Introduction

Grasslands account for approximately 20% of global land area. Grasslands are important as the production base of material products, but also because of their function to protect biodiversity and provide ecosystem services. Grasslands play an essential role in regional ecological security and socio-economic development1,2,3. However, as extreme weather events coupled with negative anthropogenic activities have risen over the past decades, ecological problems such as grassland degradation have severely increased in many regions globally3,4. Nearly 39.06% of global grassland area experienced degradation between 2000 and 20193. In China, degraded grassland accounted for 22.7% of total grassland area between 1982 and 20102. Approximately 38.8% of the grasslands on the Qinghai–Tibet Plateau of China also suffered degradation between 2001 and 20135. This indicates that grassland degradation has become a serious environmental and socio-economic problem, which will gradually threaten the sustainable utilization of grassland resources and the regional biogeochemical cycle4,5.

Climate change and human activities are the two leading factors driving terrestrial ecosystem change6,7. Generally, they separately or together affected grassland changes on a regional level8. Zhou et al. and He et al. distinguished the effects of climate and human factors on grassland degradation. The former determined that the contributions of the two factors to grassland degradation in China were almost at an equilibrium from 1982 to 2010. He et al. demonstrated that the grassland degradation in the Liao River Basin from 1999 to 2009 was driven by both natural processes and human activities2,6. In particular, climate change, especially warming, extreme weather phenomena, and altered precipitation patterns will affect the vegetation growing season length, physiological processes, primary productivity, and biodiversity, ultimately impacting the growth and causing degradation of grasslands9. In addition, human activities are also the key factors affecting grassland degradation. Among them, overgrazing is regarded as the leading human cause10. Overgrazing can lead to a decrease in primary productivity of vegetation and changes in soil characteristics2. Therefore, identifying and evaluating the impacts of climate change and human activities on grassland ecosystems is important and may facilitate the development of restoration mechanisms to prevent further grassland degradation1,6.

Grasslands degradation monitoring has typically been conducted by field surveys and field control experiments, which can determine the degradation levels and contributing factors11. However, this method primarily focuses on small-scale and short-term research and it is inefficient in some areas where field investigations are difficult to conduct12, whereas grassland degradation is a protracted and progressive process. As a result, studying grassland degradation on a larger and long-term scale may present some limitations13. Remote sensing data can be obtained for larger-scale and longer-term evaluation and can reveal the historical dynamic change law of grassland dynamics. Moreover, the grassland in our study area covered a large spatial region and remote sensing methods are more effective at monitoring grassland degradation14, for example, the normalized difference vegetation index (NDVI) and net primary productivity (NPP)15,16. The influence of climate change and human activities on grassland changes at the regional17,18,19, watershed20 and national scales9,21 have recently been studied using remote sensing technology. However, there is no consensus on the primary causes and their different impacts on grassland dynamics1.

Traditional quantitative evaluation methods such as multivariate analysis and principal component analysis simply establish the statistical relationship between grassland change and driving factors, but these methods rely primarily on statistical models and ignore the true ecological significance of grassland change9,22. Evans and Geerken proposed a residual trend analysis (RESTREND) based on NDVI providing a new method to distinguish the impact of climate change and human activities on grassland change23. This method assumes that climate change is the only driving factor, establishes the regression equation between climate change and NDVI, calculates the predicted NDVI, and uses the residual between the actual value and predicted NDVI as the anthropogenic impact to identify the main factors affecting grassland degradation6. The RESTREND method has since been frequently utilized to assess the influence of climate and human factors on grasslands on a regional scale and the spatial heterogeneity of the driving factors influencing grassland degradation on a pixel scale24,25,26. Meng et al. used RESTREND and showed that the growth of grasslands and shrublands in the Mongolian Plateau was predominantly affected by human activities24, while He et al. found that natural processes and human activities both contributed to grassland degradation in the Liao River Basin6. Arden et al. proposed using time series segmentation and RESTREND to detect dryland degradation25, and Melakeneh et al. applied this method to assess the rangeland degradation in New Mexico26. However, when this methodology was used to fit the multiple regression equation between climate elements and NDVI, there is no clear approach on how to select climate elements.

The previous studies are confined to separate the influence of climate change and human activities on grassland degradation or restoration at the regional scale, but did not account for the differences in the response to these impacts of grassland with different degradation levels on a large scale. Different degraded grasslands have different vegetation and soil characteristics, such as species richness, soil water holding capacity, water use strategies, and soil organic carbon27, coupled with the high spatial heterogeneity in the impact of climate change on ecosystem dynamics, particularly in fragile ecosystems at higher altitudes28,29. Therefore, different degraded grasslands respond differently to climate change. Song et al. studied the influence of temperature change on nitrogen mineralization in alpine wetlands soils with varying degradation gradients30. Additionally, Dai et al. proposed that grazing management had varied effects on alpine grassland at different stages of degradation31. However, the degree of grassland degradation caused by climate change and human activities at the regional scale remains uncertain. In addition, researchers also study and predict future changes in grasslands, so that management plans can be prepared in advance. Liu et al. superimposed the Hurst index with the changing trend of NPP and identified the regions in China (e.g., Gannan and Ningxia) with a consistent decrease in grassland NPP, accounting for 17.59% of the overall grassland area32. This approach has been used to determine the dynamic change of grassland in the forecasted future based on the superimposed results of the overall change in NDVI and Hurst exponent of NDVI in Central Asia. Areas with continuous improvement and those with continuous degradation accounted for 51.02% and 30.15%, respectively, of the total vegetated area8.

Vegetation fragmentation is a common feature of arid-ecosystem degradation, that lead to the presence of bare-soil patches and results in changes in vegetation cover and spatial heterogeneity33,34. The ecosystems of the Qinghai–Tibet Plateau are fragile and sensitive to climate change and human activities28, and the grassland ecosystem has been degraded4. In addition, the natural grassland has been encroached by a large number of invasive species, and the bare-soil patches eventually turned into large areas of “black soil beach,” resulting in the spatial heterogeneity that characterizes other arid ecosystems34. The Qilian County is located in the northeast of Qinghai- Tibet Plateau. The grassland covers most of the area, and it is very vulnerable to the influence of extreme weather conditions. There are few studies on the law of grassland dynamic change and its dominant factors, as well as the grassland degradation or restoration.

Therefore, this study selects the NDVI and its coefficient of variation (CVNDVI) to quantify the spatial heterogeneity and evaluate grassland degradation. Using Qilian County as the study region, the geographical distribution of grassland degradation or restoration is determined by analyzing the changing trend in NDVI and CVNDVI during a long-term time series (i.e., between 2000 and 2019). Combined with the random forest model and stepwise regression method, the dominant climate variables affecting NDVI change are selected, and the regression equation between NDVI and climate elements is established to improve the residual trend analysis. According to the relative effects of climate change and human activities on the NDVI and CVNDVI, twelve possible scenarios are established to redefine the driving factors of grassland degradation. Finally, the Hurst index was used to forecast the change of grassland degradation or restoration. The findings of the study can provide new insights for separating the influence of climate change and human activities on different degraded grasslands and in clarifying the degradation degrees caused by climate change and human activities.

Materials and methods

Study area

The Qilian County is located in the northeast of the Qinghai–Tibet Plateau and the middle part of Qilian Mountain (Fig. 1a). The mountainous terrain of Qilian County is complex and the vegetation cover types are diverse. The average altitude is 3169 m (Fig. 1b). The vertical changes of climate, soil and vegetation are obvious, and the temperature decreases with the gradual increase of altitude. The annual average temperature is—1.8– 1.0 °C, and the annual precipitation of 406.7 mm35. Total grassland area is 1.12 × 106hm2, accounting for 75.74% of the total area of the county. The grassland types mainly include Meadow, Plain grassland, Desert grassland, Alpine and sub-alpine meadow, and Alpine and sub-alpine grassland (Fig. 1c).

Figure 1
figure 1

Geographical location (a), altitude (b), and the distribution of grassland types (c) in Qilian County. The map was generated by ArcGIS 10.2, URL: https://www.esri.com.

Full size image

Data source and preprocessing

The NDVI data of Qilian Mountain for 2000–2019 were obtained from the National Cryosphere Desert Data Center (http://www.crensed.ac.cn/portal/metadata/a9f71fb8-1b0a-4607-9526-88930d04d134) of China, with a spatial resolution of 30 m. The NDVI of the study area was obtained after clipping. The land cover data were acquired from the Global Land Cover 2000 (GLC2000) product (https://forobs.jrc.ec.europa.eu/products/glc2000/products.php), with a spatial resolution of 1 km, which was utilized to obtain the grassland cover data. Road and river network data from OpenStreetMap (http://download.geofabrik.de/asia/china.html) were used to conceal greater spatial heterogeneity values generated by roads and rivers. The monthly meteorological data used herein included temperature, precipitation, sunshine hours, wind speed, and solar radiation. The data of temperature, precipitation, sunshine hours, and wind speed were obtained from the National Earth System Science Data Center (http://www.geodata.cn/data/index.html?publisherGuid=126744287495931&categoryId=4) with a spatial resolution of 1 km. The solar radiation data were from the National Science & Technology Infrastructure (http://data.tpdc.ac.cn/zh-hans/), with a spatial resolution of 10 km. All meteorological data were processed and resampled to obtain annual scale data with 30 m resolution, which was used to judge the dominant climate factors affecting the NDVI.

Data analysis

NDVI coefficient of variation (CVNDVI)

The CVNDVI is a measurement of the relative fluctuation stability of the change in a variable32. To assess the spatial heterogeneity of vegetation cover, the CVNDVI inside a 3 × 3-pixel moving window of 90 × 90 m area was determined. River and road networks were removed from the computation of the CVNDVI since the substantial spatial heterogeneity surrounding them may not imply degradation as assumed13. In this paper, the CVNDVI was employed to examine the stability of changes in grassland NDVI over a long period, computed as follows:

$${CV}_{NDVI}=\frac{\sqrt{\frac{\sum_{i=1}^{n}({NDVI}_{i}-\overline{NDVI })}{n-1}}}{\overline{NDVI} }$$
(1)

where \({CV}_{NDVI}\) is the coefficient of variation of NDVI; n is the duration of the study; \({NDVI}_{i}\) indicates NDVI in year \(i\), and \(\overline{NDVI }\) is the average NDVI value during the study period.

Trend analysis

Based on the pixel scale, the Sen’s slope estimation method36 was used to estimate the spatial trend of NDVI and CVNDVI variation between 2000 and 2019 in Qilian County. The Mann Kendall statistical test (M–K) was used for the significance of the changes37,38. The Sen’s slope in Eq. (2), and the M–K formulae are shown in Eqs. (3)– (5).

$$M= Median \left(\frac{{a_{j} - a_{i} }}{j - i} \right),\forall j > i$$
(2)

where M is the slope, Median is the median function, and i and j represent the year. If M > 0, it shows a positive trend of the NDVI or CVNDVI in the pixel; if M < 0, it means that the NDVI or CVNDVI in the pixel has a negative trend.

$${Z}_{c}=\left\{\begin{array}{ll}\frac{S-1}{\sqrt{Var\left(S\right)}}, & \quad S>0\\ 0, & \quad S=0\\ \frac{S+1}{\sqrt{Var(S)}} , & \quad S<0\end{array}\right.$$
(3)
$$S=\sum_{i=1}^{n-1}\sum_{j=j+1}^{n}sgn({a}_{j}-{a}_{i})$$
(4)
$$sgn\left({a}_{j}-{a}_{i}\right)=\left\{\begin{array}{ll}1, &\quad {a}_{j}-{a}_{i}>0\\ 0,&\quad{ a}_{j}-{a}_{i}=0\\ -1,&\quad{ a}_{j}-{a}_{i}<0\end{array}\right.$$
(5)

where \({a}_{i}\) and \({a}_{j}\) are the \(i\)-th and \(j\)-th data values, respectively, and \(n\) is the length of the time series. Z is the statistical value of the M–K test, and \(sgn\) is the signum function. For a confidence level \(p\), the Z value should satisfy the formula of \(\left|{Z}_{c}\right|>{Z}_{1-p/2}\).

Grassland degradation levels

The annual median NDVI and CVNDVI inside 3 × 3-pixel moving windows were calculated for each year between 2000 and 2019, and the trend analysis method (2.2.2) was used to study long-term change trends in grassland cover. The slope of NDVI and CVNDVI were used to determine grassland degradation levels, as shown in Table 1. A detailed description of the grassland degradation or restoration levels can be found in Li et al.13.

Table 1 Degrees of grassland degradation or restoration.
Full size table

Grassland degradation hot plots

The hot plots of grassland degradation were classified as places where the NDVI was highly concentrated in grassland degradation area. The hot plots were detected using the Anselin Local Moran’s indicator clustering approach, an index frequently used to analyze the spatial clustering properties of a variable36,39.

Improved residuals analysis

The residual analysis method can discriminate between the impact of climate change and human activities on ecological parameters23. A regression model of climate parameters and the annual maximum NDVI values is first created using multiple correlation regression. This connection may then be used to calculate the predicted NDVI. However, when establishing the multiple regression equation of the NDVI and climate elements, most studies directly use precipitation and temperature as independent variables24. Therefore, to improve the fitting degree of the regression equation, the random forest regression model and stepwise regression method were combined to select the dominant climate variables affecting the NDVI variations. The random forest regression model can rank the importance of multiple independent variables according to dependent variables, with larger values indicating the greater importance40. The stepwise regression method is effective for analyzing multiple variables in multiple linear regression analysis. By introducing independent variables one by one, the variables with significant influence are retained, and the variables with insignificant influence are eliminated. Finally, the fitting degree of the regression equation tends to be perfect41,42.

According to the results shown in Fig. S1 and Table S1, temperature, precipitation, and sunshine hours were finally selected as the climatic factors for predicting the NDVI. Next, a three-element linear regression equation for the annual NDVI and climatic factors was established pixel by pixel between 2000 and 2019. The NDVI affected by climate on the pixel scale of Qilian County for this period can be obtained from the following regression model:

$${NDVI}_{C}=aT+bP+cS+d$$
(6)

where \({NDVI}_{C}\) is the predicted NDVI; a, b, and c are the regression coefficients of T, P, and S, respectively; T, P and S are the annual average temperature (℃), annual accumulated precipitation (mm), and annual accumulated sunshine hours (h), respectively; d is a constant.

The NDVI residuals were calculated as the differences between the predicted and observed NDVI values, which can represent the impact of human activities on the NDVI23. To distinguish the CVNDVI change caused by climate variation from that caused by human activities, the CV of the predicted NDVI was calculated to obtain the predicted CVNDVI, and then the CVNDVI residuals.

$${NDVI}_{H}={NDVI}_{obs}-{NDVI}_{C}$$
(7)
$${CV}_{NDVI(H)}={CV}_{NDVI(obs)}-{CV}_{NDVI(C)}$$
(8)

where \({NDVI}_{H} \;\;or \;\;{CV}_{NDVI(H)}\) is the NDVI or CVNDVI residuals; \({NDVI}_{obs} \;\;or \;\;{ CV}_{NDVI(obs)}\) is the observed NDVI or CVNDVI values based on remote sensing images.

Grassland degradation scenario analysis and quantitative assessment method

The trend of \({NDVI}_{C}\), \({CV}_{NDVI(C)}\), \({NDVI}_{H}\), and \({CV}_{NDVI(H)}\) between 2000 and 2019 were calculated, which represented the change trends in NDVI and CVNDVI under the influence of climate variation and human activities. If the trend is positive, climate change or human activities could contribute to an increase in NDVI or spatial heterogeneity. Conversely, it will lead to a decrease in NDVI or spatial heterogeneity. The classification of the main driving factors of NDVI or CVNDVI change are listed in Table. 2, with the calculated relative contribution rate of climate change and human activities to NDVI or CVNDVI change.

Table 2 Classification criterion and contribution rate of driving factors of NDVI or CVNDVI.
Full size table

The driving factors of grassland degradation or restoration were redefined (Table 3) based on the respective roles of climate and human activities on NDVI and CVNDVI, to develop twelve scenarios (scenarios 1‒12). There are three scenarios (scenarios 1‒3) for each restored grassland pixel: climate-driven grassland improvement (scenario 1, CDI), grassland improvement because of human activity (scenario 2, HDI), and the combined effect of both drivers on grassland improvement (scenario 3, BDI).

Table 3 The twelve scenarios for determining the roles of climate and human activities in grassland dynamic.
Full size table

There are nine scenarios (scenario 4‒12) for each degraded grassland pixel: climate-driven grassland with slight degradation or re-growing (scenario 4, CDSR), grassland with slight degradation or re-growing because of human activity (scenario 5, HDSR), effect of both drivers on grassland with slight degradation or re-growing (scenario 6, BDSR), climate-driven grassland with moderate degradation (scenario 7, CDMD), grassland moderately degraded by human activities (scenario 8, HDMD), effect of both drivers on moderately degraded grassland (scenario 9, BDMD), grassland with severe degradation or desertification caused by climate change (scenario 10, CDSD), grassland with severe degradation or desertification as a result of human activity (scenario 11, HDSD), and effect of both drivers on grassland with severe degradation or desertification (scenario 12, BDSD).

Hurst exponent (H)

Since the Hurst exponent can effectively depict self-similarity and long-term reliance, it is commonly employed in climatology and vegetation studies to analyze the durability of long-term changes in time series data43,44. In this study, the Hurst exponent was estimated by R/S analysis, and the change characteristics of NDVI and CVNDVI were analyzed pixel by pixel to reflect the persistence of the changing trend as follows:

  1. 1.

    The time series is A [\({A}_{i}\)] (i = 1,2, 3,., n), and for each positive integer m ≥ 1:

    $$\overline{{A }_{i}}=\frac{1}{m}\sum_{i=1}^{m}{A}_{i} i=\mathrm{1,2},\dots ,n$$
    (9)
  2. 2.

    Cumulative deviation

    $${X}_{(i,m)}=\sum_{i=1}^{i}({A}_{i}-\overline{{A }_{m}}) 1\le i\le m$$
    (10)
  3. 3.

    Range

    $${R}_{(i)}=max1\le i\le m{X}_{\left(i,m\right)}-min1\le i\le m{X}_{(i,m)}$$
    (11)
  4. 4.

    Standard deviation

    $${S}_{m}={\left[\frac{1}{m}\sum_{i=1}^{m}({A}_{\left(i\right)})-{A}_{m}^{2}\right]}^{1/2}$$
    (12)
  5. 5.

    Calculating the Hurst exponent index

    $$\frac{{R}_{i}}{{S}_{m}}=(c\tau )$$
    (13)
  6. 6.

    Calculating the H value:

    $$\mathrm{log}\left(R/S\right)n=a+H\times \mathrm{log}(n)$$
    (14)

The Hurst exponent has a distribution range of 0 to 1; if 0.5 < H < 1, the future trend will be consistent with the previous one; if H = 0.5, the future trends cannot be forecast; If 0 < H < 0.5, it implies that the changing trend of the variable will be incompatible with the prior one32.

The Hurst exponent of NDVI and CVNDVI was respectively superimposed with their dynamic trend to obtain the corresponding future dynamic trends, and the degradation levels were divided and compared with the degradation levels of 2019 to obtain the forecasted future dynamic change trend of Qilian County.

Results

Dynamic variations in the NDVI and CVNDVI between 2000 and 2019

As shown in Fig. 2a, the mean annual NDVI of grasslands in Qilian County is 0.49, with the highest and lowest NDVI in 2013 (0.55), and 2008 (0.42), respectively, increasing at a rate of 0.005 between 2000 and 2019 (R2 = 0.52). In terms of spatial distribution, the total area with a rising tendency in grassland NDVI is 11.47 × 103 km2, accounting for 93.6% of the total grassland area, with the increasing trend being most visible in Babao Town, Zhamashi Township, and the northwestern part of Yanglong Township. The area with a downward trend in NDVI accounts for 6.4%, located around the northwest of Yeniugou Township and along roads and rivers. The changing trend in grassland NDVI accounted for 58% of the areas passing the significance test (Fig. 2c).

Figure 2
figure 2

(a) and (b) Temporal variation in normalized difference vegetation index (NDVI) and coefficient of variation of NDVI (CVNDVI) between 2000 and 2019. (c) and (d) Spatial variation in normalized difference vegetation index (NDVI) and coefficient of variation of NDVI (CVNDVI) between 2000 and 2019. The map was generated by ArcGIS 10.2, URL: https://www.esri.com.

Full size image

Regarding the CVNDVI (Fig. 2b), it also shows an uptrend between 2000 and 2019 (slope = 0.0006; R2 = 0.65). The area CVNDVI with an increasing trend is 7.10 × 103 km2, accounting for 58.0% of the total grassland area, which is mainly distributed in the northwest of Yeniugou Township and Yanglong Township, indicating that the stability of grassland patches in these areas is reduced. Moreover, 42.0% of the grassland area shows a decreasing trend in grassland CVNDVI, especially around Arou Township and Ebao town in the southeast of Qilian County. In the variation trend of grassland CVNDVI, 20.2% can pass the significance test, which is mainly located in the northwest of Qilian County (Fig. 2d).

Spatial distribution of grassland degradation and degradation hot plots

The spatial distribution of grassland degradation or restoration in Qilian County between 2000 and 2019 was obtained from the trend of NDVI and CVNDVI (Fig. 3a). Approximately 55.0% of the grassland of Qilian County has been degraded, of which the area of slight degradation accounts for 48.6% of the total grassland area. These regions are concentrated in Qilian County’s northwest. The moderate degradation is largely scattered around Yeniugou Township, Babao Town, and Arou Township, which occupies 4.3% of the grassland. Only 1.4% of the whole region has suffered severe degradation, with almost no desertification (0.7%). On the contrary, the restored grasslands area accounts for 45.1% of the total grassland area, and the regions with improving conditions are mainly concentrated in the southeast of Qilian County, such as Arou Township, Ebao Town, and Mole Town, accounting for 40.3%. In conclusion, the spatial distribution pattern of grassland degradation or restoration in Qilian County demonstrated great spatial heterogeneity, with less restoration area than the degradation area (45.1% < 55%); while the southeast was dominated by restoration, the northwest was dominated by degradation.

Figure 3
figure 3

Spatial variation of grassland degradation or restoration for the 2000–2019 period. The map was generated by ArcGIS 10.2, URL: https://www.esri.com.

Full size image

The results of the cluster analysis further reveal that the degraded hot plots are mostly in a point distribution pattern, with no significant clustering characteristics (Fig. 3b). The statistics indicate that 33.7% of grassland degradation had no clustering characteristics, mainly scattered in the northwest of Yanglong Township and Yeniugou Township, and some regions of Babao Town. Approximately 66.3% of grassland degradation was clustered and densely distributed in the northwest of Qilian County.

The impact of climate change and human activities on grasslands dynamics

The impact of climate change on NDVI and CVNDVI

As shown in Fig. 4, there is large spatial heterogeneity in the impact of climate variation on NDVI and CVNDVI dynamic changes in Qilian County between 2000 and 2019 (Fig. 4). The contribution of climate variation to NDVI is mostly above 40%, and the area with a contribution of 40–60% accounts for 37.74% of the total grassland area. Most areas of Yanglong Township and some of Yeniugou Township in the northwest of Qilian County contribute more than 60% to climate variation, indicating that the northwest has been more vulnerable to climate change (Fig. 4a). The contribution of climate variation to CVNDVI is similar to that of NDVI in spatial distribution. Overall, the contribution of climate variation in the northwest is higher than that in the southeast. The area with 80–100% contribution accounts for 29.76% of the total grassland area, mostly around Yanglong Township and Yeniugou Township in the northwest, while Ebao Town, Mole Town, and Arou Township in the southeast contributed less than 40%, indicating that CVNDVI affected by human activities in the southeast is above 60% (Fig. 4b).

Figure 4
figure 4

Contribution of climate change to the NDVI (a) and CVNDVI (b) in Qilian County between 2000 and 2019. The map was generated by ArcGIS 10.2, URL: https://www.esri.com.

Full size image

Effects of climate and human activities on grassland dynamic

Based on the twelve grassland dynamic change scenarios, the spatial distribution of grassland degradation or restoration caused by climate change, human activities, and their joint effect between 2000 and 2019 was determined (Fig. 5a). The grassland improvement regions (BDI) induced by the two factors account for 16.6%, mainly located around Arou Township and Ebao Town in the southeast of Qilian County. The human activities (HDI) and climate-driven (CDI) grassland improvement regions account for 12.2% and 4.2%, respectively, which were scattered across Qilian County. Spatially, the regions of climate-driven grassland with slight degradation or re-growing (CDSR) were observed around Yanglong Township, the northwest of Yeniugou Township, and the north of Arou Town, accounting for 12.3% (Figs. 5b, 6). The regions where human activities-driven grassland with slight degradation or re-growing (HDSR) account for 9.5%, with scattered distribution (Figs. 5b, 6). The regions of grassland with slight degradation or re-growing were dominated by both drivers (BDSR) account for the largest proportion, reaching 34.3%, and are mainly located around Zhamashi Township and Babao Town in the middle of Qilian County, as well as the northwest of Yanglong Township and Yeniugou Township. Furthermore, the grassland with moderate degradation caused by climate change (CDMD), human activities (HDMD), and both drivers (BDMD) are mainly concentrated in the northwest of Qilian County (Yeniugou Township, the border region between Yeniugou Township and Yanglong Township), accounting for 7.1%. The grassland with severe degradation is mainly controlled by climate change (CDSD), human activities (HDSD), and both drivers (BDSD) account for a relatively modest proportion, that is, 1.2%, 1.1%, and 1.3% respectively, located around Yeniugou Township (Figs. 5b, 6).

Figure 5
figure 5

Effects of climate and human activities on grassland degradation (b) and restoration (a). The map was generated by ArcGIS 10.2, URL: https://www.esri.com.

Full size image
Figure 6
figure 6

Area proportion of driving factors for grassland degradation and restoration.

Full size image

In summary, climate change and human activities jointly affected grassland restoration or degradation in Qilian County between 2000 and 2019. The contribution of the CDI region was less than that of the HDI region (4.2% < 12.2%), implying that human activities dominated grassland restoration in Qilian County for the 2000–2019 period. Secondly, regardless of the degree of degradation, the area of climate-driven grassland degradation was larger than that of human activities-driven, such as slight degradation or re-growing (12.3% vs. 9.5%), moderate degradation (2.3% vs. 1.5%), and severe degradation or desertification (1.2% vs. 1.1%), indicating that climate change played a leading role in grassland degradation.

Forecasted patterns of grassland dynamics

The dynamic change trend in grassland in Qilian County shown in Fig. 7 indicate that the area of grassland improvement accounts for only 4.31% of the total grassland area, of which the area of continuous improvement accounts for 2.86%. The area from degradation to improvement accounts for 1.45%, mainly in the south of Babao Town and Arou Town, as well as near the junction of alpine and sub-alpine meadow and bare rocks. Desert grassland and alpine and sub-alpine meadow are the main grassland types (Fig. 8), indicating that the grassland in these areas have been reasonably protected and gradually improved from degradation. The area with grassland degradation accounts for 95.69% of the total grassland, with continuous degradation accounting for 54.78% and the area from improvement to degradation accounting for 40.91%. The region of continuous degradation is mainly distributed in the northwest of Qilian County, especially in the northwest of Yanglong Township and Yeniugou Township. In addition, 65.7% and 59.9% of alpine and sub-alpine plain grassland and alpine and sub-alpine meadows, respectively, have been continuously degraded (Fig. 8). The region from improvement to degradation is mainly Arou Township and Ebao Town. In terms of grassland types, except for the 16.0% and 9.7% of desert grassland and alpine and sub-alpine meadow, respectively, improvement, the other grassland types are degraded. The area of continuous degradation is greater than that from improvement to degradation, which means that most grasslands of Qilian County will continue to degrade, including the previously degraded areas.

Figure 7
figure 7

Spatial distribution of the forecasted grassland dynamic. The map was generated by ArcGIS 10.2, URL: https://www.esri.com.

Full size image
Figure 8
figure 8

Dynamic changes in grassland types.

Full size image

Discussion

Methodology

The spatial heterogeneity linked to vegetation and soil patches is a valid indicator for assessing degradation in dry environments45. Similarly, mixtures of soil and vegetation patches can be found in degraded grasslands on the Qinghai–Tibet Plateau46,47. The formation of “black soil patches” during degradation results in the spatially discontinuity of grassland cover, resulting in changes in spatial heterogeneity. The change in grassland cover and spatial heterogeneity was the early warning signal of grassland degradation in the Qinghai–Tibet Plateau34. These two indicators can be utilized to determine the degree of grassland degradation. In this study, considering that the vegetation cover in some degraded areas may be higher than that in non-degraded areas such as the higher vegetation cover of invasive species than that of native species48, we calculated the coefficient of variation in NDVI within 90 m × 90 m areas consisting of nine 30 m × 30 m pixels defined spatial heterogeneity, which is used to quantify the soil patches in degraded grasslands. In combination with NDVI, the degradation level of Qilian County in 2019 was divided. According to our findings, approximately 55.0% of the grassland in the study area has been degraded, lower area than the 69.05% estimated by Lu et al. for Qilian County49. The differences might have occurred because this study used remote sensing data to identify grassland degradation areas from spatial, while the grassland degradation area studied by Lu et al.49 was obtained based on statistical data, and over a different research period. In addition, NDVI and NPP showed increasing trends in the Qinghai–Tibet Plateau and most regions of the Qilian Mountains50,51,52, coupled with more volatile precipitation changes, resulting in increasing spatial heterogeneity39. This is consistent with the changing trend of NDVI and spatial heterogeneity in Qilian County.

According to the findings of this study, grassland restoration or degradation in Qilian County between 2000 and 2019 was mostly jointly caused by climate change and human activities. The RESTREND method simplifies an otherwise complicated mechanism of grassland degradation or restoration by spatially identifying and quantifying the contribution degree, affected regions, and locations affected by climate change and human activities. It has been frequently used to separate and assess the relative roles of climate change and human activities in grassland degradation on a regional scale25,26. Further, this method was also applicable to arid and semiarid regions35,54. For example, Li et al. and Jiang et al. used it to differentiate vegetation changes caused by climate and human factors in the Inner Mongolia Xilingol grassland region and Central Asia, respectively8,53. Wu et al. examined both the magnitude and direction of climate and human activities influences on NDVI variation over the alpine meadows, steppes, and desert-steppes of the Qinghai–Tibet Plateau54. Therefore, we selected the RESTREND method to evaluate the relative contribution of climate change and human activities in grassland degradation in Qilian County. However, when fitting the multiple regression equation between climate variation and NDVI, most studies employed only precipitation and temperature as climate parameters, which may affect the results24. Therefore, to select reasonable climate elements and improve the fitting degree of the regression equation, we used stepwise regression and random forest regression model to select the dominant climate elements that have the most significant impact on NDVI variations, and then establish a regression equation with NDVI to predict NDVI. The results showed that when establishing the regression equation between climate elements and NDVI, the accuracy of selecting precipitation, temperature, and sunshine hours was higher than that of selecting only two climate elements of precipitation and temperature, as R2 increased by 17% (Table S2). Furthermore, we designed twelve scenarios to distinguish the driving forces of climatic variation and human activities on different degraded grasslands, which may accurately identify the degree of grassland degradation induced by climate change and human activities in a region.

Impacts of climate change on grassland degradation or restoration

Climate change are changing grassland ecosystems4. Temperature and precipitation are the two most important climatic factors influencing the presence and distribution of grassland ecosystems55. Therefore, climate change affects the changes in precipitation and temperature by changing and affecting soil moisture, soil microorganisms, photosynthesis, and plant respiration, thereby further controlling grassland growth and ecosystem productivity and affecting degradation or restoration3. Climate change was an important driving force affecting grassland degradation in this study. The area of climate-driven grassland degradation was larger than that of human activities-driven, particularly in the northwest of Qilian County. Similar to a previous study56, highlighting climate change as the principal driver of nearly half of the global grassland degradation.

In the 20 years study period, the temperature in Qilian County significantly increases (slope = 0.03; P < 0.05), while precipitation shows an insignificant increasing trend (Fig. 9). Water is generally regarded as the most critical ingredient influencing plant growth, hence water shortage may have a significant impact on vegetation production, especially in arid and semi-arid regions3,57,58. Temperature is also an indirect indicator of the available energy for vegetation growth, thus a rise in temperature could affect the greenness of the vegetation and severely limit the growth of shrubs and sparse vegetation59. Ji et al. discovered that change in forest NPP in China were mostly induced by changes in precipitation, followed by temperature60. Considering the geographical differences of climate change, we further computed the correlation coefficients between NDVI and temperature and precipitation on a pixel scale to study how climate change affects grassland ecosystems (Fig. 10). NDVI is positively correlated with temperature and precipitation in most areas. Specifically, the areas where NDVI show a positive correlation with precipitation are concentrated in the northwest of Qilian County (Fig. 10a). The precipitation in these areas is relatively scarce, and the annual precipitation is less than 400 mm (Fig. 11a). Therefore, moisture is very important to grass growth in the area, even when NDVI is negatively correlated with temperature (Fig. 10b). From the spatial variation in temperature, the rate of temperature rise is greater than 0.02 °C/a (Fig. 11d). The rise of temperature usually increases the evaporation of surface water, which can have adverse effects on grass growth8, affect grassland coverage, and lead to degradation, which also explains to some extent why grassland degradation regions are primarily scattered in the northwest. Our findings are consistent with those of Zhou et al.2, who found that a wetter environment is conducive to vegetation development, yet in areas with less precipitation, warming can cause drought, which has a negative impact on plant growth61. Wu et al. and Duan and Xiao found that warming was detected over more than three-quarters of the land surface of the Qinghai–Tibet Plateau, with warming as the most common driving factor for alpine grassland dynamics54,62.

Figure 9
figure 9

Variation in precipitation (a) and temperature (b) in the grasslands of Qilian County during the 2000–2019 period.

Full size image
Figure 10
figure 10

Correlation coefficients between NDVI and precipitation (a) and temperature (b). The map was generated by ArcGIS 10.2, URL: https://www.esri.com.

Full size image
Figure 11
figure 11

Spatial distribution (a,b) and variation (slopes) (c,d) in precipitation and temperature. The map was generated by ArcGIS 10.2, URL: https://www.esri.com.

Full size image

Due to the decrease in groundwater evaporation and enhancement of soil water retention, the slightly lower temperature (Fig. 11b) and increasing precipitation (Fig. 11a) have improved the grassland growth in the southeast of Qilian County63,64. Therefore, the regions of vegetation with improving conditions were mainly concentrated in these areas, such as Arou Township, Ebao Town, and Mole Town. This is because the soil water content is the key element affecting net primary productivity9. Increased precipitation (Fig. 11c) could increase soil water content and benefit the growth of grassland. On the other hand, from the spatial distribution of grassland degradation (Fig. 3a), some areas in the southeast of Qilian County have undergone grassland degradation to varying degrees, such as Babao Town and Mole Town. The possible reason is that the increase in precipitation exceeds the demand of vegetation growth, and it decreased solar radiation and increased relative humidity, which inhibited grassland photosynthesis65.

The different responses of different degraded or restored grasslands to climate factors are also attributed to changes in the water and heat cycles50,66. In the process of grassland degradation, vegetation coverage decreased, the influence of vegetation on water evaporation weakened, the rainfall infiltration rate decreased and bare grassland increased50. Bare soil enables a large amount of water and heat to enter the atmosphere, resulting in the instability of a grassland ecosystem, which in turn leads to the aggravation of degradation and changes in air exchanges between water and heat66. In addition, due to the differences of vegetation community structure and soil properties in different degradation stages, changes in vegetation productivity, soil moisture and the nitrogen cycle exhibit different responses to climate factors67,68,69. Xu et al. found that surface transpiration and infiltration may cause the decrease of soil moisture on the surface of grassland in the slight and moderate degradation stage. Low vegetation coverage, low plant activity level and reduced transpiration may cause the decrease in soil moisture in severe and extreme degradation stages70.

Furthermore, grassland degradation accounts for the largest proportion of degraded area at elevations of 3500–4500 m between 2000 and 2019 (Fig. S2), reducing the NDVI while the spatial heterogeneity increases, which is consistent with a prior study48. Above 4500 m altitude, 81.08% of the grassland has improved, indicating an increase in vegetation covering and a decrease in spatial heterogeneity. A recent study also discovered higher vegetation cover in the Himalayas at high altitudes71.

Anthropogenic impacts on grassland degradation or restoration

Human activities are the principal drivers of grassland degradation or restoration12. This study found that the contribution of climate-driven grassland restoration was smaller than that of human activities-driven. According to Zhou et al.2, human activities-caused grassland restoration accounts for 78.1% of the total grassland area in China, including Qinghai, southeast of Tibet, Tianshan Mountains. On the one hand, this is mainly due to the control of grazing capacity and the implementation of intervention measures that have a positive impact on the grasslands in Qilian County. Grazing is the main factor influencing grassland change72. Moderate grazing can help to maintain the biodiversity of the grassland ecosystems73. According to Fig. S3, the overload rate of Qilian County shows a decreasing trend between 1982 and 2017, implying that overgrazing has declined. Generally, reducing grazing may have a positive impact on the soil environment, such as reducing the compaction of the surface soil, thus promoting restoration74. Duan et al. also highlighted that overgrazing in the Qilian Mountains has been greatly alleviated75. In addition, grazing prohibition and exclusion, forage-livestock balance, and other policies on grasslands can avoid degradation and mitigate the damage. These may also explain why climate change was the primary driver of grassland degradation12. Wang et al. found that 28.6% of grassland restoration was attributed to grazing prohibition in the Qinghai–Tibet Plateau4. On the other hand, it may also be related to various ecological protection measures implemented by the government, such as protecting natural forests76, converting farmland to forests or grasslands77, and establishing the Qilian Mountain National Park Nature Reserve, which have had a positive influence on the vegetation75. Furthermore, considering the moderately degraded areas around Babao Town and Arou Township with point-distribution, this may be related to the overgrazing surrounding residential plots, resulting in a loss in vegetation cover78. Reduced vegetation cover fosters the invasion of pikas (Ochotona curzoniae), resulting in the formation of soil patches and grassland degradation79. Liu et al. found that overgrazing surrounding the resident plots could be the reason for the point-distribution characteristics of grassland degradation in the Qinghai–Tibet Plateau36.

From the dynamic change trend, 16.0% and 9.7% of desert grassland and alpine and sub-alpine meadow, respectively, would have been improved, while the other grassland types would have undergone continuous degradation or changed from improvement to degradation, indicating that the desert grassland and alpine and sub-alpine meadow with decreasing trends in human impacts, and climate change promoted grassland improvement. This shift in grassland types may be caused by differences in plant species combinations and functional properties54, while the increased precipitation may improve the vegetation coverage in desert and grassland areas80.

In summary, the grassland restoration or degradation were jointly affected by climate change and human activities in Qilian County between 2000 and 2019. Different types of degraded grassland should implement different restoration strategies. Specifically, grassland degradation caused by climate change is difficult to control, because it is the normal response of grassland to environmental change7. The climate-driven grassland with slight degradation or re-growing (CDSR) regions were observed around Yanglong Township, the northwest of Yeniugou Township, and the north of Arou Town, owing to increasing warmth and decreased radiation. On the one hand, native grasses can be genetically modified to adapt to the changing climates, and aridity- and cold-resistant grasses should be introduced in these areas to mitigate the impact of unfavorable climate changes on grasslands6,81. In addition, irrigation is commonly employed to fulfill the water needs of plants and to ensure grass production82. On the other hand, the government should also establish a series of appropriate development models of tree-shrub-grasses configuration to further minimize grassland degradation induced by climate change10. In severely degraded regions, grassland biodiversity should be preserved through alternative measures such as rotational grazing and seasonal fencing81.

Uncertainties and limitations

The residual trend analysis (RESTREND) has some drawbacks and limitations in assessing the impact of climate change and human activities on grassland dynamics. In the current study, the stepwise regression analysis and random forest regression models were used to select the dominant climate factors significantly affecting the NDVI to fit the regression equation with climate variables. However, this study only considered the impact of precipitation, temperature, sunshine hours, solar radiation, and wind speed on the NDVI, and did not consider other climate factors such as drought and extreme climate. However, this study could provide a reference method to reasonably select climate factors. In future research, a selection of the climate factors significantly related to NDVI in the study area as the climate variables for predicting NDVI is warranted and compare the applicability of different methods. And the cumulative and lagging effects of climate factors on NDVI will also be focused on in our future research. In addition, the variation in grassland was affected by grassland species, altitude, soil properties, grassland rodents, diseases and pests, and other factors in addition to climate factors and human activities81, which were not considered in this study because the quantitative methods for studying these factors at the macro scale are not mature.

In this study, we designed an evaluation method based on NDVI and CVNDVI, which distinguished the influence of climate change and human activities on different degraded grasslands, compensates for the shortcomings of previous studies, and is more convenient to evaluate the degree of grassland degradation caused by climate change and human activities in spatial terms. Nonetheless, this method also has some drawbacks. Although this study identified the driving factors of grassland degradation based on the relative roles of climate change and human activities on vegetation cover and spatial heterogeneity (ecosystem stability), the CVNDVI calculation remains with great uncertainty. However, we provide a useful reference for distinguishing the impacts of climate change and human activities on different degraded grasslands. Moreover, this study aimed to assess the effects and relative roles of climate change and human activities on grassland degradation or restoration in Qilian County on a macro scale. Climate change and human activities need further study at a finer scale in combination with ground truth data.

Conclusion

In the current study, NDVI and CVNDVI were selected as indicators of grassland degradation and used the improve the commonly used residual trend analysis method to evaluate the driving factors of different degraded grasslands between 2000 and 2019; further, the Hurst index was used to analyze the forecasted change in degradation or restoration in Qilian County. The results showed that nearly 55.0% of the grasslands between 2000 and 2019 are degraded, mainly affected by slight degradation. Grassland restoration or degradation was jointly induced by climate change and human activities. The regions dominated by both factors presenting grassland improvement, slight degradation or re-growing, moderate degradation, and severe degradation or desertification accounted for 16.6%, 34.3%, 3.3%, and 1.3%, respectively, of the total grassland area. In addition, the area of climate-driven grassland improvement was less than that of human activities-driven (4.2% < 12.2%). Secondly, regardless of the degree of degradation, the grassland degradation area caused by climate change was larger than that caused by human activities, such as slight degradation or re-growing (12.3% vs. 9.5%), moderate degradation (2.3% vs. 1.5%), and severe degradation or desertification (1.2% vs. 1.1%).

Nearly 95.69% of the grassland in Qilian County will continue to degrade, including the previously degraded areas. To overcome some of the limitations of this study, future research should improve the evaluation methods, combined with ground truth data, identify the driving factors of different degraded grasslands on a finer scale, and predict the trend of grassland degradation or restoration. Our findings will not only help us understand the drivers of different degraded grasslands but also help decision-makers formulate grassland ecological management policies and guidelines.