The study site is located at the Station of Chunlan Bai Desertification Control, Yanchi County, Ningxia Province, China (107°27′ E, 37°54′ N), on the southwestern edge of the Mu Us Desert. The region has a typical temperate continental monsoon climate with an elevation of 1308 m. The mean annual precipitation is 275 mm, with 73% occurring in summer and autumn. The mean annual temperature is 7°C. The average relative humidity is 51% and the frost-free period lasts for 128 days. According to the US Soil Taxonomy system, the soil type is quartisamment (Gao et al., 2014), with a pH range of 8.0 to 9.0. In the 1980s, the landscape of the research area was dominated by SL, which comprised many connected active sand dunes devoid of any vegetation. At that time, the groundwater level was high enough (2 m) to supply water for tree growth. Afforestation with poplar (Populus alba) on SL was successively performed by Chunlan Bai and her family to restrict sand movement and to protect their homeland. At present, forestlands with different plantation ages have been established at the study site. Additionally, areas of SL at some distance from human habitation have not been managed, and have remained active. Previous studies have confirmed that the soil properties in the SL do not vary over a prolonged period of time (Su and Zhao, 2003; Su et al., 2010), suggesting that the soil properties prior to the start of the experiments can be represented by those in the SL at the time of the study. Therefore, the present-day SL can be used as a control for investigating the changes in SIC and soil stable carbon isotopes following afforestation. In this study, we used different-aged forestlands to explore the dynamics of SIC along a chronosequence of afforestation, because there had been no related study in this region and there was a lack of historical data. Within the scope of the 2 km × 1 km in the study site, we selected an SL and its nearby three different-aged forestlands as the four treatments: (1) the SL (control), (2) an 8-year-old poplar land, (3) a 20-year-old poplar land, and (4) a 30-year-old poplar land. For each treatment, we selected one sample plot. The distribution of the four sample plots within the study site is illustrated in Figure 1, and information on the four sample plots is presented in Table 1.

Thirteen 20 m × 20 m subplots were randomly selected within each sample plot for soil sampling. In each subplot, five holes (100 cm in depth) along an S-shaped curve were drilled using a soil auger (10 cm in diameter) after removing litter (the relationships between sample plot, subplot and hole are shown in Figure 1). The soil samples were obtained at a depth interval of 20 cm from 0 to 100 cm. In each subplot, five soil samples obtained from five holes at the same layer were mixed into a composite sample (approximately 500 g), and five composite samples were achieved at a depth interval of 20 cm from 0 to 100 cm within each subplot. Sixty-five composite samples from the 13 subplots within each sample plot were obtained. After the samples were air-dried, roots were removed from all the 260 composite samples from the four sample plots. For each air-dried composite sample, approximately 50 g soil was taken and retained for measuring particle size distribution using a particle size analyzer (Malven Laser Mastersizer 2000, England). The remaining air-dried composite samples were fully ground in an agate mortar and passed through a 0.1 mm sieve for SIC content, SOC content and soil δ¹³C analyses.

After obtaining the 260 composite samples, a soil profile at 0–100 cm was excavated within each subplot. A metal corer (100 cm³ in volume) was driven into the soil at a depth interval of 20 cm from 0 to 100 cm, and then soil samples were oven dried at 115°C for 24 h and weighed to determine bulk density. From the excavated soil profile in each subplot, additional soil samples were obtained at a depth interval of 20 cm from 0 to 100 cm for measuring Soil pH, using a 2.5:1 ratio of deionized water/soil mass. SOC content was determined using the dichromate oxidation procedure described by Walkley and Black (1934). SIC content was determined using the pressure calcimeter method (Wang et al., 2012). The stocks of SIC were calculated as follows:

where M is soil carbon stock per unit area (Mg ha^-1); D is soil depth (cm); B is bulk density (g cm^-3); Z is carbon content (g kg^-1) and G is the relative amount of gravel (%). The gravel content was 0 because there was no gravel in the soil.

The detailed methods for determining δ¹³C-SOC and δ¹³C-SIC have been described previously by Jin et al. (2014). For the determination of δ¹³C-SOC, 5 g of ground and sieved soil was steeped in 2 M HCl for 24 h to remove SIC. The treated soil was then washed with distilled water until the pH exceeded 5, and was subsequently dried at 40°C. From each dried soil sample, approximately 30 mg soil was packed in a tin cup and analyzed with an elemental analyzer (Flash EA 1112, Thermo Fisher Scientific, Inc.) and an isotope ratio mass spectrometer (IRMS) (Finnigan MAT Delta plus XP, Thermo Fisher Scientific, Inc.). The contents of the tin cup were combusted at 1000°C in the EA, and then the SOC of the sample in the tin cup was converted to CO₂. The CO₂ from the EA was ionized and its δ¹³C value was measured by IRMS. The working standards used for determining δ¹³C-SOC were Protein (Elemental Analyses, Inc., Beijing, China, -26.98‰) and NBS-19 (National Institute of Standards and Technology, Gaithersburg, MD, United States; +1.95‰).

To determine ¹³C-SIC, approximately 100 mg sieved soil was reacted with 5 mL 100% H₃PO₄ for 2 h at 75°C in a 12 mL sealed vessel of Gas Bench II (Thermo Fisher Scientific, Inc.) to generate CO₂, and the generated CO₂ was measured by IRMS (Finnigan MAT Delta plus XP, Thermo Fisher Scientific, Inc.). The working standards used for determining δ¹³C-SIC were NBS-18 (National Institute of Standards and Technology, Gaithersburg, MD, United States; -5.01‰) and NBS-19.

The stable isotope compositions of the SOC and SIC, expressed in delta (δ) notation, were both calculated as follows (Coplen, 2011):

where (¹³C/¹²C)_sample and (¹³C/¹²C)_standard are the atomic ratio of ¹³C to ¹²C in the sample and in the Vienna Pee Dee Belemnite (VPDB) standard, respectively. All samples were measured in triplicate. In the three measurements for each sample, the standard deviation of the reported δ¹³C-SOC and δ¹³C-SIC in this study was within 0.4 and 0.3‰, respectively.

Statistical analyses were performed using version 16.0 of the SPSS software (SPSS, Chicago, IL, United States). Two-way analysis of variance was conducted to test the effects of soil depth and plant age, as well as their interactions with soil carbon contents and soil δ¹³C values (Table 2). Multiple comparisons and one-way analysis of variance procedures were used to compare the differences in soil carbon contents and soil δ¹³C values between different treatments within the same depth, and between different soil depths within the same treatment. Mean comparisons were performed using the least-significant-difference test. Linear regression analyses were carried out to evaluate the relationships between various carbon variables (SOC vs. SIC, δ¹³C-SIC vs. SIC, δ¹³C-SIC vs. δ¹³C-SOC, SIC vs. silt particle, SIC vs. clay particle).

Afforestation was found to cause a variation in bulk density and fine particles at 0–40 cm soil layer (Table 3). Within this depth, the bulk densities in P-20 land and P-30 land were significantly lower than in SL, but there was no significant difference between P-8 land and SL. The silt and clay particle contents at 0–20 cm in the three forestlands were significantly higher than in SL. At the depth of 20–40 cm, the silt particle content in P-30 land was significantly greater than that in SL, but there was no significant difference between P-8 land and SL or between P-20 land and SL. The clay particle content in P-20 land was remarkably greater than in SL, but there was no significant difference between P-8 land and SL or between P-30 land and SL. Within the 40–100 cm depth layer, no differences in bulk density or fine particles were observed between the four sample plots (Table 3). Additionally, soil pH at 0–100 cm in P-20 land and P-30 land was considerably lower than that in SL, but there was no significant difference between P-8 land and SL within the 60–100 cm depth layer (Table 3).

Soil inorganic carbon content was enhanced by afforestation. Within the 0–100 cm depth, the SIC content in each 20 cm depth interval in P-8, P-20, and P-30 lands was significantly higher than in SL (Table 4). Among the three forestlands, the SIC content increased with plantation age. Within the 0–40 cm layer, the SIC content in P-30 land was considerably higher than in P-20 land, but there was no significant difference between P-20 land and P-8 land. Within the 40–100 cm layer, the SIC content in P-30 land was significantly greater than that in P-20 land, which in turn was greater than that in P-8 land. Afforestation also elevated SIC stocks. The SIC stock at 0–100 cm in SL was 34.2 Mg ha^-1, which increased to 42.5, 49.2, and 68.3 Mg ha^-1 in P-8, P-20 and P-30 lands, respectively (Figure 2). The SIC contents in SL, P-8 land and P-20 land were almost evenly distributed among the five 20 cm soil intervals from 0 to 100 cm (Table 4). The SIC content in P-30 land at 0–20 cm was significantly higher than at 80–100 cm; however, no differences were observed among the 0–80 cm layers or among the 20–100 cm layers. In addition, the SOC content in the three forestlands was significantly higher in each soil layer than at the same depth in SL (Table 4).

In P-8, P-20, P-30 and SL lands, the δ¹³C-SIC values showed little vertical variation throughout the 0–100 cm soil layers (Table 5). Among the four sample plots, the δ¹³C-SIC values in SL land were the highest in all five soil layers, and δ¹³C-SIC value decreased with plantation age after afforestation. At 0–80 cm, the δ¹³C-SIC values in P-30 land were significantly lower than those in P-20 land, which in turn were lower than those in P-8 land. At 80–100 cm, δ¹³C-SIC value in P-30 land was also the lowest, but no difference was observed at this layer between P-20 land and P-8 land. The δ¹³C-SOC values within the 0–60 cm depth showed a gradual decrease with plantation age after afforestation. At 60–100 cm, the δ¹³C-SOC values were not significantly different between SL land and P-8 land, but these values in the both plots were dramatically higher than those in P-20 land and P-30 land (Table 5).

Figure 3 shows a strong correlation between SIC content and δ¹³C-SIC. Using all 260 samples, the relationship between δ¹³C-SIC content and SIC was shown to fit a linear model, and δ¹³C-SIC was observed to explain more than 70% of the variation in SIC (R² = 0.71, P < 0.01). Our data also showed that the variations in SIC and δ¹³C-SIC were related to SOC and δ¹³C-SOC. There was a positive linear relationship (R² = 0.52, P < 0.01) between SOC and SIC content for all soil samples (Figure 4). The entire dataset (260 samples) exhibited a positive correlation between δ¹³C-SOC and δ¹³C-SIC (R² = 0.63, P < 0.01, Figure 5). Additionally, there was no obvious correlation between silt particle content and SIC content (Figure 6A) or between clay particle content and SIC content (Figure 6B).

Our results showed that the SIC stock at depth of 0–100 cm in SL was 34.2 Mg ha^-1 and that it gradually increased along the chronosequence of afforestation (Figure 2). The results were consistent with those reported by Su et al. (2010) and Li Y.Q. et al. (2013), who also observed that SIC increased markedly with plantation age after afforestation on SL. However, our findings were in disagreement with some earlier reports in semiarid regions. In the Columbia Plateau, Oregon, United States, after 10 years, poplar plantations in a desert reduced the SIC concentration from 2.6 to 1.2 g kg^-1 in the surface layer (Sartori et al., 2007). In the Chinese Loess Plateau, Wang et al. (2016) reported that the SIC storage at depth of 0–100 cm in the farmland was significantly lower than that in the restored artificial forestland, with a difference of 16.8 Mg ha^-1. The SIC reduction in these inconsistent findings was mainly caused by irrigation or surface runoff, which can remove mass containing dissolved inorganic carbon. In the present study, similar processes would not be applicable because there was no irrigation or heavy rainfall. Therefore, our findings indicate that afforestation on shifting SL has a high potential to sequester SIC in degraded semiarid regions.

One theory posits that soil fine particles may play an important role in SIC sequestration following afforestation (Li et al., 2012). Plant canopies can intercept and deposit fine particles from the wind-sand flow after afforestation. This sediment contains rich carbonate sources, such as calcite, and causes a rapid SIC accumulation in surface soil (0–20 cm) (Wang et al., 2006). However, we found that this theory could not provide a complete explanation for SIC accumulation. Afforestation on SL not only elevates SIC stock in the surface soil layer, but also increases SIC levels in the deeper layers (Table 4; Li Y.Q. et al., 2013). Nevertheless, afforestation enhanced fine particles only at a depth of 0–40 cm, but not in the 40–100 cm depths (Table 3). In the deep layers (>40 cm), soil fine particles stack at an exceptionally slow rate and contribute little to SIC sequestration (Li et al., 2007). Moreover, we detected no correlation between fine particles and SIC content in the present study (Figure 6), further suggesting that the contribution of soil fine particles by the canopy to SIC sequestration is limited for the 0–100 cm soil layer. This phenomenon indicates that SIC sequestration is not exclusively derived from fine particle deposition and that other SIC accumulation processes may be occurring after afforestation.

We found that δ¹³C-SIC decreased with plantation age in forestlands (Table 5). Wang J.P. et al. (2015) found that the δ¹³C-SIC for desert soil was significantly higher than that for shrubland soil on the northeastern edge of the Taklamakan Desert, China. Liu et al. (2014) also pointed out that the δ¹³C value of soil carbonate along a chronosequence decreased gradually with vegetation restoration. SIC is composed of the LIC and PIC, which have distinct δ¹³C-SIC values. The changes in δ¹³C-SIC following vegetation rehabilitation can be used to explain the reason for SIC variation (Stevenson et al., 2005). There is sufficient evidence that the decrease in δ¹³C-SIC indicates PIC formation when land use patterns change (Jin et al., 2014; Liu et al., 2014; Wang et al., 2014, 2016; Wang J.P. et al., 2015; Bughio et al., 2016). Accordingly, the decrease in δ¹³C-SIC with plantation age in our study indicates that afforestation induced abundant PIC formation. Furthermore, a strong negative linear relationship between δ¹³C-SIC and SIC content in our study (Figure 3), which was also observed by Wang X.J. et al. (2015) in the northwest China, suggests that a decreasing δ¹³C-SIC is associated with SIC sequestration following afforestation. Specifically, PIC formation is accompanied by SIC sequestration, as the decrease in δ¹³C-SIC is indicative of the formation of PIC. Therefore, the carbon isotope data in this study indicate that SIC sequestration is probably caused by PIC formation after afforestation on SL. Additionally, an estimation of the amount of PIC would be very important to better understanding the contribution of PIC to SIC sequestration. Based on the precise δ¹³C-SIC, δ¹³C-PIC and δ¹³C-LIC values and empirical formulas, Wang et al. (2014) successfully estimated the accumulation rate of PIC under fertilization for loess soil. This method can ostensibly be used to calculate the amount of PIC in the forestlands in our study. However, an accurate δ¹³C-LIC value in the desert cannot be measured with the current technology, so we cannot supply values for the PIC stocks in this study. The δ¹³C-LIC of desert soil should be precisely identified in future studies because it is crucial for quantifying PIC stock.

In this study, afforestation simultaneously enhanced SIC and SOC contents (Figure 2), and SIC content was positively correlated with SOC content (Figure 4). Similar relationships have also been identified in other arid and semiarid regions in China (Zhang N. et al., 2010; Wang X.J. et al., 2015; Guo et al., 2016). These results suggest that the increase of SIC following afforestation may be related to SOC accumulation. Furthermore, our results showed that there was a decrease in both δ¹³C-SIC and δ¹³C-SOC with plantation age. δ¹³C-SIC was strongly positively correlated with δ¹³C-SOC (Figure 5), a finding that is consistent with the observations of Landi et al. (2003). In other words, the decrease in δ¹³C-SIC was accompanied by a decrease in δ¹³C-SOC. The decrease in δ¹³C-SIC indicates PIC formation, and SOC accumulation invariably leads to a decrease in δ¹³C-SOC due to plant litter input (Trolier et al., 1996; Jin et al., 2014). These results further imply that the PIC formation following afforestation is correlated with SOC accumulation. Soil organic matter affected PIC formation by regulating soil CO₂ concentration and the precipitation of carbonate in the alkaline environment (Monger et al., 2015). PIC accumulation involves two main reactions:

A mass of CO₂ is released into the soil following shrub and tree plantation in deserts, mainly due to the decomposition of the increased amount of organic matter (Zhang Z.S. et al., 2013). In general, an increase in soil CO₂ concentration would lead to the production of HCO₃^-. The accumulated HCO₃^- can drive reaction (4) to the right, resulting in the precipitation of carbonate (Wang X.J. et al., 2015; Zamanian et al., 2016). When 2 mole of CO₂ is consumed, 1 mole of CaCO₃ is generated. At our study site, the soil has a pH greater than 8 (Table 3) and is rich in available Ca²⁺ and Mg²⁺ (Table 1). The decomposition of the increased SOC in forestlands would dramatically elevate the soil CO₂ concentration and facilitate the occurrence of reaction (3). The alkaline environmental conditions could neutralize the H⁺ from reaction (3), which may be the reason for the decline in pH in forestlands (Table 3). These conditions also continuously promote the formation of HCO₃^-. The newly generated HCO₃^- combined with available cations may cause PIC accumulation following afforestation (Meyer et al., 2014; Monger et al., 2015). In addition to the CO₂ emitted via decomposition of the increased SOC, soil CO₂ respired by the roots of poplar trees (autotrophic respiration) would affect the formation of PIC. The effects of autotrophic respiration on PIC formation in plantation lands need to be studied in future. Additionally, a long-term study by observing SIC, SOC, soil carbon isotopes, soil CO₂ concentration and available cations in the same forestland is required, which could more directly and precisely characterize the mechanisms of SIC variation along a chronosequence of afforestation.