The study was conducted in a degraded desert steppe of Choir in Govisumber Province, Central Mongolia (45°52′-46°59′N, 107°53′-109°07′E). Figure 1 shows the sampling locations for the field surveys, climate characteristics during the sampling period, and the structure of hanging grass fences in the region. The regional climate has considerable daily and annual temperature fluctuations and irregular precipitation, belonging to a continental climate. With a mean annual temperature of 2.0°C, the monthly temperature variations range from -9.7°C (coldest in January) to 21.2°C (warmest in July). The annual precipitation averages 150.8 mm, mainly falling during the vegetation growing season, which accounts for approximately 70% of the total annual precipitation. The mean annual wind speed is 3.9 m/s. This is attributed to the flat terrain and lack of steep mountains in the region. The soils are categorised as Kastanozem, a classification based on the World Reference Base for Soil Resources system, and distinguished by texture, including loam, sandy loam, and gravelly light loam (Li et al., 2006). The dominant plant species mainly consist of Stipa krylovii and Artemisia frigida, which are interspersed with Caragana shrubs. Spring is characterised by increased windiness, rising temperatures, and sporadic late precipitation (Fig. 1c). This results in arid ground surfaces and hinders vegetation growth. Fine particles from the ground surface are easily carried away by wind erosion, which is a natural driving force for the degradation of regional grasslands (Meng et al., 2020). However, the grazing activity in this region has steadily increased, reaching a grazing intensity of 78 heads/km² by 2021. Therefore, overgrazing has become the most impactful socioeconomic factor in the degradation of desert steppe (Yan et al., 2023).

In the early spring of 2019, a 20.0-hm² enclosure was established in the northern suburban grazing area (Fig. 1b), surrounded by fencing facilities (mesh size of 20.0 cm×7.0 cm) 1.8 m above the ground. During the grass-withering season, fencing effectively inhibited the spread of tumbleweeds. Subsequently, a substantial amount of tumbleweed residue hanging and accumulating on the fences perpendicular to the prevailing wind direction (northwest) formed hanging grass fences with a low density and sparse upper porosity of approximately 40% (measured in May 2019) (Fig. 1e). However, following the regreening period of the desert steppe, a distinct natural green belt emerged on the leeward side of the fences (Fig. 2). Meanwhile, the fences in other directions did not exhibit such a gradient of vegetation recovery.

To quantify the gradual transition between stable vegetation states on either side of the fencing facility, we conducted field surveys and collected vegetation and soil samples in August 2019 and August 2021 using the line transect method (Michels et al., 2017). We set up transects within each of the fencing facility segments using systematic random sampling, ensuring a minimum distance of 30.0 m between segments to account for spatial variability. A mixed effect model analysis was conducted, with the duration of GF and the presence of hanging grasses on the fences as fixed effects and the transect segments of the fencing facilities as random effects to avoid pseudo-replication.

This study was initiated in 2019 and marked the first year of enclosure with hanging grass fences (abbreviated as GF1; Fig. 2c). Three transects, each extending 22.0 m, were established on the leeward side of the enclosure area. Sample plots of 1.0 m×1.0 m in size were positioned starting 1.5 m from the fences, maintaining a regular interval of 5.0 m between plots. Five sample plots were randomly selected within the grazing area on the windward side to observe the preliminary effects of hanging grass fences on grazing exclusion. In 2021, marking the third year of enclosure with hanging grass fences (abbreviated as GF3; Fig. 2d), the green belt extended beyond 22.0 m. To monitor the edge effects of hanging grass fences, we extended the transects to 44.0 m. The sample plot centres ranged from 16.5 m on the windward side of the fences to 26.5 m on the leeward side, maintaining consistent plot sizes and spacing with those in 2019. Sand accumulation occurred at the fences on the side with hanging grasses, leading us to refine the area immediately adjacent to the fences by adding plots at 0.5 m from the fence edge. The third year of enclosure with non-hanging grass fences (abbreviated as NF3; Fig. 2f), which were aligned with the direction of the prevailing wind and lacked hanging grass, was utilized as the control group to investigate the effects of hanging grass fences on edges. Nine transects, comprising 92 sample plots, were established during the sampling process.

The taxa, coverage, number of plants, and average height of each taxon were measured for each sample plot in the field. We measured aboveground biomass by clipping the aboveground parts of the plants, drying them at 65.0°C, and subsequently weighing them. We used stainless-steel augers to collect composite samples from different positions (corners and centres) within the pre-surveyed vegetation plots. In 2019, topsoil samples were collected from the depth of 0-10 cm because of the high root content of grassland plants. This layer of soil is extremely vulnerable to environmental factors and grazing demands (Bell et al., 2011). However, in 2021, soil samples were collected from the depths of 0-10 and 10-30 cm after frequent wind and sand disturbances. The collected soil samples were air-dried, passed through a 2-mm sieve to remove stones and roots, and stored at -20.0°C before being assessed through standard methods (Bao, 2000). Soil pH and electrical conductivity (EC) were measured in a 1:5 soil:water suspension using a glass electrode acidimeter (Leici PHSJ-5, INESA Scientific Instrument Co., Ltd., Shanghai, China) and an EC metre (Leici DDSJ-318, INESA Scientific Instrument Co., Ltd., Shanghai, China), respectively. Soil water content (SWC) was assessed using the oven drying method. SOC was quantified using the dichromate oxidation technique. Available phosphorus (AP) was estimated using the 0.5 mol/L NaHCO₃ molybdenum antimony colorimetric method. Meanwhile, available potassium (AK) was evaluated using NH₄OAC atomic absorption spectrophotometry. Particle size analysis was conducted using a laser particle-size analyzer (BT-2001, Bettersize Instruments Ltd., Dandong, China) to determine the proportions of clay (<0.002 mm), silt (0.002-0.050 mm), and sand (0.050-2.000 mm).

Before data analysis, we checked the normality, homogeneity, and overdispersion of the data. All the statistical analyses were conducted using R v.3.6.1 (R Core Team, 2023), with the significance level set at less than 5%. The ''lme4'' package was used for linear mixed model (LMM) analyses (Bates et al., 2015), and the ''car'' (Fox and Weisberg, 2019) and ''lsmeans'' (Lenth, 2016) packages were used for least-squares mean estimation and contrast tests, respectively. The ''vegan'' package (Oksanen et al., 2020) was used for nonmetric multidimensional scaling (NMDS), permutational multivariate analysis of variance (PERMANOVA), and canonical correspondence analysis (CCA). The ''lavaan'' package (Rosseel, 2012) was used for structural equation modelling (SEM) analysis.

We used the method of Luo et al. (2019) to calculate the importance values of plant species within each sample plot. During the application of the three distinct enclosure treatments (GF1, GF3, and NF3), NMDS was used to investigate the changes in the vegetation community at different distances from the fences. PERMANOVA was conducted to assess the statistical significance of observed variations. These approaches enabled us to determine the magnitude and significance of the observed changes in vegetation composition across different enclosure treatments and fence proximities. LMM analyses were used to investigate the effects of the duration of hanging grass fence enclosure (GF1 vs. GF3) and the presence of hanging grasses (GF3 vs. NF3) on vegetation parameters (including vegetation coverage, plant height, aboveground biomass, species richness, and plant density) and soil physical and chemical properties. Vegetation and soil parameters were used as response variables, and count data were logarithmically transformed. Transect segments were treated as random effects. Meanwhile, year (2019 and 2021), presence of hanging grasses (GF and NF), distance from the fence, sand accumulation thickness recorded as 1.0 cm for flat ground without sand accumulation, and soil depth (considered only for soil physical and chemical properties, where 1 represents topsoil and 2 represents deeper soil) were regarded as fixed effects. The NMDS model included interactions between distance and year, or the presence of hanging grasses. We separately modelled the impact of distance from the fence on vegetation and soil parameters under three different enclosure treatments (GF1, GF3, and NF3). LMM analyses, incorporating various vegetation and soil parameters as response variables, treating transect segments as random effects, and considering (1) distance from the fence (positive distances indicate sample plots within the enclosure area on the leeward side of the fences, and negative distances indicate sample plots within the grazing area on the windward side of the fences), (2) squared distance from the fence, (3) cubed distance from the fence, (4) sand accumulation thickness, and (5) soil depth (considered only for soil physical and chemical properties) as fixed effects, were executed. Anticipating nonlinear responses to the edge (Ewers and Didham, 2006), we added squared and cubed edge distances to the models and determined the final model formula by comparing the Akaike information criterion (AIC) values and goodness of fit. We hypothesized an interaction (linear, squared, and cubed) between sand accumulation thickness and distance, suggesting that the rate of changes in vegetation and soil parameters occurred faster on sandy dunes than on flat ground surfaces. To elucidate the mechanisms driving vegetation variation, we used CCA coupled with variance partitioning to analyze the significant variables influencing the composition and distribution of plant species. SEM was used to identify key environmental factors affecting vegetation coverage and plant density.

A total of 36 taxa from 13 families were recorded in the field surveys of 2019 and 2021. The most species-rich families were Asteraceae (seven species), Gramineae (six species), and Chenopodiaceae (six species), collectively representing 54.28% of all the taxa sampled. A. frigida and S. krylovii were the most frequently recorded plant species (Fig. 3a; Tables S1-3). NMDS analysis indicated that the duration of GF and the presence of hanging grasses significantly influenced species composition. This was further emphasized by the results of PERMANOVA (R²=0.187, P=0.001; Fig. 3b). Compared to GF1 and NF3, the importance values for annual and perennial herbaceous plants, such as Eragrostis minor, Leymus chinensis, Teloxys aristata, Tribulus terrestris, and Chenopodium acuminatum, gradually increased in GF3 (Table S3). Species composition exhibited high similarity among sample plots located at adjacent geographical distances from the fences. However, the distances between sample plots on the NMDS axes did not necessarily correlate directly with their geographical proximity. For instance, the sample plots at 26.5 m within the enclosure area and -16.5 m inside the grazing area showed more similar species composition even though they are positioned farther apart geographically. While, the species composition of the plots in the intermediate zone, especially the plots in the edge area between -1.5 and 1.5 m (Fig. 3b), were less similar (Fig. 3b). GF3 demonstrated the greatest separation in terms of distance from the fence, primarily attributed to the dominant plant species of tumbleweeds such as Salsola collina, Grubovia dasyphylla, and Corispermum mongolicum found in the range from -1.5 to 1.5 m at the fence edge (Fig. 3).

Vegetation coverage, aboveground biomass, species richness, and plant density were all significantly affected by the duration of GF (Table S4). Specifically, vegetation coverage, aboveground biomass, and species richness increased significantly in GF3 compared to GF1 by 8.15%, 109.238 g/m², and 1.408 species/m², respectively. Conversely, compared to NF3, these increases were 4.98%, 42.495 g/m², and 1.015 species/m², respectively, and were not statistically significant (Table S4).

The confirmed existence of the edge effects of fences, which caused heterogeneity in vegetation recovery, led us to undertake a quantitative analysis. Vegetation exhibited a nonlinear response to the distance from the fence, showing a consistent trend as it moved from the windward side to the leeward side, with most influencing initially increasing before gradually declining (Fig. 4). The fitted coefficient of distance in GF3 showed the greatest magnitude, indicating the most profound edge effects, with a significant increase stopping at 6.5 m (plant height), 11.5 m (vegetation coverage and aboveground biomass), and 26.5 m (species richness) on the leeward side. Vegetation coverage and aboveground biomass in GF1 gradually declined from the edge of the leeward side, with no significant difference observed at 16.5 and 11.5 m, respectively. However, no significant differences were observed among vegetation parameters at various distances in NF3 (Fig. 4).

The duration of GF and the presence of hanging grasses significantly influenced all the investigated soil physical and chemical properties, except SOC and SWC (Table S4). Along the transects, soil texture exhibited distinct patterns in response to the presence of hanging grasses, particularly evident in the region ranging from -1.5 to 6.5 m (Fig. 5). Here, sand content had positive edge effects when grasses hanging on the fences and negative edge effects when no grasses hanging on the fences. Far away from the fences, no significant differences were observed between the windward side interior (-16.5 m) and the leeward side interior (26.5 m). SOC and AP contents showed substantial distance impacts in GF3, with greater effects on the leeward side interior than on the windward side interior. However, SOC declined in the fence edge area. Meanwhile, AP content was significantly enriched. Furthermore, significant edge effects were observed for soil pH and AK content in NF3.

The CCA results showed that the combined constrained variables explained 37.66% of the total variation in species composition, with the first two axes explaining 32.55% and 17.52%, respectively (Fig. 6a). The relationship between species composition and environmental factors had a strong correlation (adjusted R²=0.27), but to varying degrees among the different enclosure treatments. Species composition in GF1, NF3, and areas distant from GF3 exhibited a positive correlation with SOC and AP contents. However, in GF3 closer to the fence (-6.5 to 11.5 m), where surface sand had higher accumulation, there was a stronger correlation of species composition with EC, pH, and sand content. The variation partitioning results highlighted that plot factors, including the distance from the fence, the duration of GF, and the presence of hanging grasses, affected species composition (15.04%) and statistically outweighed the soil physical and chemical properties (3.22% and 1.65%, respectively) (Fig. 6b).

SEM analysis was used to investigate the direct and indirect effects of enclosure measures and soil physical and chemical properties on vegetation coverage and plant density (Fig. 7a and b, respectively). The distance from the fence, sand accumulation thickness, and soil physical and chemical properties directly affected vegetation coverage, exhibiting a significant positive correlation. Different enclosure treatments did not directly affect vegetation coverage but exerted an indirect influence by altering soil physical and chemical properties (Fig. 7a). The only factor that was significantly negatively correlated with plant density was sand accumulation thickness. Soil physical and chemical properties did not considerably affect plant density, although sand accumulation thickness and enclosure treatments had a significant influence on this vegetation parameter (Fig. 7b).

Fencing facilities such as edges for grassland enclosures lead to changes in richness, distribution, and growth rates of plant species (Wang et al., 2021). Tumbleweed accumulation and suspension on the fences would significantly intensify the biological differences between the grassland's interior and edge areas. This is because of the exclusion of grazing disturbances and the edge effects created by linear barriers. Taxa that are reduced or degraded during grassland degradation, such as Chenopodium album and Artemisia scoparia, can regenerate once harmful factors, such as grazing and animal trampling are removed (McDonald et al., 2019). Perennial palatable grasses such as Leymus chinensis had increased dominance within the enclosed area. Conversely, prostrate plants, such as Haplophyllum dahuricum and Heteropappus hispidus, had increased dominance outside the fences. This is consistent with the findings of Díaz et al. (2007) on the adaptive responses of low-growing plants to grazing. The higher vegetation coverage within the fence enclosure may enhance seed deposition, primarily facilitated by wind dispersal (Soons et al., 2004; Rotundo et al., 2015). This can benefit light and winged seeds such as Salsola collina, Corispermum mongolicum, and Eragrostis minor, which are more prevalent in the edge areas adjacent to the fences because the hanging grasses on the fences reduce wind speed on the leeward side. Nonetheless, following three years of hanging grass fence enclosure, the edge areas underwent a sharp decrease in species richness as a result of frequent disruptions caused by the entangled grasses on the fences and wind-blowing sand movement, with pioneer species such as Bassia dasyphylla dominating these areas. Sizer et al. (2000) also noted that the primary source of edge-recolonising species is the pioneer species within the seed bank established before edge formation.

Short-term enclosures with various fencing treatments have positively impacted parameters reflecting vegetation growth, such as vegetation coverage and aboveground biomass, a typical response to excluding grazing (Zhang et al., 2019; Liu et al., 2023). However, most vegetation features in the edge areas of fences have shown more significant responses to the presence of hanging grass fences and the duration of the enclosure than to the internal environment on both sides of fences. The edge effects of artificially constructed linear barriers, such as shelterbelts and hedgerows, have been documented in some ecosystems (Sitzia et al., 2013; Schopke et al., 2023). However, quantifying the edge effects of hanging grass fences is still rare (Wang et al., 2021). The changes in vegetation may be due to substantial structural modifications in the original fences caused by hanging grasses, which reduce the fences' porosity. This alteration could create gradient changes along the distance from the fence for microclimatic factors such as wind and light, potentially leading to differences in vegetation recovery at smaller scales (Pohlman et al., 2009; Wilkerson, 2014). Increased light on the leeward side creates favourable conditions for vegetation growth (Wang et al., 2021). Lower wind speeds establish a relatively stable microclimate around the vegetation, maintaining suitable temperature and humidity levels that support the growth and development of plants (Hamm and Drossel, 2017). Conversely, high wind speeds can disrupt these microclimates, leading to unstable environmental conditions.

Enclosure measures show considerable potential for rehabilitating degraded grasslands, particularly in environments susceptible to soil erosion, such as sandy grasslands (Matthews et al., 2017; Zhang et al., 2019). Inter-annual changes in wind speed can profoundly affect the occurrence and intensity of wind erosion and dust emissions (Shinoda et al., 2011). With increased wind speed, the likelihood of more frequent windblown sand activity increases, driven by the heightened sensitivity of temperate grasslands to climate variations. Conducted in moderately to severely degraded desert steppe in Central Mongolia (Meng et al., 2021), our study has shown changes in soil texture during the early stages of GF. We observed a significant increase in sand content within a 6.5 m range on the leeward side. This result contradicts the conclusions of Wang et al. (2021), who found no substantial impact of wind erosion or dust storms on the spatial distribution of soil texture components around fencing facilities, potentially because of the less prominent effect of fencing facilities. This also highlights how the structural changes in hanging grass fences from the original fencing structure resemble a sand barrier with a finer density in the lower part and a coarser density in the upper part. This potentially influences nearby soil physical and chemical properties by decelerating wind speed and trapping windblown sand and tumbleweeds. Significant variations in SOC, EC, and AP were observed at different distances from the fence. During the initial enclosure, the hanging grass fences contributed to a slight accumulation of SOC in the desert steppe (Wang et al., 2021). This is potentially due to the increased litter and nutrient release from vegetation restoration (Zhang et al., 2018; Wang et al., 2023) and tumbleweed remnants on the fences. However, sand accumulation may affect the transformation of plant debris into organic carbon within the soil (Pulido et al., 2018). The reduced wind speed on the leeward side of hanging grass fences may have diminished soil moisture evaporation (Teng et al., 2014) and enhanced soil water retention (Liang et al., 2021). However, this process is complex and warrants further investigation.

Non-biological edge effects, such as microclimate and soil physical and chemical properties, play a significant role in causing heterogeneity in vegetation recovery (Matlack, 1993; Yates et al., 2000). However, the primary environmental factors with the strongest influence in this scenario remain unclear. In this study, the influence of hanging grass fences on the topsoil led to changes in soil texture, nutrients, and organic matter content, thereby affecting the quantity and distribution of vegetation. The microclimate generated by linear barriers might be an underlying factor driving vegetation changes (Cleugh and Hughes, 2002), potentially enhancing vegetation habitats near fences owing to temperature and humidity alterations (Gasperini et al., 2021). Microclimate measurements were not directly obtained in this study, while indirect evidence was derived from the profound modification of the microtopography on both sides of the fences that resulted from the substantial shift in windblown sand retention caused by hanging grass fences. Environmental factors such as sand content, EC, and sand accumulation thickness are pivotal determinants of changes in species composition and plant density. The leeward side of sand barriers often exhibits significant sand accumulation, which can reduce wind erosion on the ground surfaces (Zhang et al., 2016; Xi et al., 2023). Similar to the movement observed in gas-solid two-phase flow, hanging grass fences trap sand grains but may also cause wind-dispersed seeds commonly found in grasslands to settle on the wind shadow area of the leeward side owing to its wind attenuation effect, potentially increasing the local soil seed bank (Luo et al., 2019). A relevant study confirmed a significant positive correlation between the richness of wind-dispersed plant species and vegetation hedges composed of grass matrices (Wilkerson, 2014).

Considering the extensive environment of Mongolian grasslands, hanging grass fences have a relatively limited range of effects. The observed ecological effects appear to be short term and are primarily influenced by wind-blown sand deposition. Long-term ecological implications require further observations and assessment. However, the immediate and significant effects of hanging grass fences on local ecological restoration and soil conservation should not be overlooked. Due to the constraints placed on pastoralists' efforts and the organization and administration of ecological restoration projects due to the tragedy of the commons, special consideration is required in Mongolia's community-based institutions for rangeland management (Bruegger et al., 2014). Emphasis should be placed on severely degraded areas, such as settlement peripheries, roadside areas, and water sources, which are commonly frequented by grazing activities and are often subjected to trampling by cattle and sheep (Fernandez-Gimenez, 2000). Adjusting the fence direction and grid density to suit the prevailing wind patterns and creating hanging grass fences is recommended in these areas. This approach supports wind erosion mitigation, ecological restoration, and economic savings. During severe weather that causes extensive livestock losses, hanging grass fences may be essential sites for grazing, providing shelter from wind and snow and offering some fodder once vegetation recovers.