Spatiotemporal Dynamics of Aboveground Biomass and Its Influencing Factors in Xinjiang’s Desert Grasslands

Abstract

Grassland biomass is a significant parameter for measuring grassland productivity and the ability to sequester carbon. Estimating desert grassland biomass using the best remote sensing inversion model is essential for understanding grassland carbon stocks in arid and semi-arid regions. The present study constructed an optimal inversion model of desert grassland biomass based on actual biomass measurement data and various remote-sensing product data. This model was used to analyze the spatiotemporal variation in desert grassland biomass and climate factor correlation in Xinjiang from 2000 to 2019. The results showed that (1) among the established inversion models of desert grasslands aboveground biomass (AGB), the exponential function model with the normalized differential vegetation index (NDVI) as the independent variable was the best. Furthermore, (2) the NDVI of desert grasslands in Xinjiang showed a highly significant increasing trend from 2000 to 2019 with a spatially concentrated distribution in the north and a more dispersed distribution in the south. In addition, (3) the average AGB value was 52.35 g·m⁻² in Xinjiang from 2000 to 2019 and showed a spatial distribution with low values in the southeast and high values in the northwest. Moreover, (4) the low fluctuation in the coefficient of desert grassland variation accounted for 65.26% of overall AGB fluctuation (<0.10) from 2000 to 2019. Desert grassland AGB in most areas (88.65%) showed a significant increase over the last 20 years. Lastly, (5) the correlation between desert grassland precipitation and AGB was stronger than that between temperature and AGB from 2000 to 2019. This study provides a scientific basis and technical support for grassland livestock management and carbon storage assessments in Xinjiang.

1. Introduction

Grassland ecosystems are an important part of terrestrial ecosystems and a significant material basis for maintaining ecosystem balance [1,2,3]. Xinjiang has a vast area covered with natural and abundant grassland resources and is a crucial pastoral area in China [4,5,6,7,8]. In recent years, remote sensing technology has played an enormous role in large-scale grassland monitoring and yield estimation. Using remote sensing technology to estimate aboveground biomass (AGB) in grassland and analyze the spatiotemporal distribution characteristics of natural grasslands is extremely important for grassland management and balance analysis [9,10]. Therefore, an accurate assessment of grassland biomass and its variation is important for the carbon cycle and sustainable utilization of natural grasslands in the study area [11,12].

Arid deserts are ecoregions typically affected by scarce precipitation, overgrazing, and social factors associated with human activity. The distribution and growth of desert grasslands are reflected differently in individuals, populations, and communities [13,14,15]. Their fragile ecosystem structure makes desert grasslands a sensitive area for global change response. Despite this, desert grasslands still act as an ecological barrier against environmental degradation [16]. Desert grasslands are the most common grassland type in China, accounting for 47.4% of the total northern grassland area. Not only are desert grasslands a significant barrier but they are also important for maintaining the ecological balance of arid desert areas and developing livestock farming [17]. In addition, these grasslands have high carbon sequestration potentials and are important areas for future CO2 fixation [18]. We conducted a field survey to understand the spatial distribution pattern of desert grassland biomass through measured data. This is extremely important for accurately estimating the carbon reservoir of grassland ecosystems. Currently, many methods are used to estimate grassland biomass. Owing to factors such as differences in study areas, sampling methods, and measurement errors, the estimated grassland biomass differs [19,20]. However, extensive research has already been conducted on large-scale grassland biomass estimation [21,22,23].

With the development and advancement of remote sensing technology, the number of available remote sensing data sources has significantly increased. For example, satellite data such as MODIS, Landsat series, and Gaofen series are widely used because of their high spatial and temporal resolutions. Many researchers have used remote sensing and ground truth data to establish inversion models for monitoring and estimating the yield of grassland vegetation [24,25,26]. Xun et al. [27] studied the relationship of grasslands with the normalized differential vegetation index (NDVI), the modified soil-adjusted vegetation index (MSAVI), and the perpendicular vegetation index (PVI) in Xinjiang and established a multiple regression model. The inverse model was used to estimate the aboveground biomass of Xinjiang grasslands and analyze the spatial and temporal distribution of AGB. Jing et al. [28] combined grassland biomass data with remote sensing data to establish an AGB model and explore the spatiotemporal dynamics of the AGB and its driving factors in northern China over the past 20 years. Ding et al. [29] used vegetation indices, digital elevation models (DEMs), and climatic factors to establish an inverse biomass model. They used this model to analyze the relationship between precipitation, temperature, and AGB, and they found that the main limiting factor for AGB growth was a water deficit. The response of plant growth dynamics to meteorological factors is a popular research topic in this field. Temperature and precipitation are significant factors influencing grassland biomass, with changes in precipitation having a more complex impact on grassland ecosystems, where an increase in precipitation increases the grassland growth force [30,31]. Precipitation is a key factor governing the growth of vegetation in arid and semi-arid regions and an important driver of primary productivity in grassland ecosystems [32]. However, some researchers have argued that elevated temperatures also have a significant effect on the biomass of different grasslands [33]. Therefore, it is necessary to analyze the effects of temperature and precipitation changes on grassland biomass.

In this study, multi-source remote sensing and measured data were used to establish an optimal inversion model. In turn, this model was used to estimate the aboveground biomass of desert grasslands in Xinjiang. We analyzed the variation in the characteristics and spatial distribution of desert grassland biomass with different combinations of climatic factors. This study provides technical tools and theoretical support for the scientific understanding of carbon stocks in the desert grasslands of Xinjiang.

2. Materials and Methods

2.1. Study Area

Xinjiang covers an area of 1.66 million km². Xinjiang lies on the northwestern border of China, between 34°22′00″ N and 49°33′00″ N and 73°22′00″ E and 96°21′00″ E (Figure 1). The grasslands of Xinjiang were classified into three grassland vegetation types: Steppe, meadow, and desert. Xinjiang has a typical continental arid climate with scarce precipitation and dry weather. The average annual amount of rain is approximately 150 mm, which is mainly concentrated from June to August; the mean annual temperature is approximately 6.2 to 9.0°C. The specific geomorphological conditions in Xinjiang have created a complex and diverse range of grassland types.

The three selected desert grassland categories, namely warm steppe desert, warm desert, and alpine desert, account for approximately 47% of the total grassland area in Xinjiang. Desert grassland vegetation in Xinjiang is characterized by a sparse distribution, a strong drought adaptation ability, and developed root systems. The main soil types are calcareous and desert soils in Xinjiang’s desert grasslands. In northern Xinjiang, the vegetation of desert grasslands is mainly composed of small semi-shrub and semi-shrub desert vegetation, such as Anabasis salsa, Reaumuria soongonica, and Nitraria sibirica, whereas in southern Xinjiang, the vegetation of desert grasslands includes Ephedra sp., Kalidium foliatum, and Anabasis aphylla. Because of the wide distribution of desert grasslands in Xinjiang (Figure 2), dynamic monitoring of vegetation changes in these areas is necessary.

2.2. The Inversion of Grassland Biomass

We used a one-dimensional regression as an inversion model to estimate the aboveground biomass of the grasslands. This estimation method uses remote sensing images of a time and place along with their corresponding vegetation index and that of the ground-measured data to establish a linear or nonlinear statistical model. In addition, the statistical model is mainly constructed by combining various remote sensing and measured data. As the calculation is simple and relatively easy to implement, it is a widely used method in biomass estimation. Model structures are mainly linear, logarithmic, power, exponential, and polynomial regression equations; these regression models are widely used for estimating grassland ecosystems in large regions. The method used has the advantages of being accurate, reliable, rapid, and non-destructive [34,35,36].

To establish the optimal inversion model, several vegetation index data were required, including MODIS-NDVI and MODIS-EVI data with a spatial resolution of 250 m and SPOT-NDVI and NOAA CDR NDVI data with spatial resolutions of 1 and 5 km, respectively.

Based on the measured data of desert grasslands in Xinjiang and remote sensing images from different data sources, we developed a model for estimating the biomass of desert grasslands. We established the relationship between vegetation indices and aboveground biomass of desert grasslands and selected the optimal inverse model among linear, logarithmic, power, and exponential functions. Then, we analyzed the accuracy of the estimated model.

First, various statistical measures were calculated to assess the prediction agreement and accuracy of the models, including the R², Lin’s concordance correlation coefficient (LCCC), and root mean square error (RMSE). Then, we selected a set of regression models with the highest inversion accuracy and used the grid calculator tool of the ArcGIS software to estimate the aboveground biomass of desert grasslands in Xinjiang.

2.3. Data Sources

2.3.1. Field Sampling Data

AGB data of the desert grasslands were obtained from the 2019 field investigation in which 149 sample plots were investigated from June to October. Field sampling sites were selected to represent the characteristics of the entire sample area, and their locations were determined according to the different grassland types and vertical distribution characteristics. Each sample plot consisted of three 1 m² sample squares. The latitude, longitude, altitude, and biomass of each sampling point were recorded. All plants in the sample squares were mown flush with the ground, with scrub and tall shrubs cut back to only current year branches, placed into different types of envelopes, and dried in an oven at 65°C to a constant weight. The samples were then weighed, and the average AGB of the three sample squares was calculated.

2.3.2. Remote Sensing and Climate Dataset

MOD13Q1 remote sensing data, launched by the National Aeronautics and Space Administration (NASA), were selected as the data source. The data consisted of 16-day synthetic vegetation index measurements (NDVI/EVI) at a 250 m resolution. The time range was from 2000 to 2019. MODIS data have cloud detection capabilities that are sensitive and suitable for monitoring vegetation changes. First, we used the MODIS reprojection tool (MRT) to convert MOD13Q1 data into the Albers map projection. Then, we used ArcGIS software to process the data with a scale factor of 10,000 (we multiplied 0.0001 on the basis of the given value). Finally, the NDVI and EVI data were analyzed, and we calculated their average values from June to October.

The SPOT VEGETATION NDVI data included the vegetation cover data obtained from the European Union. The data provided an estimate of the surface spectral reflectance, which was corrected for atmospheric conditions. The monthly NDVI maximum was calculated using the maximum value composite (MVC). The spatial resolution of the data was 1 km. The study period was from April 1998 to June 2020.

The NOAA CDR NDVI product dataset was based on NOAA CDR AVHRR NDVI V5 data. We used the R language rgee package to employ the Google Earth Engine for processing and the terra package for band fusion, cropping, and other functions to obtain the data. This dataset contained monthly MVC NDVI data for the Chinese region from 1982 to 2020, with a spatial resolution of 5 km, which can provide long-term data support for studying vegetation changes in China.

The grassland data are based on the 1995 1:1 million grassland-type map from the National Earth System Science Data Center. This map uses data from MSS and TM satellite remote sensing photographs from the mid-1980s and early 1990s, as well as various electronic geographic data sources, to construct an ecological database of Xinjiang.

Temperature and precipitation data were obtained from the National Tibetan Plateau Scientific Data Centre; this included the mean monthly temperature and precipitation data from 2000 to 2019 collected from 105 meteorological stations in Xinjiang. Using the ANUSPLIN interpolation method and adding an elevation factor to improve the accuracy, meteorological data with a spatial resolution of 1 km were obtained. The data sources are presented in

Table 1. The data sources used in the present study.

Data	Year	Spatial Resolution	The Data Source
MODIS-NDVI	2000–2019	250 m	http://ladsweb.modaps.eosdis.nasa.gov/ (8 December 2021)
MODIS-EVI	2019	250 m	http://ladsweb.modaps.eosdis.nasa.gov/ (8 December 2021)
SPOT-NDVI	2019	1 km	https://www.resdc.cn/ (3 July 2021)
NOAA-CDR-NDVI	2019	5 km	http://www.geodata.cn/ (3 July 2021)
Grassland data	1995	/	http://www.geodata.cn/ (8 September 2021)
Land use data	2018	1 km	http://www.resdc.cn/ (20 September 2021)
Temperature Precipitation	2000–2019	1 km	http://data.tpdc.ac.cn/zh-hans/ (8 June 2021)
	2000–2019	1 km	http://data.tpdc.ac.cn/zh-hans/ (8 June 2021)

.

2.4. Data Analysis

2.4.1. Model Establishment and Verification

In the present study, 149 AGB samples were obtained from desert grasslands in Xinjiang, China. The estimation model was established using 70% of the sample data, and accuracy was verified using 30% of the sample data. The Origin 2018 software was used to establish the functional relationship between the vegetation index and the measured data, including linear, logarithmic, power, and exponential relationships.

Model accuracy was verified using the root mean square error (RMSE), mean relative error (MRE), Nash–Sutcliffe efficiency coefficient (NSE), Lin’s concordance correlation coefficient (LCCC), and the ratio of performance to deviation (RPD). The calculation formula was as follows:

$$RMSE=\sqrt{\frac{{\left(Y_i-{Y^{\prime }}_i\right)}^2}{N}}$$

(1)

$$MRE=\frac{{\displaystyle {\sum }_i^N\frac{\left(Y_i-{Y^{\prime }}_i\right)}{{Y^{\prime }}_i}}}{N}$$

(2)

$$NSE=1-\frac{{\displaystyle {\sum }_{i=1}^N{\left(Y_i-{Y^{\prime }}_i\right)}^2}}{{\displaystyle {\sum }_{i=1}^N{\left(Y_i-\overline{Y}\right)}^2}}$$

(3)

where $Y_i$ is the measured desert grassland biomass, $Y_i{}^{\prime }$ is the inverse desert grassland biomass, $\overline{Y}$ is the mean measured desert grassland biomass, and N is the total number of sample points. NSE ranges from negative infinity to 1; the closer the NSE is to 1, the higher the accuracy and credibility of the estimated model.

The LCCC was chosen because it measures the agreement between predicted and measured values and determines how close the predictions fall along a 45-degree line from the origin to the measured data [37,38,39,40].

$$LCCC=\frac{2s_{xy}}{s_x{}^2+s_y{}^2+{\left(\overline{x}-\overline{y}\right)}^2}$$

(4)

$$s_{xy}=\frac{1}{N}{\displaystyle \sum _{n=1}^N\left(x_n-\overline{x}\right)\left(y_n-\overline{y}\right)}$$

(5)

where $\overline{x}$ and $\overline{y}$ are the means for the measured and predicted variables, respectively, and $s_x{}^2$ and $s_y{}^2$ are the corresponding variances. N is the number of samples in the dataset.

$$RPD=\frac{SD}{RMSE}$$

(6)

where SD is the ratio of standard deviation. When LCCC = 1, it indicates a perfect agreement. The agreement can also be excellent (LCCC > 0.9), good (0.80 < LCCC < 0.90), moderate (0.65 < LCCC < 0.80), or poor (LCCC < 0.65). The RPD evaluates prediction accuracy, which can be excellent (RPD > 2.5), very good (2.0 < RPD < 2.5), good (1.8 < RPD < 2.0), fair (1.4 < RPD < 1.8), or poor (RPD < 1.4) [41,42].

2.4.2. Grassland Change Rate

The annual rate of change in NDVI was calculated using a slope to analyze the changes in AGB biomass of the desert grasslands in Xinjiang from 2000 to 2019. The calculation equation was as follows:

$$Slope=\frac{n\times {\displaystyle {\sum }_{i=1}^{\mathrm{n}}i\times x_i-\left({\displaystyle {\sum }_{i=1}^n{x}_i}\right)\left({\displaystyle {\sum }_{i=1}^{n}i}\right)}}{n\times {\displaystyle {\sum }_{i=1}^n{i}^2-{\left({\displaystyle {\sum }_{i=1}^{n}i}\right)}^2}}$$

(7)

where $x_i$ is the mean value of NDVI in year $i$. $n$ = 20 and $i$ ranges from 1 to 20. When Slope is greater than 0, the NDVI and AGB of the desert grassland show an increasing trend and vice versa. For an accurate analysis of grassland growth, we referred to Yang [43] and the characteristics of grassland cover in our study area. The slope was classified into five classes: (<−0.01) deterioration, (−0.01–0) mild degradation, (0–0.01) stabilization, (0.01–0.1) mild improvement, and (>0.1) improvement. The rates of change in temperature, precipitation, and desert grassland biomass were used in this formula.

2.4.3. Effects of Climatic Factors on Aboveground Biomass

Desert grasslands grow in arid and low-rainfall environments where temperature and moisture are the main limiting factors for plant growth. The response of plants to changes in water availability and heat can have far-reaching effects on plant physiological and ecological processes, dry matter accumulation, and ecosystem structure and function. To understand the relationship between desert grassland growth and climatic factors, we used temperature, precipitation, and grassland aboveground biomass data to analyze the correlation between them and explore the effects of temperature and precipitation on the aboveground biomass of desert grasslands in Xinjiang.

To study the correlation of the AGB of the desert grasslands with temperature and precipitation factors, we used the following equation:

$$R_{xy}=\frac{{\displaystyle {\sum }_{i=1}^n\left(x_i-\overline{x}\right)\left(y_i-\overline{y}\right)}}{\sqrt{{\displaystyle {\sum }_{i=1}^n{\left(x_i-\overline{x}\right)}^2{\displaystyle {\sum }_{i=1}^n{\left(y_i-\overline{y}\right)}^2}}}}$$

(8)

where $R_{xy}$ is the correlation coefficient of variables $x$ and $y$; $x_i$ is the AGB value in year $i$, $y_i$ is the value of temperature and precipitation in year $i$, and $\overline{x}$ and $\overline{y}$ are the mean values of temperature and precipitation, respectively.

2.4.4. Mann–Kendall Mutation Test

The Mann–Kendall test is a non-parametric statistical test. For mutation testing of time series ($x$), we need to define a statistic. The calculation equation was as follows:

$$S_k={\displaystyle \sum _{i=1}^k{\displaystyle \sum _j^{i=1}{M}_{ij}}},(k=2,3,\dots ,n)$$

(9)

where $n$ is the sample size. When $x_i>x_j$, $M_{ij}=1$; when $x_i<x_j$, $M_{ij}=0$, $(j=1,2,\dots ,i)$.

Assuming that the time series are randomly independent, we define the statistics:

$$U{F}_k=\frac{S_k-{\overline{S}}_k}{\sqrt{Var(S_k)}},(k=1,2,\dots ,n)$$

(10)

where ${\overline{S}}_k=k(k+1)/4$; $Var(S_k)=k(k-1)(2k+5)/72$.

$U{F}_k$ is the standard normal distribution, which is a sequence of statistics calculated in the order of the time series. We are then given the significance level a. If $\left|U{F}_k\right|>U_{\alpha }$, then there is a significant change in the trend. We sort the time series into reverse order and repeat the above equation and make $U{B}_k=-U{F}_k$, $(k=n,n-1,n-2,\dots ,2,1)$, $U{B}_1=0$. When $U{F}_k>0$, the series is in an upward trend; when $U{F}_k<0$, the series is in a downward trend, and exceeding the threshold indicates a clear trend in the series. The intersection of the two curves, $U{F}_k$ and $U{B}_k$, is the point at which the mutation begins.

3. Results

3.1. Model Development and Validation of Aboveground Biomass Estimation in Desert Grasslands

As shown in

Table 2. Inversion model of desert grasslands with different vegetation indices.

Vegetation Index	Model	Formula	R2	RMSE (g·m−2)	NSE	LCCC	RPD
MODIS-EVI	Linear	y = 415.96x + 16.43	0.54	30.42	0.35	0.62	1.23
	Logarithmic	y = 47.53ln(x) + 173.34	0.41	28.98	0.41	0.58	1.29
	Power	y = 225.48x0.61	0.50	28.26	0.44	0.56	1.33
	Exponential	y = 30.78e5.23x	0.54	26.60	0.50	0.69	1.41
MODIS-NDVI	Linear	y = 424.21x + 9.50	0.62	30.00	0.37	0.66	1.25
	Logarithmic	y = 66.99ln(x) + 186.97	0.50	26.64	0.50	0.66	1.41
	Power	y = 306.68x0.93	0.61	25.01	0.56	0.67	1.50
	Exponential	y = 20.95e5.66x	0.68	21.73	0.67	0.78	1.72
SPOT-NDVI	Linear	y = 410.24x + 6.37	0.52	26.22	0.52	0.67	1.43
	Logarithmic	y = 61.23ln(x) + 189.45	0.46	27.63	0.46	0.63	1.36
	Power	y = 321.07x0.85	0.51	27.03	0.49	0.63	1.39
	Exponential	y = 25.70e5.52x	0.48	27.76	0.46	0.66	1.35
NOAA CDR NDVI	Linear	y = 334.02x + 22.19	0.41	29.09	0.40	0.57	1.29
	Logarithmic	y = 26.63ln(x) + 125.49	0.22	33.20	0.22	0.36	1.13
	Power	y = 124.34x0.35	0.31	33.46	0.21	0.31	1.12
	Exponential	y = 32.56e4.32x	0.41	29.62	0.38	0.55	1.26

Note: RMSE = Root mean square error; NSE = Nash–Sutcliffe efficiency coefficient; LCCC = Lin’s concordance correlation coefficient; RPD = Ratio of performance to deviation. NDVI = Normalized difference vegetation index, EVI = Enhanced vegetation index. The numbers of desert grassland training and test samples were 105 and 44, respectively; x is the vegetation index value (MODIS-NDVI and MODIS-EVI represent normalized difference vegetation and enhanced vegetation indices at 250 m spatial resolution, respectively; SPOT-NDVI and NOAA CDR NDVI represent normalized difference vegetation indices at 1 km and 5 km spatial resolutions, respectively), and y is the estimated aboveground biomass value (g·m⁻²).

, the MODIS-NDVI index function biomass model had the highest accuracy (R² = 0.68) and agreement (LCCC = 0.85) among the studied estimation models. In addition, this model had the best estimation, with a high ratio of performance to deviation (RPD = 1.72), low values of error (RMSE = 21.73 g·m⁻²), and a Nash efficiency coefficient of 0.67 and the highest model credibility. Therefore, we concluded that this model is suitable for estimating the AGB in Xinjiang desert grasslands.

As shown in Figure 3, a significant correlation (p < 0.01) was observed between MODIS-NDVI and AGB. To validate the accuracy of the AGB estimation model for desert grasslands, we used 30% of the sample data. Figure 3 shows that the correlation coefficient between the measured AGB and inverse AGB was 0.74. The RMSE and MRE values of the estimated equations were 15.29 g·m⁻² and 8.45%, respectively. This indicated that the exponential function model of NDVI and AGB was the best at representing the field conditions and can be used to estimate the AGB of desert grasslands in Xinjiang.

3.2. Characteristics of NDVI Changes in Desert Grasslands in Xinjiang

The mean value of NDVI for the desert grasslands in Xinjiang was 0.135, the maximum annual mean value of NDVI in 2019 was 0.152, and the minimum annual mean value of NDVI in 2010 was 0.121 (Figure 4). The trend of NDVI in the desert grasslands in Xinjiang over the last 20 years showed a highly significant increase (p < 0.001), with an average annual increase rate of 0.013. The UF curve from the MK test showed that the NDVI in desert grasslands significantly increased (p < 0.05), and the UF and UB curves of NDVI in desert grasslands intersected in 2015. This indicated that the NDVI of the studied desert grasslands underwent a sudden change in 2015 (Figure 5). Apparent spatial differences in the NDVI change rates of the desert grasslands were observed, with the highest increase of 0.055 a⁻¹ and the largest decrease of 0.040 a⁻¹ in the last 20 years.

Most of the annual mean NDVI values in Xinjiang desert grasslands were between 0 and 0.2, accounting for 85.43% of the total NDVI values. Among them, the largest proportion (46.66%) of desert grassland NDVI was between 0.10 and 0.20. This indicated that the average NDVI value of the Xinjiang desert grasslands remained stable from 2000 to 2019 (Figure 6a). The NDVI of desert grasslands had higher values in the north and west of Xinjiang, showing an increasing trend from south to north and from east to west, with a concentrated distribution in the north and a more dispersed distribution in the south. The proportion of deterioration in the Xinjiang desert grasslands was the lowest NDVI change rate at 0.12% and was mainly distributed in the Yili area of western Xinjiang. The proportion of mild degradation was 2.66%, mainly distributed in the Altay region of northern Xinjiang, whereas the percentage of stabilization in Xinjiang desert grasslands was the highest at 61.59%, mostly concentrated in the north and east of Xinjiang. The proportion of mild improvement was 35.03%, which is the most dispersed and predominately distributed change rate in northern Xinjiang. The proportion of improvement was only 2.66%, mainly in western and northern Xinjiang. The main reasons for the stabilization and improvement of desert grasslands are the strengthening of infrastructure facilities, the implementation of the grass–livestock balance system, and the protection measures for grassland resources. In general, the NDVI of Xinjiang desert grasslands was stable and improved over time, indicating that the vegetation condition has improved over the past 20 years.

3.3. Spatiotemporal Dynamics of AGB in Xinjiang’s Desert Grasslands

The AGB of desert grasslands in Xinjiang was the lowest in 2010 at 45.75 g·m⁻² and reached a maximum of 61.82 g·m⁻² in 2019 (Figure 7). For the whole study area, the AGB showed a slowly decreasing trend from 2000 to 2010, with a decrease of 0.30 g·m⁻²·a⁻¹; on the other hand, from 2010 to 2019, the AGB showed an increasing trend, with an increase of 1.26 g·m⁻²·a⁻¹. Overall, the AGB of desert grasslands in Xinjiang showed a highly significant increasing trend from 2000 to 2019 (p < 0.001), with an average annual increase of 0.53 g·m⁻²·a⁻¹. Altogether, the AGB of the studied area showed a fluctuating increasing trend over the last 20 years.

In Xinjiang, AGB was high in the northwest and low in the southeast (Figure 8a). The average AGB of desert grasslands in Xinjiang ranged from 30 to 50 g·m⁻² at 65.50%, with an average AGB of 52.35 g·m⁻². High AGB areas were mainly located in the eastern and northern parts of Xinjiang and in the Yili River Valley. Desert grassland AGB generally showed spatially greater fluctuations in the northern slopes of the Tianshan Mountains, the Yili River Valley, and the eastern part of Xinjiang, and fewer fluctuations in the other areas. The proportion of the AGB fluctuation degree with small fluctuation variation was 23.35% (<0.05); the relatively low fluctuation variation was 41.91% (0.05–0.10); the medium fluctuation variation was 14.91% (0.10–0.15); and the relatively high and high fluctuation variations were 8.26% (0.15–0.20) and 11.57% (>0.20), respectively (Figure 8b). Overall, low fluctuation in the coefficient of variation in desert grassland AGB accounted for 65.26% of overall AGB fluctuation (<0.10) from 2000 to 2019. In addition, the AGB of Xinjiang desert grasslands is proportional to the coefficient of variation. Areas with higher AGB values have a larger coefficient of variation, indicating a greater dispersion of AGB; areas with low AGB values have a lower coefficient of variation, showing a smaller dispersion of AGB. In general, the AGB of Xinjiang’s desert grasslands has less fluctuation, less dispersion, and better resistance to a harsh environment.

In the entire study region, 81.35% and 18.65% of desert grassland AGB showed increasing and decreasing trends, respectively (Figure 9a). The area with a significantly decreasing or decreasing AGB was small, accounting for only 0.14% of the total desert grassland area; it was mainly distributed in northern Xinjiang (Figure 9b). In general, the AGB of desert grasslands in Xinjiang showed an increasing trend from 2000 to 2019.

3.4. Relationship between AGB and Climatic Factors

To explore the effects of climatic variables on AGB, we use a spline function interpolation function to interpolate the temperature and precipitation data. We analyzed the characteristics of temperature and precipitation variability as well as the spatial variation in annual temperature and precipitation (Figure 10 and Figure 11). As shown in Figure 10a, from 1960 to 2019, temperature anomalies showed an increasing trend in Xinjiang, with most of the negative values and a trend of decreasing temperature within the 1960–1996 period. The temperature anomaly was mostly positive from 1997 to 2019, with an increasing trend, reaching an increase of 0.32 °C (10 a)⁻¹ over the last 20 years. Further, precipitations showed an increasing trend in Xinjiang from 1960 to 2019. The percentages of precipitation anomalies were mostly negative from 1960 to 1999, when the precipitation showed a decreasing trend The precipitation anomaly percentage was mostly positive from 2000 to 2019, with an increasing trend, reaching an increase of 2.18 mm (10 a)⁻¹ over the last 20 years (Figure 10b). In general, the climate of Xinjiang showed overall warming and humidification in the period from 2000 to 2019. The average temperature and precipitation of desert grasslands in Xinjiang from 2000 to 2019 were 4.27°C and 232.13 mm, respectively. In addition, the high-value areas of temperature were distributed in the north and west and the high-value areas of precipitation were mainly distributed in the west and southernmost parts (Figure 11).

Annual averages may not reflect the real trends of temperature and precipitation in desert grasslands. Therefore, we carried out meteorological trend analysis and correlation analyses. As illustrated in Figure 12, the temperature in most areas showed an increasing trend in the past 20 years. The largest increases were observed in the southernmost and northernmost regions of Xinjiang, with a maximum increase of 0.04 °C·a⁻¹. Precipitations showed an upward trend in eastern and southwestern Xinjiang, with a maximum increase of 0.2 mm·a⁻¹. The area of positive correlation between AGB and precipitation accounted for 77.10% of the total desert grassland area, with 12.48% being significantly positively correlated. However, the area of positive correlation between the AGB and temperature was only 49.81% (Figure 13). These results showed that precipitation variation was the main climatic factor affecting the growth of desert grasslands in Xinjiang. The correlation between precipitation and AGB in the studied desert grasslands was greater than that between the temperature and AGB. Overall, the growth of desert grasslands in Xinjiang was shown to be closely related to temperature and precipitation.

4. Discussion

4.1. Comparison with Previous Findings

Remote sensing techniques are widely used in biomass inversion and have been well explored. However, this paper is based on a large amount of measured data and combines vegetation index data from multiple data sources to build an optimal inversion model, which estimates 20 years of desert grassland biomass and has a longer time-series analysis. The average AGB of desert grasslands estimated in the present study was 52.35 g·m⁻², which was similar to the estimated value reported in previous studies. Numerous studies have been conducted on grassland biomass in China. For example, Fang et al. [44] estimated the AGB of desert grassland in China to be 34.2 g·m⁻² based on grass production data, while Ni [45] estimated it to be 45.6 g·m⁻². Both of these estimates were lower than the estimates of the present study. The aboveground biomass of temperate desert grasslands in Inner Mongolia estimated by Hu et al. [46], Ma et al. [47], and Wang et al. [48] was 50.44 g·m⁻², 56.6 g·m⁻², and 57.1 g·m⁻², respectively, which was close to that estimated in the present study. Zhang et al. [49] estimated the annual AGB of grasslands in the arid desert region of Northwest China to be above 100 g·m⁻², and Ma et al. [50] estimated the average AGB of temperate desert grasslands in China to be 69.1 g·m⁻², which was apparently higher than the value estimated in the present study. These differences are caused by different remote sensing data sources, different temporal and spatial resolutions of remote sensing data, and different acquisition methods, resulting in differences in the estimation accuracy. However, uncertainty in the estimated grassland biomass is possible because of factors such as differences in study areas, human error, and the equation structure of the model. First, grassland growth conditions are different in various environments, which results in grassland biomass varying between regions. Second, the sampling process can affect the accuracy of the biomass measurements; human error in the collection, drying, and weighing steps of the actual data is inevitable. In addition, the conventional models were built by drawing on previous studies. However, the constructed models need to be based on the characteristics of the study area. Most previous studies have used one remote sensing data source to build the inverse model, whereas the present study was based on multiple remote sensing data sources, which were used to build an optimal model. This improves the accuracy of grassland AGB estimation. However, the limitation of the present study is that it only covered the desert grasslands in Xinjiang. For these reasons, the desert grassland biomass estimated in the present study is somewhat different from that reported in previous studies.

4.2. Impact of Climatic Factors on Biomass

The growth of grassland vegetation is not only influenced by the internal environment and community characteristics but also by the external climatic environment. Many studies have shown that temperature and precipitation are significant factors influencing grassland biomass, especially in areas with a high correlation between temperature and precipitation [51,52]. To understand the response of desert grassland biomass to hydrothermal factors in Xinjiang, we used a correlation approach to analyze the effects of temperature and precipitation on desert grasslands. Precipitation and temperature showed a significant upward trend in Xinjiang in the period from 2000 to 2019. The correlation between desert grassland biomass and the temperature was weak, and the effect of temperature on desert grasslands was not significant. This showed that temperature fluctuations were not the main factor limiting vegetation growth in areas with sufficient heat. Precipitation was the main climatic factor affecting the AGB of desert grasslands in Xinjiang, and the correlation between the precipitation and AGB of desert grasslands was higher than that between temperature and AGB of desert grasslands. In other words, the contribution of precipitation to AGB was greater than that of temperature in the investigated desert grasslands.

Many studies have shown that grassland biomass is sensitive to precipitation and temperature on a regional or broader scale [53,54,55], which was further validated in the present study. The estimated AGB of desert grasslands was highly correlated with climate, and the area of positive correlation between AGB and annual precipitation of desert grasslands accounted for 77.10% of the total area of desert grasslands, with a significant positive correlation of 2.48%. In large-scale semi-arid grasslands, precipitation has a greater effect on grassland biomass than temperature [56,57,58]. For example, Ni [59] found that precipitation was an important factor affecting temperate grasslands in northern China, and Chen [60] showed that precipitation is a determinant of desert grassland dynamics in Xinjiang. Because Xinjiang is located in the center of Eurasia, it is a typical arid and semi-arid region with a climate characterized by aridity and low rainfall. These climatic characteristics mean the influence of temperature on grassland is less. The growth of desert grasslands in Xinjiang is mainly influenced by precipitation, and the area of desert grasslands positively correlated with precipitation is higher than that of temperature. In addition, desert vegetation is sparse in arid areas, with little precipitation and large evaporation. It is difficult for desert grasslands to save water. Thus, they rely mainly on precipitation to replenish the water needed for plant growth. In general, different grassland types differ in their sensitivity to moisture, and an increase in precipitation during the growing season improves the survival and soil conditions of grassland ecosystems.

4.3. Limitations and Future Research Directions

In the present study, based on the distribution range of desert grasslands in Xinjiang, we developed an index inversion model using measured and remote sensing data. The results showed that the MODIS-NDVI index estimation model works best for monitoring the dynamics of desert grasslands in Xinjiang. Furthermore, we used the optimal inversion model to generate a spatial distribution map of desert grasslands, and we analyzed their spatiotemporal dynamics and driving factors. Our findings provide technical support for grassland livestock management and carbon storage assessments in Xinjiang.

When vegetation growth reaches a certain threshold, the NDVI will be saturated. There are two reasons for this saturation. First, the NDVI is highly sensitive to chlorophyll absorption. When the chlorophyll-a concentration exceeds a certain amount, red light is no longer sensitive to chlorophyll, and saturation is quickly reached. Second, the NDVI equation is non-linear. As vegetation cover increases, the NDVI becomes overly sensitive to red signals. Thus, in areas with high vegetation cover or high biomass content, the NDVI will have different degrees of saturation [61]. However, the NDVI is more sensitive to changes in low-cover vegetation, where saturation values are lower. The subject of this study is desert grassland. The mean value of the NDVI for the desert grasslands in Xinjiang was 0.135, while the maximum annual mean value in 2019 was 0.152, and the minimum annual mean value in 2010 was 0.121, which does not reach the saturation threshold. The NDVI has good applicability and high sensitivity in arid and semi-arid desert grasslands. Because of the sparse growth and low vegetation cover of the desert grasslands in Xinjiang, the NDVI does not appear to be saturated. Nevertheless, these estimation models require further improvements. The relationship between remote sensing data and grassland biomass is complex and needs to be validated using additional sample data. The number of sampling points in the study area was small. Therefore, we studied different estimation models for the same area to improve model accuracy. In addition, we collected samples from June to October, considering seasonal differences.

Numerous studies have shown uncertainty in the estimation of grassland biomass production. On the one hand, the parameters and resolution of the remote sensing data used can vary. On the other hand, interference from external factors, such as grazing disturbances and human activities, also occurs. In general, the factors considered were too homogeneous and the accuracy of the remote sensing data was low, which indicates the need for further improvement. Therefore, in the future, the influence of high-resolution image data and factors such as population density, livestock density, and soil organic matter should be considered. In addition, the relationship between the vegetation index and grassland biomass should be further investigated to improve model accuracy.

5. Conclusions

The biomass of desert grasslands in Xinjiang and its spatial and temporal patterns were investigated here. Using AGB and different remote sensing data of Xinjiang desert grasslands, a suitable biomass estimation model was established. The results showed that the NDVI of the area of the studied grasslands showed a highly significant increasing trend from 2000 to 2019 (p < 0.001, 0.013 a⁻¹), changing abruptly in 2015. Desert grassland areas were found to be more concentrated in the north and more dispersed in the south. The desert grasslands in Xinjiang are less volatile and more resistant to harsh environments. The average desert grassland biomass was 52.35 g·m⁻², and it showed an increasing trend, with a spatial distribution indicating it is high in the northwest and low in the southeast. Finally, the influence of meteorological factors on the biomass of desert grasslands was discussed. In general, the climate of Xinjiang showed overall warming and humidification in the period. The average temperature and precipitation of desert grasslands in Xinjiang from 2000 to 2019 were 4.27 °C and 232.13 mm, respectively. The correlation analysis showed that desert grassland biomass was significantly correlated with precipitation and temperature. The correlation between precipitation and AGB in desert grasslands was better than that between the temperature and AGB, and precipitation contributed more to the biomass of the studied desert grasslands.

One limitation of the present study is that only the desert grasslands in Xinjiang were investigated. In our follow-up work, we will build an inverse model suitable for Xinjiang grasslands. The influence of drivers on AGB is not unique, but they interact with each other. Therefore, future studies should be conducted on the effects of soil moisture, livestock density, and human activities on grassland distribution to fully reveal the trends and spatial and temporal distribution patterns of biomass in the desert grasslands of Xinjiang.