**3. Results**

#### *3.1. Characteristics of the Temperate Savanna Ecosystem*

There were nine woody plant species, *S. linearistipularis*, *S. microstachya var. bordensis*, *S. gordejevii*, *S. myrtilloides*, *B. fruticosa*, *C. microphylla*, *Spiraea aquilegifolia*, *R. diacanthum*, and *U. pumila*, which were identified in the plot. The trees in the plot mainly grew as individual trees rather than in groups (Table 1).

*U. pumila* was the dominant tree species, with a coverage of 10–15% and a density of 56 trees/ha. The densities of old trees (DBH ≥ 20 cm), medium trees (5 cm ≤ DBH < 20 cm), and juvenile trees (DBH < 5 cm) were 30, 10, and 16 trees/ha, respectively. The tree heights of these three categories varied, i.e., old trees, medium trees, and juvenile trees had average heights of 8.4 ± 0.94, 5.3 ± 0.93, and 1.7 ± 0.30 m, respectively. Although the coverage of the shrub layer was only 8–15%, eight shrub species were observed, and their density varied from 6.7 to 96.7 plants/ha; the shrubs were shorter than 0.3–1.3 m on average. These shrubs normally grew in clusters distributed in the canopy gaps among *U. pumila* trees.

The herbs were also unevenly distributed and were rarely found growing under the old trees. However, the coverage of the herb layer, 30–60%, was greater than that of the tree and shrub layers. The herbs were also very diverse, with a total of 195 species from 40 families identified in our research plot, representing approximately 30% of the total number of higher plants observed in the Otindag Sandy Land [18,39]. The herbs mainly belonged to the *Compositae* (38 species), *Gramineae* (30 species), and *Leguminosae* (14 species) families.

#### *3.2. Spatial Distribution Patterns of U. pumila Trees—A Random Distribution of Old Trees and an Aggregated Distribution of Juvenile Trees*

As shown in Figure 2, *U. pumila* trees showed significant aggregation at the 0–7 m scale. However, their distribution patterns varied with tree size. The old *U. pumila* trees were randomly distributed at all scales (0–50 m); in contrast, the medium *U. pumila* trees (5 cm ≤ DBH < 20 cm) exhibited significant aggregation at the 0–5 m scale, and juvenile trees (DBH < 5 cm) were significantly aggregated at the 0–9 m scale. However, all three tree categories were randomly distributed at other scales in the 25-ha permanent plot.

As shown in Figure 3, there were intraspecific relationships between the different DBH classes of *U. pumila* trees. There was a significant negative relationship between old trees and medium trees at the 2–10 m scale, but no obvious relationships between old trees and juvenile trees. On the other hand, there was a significantly positive relationship between medium trees and juvenile trees at the 2–12 m scale, but no significant relationship at any other scale.

#### *3.3. Aggregated Distribution of Shrubs at Specific Scales*

As shown in Figure 4, most shrub species exhibited spatial aggregation at small scales. For example, *S. aquilegifolia*, *S. microstachya* var. *bordensis*, *S. gordejevii*, and *C. microphylla* were significantly aggregated at scales of 0–6, 1–7, 0–2, and 0–6 m, respectively. However, they exhibited random distributions at other scales. *S. myrtilloides and B. fruticosa* exhibited similar distribution patterns; they were significantly aggregated only at scales of 2–4 and 2–5 m, respectively, and had random distributions at other scales. The exceptions to these findings were *S. linearistipularis* and *R. diacanthum*, which were randomly distributed at all scales. These results indicate that spatial aggregation at specific scales was common among most shrub species.

#### *3.4. Spatial Interactions of Woody Plants—Competition between Shrubs and Adult Trees and Mutualism between Dominant Shrub Species and Juvenile Trees*

As shown in Figure 5, there was a negative relationship between the presence of old trees and most shrub species. For example, there was significant negative relationship between old trees and some shrubs, e.g., *B. fruticosa* and *S. aquilegifolia*, at all scales (0–50 m). However, other shrubs had significant negative relationships with old trees only at certain scales—for example, *S. myrtilloides*, at the 10–50 m scales; *S. linearistipularis*, at the 10–20 m scales; *S. gordejevii*, at the 0–30 m scales; and *C. microphylla*, at the 0–25 m scales. In addition, there was a positive relationship between *R. diacanthum* and old trees at the 3–6 m scales. These results showed that there was a competitive relationship between most shrubs and the old trees.

**Figure 2.** Univariate spatial patterns of the three *U. pumila* tree categories with the null model.

**Figure 3.** Bivariate spatial association between three *U. pumila* tree categories with both the toroidal shift and antecedent condition null models.

**Figure 4.** Univariate spatial patterns of shrub categories with the null model of complete spatial randomness.

**Figure 5.** Bivariate spatial relationship between old *U. pumila* trees and shrubs with the independent null model.

There were negative relationships between the medium trees and most shrub species (Figure 6). For example, the presence of the shrubs *S. gordejevii*, *B. fruticosa*, *S. aquilegifolia*, and *S. microstachya* var. *bordensis* was significantly negatively associated with medium trees at all scales (0–50 m). In addition, *S. linearistipularis* was significantly negatively correlated with medium trees only at the 15–50 m scales, and no obvious relationships were observed at other scales. For *C. microphylla*, *S. gordejevii*, and *R. diacanthum*, no obvious relationships with medium trees were observed. These results also showed that there was a competitive relationship between most shrubs and medium trees.

**Figure 6.** Bivariate spatial relationship between medium *U. pumila* trees and shrubs with the independent null model.

However, there were positive relationships between juvenile trees and some shrub species (Figure 7). *S. myrtilloides*, *S. aquilegifolia*, *S. microstachya* var. *bordensis*, and *C. microphylla* showed significant positive interactions with juvenile trees at small scales (less than 10 m). However, no other obvious relationships were observed between juvenile trees and the other shrub species.

**Figure 7.** Bivariate spatial relationship between juvenile *U. pumila* trees and shrubs with the independent null model.

## **4. Discussion**

#### *4.1. Spatial Patterns of U. pumila Trees and Their Mechanism of Formation in the Temperate Savanna -Like Ecosystem*

From the viewpoint of ecological niche di fferentiation, the large crown breadth of adult *U. pumila* trees indicates a strong demand for a broad niche [46], and its random distribution indicates strong competition among *U. pumila* trees. *U. pumila* trees of the temperate savanna normally have a larger crown width (6–12 m) and deeper roots (10–20 m depth) than *U. pumila* trees in temperate forests (4–8 m crown width, 3–5 m root depth) [42]. In this study, the old *U. pumila* trees were randomly distributed at all scales. This indicates that *U. pumila* exhibits strong self-thinning due to the intense competition among individual trees [47]. In temperate forests, self-thinning normally occurs due to light competition [48,49]. However, in this study, light was not a limiting factor, as indicated by the large tree crowns. Thus, soil moisture can be deduced to be the key limiting factor that results in trees being randomly distributed far from each other through self-thinning [50]. This deduction can also be indirectly supported by the small amount and high variability of the annual precipitation in this area, which is typically 200–350 mm [37,38]. Drought has been increasing continuously in this area since 1960, and the most severe drought in recorded history in this area occurred during 1999–2011 [51]. In addition, *U. pumila* seedlings often su ffer from severe water stress during dry summers that is caused by repeated cycles of drying in the upper soil layers [52]. Wang et al. [53] noted that more than 90% of the current-year seedlings died in fenced plots because of their vulnerability to drought. Therefore, light limitation is generally replaced by increasing water stress in drought-modulated ecosystems, such as semiarid forests and temperate savanna [54].

#### *4.2. Spatial Interactions among Trees and Shrubs—Old U. pumila Trees Inhibit the Survival of Juvenile Trees and Shrubs*

In this study, a negative relationship between old *U. pumila* trees and juvenile trees was found, and significant competition between most shrub species and old *U. pumila* trees was also found. From these findings, we can conclude that the environment around old *U. pumila* trees is not suitable for the survival of juvenile trees or shrubs [53,55]. This further demonstrates that seedling regeneration in the *U. pumila*-dominated savanna-like ecosystem depends strongly on medium trees rather than on old trees. This situation is likely due to the influence of grazing. Historically, the *U. pumila*-dominated savanna-like ecosystem of the Otindag Sand Land was an important pasture source, providing a large amount of forage for livestock [51,56]. Although precipitation and soil moisture have been the main limiting factors a ffecting population regeneration of woody plants in arid and semiarid areas [57], grazing has become an important factor a ffecting population regeneration; long-term continuous heavy grazing has resulted in the destruction of topsoil and degradation of the rangeland [58]. In this savanna-like ecosystem, the area under the tree crown is usually occupied by resting livestock taking shelter from the summer heat and the intense ultraviolet radiation. Thus, the soil under the crown may be destroyed and become hardened due to the animals' trampling and reclining, and a "bare soil circle" will sometimes form under these trees [59]. This is similar to the phenomenon of piospheres caused by grazing. It was found by some studies that the soil moisture and nutrients in the piospheres are significantly lower than those in other areas; this may be the main reason for the change of vegetation composition and pattern, such as reduction in the density and production of forage, changes in the species composition of forage vegetation, shrub encroachment, and so on [59]. Some studies focus on the "bare soil circle" formed under these trees in this savanna-like ecosystem and also prove that the soil moisture and nutrients on the "bare soil circle" are lower than those outside the crown of elm [60]. These factors are unfavorable to the survival of any individual plant, including juvenile trees, shrubs, and grasses. Therefore, the disturbance of livestock caused by overgrazing may be one of the main reasons for the negative relationships between old *U. pumila* and juvenile trees and shrubs in this region.

In contrast, medium trees exhibited a significant positive association with juveniles at smaller scales. This suggests that the environmental conditions, such as light availability and soil moisture, around medium trees are more favorable for juvenile trees than those around adult trees. Some studies also show that the soil nutrients (organic matter, total nitrogen, available phosphorus) and soil moisture under the crown of medium trees are significantly higher than those outside the crown, and there are better microenvironmental conditions (light, temperature, humidity, etc.) under the crown [60]. In addition, the disturbance of livestock to medium trees is less than that to old trees.

Different shrub species showed different spatial relationships with *U. pumila* in this study. There were positive relationships between juvenile trees and some shrub species, such as *C. microphylla.* This may be because the emergence of shrubs improves soil moisture and nutrient conditions [61,62], and *C. microphylla* surely can provide some fertilization of the area due to the symbiosis with Azotobacter spp. This demonstrates that some shrubs provide some fertilization to the area in which they grow, thereby providing a favorable environment for *U. pumila* seedling growth. These results also showed that there were mutualistic relationships between juvenile trees and the shrubs *S. gordejevii* and *S. aquilegifolia*; this might be because *S. aquilegifolia* and *C. microphylla*, as spiny shrubs, can provide shelter for elm seeding, such as holding back sand and reducing damage from livestock and wind, which would protect the juvenile *U. pumila* trees [63]. Our results were similar to those of Tölgyesi et al. [9], who also found that thorny shrubs were important for supporting the biodiversity of wooded rangelands as well as facilitating the regeneration of trees by acting as nurse species.

#### *4.3. The Development of the Original Tree-Dominated Temperate Savanna-Like Ecosystem after Shrub Encroachment*

*U. pumila*-dominated temperate savanna-like ecosystem, once widely distributed throughout the forest-steppe ecotone on the Mongolian Plateau, is a relatively stable woody-herbaceous complex ecosystem in Northern China. Unfortunately, by the late 1990s, only a few degraded relic stands remained as scattered and fragmented patches that showed apparent regeneration failure [64]. This degradation was mainly due to irrational human land use, such as overgrazing.

Many studies clearly showed the indicative significance of shrub encroachment, which mainly depends on different factors, such as the type of ecosystem, past managemen<sup>t</sup> practices, the current level of biodiversity, climate change, etc. [13,14]. In this region, shrubs have occupied the grassland habitat. This encroachment has led to a decrease in grass coverage and a decline in pasture quality [51]. This change is likely to become more serious with the increase in grazing pressure and the impacts of global warming. These outcomes sugges<sup>t</sup> that shrub encroachment is an indicator of savanna ecosystem degradation to some extent. On the other hand, in this study, shrub encroachment was shown to promote the survival and growth of juvenile trees to some extent, which is beneficial to the reestablishment of the *U. pumila* community. However, mutualism was not observed between most shrub species and medium *U. pumila* trees in this study. This lack of mutualism may have occurred because the seedlings under the shrubs had not ye<sup>t</sup> developed into medium trees at the time of this study, because the seedlings were excluded by the shrubs before developing into medium trees, or because the older seedlings may have excluded the shrubs from the habitat. If the shrubs were excluded by the older seedlings, this suggests that the emergence of shrubs in this habitat is only a phase. Conversely, if the small trees were excluded, this would be catastrophic for the ecosystem, and it would be difficult for the ecosystem to return to its original state. To determine the exact causes of these phenomena, further experimental observation is greatly needed. Regardless, this study demonstrated that shrub encroachment in this area has resulted in changes in the composition and spatial pattern of the original tree-dominated savanna.
