Divergent recovery trajectories of soil organic and inorganic carbon following artificial grassland restoration in semi-arid mining ecosystems

In arid mining ecosystems, the recovery of soil carbon pools following vegetation restoration is complex, particularly regarding the often-overlooked interplay between soil organic carbon (SOC) and inorganic carbon (SIC). Here, this study investigated a 4-year chronosequence of artificial grasslands sown with legumes to assess the restoration trajectories of SOC and SIC. Integrated Bayesian and correlation analyses were used to disentangle the driving mechanisms. The results showed that SOC and SIC exhibited increasing trends at 0–30 cm soil layer but followed different trajectories. The SOC accumulated gradually with increasing below-ground biomass, which promoted soil organic matter inputs. Conversely, the SIC exhibited a unimodal trend regulated by the soil moisture content and rhizosphere-mediated pH shifts rather than direct plant biomass inputs. These findings indicate that SOC and SIC show close coupling but divergent adjustment pathways during the 4-year restoration period, rather than synchronous and linear increase. This study highlights that assessing soil carbon pool restoration in arid mining ecosystems requires consideration of multiple soil carbon reservoirs, improving the understanding of soil carbon responses to vegetation restoration during early restoration.

Soil carbon is an important component of terrestrial ecosystems, supporting both soil functioning and climate regulation (Delgado-Baquerizo et al., 2017). As a key determinant of structure and fertility, it promotes aggregate stability, enhances water retention, and sustains microbial activity, driving vegetation growth and ecosystem stability (Hatano et al., 2024; Narayanan et al., 2024). In arid and semi-arid ecosystems, soil structure is vulnerable to disruption due to limited water availability, and soil carbon is critical for maintaining physical integrity and resisting degradation (Ayangbenro and Babalola, 2021; Niu et al., 2025). Even slight changes in carbon pools can induce a series of effects, including soil environment, plant performance (growth and development), and erosion susceptibility. Therefore, soil carbon contents and stocks are widely used to evaluate ecosystem state, function, and evolutionary trends toward a natural reference state.

Mining activities induce severe disturbance to soil carbon pools by impacting physicochemical properties. For example, mining terminates organic inputs due to vegetation and topsoil stripping, resulting in soil carbon being exposed to oxidation, leaching, and erosion (Moghimi Dehkordi et al., 2024; Punia and Bharti, 2023). Large-scale excavation and reconstruction often compact the soil into a dense and highly heterogeneous base. Such scenarios not only disrupts the soil pore network, but also hinders growth of plant roots and microbial activity (O’Brien et al., 2017; Wang et al., 2022). Studies have shown that due to heavy machinery transportation, soil bulk density can increase to 1.5 g cm –3 and porosity can decrease by over 15%, which creates severe physical barriers to natural restoration processes (Zhao et al., 2013). Disturbance induces aggregate breakdown and destruction of mineral-organic associations, accelerating SOC mineralization and destabilization, leading to rapid carbon pool depletion (Six and Paustian, 2014). The SOC losses in disturbed ecosystems exceed 60% compared to undisturbed ecosystems (Akala and Lal, 2001). Furthermore, bare surfaces in mining regions are highly susceptible to wind and water erosion. Fine sediments are first removed, resulting in a thinned soil profile and soil carbon depletion, such as particulate organic carbon and mineral-associated organic carbon (Ramli et al., 2020; Song et al., 2024). Soil inorganic carbon (SIC) dynamics is also profoundly affected by mining activities and plays an important yet often overlooked role in dryland carbon budgets. After mining, the changes in soil moisture, pH, and salinity impact the carbonate dissolution-precipitation balance, thereby reshaping SIC’ vertical and lateral redistribution (Shokri et al., 2024; Walter, 2018). These processes not only induce carbon loss but also drive complex reorganization, compromising soil buffering capacity and long-term stability. Consequently, mining disturbances trigger substantial declines or redistribution of both SOC and SIC, undermining soil multifunctionality (Li et al., 2024).

Due to water limitations, natural recovery in arid mining areas is extremely difficult. Therefore, artificial restoration provides a feasible strategy for initiating a return to natural soil functions (Dong et al., 2024). Beyond site-level benefits, such restoration activities are increasingly critical within the framework of global climate policy. Under Article 5 of the Paris Agreement, nations are encouraged to conserve and enhance natural carbon sinks and reservoirs. Similarly, the UN Decade on Ecosystem Restoration (2021–2030) underscores the necessity of reclaiming degraded landscapes to facilitate atmospheric CO 2 capture. Post-mining lands, which occupy vast areas globally, represent significant but underutilized potential for the achievement of regional carbon neutrality goals. Establishing artificial grasslands (AG) can rapidly increase vegetation coverage and above-ground biomass, and regulate soil microclimates via litter deposition (She et al., 2024). Notably, this recovery is achieved through internal biological mechanisms, such as leguminous nitrogen fixation, without the need for additional external mineral or organic amendments, allowing the soil to follow a natural successional trajectory. The penetration and expansion of plant roots reduce soil compaction and increase porosity, optimizing water infiltration and retention (Pandey and Bennett, 2024). Such physical ameliorations promote aggregate formation and physical protection, conducive to SOC accumulation. The restoration of the soil environment effectively enhances microbial activity, which ensures the rapid decomposition of organic residues and promotes the accumulation of organic carbon. Restored grasslands exhibit higher SOC content than bare mine land, with strong depth gradients reflecting biological inputs in surface layers (Pan et al., 2024; Wei et al., 2012). Furthermore, vegetation restoration also regulates SIC dynamics by regulating the rhizosphere microenvironment, including carbon dioxide concentration, pH, and ion composition (Dong et al., 2024; Li C. et al., 2023). A study indicates that grassland transformation increases SIC content by approximately 6% in a mining area on the Loess Plateau (Zhao et al., 2024). These SIC responses are not necessarily monotonic over time, because the balance between root-mediated carbonate dissolution and moisture-facilitated precipitation can vary with grassland development, potentially producing a non-linear, unimodal trajectory during early restoration. Consequently, vegetation restoration modifies not only SOC stocks but also SIC distribution, especially in carbonate-rich arid mining substrates (Ascenzi et al., 2025). Leguminous plants are frequently employed as ecological catalysts. They enhance productivity by improving nitrogen supply and stimulating microbial growth, promoting the site’s transition toward an independent, natural state (Jia et al., 2020; Tao et al., 2024). Existing restoration studies often focus on surface productivity and improvement of soil quality and aggregate stability. However, the dynamics and potential mechanisms by which artificial grasslands facilitate a return to natural SOC and SIC trajectories remain poorly understood, especially during the critical early stages of restoration in arid mining environments.

Therefore, newly reclaimed land (CK) and 1-year, 2-year, 3-year, and 4-year artificial grasslands (AG1, AG2, AG3, AG4) sown with leguminous plants (Astragalus adsurgens and Medicago sativa) were selected to evaluate the restoration of soil carbon pools in an arid mining area. Given soil and vegetation properties are often intercorrelated during early restoration, we combined a Bayesian multivariable framework with Pearson correlation analysis to explore the associations between carbon pools and their potential drivers. The study aimed to: (1) quantify the SOC and SIC contents and stocks across the restoration chronosequence; (2) identify the key drivers influencing soil carbon pools; and (3) discuss the ecological implications of vegetation restoration in mining areas.

Study area

The study area is located in Jungar Banner, Ordos City, Inner Mongolia Autonomous Region, China (39°41′5″ N, 110°16′3″ E) (Figure 1). It lies along the western bank of the Yellow River at the northern margin of the Loess Plateau. The region is a representative coal mining resource area containing extensive waste dumps formed by long-term mining activities. Many waste dumps have been reclaimed by planting forage grass. The waste dumps used here were formed by the stacking and leveling of overburden and waste rock from open-pit coal mining. The restoration age was defined as the number of years since sowing, so that 1-, 2-, 3- and 4-year artificial grasslands (AG1, AG2, AG3, and AG4) correspond to plots sown 1, 2, 3, and 4 years before sampling, respectively, and CK denotes newly reclaimed land prepared for sowing. Elevation of study area ranges from 1,026 to 1,304 m, decreasing gradually from northwest to southeast. The area exhibits a semi-arid mid-temperate continental climate. The mean annual temperature is 7.2 °C, with a maximum of 38.3 °C, and a minimum of −30.9 °C. Mean annual precipitation is 404.1 mm, whereas potential evaporation reaches 2082.2 mm. Seasonal freezing persists from late October to late April, with a maximum frost depth of approximately 1,490 mm and snow depth of about 150 mm. Winters and springs are cold, dry, and windy, summers are characterized by abundant but brief precipitation, and autumns are cool and short. Vegetation is classified as warm-temperate grassland, characterized by sparse and low-statured plant coverage. Shrubs are predominantly Caragana intermedia, Hippophae rhamnoides, and Artemsia gmelinii, while herbaceous plants are primarily established through artificial seeding, dominated by Astragalus adsurgens, Medicago sativa, and Artemisia desertorum. Dominant soils are Arenosols and Cambisols (WRB, 2022), exhibiting a uniform distribution across the study area. These soils are nutrient-poor, structurally unstable, and highly susceptible to erosion, driving severe soil loss. Based on meteorological records during this research period, the temperature and precipitation patterns were basically consistent with the regional average over the past 30 years. During this period, no catastrophic weather events were recorded, such as unprecedented floods or extreme temperature anomalies, which could disrupt the succession and development of the soil carbon pool.

Field investigation and sampling

The vegetation survey and soil sampling were carried out in CK, and in AG1, AG2, AG3 and AG4. All restoration stages were selected from the same abandoned mining area (within 2 km), and they had similar topography, soil conditions, as well as parent material and substrate composition. Because the restoration stages are represented by spatially separated plots, the chronosequence design substitutes space for time, and site-to-site variability cannot be entirely excluded. The AG1 to AG4 were sown primarily with Astragalus adsurgens and Medicago sativa, which are widely applied for the restoration of mining areas. No topsoil was applied, and no external fertilizers, irrigation, or other soil amendments were used during the restoration period, so that the observed soil carbon dynamics reflect internal biological processes rather than exogenous inputs. All restored plots were established using a manual strip sowing method with consistent row spacing to ensure uniform planting density. The seeds of the two plants were sown in a 1:1 ratio at a combined seeding rate of 66.7 g m –2. This relatively high seeding rate was specifically implemented to counteract the low germination and survival rates associated with the severely disturbed, structurally damaged, and nutrient-poor soil conditions characteristic of arid mining areas.

The growing season of the study area was from April to September. Therefore, the vegetation survey and soil sampling were conducted at the peak of biomass (in early September). On the newly reclaimed land and each artificial grassland, five sample plots (1 m × 1 m) were randomly selected, with an interval of approximately 200 m between the plots, to investigate the coverage (VC) and height. Subsequently, all aboveground plant material within each sample plot was collected and then labeled to calculate above-ground biomass (AGB). To determine below-ground biomass (BGB), soil cores containing roots were collected using a soil auger (diameter: 9 cm) at 0–10, 10–20, and 20–30 cm depths. Undisturbed soil samples at the same soil depths were obtained using the cutting-ring method (100 cm 3) to determine soil moisture content (SMC), total porosity (TP), capillary porosity (CP), non-capillary porosity (NCP), and bulk density (BD). Additionally, soil samples were collected to analyze soil pH and soil carbon content, including soil organic carbon (SOC) and inorganic carbon (SIC). During the sampling process, visible roots, stones, and other debris were removed by sieving through a 2-mm mesh. Therefore, coarse fragments (>2 mm) were virtually absent in the reclamation substrate, the reported carbon contents and storage values refer to the fine-earth (<2 mm) fraction. In this study, five sampling plots were investigated for each restoration stage, and soil samples were collected from three depth intervals at each plot.

Parameter testing

In the laboratory, soil samples containing roots were carefully washed with clean water to remove adhering soil particles and obtain the plant roots. The isolated root tissues and aboveground plant materials were then oven-dried at 65 °C for 72 h until constant weight to determine AGB and BGB. Within 3 days after sampling, the BD, SMC, and soil porosity indices were determined using the cutting ring method (Equations 1–5), following the procedures described by Zhao et al. (2025). Specifically, cutting rings with fresh soil were weighed first after removing the upper covers (W1). The cutting rings were then immersed in water (48 h), ensuring the water did not exceed the upper edge to achieve full soil saturation. After saturation, the cutting rings were taken out, the bottom covers were detached, and the samples were reweighed. Subsequently, the rings were placed on dry soil of the same texture, and the corresponding weights were recorded after drainage for 2 h and 6 h. Based on the above procedures, the TP, CP, and NCP were calculated. Finally, the soil cores within cutting rings were oven-dried at 105 °C to a constant weight (W2). The BD was calculated from the oven-dried soil weight and the known cutting ring volume. The SMC was determined from the difference between the fresh weight and the oven-dried weight. The relevant formulas are as follows: where W 1 is the weight of ring + wet soil, g; W 2 is the weight of ring + dried soil, g; W 0 is the weight of cutting ring, g; and C W C is the capillary water capacity measured when cutting ring was placed on dry soil for 2 h; ds is the soil particle density (2.65 g cm –3).

Soil pH was measured using a soil-water suspension (1:5, w/v). Air-dried soil (10 g) was placed in a beaker and distilled water was added. The soil-water mixture was agitated vigorously and equilibrated for 30 min to measure soil pH via a calibrated pH meter (PB-10, Sartorius). For carbon analyses, dried soil samples were ground and sieved through a 0.15 mm sieve to obtain target samples. The SOC was determined by the potassium dichromate oxidation with external heating. The SIC was measured using the calcimeter method, based on the volumetric quantification of CO 2 released after acidification with HCl solution (0.5 mol L –1). Furthermore, the soil organic carbon storage (OCS) and inorganic carbon storage (ICS) were calculated following Equation 6: where C S i indicates the carbon storage at ith soil layer (t hm –2); C i represents the organic and inorganic carbon content at ith soil layer (g kg –1); B D represents the soil bulk density at ith soil layer (g cm –3); and d i represents the soil depth at ith soil layer (cm).

Statistical analysis

In this study, analyses and graphical visualization were completed using R (v. 4.5.1, R Core Team). One-way ANOVA followed by Tukey’s HSD test was performed to evaluate the significant differences in vegetation characteristics and soil physicochemical properties among CK, AG1, AG2, AG3, and AG4. All data are expressed as mean ± standard error (SE), and significance was set at P < 0.05. This analysis was conducted using the “agricolae” package (v. 1.3–9). Pearson correlation analysis was used to examine the bivariate associations between each carbon pool and the soil and vegetation parameters. However, bivariate correlation analysis, although widely used, assesses each driver in isolation and cannot control for the strong collinearity among vegetation and soil properties that typifies short restoration chronosequences, and may therefore misattribute effects to spurious pairwise associations. Therefore, this study applied a Bayesian multivariable modeling framework, which jointly estimates the independent influence of each predictor on a given carbon pool after accounting for all other covariates. After standardizing all variables using the Z-score, a Bayesian linear model with a Gaussian response distribution was fitted separately for OCS and ICS pool using the “brms” package (v. 2.23.0), with the plant and soil parameters (e.g., biomass, BD, pH, SMC) included as predictors. Posterior sampling was performed with Markov Chain Monte Carlo (MCMC) methods using four independent chains with 2000 iterations each. For each predictor, the posterior mean, the 95% credible interval, and the probability of direction (Pd), defined as the posterior probability that a coefficient is positive or negative, were computed, and predictors were ranked by the absolute value of the posterior mean. These analyses and figures were accomplished using the “tidyverse” (v. 2.0.0), “ggridges” (v. 0.5.7), and “bayestestR” (v. 0.17.0) packages.

Changes in vegetation characteristics and soil properties across the restoration chronosequence

Clear variation in vegetation characteristics and soil properties was observed across the restoration chronosequence (Figures 2, 3). Vegetation height reached 90.6 cm in 1-year artificial grasslands (AG1), but decreased by 68.0% in 2-year artificial grasslands (AG2; Figure 2a) (P < 0.05). Subsequently, height in 3-year artificial grasslands (AG3; 42 cm) and 4-year artificial grasslands (AG4; 64 cm) increased by 44.8% and 120.7% compared to AG2, respectively (P < 0.05). After vegetation restoration, vegetation coverage (VC) was lowest in AG1, reaching 22% (Figure 2b). VC in AG2, AG3, and AG4 gradually increased (52%, 62%, and 70%, respectively), increasing by 136.4%, 181.8%, and 218.2% compared to AG1, respectively. Above-ground biomass (AGB) in AG1 and AG2 was 64.1 g m –2 and 147.9 g m –2, respectively (Figure 2c). Compared to AG1 and AG2, AGB was 8.7- and 3.8-fold in AG3 (558.0 g m -2) and 8.2- and 3.6-fold in AG4 (525.1 g m -2), respectively (P < 0.05). Below-ground biomass (BGB) declined from 391.9 g m -2 in AG1 to 225.9 g m -2 in AG3, corresponding to a cumulative decrease of 42.4%, although the differences were not significant (P > 0.05). By AG4, BGB increased significantly to 770.31 g m –2 (P < 0.05), a 241.0% increase compared to AG3 (Figure 2d).

The bulk density (BD) in AG1 to AG4 at 0–30 cm was lower than that of the newly reclaimed land (CK, 1.59–1.68 g cm –3) (Figure 3a). With increasing restoration age, BD showed a gradual upward trend. In AG4, BD was slightly higher than that in AG1. The temporal variation of soil moisture content (SMC) was consistent with that of BD (Figure 3b). After restoration, soil pH showed an overall increase, while no significant difference was observed among artificial grasslands (Figure 3c; P > 0.05). Total porosity (TP) increased following vegetation restoration across the study plots, with a more pronounced increase at 0–10 cm, ranging from 5.5% to 14.3% (Figure 3d; P < 0.05). Capillary porosity (CP) showed a unimodal pattern along the restoration chronosequence. At 0–10 cm layer, CP in AG4 was slightly lower than in CK, showing a 4.2% decrease (Figure 3e). The variation pattern of non-capillary porosity (NCP) was consistent with CP, but at 20–30 cm, NCP in AG2 was 12.4% lower than that in CK, despite no significant differences (Figure 3f; P > 0.05).

Changes in soil organic carbon and inorganic carbon across the restoration chronosequence

The soil organic carbon (SOC) content increased across 0–30 cm depth compared to CK after restoration, but the differences were not statistically significant (P > 0.05). Across different soil depths of the study area, SOC in artificial grasslands increased by 46.2%–146.2% at 0–10 cm, 6.2%–115.1% at 10–20 cm, and 3.9%–107.7% at 20–30 cm (Figure 4a). Soil inorganic carbon (SIC) increased following restoration and presented a unimodal distribution. In AG4, SIC (0.59 g kg –1) was 18.6% and 10.5% lower than in AG1 at 0–10 cm and 20–30 cm soil depths, respectively (Figure 4b). Both organic carbon storage (OCS) and inorganic carbon storage (ICS) in the 0–30 cm layer increased after vegetation restoration (Figure 5). Compared with CK (17.6 t ha -1 and 2.8 t ha -1), soil OCS increased by 7.4%–100.6% (Figure 5a), while soil ICS increased by 7.1%–53.6% (Figure 5b).

Association patterns between carbon pools and vegetation–soil variables

Bayesian posterior analysis and Pearson correlation analysis were used to characterize association patterns between vegetation-soil variables and carbon pools within the restoration chronosequence (Figure 6). For OCS, the Bayesian model provided the strongest posterior support for a positive association with pH (Pd = 96.4%) and a negative association with CP (Pd = 95.8%). AGB showed weaker posterior support for a negative association, whereas SMC showed moderate support for a positive association (Figure 6a). Pearson correlation analysis showed that OCS was positively correlated with BGB and pH, but negatively correlated with CP (Figure 6b). For ICS, the Bayesian model showed the posterior support for a negative association with BGB (Pd = 79.5%) and a positive association with SMC (Pd = 73.0%) (Figure 6c). Correlation analysis revealed that ICS was negatively correlated with BGB and BD, while positively correlated with SMC and AGB (Figure 6d). Some relationships, such as that between BGB and OCS, showed inconsistent directions between the multivariable Bayesian model and the bivariate correlations, suggesting that the simple positive correlation may partly reflect confounding with other covariates rather than a direct effect. The two analyses consistently indicated that OCS and ICS responded differently to vegetation and soil factors during the four-year restoration period.

Vegetation input promotes the accumulation of organic carbon

In many severely disturbed landscapes, vegetation restoration increases soil organic carbon (SOC) content, reflecting the sensitivity of carbon pools to plant establishment (Chen et al., 2024; Xu et al., 2025). In this study, compared to newly reclaimed land, the establishment of artificial grasslands was associated with increases in SOC content and storage, although these were not statistically significant within the 0–30 cm layer (Figures 4, 5; P > 0.05). SOC increases during the early restoration stages were more closely linked to vegetation-derived carbon inputs (Figure 2). This result indicates that biological productivity is the fundamental engine of organic carbon sequestration in the early restoration phase. These scenarios are consistent with patterns observed in mining and other degraded lands where SOC responded rapidly to vegetation cover reestablishment (Ma et al., 2024a; Misebo et al., 2025). As above-ground biomass increases, litter deposition accumulates. Concurrently, roots expand and contribute organic substrates through turnover and rhizodeposition (Wang et al., 2021; Xu et al., 2021). The positive correlation between OCS and below-ground biomass reinforces the role of root-derived carbon in stabilizing the organic pool (Figure 6B). The transient decline in BGB from AG1 to AG3 may reflect the natural replacement of the initial sown cohort by more productive perennial growth, after which below-ground biomass increased markedly in AG4. Notably, the introduction of legumes, such as alfalfa and Astragalus adsurgens, likely acted as catalysts in carbon accumulation. Leguminous plants not only produce high-quality debris but also promote symbiotic nitrogen fixation, which helps alleviate the severe nitrogen deficiency in mining areas (Man et al., 2022). This mechanism may reduce the soil carbon-nitrogen ratio and stimulate microbial activity, potentially promoting the conversion of plant-derived carbon into stable soil organic matter (Guo et al., 2025; Wang et al., 2025). The Bayesian model and correlation analysis identified soil pH as the most potent positive driver of OCS accumulation (Figures 6a,b). This suggests that in these alkaline soils, the moderation of pH levels associated with plant-soil feedbacks may create a more favorable environment for microbial carbon processing and the persistence of organic matter. Numerous studies further indicate that vegetation restoration consistently increases SOC and associated nutrient concentrations (i.e., nitrogen, phosphorus) across post-mining lands (Xiao et al., 2022; Xu et al., 2024). Therefore, the increase in soil organic carbon content and reserves is more closely related to the input brought by vegetation (especially underground biomass).

The improvement of the soil environment induced by vegetation restoration has a potential positive effect on soil organic carbon accumulation. Changes in soil physical properties, such as aggregate formation and mineral organic binding, can support soil organic carbon restoration in the long term (Ponyane et al., 2025). Although root penetration changes porosity and bulk density, these short-term physical changes do not directly promote SOC storage (Robinson et al., 2022). The gradual increase in bulk density with restoration age likely reflects the continued settling and consolidation of the reconstructed waste-dump substrate rather than a decline in soil quality, because bulk density in all artificial grasslands remained below that of the newly reclaimed land. The Bayesian analysis revealed that capillary porosity acted as a dominant negative constraint on OCS, which was further supported by their significant negative correlation (Figures 6a,b). While restoration increases soil porosity, excessive capillary pores might facilitate higher mineralization rates or oxygen diffusion, potentially counteracting carbon sequestration in the early stages (Kravchenko and Guber, 2017). This indicates that the initial reorganization of soil structure may prioritize microbial decomposition pathways over immediate physical protection. Nevertheless, the evolution of soil pore structure can alter moisture and aeration conditions, influencing microbial turnover and promoting the transformation of unstable carbon into more persistent aggregates over longer temporal scales (Schluter et al., 2022; Xia et al., 2022). Exudates and fine debris can serve as substrate for microbial communities, stimulating enzymatic processes related to carbon conversion and stabilization during plant growth (Chen and Liu, 2024; Wu et al., 2024). Mine reclamation studies have proven that vegetation restoration increases functional enzyme activities along with SOC, highlighting the coupling between biological activity and carbon accumulation (Ning et al., 2022; Sun et al., 2021). In long-term desert revegetation studies, vegetation restoration has been shown to change SOC fraction composition, with mineral-associated organic carbon becoming more prominent over time (Li Y. et al., 2023; Zhang et al., 2023). These results indicate a progression from plant-derived to more structurally protected carbon pools. Early plant inputs dominate surface SOC increases, while longer-term processes facilitate deeper and more stable carbon storage.

After vegetation reconstruction, although soil organic carbon increased in the initial few years (Figure 4a), the balance between input and conversion processes may change with restoration progresses. A study reveals that SOC generally increases with the restoration sequence, though the change extent decreases with depth, reflecting interactions among vegetation type, climate, and soil conditions (Pan et al., 2025). Our findings indicate that in arid mining areas, SOC dynamics are not only determined by biomass, but are regulated by the interplay between pH-mediated chemical environments and the soil pore network. Therefore, understanding early SOC dynamics requires combining plant carbon input pathways with soil structure across temporal scales.

SIC variation was associated with rhizosphere and moisture condition

Soil inorganic carbon serves as an important but often overlooked component of soil carbon pools in arid and semi-arid regions, where carbonate accumulation and dissolution can substantially influence total carbon storage. In this study, after vegetation restoration, the soil inorganic carbon at 0–30 cm depth showed a unimodal trend along the restoration chronosequence (Figure 4b). The Bayesian model identified BGB as the negative predictor of ICS, whereas SMC was the primary positive predictor (Figures 6c,d). Such patterns are often reported in dry-land ecosystems, where soil inorganic carbon dynamics are strongly regulated by biological activity and soil chemical conditions (Li C. et al., 2023; Zhang et al., 2020).

Vegetation restoration can influence SIC dynamics through multiple and sometimes opposing pathways. Root growth and associated belowground biomass inputs may change the soil chemical environment by altering rhizosphere CO 2 concentrations and pH (Bach and Gojon, 2023; Yang et al., 2023). Although high root activity may initially promote carbonate dissolution via respiratory CO 2 and organic acid release, it may simultaneously facilitate carbonate precipitation in the later stages by increasing alkalinity (Gatz-Miller et al., 2022; Hinsinger et al., 2003). Meanwhile, enhanced soil moisture conditions, could promote carbonate re-deposition and transport, contributing to the temporal variability of soil inorganic carbon responses observed during vegetation restoration (Hao et al., 2025; Ni et al., 2024). Similar relationships between root-derived CO 2, soil pH, and soil inorganic carbon have been recorded in restored grasslands and reclaimed lands, where intensified rhizosphere processes can accelerate carbonate turnover (Malhotra et al., 2025). However, the direction and magnitude of soil inorganic carbon changes exhibit significant differences among various studies. Some studies reveal that net soil inorganic carbon accumulation follows vegetation reestablishment, whereas others observe soil inorganic carbon depletion, which has been attributed to enhanced carbonate dissolution driven by biological activity (Ma et al., 2024b; Zhao et al., 2024). These results suggest that early-stage SIC variation in the chronosequence was more closely associated with rhizosphere chemical conditions and soil moisture than with short-term changes in soil physical structure.

The observed soil physical properties, such as capillary porosity, have a relatively weak influence on SIC (Figure 6). This is reinforced by the Bayesian result, in which physical factors like CP showed lower posterior probabilities (Pd = 52.1%) for ICS. Such results suggest that, over short restoration timescales, inorganic carbon dynamics are less constrained by physical structure than by chemical and biological drivers. In arid mining lands, where carbonate minerals are often abundant but poorly integrated into stable soil matrices (Batool et al., 2024), SIC may respond more readily to shifts in rhizosphere chemistry than to gradual improvements in soil structure (Qu et al., 2022). This result contrasts with organic carbon dynamics, which are more directly linked to biomass inputs and biological turnover.

Implications of vegetation restoration in mining areas

Early vegetation restoration was associated with gains in several measured soil properties, including carbon content and porosity, in degraded mining lands. Over a four-year period, through the introduction of leguminous species (e.g., Astragalus adsurgens and Medicago sativa), the organic and inorganic carbon in the 0–30 cm soil layer increased to varying degrees (Figure 4), but there were clear differences in the regulatory pathways between root activities and the soil environment. Meanwhile, gains were observed in some soil physical properties (e.g., porosity and moisture), indicating that biological restoration may contribute to soil carbon accumulation and the recovery of ecosystem services. This implies that even young plant cover may act as a shield against erosion and lay the groundwork for long-term recovery. However, the directions between BGB and OCS in the multivariate Bayesian model and bivariate correlation analysis are inconsistent (Figure 6), highlighting the complexity of carbon dynamics in the arid mining area ecosystem. This discrepancy also illustrates the analytical value of the Bayesian multivariable framework. Bivariate correlation captures only unadjusted pairwise associations, whereas the Bayesian model isolates the net effect of each driver after controlling for intercorrelated covariates, indicating that simple pairwise relationships can partly reflect confounding rather than direct regulation. Importantly, the introduction of legumes may accelerate early organic carbon accumulation while also improving vegetation establishment under harsh post-mining conditions. Artificial grasslands have the potential to enhance soil carbon retention and support early ecosystem recovery in arid mining lands. From a management perspective, the results support three strategies for post-mining reclamation in semi-arid areas. First, legume-based species mixtures can sustain early organic inputs and root development. Second, rhizosphere pH should be monitored to track the balance between SOC and SIC accumulation. Third, moisture-conservation practices such as appropriate sowing density and mulching can stabilize the rhizosphere conditions that favor carbon retention.

Furthermore, artificial grasslands may act as ecological catalysts, yet their long-term sustainability may depend on the balance between carbon sequestration and nutrient availability. Our findings indicate that the rhizosphere environment, particularly root-mediated pH shifts, may play a central role in carbon dynamics. However, the alkaline nature of these Arenosols and Cambisols may constrain nutrient solubility. While specific data for available phosphorus (P), potassium (K), and magnesium (Mg) were not quantified in this short-term chronosequence, regional studies on the Loess Plateau confirm that high pH levels typically lead to strong P-fixation and reduced bioavailability of essential micronutrients (e.g., iron and zinc) (Zhang et al., 2015). For leguminous species like alfalfa, the biological activity that drives carbon storage, including BGB-driven pH increases, could eventually impair symbiotic nitrogen fixation efficiency and biomass production, potentially decelerating the observed soil carbon accumulation rate in later restoration stages. Therefore, although the current study suggests substantial early gains in soil carbon pools, the potential constraints imposed by nutrient availability remain unresolved and should be considered when interpreting the sustainability of carbon sequestration.

In summary, this study tracked soil organic and inorganic carbon dynamics during the initial restoration phase, but there were also some limitations. The four-year observation period ignores the sustainability and long-term development trajectory of the carbon pool. Moreover, because the restoration stages are represented by spatially separated plots rather than by temporal monitoring of the same plots, the chronosequence substitutes space for time, and site-to-site variability may partly confound the inferred trajectories. The study evaluated the carbon changes in shallow soil (0–30 cm), ignoring the possible influence of deep root activities on soil carbon redistribution or mineral interactions. In addition, carbonate mineralogy, microbial processes, nutrient availability, and carbon fractions were not directly measured, so the mechanistic interpretations should be regarded as hypotheses pending direct measurements. Therefore, longer-term monitoring is needed to determine whether the carbon accumulation during the restoration process is continuous, decreasing, or stable. Expanding the sampling range to deeper soil layers helps capture the vertical redistribution of organic and inorganic carbon and more accurately quantify the total soil carbon pool. This will help obtain a more comprehensive understanding of how plant traits and root activities interact with soil processes in arid mining areas, thereby influencing short-term recovery and long-term carbon sequestration.

This study indicates that the establishment of artificial grasslands promoted the recovery of SOC and SIC pools in the 0–30 cm soil layer of an arid mining area over a four-year restoration chronosequence. Although the increases in SOC were not significant (P > 0.05), the consistent positive trends in both content and storage support an early trajectory of organic carbon recovery. SOC increases were primarily associated with vegetation-derived carbon inputs, particularly belowground biomass accumulation, indicating that biological inputs were the main driver of the short-term. In contrast, inorganic carbon followed a different recovery pathway and was more closely linked to soil moisture and root-mediated rhizosphere processes, especially pH changes, than to direct biomass inputs. These results suggest that OCS and ICS did not recover synchronously, but instead followed distinct yet interrelated pathways during early restoration. Therefore, evaluating carbon recovery in disturbed arid mining ecosystems should not rely solely on total soil carbon, but should explicitly distinguish between organic and inorganic carbon pools. Given the short restoration period and the restriction to shallow soil (0–30 cm), longer-term monitoring and deeper soil profiling are still needed to determine the persistence, vertical redistribution, and long-term stability of these restored carbon pools.