The study area is located in the eastern part of the agro-pastoral ecotone in North China (41.70°N to 44.54°N, 115.13°E to 122.37°E). Administratively, it includes Chifeng City, Tongliao City and a part of Xilin Gol League in Inner Mongolia, China (Figure 1). The area has a semi-arid temperate continental monsoon climate. The mean annual temperature ranges between 2.0°C and 6.7°C and the mean annual precipitation ranges between 320 mm and 470 mm, over two thirds of which falls in the growing season from May to September. The soils are Kastanozem. The original vegetation is sparse forest grassland, and stems from the Eurasian steppe (Zhao et al., 2002). The major grass species include Leymus chinensis and Stipa baicalensis. The main tree species are Ulmus pumila, Populus simonii, Pinus sylvestris var. mongolica, and Pinus tabulaeformis. The most common crop is Zea mays, accompanied by other crops such as Triticum aestivum and Avena nuda. Grassland, forest and cropland coexist due to frequent human activities such as cultivation, grazing, and the implementation of ecological projects in recent decades. In the agro-pastoral ecotone of Inner Mongolia, vegetation coverage is 29% for grassland, 9% for forest, and 8% for cropland (Zhao et al., 2002).

2.2.1 Literature survey

We collected SOC stock data of grasslands, forests and croplands in the study area from literature published between 2000 and 2014 in the Web of Science and China National Knowledge Infrastructure. The soil samples were collected from not less than three depths. When only soil organic matter was available, a factor of 0.58 was used to convert it to SOC (Poeplau et al., 2011). When soil bulk density was not available, the soil bulk density of the closest site was used. The raw data were obtained from tables or extracted from graphs in the papers using the Get Data Graph Digitizer (Version 2.20, Russian Federation). In total, there were 71 grassland sites and 13 cropland sites.

2.2.2 Field investigation

Soil samples were collected in 2011 from 0-10, 10-20, 20-30, 30-50 and 50-100 cm depths by digging a hole with a shovel from three quadrats (1 m × 1 m) in a plot of at least 40 m × 40 m at each cropland site, and from three quadrats (20 m × 50 m) at each forest site (>200 m distance between adjacent sites). Soil bulk density was determined with the core method (using a cylindrical metal 100 cm³ corer) with three replicas at each depth and the soil samples were dried at 65°C for more than 24 h until the weight remained unchanged. SOC content was measured using the modified Mebius method (Nelson and Sommers, 1982). Briefly, a 0.5 g soil sample was digested with 5 ml of 1 N K₂Cr₂O₇ and 10 ml of concentrated H₂SO₄ at 180°C for 5 min, followed by titration of the digests with standardized FeSO₄. We investigated 23 cropland sites and 43 forest sites.

2.2.3 Data calculation and analysis

The SOC stock at each depth (layer) was calculated using Eq. (1):

where D_SOCi, BD_i, and h_i represent SOC stock (kg m^-2), soil bulk density (g cm^-3), and thickness (Eclesia et al., 2012) of the i layer (cm), respectively; C_i is the SOC content (g kg^-1).

Sampling depths for SOC varied in different studies. To compare SOC data collected from different studies and field investigations, the original SOC data were converted to SOC stock in the top 100 cm depth at each 20-cm interval using the SOC depth distribution functions developed by Yang et al. (2007). SOC data in the upper three depths from literature survey and in the five depths (0-100 cm) from field investigation were used to fit Eq. (2). At each 20-cm interval depth, we obtained the SOC content in volume at h using Eq. (2) and then calculated the SOC stock using Eq. (3) (Yang et al., 2007):

where a and b are the coefficients, h, h₁, and h₂ are soil depths, and C_soch and D_soch are SOC content in volume (g cm^-3) and SOC stock (kg cm^-2) at each depth. The SOC stock dataset was compiled with five fixed depths: 0-20, 20-40, 40-60, 60-80, and 80-100 cm. The coefficient of determination of fit for most SOC sites was above 70% (all sites for grassland, 36 of 43 sites for forest, and 26 of 36 sites for cropland).

In this study, grasslands were split into grazed (29 sites) and grazing-excluded (GE) (42 sites) sites. The latter was further divided into two groups (25 sites for GE<20a and 17 sites for GE≥20a) according to the duration for grassland restoration of He et al. (2008). Forests were classified either as natural forest (16 sites) and plantation (27 sites) or as coniferous forest (14 sites) and broad-leaf forest (29 sites). Stand ages of the forests, determined by drilling three tree-ring cores, were grouped into three classes: 0-15a (9 sites), 16-30a (17 sites), and >30a (17 sites). Croplands were either under conventional cultivation (tillage) (21 sites) or under conservation cultivation (no-tillage) (15 sites). All statistical analyses were conducted using one-way analysis of variance (ANOVA) with a significance level of 0.05. ANOVA and correlation analyses were performed using SPSS 16.0 (Chicago, IL, USA).

On average, SOC stocks in the top 1 m differed significantly among the three land use types (p<0.05), of which grassland (10.8±0.87 kg m^-2) had significantly higher SOC stocks than forest (7.9±0.89 kg m^-2, p<0.05) and cropland (6.4±0.66 kg m^-2, p<0.01), but forest SOC stocks did not differ significantly from those of cropland (p=0.296) (Figure 2a). In addition, spatial variation in SOC stocks (SD=5.78, CV=73%) was higher in forest than in grassland (SD=7.26, CV=67%) and cropland (SD=3.88, CV=60%).

SOC stocks decreased with depth for each land use type. The difference in SOC stocks between different land use types narrowed with soil depth (Figure 2b). At the 0-40 cm depth, grassland SOC stock was significantly higher by 52% than that of forest (p<0.05); at the 40-100 cm depth, grassland SOC stock was more than that of forest by 13.5% but was not significantly higher. Compared to cropland, grassland SOC stocks were higher by 79% (p<0.05) at the 0-60 cm depth and by 30% (p>0.05) at the 60-100 cm depth. Vertical distribution was described by the proportion of SOC content in the 0-20 cm depth relative to that in 0-100 cm depth (Jobbágy and Jackson, 2000). The vertical distribution patterns of SOC stocks indicate that less SOC is stored in the upper layer of soil (0-20 cm) in forest (34%) and cropland (34%) than in grassland (41%) in the study area.

SOC stock was on average 12.2 kg m^-2 for grazing-excluded (GE) grassland, which is marginally significantly higher than the 8.7 kg m^-2 for grazed grassland (p=0.051). SOC stocks increased significantly with the duration of GE (R²=0.19, p<0.01). When the duration of GE was less than 20 years, the SOC stock of GE grassland was 10.5 kg m^-2, being higher than that of grazing grassland, but not significantly. However, carbon stock (14.7 kg m^-2) was significantly increased when the duration of GE exceeded 20 years (p<0.01) (Figure 3a). Such an increase was still observed even when the duration of GE was 30 years, which is the longest GE duration in this dataset.

SOC stock declined with depth for grazed and GE grassland (Figure 3b). At the 0-40 cm depth, the SOC stock in ≥20a GE grasslands was significantly higher than that in grazed grassland (p<0.05). Throughout the soil profile up to 1 m, no significant difference in SOC stocks was found between the <20 year GE grassland and grazed grassland (p=0.061), although slightly higher SOC stocks were observed at each layer in the GE grassland (<20a) than in grazed grassland (Figure 3b). Larger differences in SOC stocks were detected in the upper layers than in deeper layers among the studied grasslands: e.g., SOC stocks at the 0-20 cm depth in the ≥20 year GE grassland were 71% and 64% higher, respectively, than in the <20 year GE grassland and grazed grassland, and these percentages declined to 41% and 40%, respectively, at the 20-40 cm depth (Figure 3b).

SOC stocks increased evidently but not significantly with forest stand age (p>0.05). The average SOC stocks were 5.2, 8.0, and 9.3 kg m^-2 in the 0-15, 16-30 and >30 year age classes, respectively (Figure 4a). Irrespective of tree species, significant linear positive correlations were found between SOC stocks and stand ages for plantations (R²=0.21, p<0.05), but not for natural forests (p=0.27). Similar to the grasslands, SOC stocks declined with depth, and increased with forest age for each depth (Figure 4b).

No significant differences in SOC stocks were found between coniferous (8.1 kg m^-2) and broadleaf forests (7.8 kg m^-2) (p=0.90) (Figure 5a) and between natural forests (8.7 kg m^-2) and plantations (7.4 kg m^-2) (p=0.50) (Figure 5c). Vertical distribution of SOC stocks was

similar among the forest types (Figures 5b and 5d), but the SOC stock at each depth was slightly higher for natural forests than for plantations (Figure 5d).

In cropland, the SOC stock of conservation cultivated land (no-tillage, 8.4 kg m^-2) was significantly higher than that of conventionally cultivated land (tillage, 5.1 kg m^-2) (p<0.05) (Figure 6a). On average, no-tillage significantly increased cropland SOC stock by 65% (3.3 kg m^-2, p<0.05) compared to tillage (Figure 6a). In terms of vertical distribution, the effect of no-tillage on SOC content largely occurred in the 0-40 cm depth (p<0.05) (Figure 6b). Compared to tilled croplands, the SOC stocks of no-tillage croplands were greater on average by 1.5, 0.9, 0.5, 0.3, and 0.2 kg m^-2 at the 0-20, 20-40, 40-60, 60-80, and 80-100 cm depths, respectively. Specifically, the average SOC stocks at the 0-20 cm and 20-40 cm

depths were 96% and 71% significantly greater, respectively, in the no-tillage croplands than in the tilled croplands (p<0.05).

Among the three land use types in the agro-pastoral ecotone in Inner Mongolia, grassland exhibited the highest SOC stock, followed by forest and cropland. Grassland that stored more SOC than forest is consistent with the findings of Wei et al. (2012). They reported that grassland ecosystems would have more SOC than planted forests on the Loess Plateau, which is located in a similar climate zone relative to our study area. The higher SOC stock in the grassland compared to the forest is likely ascribed to: (1) low forest stand density leading to less carbon input into the soil from litter; (2) plantations being mainly established on abandoned farmland, which had low SOC stock due to frequent tillage - restoring SOC takes several years to decades. A meta-analysis also shows that, in temperate zones, grassland ecosystems have more SOC than planted forests converted from croplands (Poeplau et al., 2011). Two reasons likely explain the higher SOC stocks in grassland versus cropland systems. Firstly, grassland is not subjected to regular crop harvest (Solomon et al., 2000) and is no-tillage, which increases the biomass inputs and enhances the physical protection of SOC stocks (Post and Kwon, 2000). Secondly, cropland upper soil carbon is easily taken away by the wind, especially during spring, due to surface bareness. Our result is in good agreement with a previous study conducted in the Xilin River Basin of Inner Mongolia that reported the stored total organic carbon in the uppermost 30 cm soil layer was greater in the grassland than in the cropland (Qi et al., 2012).

Afforestation is supposed to be the plausible pathway to promote terrestrial ecosystem carbon stocks (Lal, 2004; Li et al., 2012); however, in arid and semi-arid regions, the carbon sequestration potential via afforestation is disputable due to water shortage and drought, which have been shown to be unfavorable for forest sustainability and SOC accumulation (Cao, 2008; Cao et al., 2010; Wu et al., 2014). A synthetic analysis shows that the establishment of grassland in temperate zones renders a long lasting carbon sink whereas afforestation of grasslands leads to mostly SOC losses or lack of an SOC sink even after 100 years (Poeplau et al., 2011). This suggests that afforestation or cultivation on original grassland would likely decrease the SOC stock and that grassland restoration would be helpful for increasing SOC stock in semi-arid regions.

In this study, over one third of the SOC was stored in the top 20 cm depth relative to the total depth studied (0-100 cm), being higher on average in grasslands (41%) than in forests (34%) and croplands (34%). This SOC vertical distribution pattern in grassland is similar to that across the world (39%) (Jobbágy and Jackson, 2000)_, but is slightly higher than that averaged over China (34%) (Yang et al., 2007). Although they are located in the same climate zone, the grassland in our study area had a much higher percentage of SOC (41%) in its 0-20 cm layer than the Loess Plateau (26.8%) (Wei et al., 2012). This implies that conversion from grassland to other land use types would influence SOC content more remarkably in the upper soil layers than in the deeper soil layers. Deng et al. (2014) also found that land use conversion in China’s ‘Grain-for-Green’ project led to greater SOC changes in upper soil (0-20 cm) than in deeper soil. Grassland stored more SOC in the 0-20 cm depth than forest and cropland, and thus would give rise to more SOC loss than the deeper depth when disturbed. Thus, protecting the surface SOC of grassland from destruction plays a major role in ensuring SOC stability in terrestrial ecosystems. This implies that grassland restoration or conversion from croplands and forests to grassland systems may increase SOC sequestration in the agro-pastoral ecotone in Inner Mongolia.

The SOC stock of GE grassland increased by 40% on average relative to the grazed grassland in this study. Grazing can lower vegetation coverage and reduce carbon input (Dai et al., 2014). GE is considered to be an effective way to improve carbon sequestration (Hu et al., 2015; Pei et al., 2008; Steffens et al., 2008; Su et al., 2005). GE can promote grass growth and biomass accumulation and thereby increase litter-derived carbon input to the soil (Hu et al., 2015). Mostly, when vegetation is restored, the avoidance or decrease of wind or water erosion can effectively reduce direct carbon loss from the soil (Hu et al., 2005; Wiesmeier et al., 2015; Zhou et al., 2011). Our study demonstrated that compared to grazed grassland, SOC stocks showed a significant increasing trend for more than 20 years after GE. An obvious increase in SOC stocks was found even after 30 years of GE, the longest GE time in our dataset. However, our result disagrees with the finding of Hu et al. (2015), who reported that, in North China, the achievement of an SOC stable state takes about 15 years on average after the grassland is fenced against grazing. A major reason for this discrepancy is likely that SOC accumulates slowly in arid and semi-arid regions (Werth et al., 2005) due to droughts. Therefore, more time is required for SOC equilibrium in our study area than in North China. Additionally, in our study, GE grassland is mainly distributed in the agro-pastoral regions, and although fenced, is influenced occasionally by humans or livestock. Further, in a study performed in a region adjacent to our study area, Zhang et al. (2014) suggested that it takes at least 50 years to attain the equilibrium of SOC stock for grasslands recovered from degradation. Lastly, different research scales and grassland types are probable reasons for this discrepancy.

In forests, SOC stocks either in the entire profile up to 100 cm or in each depth (20 cm intervals) was observed to increase with stand age. Globally, there is an overall increasing trend in SOC stocks through time across all biomes (Pregitzer and Euskirchen, 2004). Tree species and forest types did not present obvious impacts on forest SOC stocks probably because most of the forests studied were relatively young. The average stand age was 26 and 31 years for coniferous and broad-leaf forests, and 23 and 39 years for plantation and natural forest.

In cropland, compared with conventional cultivation (tillage), conservational cultivation (no-tillage) increased SOC stock by 65% and 48% at 0-100 cm and 0-20 cm depths, respectively. Many studies have reported that no-tillage can largely increase SOC stock; tillage disturbs the soil physical aggregate structure, which can increase microbial activity and improve organic matter decomposition (Ogle et al., 2012; Post and Kwon, 2000). However, this increase is primarily limited to the top soil layer (e.g., 0-30 cm) (Ogle et al., 2005; Powlson et al., 2014). The mechanisms of SOC stock increase are not yet fully understood (Ogle et al., 2012). Other than the direct increase in carbon input to the soil through litter and/or stubble mulching, the reduction of erosion and runoff and amelioration of water holding capacity might be also important contributors (Palm et al., 2014). Apart from tillage, factors such as crop type, crop rotation, fertilization, and irrigation have direct or indirect effects on SOC dynamics.