cumulative value of the difference between $P$ and $PET$ for the preceding month of calculation $n,$ and $k$ the selected timescale (e.g., monthly, semi-annually).

To fit the sequence we use a three-parameter log-logistic probability distribution function (PDF):

$$\mathrm{F}\left(x\right)= {\left[1+{\left({\displaystyle \frac{\alpha }{x-\gamma }}\right)}^{\beta }\right]}^{-1},$$

(4)

where are the PDF-specific parameters representing the series’ distribution scale, shape, and origin, respectively.

We standardize $\mathrm{F}\left(x\right)$ to generate the corresponding sequence of SPEI-values:

$$P_r=1-\mathrm{F}\left(x\right).$$

(5)

Viable cumulative probabilities $P_r$’s were determined whenever $P_r$ ≤ 0.5 and

$$\mathrm{S}\mathrm{P}\mathrm{E}\mathrm{I}=\omega -{\displaystyle \frac{{\mathrm{c}}_0+{\mathrm{c}}_1\mathrm{\omega }+{\mathrm{c}}_2{\mathrm{\omega }}^2}{1+{\mathrm{d}}_1\mathrm{\omega }+{\mathrm{d}}_2{\mathrm{\omega }}^2+{\mathrm{d}}_3{\mathrm{\omega }}^2}},$$

(6)

where

$$\omega =\sqrt{-2\ln P_r}.$$

(7)

Whenever $P_r>0.5$, $P_r$ is substituted by ($1-P_r$), and SPEI is calculated as follows:

$$\mathrm{S}\mathrm{P}\mathrm{E}\mathrm{I}=-\left(\omega -{\displaystyle \frac{{\mathrm{c}}_0+{\mathrm{c}}_1\mathrm{\omega }+{\mathrm{c}}_2{\mathrm{\omega }}^2}{1+{\mathrm{d}}_1\mathrm{\omega }+{\mathrm{d}}_2{\mathrm{\omega }}^2+{\mathrm{d}}_3{\mathrm{\omega }}^2}}\right).$$

(8)

Coefficients in Equations (6) and (8) are set to the following values: ${\mathrm{c}}_0 = 2.515517,{ \mathrm{c}}_1 = 0.802853, {\mathrm{c}}_2 = 0.010328, {\mathrm{d}}_1 = 1.432788, {\mathrm{d}}_2 = 0.189269,{ \text{and} \mathrm{d}}_3 = 0.001308$ [33].

To facilitate the analysis of the response of FVC with respect to corresponding variations in SPEI at different timescales, we calculated drought indices for five individual periods, namely monthly (represented by SPEI1), quarterly (SPEI3), semi-annually (SPEI6), nine-months (SPEI9), and annually (SPEI12) from 2003 to 2022. Since SPEI1 is calculated from the precipitation and temperature of the current month, SPEI3 is derived from the precipitation and temperature of the preceding three months, and so on. Therefore, SPEI effectively represents the lagged response of semi-arid ecosystems. Spatiotemporal calculations of SPEI were performed within the R-programming software v. 4.1.3 (https://www.r-project.org) (accessed on 22 April 2023) using the SPEI-package (v. 1.8.1). Computed, spatiotemporal values were resampled at a 500-m resolution and projected to WGS 1984 UTM Zone 49N coordinates to complement the spatial resolution of all MODIS-based images.

2.3.3. Trend Analysis

We conducted mutation analysis on SPEI-values using Pettitt Text [34], and then analyzed the trend of changes before and after mutation. Trends in FVC- and SPEI-values were analyzed using the Sen-Median trend method. Compared to linear regression, the Sen-Median trend method is more robust against errors, less sensitive to outliers, and does not require the data to follow a specific distribution type [35,36,37]. These procedural characteristics make it particularly useful for analyzing environmental data, which can often be irregular and contain anomalies. The Sen-Median trend formula is given as:

$$\beta =\mathrm{m}\mathrm{e}\mathrm{d}\left({\displaystyle \frac{x_j-x_i}{j-i}}\right), 1<j<i<n$$

(9)

where $i$ and $j$ are integers from 1 to n, med the ‘median operator’, and $\beta$ the Sen-Median slope indicating either an increasing ($\beta$ > 0) or decreasing ($\beta$ < 0) trend.

The Mann–Kendall significance test is used to identify regions in gridded areas where changes in long-term trend are statistically significant [38]. The Mann–Kendall significance test is based on the following set of equations:

$$S=\sum _{i=1}^{n-1}\sum _{j=i+1}^n\mathrm{s}\mathrm{i}\mathrm{g}\mathrm{n}\left(x_j-x_i\right),$$

(10)

$$\mathrm{s}\mathrm{i}\mathrm{g}\mathrm{n}\left(x_j-x_i\right)=\left\{\begin{array}{c}1,x_j-x_i>0\\ 0,x_j-x_i=0\\ -1,x_j-x_i<0\end{array}\right.,$$

(11)

$$\mathrm{v}\mathrm{a}\mathrm{r}\left(S\right)={\displaystyle \frac{n\left(n-1\right)\left(2n+5\right)}{18}}, \mathrm{a}\mathrm{n}\mathrm{d}$$

(12)

$$Z=\left\{\begin{array}{c}{\displaystyle \frac{S-1}{\sqrt{\mathrm{v}\mathrm{a}\mathrm{r}\left(S\right)}}}, S>0\\ 0, S=0\\ {\displaystyle \frac{S+1}{\sqrt{\mathrm{v}\mathrm{a}\mathrm{r}\left(S\right)}}}, S<0\end{array}\right.$$

(13)

where $n$ is the length (number of records) of the timeseries, ‘sign’ the sign function, $\mathrm{v}\mathrm{a}\mathrm{r}\left(S\right)$ the variance of the test statistic $S$, and $Z$ the standardized normal test statistic. Changes in FVC-trends are determined to: (i) increase significantly (with obvious visual improvement) when $\beta >0$ and $\left|Z\right|>1.96$; (ii) increase slightly when $\beta >0$ and $\left|Z\right|\le 1.96$; (iii) decrease significantly (with obvious visual degradation) when $\beta <0$ and $\left|Z\right|>1.96$; (iv) slightly decrease when $\beta <0$ and $\left|Z\right|\le 1.96$, and (v) exhibit no change when $\beta =0$. Changes in trend are determined to be statistically significant at a critical p-value (α) of 0.05.

2.3.4. Correlation

Spearman’s correlations (r) between monthly FVC from April to October and SPEI at different timescales (i.e., SPEI1, SPEI3, SPEI6, SPEI9, and SPEI12) were calculated. Spearman’s correlation does not depend on any particular distribution of the variables, making it particularly suitable for a wide range of applications [39,40].

$$r_{a,b}=\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{r}\left({\mathrm{F}\mathrm{V}\mathrm{C}}_a,{\mathrm{S}\mathrm{P}\mathrm{E}\mathrm{I}}_{a,b}\right) 4\le a\le 10, \mathrm{w}\mathrm{i}\mathrm{t}\mathrm{h} b=1,3,6,9, \mathrm{a}\mathrm{n}\mathrm{d} 12$$

(14)

$$r_{\mathrm{m}\mathrm{a}\mathrm{x}}=\mathrm{m}\mathrm{a}\mathrm{x}\left(r_{a,b}\right)$$

(15)

where $a$ is the April-to-October period, $b$ the timescale of associated drought over 1- through 12-month intervals, ${\mathrm{F}\mathrm{V}\mathrm{C}}_a$ the FVC (and NDVI) of the $a$^th month, ${\mathrm{S}\mathrm{P}\mathrm{E}\mathrm{I}}_{a,b}$ the corresponding drought index for month $a$ with a timescale of $b$ months, and $r_{a,b}$ and $r_{\mathrm{m}\mathrm{a}\mathrm{x}}$ the Spearman correlation coefficients between ${\mathrm{F}\mathrm{V}\mathrm{C}}_a$ and ${\mathrm{S}\mathrm{P}\mathrm{E}\mathrm{I}}_{a,b}$ and MCC of available relationships.

The procedure accounted for the accumulation of 35 images (grids) of $r_{a,b}$-based values corresponding to seven months of images over five different timescales. To assess responses of FVC to drought during spring, summer, autumn, and the entire growing season, the generated images of MCC-values were based on monthly calculations for each season. Statistical significance of spatially, gridded correlations was assessed by means of t-tests. Pixel-based MCC-values for the different land covertypes were graded into three statistical significance categories, namely (i) non-significant (with p > 0.05, r < 0, or 0 < r < 0.379), (ii) significant (p < 0.05, 0.379 < r < 0.515), and (iii) highly significant (p < 0.01, r > 0.515).

2.3.5. Interquartile Range

The interquartile range (IQR) is a robust statistical measure commonly used to assess the spread in data. We evaluated the stability of vegetation response to drought for different FVCs by means of the IQR. High IQRs indicate enhanced variance (indicative of low stability) in vegetation response to drought, and vice versa. Kruskal–Wallis tests were employed to determine whether specific IQR-values were significantly different for each land covertype. When IQR-values were determined to be significantly different, Dunn’s multiple comparison tests were then performed to identify differences at different levels of FVC [41,42]. In addition, the median absolute deviation (MAD) was calculated as a complementary robustness measure to further quantify variability in vegetation responses. To characterize the distributional patterns of the MCC, frequency histograms were constructed separately for each major land covertype across the study area.

3. Results

3.1. Spatiotemporal Trends in FVC and SPEI

Areas of grassland, sand land, cultivated land, and forestland constituted 17,100 (44.4%), 12,600 (32.7%), 2700 (6.9%), and 800 km² (1.7%) of the total land surface area of the study area (

Table 2. Land surface area (in km²) for each land covertype and % surface cover as a function of fractional vegetation cover (FVC) classes, I through V.

Land Covertype	Area	I	II	III	IV	V
Cultivated land	2700	0	0.17	2.77	3.51	0.46
Forestland	800	0	0.11	1.12	0.47	0.01
Grassland	17,100	0.04	11.81	28.25	4.21	0.11
Sand land	12,600	0.13	18.24	12.98	1.28	0.04
Other	5200	0.40	4.3	7.49	1.95	0.14
Subtotal	-	0.57	34.63	52.61	11.42	0.76

). Figure 3 provides the multi-year average FVC for the entire growing season from 2003 to 2022. The medium-high coverage and high coverage (FVC > 42%) accounted for 12.2% of the total area. These areas were mainly found to occur on cultivated lands and in forestlands, predominantly in the southern and northeastern regions of the Mu Us Desert, such as in Jingbian County, Hengshan District, Yuyang District, and Shenmu County in Shaanxi Province (Figure 1a). The very low, low, and medium vegetation coverage (FVC < 42%) accounted for 87.8% of the total area. These areas were found to occur mostly throughout the sand lands and grasslands, predominantly in the northwestern part of the study area, including Otog Banner, Otog Front Banner, and Uxin Banner in Inner Mongolia (Figure 1a). Specially, medium-high vegetation coverage was most prominent on cultivated lands, i.e., 3.5% of the total land surface area. Medium vegetation coverage was observed to occur mostly in grasslands (28.3% of the study area) and forestlands (1.1%). Sand lands, comprising 18.2% of the study area, commonly exhibited low vegetation coverage.

Figure 4 gives the growing-season-specific, multi-year trend in FVCs across the Mu Us Desert from 2003 to 2022. The mean FVC from 2003 to 2022 was 32%, with an increasing trend from 26 to 38% at a rate of 6% (95% confidence interval [CI] of 4.8–7.5%) per decade. There were distinct seasonal variations, with the lowest FVC in spring (23%), highest in summer (39%), and medium in autumn (29%). FVC increased for all growing seasons, with rates of 3% (95% CI of 1.9–4.1%) in spring, 7% (95% CI of 5.4–8.6%) in summer, and 5% (95% CI of 3.7–6.3%) per decade in autumn, respectively.

Although FVC exhibited an overall increasing trend across the entire study area from 2003 to 2022, some areas remained unchanged whereas others underwent some level of decline (Figure 5). The improved area accounted for 97.3% of the total land surface area, while the unaffected and degraded areas accounted for 1.7 and 1.0% of the study area, respectively. Within the improved area, 9.3% underwent a slight improvement, primarily in the grasslands and sand lands of the central part of the Otog Front Banner and the southern part of the Uxin Banner (Figure 1a). The remaining 88% of the study area underwent significant improvement, distributed in other regions of the study area (p < 0.05; Figure 1a). Seasonal patterns were largely consistent with the overall trend. During spring, summer, and autumn, more than 94% of the study area showed improvement, with significant improvement observed in the majority of those regions (

Table 3. Area ratio (%) of fractional vegetation cover (FVC) trend in growing season, spring, summer, and autumn, 2003–2022.

FVC Change Category	Change Intensity	Growing Season	Spring	Summer	Autumn
Improvement	Significant improvement	88	73.9	80.3	85.3
	Slight improvement	9.3	20.9	15.9	11.8
unaffected	No change	1.7	3.5	1.8	1.8
Degradation	Significant degradation	0.7	1.4	1.7	0.9
	Slight degradation	0.3	0.3	0.3	0.2

).

Figure 6 shows the average SPEI for the Mu Us Desert for the five timescales (i.e., 1, 3, 6, 9, and 12 months). Over the past 20 years, trends in SPEI have increased slightly, indicating a weakening of drought intensity. The Pettitt test for the SPEI time series showed an abrupt change time in October 2011. Overall, SPEI1, SPEI3, and SPEI6 showed an upward trend until abrupt change time, followed by a downward trend thereafter; SPEI9 and SPEI12 showed a downward trend both before and after abrupt change time, and both indices were abruptly increased at the abrupt change time. Specifically, SPEI12 increased at a rate of 0.02 per decade over the 20-year period, with a rate of −0.02 per decade before 2012 and a rate of −0.08 per decade after 2012. Prior to 2012, drought intensified from 2004 to 2006, then weakened from 2012 to 2013, with occasional prolonged and sustained drought (SPEI12 < 0) occurring from 2004 to 2008. After 2012, droughts became less frequent until 2020. However, their intensity increased after 2020. Although the region experienced an overall wetting trend, with SPEI increasing at a rate of 0.02 per decade, drought events still occurred in some years.

3.2. Correlation Between FVC to Drought for Variable Timescales

The spatial distribution of MCC-values is provided in Figure 7 and their percentage area coverage is summarized in

Table 4. Area ratio (%) of maximum correlation coefficients (MCC) for variable significance levels (as defined in Section 2.3.4) for the growing season, spring, summer, and autumn.

Season	Significance Level
	<0	0–0.379	0.379–0.515	>0.515
Growing season	0.4	33.3	26.6	39.7
Spring	1.5	48.4	29	21.1
Summer	5.9	43.6	19.4	31.1
Autumn	19.9	65.6	10.4	4.1

as per statistical significance categories identified in Section 2.3.4. As shown, the area with positive MCC accounted for 99.6% of the study area for the entire growing season, while the remaining 0.4% was characterized by negative MCC (Figure 7a, Table 4). Positive MCC-values accounted for 98.5% of the total surface area in spring, 94.1% in summer, and 80.1% in autumn (Figure 7b–d and Table 4). During the growing season, significant positive MCC-values were observed in 66.3% of the study area, with 26.6% of the area being significant and 39.7% highly significant. Drought-impacted areas were concentrated in the central, western, southern, and eastern parts of the Mu Us Desert. In spring, 50.1% of the areas showed significant positive MCC, with 29% of the area significant and 21.1% highly significant. Drought-impacted areas were mostly located in the central, western, and eastern parts. In summer, 50.5% of the areas showed significant positive MCC, with 19.4 and 31.1% of that share being significant and highly significant, respectively. Drought-impacted areas were similar to those in spring. In autumn, only 14.5% of the area showed positive MCC, with 10.4 and 4.1% being regarded as significant and highly significant. These areas were sparsely distributed. Non-significant, positive MCC occurred on 48.4, 43.6, and 65.6% of the study area during spring, summer, and autumn, respectively. Non-significant, negative MCC occurred on 19.9, 5.9, and 1.5% of the land area in autumn, summer, and spring, respectively, from highest to lowest.

Figure 8 and

Table 5. Area ratio (%) of the timescale corresponding to the maximum correlation coefficients (MCC) for the growing season, spring, summer, and autumn.

Season	Timescale
	SPEI1	SPEI3	SPEI6	SPEI9	SPEI12
Growing season	49.9	11.4	8.0	11.0	19.7
Spring	39.6	14.9	6.8	9.8	28.9
Summer	45.9	10.1	7.4	11.0	25.6
Autumn	72.4	8.9	1.8	1.8	15.1

illustrate the spatial distribution of MCC throughout the growing season as well as within each individual season. During the growing season, FVC was most sensitive to SPEI1 (49.9%) and SPEI12 (19.7%). The areas sensitive to SPEI1 were mainly located in the southeastern region, whereas those sensitive to SPEI12 were in the central and eastern regions (Figure 8a, Table 5). In spring, FVC was most sensitive to SPEI1 (39.6%) and SPEI12 (28.9%). The areas sensitive to SPEI1 were mainly in the northern and central-eastern regions, while the areas sensitive to SPEI12 were in the central and southern regions (Figure 8b, Table 5). In summer, FVC was most sensitive to SPEI1 (45.9%) and SPEI12 (25.6%). The areas sensitive to SPEI1 were in the southeastern and central regions, while those sensitive to SPEI12 were in the northern region (Figure 8c, Table 5). In autumn, FVC was most sensitive to SPEI1 (72.4%) and SPEI12 (15.1%). The areas sensitive to SPEI1 were in the northern, southern, and east-central regions, while those sensitive to SPEI12 were in the western region (Figure 8d, Table 5). It indicated that during the growing season, FVC was most sensitive to both current-month and annual droughts. The responses of FVC in spring, summer, and autumn to droughts at different timescales followed a similar pattern to that observed during the growing season.

3.3. Land Covertype Response to Drought for Variable Timescales

Figure 9 shows the response of cultivated land, forestland, grassland, and sand land to drought for the different timescales. The response of all land covertypes to drought exhibited an increasing trend followed by a decreasing trend during the growing season. The order of response of land covertype to drought, from greatest to lowest, was grassland, cultivated land, forestland, and sand land. Cultivated land and grassland exhibited heightened sensitivity to short-term drought between April to July, with maximum response in July, yielding MCC of 0.60 and 0.63, respectively (Figure 9a,c). In contrast, forestland and sand land showed greater sensitivity to long-term drought over the same period, with the highest MCC of 0.58 and 0.56, also in July (Figure 9b,d). From August to October, the response of cultivated land, forestland, and grassland to SPEIs at all timescales revealed weak correlation (either positive or negative), only sand land displayed a strong negative correlation (Figure 9d).

3.4. Correlation Between FVC of Similar Land Covertypes to Drought for Variable Timescales

Figure 10 shows the response of different FVC levels to drought within the same land covertypes. For cultivated land and forestland, as FVC increased from class II to V, the median of the MCC increased from 0.45 to 0.51 and from 0.47 to 0.53, respectively (Figure 10a,b). For grassland, as FVC increased from class I to V, the median MCC increased by 1.8 times, from 0.29 to 0.52 (Figure 10c). For sand land, as FVC increased from class I to V, the median MCC increased by 1.9 times, from 0.28 to 0.52 (Figure 10d). Overall, the response to drought stress increased with FVC in grassland and sand land, whereas the increase in drought stress was more modest in cultivated land and forestland.

The Kruskal–Wallis test revealed significant differences in IQR among cultivated land, forestland, grassland, and sand land (p < 0.0001; Figure 10). Subsequent Dunn’s multiple comparison tests confirmed that significant differences in IQR existed among different FVC levels within each land covertype (p < 0.0001; Figure 10). Additionally, IQR decreased with increasing FVC, particularly on sand land, where it declined from 0.24 to 0.16 (

Table 6. Interquartile range (IQR) and median absolute deviation (MAD) of maximum correlation coefficients (MCC) for each land covertype.

Land Covertype	I		II		III		IV		V
	MAD	IQR	MAD	IQR	MAD	IQR	MAD	IQR	MAD	IQR
Cultivated land	-	-	0.19	0.26	0.18	0.24	0.14	0.20	0.13	0.18
Forestland	-	-	0.2	0.28	0.19	0.16	0.14	0.21	0.13	0.18
Grassland	0.15	0.19	0.2	0.26	0.18	0.24	0.15	0.19	0.11	0.15
Sand land	0.2	0.24	0.19	0.24	0.17	0.25	0.17	0.2	0.14	0.16

), indicating that areas with high FVC responded to drought more reliably, suggesting greater stability. The MAD revealed consistent patterns with the IQR (Table 6), indicating that vegetation responses to drought were more stable in areas with higher FVC. The MCC frequency histograms (Figure 11) revealed distinct distributional differences among land covertypes. For cultivated land, forestland, and grassland, MCC values were mainly concentrated in the higher range (>0.4, accounting for approximately 73.5%, 74.3%, and 71% of pixels, respectively), indicating that vegetation responses to drought in these areas were more sensitive and stable. In contrast, MCC values for sand land were largely concentrated in the lower range (<0.4, about 53.4% of pixels), reflecting weaker and less stable vegetation responses to drought.

4. Discussion

4.1. Temporal Trends in Vegetation Cover and Drought

In the southern and northeastern regions, abundant precipitation satisfies the substantial water requirements of cultivated land and forestland, allowing medium-to-high and high FVC vegetation to flourish (Figure 2a and Figure 12a,b). Conversely, in the northwest regions, receiving relatively low precipitation is generally sufficient to meet the modest water requirements of the vegetation, enabling low to medium FVC ecosystems to persist (Figure 2a and Figure 12c,d). In arid and semi-arid regions, vegetation is highly sensitive to climate change [43]. The spatial heterogeneity of precipitation, land covertype, and FVC within the study area provides a robust foundation for investigating vegetation dynamics in response to climate change.

Fractional vegetation cover demonstrated a consistent upward trend throughout the entire growing season, which is consistent with the findings of Shi et al. [44] and Sun et al. [19]. The increase in FVC is attributed to several factors, including rising rainfall, air temperature, atmospheric CO₂ concentration, and nitrogen deposition [19,45]. Additionally, government initiatives such as the Three-North Protective Forest Program, Returning Cultivated Land to Forests and Grasslands, and the Beijing-Tianjin Sand and Dust Source Control Project have played a significant role in enhancing FVC across the region.

The SPEI exhibited a slight upward trend, suggesting a reduction in drought severity (Figure 6). This observation aligns with the findings of Su et al. [46] and Wan et al. [47], who reported a modest decline in drought conditions in northwestern Inner Mongolia since 2000. Driven by climate change, both air temperature and precipitation have increased in the study area; however, the rate of increase in precipitation has outpaced the warming effect on the lower atmosphere, contributing to the slight upward trend observed in SPEI. Notably, the region experienced severe drought events—indicated by low SPEI12—during 2004–2006 and 2020–2022, primarily driven by rising temperatures and declining precipitation (Figure 13). This suggests that the region experienced increased moisture during the study period, which contributed to the rise in FVC. In turn, the higher FVC led to greater water consumption. When precipitation fails to meet this elevated water demand, drought conditions usually develop.

4.2. Selection of Drought Timescale

The impact of drought on vegetation exhibits a lag effect, and employing a multi-timescale analysis helps capture this delayed response [48]. Short-term drought, as indicated by SPEI1, reflects vegetation’s response to current-month water conditions. This is particularly relevant during the spring germination period, when vegetation is highly sensitive to short-term water shortages, and for herbaceous plants, which are especially dependent on short-term water availability. At the seasonal scale, drought as indicated by SPEI3 captures the effects of seasonal water variability on vegetation. This includes the influence of 3–6 months of water availability on crop development, as well as the response of vegetation during its peak summer growth to prolonged drought conditions from March to June. Medium- to long-term drought, as indicated by SPEI6, SPEI9, and SPEI12, reflects vegetation responses to prolonged water deficits. Deep-rooted vegetation, such as trees and shrubs, tends to be more sensitive to drought conditions over these extended timescales, particularly those spanning 6 to 12 months [49,50,51]. Longer-term drought (e.g., 12–24 months) further affects deep soil moisture, reinforcing stress on deep-rooted species. Such lagged responses are characteristic of semi-arid ecosystems, where limited water availability and deep soil moisture dynamics cause vegetation to respond to drought with delays of several months [52]. These delayed responses vary by land covertype, season, and soil depth, making multi-timescale analyses essential for understanding vegetation dynamics and ecosystem resilience in semi-arid regions.

4.3. Impact of Drought on Regional Vegetation

Vegetation growth was strongly impacted by drought in spring and summer but weakly affected in autumn (Figure 7). This pattern aligns with the findings of Tong et al. [24] and Wang et al. [53]. During spring and summer, most vegetation is in the critical stages of growth and development, making it particularly vulnerable to drought stress. In contrast, by autumn, much of the vegetation begins to wither and decline, and the high precipitation from the monsoon during this period mitigates drought impacts [54]. FVC was most sensitive to current-month drought (49.9% of the total area) and annual drought (19.7%; Table 4), which is consistent with studies by Tong et al. [24] and Wang et al. [53]. This sensitivity can be attributed to the differing root depths of vegetation: shallow-rooted herbaceous plants are more responsive to current-month precipitation, while deep-rooted woody plants are more affected by annual precipitation.

Cultivated land and grassland exhibited heightened sensitivity to short-term drought; forestland and sand land showed greater sensitivity to long-term drought (Figure 9). This finding aligns with studies conducted in other arid and semiarid regions of China [55,56]. Vegetation in cultivated land and grasslands primarily consists of shallow-rooted plants that absorb water mostly from the topsoil. During short-term droughts, moisture in the shallow soil layer declines rapidly, leading to reduced photosynthetic activity and, in severe cases, plant mortality. The root system of crops is shallower than that of grass, and the response of cultivated land to drought should be greater than that of grassland. However, our data shows that the response of cultivated land to drought is not as great as grassland. It is likely due to the mitigating effects of irrigation practices in some cultivated land, which alleviates the drought [57]. In contrast, vegetation in forestland and sand land are dominated by deep-rooted woody plants. These plants can access water from deeper soil layers during prolonged droughts. Moreover, their high drought resistance—supported by the ability to store water and carbohydrates in their tissues—further enhances their resilience under long-term water stress [58,59,60]. Sand land exhibited the lowest sensitivity to drought, largely due to the presence of naturally drought-tolerant woody species, such as Artemisia ordosica, Caragana microphylla, and Juniperus sabina. These species adopt adaptive strategies—such as reducing stomatal conductance and developing deep root systems—that enable them to efficiently absorb and conserve water under arid conditions [50,61]. However, the strong negative correlation observed in sand land during August–October may be related to vegetation entering the senescence stage at the end of the growing season. In addition, in sparsely vegetated sandy areas, the noise effect of NDVI may also contribute to the negative correlation [62]. Deng and Xu’s [63,64] studies only analyzed the response of different land covertypes to drought based on the relationship between NDVI and SPEI, without analyzing the sensitivity of different FVCs to drought at different time scales under the same land covertype. This study analyzed the vegetation drought sensitivity based on stratified FVC for each land covertype and calculated the MCC for different FVC and SPEI at different time scales, revealing how vegetation drought sensitivity changes with increasing FVC.

For cultivated and forestland, drought stress increased slightly with rising FVC (Figure 10a,b), whereas in grassland and sand land, drought stress showed a more pronounced increase with increasing FVC (Figure 10c,d). Regions with high FVC, corresponding to increased aboveground biomass and leaf area, tend to be more prone to drought, as increased FVC enhances evapotranspiration capacity, leading to greater soil moisture depletion and reduced water use efficiency. This can result in vegetation exceeding the vegetation carrying capacity of soil water (VCCSW), thereby intensifying drought stress.

Variations in the response to drought among different levels of FVC across various land covertypes are primarily driven by differences in the vegetation-climate-soil-water (VCSW) interactions within a given habitat. In such habitats, the availability of soil water resources for plants may originate from precipitation, the soil’s field capacity within the root zone, lateral soil interflow from upslope areas, or shallow groundwater sources. Within a watershed, concentrated soil interflow from upper slopes accumulates in the lower slopes. In sandy landscapes, excess water from sand dunes tends to collect in the interdunal lowlands. Additionally, shallow groundwater plays a crucial role in replenishing soil moisture, thereby supporting vegetation growth. Beyond soil water availability, VCSW interactions are also influenced by plant-specific water-saving traits—such as efficient leaf structures that minimize transpiration—and by environmental conditions, including atmospheric temperature and solar radiation, which directly affect evapotranspiration rates.

In a given habitat, the vegetation carrying capacity could be explained by water balance between water supply, water demand as the principles of Eagleson model [65]. A conceptual sketch illustrated the relationship between soil water resources and FVC (Figure 14).

(1) Below the VCCSW threshold: When the habitat has not yet reached its VCCSW limit, FVC has the potential to increase. As FVC rises, rainfall interception by vegetation increases, while soil water recharge decreases, and evapotranspiration intensifies. This leads to a reduction in soil moisture within the root zone, decreasing water availability for plants and increasing their vulnerability to drought stress.

(2) At the VCCSW threshold: When the habitat reaches its VCCSW limit—whether that threshold is low or high—the maximum sustainable FVC is constrained accordingly. At this point, there is no surplus, stored soil water. Consequently, regardless of whether the FVC is high or low, vegetation becomes equally sensitive to drought due to the lack of buffer water resources.

When a habitat reaches its VCCSW threshold, drought stress leads to a decline in FVC. After the drought ends, vegetation may exhibit two distinct responses. In one scenario, FVC gradually recovers following the end of the drought, indicating that interannual variations in hydroclimate and FVC remain coupled (i.e., high MCC). In the other scenario, FVC fails to recover after the drought (e.g., wilting, reduced productivity, or vegetation mortality), resulting in FVC being consistently below the VCCSW threshold in subsequent years, and vegetation being insensitive to water (i.e., low MCC).

In grassland and sand land ecosystems, drought stress tends to increase with rising FVC. This is because, at low FVC levels—when the vegetation has not yet reached the VCCSW threshold—these landscapes are less prone to drought stress. However, as FVC increases, so does water demand, leading to increased drought sensitivity. In contrast, cultivated land and forestland exhibit sensitivity to drought stress regardless of whether FVC is low or high. This is because these land covertypes have already reached their VCCSW thresholds.

In habitats with lower FVC, IQR of vegetation response to drought was significantly greater, indicating greater variability and instability. This variability arises because some of these habitats had already reached their VCCSW threshold and were sensitive to drought, while others had not and remained relatively resilient. In contrast, habitats with greater FVC exhibited a smaller IQR, reflecting a more stable vegetation response to drought. This is because these habitats were uniformly near or at their VCCSW thresholds, rendering them consistently sensitive to drought stress.

4.4. Implications and Limitations

In grassland and sandy land, drought sensitivity increased significantly with rising FVC. Nevertheless, this does not negate the need for ecological restoration in low-FVC regions. The key challenge lies in scientifically balancing the objectives of “increasing FVC” and “avoiding increased sensitivity”. Grassland and sand land, primarily located in the northwestern part of the study area, showed only light drought stress when FVC was low. This suggests that these ecosystems have not yet reached their VCCSW threshold, indicating potential for further increases in FVC without immediate drought vulnerability. Therefore, it is recommended that ecological restoration in grassland and sand land areas be guided by the VCCSW theoretical framework. Efforts should avoid the indiscriminate pursuit of increasing FVC and instead focus on scientifically controlling vegetation density. Priority should be given to drought-tolerant, deep-rooted, or water-efficient native species (such as Caragana korshinskii and Salix psammophila). These measures can enhance vegetation water use efficiency and drought adaptability, thereby preventing the system from exceeding its VCCSW threshold. In contrast, forestland and cultivated land—mainly distributed in the southern and northeastern regions—experienced drought stress at both low and high FVC levels. This pattern indicates that these land covertypes are already near or at their water carrying capacity. As a result, further increases in vegetation density could exacerbate drought sensitivity, and high-density planting strategies should be approached with caution. Vegetation reconstruction remains a vital strategy for combating desertification. However, given the differing drought responses among land covertypes (Figure 9), it is essential to select appropriate land covertypes and assess their ecological stability. Understanding how various land covertypes respond to drought provides valuable insights into ecosystem resilience and long-term sustainability.

The Thornthwaite method may underestimate PET in semi-arid regions, as it does not explicitly account for radiation or wind speed. Due to its relatively low data requirements, Thornthwaite was widely applied in data-scarce regions such as our study area, where long-term and spatially consistent radiation and wind speed data are limited [66]. Previous studies have successfully applied Thornthwaite in similar semi-arid regions [66,67]. Moreover, our analysis focuses on the relative temporal variations and correlations between SPEI and FVC rather than absolute PET values. Beyond drought, other natural and anthropogenic factors—such as climate change, soil moisture variability, topography, pest outbreaks, disease, logging, wildfires, and urban expansion—also influence vegetation dynamics. These factors must be considered when evaluating the impacts of drought on vegetation health and stability. Finally, the use of low spatial-resolution datasets (e.g., semi-monthly NDVI, monthly temperature and precipitation, and IDW interpolation from national meteorological networks) may limit the precision of the findings. To enhance the reliability and accuracy of the results, future studies should incorporate high-resolution remote sensing data and field-based observations.

5. Conclusions

In this study, we assessed the Mu Us Desert susceptibility to hydrometeorological drought by analyzing the MCC derived from the spatiotemporal relationships between FVC and estimates of SPEI for the region. The main findings are as follows:

(1)

FVC showed a consistent increasing trend throughout the growing seasons from 2003 to 2022. Despite the overall trend toward increased wetting, droughts still occurred intermittently across the region. Cultivated land and grassland exhibited heightened sensitivity to short-term drought; forestland and sand land showed greater sensitivity to long-term drought. With an increase in FVC, the response of grassland and sand land to drought stress increased, while the response of cultivated land and forestland was marginally increased.

(2)

The MCC-based approach could be applied to monitor vegetation responses to drought and evaluate vegetation stability under water limited conditions.

(3)

Vegetation with high FVC values had a more stable response to drought. Low FVC grassland and sand land have not yet reached the VCCSW threshold and can support moderate vegetation restoration. In contrast, forestland and cultivated land exhibit drought sensitivity regardless of FVC levels, indicating that increasing vegetation should be approached with caution. These findings offer valuable insights for planning water-limited ecological restoration strategies aimed at promoting sustainable landscape management in arid regions.