Next Article in Journal
Evaluation of Fir Forest Die-Back and Regeneration After a Severe Bark Beetle Disturbance Using UAV-Based Remote Sensing
Previous Article in Journal
A Novel Framework Integrating Spectrum Analysis and AI for Near-Ground-Surface PM2.5 Concentration Estimation
error_outline You can access the new MDPI.com website here. Explore and share your feedback with us.
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Remote Sensing
Volume 17
Issue 22
10.3390/rs17223781
remotesensing-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Estimating Grazing Pressure from Satellite Time Series Without Reliance on Total Production

by
Yan Shi
1,
Jay Gao
2,
Gary Brierley
2,
Xilai Li
3 and
Jin-Sheng He
1,4,*
1
State Key Laboratory of Herbage Improvement and Grassland Agro-Ecosystems, College of Pastoral Agriculture Science and Technology, Lanzhou University, Lanzhou 730000, China
2
School of Environment, The University of Auckland, Auckland 1010, New Zealand
3
State Key Laboratory of Plateau Ecology and Agriculture, Qinghai University, Xining 810016, China
4
Institute of Ecology, College of Urban and Environmental Sciences, Key Laboratory for Earth Surface Processes of the Ministry of Education, Peking University, Beijing 100871, China
*
Author to whom correspondence should be addressed.
Remote Sens. 2025, 17(22), 3781; https://doi.org/10.3390/rs17223781
Submission received: 14 September 2025 / Revised: 4 November 2025 / Accepted: 18 November 2025 / Published: 20 November 2025
(This article belongs to the Section Ecological Remote Sensing)
Browse Figures
Versions Notes

Highlights

What are the main findings?
  • Multi-period vegetation index datasets capture seasonal grazing dynamics better than snapshot data.
  • Only 38.8% of high intensity areas identified as under high grazing pressure.
  • More than 40% of high intensity area exhibiting substantial aboveground biomass.
What is the implication of the main finding?
  • Grazing intensity alone poorly explains grazing pressure and pasture degradation risk, but uncertainty is reduced when estimates are integrated with residual biomass.
  • The proposed method produces more realistic appraisal of grazing pressure than total-production-based methods.

Abstract

Accurately assessing grazing impacts is essential for sustaining alpine grasslands. Conventional approaches often rely on total forage productivity, an indirect and uncertain proxy for forage availability. In this study, we propose a novel framework for estimating grazing pressure that integrates residual biomass with grazing intensity, thereby overcoming the limitations and uncertainties inherent in total forage-based assessments. Our results reveal pronounced spatiotemporal variation in grazing intensity: lowland areas experienced the highest intensity early in the growing season, whereas upland areas became more heavily grazed later in the season. However, grazing intensity alone proved insufficient to explain grazing pressure or predict pasture degradation risk. Overlay analyses demonstrated that only 38.8% of high intensity areas identified as under high grazing pressure, and more than 40% of high intensity area exhibiting substantial aboveground biomass. These findings highlight the limited explanatory power of grazing intensity when considered in isolation. By explicitly incorporating standing biomass rather than relying merely on total production, the proposed framework reduces estimation uncertainty, enhances ecological realism, and provides a scalable, more accurate and practical tool for monitoring grassland utilization and degradation.

1. Introduction

Grasslands, particularly those in alpine and highland regions, are essential and important socio-ecological systems that support pastoral livelihoods and provide a range of ecosystem services, including biodiversity conservation, climate regulation, and carbon sequestration [1,2]. The fragility of alpine meadows reflects their harsh climates, short growing seasons, and limited vegetation resilience [3,4]. As the dominant land use form in the Qinghai–Tibetan Plateau (QTP) and surrounding regions, livestock grazing plays a crucial role in shaping ecosystem dynamics [5,6]. However, excessive or poorly managed grazing imposes a substantial pressure on meadow biomass, alters surface energy and carbon exchange processes, and contributes increasingly to widespread meadow degradation [7,8]. Alarmingly, approximately 40% of grasslands in the Three-River Source Zone of the QTP have been reported to be degraded or fragmented [9,10]. Grassland degradation undermines not only the ecological functions but also the long-term sustainability of pastoral systems. Therefore, there is an urgent need to develop competent and efficient methods to monitor grazing impacts and detect early signs of degradation in order to prevent irreversible damage in an effective manner.
As grazing pressure is gauged by the capacity of a pasture to support grazing without inducing degradation, it provides an integrative perspective on livestock-ecosystem interactions [11,12]. It is defined as the relationship between livestock forage demand and the total availability of forage yield, commonly approximated as aboveground net primary productivity (ANPP) [13]. For large scale studies, ANPP is generally quantified from satellite images. However, this method of measuring grazing pressure is critically constrained as it does not consider biomass already consumed by herbivores, which also constitutes part of the forage demand, as many remote-sensing-based estimates of ANPP refer specifically to peak standing biomass at a given moment, not the overall biomass yield over a period (e.g., the growing season) [14]. Consequently, systemic underestimation of available forage leads to an overestimation of actual grazing pressure and, in turn, provides an exaggerated measure of ecosystem resilience [1,15,16,17]. Also, as grazing activities vary across landscapes, spatial grazing pressure models are required to capture heterogeneity and avoid oversimplified assumptions. Without quantifying grazing pressure, it is difficult to assess ecosystem functioning. Thus, it is imperative to develop a direct and effective method for realistically estimating grazing pressure.
Traditionally, grazing intensity, defined as livestock population per unit area over a given time period [18], is used to assess rangeland condition and degradation. This metric has been widely adopted due to its simplicity and availability in livestock statistics from administrative sources [19,20,21]. However, grazing intensity alone fails to reflect the ecological stress imposed on forage vegetation, as it does not take into account the spatial variability of forage availability, vegetation productivity, or recovery potential [22] in case of disturbance or impairment. Consequently, pastures with a high grazing intensity may remain ecologically stable if primary productivity is maintained at a sufficient level, while grasslands with a low grazing intensity may still experience degradation due to inadequate vegetation resilience or unfavorable environmental conditions (e.g., shady slopes) [23,24,25]. Generally, grazing intensity over extensive areas has been widely modeled by varied machine learning models (such as support vector machines (SVM), artificial neural networks (ANN)) based on changes in vegetation index (VI) derived from remotely sensed images [18,20,26,27]. While such a method is straightforward and easy to implement, and is able to produce replicable and consistent estimates, it faces several inherent limitations in light of spatial heterogeneity in the grazed pasture and seasonal fluctuations in the local environment [28,29]. For instance, marked spatial variability in forage production is evident in mountainous areas. Livestock grazing is a highly dynamic and variable process and activity that varies spatially, temporarily, and even seasonally as livestock roam the pasture to search for the most palatable forage available. Even the same productive pasture will cease to yield new forage in deep winter in some harsh high-altitude regions of the world [30]. Consequently, a single VI obtained from a snapshot remote sensing image is unable to capture all of the localized or transient grazing variability. More profoundly, such procedures fail to capture the cumulative and heterogeneous effects of grazing across the entire growing season from time-series images. On the other hand, grazing intensity is strongly correlated with grazing pressure, and areas of a higher grazing intensity generally correspond to a higher grazing pressure [11]. Thus, grazing intensity can serve as an effective surrogate for grazing pressure under certain conditions.
Residual biomass, the aboveground biomass remaining at the end of the growing season after grazing has terminated, provides an underutilized yet ecologically critical variable related to grazing pressure [31,32]. As residual biomass captures the cumulative impact of herbivory, it can be used to accurately depict pasture vulnerability [33], as it not only reflects the actual production capacity of a grassland ecosystem but also indicates whether that productivity has been sustainably utilized. For instance, low residual biomass levels are a clear signal of overgrazing [34].
Remote sensing offers a non-destructive means of estimating grassland biomass by relying on indirect indicators such as vegetation indices (VIs). These indices can be integrated with additional environmental variables—such as geographic location, topography, and meteorological factors—to improve the estimation of standing biomass [35,36,37]. Incorporating such variables has been shown to reduce the prediction error of VI-based models, thereby enhancing their stability and general applicability [38,39,40,41]. Among the various indices, the normalized difference vegetation index (NDVI) has been most extensively evaluated and consistently emerges as the most accurate and widely used indicator of grassland standing biomass [42,43,44]. In recent years, diverse machine learning techniques—including SVM, ANN, and deep neural networks (DNN)—have been increasingly applied to improve the accuracy of biomass estimation [45,46,47].
Both grazing intensity and residual biomass are closely correlated with grazing pressure. Building on this relationship, the present study proposes a new framework to estimate spatial variability in grazing pressure that integrates grazing intensity with residual biomass at a catchment scale, thereby overcoming the limitations associated with the indirectly measurable ANPP derivable from remotely sensed data. Focusing on an alpine meadow catchment in the source region of the Yellow River as a case study, this paper evaluates the capability of the proposed framework to quantify the vulnerability of a grassland ecosystem to degradation. The specific objectives are: (1) To estimate grazing intensity across the growing season by incorporating multi-period vegetation index (VI) changes, thereby capturing the temporal variability of grazing activities; (2) To develop a non–forage-production-based method for estimating grazing pressure using residual biomass as a direct indicator in combination with grazing intensity; and (3) to assess the strength of this method comparatively against the estimation method based on ANPP.

2. Materials and Methods

2.1. Study Area

This study was conducted in the Yongqu River Basin, located inside Henan Mongolian Autonomous County in the southeastern Yellow River Source Region (Figure 1a,b). The area represents a typical alpine meadow ecosystem, with elevations ranging from 3100 to 4600 m above sea level and a diverse array of geomorphic features shaped by complex topography. The region experiences a plateau continental climate characterized by marked seasonal variations in precipitation. Annual rainfall ranges from 597 mm to 616 mm, with approximately 75% occurring between May and September (http://data.cma.cn/ (accessed on 1 February 2020)). The average annual temperature ranges from 9 °C to 15 °C, accompanied by substantial diurnal temperature fluctuations. The two vegetation types in the study area are grasses and shrubs. The former consists mostly of Kobresia spp., Elymus nutans, the latter are mainly Potentilla fruticose and Salix oritrepha [48]. This area generally is under moderate grazing intensity. Livestock grazing in the region follows a seasonal migratory pattern, driven primarily by topographic conditions and the availability of water and forage resources [49]. Herders typically move livestock from winter pastures in valley bottoms to higher-elevation summer pastures in May and June, returning between late August and October [50]. These rotational timings rely heavily on the experiential knowledge of herdsmen, which, while adaptive, can inadvertently lead to mismatches between forage availability and livestock demand, thereby increasing the risk of both overgrazing and underutilization.

2.2. Field Data Collection and Processing

A total of 187 sampling sites, including 129 grassland sites and 58 shrubland sites, was established in 81 paddocks along several cross-sections in the Yongqu River catchment (Figure 1). Shrub sampling sites were located in areas where shrub cover exceeded 60%, while grassland sampling sites were selected in areas entirely devoid of shrubs. Aboveground biomass sampling was conducted at the end of the growing season from 27 August to 9 September 2019. Field observations indicate that by early September, night-time temperatures frequently fall below 0 °C, and growth had effectively stopped at selected field sites. At each site, a 10 m × 10 m plot was delineated on the ground, within which two 1 m × 1 m subplots were randomly selected for biomass collection. One-quarter of the standing biomass within each subplot was clipped and subsequently oven-dried at 60 °C for 48 h in the laboratory to obtain ground truth data. The geographic coordinates of each subplot (centroid) were logged using a GPS device (Garmin Rino650, Shanghai, China). Livestock population and foraging days were recorded during monthly field campaigns from June to September.
Various types of remote sensing data were acquired to support spatial analysis including VIs Changes, landform and vegetation classification, grazing intensity modeling and aboveground biomass modelling. A Digital Elevation Model (DEM) was obtained from the Alaska Satellite Facility (https://asf.alaska.edu/ (accessed on 1 June 2019)). Five time-series Sentinel-2 satellite images (free dataset with high frequency and resolution) were initially acquired throughout the growing season of 2019, from which cloud-covered areas were manually masked and excluded from analysis. Due to substantial cloud coverage in August, two images on 10 and 16 of August had to be merged to generate one cloud-free composite image; approximately 80 km2 (11.3% of the total area) of the 16 August image was substituted with data from the 10 August. As a result, four usable images were analyzed in this study, captured on 12 June, 27 July, 16 August, and 5 September.

2.3. Research Method

2.3.1. Overall Workflow

Figure 2 presents the study workflow, comprising three modules: (i) step-by-step derivation, which develops and refines the equations for the Grazing Pressure Index (GPI); (ii) modeling process, which details the estimation procedures for the two GPI components—grazing intensity and residual biomass; and (iii) spatial analysis, which evaluates spatial heterogeneity and performs overlay analyses to assess patterns and concordance among grazing intensity, residual biomass, and the resulting GPI.

2.3.2. Grazing Pressure Index Estimation

Grazing pressure index (GPI) is generally used to represent grazing pressure. GPI is generally measured by the ratio of forage demand to forage supply [13,51], Figure 2a demonstrates the derivation of GPI, and it is calculated as:
G P I = F o r a g e   d e m a n d F o r a g e   s u p p l y
where Forage supply refers to the total available biomass, also known as ANPP. Many ANPP estimates rely solely on peak standing biomass, yet true ANPP comprises both peak standing production and biomass consumed [14]. To improve accuracy, we estimated ANPP in grazed grasslands as the sum of two components: forage demand (biomass consumed by livestock during the grazing period) and residual biomass (standing biomass remaining after grazing) shown as:
F o r a g e   s u p p l y = F o r a g e   d e m a n d + R e s i d u a l   b i o m a s s
where Residual biomass represents the standing biomass left at the end of the growing season when plant growth ceases (i.e., no further biomass accumulation), it can be modeled based on remote sensing imagery and ground observations (biomass); Forage demand denotes the biomass consumed by livestock prior to the termination of the growing season, it can be calculated based the following equation:
F o r a g e   d e m a n d = S t o c k i n g   r a t e × G r a z i n g   d a y s   × D a i l y   i n t a k e          
G r a z i n g   i n t e n s i t y = S t o c k i n g   r a t e × G r a z i n g   d a y s
where SU denotes sheep units, one adult yak is equivalent to 5 SU (NY/T 635–2015; Calculation of Reasonable Livestock Carrying Capacity of Natural Grassland. Ministry of Agriculture of the People’s Republic of China: Beijing, China, 2015) [52]. Grazing days is the number of days during which livestock graze. Daily intake is the biomass consumed per SU per day; for alpine meadow on the Qinghai–Tibet Plateau, it is 1.8 kg SU−1 day−1 (NY/T 635–2015) [52].
Thus, according to Equations (1)–(4), GPI is expressed as:
G P I = 1 1 + R e s i d u a l   b i o m a s s G r a z i n g   i n t e n s i t y × D a i l y   i n t a k e
where Area and Daily intake are considered constants. Thus, GPI is dictated jointly by two variables: Grazing intensity and Residual biomass.

2.3.3. Grazing Intensity Modelling

As shown in Figure 2b, grazing intensity was modeled separately for each time interval, and the cumulative sum across the three intervals was used to represent grazing intensity for the entire growing season. In this study, the three intervals, 12 June–27 July, 27 July–16 August, and 16 August–5 September, were defined by the gaps between four satellite acquisition dates.
Ground-truth grazing intensity data (paddock area, livestock number and grazing days) was derived from field observations and Henan County Grassland Station. To align the satellite-derived indicators with field data, all the inputs were averaged spatially over each paddock area (polygon shapefile), ensuring that the image-derived metrics matched exactly the ground-based grazing observations in a paddock. Ground-truthing grazing intensity data were obtained in 70, 65, and 59 paddocks during the three respective periods of the growing season. The trained model is then applied to the original Sentinel-2 raster at 10 m × 10 m resolution to produce spatially explicit prediction maps. Seasonal grazing intensity was computed as the sum of the three period estimates. Because each period-specific model was validated against its corresponding ground truth, we did not conduct a separate validation of the seasonal sum.
Several multi-variable modeling approaches were adopted to upscale the paddock-level grazing intensity to the landscape level, including traditional multiple linear regression (MLR), support vector machines (SVM), artificial neural networks (ANN), and deep neural networks (DNN). The modelling process were based on our previous study [53]. Models’ performance may decline with other input datasets due to the overfitting problem. Weight decay and early stopping were used to interrupt training at an appropriate point to minimize the overfitting problems [42]. In the modeling, field-observed grazing intensity served as the response variable, while spatial metrics (time-series NDVI), topographic attributes (e.g., elevation, slope, aspect), climatic variables (temperature and precipitation), and categorical variables (e.g., vegetation type, and observation period) encoded using a one-hot scheme were treated as the predictor variables [54]. The 10-fold cross-validation method was used to evaluate model accuracy. The best-performing model was subsequently selected to generate spatially explicit grazing intensity maps for further ecological and management analysis.

2.3.4. Residual Biomass Modelling

As shown in Figure 2b, residual biomass is the standing biomass at the end of growing season. in this study, it was modeled from a Sentinel-2 image acquired on 5 September based on a total of 187 biomass samples (129 grass samples and 58 shrub samples) collected in early September. Spectral bands (red, green, blue and near infrared), NDVI, topographic attributes (elevation, aspect and slope), climatic variables (temperature and precipitation) and vegetation type (grass, shrub and non-vegetated) were used as inputs. None of the sampling sites were contaminated by cloud cover, and hence were fully usable for model development. The modelling framework for residual biomass was implemented using Google Earth Engine based on our previous study [53]. The 10-fold cross-validation method was used to evaluate model accuracy. The best performing model was used to map the landscape-level distribution of grassland aboveground biomass.
Given that the grassland utilization ratio and the proportion of edible forage both vary with vegetation type and geography [55,56], vegetation types were categorized into three classes: grassland, shrubland, and non-vegetated areas from the Sentinel satellite images using a support vector machine (SVM) algorithm in ArcMap 10.8. Sentinel-2 satellite images (10 m) were used to visually identify and randomly select training samples to ensure the highest reliability of vegetation mapping. Throughout image classification, non-vegetated areas and cloud-covered pixels were consistently masked to minimize noise and improve the robustness of standing biomass estimates and mapping outputs. The classification was accomplished at an overall accuracy in excess of 90.0% and a Kappa coefficient greater than 0.84. The classified vegetation types were deemed acceptable, and subsequently integrated into the modeling process.

2.3.5. Spatial Analysis

As shown in Figure 2c, based on the maps derived from the best models, spatial heterogeneity and overlay analysis was the final component of this study. Landscape heterogeneity was characterized to support spatial analyses of the patterns of grazing intensity, residual biomass and grazing pressure. Landforms of the study area were classified from the DEM using the Geomorphons landforms classification method in ArcGIS Pro 3.5 [57]. To facilitate analysis of terrain-related effects, detailed geomorphon landform classes were grouped into three broader geomorphic units based on topographic position and functional similarity: upland (ridge, peak, spur), mid-slope (shoulder, slope), and lowland (flat, hollow, valley, pit, footslope). These categories accounted for 21.1%, 40.0%, and 38.9% of the total study area, respectively (Figure 1c). This reclassification enabled a more systematic comparison of vegetation dynamics and grazing intensity across distinct landforms, and facilitated the inference of local topographic variations influencing grazing activity and vegetation distribution [58,59].
For raster maps of grazing intensity, residual biomass, and the composite grazing pressure, the Natural Breaks (Jenks) method is preferable because it explicitly optimizes class boundaries to minimize within-class variance and maximize between-class variance, thereby aligning class cuts with real distributional gaps common in skewed, multi-modal environmental data (Data classification methods—ArcGIS Pro 3.5 documentation). In contrast, Quantile forces equal counts per class, often mixing dissimilar values in dense portions of the distribution; Equal Interval ignores distributional shape and can obscure hotspots when extreme values stretch the range; and Standard Deviation centers interpretation on departures from the mean, which is ill-suited when data are non-normal or heavy-tailed. Empirical and cartographic evaluations have long recommended data-adaptive schemes like Jenks for thematic rate/intensity mapping because they preserve meaningful clusters and enhance visual discrimination, improving both map reading and cross-class separability. In ArcGIS Pro 3.5, Natural Breaks implements Jenks’ optimization directly and is documented as the default data-driven classifier for graduated symbology [60].
Thus, the Natural Breaks (Jenks) classification method was employed to identify four natural groupings (very low, low, moderate, and high) in grazing intensity, residual biomass, and grazing pressure datasets. This approach minimizes variance within classes while maximizing variance between classes, thereby revealing inherent and genuine patterns in the input data. It is particularly effective for skewed or clustered datasets and produces visually intuitive maps that highlight meaningful spatial disparities [61]. For the grazing pressure dataset, the three periods differed in length (45, 20, and 20 days, respectively). To ensure full comparability, the results of the first period were standardized to a 20-day equivalent, thereby improving the accuracy and consistency of assessment.
Overlay analysis in ArcGIS Pro 3.5 was employed to examine the spatial correspondence between different variables including grazing pressure, grazing intensity and residual biomass. This method involves the integration of multiple thematic layers within a GIS environment, allowing quantification and visualization of the degree of overlap between the two variables [62]. This overlay approach provided a clear spatial framework to evaluate consistency. It was able to identify areas where high levels of both variables coincided, as well as zones of divergence, thereby offering valuable insights for sustainable grassland management.
For statistical analysis, we evaluated treatment effects in Python 3.9 using a multiple-comparison framework. For approximately normal outcomes, we fit linear models and conducted pairwise comparisons with Tukey or Holm-adjusted t-tests; when assumptions were uncertain, we applied rank-based alternatives. We report mean differences with 95% confidence intervals and adjusted p-values, interpreting p < 0.05 as significant and p < 0.001 as highly significant. Analyses were performed with packages of statsmodels 0.14.4 and SciPy 1.16.3.

3. Results

3.1. Modelled Grazing Intensity and Its Spatial Variation

Figure 3 shows slight variations in the performance of the four modeling approaches. Although SVM achieved the lowest root mean square error (RMSE = 48.37 SU·day ha−1), its accuracy (R2 = 0.827) was lower than that of DNN (R2 = 0.875). Given the high accuracy of DNN, it was selected for generating aboveground biomass maps of the study area.
Grazing intensity over the three growing periods and the whole growing season was assessed via pairwise comparisons (Figure 4). There were marked differences in grazing intensity among landforms, although the patterns varied across the three periods. In the period of 12 June–27 July, lowlands exhibited significantly higher values than uplands (p < 0.01), while the contrasts between lowlands and mid-slopes and between uplands and mid-slopes were not significant. In the period of 27 July–16 August, grazing intensity did not differ significantly among the three landforms. By the period of 16 August–5 September, the pattern shifted, with uplands showing significantly higher values than lowlands (p < 0.05), whereas the comparisons involving mid-slopes remained insignificant. However, no significant differences were observed between landforms throughout the growing season.
Areas of a low grazing intensity accounted for more than half of the total area across the growing season. However, the extent of very low grazing intensity declined sharply from 25.7% to less than 5%, while the proportion of moderately grazed areas increased from 20.0% to 39.2%. The area under highly intensive grazing remained relatively stable at around 5%. Overall, over the entire growing season, the distribution of grazing intensity ranked as low > moderate > very low > high. Notably, the total area classified as high grazing intensity during the full season (Figure 4d) was approximately double that observed in any of the individual periods (Figure 4a–c).

3.2. Residual Biomass Variability

Figure 5 illustrates the performance of the four models in predicting residual biomass. The DNN model achieved the highest accuracy (R2 = 0.818) and the lowest root mean square error (RMSE = 107.80 g/m2), outperforming the other three models. All of them exhibited an accuracy below 0.80. Although the MLR model achieved a relatively high R2 value of 0.798, its performance was compromised by a substantially higher RMSE of 130.40 g/m2, compared to the SVM and ANN models. Given its superior accuracy and predictive power, the DNN model was adopted for generating aboveground biomass maps of the study area.
Grasses and shrubs occupied a comparable proportion of the study area to each other, with grasses predominantly distributed in lowland areas, followed by mid-slope regions (Figure 6). In contrast, shrubs were primarily found on mid-slopes, followed by lowlands, and had a greater coverage than grasses in upland areas.
Although shrubs exhibited a significantly higher total biomass density (p < 0.01), approximately 715.6 g/m2 compared to 347.5 g/m2 for grasses, the available biomass was significantly higher in grasses (p < 0.05). Grasses provided around 165.5 g/m2 of available biomass, whereas shrubs contributed only about 102.9 g/m2.

3.3. Grazing Pressure and Its Spatial Variability

The grazing pressure index map (Figure 7) generated from grazing intensity integrated with residual biomass showed that lowland areas experienced the highest grazing pressure, even though the differences among different landforms were not statistically significant. Similarly, shrublands faced a higher grazing pressure than grasslands, but the difference between them was also insignificant. Spatially, areas under a moderate grazing pressure dominated the landscape, accounting for 62.2% of the total, followed by areas with a low grazing pressure (26.6%) and a high grazing pressure (10.3%). In contrast, very low grazing pressure was rarely encountered, comprising less than 1% of the study area.
Figure 8 illustrates the overlap ratios among grazing pressure, grazing intensity, and residual biomass. Of the areas classified as having a high grazing pressure (Figure 8a), only 37.3% of them also experienced a high grazing intensity, while 34.4% were subject to moderate grazing and 21.9% to low intensity grazing. Conversely, of the areas classified as having a high grazing intensity (Figure 8c), 38.8% of them overlapped with those having a high grazing pressure, whereas more than half corresponded to those of a moderate grazing pressure.
Moreover, a high grazing pressure was strongly associated with a very low residual biomass, over 99.8% of the corresponding area (Figure 8b). In contrast, only 59.3% of areas under a high grazing intensity exhibited a very low residual biomass (Figure 8d). This further highlights the limitations of grazing intensity as an indicator for assessing grassland utilization and degradation status.

4. Discussion

4.1. Variability of Grazing Intensity During Growing Season

Our analysis reveals pronounced spatiotemporal variability in grazing intensity across multiple landforms, with statistically significant pairwise differences that shift over the growing season. These dynamic changes are consistent with landscape-ecological theory and rangeland science in that topography and resource distribution govern the structure of herbivore use patterns, producing fine-grained spatial heterogeneity and temporal turnover in use [63,64,65]. In alpine grassland ecosystems, shifting water availability, phenology, and access constraints commonly force livestock to graze valley bottoms first, where soils are deeper and forage greens-up earlier, followed by seasonal migration to higher grounds after lowland forage has been mostly devoured or its quality deteriorated [58,66].
These results also underscore a key methodological limitation of deriving grazing intensity from single-period VI change. VI signals are sensitive to short-term precipitation pulses, and phenological stage, which can mask or mimic grazing effects if assessed at a single interval [67,68]. A single snapshot is therefore prone to capturing localized or transient impacts (e.g., a rain-driven green-up in uplands or a momentary concentration of animals near water) rather than the cumulative, heterogeneous use that unfolds across multiple months. Instead, multi-period VI trajectories better represent seasonal utilization by integrating phenological phases and reducing climate-noise aliasing [67,69]. Our three-period assessment aligns closely with this rationale, revealing that landform hierarchy of grazing intensity is not static but flips as the season elapses—information that a single-period approach would have missed.
Ecologically, the observed landform-dependent shifts in grazing intensity are expected to translate into uneven risks of vegetation depletion and soil disturbance over time, particularly where repeatedly intensive grazing coincides with sensitive substrates or shallow soils [31,70]. Lowlands, often the earliest and most consistently used pasture, can accumulate a grazing pressure detrimental to residual biomass and recovery potential; later-season migration of livestock to uplands broadens the footprint of pasture utilization to steeper and more erosion-prone areas [58,59]. Such spatiotemporal mosaics of use central to the patch-dynamic tenet of rangelands accentuate why multi-period indicators should complement intensity estimates derived from snapshot images [64,65].
By demonstrating that grazing intensity varies substantially across meadow landforms and throughout the growing season, our findings caution against the prevalent reliance on single-period VI observations, which risk misrepresenting the dynamics of grazing pressure due to their sensitivity to transient signals. In contrast, our approach offers a more robust and ecologically meaningful assessment of cumulative grazing use and associated degradation risks. This integration not only improves the accuracy of grazing pressure estimation in topographically heterogeneous landscapes but also advances the capacity of remote sensing to support sustainable pasture management and conservation planning.

4.2. Methodological Framework for Estimating Grazing Pressure

The key innovation of this study lies in its departure from traditional methods that rely exclusively on ANPP to estimate grazing pressure [11]. Our approach reduces the uncertainties associated with ANPP in quantifying grazing pressure by refining the mathematical formulation to incorporate two easily quantifiable factors: residual biomass and grazing intensity. According to our previous study [14], this omission is the dominant source of bias—leading to an underestimation of ANPP by ~31%. Given common formulations of grazing pressure (which scale inversely with ANPP), such underestimation inflates the estimated pressure; hence, ANPP-based approaches tend to overestimate grazing pressure when consumed biomass is ignored. Residual biomass can be directly captured by satellite images and modeled at a relatively high accuracy. By contrast, ANPP represents a conceptual construct rather than a precise physical entity that can only be inferred indirectly through surrogate measurements [71,72,73]. Studies that use the peak standing biomass as the ANPP to estimate the grazing pressure [11,74] overestimate actual grazing pressure. This, in turn, provides an exaggerated measure of ecosystem resilience. Our proposed framework overcomes these limitations by directly integrating remotely sensed grazing intensity with residual biomass. By leveraging residual biomass rather than ANPP, our method reduces reliance on difficult-to-validate process-based models and avoids the error-prone assumptions associated with peak biomass equating ANPP. It also ensures that livestock consumption is explicitly considered, rather than being inferred indirectly or ignored altogether.
Overall, this integrated approach not only improves the ecological fidelity of grazing pressure estimation but also enhances its practical applicability for land managers and policy makers. It enables targeted interventions at appropriate spatial scales and provides a robust basis for monitoring the sustainability of grazing practices. More broadly, the work establishes a methodological framework that can be applied to other alpine and semi-arid ecosystems worldwide, strengthening the role of Earth observation in ecological monitoring under conditions of climatic and anthropogenic stress.

4.3. Overlaps Between Grazing Intensity and Grazing Pressure

Our spatial overlays reveal a pronounced mismatch between locations of high livestock use and areas experiencing genuinely high grazing pressure. Only 38.8% of highly intensively grazed pastures coincided with high-pressure zones, and more than 40% overlapped with sites retaining substantial residual biomass. This pattern indicates that instantaneous use (intensity) is an unreliable proxy for the net balance between forage demand and supply (pressure), which ultimately governs vegetation stress, recovery, and degradation risk.
Several ecological mechanisms underpin this decoupling. First, fine-scale heterogeneity in forage quantity and quality drives selective foraging: animals briefly concentrate on palatable patches before moving on as marginal intake declines [75,76]. Second, temporal lags in plant regrowth and phenology modulate effective pressure independently of short-term intensity; identical use rates can yield very different outcomes depending on growth stage and moisture, with consequences for carbohydrate reserves and bud banks [77,78]. Third, terrain and water proximity channel animals into predictable corridors early in the season, whereas topographic moisture gradients and thermal constraints later redistribute use upslope; such movements can produce high measured intensity without sustained pressure where forage production is simultaneously high [79,80]. Finally, residual biomass functions as a biophysical “shock absorber,” moderating microclimate, limiting trampling and erosion, and safeguarding tiller recruitment [81,82]—a buffering component that intensity metrics, by definition, fail to capture.
These mechanisms have direct implications for degradation indicators. Grazing intensity is widely used because it is comparatively easy to measure or infer from livestock counts and animal tracking [13,83], and it aligns with traditional monitoring and remote sensing of animal presence [18,20]. However, grazing pressure—forage demand relative to contemporaneous supply, often operationalized via residual biomass or utilization thresholds—more accurately predicts vegetation decline, soil exposure, and long-term productivity losses because it integrates plant growth capacity, seasonality, and recovery potential [74,84]. In practice, intensity can over-flag productive patches that tolerate brief heavy use, while pressure identifies those patches where demand persistently exceeds regrowth, a condition more tightly linked to degradation trajectories [13,51].
Seasonal rotational movements further structure the spatial separation between intensity and pressure. In many alpine and temperate systems, herds occupy lowlands early in the growing season, shift to highlands mid-season, and return downslope at season’s end [85,86]. This sequence generates pulses of intensity that reflect herding logistics and phenology rather than a steady demand-over-supply ratio. Early-season lowlands may exhibit high intensity but low pressure due to rapid regrowth and ample soil moisture; mid-season uplands can display moderate intensity yet elevated pressure if growth slows or soils are shallow; late-season returns to lowlands may elevate pressure despite modest intensity when regrowth stalls and residual biomass has already been depleted. Consequently, management strategies that rely solely on intensity risk misallocating rest or destocking. Monitoring should prioritize pressure-based metrics stratified by landform and season, complemented by adaptive movement schedules that maintain minimum residuals and align spatial grazing demand with forage production [87,88].

4.4. Implications and Limitations

Several important implications for rangeland sustainability emerge from this study, particularly in alpine and topographically heterogeneous environments. First, the proposed non-forage-based framework for estimating grazing pressure enables more precise rangeland management by incorporating residual biomass rather than ANPP as a key driver. This substitution simplifies the mathematical calculation and makes use of more accurate, directly quantifiable input variables for estimating grazing pressure. Second, the results reveal that even areas under a moderate grazing intensity can experience a high grazing pressure and potential degradation. This highlights the need to reconsider what constitutes “moderation” in grazing management. Defining acceptable use solely in terms of livestock numbers risks underestimating the long-term ecological footprint and trajectory of impacts, especially in fragile ecosystems. Third, the framework provides a scalable and transferable tool for policymakers and land managers, particularly in data-sparse regions where field-based monitoring is logistically challenging. By integrating satellite derived indicators with machine learning–based standing biomass (rather than ANPP) predictions, this approach generates actionable spatial maps that can guide restoration planning, rotational grazing schemes, or seasonal exclusions.
Accurate grazing-intensity data are critical for estimating grazing pressure. Although multi-period vegetation indices improved grazing-intensity accuracy, they require frequent ground observations synchronized with satellite acquisitions, increasing field effort and time costs. Accurate data on livestock daily intake remain limited, posing major challenges for reliably estimating livestock-consumed biomass. Standardized daily feed intake values for sheep units can differ across regions and livestock types [52], introducing potential errors. Despite employing robust machine-learning algorithms, our estimates of grazing pressure—specifically grazing intensity and residual biomass—remain subject to several limitations and sources of uncertainty. Even with identical inputs, predictions can vary modestly across model realizations, particularly under cross-validation. Because we report a single realization in this study, alternative runs of the same model could yield slightly different outcomes. Additional uncertainty stems from pronounced spatiotemporal heterogeneity in landscape and hydrological conditions [89,90]. Although we used weight decay and early stopping to mitigate overfitting, further ground truth observations are needed to enhance model performance. Going forward, sampling should prioritize broader geographic coverage rather than a few focal sites to strengthen out-of-sample generalizability and improve the transferability of results from individual watersheds to grazing–adapted alpine grasslands across the Qinghai–Tibet Plateau.

5. Conclusions

This study proposes a new framework for estimating grazing pressure and explicitly distinguishes between grazing intensity and grazing pressure. The approach addresses a key limitation of traditional ANPP-based methods, which typically omit biomass consumed by livestock. Our results reveal pronounced spatiotemporal variability in grazing intensity across landforms, with seasonal shifts between lowlands and uplands. We further show that single-period vegetation-index changes are inadequate, as they capture localized or transient impacts rather than cumulative seasonal patterns.
Crucially, grazing intensity alone does not reliably indicate degradation risk: high intensity did not consistently coincide with high grazing pressure or low residual biomass. By contrast, elevated grazing pressure was strongly associated with forage depletion, underscoring its greater ecological relevance as a direct indicator of grassland utilization and degradation. By substituting ANPP with residual biomass and explicitly accounting for grazing intensity, the proposed framework offers a more accurate and reliable basis for grazing-pressure monitoring. The method is scalable and actionable for policymakers and land managers, supporting targeted interventions such as rotational grazing, seasonal exclusion, and restoration planning in fragile alpine ecosystems.

Author Contributions

Y.S. conceived the ideas and designed methodology; J.G., G.B., X.L. and J.-S.H. contributed to the original draft preparation, review and editing; All authors have read and agreed to the published version of the manuscript.

Funding

This study was supported by the Postdoctoral Fellowship Program of CPSF under grant number GZC20230998. It was also supported by the China Postdoctoral Science Foundation (grant number 2024M751260), Science and Technology Department of Gansu Province (grant number 24JRRA522), National Natural Science Foundation of China (grant number U23A20159 and U21A20191) and Qinghai Science and Technology Department (grant number 2023-QY-210) and CAS-ANSO Fellowship (CAS-ANSO-FS-2024-38).

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Acknowledgments

We thank Yanbiao Chen, Qie-jiang-zha-xi, Gong-bao-ang-de, Peng Han, Yupeng Zhang, Zhixiang Gao, Qian Li, Yongsheng Liu, Faxu Bai, and Rui Ma for assisting the field data collection. We also thank Ka-zhao-cai-rang for the help of field investigation.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Piipponen, J.; Jalava, M.; de Leeuw, J.; Rizayeva, A.; Godde, C.; Cramer, G.; Herrero, M.; Kummu, M. Global trends in grassland carrying capacity and relative stocking density of livestock. Global Change Biol. 2022, 28, 3902–3919. [Google Scholar] [CrossRef]
  2. Li, X.L.; Gao, J.; Brierley, G.; Qiao, Y.M.; Zhang, J.; Yang, Y.W. Rangeland Degradation on the Qinghai-Tibet Plateau: Implications for Rehabilitation. Land. Degrad. Dev. 2013, 24, 72–80. [Google Scholar] [CrossRef]
  3. Chen, Z.; Zhang, X. Value of ecosystem services in China. Chin. Sci. Bull. 2000, 45, 870–876. [Google Scholar] [CrossRef]
  4. Dong, S.; Shang, Z.; Gao, J.; Boone, R.B. Enhancing sustainability of grassland ecosystems through ecological restoration and grazing management in an era of climate change on Qinghai-Tibetan Plateau. Agric. Ecosyst. Environ. 2020, 287, 106684. [Google Scholar] [CrossRef]
  5. Sun, J.; Fu, B.; Zhao, W.; Liu, S.; Liu, G.; Zhou, H.; Shao, X.; Chen, Y.; Zhang, Y.; Deng, Y. Optimizing grazing exclusion practices to achieve Goal 15 of the sustainable development goals in the Tibetan Plateau. Sci. Bull. 2021, 66, 1493–1496. [Google Scholar] [CrossRef] [PubMed]
  6. Zhao, J.X.; Luo, T.X.; Li, R.C.; Li, X.; Tian, L.H. Grazing effect on growing season ecosystem respiration and its temperature sensitivity in alpine grasslands along a large altitudinal gradient on the central Tibetan Plateau. Agric. For. Meteorol. 2016, 218, 114–121. [Google Scholar] [CrossRef]
  7. Yu, C.; Zhang, J.; Pang, X.P.; Wang, Q.; Zhou, Y.P.; Guo, Z.G. Soil disturbance and disturbance intensity: Response of soil nutrient concentrations of alpine meadow to plateau pika bioturbation in the Qinghai-Tibetan Plateau, China. Geoderma 2017, 307, 98–106. [Google Scholar] [CrossRef]
  8. Gao, Q.; Guo, Y.; Xu, H.; Ganjurjav, H.; Li, Y.; Wan, Y.; Qin, X.; Ma, X.; Liu, S. Climate change and its impacts on vegetation distribution and net primary productivity of the alpine ecosystem in the Qinghai-Tibetan Plateau. Sci. Total Environ. 2016, 554–555, 34–41. [Google Scholar] [CrossRef]
  9. Li, X.-l.; Perry, G.L.W.; Brierley, G.; Gao, J.; Zhang, J.; Yang, Y.-w. Restoration prospects for Heitutan degraded grassland in the Sanjiangyuan. J. Mt. Sci. 2013, 10, 687–698. [Google Scholar] [CrossRef][Green Version]
  10. Li, L.; Zhang, Y.; Wu, J.; Li, S.; Zhang, B.; Zu, J.; Zhang, H.; Ding, M.; Paudel, B. Increasing sensitivity of alpine grasslands to climate variability along an elevational gradient on the Qinghai-Tibet Plateau. Sci. Total Environ. 2019, 678, 21–29. [Google Scholar] [CrossRef]
  11. Cai, Z.; Song, P.; Wang, J.; Jiang, F.; Liang, C.; Zhang, J.; Gao, H.; Zhang, T. Grazing pressure index considering both wildlife and livestock in Three-River Headwaters, Qinghai-Tibetan Plateau. Ecol. Indic. 2022, 143, 109338. [Google Scholar] [CrossRef]
  12. Li, X.; Lyu, X.; Dou, H.; Dang, D.; Li, S.; Li, X.; Li, M.; Xuan, X. Strengthening grazing pressure management to improve grassland ecosystem services. Glob. Ecol. Conserv. 2021, 31, e01782. [Google Scholar] [CrossRef]
  13. Matthew, C. A note on the reporting of stocking rate, stocking density, and “grazing intensity” in pasture and rangeland research. Grassl. Res. 2024, 3, 303–305. [Google Scholar] [CrossRef]
  14. Shi, Y.; Brierley, G.; Perry, G.L.W.; Gao, J.; Li, X.; Prishchepov, A.V.; Li, J.; Han, M. An Improved Approach to Estimate Stocking Rate and Carrying Capacity Based on Remotely Sensed Phenology Timings. Remote Sens. 2024, 16, 1991. [Google Scholar] [CrossRef]
  15. Yang, Y.; Zhao, D.; Chen, H. Quantifying the ecological carrying capacity of alpine grasslands on the Qinghai-Tibet Plateau. Ecol. Indic. 2022, 136, 108634. [Google Scholar] [CrossRef]
  16. Zhang, X.; Li, M.; Wu, J.; He, Y.; Niu, B. Alpine Grassland Aboveground Biomass and Theoretical Livestock Carrying Capacity on the Tibetan Plateau. J. Resour. Ecol. 2022, 13, 129–141. [Google Scholar] [CrossRef]
  17. Zhang, J.; Zhang, L.; Liu, W.; Qi, Y.; Wo, X. Livestock-carrying capacity and overgrazing status of alpine grassland in the Three-River Headwaters region, China. J. Geogr. Sci. 2014, 24, 303–312. [Google Scholar] [CrossRef]
  18. Ma, Q.; Chai, L.; Hou, F.; Chang, S.; Ma, Y.; Tsunekawa, A.; Cheng, Y. Quantifying Grazing Intensity Using Remote Sensing in Alpine Meadows on Qinghai-Tibetan Plateau. Sustainability 2019, 11, 417. [Google Scholar] [CrossRef]
  19. Saizen, I.; Maekawa, A.; Yamamura, N. Spatial analysis of time-series changes in livestock distribution by detection of local spatial associations in Mongolia. Appl. Geogr. 2010, 30, 639–649. [Google Scholar] [CrossRef]
  20. Wang, D.; Peng, Q.; Li, X.; Zhang, W.; Xia, X.; Qin, Z.; Ren, P.; Liang, S.; Yuan, W. A long-term high-resolution dataset of grasslands grazing intensity in China. Sci. Data 2024, 11, 1194. [Google Scholar] [CrossRef]
  21. Zhao, L.; Liu, Z.; Hu, Y.; Zhou, W.; Peng, Y.; Ma, T.; Liu, L.; Li, S.; Wang, L.; Mao, X. Evaluation of Reasonable Stocking Rate Based on the Relative Contribution of Climate Change and Grazing Activities to the Productivity of Alpine Grasslands in Qinghai Province. Remote Sens. 2022, 14, 1455. [Google Scholar] [CrossRef]
  22. Török, P.; Penksza, K.; Tóth, E.; Kelemen, A.; Sonkoly, J.; Tóthmérész, B. Vegetation type and grazing intensity jointly shape grazing effects on grassland biodiversity. Ecol. Evol. 2018, 8, 10326–10335. [Google Scholar] [CrossRef] [PubMed]
  23. Ahmed, A.I.; Hou, L.; Yan, R.; Xin, X.; Zainelabdeen, Y.M. The joint effect of grazing intensity and soil factors on aboveground net primary production in Hulunber grasslands meadow steppe. Agriculture 2020, 10, 263. [Google Scholar] [CrossRef]
  24. Schönbach, P.; Wan, H.; Gierus, M.; Bai, Y.; Müller, K.; Lin, L.; Susenbeth, A.; Taube, F. Grassland responses to grazing: Effects of grazing intensity and management system in an Inner Mongolian steppe ecosystem. Plant Soil. 2011, 340, 103–115. [Google Scholar] [CrossRef]
  25. Hu, L.-J.; Wang, W.; Cheng, Y.; Guo, Y. Effects of grazing livestock on grassland functioning may depend more on grazing intensity than livestock diversity. Proc. Natl. Acad. Sci. USA 2019, 116, 18762–18763. [Google Scholar] [CrossRef]
  26. Wang, M.J.; Sun, R.; Liu, Z.; Xin, X.P.; Liu, G.; Zhang, L.; Qiao, C. A study of grazing intensity in the Hulunbuir grasslands using remote sensing. Acta Prataculturae Sin. 2017, 26, 28–36. [Google Scholar] [CrossRef]
  27. Li, F.; Bai, Y.; Wan, H.; Zheng, J.; Luo, J.; Zhao, D.; Liu, P. Quantifying Grazing Intensity in China Using High Temporal Resolution MODIS Data. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2017, 10, 515–523. [Google Scholar] [CrossRef]
  28. Li, F.; Zheng, J.; Wang, H.; Luo, J.; Zhao, Y.; Zhao, R. Mapping grazing intensity using remote sensing in the Xilingol steppe region, Inner Mongolia, China. Remote Sens. Lett. 2016, 7, 328–337. [Google Scholar] [CrossRef]
  29. Lei, G.; Li, A.; Zhang, Z.; Bian, J.; Hu, G.; Wang, C.; Nan, X.; Wang, J.; Tan, J.; Liao, X. The Quantitative Estimation of Grazing Intensity on the Zoige Plateau Based on the Space-Air-Ground Integrated Monitoring Technology. Remote Sens. 2020, 12, 1399. [Google Scholar] [CrossRef]
  30. Cao, J.; Li, Y.; Biswas, A.; Holden, N.M.; Adamowski, J.F.; Wang, F.; Hong, S.; Qin, Y. Grassland biomass allocation across continents and grazing practices and its response to climate and altitude. Agr. Forest Meteorol. 2024, 356, 110176. [Google Scholar] [CrossRef]
  31. Milchunas, D.G.; Lauenroth, W.K. Quantitative effects of grazing on vegetation and soils over a global range of environments: Ecological Archives M063-001. Ecol. Monogr. 1993, 63, 327–366. [Google Scholar] [CrossRef]
  32. Vallentine, J.F. Grazing Management; Elsevier: Amsterdam, The Netherlands, 2000. [Google Scholar]
  33. Hamelin, L.; Borzęcka, M.; Kozak, M.; Pudełko, R. A spatial approach to bioeconomy: Quantifying the residual biomass potential in the EU-27. Renew. Sustain. Energy Rev. 2019, 100, 127–142. [Google Scholar] [CrossRef]
  34. Liu, S.; Zamanian, K.; Schleuss, P.-M.; Zarebanadkouki, M.; Kuzyakov, Y. Degradation of Tibetan grasslands: Consequences for carbon and nutrient cycles. Agric. Ecosyst. Environ. 2018, 252, 93–104. [Google Scholar] [CrossRef]
  35. Che, M.; Gong, Y.; Xu, M.; Kang, C.; Lv, C.; He, S.; Zheng, J. Effects of elevation and slope aspect on the distribution of the soil organic carbon associated with Al and Fe mineral phases in alpine shrub–meadow soil. Sci. Total Environ. 2021, 753, 141933. [Google Scholar] [CrossRef] [PubMed]
  36. Ge, J.; Hou, M.; Liang, T.; Feng, Q.; Meng, X.; Liu, J.; Bao, X.; Gao, H. Spatiotemporal dynamics of grassland aboveground biomass and its driving factors in North China over the past 20 years. Sci. Total Environ. 2022, 826, 154226. [Google Scholar] [CrossRef]
  37. Bretas, I.L.; Valente, D.S.M.; Silva, F.F.; Chizzotti, M.L.; Paulino, M.F.; D’Áurea, A.P.; Paciullo, D.S.C.; Pedreira, B.C.; Chizzotti, F.H.M. Prediction of aboveground biomass and dry-matter content in Brachiaria pastures by combining meteorological data and satellite imagery. Grass Forage Sci. 2021, 76, 340–352. [Google Scholar] [CrossRef]
  38. Liu, Y.; Zha, Y.; Gao, J.; Ni, S. Assessment of grassland degradation near Lake Qinghai, West China, using Landsat TM and in situ reflectance spectra data. Int. J. Remote Sens. 2004, 25, 4177–4189. [Google Scholar] [CrossRef]
  39. Marsett, R.C.; Qi, J.G.; Heilman, P.; Biedenbender, S.H.; Watson, M.C.; Amer, S.; Weltz, M.; Goodrich, D.; Marsett, R. Remote sensing for grassland management in the arid Southwest. Rangel. Ecol. Manag. 2006, 59, 530–540. [Google Scholar] [CrossRef]
  40. Zhu, L. Remote Sensing Based Vegetation Dynamics in Southern Africa: Physiographic Gradients Determine the Relative Importance of Environmental Controls on Savanna Vegetation. Ph.D. Thesis, University of Florida, Ann Arbor, MI, USA, 2014. [Google Scholar]
  41. Shi, Y.; Gao, J.; Li, X.; Li, J.; dela Torre, D.M.G.; Brierley, G.J. Improved Estimation of Aboveground Biomass of Disturbed Grassland through Including Bare Ground and Grazing Intensity. Remote Sens. 2021, 13, 2105. [Google Scholar] [CrossRef]
  42. Yu, H.; Wu, Y.; Niu, L.; Chai, Y.; Feng, Q.; Wang, W.; Liang, T. A method to avoid spatial overfitting in estimation of grassland above-ground biomass on the Tibetan Plateau. Ecol. Indic. 2021, 125, 107450. [Google Scholar] [CrossRef]
  43. Zhou, W.; Li, H.R.; Xie, L.J.; Nie, X.M.; Wang, Z.; Du, Z.P.; Yue, T.X. Remote sensing inversion of grassland aboveground biomass based on high accuracy surface modeling. Ecol. Indic. 2021, 121, 107215. [Google Scholar] [CrossRef]
  44. Zhang, J.; Fang, S.; Liu, H. Estimation of alpine grassland above-ground biomass and its response to climate on the Qinghai-Tibet Plateau during 2001 to 2019. Glob. Ecol. Conserv. 2022, 35, e02065. [Google Scholar] [CrossRef]
  45. Ali, I.; Cawkwell, F.; Dwyer, E.; Barrett, B.; Green, S. Satellite remote sensing of grasslands: From observation to management. J. Plant Ecol. 2016, 9, 649–671. [Google Scholar] [CrossRef]
  46. Sharma, P.; Leigh, L.; Chang, J.; Maimaitijiang, M.; Caffé, M. Above-Ground Biomass Estimation in Oats Using UAV Remote Sensing and Machine Learning. Sensors 2022, 22, 601. [Google Scholar] [CrossRef] [PubMed]
  47. Morais, T.G.; Teixeira, R.F.M.; Figueiredo, M.; Domingos, T. The use of machine learning methods to estimate aboveground biomass of grasslands: A review. Ecol. Indic. 2021, 130, 108081. [Google Scholar] [CrossRef]
  48. Yang, C.; Zhang, Y.; Hou, F.; Millner, J.P.; Wang, Z.; Chang, S. Grazing activity increases decomposition of yak dung and litter in an alpine meadow on the Qinghai-Tibet plateau. Plant Soil. 2019, 444, 239–250. [Google Scholar] [CrossRef]
  49. Wei, D.; Zhao, H.; Zhang, J.; Qi, Y.; Wang, X. Human activities alter response of alpine grasslands on Tibetan Plateau to climate change. J. Environ. Manag. 2020, 262, 110335. [Google Scholar] [CrossRef]
  50. Wang, P.; Wang, Z.; Zhang, X.; Feng, Q.; Jiu, C.; Chen, Q. GIS-based classification of seasonal pasture in Qinghai province. Pratacult. Sci. 2010, 27, 119–128. [Google Scholar]
  51. Smart, A.J.; Derner, J.D.; Hendrickson, J.R.; Gillen, R.L.; Dunn, B.H.; Mousel, E.M.; Johnson, P.S.; Gates, R.N.; Sedivec, K.K.; Harmoney, K.R.; et al. Effects of Grazing Pressure on Efficiency of Grazing on North American Great Plains Rangelands. Rangel. Ecol. Manag. 2010, 63, 397–406. [Google Scholar] [CrossRef]
  52. Cao, Y.; Wu, J.; Zhang, X.; Niu, B.; He, Y. Comparison of methods for evaluating the forage-livestock balance of alpine grasslands on the northern Tibetan Plateau. J. Resour. Ecol. 2020, 11, 272–282. [Google Scholar] [CrossRef]
  53. Shi, Y.; Gao, J.; Brierley, G.; Li, X.; Perry, G.L.W.; Xu, T. Improving the accuracy of models to map alpine grassland above-ground biomass using Google earth engine. Grass Forage Sci. 2023, 78, 237–253. [Google Scholar] [CrossRef]
  54. Zhang, W.; Du, T.; Wang, J. Deep learning over multi-field categorical data. In Proceedings of the European Conference on Information Retrieval, Padua, Italy, 20–23 March 2016; pp. 45–57. [Google Scholar]
  55. Wei, P.; Xu, L.; Pan, X.; Hu, Q.; Li, Q.; Zhang, X.; Shao, C.; Wang, C.; Wang, X. Spatio-temporal variations in vegetation types based on a climatic grassland classification system during the past 30 years in Inner Mongolia, China. CATENA 2020, 185, 104298. [Google Scholar] [CrossRef]
  56. Wang, J.; Li, W.; Cao, W.; Wang, S. Effects of different intensities of long-term grazing on plant diversity, biomass and carbon stock in alpine shrubland on the Qinghai-Tibetan Plateau. PeerJ 2022, 10, e12771. [Google Scholar] [CrossRef]
  57. Jasiewicz, J.; Stepinski, T.F. Geomorphons—A pattern recognition approach to classification and mapping of landforms. Geomorphology 2013, 182, 147–156. [Google Scholar] [CrossRef]
  58. Brierley, G.; Li, X.; Fryirs, K.; Gao, J.; Shi, Y.; Perry, G.L.W.; Cullum, C. Development of place-based catenal models for grassland ecosystems of the Upper Yellow River, Western China. Catena 2022, 213, 106193. [Google Scholar] [CrossRef]
  59. Brierley, G.; Li, X.-l.; Fryirs, K.; Gao, J.; Shi, Y.; Perry, G.L.W.; Cullum, C. Degradation and recovery of alpine meadow catenas in the source zone of the Yellow River, Western China. J. Mt. Sci. 2022, 19, 2487–2505. [Google Scholar] [CrossRef]
  60. Slocum, T.A.; McMaster, R.B.; Kessler, F.C.; Howard, H.H. Thematic Cartography and Geovisualization; CRC Press: Boca Raton, FL, USA, 2022. [Google Scholar]
  61. Sui, D.Z. A Review of “Geospatial Analysis: A Comprehensive Guide to Principles, Techniques and Software Tools” Michael J. de Smith, Michael F. Goodchild, and Paul A. Longley. Leicester, UK: Troubador Publishing, 2007. xxii and 491 pp., maps, diagrams, notes, appendices, and index. $69.95 paper (ISBN 978-1-906221-980). Ann. Assoc. Am. Geogr. 2009, 99, 440–442. [Google Scholar]
  62. Scott, L.M.; Janikas, M.V. Spatial statistics in ArcGIS. In Handbook of Applied Spatial Analysis: Software Tools, Methods and Applications; Springer: Berlin/Heidelberg, Germany, 2009; pp. 27–41. [Google Scholar]
  63. Turner, M.G. Landscape ecology: The effect of pattern on process. Annu. Rev. Ecol. Syst. 1989, 20, 171–197. [Google Scholar] [CrossRef]
  64. Fuhlendorf, S.D.; Engle, D.M. Restoring heterogeneity on rangelands: Ecosystem management based on evolutionary grazing patterns: We propose a paradigm that enhances heterogeneity instead of homogeneity to promote biological diversity and wildlife habitat on rangelands grazed by livestock. BioScience 2001, 51, 625–632. [Google Scholar]
  65. Adler, P.; Raff, D.; Lauenroth, W. The effect of grazing on the spatial heterogeneity of vegetation. Oecologia 2001, 128, 465–479. [Google Scholar] [CrossRef]
  66. Fetzel, T.; Havlik, P.; Herrero, M.; Erb, K.-H. Seasonality constraints to livestock grazing intensity. Glob. Change Biol. 2017, 23, 1636–1647. [Google Scholar] [CrossRef]
  67. Pettorelli, N.; Vik, J.O.; Mysterud, A.; Gaillard, J.-M.; Tucker, C.J.; Stenseth, N.C. Using the satellite-derived NDVI to assess ecological responses to environmental change. Trends Ecol. Evol. 2005, 20, 503–510. [Google Scholar] [CrossRef]
  68. Evans, J.; Geerken, R. Discrimination between climate and human-induced dryland degradation. J. Arid Environ. 2004, 57, 535–554. [Google Scholar] [CrossRef]
  69. Zhang, X.; Friedl, M.A.; Schaaf, C.B.; Strahler, A.H.; Hodges, J.C.; Gao, F.; Reed, B.C.; Huete, A. Monitoring vegetation phenology using MODIS. Remote Sens. Environ. 2003, 84, 471–475. [Google Scholar] [CrossRef]
  70. Eldridge, D.J.; Poore, A.G.; Ruiz-Colmenero, M.; Letnic, M.; Soliveres, S. Ecosystem structure, function, and composition in rangelands are negatively affected by livestock grazing. Ecol. Appl. 2016, 26, 1273–1283. [Google Scholar] [CrossRef] [PubMed]
  71. Scurlock, J.M.O.; Johnson, K.; Olson, R.J. Estimating net primary productivity from grassland biomass dynamics measurements. Glob. Change Biol. 2002, 8, 736–753. [Google Scholar] [CrossRef]
  72. Sala, O.E.; Austin, A.T. Methods of estimating aboveground Net Primary Productivity. In Methods in Ecosystem Science; Sala, O.E., Jackson, R.B., Mooney, H.A., Howarth, R.W., Eds.; Springer: New York, NY, USA, 2000; pp. 31–43. [Google Scholar]
  73. Lauenroth, W.K.; Hunt, H.W.; Swift, D.M.; Singh, J.S. Estimating aboveground net primary production in grasslands: A simulation approach. Ecol. Model. 1986, 33, 297–314. [Google Scholar] [CrossRef]
  74. Wang, Y.; Geng, Y.; Wang, X.; Xu, T.; Zhang, Q.; Liu, H.; Zhao, N.; Xu, S. Seasonal Herbage–Livestock Balance and Grassland Pressure Index Analysis in the Yellow River Source Park, Tibet Plateau, China. Ecol. Evol. 2024, 14, e70586. [Google Scholar] [CrossRef]
  75. Li, W.; Hooper, D.U.; Wu, L.; Bakker, J.D.; Gianuca, A.T.; Wu, X.B.; Taube, F.; Wang, C.; Bai, Y. Grazing regime alters plant community structure via patch-scale diversity in semiarid grasslands. Ecosphere 2021, 12, e03547. [Google Scholar] [CrossRef]
  76. Wen, L.; Dong, S.K.; Li, Y.Y.; Sherman, R.; Shi, J.J.; Liu, D.M.; Wang, Y.L.; Ma, Y.S.; Zhu, L. The effects of biotic and abiotic factors on the spatial heterogeneity of alpine grassland vegetation at a small scale on the Qinghai–Tibet Plateau (QTP), China. Environ. Monit. Assess. 2013, 185, 8051–8064. [Google Scholar] [CrossRef]
  77. Li, Y.; Li, F.Y.; Shi, C.; Wang, H.; Wu, L.; Wang, Y.; Minggagud, H. Precipitation and grazing intensity jointly shape plant compensatory growth and productivity in a semi-arid steppe ecosystem. Agric. Ecosyst. Environ. 2025, 393, 109834. [Google Scholar] [CrossRef]
  78. McNaughton, S.J. Compensatory plant growth as a response to herbivory. Oikos 1983, 40, 329–336. [Google Scholar] [CrossRef]
  79. Milchunas, D.; Lauenroth, W. Three-dimensional distribution of plant biomass in relation to grazing and topography in the shortgrass steppe. Oikos 1989, 55, 82–86. [Google Scholar] [CrossRef]
  80. Yao, Y.; Ren, H. Estimation of grassland aboveground biomass in northern China based on topography-climate-remote sensing data. Ecol. Indic. 2024, 165, 112230. [Google Scholar] [CrossRef]
  81. Teng, Y.; Zhan, J.; Agyemang, F.B.; Sun, Y. The effects of degradation on alpine grassland resilience: A study based on meta-analysis data. Glob. Ecol. Conserv. 2020, 24, e01336. [Google Scholar] [CrossRef]
  82. Vogel, A.; Scherer-Lorenzen, M.; Weigelt, A. Grassland resistance and resilience after drought depends on management intensity and species richness. PLoS ONE 2012, 7, e36992. [Google Scholar] [CrossRef] [PubMed]
  83. Chang, C.; Wang, J.; Zhao, Y.; Cai, T.; Yang, J.; Zhang, G.; Wu, X.; Otgonbayar, M.; Xiao, X.; Xin, X.; et al. A 10-m annual grazing intensity dataset in 2015–2021 for the largest temperate meadow steppe in China. Sci. Data 2024, 11, 181. [Google Scholar] [CrossRef]
  84. Pringle, H.J.; Landsberg, J. Predicting the distribution of livestock grazing pressure in rangelands. Austral. Ecol. 2004, 29, 31–39. [Google Scholar] [CrossRef]
  85. Li, Y.; Dong, S.; Gao, Q.; Zhang, Y.; Liu, S.; Ganjurjav, H.; Hu, G.; Wang, X.; Yan, Y.; Wu, H.; et al. Rotational grazing promotes grassland aboveground plant biomass and its temporal stability under changing weather conditions on the Qinghai-Tibetan plateau. Land Degrad. Dev. 2020, 31, 2662–2671. [Google Scholar] [CrossRef]
  86. Yu, H.Q. The Impact of Different Rotational Grazing Systems on the Physicochemical Properties of Desert Grassland Soils and Carbon Balance. Master’s Thesis, Ningxia University, Yinchuan, China, 2014. [Google Scholar]
  87. Briske, D.D.; Derner, J.; Brown, J.; Fuhlendorf, S.D.; Teague, W.; Havstad, K.; Gillen, R.L.; Ash, A.J.; Willms, W. Rotational grazing on rangelands: Reconciliation of perception and experimental evidence. Rangel. Ecol. Manag. 2008, 61, 3–17. [Google Scholar] [CrossRef]
  88. Teague, W.; Dowhower, S.; Baker, S.; Haile, N.; DeLaune, P.; Conover, D. Grazing management impacts on vegetation, soil biota and soil chemical, physical and hydrological properties in tall grass prairie. Agric. Ecosyst. Environ. 2011, 141, 310–322. [Google Scholar] [CrossRef]
  89. Jia, W.; Liu, M.; Wang, D.; He, H.; Shi, P.; Li, Y.; Wang, Y. Uncertainty in simulating regional gross primary productivity from satellite-based models over northern China grassland. Ecol. Indic. 2018, 88, 134–143. [Google Scholar] [CrossRef]
  90. Jia, W.X.; Liu, M.; Yang, Y.H.; He, H.L.; Zhu, X.D.; Yang, F.; Yin, C.; Xiang, W.N. Estimation and uncertainty analyses of grassland biomass in Northern China: Comparison of multiple remote sensing data sources and modeling approaches. Ecol. Indic. 2016, 60, 1031–1040. [Google Scholar] [CrossRef]
Remotesensing 17 03781 g001
Figure 1. Location and landforms of the study area and the distribution of sample sites. (a) Location of Yellow River Source Zone on the Qinghai Tibetan Plateau. (b) Location of study area. (c) Distribution of sample sites and landform types within the study area, along with the area percentage represented by each landform category (percentages below 5% are not shown).
Figure 1. Location and landforms of the study area and the distribution of sample sites. (a) Location of Yellow River Source Zone on the Qinghai Tibetan Plateau. (b) Location of study area. (c) Distribution of sample sites and landform types within the study area, along with the area percentage represented by each landform category (percentages below 5% are not shown).
Remotesensing 17 03781 g001
Remotesensing 17 03781 g002
Figure 2. Workflow of this study. (a) step-by-step derivation for grazing pressure index. (b) modelling processes for the components of grazing pressure index. (c) Spatial analysis. GI—grazing intensity, i—number of observed periods, MLR—multiple linear regression, SVM—support vector machine, ANN—artificial neural network, and DNN—deep neural network.
Figure 2. Workflow of this study. (a) step-by-step derivation for grazing pressure index. (b) modelling processes for the components of grazing pressure index. (c) Spatial analysis. GI—grazing intensity, i—number of observed periods, MLR—multiple linear regression, SVM—support vector machine, ANN—artificial neural network, and DNN—deep neural network.
Remotesensing 17 03781 g002
Remotesensing 17 03781 g003
Figure 3. Scatterplots of measured and predicted grazing intensity. (a) MLR—multiple linear regression, (b) SVM—support vector machine, (c) ANN—artificial neural network, and (d) DNN—deep neural network. Fitted curve confidence interval is 90%.
Figure 3. Scatterplots of measured and predicted grazing intensity. (a) MLR—multiple linear regression, (b) SVM—support vector machine, (c) ANN—artificial neural network, and (d) DNN—deep neural network. Fitted curve confidence interval is 90%.
Remotesensing 17 03781 g003
Remotesensing 17 03781 g004
Figure 4. Grazing intensity during the growing season. (a–d) GI maps produced by DNN model, 12 June–27 July (a), 27 July–16 Aug (b), 16 Aug–5 Sep (c) and growing season (d), along with the boxplot of GI in different landforms and pie chart of GI levels’ coverage (percentages below 5% are not shown). Uppercase letters indicate significant differences at p < 0.001; lowercase letters indicate the significance level of p < 0.05.
Figure 4. Grazing intensity during the growing season. (a–d) GI maps produced by DNN model, 12 June–27 July (a), 27 July–16 Aug (b), 16 Aug–5 Sep (c) and growing season (d), along with the boxplot of GI in different landforms and pie chart of GI levels’ coverage (percentages below 5% are not shown). Uppercase letters indicate significant differences at p < 0.001; lowercase letters indicate the significance level of p < 0.05.
Remotesensing 17 03781 g004
Remotesensing 17 03781 g005
Figure 5. Scatter plots of measured and predicted aboveground biomass using four methods. (a) MLR—multiple linear regression, (b) SVM—support vector machine, (c) ANN—artificial neural network, (d) DNN—deep neural network. Fitted curve confidence interval is 90%.
Figure 5. Scatter plots of measured and predicted aboveground biomass using four methods. (a) MLR—multiple linear regression, (b) SVM—support vector machine, (c) ANN—artificial neural network, (d) DNN—deep neural network. Fitted curve confidence interval is 90%.
Remotesensing 17 03781 g005
Remotesensing 17 03781 g006
Figure 6. Availability and variability of residual biomass. (a–c) Vegetation type map, area proportion of each vegetation type, and distribution of vegetation types across different landforms; (d,e) Maps of total aboveground biomass (produced by DNN model) and its distribution by vegetation type and landform; (f,g) Maps showing available aboveground biomass and its corresponding distribution by vegetation type and landform. Uppercase letters indicate significant differences at p < 0.001; lowercase letters indicate significance at p < 0.05.
Figure 6. Availability and variability of residual biomass. (a–c) Vegetation type map, area proportion of each vegetation type, and distribution of vegetation types across different landforms; (d,e) Maps of total aboveground biomass (produced by DNN model) and its distribution by vegetation type and landform; (f,g) Maps showing available aboveground biomass and its corresponding distribution by vegetation type and landform. Uppercase letters indicate significant differences at p < 0.001; lowercase letters indicate significance at p < 0.05.
Remotesensing 17 03781 g006
Remotesensing 17 03781 g007
Figure 7. Grazing pressure index map derived from grazing intensity and residual biomass. (a) Spatial distribution map of GPI; (b) GPI in relation to different landform positions and vegetation types; (c) Coverage ratio of different GPI levels. Percentages below 1% are not displayed in the pie charts.
Figure 7. Grazing pressure index map derived from grazing intensity and residual biomass. (a) Spatial distribution map of GPI; (b) GPI in relation to different landform positions and vegetation types; (c) Coverage ratio of different GPI levels. Percentages below 1% are not displayed in the pie charts.
Remotesensing 17 03781 g007
Remotesensing 17 03781 g008
Figure 8. Heatmap illustrating the overlap ratio between grazing pressure, grazing intensity and residual biomass. (a,b) Grazing pressure overlapped with grazing intensity and residual biomass; (c,d) Grazing intensity overlapped with grazing pressure and residual biomass.
Figure 8. Heatmap illustrating the overlap ratio between grazing pressure, grazing intensity and residual biomass. (a,b) Grazing pressure overlapped with grazing intensity and residual biomass; (c,d) Grazing intensity overlapped with grazing pressure and residual biomass.
Remotesensing 17 03781 g008
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Shi, Y.; Gao, J.; Brierley, G.; Li, X.; He, J.-S. Estimating Grazing Pressure from Satellite Time Series Without Reliance on Total Production. Remote Sens. 2025, 17, 3781. https://doi.org/10.3390/rs17223781

AMA Style

Shi Y, Gao J, Brierley G, Li X, He J-S. Estimating Grazing Pressure from Satellite Time Series Without Reliance on Total Production. Remote Sensing. 2025; 17(22):3781. https://doi.org/10.3390/rs17223781

Chicago/Turabian Style

Shi, Yan, Jay Gao, Gary Brierley, Xilai Li, and Jin-Sheng He. 2025. "Estimating Grazing Pressure from Satellite Time Series Without Reliance on Total Production" Remote Sensing 17, no. 22: 3781. https://doi.org/10.3390/rs17223781

APA Style

Shi, Y., Gao, J., Brierley, G., Li, X., & He, J.-S. (2025). Estimating Grazing Pressure from Satellite Time Series Without Reliance on Total Production. Remote Sensing, 17(22), 3781. https://doi.org/10.3390/rs17223781

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop