Impacts of Hydrothermal Factors on the Spatiotemporal Dynamics of Alpine Grassland Aboveground Biomass During the Pre-, Mid-, and Post-COVID-19 Pandemic Periods

Abstract

Aboveground biomass (AGB) is a key parameter for studying the carbon cycle, evaluating grassland growth, and assessing the grass–livestock balance. In this study, we established an optimal inversion model for alpine grassland AGB and estimated the growing-season (July–September) AGB from 2018 to 2022 based on field survey data and remote sensing data. We aimed to analyze the spatiotemporal dynamics of AGB in alpine grasslands and its response mechanisms to hydrothermal factors, as well as to explore the indirect impacts of changes in human activities during the COVID-19 pandemic on the grassland ecosystem. The results showed the following: (1) Alpine grassland AGB was high in the southwest and low in the northeast of the studied area, initially increasing and then decreasing over time. This pattern was largely consistent with the spatial distribution and interannual variations in precipitation and temperature, with a significant positive correlation being observed between precipitation and AGB, indicating that hydrothermal factors are key drivers of grassland AGB dynamics. (2) The grasslands demonstrated a trend of slight decrease in AGB overall, with some local areas showing a slight increase. Compared with before 2018, grasslands showed a gradual recovery trend, which may be related to grazing policies and conservation management measures. (3) An increase in grazing intensity in local areas decreased grassland AGB and vice versa, indicating that the restrictive measures led to changes in grazing intensity, which indirectly affected grassland AGB during the pandemic. This study reveals the general patterns of hydrothermal factors’ influence on alpine grassland AGB dynamics during the pre-, mid-, and post-COVID-19-pandemic periods, providing a scientific basis for formulating sustainable grassland management strategies.

1. Introduction

As one of the largest terrestrial ecosystems, grassland occupies nearly 40% of the Earth’s land area [1,2]. It plays a key role in the global carbon cycle, maintains biodiversity, and sustains livestock grazing [3,4,5]. Due to climate change, overgrazing, and forest fires, grassland ecosystems have been degraded to varying degrees, which has reduced their productivity and carbon sequestration capacity [6,7,8]. The implementation of prevention and control measures during the COVID-19 pandemic might have affected grassland vegetation productivity and recovery. This impact is unclear and must be analyzed case by case [9,10]. Alpine grasslands in southwest China have significant ecological and economic value. However, most grasslands in this region are scattered in mountainous areas, making it challenging to monitor their resources [11]. Additionally, climate change has altered hydrothermal conditions, and the COVID-19 pandemic reduced human activities, which may have affected the growth and recovery of alpine grassland vegetation [12,13]. Thus, analyzing the impacts of hydrothermal factors on the temporal dynamics of alpine grassland AGB during the pre-, mid-, and post-COVID-19 pandemic periods is necessary to evaluate grassland productivity, growth status, and ecological benefits.

The prediction of grassland AGB using machine learning algorithms, combined with field survey data and remote sensing data, has been frequently applied in previous research [14,15]. Machine learning algorithms are efficient because they automatically select the optimum variables for AGB estimation. They are suitable for analyzing complex data and provide accurate results [16]. Liu et al. [17] found that among the models evaluated for estimating AGB in the mountain grasslands of southwest China (Yunnan and Guizhou Provinces), random forest (RF) achieved the highest accuracy (R² = 0.75), outperforming multiple stepwise regression (MSR), support vector machine (SVM), and convolutional neural network (CNN). There is a lack of research on using remote sensing technology to estimate AGB in the alpine grassland of southwest China. Moreover, substantial differences in training sample requirements and hyperparameter settings exist among various machine learning algorithms. Thus, evaluating the accuracy, performance, and applicability of different algorithms in the study of the remote sensing inversion of alpine grassland AGB is crucial.

The long-term and continuous monitoring of grassland ecosystems is crucial for assessing grassland health and growth [18]. Zeng et al. [19] observed that the grassland AGB in the Three-River-Source (TRS) National Park from 2000 to 2018 exhibited a spatial pattern of higher values in the southeast and lower values in the northwest, and the interannual variation showed a non-significant upward trend in most areas. Understanding the dynamic trends and future development of grassland AGB is required to formulate sustainable development strategies and reasonable grazing policies for alpine grassland. Therefore, it is necessary to perform the continuous and systematic AGB monitoring of this ecosystem.

Increased warming trends and wet conditions directly affect the growth and biomass accumulation of grassland vegetation. A significant positive correlation exists between aboveground net primary productivity (NPP) and precipitation at the global scale (R² = 0.84, p < 0.05), and fluctuations in precipitation could change the growth of grassland vegetation [20]. Additionally, He et al. [21] observed that the rate of increase in grassland AGB was higher under conditions of 100–120 mm of precipitation and 9–12 °C. Due to the complex environmental conditions and fragile grassland ecosystems, changes in hydrothermal factors significantly affect the biomass of alpine grasslands in southwest China. Studying how grassland AGB responds to temperature and precipitation under climate warming and increased humidity is crucial to revealing its response mechanisms to hydrothermal conditions.

The COVID-19 pandemic had profound socio-economic, human health, and environmental impacts on many areas [22,23,24]. Grassland is a natural resource with important ecological, economic, and social value, where its economic value is mainly reflected in providing grazing land and livelihood sources for herders, while its social value is reflected in satisfying the leisure and recreation needs of tourists [25]. However, the government implemented restrictive measures that significantly reduced human activities during the COVID-19 pandemic [26]. For example, the visitor numbers at Pudacuo National Park, located in the southwest of Yunnan Province, China, showed a fluctuating trend from 2018 to 2022: about 1.11 million and 1.37 million visitors in 2018 and 2019, dropping to 0.5 million and 0.7 million in 2020 and 2021 due to the pandemic, and rebounding to about 1.25 million in 2022 [27]. This decrease in activity lessened the disruption of grassland vegetation caused by trampling, indirectly promoting its growth and increasing productivity [28,29]. Studies on alpine grasslands have focused on analyzing the trend and spatial distribution of AGB, whereas limited research has been performed on the impact of the pandemic on grasslands [17,30]. Therefore, it is worth examining the indirect impact of the COVID-19 pandemic on the AGB of alpine grasslands to provide insights for grassland management in this region.

Research indicated that the sustained increase in precipitation in north China prior to 2019 significantly contributed to the accumulation of grassland AGB in the region [12]. However, the increase in temperature caused by global warming will accelerate soil moisture loss, thereby inhibiting the accumulation of grassland AGB [31]. In this study, we aimed to comprehensively employ multiple machine learning algorithms to accurately estimate the alpine grassland AGB during the growing seasons (July to September) from 2018 to 2022, systematically determine its spatiotemporal dynamics, and predict future trends. The most important purpose of this study was to quantitatively assess the mechanistic impacts of hydrothermal factors on alpine grassland AGB during the pre-, mid-, and post-COVID-19 pandemic periods and explore the indirect influence of the disruption in human activities caused by the pandemic on grassland AGB. The findings will provide a scientific basis for the management and sustainable development of alpine grassland ecosystems. Our hypothesis is based on the consideration that changes in grazing during the pandemic may have had an impact on local grassland AGB and that hydrothermal conditions are dominant factors influencing grassland AGB on the whole, because they directly affect plant growth and nutrient uptake.

2. Materials and Methods

2.1. Study Area

Shangri-La City (99°22′ E to 100°19′ E, 26°52′ N to 28°52′ N) in Yunnan Province, southwest China, was selected as the study area. It has a grassland area of 3265.7 km², dominated by natural alpine grasslands with high biodiversity. The elevation difference is 3872 m above mean sea level. The southwest Indian monsoon affects the local climate, resulting in the region having a dry season (June to October) and a wet season (November to May). The climatic conditions in the study area include an average yearly precipitation of 620 mm, an annual mean temperature of 6 °C, and an average relative humidity of 70%. The area also receives roughly 2186.6 h of sunshine and experiences a frost-free period of approximately 124 days annually [32]. According to our field investigation, the dominant species in the study area include Blysmus sinocompressus Tang & F.T.Wang, Carex nubigena D.Don ex Tilloch & Taylor, Carex parvula O.Yano, Juncus allioides Franch, and Argentina lineata (Trevir.) Soják (see Figure 1). Animal husbandry and tourism are the dominant industries.

2.2. Data Collection

2.2.1. Field Survey Data

We conducted systematic field investigations on grassland resources in Shangri-La during the growing seasons of 2021 and 2022. In the wet season, grassland AGB undergoes growth in an allometric and nonlinear manner [33], while in the dry season, its accumulation is at its peak [34]; therefore, the relevant data were collected in the dry season to better reflect interannual AGB differences under different influencing factors. We used a random sampling method to select typical, undisturbed or minimally disturbed areas and established 112 sample plots, where each plot was more than 1000 m away from the others [35] (see Table S1 of the Supplementary Materials) and had an area of 10 × 10 m. Five 1 × 1 m quadrats were established in the middle and around each plot, and all vegetation in each was harvested. The fresh weight was measured on site before bringing all the samples back to the laboratory, where they were oven-dried at 85 °C until constant weight. The dry weight of the vegetation in each quadrat was determined, and the mean value of the five quadrats was used as the AGB for the 10 × 10 m grassland plot [36]. A Trimble GeoXH 3000 handheld GPS device (China) was used to record the geographic coordinates and elevation of each plot [17].

2.2.2. Remote Sensing Data

The remote sensing data were obtained from the Sentinel-2A satellite imagery “https://sentiwiki.copernicus.eu/web/sentinel-2 (accessed on 15 October 2024)” released by the European Space Agency (ESA) on 23 June 2015. For this study, the satellite imagery data covering the whole study area from 2018 to 2022 during the growing season were downloaded on the Google Earth Engine (GEE) cloud platform. Post-processing included cloud removal, cropping, and mosaicking. The vegetation greenness value is higher during the growing season of grassland, so remote sensing imagery makes it easier to identify and extract features [37]. We obtained 9 vegetation indices and 10 texture features by using band calculation and the gray-level co-occurrence matrix (GLCM) (see Appendix A) [38].

The terrain data were derived from the Advanced Land Observing Satellite (ALOS) digital elevation model (DEM) published by the Japan Aerospace Exploration Agency (JAXA) “https://search.asf.alaska.edu (accessed on 17 October 2024)”. The DEM had a spatial resolution of 12.5 m and was analyzed by using the Spatial Analyst tool in ArcGIS to obtain slope and aspect data.

Precipitation data were derived from the Climate Hazards Group Infrared Precipitation with Stations (CHIRPS) dataset created by the USGS Earth Resources Observation and Science (EROS) Center “https://www.chc.ucsb.edu/data/chirps (accessed on 17 October 2024)”. We used the GEE cloud platform to extract the total precipitation in grasslands in the growing seasons from 2018 to 2022 [39]. The temperature data were downloaded from the National Earth System Science Data Center of China “https://www.geodata.cn (accessed on 17 October 2024)”. We used monthly average temperature data in the growing season derived by using bilinear interpolation in ArcGIS [40].

2.2.3. Other Data

Other data included the number of grazing animals, and tourists from 2018 to 2022. The data were obtained from the statistical bulletin on national economic and social development issued by the Yunnan Provincial Bureau of Statistics “http://stats.yn.gov.cn/ (accessed on 21 October 2024)”.

2.3. Data Analysis

2.3.1. Biomass Estimation

In this study, we employed four advanced machine learning algorithms, including RF, support vector machine (SVM), artificial neural network (ANN), and gaussian process regression (GPR), to develop predictive models for grassland AGB [41,42,43,44]. Recursive feature elimination was used to screen the optimal variables. The interactions among these variables were considered in the regression model to evaluate their combined effect on grassland AGB [45]. We employed 10-fold cross-validation to evaluate the effects of different training–validation set ratios and to select the optimal ratio for model building.

The coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE) between observed and predicted grassland AGB values were used to evaluate the predictive accuracy of different dataset ratios and models. The model with the highest accuracy was selected to estimate the growing-season AGB of alpine grasslands from 2018 to 2022. The metrics were calculated by using Equations (1)–(3) [46]. We selected the period from 2018 to 2022 because previous studies had focused on the trend in grassland AGB dynamics in the study area from 2008 to 2017 [30]. This allowed us not only to analyze the trends in grassland AGB dynamics during this period but also to compare them with those in the previous period, determining whether grassland conditions had improved or degraded over time. The peak period of the COVID-19 pandemic was from 2020 to 2021. We focused on the pre-pandemic (2018 and 2019), mid-pandemic (2020 and 2021), and post-pandemic (2022) periods to compare trends and assess the potential impact of the pandemic on grassland ecosystems. Although the survey data were concentrated in 2021 and 2022 and pre-pandemic field measurements for direct comparison are lacking, we could verify the accuracy between the predicted values for 2021 and 2022 inverted by the optimal machine learning model and the measured values, so as to evaluate whether the model also had a high accuracy in predicting grassland AGB dynamics from 2018 to 2020 [47,48]. The variations in grassland AGB pre- and post-pandemic were analyzed based on the models’ predictions:

$$R^2=1-{\displaystyle \frac{\sum _{i=1}^n{(y_i-x_i)}^2}{\sum _{i=1}^n{(y_i-\overline{y})}^2}}$$

(1)

$$\mathit{RMSE}=\sqrt{{\displaystyle \frac{\sum _{i=1}^n{(y_i-x_i)}^2}{n}}}$$

(2)

$$\mathit{MAE}={\displaystyle \frac{\sum _{i=1}^n\left|y_i-x_i\right|}{n}}$$

(3)

where x_i, y_i and $\overline{y}$ are the predicted, measured, and average measured AGB, respectively. n is the number of sample plots. R² is used to evaluate the accuracy of a model’s fit to the data, with a value range of 0 to 1. The closer R² is to 1, the stronger the model’s explanatory power is, and the better the fit is. The RMSE is used to quantify the deviation between predicted and actual values, with a range of (0, +∞). A smaller RMSE indicates that the predicted values are closer to the actual values, reflecting higher predictive accuracy. These two metrics, R² and RMSE, are complementary: an ideal predictive model should simultaneously achieve an R² close to 1 (indicating strong explanatory power) and an RMSE close to 0 (indicating minimal prediction error). The MAE represents the average of the absolute error. The smaller it is, the higher the model accuracy and stability are.

2.3.2. Biomass Trend

We estimated the spatiotemporal dynamics of alpine grassland AGB during the growing season by using Theil–Sen median trend analysis and the Mann–Kendall nonparametric test. Since our study focused on the growing season, the trends we observed reflect the changes in grassland AGB dynamics during this critical period, minimizing the influence of seasonal variability outside this timeframe. The Theil–Sen median method is a nonparametric trend estimation technique known for its computational efficiency and outlier resistance [49,50]. The trend was calculated by using Equation (4). The Mann–Kendall test is a nonparametric statistical method used to assess trends in time-series data. It does not assume a normal distribution and is robust against missing values and outliers [51]. The Mann–Kendall test was performed by using Equations (5)–(7). These methods are widely used to assess continuous trends in grassland AGB dynamics, especially when dealing with complex and noisy data. We used the Hurst index, which measures the continuity and stability of a time series [52], to explore the future development trend of grassland AGB dynamics in our study area:

$$\beta =\mathit{median}\left({\displaystyle \frac{x_j- x_i}{j-i}}\right)$$

(4)

$$Z=\left\{\begin{array}{ll}\hfill {\displaystyle \frac{S-1}{\sqrt{V}}},& \mathit{if}S>0\\ \hfill 0,& \mathit{if}S=0\\ \hfill {\displaystyle \frac{S+1}{\sqrt{V}}},& \mathit{if}S<0\end{array}\right.$$

(5)

$$S={\sum }_{i=1}{\sum }_{j=i+1}\mathit{sign}(x_j-x_i)$$

(6)

$$V={\displaystyle \frac{n(n-1)(2n+5)}{18}}$$

(7)

In this context, x_j and x_i denote the AGB values for years j and i, respectively. The variable n is the length of the time series, sign refers to a symbolic function, and Z is a statistical value ranging within (−∞, +∞). AGB increases when β is positive and decreases when β is negative, and it is stable when β is greater than 0. At a given level of significance, $\left|Z\right| >{ u}_{1-\alpha ∕2}$ indicates a significant change. In contrast, $\left|Z\right| \le { u}_{1-\alpha ∕2}$ demonstrates a non-significant change. The significance level, ɑ, was 0.05, and $u_{1-\alpha ∕2}$ was ±1.96.

2.3.3. Response to Hydrothermal Factors

The correlations between grassland AGB and both total precipitation and monthly mean temperature during the growing season were calculated to evaluate the impact of hydrothermal factors on the spatial distribution of grassland AGB during the pre-, mid-, and post-COVID-19 pandemic periods. We used the raster calculator in ArcGIS, and the correlation was calculated with Equation (8). We analyzed the interannual trends in average grassland AGB, total precipitation, and monthly mean temperature during the growing season to demonstrate the direct influence of hydrothermal factors on grassland AGB:

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

(8)

where r_xy refers to the correlation between hydrothermal factors and grassland AGB and its values range from −1 to 1; x_i and y_i denote the value of grassland AGB, of temperature and precipitation in years i and j, respectively; and $\overline{x}$ and $\overline{y}$ are the values of average grassland AGB, and of temperature and precipitation, respectively.

2.3.4. The Indirect Impact of the COVID-19 Pandemic

Restrictive measures mid-pandemic resulted in changes in grazing and tourism, affecting alpine grassland AGB indirectly. We used the grazing intensity and tourism density, which were calculated by using Equations (9) and (10), to analyze the indirect effects of the pandemic on alpine grassland AGB:

GI = Ng/Tg

(9)

TD = Nt/Ta

(10)

where GI and TD are the grazing intensity and tourism density. Ng and Nt are the number of grazing animals and the number of tourists, respectively. Tg and Ta are the total area of grassland and the total land area, respectively.

3. Results

3.1. Biomass Estimation

3.1.1. Optimization of Machine-Learning Models

We analyzed models with 27 variables to determine the optimal variables to establish the AGB inversion model. The RMSE was the lowest for the model with six variables, indicating that it had the best prediction performance (see Figure 2a). We ranked the 27 variables by importance and found that the top six were elevation, the atmospherically resistant vegetation index (ARVI), the ratio vegetation index (RVI), the normalized difference vegetation index (NDVI), the infrared percentage vegetation index (IPVI), and the green normalized difference vegetation index (GNDVI) (see Figure 2b).

We assessed the effects of the interactions among optimal variables on grassland AGB. The interactions between elevation and the ARVI, the IPVI, the NDVI, and the RVI were significant. The values of these vegetation indices and AGB were significantly higher at lower altitudes. However, the effects of the NDVI and the IPVI on AGB were synergistic, with AGB being significantly higher when both indices were high (see Figure 3).

Before building the model, we compared the effects of different ratios of the training and validation sets. The highest accuracy was achieved at 70% of the training set and 30% of the validation set (see

Table 1. Model accuracy for different ratios of the training and validation sets.

Training Set (%)	Validation Set (%)	Accuracy Evaluation Index
		R2	RMSE	MAE
60	40	0.61	119.90	70.44
65	35	0.33	184.32	98.94
70	30	0.70	84.13	61.94
75	25	0.42	172.12	95.38
80	20	0.66	94.38	71.81
85	15	0.31	117.10	93.22
90	10	0.32	188.74	120.54

Notes: coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE).

).

An analysis comparing the efficiency of the four machine learning algorithms in simulating grassland AGB from 2018 to 2022 (see

Table 2. R², RMSE, and MAE for different machine learning algorithms.

Algorithm	Accuracy Evaluation Index			Optimal Machine Learning Model
	R2	RMSE	MAE
RF	0.78	65.47	41.38	119.4 + 221.9 × GNDVI + 192.8 × ARVI + 428.2 × IPVI + 214.1 × NDVI + 23.04 × RVI − 0.05145 × Elevation
SVM	0.85	55.70	38.69	152.6 + 221.9 × GNDVI + 181.4 × ARVI + 411.1 × IPVI + 205.6 × NDVI + 20.74 × RVI − 0.05766 × Elevation
ANN	0.82	59.84	42.55	57.13 + 218.4 × GNDVI + 204.3 × ARVI + 443.5 × IPVI + 221.7 × NDVI + 25.62 × RVI − 0.04264 × Elevation
GPR	0.83	59.36	43.95	215 + 250.2 × GNDVI + 178.7 × ARVI + 424.6 × IPVI + 212.3 × NDVI + 18.16 × RVI − 0.07923 × Elevation

Notes: random forest (RF), support vector machine (SVM), artificial neural network (ANN), and gaussian process regression (GPR), coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE).

) showed the following ranking based on overall prediction accuracy: SVM > GPR > ANN > RF, where the SVM algorithm had the highest inversion accuracy (R² = 0.85, RMSE = 55.7). The results showed that using field survey data from 2021 and 2022 to invert grassland AGB from 2018 to 2020 resulted in high model accuracy (see Figure 4).

3.1.2. Spatial Distribution and Interannual Dynamics

The spatial distribution of grassland AGB during the growing season in Shangri-La City from 2018 to 2022 based on the SVM algorithm is shown in Figure 5. The AGB of some grasslands in the northeast of the study area was lower and higher in the southwest in 2020 than in 2019. Some areas in the northeast had higher AGB, whereas some areas in the southwest had lower values in 2021 than in 2020. The grassland AGB exhibited a decreasing trend from the southwest to the northeast on the whole, exhibiting substantial spatial heterogeneity.

The average AGB values in the growing seasons from 2018 to 2022 was 487.33 g·m⁻², 552.42 g·m⁻², 535.13 g·m⁻², 526.24 g·m⁻², and 405.76 g·m⁻², respectively; therefore, the average AGB was the highest in 2019, with 44.25% and 16.98% of the grasslands being in high- and very high-biomass areas, respectively, and only 8.6% of the grasslands being in low-biomass areas (see

Table 3. Aboveground biomass (AGB) of grassland in different years.

Year	Low AGB Area (%)	Medium AGB Area (%)	High AGB Area (%)	Very High AGB Area (%)	Minimum Value (g·m−2)	Maximum Value (g·m−2)	Average Value (g·m−2)	Sum (t)
2018	22.29	27.43	36.15	14.13	56.91	1036.44	487.33 ± 225.91 c	1.64 × 107
2019	8.60	30.17	44.25	16.98	74.55	1099.58	552.42 ± 198.89 b	1.86 × 107
2020	25.71	17.92	27.59	28.78	40.20	1140.43	535.13 ± 274.34 bd	1.80 × 107
2021	11.95	33.47	38.44	16.14	68.98	1054.08	526.24 ± 204.88 ab	1.77 × 107
2022	19.59	55.88	20.27	4.26	94.60	913.50	405.76 ± 120.07 d	1.37 × 107

Notes: The different letters in the table represent significant differences at the 0.05 level.

). Grassland AGB showed a trend that increased initially and then decreased over time. According to analysis of variance (ANOVA), we found that the grassland AGB mid- and post-pandemic showed a downward trend compared with the pre-pandemic period. The mid-pandemic value increased significantly compared with 2018, while the post-pandemic value decreased significantly compared with the pre-pandemic period.

3.2. Biomass Trend

As shown in Figure 6a, the grasslands demonstrated a trend of slight decrease in AGB overall, with some local areas showing a slight increase. Grasslands with a slight decline in AGB were dominant, covering 2182.84 km² and making up 66.84% of the total grassland, and were situated in the northeastern part of the research area. Grasslands with slight increases in AGB covered 800.44 km², accounting for 24.51% of the total area, and were primarily concentrated in the central and southwestern zones (see

Table 4. The area and proportion in different categories of aboveground biomass (AGB) trends.

Category	β	Z	Area (km2)	Percentage (%)
Significant increase	$\mathsf{\beta }>0$	$Z>1.96$	38.21	1.17
Slight increase	$\mathsf{\beta }>0$	$Z\le 1.96$	800.44	24.51
Stable	$\mathsf{\beta }=0$	Z	66.30	2.03
Sligh decrease	$\mathsf{\beta }<0$	$Z\le 1.96$	2182.84	66.84
Significant decrease	$\mathsf{\beta }<0$	$Z>1.96$	177.98	5.45

).

As shown in Figure 6b, the future trend is dominated by an anti-continuous decrease in AGB, with an area of 1691.02 km², accounting for 51.78%, while areas with a continuous increase have a proportion of 9.6%. These results indicate that the grassland condition will improve in the future. The area of grassland with a continuous decrease in AGB would be 736.43 km² (22.55%), and that with an anti-continuous increase, 524.81 km² (16.07%) (see

Table 5. The area and proportion in different categories of future aboveground biomass (AGB) trends.

Category	β	Hurst	Area (km2)	Percentage (%)
Continuous increase	$\mathsf{\beta }>0$	$\mathrm{Hurs}t>0.5$	313.51	9.60
Anti-continuous increase	$\mathsf{\beta }>0$	$\mathrm{Hurst}<0.5$	524.81	16.07
Continuous decrease	$\mathsf{\beta }<0$	$\mathrm{Hurst}>0.5$	736.43	22.55
Anti-continuous decrease	$\mathsf{\beta }<0$	$\mathrm{Hurst}<0.5$	1691.02	51.78

). Thus, the future development trend of alpine grassland AGB is an upward trend, and the proportion of grassland AGB increase (61.38%) is predicted to be higher than that of grassland AGB decrease (38.62%).

3.3. Response to Hydrothermal Factors

As shown in Figure 7a, positive correlations between AGB and precipitation occurred in 92.26% of the study area, with 49.66% of them having correlation coefficients larger than 0.5. As shown in Figure 7b, negative correlations between AGB and temperature were observed in 83.76% of the grassland area (see

Table 6. The correlations between grassland aboveground biomass (AGB) and precipitation and temperature.

Correlations Between Grassland AGB and Precipitation	Proportion (%)	Correlations Between Grassland AGB and Temperature	Proportion (%)
0.5~1	49.66	0.5~1	1.48
0~0.5	42.58	0~0.5	14.76
−0.5~0	7.42	−0.5~0	40.94
−1~0.5	0.34	−1~0.5	42.82

). Areas with positive correlations between grassland AGB and precipitation and negative correlations with temperature were located in the central region of the research area. In general, the correlation was much higher for precipitation than for temperature.

We analyzed the correlations between average AGB and both precipitation and temperature during the growing season and the significance of their interannual variations. The correlation coefficient between average AGB and precipitation was 0.92, and that between average AGB and temperature was −0.73, with the correlation between alpine grassland average AGB and precipitation being significant (see Figure 8a,b). The highest precipitation, 551.2 mm, was recorded in 2019, and the average AGB was also the highest that year. In contrast, 2022 saw the lowest precipitation, 310.04 mm, and the average AGB was also the lowest. Thus, precipitation had the largest effect on grassland AGB. With the increase in precipitation and the decrease in temperature, the pre-pandemic grassland AGB showed an increasing trend. With the decrease in precipitation and the increase in temperature, the grassland AGB during the mid- and post-pandemic periods showed a downward trend, which was more obvious post-pandemic.

3.4. The Potential Impact of the Pandemic

The pandemic may affect the number of grazing animals and tourists due to restrictive measures, thereby indirectly influencing grassland AGB. In this study, we found that the number of grazing animals and tourists did not fluctuate much during the pandemic. However, the number of tourists mid-pandemic decreased significantly compared with the pre- and post-pandemic periods, i.e., they decreased by 61.37% in 2020 compared with 2019. Although the number of grazing animals increased during the pandemic, the growth rate in 2021 was only 2.28% compared with 2020, which is the lowest level between 2018 and 2022 (see

Table 7. Numbers of grazing animals and tourists in the study area from 2017 to 2022.

Year	Number of Grazing Animals (thou)	Growth Rate (%)	Grazing Intensity (AU/ha)	Number of Tourists (thou)	Growth Rate (%)	Tourism Density (Tourists/ha)
2017	249.2	10.50	0.76	15698.3	6.00	13.52
2018	183.2	−26.5	0.56	18147.2	15.60	15.63
2019	197.2	7.64	0.60	16514.0	−9.00	14.22
2020	210.1	6.54	0.64	6379.0	−61.37	5.49
2021	214.9	2.28	0.66	6724.3	5.41	5.79
2022	237.1	10.33	0.73	11084.2	64.84	9.54

).

We analyzed the correlation and significance of average AGB during the growing season and both grazing intensity and tourism density. The average AGB was negatively correlated with grazing intensity, with a correlation coefficient of −0.6, which was not significant. A low negative correlation existed between average AGB and tourist density, with a correlation coefficient of −0.07 (see Figure 9a,b).

We mapped the spatial distribution of grazing intensity from 2018 to 2022, finding that it increased in the northeast and decreased in the southwest in 2020. However, the grazing intensity was lower in the northeast and higher in the southwest in 2021 than in 2020. This result aligned with the spatial distribution of alpine grassland AGB during the pandemic, which decreased as the grazing intensity increased, and vice versa (Figure 5 and Figure 10). In summary, the pandemic indirectly affected grassland AGB in local areas, but the impact was not significant (p > 0.05). Therefore, our hypothesis was valid.

4. Discussion

4.1. Performance of Machine-Learning Algorithms for Biomass Estimation

Elevation, the ARVI, the RVI, the NDVI, the IPVI, and the GNDVI were the optimal variables for the grassland AGB inversion model, in agreement with previous studies indicating that vegetation indices are important feature variables [53,54,55]. Riihimaki et al. found that the greatest NDVI and biomass were observed in areas of low elevation and high solar radiation, predominantly on valley slopes facing south to southwest. This is consistent with our findings indicating that the interactions between elevation and the ARVI, the IPVI, the NDVI, and the RVI were significant. The vegetation indices and AGB were significantly higher at lower altitudes. This is primarily because lower-altitude areas are closer to the tree line and have abundant shrubs and sufficient soil moisture, resulting in relatively higher vegetation indices and grassland AGB [56]. The SVM algorithm provided the highest estimation accuracy (R² = 0.85), outperforming the RF, ANN, and GPR models. SVM has high accuracy and prediction performance on small data samples [57,58] and has been used to estimate grassland AGB accurately based on vegetation indices [59,60,61]. Therefore, SVM is the best algorithm for estimating grassland AGB, considering the sample data size, optimal variables, and computational complexity.

Although the estimation accuracy of the SVM algorithm based on the optimal variables was high, its accuracy depended on the observation data to a certain extent. The sampling locations were nonuniformly distributed, and the data sample was small due to the topography, transportation, and other conditions, affecting the accuracy of AGB estimation [14,62]. Therefore, in future studies, model accuracy could be improved through surveys of undersampled grasslands, increasing the data sample and ensuring the uniform distribution of sampling points for improving the long-term monitoring of grassland ecosystems. Additionally, we acknowledged that the addition of observational data from 2018 to 2020 can further improve the reliability of the model, especially for capturing microscale changes in grassland AGB. Our model demonstrated high accuracy (R² = 0.85, RMSE = 55.7) in predicting AGB, which supports the robustness of the current results. This suggests that we were able to accurately predict grassland AGB from 2018 to 2022, despite the limited sample size. However, in future studies, incorporating additional field survey data from multiple years should be considered to capture finer-scale variations and further enhance the model’s predictive accuracy.

4.2. Trend of Grassland AGB

The results of this study demonstrate a trend of slight decrease in AGB dynamics overall, with some local areas showing a slight increase. However, the area of severely degraded grassland was significantly smaller compared with that in the period from 2008 to 2017 [30,63]. This is likely to be ascribed to local policies. Overgrazing reduces grassland AGB and adversely affects the structure and function of grassland ecosystems [64,65]. The implementation of grassland compensation policies has improved grassland vegetation growth and productivity [66,67]. The reason why the number of grazing animals in the study area decreased by 26.5% from 2017 to 2018 was that grassland protection policies, including moderate grazing, returning grazing land to grassland, and a new round of grassland protection subsidies, were implemented by the local government starting from 2018. They have increased public awareness regarding the conservation and use of grasslands, resulting in the recovery of grasslands and the prevention of their degradation. Our prediction indicates future increases in grassland AGB, which is in accordance with findings indicating that grassland productivity will continue to increase in north China and North America [12,68]. This result clearly demonstrates that the aforementioned policies and protection measures exert positive and sustainable effects on grassland recovery. Moderate grazing can stimulate plant tillering and branching and enhance photosynthetic efficiency, thereby improving compensatory growth capacity [69]. Simultaneously, it suppresses the excessive dominance of competitive species, creating ecological niches for other plants and helping to maintain the grassland ecosystem balance [70]. Therefore, it is vital to conduct continuous monitoring and formulate reasonable grazing policies to ensure the sustainability of grassland ecosystems.

We acknowledge that more complex methods, such as time series decomposition or Fourier analysis, could be employed to capture seasonal and periodic trends in greater detail. However, given the scope of our study and the specific focus on interannual trends during the growing season, we believe that the chosen methods provide a reliable and interpretable assessment of grassland AGB dynamics. In future work, we will consider incorporating more advanced techniques to explore seasonal and periodic variations in greater depth.

4.3. Relationships Between AGB and Hydrothermal Factors

The results of the above analysis indicate that the spatial distribution pattern of grassland AGB corresponds to the spatial characteristics of hydrothermal conditions and that the interannual variation trends in hydrothermal conditions show high consistency with the interannual dynamic change in average biomass in the grassland growing season. Therefore, hydrothermal factors were the principal factors influencing the spatiotemporal dynamics of alpine grassland AGB during the pre-, mid-, and post-COVID-19 pandemic periods. Hydrothermal factors fundamentally determine the productivity level of grassland ecosystems by regulating key physiological processes, such as photosynthesis of vegetation leaf area and nutrient absorption capacity [71]. Precipitation is a source of water, while temperature affects the metabolic rate and the phenological stage [72]. Extreme weather occurrences, such as prolonged drought, are detrimental to the accumulation of grassland AGB [73,74]. On the contrary, suitable water and heat conditions are conducive to the improvement in grassland productivity.

Seasonal changes in precipitation and temperature influence plant growth status and biomass accumulation [48]. A positive relationship was observed between grassland aboveground biomass and precipitation. Grasslands with high precipitation had high AGB, and vice versa. This finding is consistent with other studies. Precipitation directly influences grassland AGB, as sufficient precipitation is necessary for plant growth, photosynthesis, and nutrient uptake, which increase AGB [75,76]. In addition, precipitation could also indirectly affect grassland AGB by altering plant functional traits and plant coverage [77,78]. A negative non-significant correlation was observed between grassland AGB and temperature. The likely reason is that temperature indirectly affects grassland AGB by increasing water evaporation. Thus, grassland AGB declines when plants have insufficient water and nutrients during the growth stage [79,80]. However, the impact of climate change on grassland biomass is a long-term process, and the continuous monitoring of grassland ecosystems is needed to inform the formulation of climate adaptation strategies.

4.4. The Potential Impact of the COVID-19 Pandemic

Some studies showed that vegetation productivity increased after restrictive measures were implemented during the COVID-19 pandemic [10,29]. The pandemic itself did not influence grassland AGB but indirectly affected it due to a series of restrictive measures which led to a change in grazing intensity and tourism density. The number of grazing animals in the study area mid-pandemic did not fluctuate much, but it still increased slightly. This is likely because animal husbandry is the pillar industry in the study area, and most of the local residents rely on grazing as a living resource [30]. However, by analyzing the spatial distribution of grazing intensity mid-pandemic, we found that an increase in grazing intensity in local areas decreased grassland AGB, and vice versa. As grazing intensity increases, it leads to excessive forage consumption, which exceeds compensatory growth capacity, resulting in reduced vegetation cover, soil exposure, intensified water evaporation, and diminished soil water retention [81,82]. Ultimately, this severely inhibits plant growth and exacerbates grassland degradation [83]. Al-Ali et al. [84] found that the government’s restrictive measures during the pandemic reduced overgrazing and improved grassland vegetation recovery. This result is in agreement with our finding. We also observed a low correlation between tourism density and grassland AGB, which is likely due to the fact that large grasslands are far from densely populated areas such as cities and scenic spots. The study area has low population movement, reducing the impact of trampling disturbance on grassland ecosystems.

5. Conclusions

In this study, we systematically analyzed the dynamic changes in alpine grassland AGB in different periods, revealing how they were influenced by hydrothermal factors and anthropogenic drivers. The main findings are as follows: (1) Hydrothermal factors are the dominant drivers in the dynamic changes in alpine grassland AGB. This study demonstrated that the spatial distribution patterns and interannual variation characteristics of grassland AGB in the study area exhibited significant spatial coupling and temporal synchronization with the hydrothermal conditions and that suitable hydrothermal conditions were conducive to the growth of grassland vegetation and the accumulation of biomass. (2) Compared with the years before the study period, the degradation of grassland decreased, and its biomass in the local area showed a slight recovery trend. This good ecological situation is attributed to the grassland ecosystem restoration measures implemented by the local government, which mainly included the scientific management of the grass–livestock balance and the strict formulation of ecological compensation mechanisms. (3) A series of control measures mid-pandemic affected human activities. Although the overall grazing scale ascribed to local residents with husbandry as the main source of livelihood remained relatively stable, changes in grazing intensity in local areas had a negative effect on alpine grassland AGB.

There are some limitations to this study. The short research period makes it difficult to capture the long-term trends and cyclical patterns of alpine grassland AGB. In addition, the synergistic effect of hydrothermal factors and human activities also had a significant impact on grassland AGB, which was not considered in this study. Therefore, the long-term continuous dynamic monitoring of grassland AGB should be carried out in future studies, and the effects of various factors on grassland biomass should be comprehensively considered, as this will help to further verify the accuracy of the present research results. From the perspective of management practices, the government should continue to formulate and implement reasonable grazing policies and grassland protection mechanisms, monitor the impact of long-term climate change on grassland AGB, and formulate climate adaptation management strategies in advance to achieve the sustainable development of alpine grasslands.