We collected peer-reviewed journal articles published before July 2021 from Web of Science and the China Knowledge Resource Integrated Database by using the following search term combinations: “grazing exclusion,” “fence,” “alpine meadow/steppe,” “Qinghai-Tibet/Tibetan Plateau.” Several criteria were set to avoid bias in paper selection: (1) the experimental data should be obtained from field studies with no other treatments (e.g., warming, precipitation change, or nutrient enrichment); (2) GE must be of a duration with at least 1 year; (3) The values without GE is necessarily included as the “reference” to enable the “effect size” analysis; (4) The experimental data in the publications had at least one of the following variables: Plant aboveground biomass (AGB), belowground biomass (BGB), species richness (SR), Shannon-Wiener index (Supplementary Table 1); (5) The mean, standard error or standard deviation, and samples size of selected variables could be acquired directly from tables, digitized graphs, or the body of the text. In total, 43 published papers (Supplementary Table 2) that conformed with these criteria were selected, covering 60 study sites (Figure 1). The GetData Graph Digitizer software (ver.2.25)¹ was used to extract data from figures in the papers. Additionally, information of study sites on grassland type, longitude, and latitude were also collected. Monthly mean precipitation (MMP) were obtained at 30-s (∼1 km²) spatial resolution from 2000 to 2020. These data were provided by National Earth System Science Data Center, National Science and Technology Infrastructure of China.² We extracted the climate data by the Kriging interpolation method using ArcGIS 10.2 (ESRI, Inc., Redlands, CA, United States) for each of sample plots according to their geographic coordinates (latitude and longitude). Mean annual precipitation (MAP) are further calculated by MMP.

Additionally, we selected 9 typical sites in a large-scale transect with different GE duration in both alpine meadows and alpine steppes and conducted the field surveys at the peak growing season in 2020. The information on duration of GE, location, and grassland type in each site were identified and recorded. We divided the duration of GE into short (1–4 years), medium (5–8 years), and long (> 8 years) term according to previous studies (Sun et al., 2020a). Three 50 cm × 50 cm quadrats were randomly selected as replicates for vegetation investigation and sampling inside and outside the fence, respectively. All plant species within each quadrat were identified and their coverage, number, and height were measured and recorded. Then, we clipped the aboveground parts of plants with scissors to determine the aboveground biomass. The belowground parts of plants (soil depth ≤ 30 cm) were obtained from 3 soil cores of 5 cm diameter which were washed to remove residual soil and gain roots with a 0.5 mm sieve. The above- and belowground biomass were determined as constant weights by oven-drying the plant samples at 65°C for 72 h.

The Shannon-Wiener diversity index (H) was calculated as follows (Ma et al., 1995):

where S represents the total number of plant species in the plot, P_i represents the relative importance value of species i, calculated as follows (Wang et al., 2014):

where RH, RC, and RB represent height, cover, and biomass of species i divided by sum of those for all species, respectively.

We conducted the meta-analysis to determine whether GE with different durations had a significant effect by using the MetaWin 2.1 software (Sinauer Associates Inc., Sunderland, MA, United States) (Hedges et al., 1999; Zhan et al., 2020), in which the response ratio (RR) of response variables was calculated to assess the effects of GE as follows:

where X¯t and X¯c represent the arithmetic mean concentrations of the concerned variable in GE group (treatment) and that without GE (control), respectively.

The variance (v) of the RR was estimated by:

where n_t and S_t represent the sample sizes and the standard deviations of the concerned variable in the treatment group, respectively; n_c and S_c are the sample sizes and the standard deviations of the concerned variable the control group, respectively. The reciprocal of the variance (w=1v) is used as the weight (W) for each RR. The mean response ratio (RR₊₊) is calculated from RR values of individual pairwise comparisons between the control and treatment group, RR_ij (i = 1,2,3…,m; j = 1,2,3…,k) as follows:

where m is the number of treatment groups, and k is the number of comparisons in the corresponding control group. The bias-corrected 95% confidence intervals (CIs) were calculated via a bootstrapping procedure (5,000 iterations) for each categorical group (Guo and Gifford, 2002). It indicates a statistically significant response of the CIs of the RR₊₊ did not overlap zero (Zhan et al., 2020). The standard error of RR₊₊ was estimated by the following equation:

The vegetation restoration efficiency of GE was illustrated by the percentage of mean annual changes in above- and belowground biomass, species richness, and Shannon-Wiener index (ΔV% yr^–1) due to GE, which were calculated using the following equations:

where V_GE and V_FG represent the values of objective variables (above- and belowground biomass, species richness, and Shannon-Wiener index) with and without GE, respectively. Y is the duration of GE.

linear-regression analysis were performed to explore the relationships between the percentage of mean annual changes (i.e., vegetation restoration efficiency) in above- and below-ground biomass, and species richness with GE duration and precipitation. The linear-regression analysis were carried out with R version 3.3.2 software (R Core Team, 2016).

The aboveground biomass significantly increased with short-, medium-, and long-term GE in alpine meadows with the mean response ratios of 0.22, 0.19, and 0.28, respectively (P < 0.05, Figures 2A–C), while only increased with medium-term GE in alpine steppes with the mean response ratio of 0.25 (P < 0.05, Figures 2D–F). Short-term GE significantly increased belowground biomass with the mean response ratio of 0.32 in alpine steppes (P < 0.05, Figure 2D), while showed no significant effect on that in alpine meadows (P > 0.05, Figure 2A) where belowground biomass was significantly increased by medium- and long-term GE, with the mean response ratios of 0.12 and 0.09, respectively (P < 0.05, Figures 2B,C). Species richness showed a slight (no-significant) increase with short-term GE, while significantly increased by medium- and long-term GE in alpine meadows with the mean response ratios of 0.12 and 0.16, respectively (P < 0.05, Figures 2B,C). By contrast, species richness was significantly increased with all durations of GE in alpine steppes, and the mean response ratios were 0.03, 0.18, and 0.20 for short-, medium-, and long-term GE, respectively (P < 0.05, Figures 2D–F). Moreover, there were increases in Shannon-Wiener index in alpine meadows with short- and long-term GE and alpine steppes with long-term GE.

The percentage of mean annual changes in aboveground biomass (Figure 3) was significantly negatively related to GE duration in both alpine meadows (R² = 0.06; P < 0.001; Figure 3A) and alpine steppes (R² = 0.20; P < 0.001; Figure 3E). Additionally, there was also a significantly negative relationship between the percentage of mean annual changes in Shannon-Wiener index with GE duration in alpine meadows (R² = 0.05; P < 0.001; Figure 3D), and a significantly negative relationship between the percentage of mean annual changes in species richness with GE duration in alpine steppes (R² = 0.07; P < 0.01; Figure 3G). Interestingly, the percentage of mean annual changes in above- and belowground biomass both showed a significantly positive relationship with MAP for short-term GE (P < 0.05; Figures 4A,B). By contrast, the percentage of mean annual changes in both species richness and Shannon-Wiener index were significantly negatively associated with MAP for long-term GE (P < 0.05; Figures 4C,D).

In this study, we found that short-term GE significantly increased plant aboveground biomass of alpine meadow, in consistent with previous studies (Shang et al., 2013; Liu et al., 2020b). It is mainly because that GE eliminates direct consumption of plant tissues by herbivore and thus facilitates a rapid restoration of vegetation after GE (Sun et al., 2018). The increase of plant biomass and litter material can enhance the input of organic matter into soil environment, reduce water loss caused by soil evaporation, and reduce soil erosion by improving soil surface roughness (Zhang et al., 2021). By contrast, aboveground biomass of alpine steppe exhibited no significant change with short-term GE, but an apparent increase with medium-term GE. The main reason might be that alpine meadow commonly experiences relatively higher precipitation than alpine steppe. Precipitation is tightly related to water availability for plant growth and has been proven a dominant factor driving vegetation restoration (Jia et al., 2015). Previous studies found that high precipitation greatly promoted grassland productivity (Liu et al., 2020a; Sun et al., 2020b). It explains that short-term GE significantly increases aboveground biomass of alpine meadow but has no significant effect on that of alpine steppe, which apparently recovers after medium GE. These findings agree with previous studies that patterns of vegetation recovery in response to GE are highly associated with precipitation (Wu et al., 2014a,2017). Precipitation may be a key factor in the restoration process of alpine steppes. This was supported by our results, as we found a significantly positive relationship between the efficiency of vegetation recovery with precipitation in this study. However, this positive relationship weakens as GE duration increases, and turns no significance in long-term GE. It implies that precipitation mainly promotes a rapid production restoration in early stage of GE, while the effect gradually loses when vegetation restores to some extent.

During short-term GE, plant belowground biomass shows no significant change in alpine meadow, while significantly increases in alpine steppe. It might be explained by the optimal partitioning hypothesis that plants maximize growth rate by allocating photosynthate among organs responding to habitat condition (Bloom et al., 1985; Sun et al., 2021b). The soils are commonly of relatively high soil nutrient availability in early stage of GE due to the fertilization effect of grazing through feces and urine (Bowman et al., 2003). Therefore, plants allocate more biomass to the aboveground part for a better access of light resources under abundant soil water and nutrient supply in alpine meadow, while they would partition more photosynthate to the belowground portion to maximize water uptake in relatively arid alpine steppe in early stage of GE (Dukes et al., 2005; Sun et al., 2021b). Additionally, the continuous increase in belowground biomass during long-term GE in this study indicates an intensifying competition for soil water and nutrient in the high-productivity grassland after long-term restoration.

We found increases in species richness with short-term GE in both alpine meadow and steppe, in consistent numerus previous studies (Tang et al., 2016; Guo et al., 2019; Liu et al., 2020b; Zhang et al., 2021). The logic is that GE eliminates trampling damage by herbivore and favors seedling establishment due to the high density of buried seeds (Ma et al., 2013), which leads to appearance of lost species induced by overgrazing, and annual species or short-lived perennials with higher colonization capacity rapidly establish during short-term GE (Zhao et al., 2019; Zhang et al., 2021). Plant interspecific facilitation plays a dominant role in species colonization during the early stage of GE (Bonet, 2004). However, an aggravating interspecific competition happens as a result of resource limitation in the high-productivity grassland after long-term GE, so that slow-growing and long-lived species with higher competitive advantages suppress or even exclude the less competitive species (Der Wal et al., 2004; Borer et al., 2014; Liu et al., 2019). Thus, there are remarked decreases in species richness and Shannon-Wiener index in long-term GE as reported in numerus previous studies (Oba et al., 2001; Wu et al., 2009; Kelemen et al., 2013).

In this study, GE is well proved as an effective way for vegetation restoration indicated by the remarked improvement of productivity and species richness. However, the patterns of vegetation restoration with GE duration were diverse between alpine meadow and steppe. It suggests that the optimal GE duration should be different in different grassland types (Sun et al., 2020a). Given that vegetation of alpine meadow and steppe can restore to relatively high level with short- and medium-term GE, respectively, an appropriate use is advised after short-term GE in alpine meadow and medium GE in alpine steppe in order to avoid waste of grassland resource. Additionally, long-term GE reduces species richness since interspecific competition intensifies when productivity reaches environmental carrying capacity (Kenkel, 1988; Chesson, 2018). Moreover, we found that the restoring efficiency of vegetation restoration turns lower with GE duration increasing. According to these findings, we argue that GE should be ceased after a long period (> 8 years) because fences can be useful tools but only when they are transitional and impermanent (Sacha, 2020).

Plant diversity is very important for the development of animal husbandry, livelihood of herdsmen, and social production in the Tibetan Plateau. Curbing biodiversity loss is a cornerstone of Sustainable Development Goal 15 of the United Nations’ 2030 Agenda for Sustainable Development. In view of this, the maintenance of biodiversity through the removal of grazing exclusion should be encouraged. Transient grazing exclusion may make sense, but we need better understanding of their impact on biodiversity is needed.

In this study, restricted by experimental conditions, there was one plot for each of the treatments (grasslands with different years of GE), which may result in an excessively high pseudo-repetition rate. In addition, the effects of climate change were not considered, which may have played an important role in plant growth and community structure. These potential problems may limit assessment of the impact of grazing exclusion on grassland restoration in different years. Therefore, climate change and other factors should be considered on a long-term time scale in the future to continuously monitor the impact of grazing exclusion duration on restoration efficiency of degraded grassland.