Unravelling depth-dependent pedoclimatic controls on measurable soil organic carbon fractions across climatic gradients in Australian agricultural soils

Abstract

Soil organic carbon (SOC) accumulation and persistence are predominantly governed by an intricate interplay among biomass inputs, microbial activity, and soil mineralogy, all of which are strongly modulated by climate-driven moisture availability. However, how climate gradients influence pedoclimatic controls on particulate organic carbon (POC) and mineral-associated organic carbon (MAOC) remains unclear, limiting reliable assessments of SOC vulnerability to climate change and the development of targeted sequestration strategies across contrasting climate zones. Here, using a continent-wide dataset from 2256 Australian farm paddocks across diverse climate zones (ranging from dry to very humid) and soil depths (0–10, 10–20 and 20–30 cm), we applied XGBoost and SHAP, coupled with piecewise structural equation modeling, to disentangle the key drivers and mechanisms governing POC and MAOC stocks in agricultural ecosystems. Our results show that land-use effects on SOC fractions varied with aridity and soil depth, with pastures (modified pastures including annual, perennial and mixed annual-perennial pastures dominated by grasses, legumes or their mixture) maintaining higher POC stocks across all depths in Mediterranean and semi-humid zones but mainly in the surface layers in both drier and wetter regions, while cropping systems (including cereals, oilseeds, grain legumes and other commercial crops) enhanced subsoil MAOC storage in humid climates. Total nitrogen (TN) was the predominant driver of both SOC fractions across most aridity-depth combinations, explaining 15–50% of spatial variation. The TN threshold for surface SOC accumulation increased from ~ 1.0 mg/g soil in dry zones to ~ 2.8 mg/g soil in very humid zones. TN was more influential in surface soils of relatively dry areas (34–50% for POC, 33–37% for MAOC) but shifted to subsoils in humid zones. POC was also strongly associated with Mean Annual Temperature (MAT) and pH under relatively dry climates and with clay content in surface soils, whereas MAOC was sensitive to both MAT and Mean Annual Precipitation (MAP) in dry zones and became increasingly influenced by oxide minerals with depth and humidity, occasionally surpassing TN in subsoils of relatively arid climates or in surface layers of humid regions. These findings underscore the need for climate-smart SOC management strategies: in dry zones, increasing organic inputs and improving soil structure are essential to overcome input limitations and minimize carbon losses, whereas in humid areas, enhancing surface MAOC formation and increasing subsoil carbon inputs is critical for promoting SOC accumulation and stability.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-38349-2.

Introduction

Soil organic carbon (SOC) plays a fundamental role in global carbon cycling through soil-climate feedback, with its distribution and stability strongly influenced by both climatic conditions and soil biogeochemical properties^1,2. Recent advances suggest that the sensitivity of SOC to climate is primarily governed by its stabilization through organo-mineral associations and physical protection within aggregates, rather than by its biochemical recalcitrance against microbial decomposition^3,4. Partitioning SOC into particulate organic carbon (POC) and mineral-associated organic carbon (MAOC) offers a valuable framework for assessing SOC vulnerability to climate change and developing targeted SOC sequestration strategies^5,6. These two fractions differ markedly in their formation pathways and persistence⁷. Specifically, POC is largely composed of plant-derived material⁸, relatively unprotected and readily accessible to microbial decomposition, though some can be stabilized through aggregate occlusion, resulting in a short mean residence time of years to decades⁹. In contrast, MAOC primarily originates from dissolved organic carbon and microbial residues that strongly associate with fine mineral particles^10,11, where mineral associations restrict microbial accessibility and contribute to its much longer persistence in soil, often lasting from decades to centuries¹². The relative distribution between these unprotected and mineral-protected carbon pools not only defines a soil’s carbon storage potential but also shapes its stability under shifting climatic conditions and anthropogenic pressures¹³.

Given their intrinsic differences in composition and persistence, SOC fractions are governed by distinct interactions among biomass inputs, microbial activity, soil biogeochemistry, climate, land use and management^14–17. POC is more associated with changes in organic matter inputs such as litterfall, crop residue incorporation and root exudation and is more vulnerable to microbial decomposition than MAOC¹, making it more sensitive to climate change, particularly warming^18,19, as well as to land use types (e.g., arable croplands, grasslands, forests and shrublands) and management practices such as residue retention, tillage intensity and fertilizer additions^16,20,21. In contrast, MAOC formation and stability rely on both organic inputs into the soil and carbon stabilization mechanisms through mineral association and physical protection^5,22. New evidence demonstrates that soluble plant-derived compounds can directly associate with mineral surfaces to form MAOC, rather than exclusively relying on the traditional pathway in which plant inputs are initially transformed by microbes before stabilization^23,24. To further understand these processes, it is critical to consider soil depth. Soil physical, chemical and biological dynamics vary with soil depth and their corresponding micro-environments, but little has been done to explicitly explore these differences that can significantly affect SOC dynamics. Process-based models such as RothPC-1, an innovative extension of RothC that simulates SOC dynamics beyond 30 cm, introduce additional depth-dependent parameters representing SOC movement and reduced decomposition with depth, enabling more accurate predictions of subsoil carbon dynamics²⁵. A recent study reported that the response of SOC fractions to climate change varies with soil depth⁵. Subsurface soils exhibited greater sensitivity to climate warming than surface soils, with accelerated decomposition rates observed for both MAOC and POC. This heightened vulnerability is likely due to energy limitations in deeper layers, where microbial activity is constrained by the lack of fresh organic inputs^26,27, as well as by oxygen limitations²⁸. The gap between the existing MAOC stock under current management practices and the maximum amount that the soil could store under unlimited carbon inputs given its finite mineral capacity is referred to as the carbon saturation deficit²⁹. Subsoils often have a larger carbon saturation deficit, which not only indicates a greater potential for additional MAOC formation but also a greater capacity to facilitate the transformation of POC into more stable MAOC^5,30. Despite growing recognition of depth-specific mechanisms, such as energy and oxygen limitations and greater carbon saturation deficits in subsoils, the depth-dependent patterns of pedoclimatic controls on measured SOC fractions remain poorly quantified across diverse environmental gradients.

Recent studies highlight the urgency of integrating aridity gradients into soil carbon stewardship frameworks to support the sustainable management and protection of SOC, as moisture is a primary constraint on both plant biomass input and microbial decomposition^31–34. Arid and humid ecosystems impose fundamentally distinct constraints on SOC dynamics. In arid regions, limited plant productivity directly restricts SOC accrual (carbon inflows into the soil system), but reduced microbial activity may allow more efficient preservation of organic inputs^31,35. Conversely, humid environments face greater SOC loss due to accelerated microbial decomposition³⁶ and a reduced capacity of soil minerals to stabilize additional organic matter compared with arid ecosystems³¹. Subsurface horizons often retain moisture longer than surface soils which are more prone to desiccation. In the temperate grain-growing regions of southern Australia, summer crop production often relies on subsoil water stored from winter rainfall due to limited in-season summer precipitation³⁷, raising questions about how deep soil water availability, oxygen limitation and low carbon inputs interact to constrain SOC accumulation at depth. Additionally, variations in mineral weathering intensity, mineralogical composition and the stability of mineral-organic associations may also contribute to the contrasting carbon dynamics observed in arid and humid environments¹⁰. Moreover, climate extremes such as drought further underscore the divergent responses of SOC fractions: in humid ecosystems, drought significantly reduces POC but not MAOC, whereas in drier regions, both fractions show minimal changes, likely due to lower baseline SOC concentrations (often < 4 kg C m^{− 2} in the top 30 cm, compared with 10–15 kg C m^{− 2} in humid systems³⁸, highlighting fraction-specific vulnerabilities along aridity gradients³². Given the contrasting constraints on SOC accrual and loss between arid and humid ecosystems, and their differential responses to climate change, aridity-informed management strategies may present a more targeted and potentially effective pathway for climate mitigation³¹. However, our understanding of the mechanisms governing different SOC fractions across aridity gradients, and quantification of these complex relationships remains limited, hindering efforts to develop robust climate-resilient carbon sequestration strategies tailored to local environmental conditions.

Here, we present a national-scale study based on the Australian National Soil Carbon Research Program (SCaRP) dataset, encompassing 6768 soil samples from 2256 farm paddocks (fenced and actively managed agricultural fields used for cropping or modified pasture systems) across diverse climatic, topographic, and edaphic conditions. Our study aimed to investigate how climatic conditions, soil properties, oxide minerals, land use and topography interact to regulate SOC fractions (POC and MAOC) across soil depths (0–10, 10–20 and 20–30 cm) and aridity gradients ranging from dry to very humid zones. We hypothesized that (1) climate variables have a stronger influence on SOC stocks in relatively arid zones due to lower soil moisture availability; (2) the nitrogen requirement for SOC accumulation increases under more humid climates, where higher productivity and decomposition rates intensify nitrogen demand; (3) soil oxide minerals play a more important role in SOC stocks in subsoil layers and humid climates, where energy limitations or high decomposition pressures may make mineral-associated protection the primary stabilization pathway. To capture both the relative importance and the functional pathways of these drivers, we applied an interpretable machine learning approach, the eXtreme Gradient Boosting (XGBoost) algorithm in combination with the Shapley Additive exPlanations (SHAP) method, to identify key drivers and quantify their relationships with SOC fractions, together with piecewise structural equation modeling (piecewiseSEM) to disentangle their direct and indirect effects in contrasting arid and humid environments. Collectively, these approaches help offer mechanistic insights into aridity-related carbon vulnerabilities and provide evidence-based guidance for tailoring soil carbon management strategies to regional conditions in the face of ongoing and future environmental change.

Materials and methods

Database preprocessing

The soil database for this study was derived from the Australian National Soil Carbon Research Program (SCaRP)³⁹, which employed the SOC fractionation approach of Baldock et al.⁴⁰. This approach quantified the following three biologically significant fractions: POC, the SOC associated with coarse particles (50–2000 μm) after subtraction of the biologically resistant poly-aryl carbon (PAC) fraction; MAOC, the SOC associated with fine particles (≤ 50 μm) after subtraction of PAC; PAC, a stable and charcoal-like aromatic and polyaromatic form present in both particle size classes, identified using solid-state¹³ C NMR spectroscopy. In the database, POC, MAOC and ROC were measured for 312 soil samples, while the remaining samples were predicted using mid-infrared (MIR) spectroscopy coupled with the partial least squares regression (PLSR) model⁴¹, given the labor- and time-intensive analysis. This study excluded ROC owing to its extreme stability and turnover times ranging from centuries to millennia, which require longer historical records than were available. Instead, the analysis focused on POC and MAOC, extracted exclusively from farm paddocks (spatial support of 25 × 25 m) under continuous cropping and pasture systems. These agricultural systems were broadly classified based on historical land management records, requiring a consistent land use for at least seven of the past ten years. Two soil depth stratification schemes were used in the database: 0–5 cm, 5–10 cm, 10–20 cm, 20–30 cm or 0–10 cm, 10–20 cm, 20–30 cm. To address inconsistencies in soil layer stratification, SOC stocks for both fractions were calculated using Eq. (1), and the SOC fraction stocks from the 0–5 cm and 5–10 cm layers were summed to obtain the SOC fraction stocks for the 0–10 cm depth interval, thereby standardizing the stratification into three uniform depth intervals: 0–10 cm, 10–20 cm, 20–30 cm. Auxiliary covariates representing soil properties, including pH, clay content, total nitrogen (TN) and oxide mineral concentrations (SiO₂, Al₂O₃, and Fe₂O₃), were also harmonized into standard depth intervals (0–10, 10–20 and 20–30 cm) using equal-area quadratic splines^42,43, with standard error-checking procedures applied to exclude unreliable entries. Where sampling coordinates were unavailable, regional center points were used to impute locations for the purpose of displaying sample distribution only. These steps follow the methodology described in our previous work², where full details are provided. The final dataset used in this study encompassed a broad range of climates, soil types, topographies and management conditions, comprising 6768 soil samples collected from 2256 farm paddocks across Australian agricultural regions. To investigate the spatial patterns of SOC fractions and their driving factors under varying climatic conditions, individual soil samples were assigned to six distinct climatic zones based on the De Martonne Aridity Index (AI_DM)⁴⁴: dry (AI_DM < 15), semi-dry (15 ≤ AI_DM < 20), Mediterranean (20 ≤ AI_DM < 24), semi-humid (24 ≤ AI_DM < 28), humid (28 ≤ AI_DM < 35), and very humid (AI_DM ≥ 35). The geographic distribution of sampling sites, along with their corresponding aridity zones and land use types, is shown in Fig. 1. The following equation⁴⁵ was used to calculate the SOC fraction stocks (POC or MAOC) involved in all subsequent analyses:

1

Fig. 1.

Geographic distribution of soil sampling sites covering diverse aridity zones in Australian agricultural systems. Different colors indicate sampling sites located in dry (red, n = 135), semi-dry (pink, n = 427), mediterranean (orange, n = 377), semi-humid (green, n = 506), humid (light blue, n = 482), and very humid (dark blue, n = 329) zones. Site markers differ by land use, with triangles representing continuous cropping system (n = 629) and plus symbols denoting pasture system (n = 1627).

where CF represents the concentration of SOC fractions (POC or MAOC), BD is bulk density, T is the soil layer thickness, P_g accounts for the gravel fraction, and P_rt accounts for the proportion of rocks or trees within the sampling.

Statistical analyses

To compare the distribution of SOC fraction stocks across different land uses and aridity zones, we employed both descriptive (Table S1) and multivariate statistical analyses. Statistical significance of group differences was assessed using non-parametric Kruskal-Wallis tests followed by Dunn’s post-hoc comparisons, given that the data did not satisfy assumptions of normality required for parametric tests. To further assess whether aridity zones and land uses structure distinct functional groupings, and to investigate the associations between SOC fractions, groupings, and environmental predictors, principal component analysis (PCA) combined with permutational multivariate analysis of variance (PERMANOVA)⁴⁶ based on Euclidean distance was employed. Homogeneity of multivariate dispersion was assessed using PERMDISP⁴⁷ and the results showed that group variances differed significantly (P < 0.05). Such heterogeneity is expected in soil datasets due to strong spatial and environmental variability, thus the PERMANOVA results are interpreted as reflecting a combination of group separation and dispersion differences. All statistical analyses were conducted using R packages “rstatix”⁴⁸, “factoextra”⁴⁹, and “vegan”⁵⁰, with visualizations generated using the “ggplot2”⁵¹ package.

Machine learning-based predictive modeling and interpretive analysis of SOC fractions

The eXtreme Gradient Boosting (XGBoost)⁵² algorithm is a high-performance boosting method that constructs an ensemble of decision trees, where each new tree learns from the residuals of the previous ones to progressively minimize prediction error. It integrates regularization to prevent overfitting and employs parallel computing with a block-based structure to accelerate training and efficiently handle large-scale datasets⁵³. Previous SOC modelling studies have reported that XGBoost can outperform simpler ensemble algorithms such as Random Forest^54,55. XGBoost was applied in this study to identify the key environmental drivers influencing measurable SOC fractions across different soil depths and aridity zones. The explanatory predictor set encompassed: (i) Climate: mean annual temperature (MAT), mean annual precipitation (MAP); (ii) Soil: Total Nitrogen (TN), pH, clay, Australian Soil Classification (major soil type categories); (iii) Soil oxide minerals: SiO₂, Al₂O₃, Fe₂O₃; (Ⅳ) Land use: pasture and continuous cropping; (Ⅴ) Topography: elevation, aspect, slope, Topographic Wetness Index (TWI)). The dataset was stratified by both aridity zone and depth, and separate models were developed for each subset to ensure reliable, context-specific interpretation. For each model, the data were randomly partitioned into 70% for training and validation and 30% as an independent testing set for final performance evaluation. Hyperparameter tuning was conducted using a grid search strategy coupled with 10-fold cross-validation to enhance model robustness and prevent overfitting. Five key hyperparameters, including max_depth (6, 8, 10), nrounds (300, 500, 700), subsample (0.6, 0.7), min_child_weight (1–10), and colsample_bytree (0.7, 0.8, 0.9), were tuned, with the learning rate (eta) fixed at 0.1. The optimal combination of hyperparameters was selected from these candidate ranges by minimizing mean squared error (MSE). Model predictive performances were quantified on the independent testing dataset using a set of statistical evaluation metrics, including mean absolute error (MAE), MSE, root mean square error (RMSE), the coefficient of determination (R²) and Lin’s concordance correlation coefficient (LCCC)⁵⁶. These metrics collectively provided a comprehensive assessment of the agreement between observed and predicted SOC fraction stocks. Detailed performance results for each model are presented in Table S3. All modelling and validation processes were performed using R packages “caret”⁵⁷ and “xgboost”⁵⁸.

To gain a deeper understanding of the factors influencing SOC fractions with soil depths across varying aridity zones, Shapley Additive exPlanations^59,60 (SHAP) were applied to decipher the final outputs of XGBoost predictive models. This game-theoretic approach offers a robust framework for quantifying the contribution of each predictor to model predictions, revealing both the magnitude and direction of their effects on the response variable and helping to uncover complex and non-linear relationships⁶¹. SHAP analyses were conducted using the R package “shapviz” (https://cran.r-project.org/package=shapviz). Higher SHAP values for each predictor indicate a stronger influence on the model output, where positive values suggest a tendency to increase the predicted response, while negative values imply a suppressing effect. The relative importance of each predictor was calculated by dividing the sum of its absolute SHAP values by the total sum of absolute SHAP values across all predictors, enabling direct comparison of their contributions to SOC fraction variability across soil depths and aridity zones. To facilitate group-level comparisons, the relative importance of each driver group (climate, soil properties, soil oxide minerals, land use and topography) was determined by aggregating the relative importance scores of all predictors within the corresponding functional group.

Structural equation modeling

We constructed piecewise structural equation modeling (piecewiseSEM) to evaluate the direct and indirect controls of key environmental components on SOC fractions (POC and MAOC) across soil depths. The analysis was focused on semi-dry and humid zones, where sufficient data coverage allowed meaningful comparison of contrasting environmental conditions and supported robust model interpretation. To reduce dimensionality and mitigate multicollinearity, principal component analysis (PCA) was applied to extract the first principal component (PC1) for each group of strongly correlated predictors: Climate (MAT and MAP), Soil (pH and clay), Topography (elevation and slope), and Secondary mineral oxides (Al₂O₃ and Fe₂O₃). Additionally, in the humid zone, where primary oxide minerals (SiO₂) and secondary oxide minerals were highly correlated (Spearman’s ρ = -0.93). To avoid multicollinearity, variable selection was guided by their relative importance indicated by SHAP analysis: secondary oxide minerals were retained for POC models, while primary oxide minerals (SiO₂) were incorporated for MAOC models. Prior to modeling, SOC fractions were log-transformed to improve normality, and all predictors were standardized using z-score transformation to remove scale bias. To account for variability associated with different soil types (Fig. S4), Australian Soil Classification (categorical variable) was incorporated as a random effect in all models using linear mixed-effects structures. PiecewiseSEM models were developed based on a priori hypotheses derived from established ecological pathways linking climate, soil properties, land use, oxide minerals, nitrogen and SOC fractions. The model fit was assessed using Fisher’s C statistic⁶², where P > 0.05 indicates an acceptable fit. Model refinement was performed based on the statistical significance of individual pathways and the improvement of overall model performance⁶³. Residual independence was assessed using the Durbin–Watson test, and minor residual dependence remained even after including Australian Soil Classification as a random effect, which is common in heterogeneous soil datasets lacking spatial coordinates due to privacy constraints. All piecewiseSEM processes were conducted using R packages ‘piecewiseSEM’⁶², ‘lme4’⁶⁴ and ‘semEff’ (https://cran.r-project.org/package=semEff).

Results

Climate- and depth-dependent variation in SOC fractions across land uses

SOC fractions exhibited distinct patterns across aridity zones and land uses (Fig. 2). Both POC and MAOC stocks generally increased as aridity decreased, with the highest average stocks observed in very humid regions across all depths. Land-use effects on SOC fractions were strongly modulated by aridity and soil depth. In dry zones, continuous cropping systems contained slightly higher POC and MAOC stocks than pasture systems across all depths. In semi-dry zones, pasture systems exceeded cropping only in surface POC stocks (0–10 cm), whereas this advantage extended to all depths in Mediterranean and semi-humid zones. In contrast, in humid and very humid regions, pasture maintained higher surface POC stocks, but cropping systems stored more at deeper layers. Regarding MAOC, pasture held greater stocks in the upper 20 cm of semi-dry zones; however, in wetter regions, this advantage was restricted to the top 10 cm. Conversely, cropping systems consistently stored more MAOC across all depths in humid and very humid areas.

Fig. 2.

Distribution of POC and MAOC stocks across six aridity zones under different land uses and soil depths. POC and MAOC stocks are shown for six aridity zones: Dry (red), Semi-dry (pink), Mediterranean (orange), Semi-humid (green), Humid (light blue), and Very humid (dark blue). Statistical differences among aridity zones within each land use and depth were tested using the Kruskal-Wallis test followed by Dunn’s post hoc comparisons. Different lowercase letters indicate significant differences at p < 0.05; bars sharing the same letter are not statistically different.

Both POC and MAOC stocks declined with depth, with distinct patterns shaped by land uses and aridity zones (Table S1). In cropping systems, POC stocks dropped sharply from 0–10 to 10–20 cm in more arid zones (49–58%), but declines were smaller in humid (27% ) and very humid (1%) regions. MAOC followed a similar but more moderate trend, with reductions lessening from about 18% in dry and semi-dry zones to 6% in humid and very humid areas. In contrast, pasture systems exhibited consistently large declines in both POC (56–73%) and MAOC (22–37%) across all zones, with relatively smaller drops in dry and very humid regions. However, from 10–20 to 20–30 cm, POC stocks declined less in arid areas (5–12% in cropping, 6–21% in pasture) but more sharply in humid regions (up to 42% in cropping, 53% in pasture). For MAOC, pasture systems also showed greater reductions in wetter zones (18–23%) than in drier zones (9–13%), whereas cropping systems exhibited less consistent patterns.

Depth-dependent variability in environmental controls and SOC fraction proportions

PERMANOVA analysis revealed that the combined set of soil and environmental variables differed significantly across aridity zones and land uses (P = 0.001, Fig. 3), indicating strong climate-land use imprints on soil profiles. PCA further showed that the dominant contributors to variability shifted with depth: in surface layers (0–10 cm), climatic variables, TN, SOC fractions, pH, clay content and topography primarily contributed to PC1, while soil oxide minerals (SiO₂, Al₂O₃, Fe₂O₃) were the main contributors to PC2. In deeper layers (20–30 cm), MAOC, clay content and oxide minerals emerged as dominant contributors to PC1, reflecting a shift from climate-nutrient to mineral-dominated controls with depths.

Fig. 3.

Principal component analysis (PCA) and relative proportions of SOC fractions across aridity zones, land uses and soil depths. (a–c) PCA results plots showing distributions of samples along PC1 and PC2 across six aridity zones, including Dry (dark pink), Semi-dry (light pink), Mediterranean (orange), Semi-humid (green), Humid (light blue), and Very humid (dark blue), across soil depths 0–10 cm, 10–20 cm, and 20–30 cm, respectively. (d–f) PCA results comparing Continuous Cropping (orange) and Pasture (green) at 0–10 cm, 10–20 cm, and 20–30 cm soil depths, respectively. Arrows represent variable loadings (contributions) on the principal components: climatic variables (MAT, MAP) in blue, SOC fractions (POC, MAOC) in yellow, soil properties (TN, pH, clay, SiO₂, Al₂O₃, Fe₂O₃) in grey, and topographic factors (elevation, slope, aspect, TWI) in red. R² and P values shown in each PCA plot are derived from PERMANOVA (Permutational Multivariate Analysis of Variance), indicating the explained variance (R²) and the significance level (P) of group differences. (g) Relative proportions of POC (purple) and MAOC (yellow) as percentages of total non-pyrogenic organic carbon (the sum of POC and MAOC).

MAOC was the dominant component of SOC stocks in the soils examined in this study, contributing 68–92%, particularly in deeper soils and under continuous cropping systems (Fig. 3). Pasture systems consistently maintained higher proportions of POC than cropping systems, particularly in surface soils (0–10 cm), though POC contributions declined with depth regardless of land use. In dry and semi-dry regions, the relative contributions of POC and MAOC showed minimal variation between land uses. In contrast, in semi-humid to very humid zones, POC proportions were substantially higher under pasture across all depths. Conversely, under continuous cropping, topsoil POC proportions declined with increasing humidity, ranging from 20.7% in dry regions to 14.9% in humid zones, with a slight increase to 18% in very humid regions.

Key environmental drivers and their relationships with SOC fractions across soil depth and aridity gradients

PCA results revealed depth-dependent patterns in the associations between environmental factors and SOC fractions (Fig. 3). In the surface layer, both fractions were strongly associated with climate, topography, and soil properties but weakly influenced by soil oxide minerals. POC exhibited strong correlations with soil pH and MAT, whereas MAOC was strongly associated with TN and MAP. At deeper soil layers (10–30 cm), the influence of climate and topography on both SOC fractions diminished. POC remained more responsive to climate and topography, and TN became more influential on POC stocks. In contrast, MAOC showed enhanced associations with clay content and pH. Notably, the contribution of soil oxide minerals increased with depth for both SOC fractions, with a more pronounced impact observed on MAOC.

Aligning with PCA results, SHAP-based analyses further revealed that soil properties (TN, pH, clay content, Australian soil classification) explained the majority of the variation in SOC fractions (27–64%), and were the most influential functional group overall, except for surface MAOC stocks (0–10 cm) in very humid zones, where soil oxide minerals contributed more to the explained variation (Figs. 4, 5 and 6, Table S2). TN was the most influential individual factor affecting both POC and MAOC stocks under most conditions, explaining 15–50% of the variation in SOC fractions. However, in certain cases, such as the 20–30 cm layer of semi-dry and semi-humid zones and the 0–10 cm layer of humid and very humid zones, SiO₂ was more influential than TN for MAOC stocks. Similarly, in the 20–30 cm layer of dry zones and the 0–10 cm layer of semi-humid zones, Fe₂O₃ and clay content had a greater influence than TN on POC, respectively. Despite these exceptions, TN was the only factor that consistently showed positive correlations with both SOC fractions across all combinations of aridity and depth (Fig. 4). Notably, the threshold TN concentration above which SOC accumulation was no longer limited increased with decreasing aridity from approximately 0.9 mg/g soil in dry zones to 2.7 mg/g soil in very humid zones for MAOC, and from 1.0 to 2.8 mg/g soil for POC. TN had a stronger influence on both SOC fractions in the surface layer than in the subsurface soils (10–30 cm) in relatively dry areas, whereas in humid and very humid zones, its influence increased with depth. Additionally, POC stocks were strongly sensitive to clay content, with lower clay content (less than 20%) favoring POC accumulation. The influence of clay content declined with soil depth. Soil pH also emerged as a key factor for both SOC fractions under specific conditions. POC was particularly responsive to pH variation in dry, semi-dry, and semi-humid zones, whereas MAOC was more strongly influenced by pH in more humid environments (Mediterranean to very humid zones). Lower pH levels were favorable for MAOC accumulation, while the response of POC to pH was more complex. In dry and semi-humid zones, lower pH also promoted POC accumulation, but in semi-dry zones, higher pH levels (above 6.6) were more conducive to POC storage in subsoils. Furthermore, Australian soil classification played a notable role in shaping SOC fractions, with a more pronounced effect observed in relatively dry regions. For example, in SHAP analysis for POC, soil classification ranked among the top three most important factors at 20–30 cm in dry zones. Similarly, for MAOC, soil class was a top three predictor at 0–10 cm and 10–20 cm depths in semi-dry zones.

Fig. 4.

Correlations and relative contributions of explanatory variables to SOC fractions across aridity zones and soil depths. (a) Correlation heatmap showing relationships between environmental factors and SOC fractions across soil depths (0–10 cm, 10–20 cm, and 20–30 cm) and aridity zones (Dry, Semi-dry, Mediterranean, Semi-humid, Humid). The continuous legend bar represents Spearman correlation coefficients ranging from − 1 (orange, negative) to 1 (blue, positive). The size of the circles indicates the relative importance of each factor, with red circles highlighting the top four most influential factors. (b) relative importance of functional groups of explanatory variables across depths and aridity zones, with color gradients denoting different categories: climate (blue), soil properties (brown), soil oxide minerals (purple), land use (yellow), and topographic attributes (green).

Fig. 5.

SHAP-derived relationships between key predictors and POC across aridity zones and soil depths. Orange points in each plot represent SHAP values, quantifying the contributions of different levels of a specific predictor to individual POC predictions. Positive and negative SHAP values indicate the magnitude of the predictor’s positive or negative influence, respectively. Blue lines (with confidence intervals) show fitted trends use polynomial regression (with R² and P values) or LOESS (Locally Estimated Scatterplot Smoothing, non-parametric). Green vertical reference lines mark threshold values where the direction of contribution shifts, representing tipping points beyond which POC accumulation is either promoted or constrained. For the categorical predictor “Australian Soil Classification”, a boxplot of observed POC stocks is shown instead.

Fig. 6.

SHAP-derived relationships between key predictors and MAOC across aridity zones and soil depths. Orange points in each plot represent SHAP values, quantifying the contributions of different levels of a specific predictor to individual MAOC predictions. Positive and negative SHAP values indicate the magnitude of the predictor’s positive or negative influence, respectively. Blue lines (with confidence intervals) show fitted trends use polynomial regression (with R² and P values) or LOESS (Locally Estimated Scatterplot Smoothing, non-parametric). Green vertical reference lines mark threshold values where the direction of contribution shifts, representing tipping points beyond which MAOC accumulation is either promoted or constrained. For the categorical predictor “Australian soil classification”, a boxplot of observed MAOC stocks is shown instead.

Soil oxide minerals represented the second most important functional group influencing SOC fractions, accounting for 8% ~ 44% of the explained variation. In dry zones, POC stocks were strongly sensitive to soil oxide minerals, whereas in other aridity zones, MAOC stocks responded more strongly. For MAOC, the influence of oxide minerals was greater in subsoils (10–30 cm) under relatively dry conditions, but shifted toward surface soils in more humid regions (humid and very humid zones). SiO₂ consistently ranked among the top two predictors of MAOC stocks in all but dry zones, even surpassing TN in surface soils of humid and very humid areas. MAOC accumulation was favored when SiO₂ content was below 67%. In contrast, POC stocks were more sensitive to Al₂O₃ and Fe₂O₃ compared to their relationship to SiO₂. The influence of Fe₂O₃ on POC varied across aridity zones: it enhanced POC accumulation in dry and semi-dry areas when concentrations exceeded 7%, but had a negative effect in humid zones, where levels above 4.7% were associated with reduced POC stocks. Al₂O₃ also exhibited a threshold-dependent effect on POC, with lower concentrations supporting accumulation and levels above 7% limiting stocks. However, higher Al₂O₃ content (> 10%) was associated with increased MAOC accumulation.

Topography also played an important role in shaping SOC fractions, with stronger influence on subsurface soil POC stocks (10–30 cm) in relatively dry regions and on surface (0–10 cm) MAOC stocks in humid and very humid zones. Elevation emerged as one of the top three predictors for SOC fractions under specific conditions. For MAOC, elevation was particularly influential in semi-humid and humid zones, while for POC, it was more important in subsurface soils (20–30 cm) in Mediterranean areas and in surface layers (0–20 cm) of semi-humid and very humid zones. Lower elevation usually favored accumulation of both fractions, although higher elevation promoted subsurface soil POC storage in Mediterranean climate areas.

Climate was a key driver for SOC fractions primarily in dry and semi-dry zones. POC stocks were especially sensitive to MAT, with its influence declining with depth. For MAOC, both MAT and MAP were important drivers in dry zones, where climate collectively explained more than 20% of the variation. Both fractions’ accumulation was constrained when MAT exceeded 18 °C, and MAP values above 350 mm promoted MAOC accumulation. Land use primarily influenced POC stocks, with a more pronounced effect observed in surface soils of humid and very humid zones.

Mediating pathways of dominant drivers controlling SOC stocks in semi-dry vs. humid zones

PiecewiseSEM analysis (Figs. 7 and 8) quantified the direct, indirect and total standardized effects of environmental drivers on SOC fractions in semi-dry and humid zones. Results are reported based on the total effects, calculated as the sum of direct and indirect pathways (standardized coefficients), where larger absolute values indicate stronger influence and the sign denotes positive or negative associations. For POC stocks, TN and soil depth exerted the strongest total effects in both semi-dry and humid zones, with TN dominant in semi-dry areas (+ 0.51, -0.39 for soil depth) and soil depth stronger in humid areas (-0.55, + 0.36 for TN). In semi-dry zones, climate had a greater impact on POC stocks (-0.21), whereas in humid zones, topography (-0.15), climate (-0.13), soil properties (-0.12), and land use (+ 0.11) contributed equally. For MAOC stocks, TN was the primary driver in semi-dry zones (+ 0.45), followed by soil depth (-0.28) and both primary (SiO₂) and secondary (Al₂O₃, Fe₂O₃) oxide minerals (-0.27, + 0.24). In contrast, in humid zones, primary oxide minerals (-0.44) and soil depth (-0.36) were more influential than TN (+ 0.25), with topography (-0.24) and climate (-0.13) also playing notable roles. Across both zones, topography, climate, land use, and soil depth could influence SOC fractions indirectly through interactions with oxide minerals, soil properties, and TN, although the direction and strength of these pathways varied by local environmental conditions.

Fig. 7.

Mediating pathways of environmental controls on SOC fractions (POC, MAOC) in semi-dry zones. The diagram displays structural equation model (SEM) pathways and standardized effects (direct, indirect, total effects) of key environmental factors based on bootstrapped SEM results. Numbers adjacent to pathway arrows denote standardized path coefficients, with their significance levels denoted as *** (P < 0.001), ** (P < 0.01), and * (P < 0.05). Red and blue paths represent positive and negative effects, respectively, with path widths scaled to effect sizes. Fisher’s C and P values indicate an overall model fit, where a non-significant P (> 0.05) suggests an adequate fit. d.f., degree of freedom. R² values represent the variance explained for the response variable (POC or MAOC stocks).

Fig. 8.

Mediating pathways of environmental controls on SOC fractions (POC, MAOC) in humid zones. The diagram displays structural equation model (SEM) pathways and standardized effects (direct, indirect, total effects) of key environmental factors based on bootstrapped SEM results. Numbers adjacent to pathway arrows denote standardized path coefficients, with their significance levels denoted as *** (P < 0.001), ** (P < 0.01), and * (P < 0.05). Red and blue paths represent positive and negative effects, respectively, with path widths scaled to effect sizes. Fisher’s C and P values indicate an overall model fit, where a non-significant P (> 0.05) suggests an adequate fit. d.f., degree of freedom. R² values represent the variance explained for the response variable (POC or MAOC stocks).

Discussion

Climate-land use-depth interactions modulate the distribution of SOC fractions

Our findings illustrate the complex interactions among climate, land use and soil depth in shaping the distribution of SOC fractions across Australian agricultural soils. Both POC and MAOC stocks increased with decreasing aridity, largely due to improved water availability that supports greater plant productivity, denser vegetation cover and increased organic inputs from litter, roots and exudates⁶⁵. Enhanced soil moisture and organic inputs not only promote microbial activity and turnover^66–68, facilitating the transformation of POC into MAOC, but also reduces carbon losses through improved aggregate stability and mineral protection^69,70. Additionally, active plant metabolism and litter decomposition can acidify the soil, accelerating mineral weathering and generating reactive secondary minerals that promote organo-mineral associations, thereby facilitating the sequestration of MAOC in humid climate soils¹⁰.

Interestingly, land-use impacts on SOC fractions varied across the aridity gradient. In dry regions, the slightly higher SOC stocks of both fractions in cropping systems compared to pastures are likely attributed to the incorporation of crop residues, irrigation and fertilization, which enhance organic matter input and nutrient availability^71–74, thereby supporting microbial activity and promoting the formation and accumulation of both POC and MAOC in low-productivity soils, relative to typically lower-input and, in some cases, intensively grazed pasture systems. While pastures are generally expected to sustain larger POC pools through continuous root turnover, the conversion of POC into MAOC is mediated by microbial activity and may be constrained by the soil’s attainable MAOC capacity²⁹. These findings highlight that improved management practices in arid regions that can maintain fractional vegetation cover and continuous carbon inputs, or can take maximum advantage of wetter, more productive seasons, have substantial potential to enhance the persistence of both POC and MAOC in agricultural ecosystems. In more mesic regions, the advantage of pasture systems in storing POC likely stems from sustained vegetation cover, less aboveground biomass removal, and minimal soil disturbance, which together promote the accumulation of POC, especially in surface soils. This aligns with recent findings identifying vegetation cover as a key local driver of space-time SOC variation across the Australian continent⁷⁵. However, cropping systems store more SOC, especially MAOC, in deeper soil layers under humid and very humid zones. This pattern suggests that management practices such as crop root architecture and fertilization may enhance carbon inputs to subsoils⁷⁶ and facilitate MAOC formation through increased POC supply and organo-mineral associations, particularly in subsurface soils where mineral saturation deficits remain high⁵. In addition, inherent environmental constraints, such as low nutrient availability and limited oxygen diffusion, further restrict microbial decomposition and contribute to SOC stabilization at depth. These findings underscore the importance of considering both surface and subsurface soil processes when evaluating SOC sequestration potential across different climatic and management regimes. In humid regions, integrating pastures into cropping systems may help optimize surface carbon retention while enabling deeper MAOC formation through managed nutrient inputs or deep-rooted vegetations. Future research should investigate the trade-offs and synergies between these systems, especially in response to climatic variability and extremes, to inform adaptive, climate-smart soil carbon management strategies.

Dominant drivers of SOC fractions across aridity and depth gradients

Combined PERMANOVA, PCA and SHAP analyses revealed a complex hierarchy of environmental drivers controlling SOC fractions across aridity zones and soil depths. Consistent with other research¹, soil properties were the most important drivers for both POC and MAOC stocks across most regions. Among the key soil properties, TN was consistently the most influential individual predictor of both POC and MAOC stocks across most aridity-depth combinations. This finding aligns with previous evidence that nitrogen availability constrains both plant biomass production and microbial carbon use efficiency^77,78, thereby directly enhancing POC accumulation and ultimately facilitating MAOC formation⁷⁹. The observed increasing TN thresholds for SOC accumulation with aridity declines likely reflect not only elevated plant biomass inputs and heightened microbial N demand in more humid environments, but also greater nitrogen losses through leaching and denitrification under wet conditions⁸⁰, which together raise the nitrogen requirement for sustained SOC formation. Furthermore, compared to its stronger influence in surface layers (0–10 cm) under relatively arid conditions, the enhanced relative importance of TN in subsoil (10–30 cm) of more humid regions suggests that deeper rooting and greater moisture penetration enable nitrogen-driven SOC formation and turnover to occur at greater depths, where nitrogen limitation of microbial decomposition may be more pronounced^26,27.

POC stocks exhibited strong sensitivity to clay content, with lower clay levels (< 20%) associated with higher POC accumulation, particularly in surface soils. This is likely due to weaker aggregation, fewer fine minerals and particles sizes, and reduced moisture retention in lower-clay soils^81–83, which limit the formation of stable organo-mineral associations and, in turn, favor the maintenance of soil carbon as POC while constraining the transformation of POC into stabilized MAOC. However, because POC is less protected and highly dynamic, its persistence depends on a continuous flow of fresh carbon inputs, and input disruptions in low-clay soils may lead to rapid losses. Furthermore, studies have shown that soils with clay content below 15% can exhibit dissolved organic carbon concentrations up to four times higher than those of high-clay soils under similar vegetation, suggesting greater mobility to deeper, more protected depths and potential for POC preservation in coarse-textured soils¹⁰. In addition, clay content, types or their relative specific surface areas have been found to have relatively low explanatory power for organic matter stabilization compared to other physicochemical factors, such as exchangeable Ca in dry, alkaline soils and Fe/Al oxides in more humid, acidic environments^84,85.

Soil pH is an important factor influencing both POC and MAOC, but its influence varies across aridity zones and SOC fractions. The high sensitivity of POC to pH in relative dry zones is likely primarily due to inhibited or altered microbial activities at low pH^86,87, potentially reducing microbial decomposition and promoting plant-derived organic matter accumulation. In contrast, the great sensitivity of MAOC to pH in more humid zones can be attributed to the role of acidified conditions in enhancing carbon sorption onto minerals via increased ionic strength and the formation of strong inner-sphere bonds with reactive mineral surfaces such as Fe and Al oxides^88,89. Specifically, as pH increases, SOC stabilization mechanisms shift from these strong organo-metal complexation toward weaker outer-sphere interactions and Ca/Mg-mediated cation bridging dominated in neutral to alkaline soils^10,90. Furthermore, organic acids released by living roots and microbial oxidative decomposition can lower pH and thereby promote mineral dissolution¹⁰. However, in dry regions, the lower abundance and weaker bonding strength of mineral-organic associations may render MAOC stocks more vulnerable to disturbance compared to the more robust associations found in humid productive soils^10,91, although decomposition can also be constrained by limited moisture availability.

Apart from soil properties, soil oxide minerals were identified as the second most important functional group influencing SOC fractions. In arid regions, the great sensitivity of POC stocks to oxide minerals suggests that under limited organic inputs and microbial transformation, POC persistence relies more on spatial inaccessibility of microbial decomposition through physical protection via metal oxide-induced aggregation processes^92,93. Additionally, less acidic soil conditions with water limitation compared to humid soils may reduce metal (Fe/Al) solubility and weathering rates⁹⁴, weakening mineral reactivity and organo-mineral associations, which in turn may limit MAOC formation and favor POC accumulation. Notably, Fe₂O₃ exhibited contrasting effects on POC across climatic zones: concentrations above 7% promoted POC accumulation in relatively dry regions but reduced it in humid zones, likely because humid, productive conditions enhance microbial decomposition, while increased carbon inputs and the greater reactivity of Fe oxides in acidic, weathered soils promote strong organo-mineral associations, facilitating the transformation of carbon from POC to MAOC. Similarly, low Al₂O₃ (< 7%) favored POC accumulation, while concentrations above 10% enhanced MAOC sequestration, contributing to an Al/Fe oxide-induced decline in POC stocks. In relatively dry zones (ranging from dry to semi-humid), oxide minerals played a larger role in subsoil MAOC accumulation likely due to greater moisture retention and deeper interactions with percolating dissolved organic matter, enabling sustained organo-mineral associations. In contrast, in humid regions, abundant surface inputs and high microbial activity promote rapid MAOC formation in the weathered topsoil, while the high potential for mineral saturation may also make oxide minerals as a key constraint on further MAOC accumulation. Additionally, MAOC stocks were more sensitive to SiO₂ content than to Al₂O₃ and Fe₂O₃, likely because silicate-dominated minerals (e.g., phyllosilicates, allophane and amorphous silica) often prevail in soil mineralogy and provide high specific surface area, persistent negative charges from isomorphic substitution, and pH-independent organo-mineral binding capacity, making SiO₂ a more reliable predictor of MAOC storage¹⁰. These findings underscore that while soil oxide minerals universally contribute to SOC dynamics, their effects are fraction-, depth- and climate-dependent, reflecting distinct stabilization mechanisms across environmental gradients.

As expected, climate was a major driver in relatively arid regions. The high sensitivity of POC stocks to MAT in surface soils is primarily driven by carbon loss processes, likely due to surface-dominated input of fresh plant material and the temperature dependence of decomposition and turnover processes in water-limited environments^5,85. In contrast, both MAT and MAP strongly influenced MAOC stocks, indicating the dual role of limited organic inputs and constrained organo-mineral associations in shaping MAOC formation and persistence, highlighting the high vulnerability of MAOC in arid conditions. Additionally, SOC fractions were also affected by topography and land use. The influence of topography likely stems from its control over plant biomass input and microbial activity through governing microclimatic conditions, erosional and hydrological processes². Consistent with previous studies^95,96, land use had a more pronounced influence on POC than MAOC, particularly in surface soils under humid conditions. In these regions, greater biomass inputs and reduced soil disturbance under pasture are likely to promote POC accumulation, while tillage and residue removal in croplands accelerate surface POC losses.

Aridity-dependent pathways regulating SOC fractions

In semi-dry zones, TN was the most influential factor controlling both POC and MAOC stocks, with a stronger influence observed compared to humid zones. This underscores the pivotal role of nutrient availability in regulating SOC dynamics under relatively arid conditions, consistent with stoichiometric constraints on SOC storage reported in previous studies^97,98. Nitrogen availability enhances plant productivity and microbial activity⁹⁹, thereby facilitating both the input-driven accumulation of POC and microbial transformation required for MAOC formation and stabilization. Nitrogen-rich carbon inputs can provide stoichiometrically balanced nutrients that improve microbial carbon use efficiency, thereby increasing the potential for carbon being stabilized in MAOC^11,100, which typically requires greater nitrogen investment than POC formation⁷. Furthermore, variations in nitrogen availability often co-occur with changes in soil pH or osmotic stress^101–103, which can alter microbial biomass responses and mask nitrogen stimulation of microbial activity¹⁰⁴. However, in humid zones, soil depth dominated over TN in controlling both POC and MAOC stocks, highlighting pronounced vertical stratification of SOC fractions with greater accumulation near the surface. Favorable conditions for plant growth and microbial activity in these regions lead to high biomass inputs concentrated in the topsoil, thereby enriching surface SOC pools. Additionally, across both zones, TN exerted a stronger influence on POC stocks compared to MAOC. This difference likely arises because POC, composed of relatively fresh plant and microbial residues, depends more directly on nitrogen-driven biomass inputs and nutrient cycling. In contrast, MAOC represents more stabilized carbon pools influenced by multiple factors such as soil minerals and physical protection, and its lower C/N ratio makes it less nitrogen-limited for microbes⁶, rendering TN’s relative impact less pronounced.

Climate had a greater influence on POC stocks in semi-dry zones, where water limitation intensifies POC sensitivity to precipitation and temperature variability, directly constraining plant productivity and litter decomposition. In contrast, in humid zones, where climate becomes a less dominant control on POC, local-scale factors such as topography, soil properties, land cover emerge as equally important drivers in regulating POC dynamics, likely due to their influence on biomass distribution, erosion patterns and soil functioning process. For MAOC stocks, oxide minerals (SiO₂, Al₂O₃, Fe₂O₃) consistently exerted a strong influence across zones, highlighting their critical role in organo-mineral associations that stabilize carbon. In humid regions, these oxides emerged as the most influential factors governing MAOC variation, reflecting a greater reliance on mineral protection mechanism. This may be attributed to the formation of more stable organo-mineral associations in highly weathered soils¹⁰, as well as the relatively high mineral saturation potential in humid environments³¹, which can limit additional MAOC formation. These findings underscore the need to address mineral saturation constraints when aiming to enhance long-term carbon storage in such systems^29,105,106.

Therefore, the vulnerability of both POC stocks and MAOC to nitrogen availability in arid conditions highlights the priority of enhancing organic inputs through improved nutrient retention strategies, such as legume incorporation and organic amendments, to support plant productivity and microbial activity essential for carbon accumulation and stabilization. The heightened sensitivity of POC stocks to climate in these environments further underscores the need to improve water retention via continuous land cover and appropriate irrigation, while reducing microbial decomposition by minimizing soil disturbance (e.g., through reduced tillage). In contrast, in humid environments, both POC and MAOC stocks tend to accumulate in the surface layer, where a high risk of mineral saturation may further limit MAOC formation. This challenge is particularly evident under pasture systems, where the large proportion of surface-concentrated POC stocks is more vulnerable to environmental disturbances and microbial decomposition, thereby posing a constraint on long-term carbon sequestration. Minimizing human intervention, such as excessive grazing or soil disturbance, alongside promoting deep-rooted vegetation can help protect surface carbon and enhance subsoil sequestration. Furthermore, given the strong reliance of MAOC on soil oxide minerals and high potential of mineral saturation under humid zones, external mineral inputs³¹, such as clay-rich amendments, iron and aluminum oxides, or silicate rock powders, may help overcome mineral saturation limitation and facilitate the transformation of POC into more stable MAOC. These findings underscore the need for SOC sequestration strategies that consider both stabilization capacity and vulnerability to carbon loss pathways. Additionally, topography, land use, and climate exerted both direct and indirect effects on SOC fractions by altering soil physicochemical properties and nutrient dynamics. These interactive pathways varied by aridity zones, indicating that SOC accumulation is shaped by a network of interdependent processes rather than isolated drivers. Recognizing these complex, context-specific interactions is essential for designing effective soil carbon management strategies tailored to regional environmental conditions.

Research limitations and uncertainties

This study has several potential limitations and uncertainties. First, as is common in large-scale analyses, although comparisons between land uses across aridity zones were conducted, the limited sample sizes for cropping systems in humid (n = 50) and very humid (n = 26) regions may reduce the reliability of these interpretations. Therefore, findings related to land-use differences in wetter zones should be further validated using larger and more balanced datasets. Furthermore, the input dataset may introduce additional sources of uncertainty into our analysis. For example, estimates of SiO₂, Al₂O₃ and Fe₂O₃ concentrations were obtained through Mid-infrared technology combined with partial least squares modeling, which were calibrated using a limited number of soil samples. Clay content and pH values were also derived from spatial predictions rather than direct measurements. Additionally, given the unavailability of detailed information on biomass inputs and fertilizer applications, this study could not fully account for management-driven variation in SOC fractions. Future research should incorporate management data including residue returns, fertilizer use and manure inputs to allow a more mechanistic assessment of how nutrient management influences SOC dynamics. Despite these limitations, the study provides valuable insights into the depth- and aridity-specific drivers of SOC fractions in Australian agricultural soils, which can guide future research and land management strategies.

Conclusion

Our study reveals a fundamental divergence in SOC stabilization pathways across aridity gradients and soil depths, driven by distinct interactions among land use, soil properties and mineralogy. In arid zones, where MAOC formation is constrained, limited moisture availability fosters POC-dominated stabilization. Here, carbon storage relies primarily on intermittent plant inputs restricted by water and nutrient stress, together with physical protection through oxide-mediated aggregation. Maintaining a continuous flow of organic inputs and stable soil structure are therefore top priorities for enhancing the persistence of both POC and MAOC in these soils. In contrast, humid zones, where surface POC pools are larger and relatively unprotected, are characterized by MAOC-dominated stabilization, supported by intensive mineral weathering, sustained microbial processing, and acid-enhanced organo-metal bonding. Accelerating the transfer of POC into MAOC in surface soils, while addressing the constraint of soil attainable MAOC capacity and developing subsoil SOC sequestration potential, is critical for enhancing long-term storage in these systems. These findings underscore the need for climate-smart, region-specific land management to sustain agricultural productivity and resilience in the face of changing climate conditions. Monitoring and modeling efforts should also account for the differing vulnerabilities of SOC fractions under shifting climate regimes. Our results provide a mechanistic quantitative foundation for predicting divergent SOC responses to climate change, with arid systems susceptible to input-driven carbon sequestration limitations and humid systems constrained by their attainable mineral-associated storage capacity. Future work should integrate long-term detailed records of management practices or experimental data, such as tillage, irrigation and stubble management, cover crops, fertilizer and organic inputs to more accurately quantify the impacts of management practices on POC and MAOC dynamics across depth and aridity gradients under current and future climates.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

H.J., A.P. and S.K. developed the conceptualization and methodology; H.J., S.K. and B.P. performed the statistical analyses and validation; H.J. and A.P. wrote the original draft; All authors contributed to improving and finalizing the manuscript.

Funding

This work was funded by the Australian Government Department of Agriculture, Fisheries and Forestry as part of the Soil Science Challenge Grants Program (ID: 4-H4SSYKU) and supported by the Graduate Research Scholarship of the University of Melbourne.

Data availability

The raw database is openly accessible at [https://data.csiro.au/collection/csiro:5883](https:/data.csiro.au/collection/csiro:5883) . The source dataset after pre-processing that supports the findings of this study is provided with this paper.

Declarations

Competing interests

The authors declare no competing interests.