A field survey was conducted in the Ansai County of Shaanxi Province, China (Figure 1). Ansai County, located in the central part of the Loess Plateau, is well known for its extensive soil erosion. This area has a typical semiarid continental climate, with a mean annual temperature of 8.8°C and a mean annual precipitation of 549.1 mm. Precipitation varies seasonally, however, with 74.3% of it falling from June to September (Chen et al., 2007b; Sun et al., 2016a). Most native trees and shrubs were replaced by agricultural crops and were retained only at field edges and on gully slopes (Jiao et al., 2007). The current major land-use types are forest, shrubland, grassland, cropland, abandoned cropland, and residential areas in a mosaic rural landscape pattern (Jiao et al., 2012; Sun et al., 2016c). The typical vegetation currently found in the study area includes trees, such as Populus davidiana Dode., Robinia pseudoacacia Linn., and Quercus wutaishanica Blume; shrubs, such as Hippophae rhamnoides Linn. and Caragana korshinskii Kom.; and herbs, such as Stipa bungeana Trin. and Artemisia sacrorum Ledeb.

We used the approach of substituting space for time—i.e., using plots of different ages in the same study area—to investigate grassland vegetation dynamics in the recovery phases during secondary succession. Data were collected in September 2011 from a total of 36 sampling plots (each 20 m × 20 m, “L-quadrats”) at nine locations and from four recovery phases after cropland abandonment: 1–5, 5–10, 10–20, and 20–30 y. Three 1 m × 1 m quadrats (108 “S-quadrats” in total) were chosen along the diagonals of each plot for detailed inventories of the grassland vegetation. All trees, shrubs, and herbs were identified and measured for their base diameter, layer coverage, total height, and growth condition. Local villagers and land documents verified the time of abandonment of each plot. Farming practices had been similar across all nine locations for >40 y before their abandonment. These abandoned croplands were allowed to undergo natural vegetation succession with no influence from human activity (Sun et al., 2016b).

All statistical analyses used the R version 3.3.2 (R Core Team, 2017). Beta diversity was analyzed using the betapart package (Baselga and Orme, 2012), and the rarefaction and extrapolation curves were compiled by using the iNEXT package (Hsieh et al., 2016). The figures were drawn and the data manipulated by using the ggplot2 (Hadley, 2007a) and reshape2 (Hadley, 2007b) packages, respectively.

We identified a total of 64 plant species, belonging to 50 genera and 23 families, among the 5634 live plants found in the 108 quadrats spanning the four recovery phases on the Loess Plateau (see Table S1). The numbers of families, genera, and species and the species abundances increased gradually through the recovery phases of 1–5 to 10–20 y, but then they abruptly decreased to a low level after 20–30 y of natural secondary succession (Table 1).

The species A. capillaris Thunb., Sonchus oleraceus Linn., S. bungeana Trin., and Bothriochloa ischaemum (Linn.) Keng had the highest IVs and were the dominant species in the 1–5y recovery phase; H. altaicus (Willd.) Novopokr., B. ischaemum, Setaria viridis (Linn.) Beauv., and Leymus secalinus (Georgi) Tzvel. were the dominant species at 5–10 y; A. sacrorum, A. leucophylla (Turcz. ex Bess.) C. B. Clarke, S. bungeana, and H. altaicus were the dominant species at 10–20 y; and A. sacrorum, A. leucophylla, Lespedeza bicolor Turcz., and S. bungeana were the dominant species at 20–30 y. The grassland at the site underwent a clear successional series, with species assemblages dominated by A. capillaris→ H. altaicus→ A. sacrorum.

The variation in species composition among the four recovery phases was estimated by the beta-diversity analysis (Figure 2). Total dissimilarity (β_SOR) and turnover dissimilarity (β_SIM) showed similar trends among the four phases: being highest for 10–20 y, but with similar values at 5–10 y and 20–30 y, but lowest for 5–10 y. The nestedness dissimilarity (β_SNE) clearly differed among the four recovery phases in this order: 10–20 > 20–30 > 5–10 > 1–5 y.

Figure 5 shows a graphic representation of the power law for the four recovery phases. The high R² values (coefficients of determination) for all phases indicated that the power law adequately explained the mapping data and successfully identified high spatial heterogeneity. All of the regression lines were mostly above the y = x line over the observed ranges of x. The spatial patterns of the four phases were judged to be more heterogeneous than would be expected from a random pattern, as they displayed an overall aggregated pattern according to the power-law graph.

Figure 6 shows the relationship between the occurrence, p, and spatial heterogeneity, δ, of the four recovery phases. Here, we used the 15 most dominant species with their maximum abundances (see Table S1) as examples to illustrate their spatial heterogeneity in the four recovery phases.

In the first 1–5 y of succession, Dracocephalum moldavica Linn. (H12), L. bicolor (H9), Potentilla bifurca Linn. (H13), and Bidens pilosa Linn. (H4) were highly heterogeneous. By contrast, L. daurica (Laxm.) Schindl. (H7), S. viridis (H2), A. argyi Levl. et Vant. (H15), and A. leucophylla (H8) were less heterogeneous and had intermediate δ values. The species A. capillaris Thunb. (H3) had an approximately random pattern. In the next phase, of 5–10 y, S. oleraceus (H10), B. pilosa (H4), L. secalinus (H11), and L. bicolor (H9) were all highly heterogeneous, whereas S. bungeana (H5), P. bifurca (H13), D. moldavica (H12), and S. viridis (H2) were less heterogeneous and had intermediate δ values. The distribution of species H. altaicus (H6) followed an approximately random pattern. In the third, of 10–20 y, S. oleraceus (H10), B. pilosa (H4), L. secalinus (H11), and S. viridis (H2) were all highly heterogeneous. Conversely, L. bicolor (H9), S. bungeana (H5), A. argyi (H15), and D. moldavica (H12) were less heterogeneous and had intermediate δ values. The species A. sacrorum (H1) had an approximately random pattern. In the final phase assessed, that of 20–30 y, D. moldavica (H12), A. argyi (H15), and L. daurica (H7) were highly heterogeneous. P. bifurca (H13), A. leucophylla (H8), S. bungeana (H5), and H. altaicus (H6) were less heterogeneous and had intermediate δ values, while A. sacrorum (H1) showed a more or less random pattern.

The spatial heterogeneity for the entire grassland community (δ_c) increased gradually for the recovery phases of 1–5 y (δ_c = 0.077), 5–10 y (δ_c = 0.071), and 10–20 y (δ_c = 0.142), was largest in the 10–20 y phase, but then abruptly decreased to a low level in the phase of 20–30 y (δ_c = 0.078).

After cropland abandonment, the grassland species composition and dominance changed markedly during the four phases of 1–30 y of secondary succession. A clear successional series emerged, whereby the species assemblage was dominated by A. capillaris→ H. altaicus→ A. sacrorum. This same successional series has been reported by multiple studies on the Loess Plateau (Zou et al., 1998; Wang et al., 2011; Sun et al., 2016b), which have generally found that the annual herbaceous species, A. capillaris (a common pioneer species), was often later succeeded by the perennial species A. sacrorum during grassland succession. The perennial species H. altaicus has always been an associated species, and though it continues to survive and grow over the long term it cannot ultimately form a climax grassland community. The importance value of A. sacrorum increased across the recovery phases; it was the dominant species in the 10–20 y phase, but obviously more so at 20–30 y. We therefore conclude that the grassland vegetation at our study site is adaptable to the environment: it was able to form a stable community after 20 years of natural recovery of abandoned cropland.

The diversity pattern of the grassland vegetation had a trend of low–high–low, with diversity greatest in the 10–20 y phase. This pattern corresponds to a hump-backed model with maximum diversity at intermediate stages (Grime, 1973). Species diversity in this model is low during the early stages of succession, which are colonized by a limited number of pioneer plants. Other species invade the area as succession progresses, so that many mid- and late-successional species compose the intermediate stages, decreasing species diversity, whereas the late stage becomes dominated by a few late arrivals (Begon et al., 2006; Madeira et al., 2009). Our results also showed that both the pioneer species (A. capillaris, S. oleraceus, and B. pilosa) and the late-successional species (A. sacrorum, S. grandis P. Smirn., and Parinia heterophylla Bunge) coexisted in the 10–20 y phase. Chen et al. (2014) also found this pattern and noted that community coverage, plant density, diversity, and biomass all gradually increased during the secondary succession of grassland vegetation, with the maximum levels reached within about 20 years. We therefore conclude that the Loess Plateau grassland had attained a stable plant community through 30 years of invasion, competition, diffusion, and ecesis.

An aggregated pattern was prevalent in the four recovery phases of the grassland, consistent with multiple findings in arid or semiarid natural grasslands in China and elsewhere (Song et al., 2005; Tsuki et al., 2005; Chen et al., 2007a; Guan et al., 2016). Condit et al. (2000) noted that wind-dispersed species were not as well dispersed as the animal-dispersed species and tended to have aggregated distributions. The majority of herbaceous plants found in our study area produce wind-dispersed seeds, and some, such as the perennial species H. altaicus and A. vestita, can propagate by root shoots, which can further contribute to a highly aggregated spatial pattern. The ecological environments in arid and semiarid areas, especially on the Loess Plateau, are harsh and complex, the habitats are highly heterogeneous, both soil moisture and fertility are unevenly distributed, and the grazing disturbance is high—this great environmental heterogeneity determines the variation and aggregated distribution of the vegetation. Jiao et al. (2007) reported that the time of abandonment, total P content, and soil-water content were the key factors affecting the establishment of vegetation on the Loess Plateau. Many studies have emphasized that the availability of soil water to plants is a critical factor controlling productivity and heterogeneity of plant distributions, and of other life forms, in arid and semiarid areas (e.g., Snyman, 2002; Noymeir, 2003; Diaz et al., 2004; Chen et al., 2010). Liu et al. (2014) and Song et al. (2005) suggested that topography was another heterogeneous factor that could affect the structure and dynamics of vegetation communities, because topography also controls soil moisture that drives plant distributional pattern and diversity. This view was supported by the findings by Moeslund et al. (2013). The pronounced aggregated pattern reported by us can thus be linked to the dispersal properties of herbaceous plants and the higher environmental heterogeneity on the Loess plateau.

Natural succession assists in the recovery of vegetation. However, natural succession over a long period (>20 years) may be needed to establish stable vegetation cover, especially in grazed areas or where erosion is still a large problem. Our results indicated that successional age was a key factor in controlling species composition and diversity, as it further affected the recovery and development of the vegetation. Furthermore, the succession of grassland vegetation may be accelerated by introducing pioneer tree species. The succession of vegetation on the Plateau is as follows: annual herbs→ perennial herbs→ shrubs→ pioneer trees→ climax trees. Our study found that the grassland communities which formed after 20 years of natural recovery on abandoned cropland were dominated by some shrub and semi-shrub species, namely A. sacrorum, A. leucophylla, L. bicolor, and L. daurica. We conclude that the grassland vegetation in our study was on a positive and progressive succession trajectory, such that we predict the plant community at our site will eventually transition to a stable and climax tree community (if given the time). We therefore recommend that some native pioneer tree species, such as Pinus tabuliformis Carr., Betula platyphylla Suk., and P. davidiana Dode., be considered for introduction to the grassland 20 years after abandonment, to accelerate the process of succession and to promote community stability.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.