Soil Inorganic Carbon Losses Counteracted Soil Organic Carbon Increases in Deeper Soil over 30 Years in North China

Abstract

Finding out the dynamics of soil organic carbon and inorganic carbon is paramount for sustaining terrestrial carbon cycling and climate change mitigation. From the 1980s to 2010s, substantial changes in land use, climate, and agricultural practices have occurred across North China. This study systematically quantified the stratified dynamics of soil carbon stocks (0–100 cm with 20 cm intervals) and their compositional shifts by using the geographically weighted regression kriging model. The model integrated soil sample data from provincial surveys across North China with key environmental covariates (e.g., elevation, precipitation, air temperature, and the vegetation index) to spatially predict and analyze vertical carbon stock changes. The results indicated that soil carbon stocks decreased considerably by 5.86 Gt in the one-meter soil profile from the 1980s to the 2010s. Significant losses in soil inorganic carbon stocks directly contributed to net soil carbon sources. These significant soil inorganic carbon losses of 7.03 Gt, originating primarily from losses of 7.35 Gt in deeper soil layers (20–100 cm), effectively offset increases of 1.17 Gt in soil organic carbon. About two-thirds of regions in North China have been categorized as carbon source regions. These are distributed for the most part in arid and semi-arid areas and the Qinghai–Tibet Plateau. The remaining one-third of regions have been classified as carbon sink regions which are primarily found in the Loess Plateau, the Huang–Huai–Hai Plain, the Middle-lower Yangtze Plain, and the Northeast China Plain. Significant losses in soil inorganic carbon stocks caused by strong carbon sources may undermine global measures aimed at enhancing terrestrial ecosystem carbon sequestration and fixation. Our results highlight the urgent need to account for vulnerable subsurface inorganic carbon pools in regional carbon sequestration strategies and climate models.

1. Introduction

The primary greenhouse gases, CO₂, CH₄, and N₂O, contribute approximately 80% of the global warming effect and are emitted annually from soils at estimated rates of 5%~20% (CO₂), 15%~30% (CH₄), and 80%~90% (N₂O) [1]. The accumulation of these gases in the atmosphere has increased exponentially, leading to elevated levels of CO₂, which perturb global carbon (C) cycling equilibrium. Soil C, which includes the soil organic carbon and inorganic carbon (SOC and SIC) forms, is a critical reservoir that governs key soil physicochemical attributes—including aggregation, hydraulic properties, ion exchange capacity, and nutrient dynamics. Additionally, soil C directly modulates atmospheric CO₂ concentrations through bidirectional C fluxes [2]. Due to their sensitivity to environmental disruptions, subtle fluctuations in SOC and SIC pools exert disproportionate impacts on terrestrial C—climate feedbacks [3,4]. Consequently, SOC–SIC dynamics represent critical determinants of both the structural resilience of terrestrial C cycles and the trajectory of climate change mitigation strategies.

Soil C stocks and their dynamics have emerged as a global scientific priority, with empirical evidence demonstrating measurable increases in terrestrial C reservoirs across both forest ecosystems and agroecosystems [5,6,7,8,9,10,11]. Notably, anthropogenic interventions such as China’s revegetation initiatives have driven significant increases in vegetation coverage and biological C sequestration, particularly in northern regions [8,10,11]. Concomitant with these efforts, nationwide surveys reveal a 30-year enhancement plan for SOC stocks [12,13,14], as exemplified by a tripling of planted forest C stocks from 675.6 ± 12.5 Tg C (1990 year) to 1873.1 ± 16.2 Tg C (2020 year) [15]. Regional analyses further corroborate this trend, with SOC concentrations in the Loess Plateau’s revegetated zones reaching 2.27~9.81 g kg⁻¹ in topsoil (0–20 cm) [16], underscoring the efficacy of ecological restoration in rebuilding soil C capital. Researchers also noticed that anthropogenic–driven soil acidification particularly from intensive fertilization threatens to destabilize pedogenic carbonates. This process accelerates their dissolution into CO₂ efflux and trigger substantial SIC losses in farmland and agroforestry ecosystem [17,18]. At larger spatial scales (e.g., terrestrial ecosystems in China, global soil C pools), some researchers have also noticed that natural environmental changes such as soil pH reduction caused by atmospheric acid deposition and climate change are the main causes of SICs loss, and the loss is mainly caused by migration to groundwater [3,17].

Soil C pools comprise both SOC and SIC stocks (SOCs and SICs), research has historically prioritized SOC dynamics and relegated SIC to a presumed state of inertness due to its perceived mineralogical stability. Emerging evidence, however, challenges this paradigm, revealing that anthropogenic activities such as land use change, climate forcing, and agronomic interventions can hinder the formation of SIC through pedogenic carbonate dissolution–reprecipitation processes, thereby modulating their role as net C sources [3,19]. Despite the critical biogeochemical linkages between SIC and atmospheric CO₂, mechanistic understanding of SIC dynamics remains fragmented, particularly in the context of its C sequestration potential and sensitivity to human disturbances [19,20].

Recent empirical evidence reveals divergent vertical trajectories in SOC content and pH over the past three decades, underscoring stratified heterogeneity in soil C dynamics across depth profiles [7,13,14,21,22]. However, the depth-resolved variations in both SOC and SIC remain inadequately indistinct. While extant research prioritizes SOC stratification [6,7,13], it accounts for merely ~30% of soil C stocks in arid and semi-arid regions, where SIC dominates as the principal C reservoir [17,23,24]. This disparity necessitates focused investigation into SIC’s vertical redistribution mechanisms, particularly in northern China—a region encompassing >50% of the nation’s arid territories—where neglecting SIC dynamics risks significant underestimation of carbon–climate feedbacks in climate-vulnerable dryland ecosystems.

The goals of this paper is to explore the vertical changes in changes in SOCs and SICs in North China, along with the soil C stocks changes that directly result from these variations. The objectives of this paper is

(1)

to acquire the vertical SOC and SIC content. The organic matter and calcium carbonate data of typical soil profiles were normalized, and a normalized profile with a depth range of 0–100 cm was constructed. The profile was divided into five layers (0–20, 20–40, 40–60, 60–80, and 80–100 cm) with 20 cm intervals, and the contents of SOC and SIC in each layer were calculated.

(2)

to obtain the density and storage of SOC and SIC in a vertical soil profile. Based on the normalized soil-layer data, the prediction model was used to generate and visualize the spatial distribution map of SOC and SIC contents. The bulk density and coarse fraction (>2 mm) content were integrated into the standard soil C storage calculation formula to construct a soil carbon storage dataset of density and stocks.

(3)

to quantify soil C sinks and sources and analyze the relationship between vertical dynamics of SOCs and SICs and total C storage dynamics. The vertical dynamics of soil C were obtained by subtracting the soil C stocks in the 2010s from the those in the 1980s. The dynamics of soil C directly determined the distribution of soil C sinks and sources. A dynamic relationship model was built to evaluate the contribution of SOCs and SICs to C sinks and sources.

Based on the above literature, we propose the following hypotheses: (i) SICs are higher than SOCs in North China; (ii) SOCs increased while SICs decreased at a depth of one meter; (iii) variations in SOCs and SICs in the vertical profile lead to the stratified heterogeneity of soil carbon stock dynamics.

2. Materials and Methods

2.1. Study Area

North China consists of 17 provinces (cities and autonomous regions, Figure 1a) (only a few areas of some provinces, such as Shaanxi, Henan, Anhui, and Jiangsu, were included based on the boundary of the Qinling mountain and the Huai River) spanning a longitude of 31°04′–53°33′ N and a latitude of 73°40′–135°02′ E, with a total area of 5.58 million km², which constitutes 58% of mainland China. The elevation range of the study area is −86 to 2345 m (Figure 1d). The landform of North China gradually decreases from west to east and exhibits diverse geomorphologies features including the Northeast China Plain, northern arid and semi-arid regions, the Huang–Huai–Hai Plain, the Loess Plateau, the Qinghai–Tibet Plateau, and the Middle-lower Yangtze Plain (Figure 1e). The annual temperature ranged from −22 to 16 °C, and precipitation ranged from 0 to 1400 mm. Among them, the climate conditions in the northern arid and semi-arid regions and the Loess Plateau are special. The annual precipitation is about 50~650 mm, and some areas receive less than 50 mm of precipitation. The average annual temperature is about 0~12 °C. On the one hand, such a climate limits the dissolution of carbonates, but concentrated seasonal precipitation will take away carbonates in greater amounts [25]. Lower precipitation and higher evaporation induced the arid and semi-arid climate features that all deserts in subregions of China have [26].

The soils in North China exhibit distinct regional heterogeneity, as they are dominated by Silt Loam from the loess parent material. The main soil texture in the important subregion includes Clay Loam in the Northeast China Plain, which consists of ~20–30% sand, ~25–35% silt, and ~40–50% clay. Sandy Loam in the northern arid and semi-arid regions consists of ~40–60% sand, ~30–40% silt, and ~10–20% clay. Silt Loam in the Loess Plateau has a high silt fraction of ~60–70%. The common texture of the Huang–Huai–Hai Plain includes Loam and Sandy Clay Loam. The soil of the Qinghai–Tibet Plateau is unique due to the influence of temperature and altitude, and its soil is characterized by high gravel and coarse bone [27]. North of the Qinling mountain and Huai River boundary, diverse soils under arid and semi-arid climates with the loess parent material and a geological history exhibit widespread calcic horizons, particularly in typical steppe and desert steppe zones. These horizons are particularly rich in calcium carbonate (CaCO₃, ~15–60%) [28].

As shown in Figure 2, 10 soil orders are distributed in North China according to Chinese soil taxonomy (total 12 soil types in China, CST), of which aridosols and cambosols are the two largest soil orders with an area of about 32.57% and 28.64% (excluding the area of unobservable land use types, such as water, deserts, and glaciers), far exceeding other orders. The Northeast China Plain has 7 soil orders, of which argosols account for almost half of the area of this subregion, about 56.31%. There are 8 soil orders in the northern arid and semi-arid regions and the Qinghai–Tibet Plateau, and the largest proportions of aridosols and cambosols are about 51.31% and 47.67%, respectively. There are 6 soil orders in the Huang–Huai–Hai Plain, and the area proportions of cambosols and argosols (50.76% and 40.49%, respectively) are far more than other orders. The area of primosols and cambosols accounted for the highest proportions in the Loess Plateau, about 47.9% and 34.44%, respectively.

2.2. Data Sources

The basic datasets used in this study are shown in

Table 1. Sources of the basic dataset in this study.

Basic Dataset	Sources
Soil sample dataset	1980s: The Second National Soil Survey 2010s: The Chinese Soil Series
Bulk density, BD/g cm−3	National Earth System Science Data Center “http://soil.geodata.cn/index.html (accessed on 8 July 2024)”
Coarse fragments (>2 mm), CF/%
Digital elevation model, DEM/m	Resources and Environmental Science Data Platform “https://www.resdc.cn (accessed on 25 June 2024)”
Annual precipitation, PRE/mm
Annual temperature, TEM/°C
Land use dataset
Normalized vegetation index, NDVI	Google Earth Engine “https://developers.google.cn/earth-engine (accessed on 20 May 2024)”

. The datasets included bulk density (BD) and coarse fragment (>2 mm, CF) data [5]. To simulate the soil organic carbon and inorganic carbon (SOC and SIC) estimations on a large scale, we referred to environmental covariates, such as DEM, PRE, TEM, and NDVI, which were used in geographically weighted regression kriging (GWRK) studies for SOC and SIC content prediction by Chen et al. [29], Ho et al. [30], Wang et al. [31], and Yigini & Panagos [32]. In addition, we used a land use dataset to analyze soil carbon (C) stocks under different land use types.

The second national soil survey was conducted in the 1980s in every province of China. The analysis of soil C stocks requires that the typical profile contains at least one of the information of SOC and SIC. Therefore, we excluded 477 (1980s) and 325 (2010s) soil profiles lacking detectable SOC and CaCO₃ information from the original 3767 and 2368 samples, respectively. Finally, a total of 3290 valid typical soil profiles were used in this study, all of which contained SOC data, and 1964 profiles contained SIC data. One of the achievements of the second soil survey is that it contains detailed profile locations as well as soil chemical and physical properties. Each typical soil profile was dug more than 100 cm deep, and 3–5 soil layers in each profile were identified based on the soil color, structure, texture, and so on. In the 2010s (mainly in 2014 and 2015), a total of 2043 typical soil profiles were revisited to compile the Chinese Soil Series, all of which contained SOC data, and 1254 profiles contained SIC data.

Table 2. Sample numbers for different soil orders in North China.

Ages	Soil Orders		Samples		Ages	Soil Orders		Samples
1980s	CST *	ST *	SOC	SIC	2010s	CST *	ST *	SOC	SIC
	Histosols	Histosols ***	2	3		Histosols	Histosols ***	5	5
	Anthrosols	—	182	43		Anthrosols	—	64	45
	Vertosols	Vertosols	21	7		Vertosols	Vertosols	26	3
	Aridosols	Aridosols	337	261		Aridosols	Aridosols	279	219
	Halosols	Aridisols **	182	121		Halosols	Aridisols **	102	73
		Alfisols **					Alfisols **
		Inceptisols **					Inceptisols **
	Gleyosols	Inceptisols ***	22	9		Gleyosols	Inceptisols ***	24	4
		Gelisols **					Gelisols **
	Isohumosols	Mollisols	228	148		Isohumosols	Mollisols	162	92
	Argosols	Alfisols ***	709	291		Argosols	Alfisols ***	576	259
		Ultisols **					Ultisols **
		Mollisols **					Mollisols **
	Cambosols	Inceptisols **	1191	840		Cambosols	Inceptisols **	621	411
		Mollisols **					Mollisols **
		Gelisols **					Gelisols **
	Primosols	Entisols ***	370	235		Primosols	Entisols ***	166	114
		Gelisols **					Gelisols **

* “CST” represents Chinese soil taxonomy, and “ST” is the soil taxonomy of America. In Table 2, “***” means the majority of the two soil orders is equivalent, while “**” means portion of the two soil orders is equivalent. The number of SOC samples in each soil order is more than that in SIC because both SOC and SIC are derived from the information of soil physical and chemical properties in typical profiles, while most of the typical profiles contain organic matter information and less inorganic carbon information. The total number of samples is different from the number of typical profiles because the typical profiles located in water, glaciers, and deserts are removed.

lists the number of typical profiles containing SOC and SIC information under different soil orders from the 1980s and the 2010s. Typical profiles of 10 soil orders in North China were sampled in both the 1980s and the 2010s. Cambosols are the soil order with the most typical profiles. They benefit from the relatively uniform distribution in the study area.

In this study, soil profiles from the 1980s and the 2010s were systematically investigated, and the surface organic matter layer (usually including the litter layer, the humus layer, and the O layer) was recorded (about 53 samples from the 1980s and 10 samples from the 2010s, respectively). However, in view of the fact that this study aims to quantify and analyze the carbon stocks and their changes within the mineral soil, following the definition criteria of the normalized mineral soil profile in the conventional soil carbon pool assessment, we excluded all the data of the surface O layer in the subsequent analysis. Therefore, the results of SOC reported in this paper are only based on the sample data of mineral soil layers (the Aa layer, the Ap layer, the P layer, the Br layer, etc.).

The same sampling and preservation methods were used in the 1980s and the 2010s. Sampling sites with typical representative environments were located, and soil profiles 1 m wide, 1.5–2 m long, and 1–1.5 m deep were dug and sampled using the ring knife sampling method after profile delineation. Fresh soil samples were preserved in cloth bags for the determination of chemical properties, sent to the laboratory on the same day to be air-dried, and stored at room temperature (<25 °C) with <70% humidity in a protected place. In the 1980s and the 2010s, the soil organic matter (SOM) content was determined by using the Walkley–Black method [33], and then SOC data were calculated by the measured SOM data using a “Van Bemmelen Factor” of 0.58 [34]. The air-dried soil samples were measured by wet oxidation using an acidified potassium dichromate solution (K₂Cr₂O₇ + H₂SO₄) after grinding, and the acidity titration method was used to determine the content of CaCO₃ equivalents.

2.3. Generating Soil Samples at a Set of Standard Depths

The SOC content showed a decreasing trend from the topsoil to the subsoil on the vertical profiles [7,13,35], and the following formula is the ideal simulation model with more applications while satisfying the vertical trend of SOC [36,37]. The relationship between SOC and soil depth on the vertical profiles was simulated according to the following formula:

$$Y={\displaystyle \frac{1}{c_1+c_{2}X}},$$

(1)

where Y is the SOC content (g kg⁻¹), X is the soil depth (cm), and c₁ and c₂ are constants calculated from two sets of observed data. In the above equation, it is necessary to calculate the Y value when X is 10, 30, 50, 70, and 90, respectively; then, we regarded the values at the five center points as the SOC content in the 0–20, 20–40, 40–60, 60–80, and 80–100 cm layers, respectively.

The SIC content is constrained by the actual data and does not satisfy the decreasing trend from topsoil to subsoil. Therefore, considering the above reasons, we normalized the SIC content data of different thicknesses of each soil layer to the standard depths, enabling the comparison with SOC, and it is expressed as [30]

$${\mathrm{C}}_{a–b}={\displaystyle \sum C_i\times \frac{H_i}{20}},$$

(2)

where C_a–b is the SIC content (g kg⁻¹) in the a–b cm layer, C_i is the SIC content of layer i before normalization, and H_i is the thickness (<20 cm) of layer i. Then, the SIC data of different thicknesses of soil layers can be normalized to the SIC content of five layers in the one-meter profile with 20 cm intervals.

2.4. Geographic Weighting Regression Kriging Analysis

Geographic weighting regression (GWR) [31] was used to predict the distribution of the SOC and SIC content from the 1980s and the 2010s. The geographically weighted regression kriging (GWRK) model is a combinatorial method consisting of the GWR method and the ordinary kriging (OK) method, which considers both the spatial non-stationarity of regression analysis and the spatial autocorrelation of regression variables. It can better reflect the spatial variabilities of SOC and SIC, and its expression [38] is as follows:

$$y_{\mathit{GWRK}}\left({\mu }_i, v_i\right)=y_{\mathit{GWR}}\left({\mu }_i, v_i\right)+{\epsilon }_{\mathit{OK}}\left({\mu }_i, v_i\right),$$

(3)

where y_GWRK(μ_i, v_i) is the predicted value of the SOC or SIC content at point i, y_GWR(μ_i, v_i) is the GWR result at point i, and ε_OK(μ_i, v_i) is the result of OK interpolation of the residuals at point i. The GWR method was implemented using GWR4.0. OK interpolation was implemented using ArcGIS Desktop10.6; meanwhile, the semi-variational function and the K-Bessel model were used as the same parameters in all interpolations.

This paper uses four commonly used regression model evaluation indicators to evaluate the accuracy of the GWRK model, including the mean squared error (MSE), the root mean squared error (RMSE), the mean absolute error (MAE) and R-squared (R²). In the expression, y_i is the real value, y_GWRK is the predicted value, $\overline{y}$ is the average value of y_i, and n represents the number of samples.

$$\mathit{MSE}={\displaystyle \frac{1}{n}}{{\displaystyle \sum _{i=1}^n\left(y_i-y_{GWRK}\right)}}^2,$$

(4)

$$\mathit{RMSE}=\sqrt{{\displaystyle \frac{1}{n}}{\displaystyle \sum _{i=1}^n{\left(y_i-y_{GWRK}\right)}^2}}=\sqrt{MSE},$$

(5)

$$\mathit{MAE}={\displaystyle \frac{1}{n}}{\displaystyle \sum _{i=1}^n\left|y_i-y_{GWRK}\right|},$$

(6)

$$R^2=1-{\displaystyle \frac{{\displaystyle \sum _{i=1}^n{\left(y_i-y_{GWRK}\right)}^2}}{{\displaystyle \sum _{i=1}^n{\left(y_i-\overline{y}\right)}^2}}},$$

(7)

2.5. Estimation Methods for Soil Carbon Stocks

Since the purpose of this paper is to calculate the “effective” C pool in the soil that actually participates in the biogeochemical cycle, interacts closely with plants and microorganisms, and directly contributes to soil fertility and ecosystem function, the calculation of the reserves of soil organic carbon stocks (SOCs) and inorganic carbon stocks (SICs) is based on the active fraction. The formulas are [39,40,41] as follows:

$$\mathit{SOCs} \left(\mathit{SICs}\right) = M \times \mathit{SOCD} \left(\mathit{SICD}\right),$$

(8)

$$\mathit{SOCD} \left(\mathit{SICD}\right) = {\rho }_k{ \times P}_k{ \times D}_k \times (1 -S_k{) \times 10}^{-2}(\times 0.12),$$

(9)

where SOCs (SICs) are the soil organic (inorganic) carbon stocks (kg), M is the objective area (m²), SOCD (SICD) is the soil organic (inorganic) carbon density (kg m⁻²), k is the number of soil layers, ρ_k is the BD (g cm⁻³), P_k is the soil organic (inorganic) content (g kg⁻¹) in layer k, D_k is the thickness of layer k (the standard thickness is 20 cm in this study), and S_k denotes the CF (>2 mm, %). Otherwise, the number of 0.12 corresponds to the molar fraction of C in CaCO₃ (~CaCO₃ equivalent) used for converting the measured carbonates into SICs [24]. Furthermore, soil C stocks at each soil layer were calculated by adding SOCs and SICs [24]:

$$\mathit{soil} C \mathit{stocks} = \mathit{SOCs} + \mathit{SICs},$$

(10)

where soil C stocks are the total soil C stocks (Gt, 1 Gt = 10¹² kg) and SOCs and SICs are the soil organic and inorganic carbon stocks (Gt). The variation in soil C stocks over a period of time was then analyzed. The calculation formula [24] is as follows:

$$\Delta Q_c={\mathit{SCs}}_{t_2}-{\mathit{SCs}}_{t_1},$$

(11)

where SCs_t1 and SCs_t2 denote the soil carbon stocks (Gt) at times t₁ and t₂ and ΔQ_c denotes the carbon stock changes (Gt) during times t₁ to t₂. We regarded the positive (ΔQ_c > 0) results as “soil C sinks”; otherwise, they were regarded (negative, ΔQ_c < 0) as “soil C sources”.

2.6. Analysis of Dynamic Contributions

To ascertain the contributions of SOC sand SIC dynamics to C sources/sinks, this study employs path coefficients derived from partial least-squares path modeling (PLS-PM) to quantify their respective effects. PLS-PM applies multiple latent variables (SOC and SIC dynamics) and response variables (soil C stock dynamics) to a causal network designed to calculate the probability of effects [41]. PLS-PM uses variance-based operations, relying on the variance of each dimension. Each latent variable is a linear combination of observed variables, which can be used as a reflection or formative indicator. In PLS-PM, recursive path models can be analyzed, and a single dimension (latent variable) with one manifest indicator is permissible. The framework comprises an outer model (measurement model) and an inner model (structural model). The path coefficient β_ji is solved by the least-squares method, which reflects the standardized influence intensity of the exogenous latent variable ξ_i (the changes in SOCs or SICs in this study) and the endogenous latent variable η_j (the changes in soil C stocks in this study). The calculation for the path coefficients formula [41] is as follows:

$${\beta }_{\mathit{ji}}={\displaystyle \frac{\mathit{Cov}\left({\xi }_i,{\eta }_j\right)}{Var\left({\xi }_i\right)}}\times {\displaystyle \frac{{\sigma }_{\xi i}}{{\sigma }_{\eta j}}},$$

(12)

where σ is the standard deviation of the latent variable score. The statistical significance of all path coefficients was tested by 1000 Bootstrap sampling tests, and a 95% confidence interval excluding 0 was considered significant (α = 0.05).

Given that each latent variable in this study contains only one manifest theme, the analysis necessarily focuses on the inner model. Therefore, this study focuses more on the stability of the internal model. The Goodness of Fit (gof) is a comprehensive index for evaluating the goodness of fit of global models in PLS-PM [42]. Its core is to integrate the fitting quality of the relationship between the outer models and the inner models. The formulas are as follows [42]:

$$gof=\sqrt{\overline{Com}*\overline{R^2}},$$

(13)

$$\overline{\mathit{Com}}={\displaystyle \frac{1}{p}}{\displaystyle \sum _{j=1}^J{\displaystyle \sum _{k=1}^{K_j}{\lambda }_{\mathit{jk}}^2}},$$

(14)

$$\overline{R^2}={\displaystyle \frac{1}{m}}{\displaystyle \sum _{j=1}^m{R}_j^2}$$

(15)

where p is the number of explicit variables and λ_jk is the normalized load of the kth explicit variable under the jth latent variable. m is the number of endogenous latent variables, and R_j is the explanatory variance of the jth endogenous latent variable. When the gof value is greater than 0.36, it can be considered that the model prediction results are excellent and statistically significant.

3. Results

3.1. Spatial–Temporal Variation in Soil Organic Carbon in North China

3.1.1. Statistical Vertical Dynamics of Soil Organic Carbon Density and Stocks

As illustrated in Figure 3, significant increases in soil organic carbon density (SOCD) and stocks (SOCs) occurred across all deeper layers (20–100 cm) in North China from the 1980s and the 2010s, contrasting with the decrease observed in topsoil (0–20 cm). The mean SOCD increased by 0.02 kg m⁻² in the 20–40 cm layer, 0.12 kg m⁻² in 40–60 cm, 0.17 kg m⁻² in 60–80 cm, and 0.15 kg m⁻² in 80–100 cm, while it decreased by 0.29 kg m⁻² in the 0–20 cm layer. Over the entire one-meter profile, total SOCs showed a net accumulation of 1.17 Gt, increasing from 57.61 ± 0.92 Gt in the 1980s to 58.79 ± 1.08 Gt in the 2010s. In terms of the depth layer, SOCs increased by 0.35 Gt (20–40 cm), 0.60 Gt (40–60 cm), 0.93 Gt (60–80 cm), and 0.77 Gt (80–100 cm) but decreased significantly by 1.48 Gt in the topsoil (0–20 cm).

3.1.2. Spatial Distribution of Vertical Soil Organic Carbon Density and Stocks

Figure 4 and Figure 5 indicate that SOCD and SOCs were generally higher in the Northeast China Plain and the southern Qinghai–Tibet Plateau. Conversely, lower SOCD and SOC values characterized southern Xinjiang and western Inner Mongolia. These figures further reveal that SOC decreased significantly within the central northern arid and semi-arid regions and the Qinghai–Tibet Plateau. Conversely, SOC increased in the Huang–Huai–Hai Plain (excluding the Shandong hills), the Loess Plateau, the Middle-lower Yangtze Plain, and northern Xinjiang. A distinct geographic partitioning of SOC and its vertical changes was found across North China.

3.2. Spatial–Temporal Variation in Soil Inorganic Carbon in North China

3.2.1. Statistical Vertical Dynamics of Soil Inorganic Carbon Density and Stocks

Figure 6 indicates that soil inorganic carbon density (SICD) and stocks (SICs) decreased in deeper soil layers (20–40, 40–60, 60–80, and 80–100 cm) but only increased in the 0–20 cm layer. The mean value of SICD showed a significant reduction of 1.13 kg m⁻² in the one-meter soil profile, decreasing from 13.02 ± 0.68 kg m⁻² in the 1980s to 11.88 ± 0.16 kg m⁻² in the 2010s. Specifically, when examining the changes within individual layers, the results revealed reductions of 0.10 kg m⁻² in the 20–40 cm layer, 0.12 kg m⁻² in the 40–60 cm layer, 0.50 kg m⁻² in the 60–80 cm layer, and 0.52 kg m⁻² in the 80–100 cm layer. In contrast, the 0–20 cm layer showed a slight increase, with a recorded increment of 0.06 kg m⁻².

Furthermore, there was a substantial overall SIC decrease of 7.03 Gt. The SICs declined by 0.93, 0.90, 2.63, and 2.89 Gt in the 20–40, 40–60, 60–80, and 80–100 cm layers, respectively. Interestingly, the 0–20 cm layer exhibited an increase of 0.33 Gt, highlighting the contrasting change trends between topsoil and deeper soil layers for SOCs and SICs.

3.2.2. Spatial Distribution of Soil Inorganic Carbon Density and Stocks

Figure 7 and Figure 8 display that higher SICD and SIC values were observed in the northern arid and semi-arid regions, particularly in southern Xinjiang, whereas, the Northeast China Plain exhibited lower SICD and SICs. Additionally, the results also indicated that the entire arid and semi-arid regions received significant decreases in SICs, while the southern region of North China received SIC increases. This distinct spatial pattern reveals a fundamental divergence in SIC dynamics, with significant losses dominating the arid and semi-arid cores, contrasting with gains occurring in specific relative humid regions.

3.3. Spatial–Temporal Variation in Soil Carbon Stocks

3.3.1. Vertical Dynamics of Soil Carbon Stocks

In North China, a significant loss of soil carbon (C) stocks was observed, amounting to 5.86 Gt in the one-meter soil profile (Figure 9a). It decreased from 126.24 ± 3.3 Gt in the 1980s to 120.38 ± 0.11 Gt in the 2010s. Soil C stocks decreased across all five soil layers from the 1980s to the 2010s, with losses recorded at 1.15 Gt in the 0–20 cm layer, 0.58 Gt in the 20–40 cm layer, 0.30 Gt in the 40–60 cm layer, 1.70 Gt in the 60–80 cm layer, and the most substantial loss of 2.12 Gt in the 80–100 cm layer.

As shown in Figure 9b and

Table 3. Vertical changes in soil organic carbon stocks, organic carbon stocks, and soil carbon stocks (Gt).

SOIL Depth	0–20 cm	20–40 cm	40–60 cm	60–80 cm	80–100 cm	One-Meter
Stocks Changes
Soil organic carbon stocks	−1.48	0.35	0.60	0.93	0.77	1.17
Soil inorganic carbon stocks	0.33	−0.93	−0.90	−2.63	−2.89	−7.03
Soil carbon stocks	−1.15	−0.58	−0.30	−1.70	−2.12	−5.86

, the magnitude of SIC losses significantly surpassed the increases in SOCs within the 20–100 cm depth interval (specifically across the 20–40, 40–60, 60–80, and 80–100 cm layers). Quantitative analysis indicated a net SOC accumulation of 1.17 Gt across the one-meter soil profile. However, this increase was insufficient to compensate for the substantial SIC losses of 7.03 Gt. Consequently, the contrasting dynamics of these carbon components led to an overall net decrease of 5.86 Gt in soil C stocks within the one-meter profile.

According to the vertical dynamics of SOCs, SICs, and soil C stocks shown in the above results, the soil depth layers were re-divided. The 0–20 cm soil layer was regarded as the topsoil, and the 20–40, 40–60, 60–80, and 80–100 cm layers were regarded as the deeper soil. The statistical data in Figure 10 show distinct vertical stratification patterns in SOC and SIC dynamics between the topsoil and the deeper soil. In the topsoil, a SOC decrease of 1.48 Gt partially compensated for SICs’ slight increase of 0.33 Gt, leading to a net carbon loss of 1.15 Gt within this layer. Conversely, across deeper soil layers, significant SIC losses of 7.35 Gt substantially exceeded an SOC accumulation of 2.65 Gt, resulting in a net decrease of 4.71 Gt. The combined effect of carbon losses in both the topsoil and deeper soil culminated in a total net reduction of 5.86 Gt within the one-meter profile.

3.3.2. Spatial Distribution of Soil Carbon Stocks

Figure 11 reveals distinct spatial patterns in soil C stocks across North China. Higher stocks characterized the northern arid and semi-arid regions and the Qinghai–Tibet Plateau, whereas lower stocks were observed in the Northeast China Plain, the Huang–Huai–Hai Plain, and the Middle-lower Yangtze Plain. Crucially, temporal analysis showed significant decreases in soil C stocks within the northern arid and semi-arid regions. Conversely, the Northeast China Plain, the Loess Plateau, the Huang–Huai–Hai Plain, and the Middle-lower Yangtze Plain exhibited substantial increases in soil C stocks from the 1980s to the 2010s.

3.4. Soil Carbon Sinks and Source Detection

North China had obvious carbon sources in the one-meter soil profile (Figure 12). This was due to the fact that 34.42% and 65.58% of the regions were recognized as C sinks and C sources in the one-meter profile, respectively. In addition, the C source area exceeded the C sink area across all soil layers. The percentages of the areas which were identified as C sinks were 43.32% in the 0–20 cm layer, 47.64% in the 20–40 cm layer, 55.83% in the 40–60 cm layer, 45.85% in the 60–80 cm layer, and 38.39% in the 80–100 cm layer. Conversely, the areas classified as C sources were 56.68% in the 0–20 cm layer, 52.36% in the 20–40 cm layer, 44.17% in the 40–60 cm layer, 54.15% in the 60–80 cm layer, and 61.61% in the 80–100 cm layer. The subregions with more C sink areas across the one-meter soil profile included the Loess Plateau, the Huang–Huai–Hai Plain, the Middle-lower Yangtze Plain, and the Northeast China Plain. The northern arid and semi-arid regions and the Qinghai–Tibet Plateau were the subregions that exhibited a greater prevalence of C sources.

4. Discussion

4.1. Uncertainty and Sensitivity of the Geographic Weighting Regression Kriging Model

We collected substantial soil samples from two soil surveys conducted in the 1980s and the 2010s. The sample density was higher than in most related studies. Therefore, our prediction results were closer to the true spatial distribution of soil carbon. The existence of outliers induces distortion in the predictive–observed value-fitting plot, resulting in pronounced blank areas. As shown in Figure 13, the geographically weighted regression kriging (GWRK) model could significantly reduce the outliers of the sample points [31,38].

From the perspective of the uncertainty and sensitivity of the model’s prediction, the mean values of the MSE, RMSE, MAE, and R² of soil organic carbon (SOC) are 15.24, 3.9, 3.45, and 0.39 (Figure A1), respectively. The mean values of soil inorganic carbon (SIC) are 16.46, 4.06, 2.82, and 0.41 (Figure A2), respectively. Combined with the target object range that the SOC content ranges from 0 to 152 and the SIC content ranges from 0 to 166, it can be considered that the prediction results of GWRK are acceptable.

4.2. Soil Carbon Stocks in the One-Meter Profile

Our research revealed that soil inorganic carbon stocks (SICs) in the one-meter profile across North China are approximately 1~1.15 times greater than soil organic carbon stocks (SOCs). The finding is at odds with other estimates yet aligns with our hypothesis (hypothesis (i)). On a global scale, the composition of soil C stocks is roughly divided into 60% for SOCs and 40% for SICs. Particularly for China, the estimated SOCs are around 83.8 Pg C, while the estimates for SICs vary between 53.3 and 77.9 Pg C, as reported by [17]. The opposite results observed in this study may stem from the predominant presence of arid and semi-arid regions within our study area, which account for 65% of North China. Previous studies [17,23,24] indicate that drought conditions tend to promote the accumulation of SICs, contributing to a scenario where SICs can be up to 2.1 times higher than SOCs. Consequently, the findings are well within a plausible estimation range. Although this study is limited to a soil depth of one meter and fails to fully describe the impact of >1 m deeper soil processes, future research urgently needs to carry out multi-factor collaborative analysis by collecting deeper continuous profiles and combining various soil types on a typical small watershed scale in order to achieve breakthroughs in key issues such as deep C dynamics.

4.3. Changes in Topsoil Carbon in North China

Soil C pools are a crucial property that plays significant roles in determining soil quality and fertility, significantly impacting soil aggregation, water retention, cation/anion exchangeability, and nutrient availability [2]. The amount of global soil C in the top 1 m depth is much larger than the C stored in plants and the atmosphere [16]. It makes sense to analyze the dynamics of soil C pools. Due to the fact that soil C pools are composed of SOC and SIC, which influence the dynamics of soil C pools directly, it is important to analyze SOCs and SICs separately and vertically. According to the estimation, it was noted that SOC decreased while SIC slightly increased in topsoil, and Figure 14a indicates an increase in the SOCs of cropland within the study area. The increasing SOC in cropland topsoil has been reported by several studies [5,6,7,8,13]. Xie et al. [7] found that topsoil and subsoil received increments in SOCs by 33% and 32%, respectively, following long-term straw return practices in the Northeast Plain. Han et al. [8] found that the surface SOC content in the North China Plain (cultivated land accounts for more than 64%, which is the largest contiguous cultivated land concentration area in China) has increased by 73% since the 1980s. In contrast, Our findings indicated that surface SOCs in North China decreased significantly, offsetting the slight increase in SICs, ultimately resulting in a reduction of total C stocks in topsoil. The difference inferred above could be briefly explained by the fact that ongoing and acute climate changes lead to SOC losses through higher temperatures and vegetation collapse at lager spatial scales [43,44]. Interestingly, these SOC losses mainly occurred in the northern arid and semi-arid regions (except the central region) and the Qinghai–Tibet Plateau in North China, which hosts 62.89% of the grasslands and unused lands (mainly the Gobi desert) with huge SOC losses (Figure 14a). The SOCs in these two land use types exhibit pronounced sensitivity to abrupt climatic perturbations, with rising temperatures and prolonged drought conditions driving significant losses of SOCs [45,46,47,48].

The process of SIC turnover with degradation and formation or inputs is related to changes in calcium (Ca²⁺), hydrogen (H⁺) and bicarbonate (HCO₃^–) concentrations, soil water content, and CO₂ pressure, which can directly stimulate the equilibrium reaction of carbonate dissolution [12,49]. Song et al. [17] observed an overall decrease in SICs, averaging 11.33 g m⁻² yr⁻¹ in the 0–30 cm layer from the 1980s to the 2010s; this decrease has been attributed to substantial nitrogen (N) inputs in intensified agriculture with large amounts of chemical N fertilizers (~0.135 t ha⁻¹ yr⁻¹) and huge N deposition from the atmosphere. Such factors greatly accelerate soil acidification and carbonate dissolution. Moreover, the rise in atmospheric CO₂ concentrations with enhancing pressure and average precipitation with increasing soil water content will facilitate the leaching and dissolution of carbonates [23,50]. However, our observations regarding the increase in SICs appear inconsistent with the findings reported by Song et al. [17]. This discrepancy can be explained by the following reasons: Firstly, our samples have a higher density, cover a larger area, consist of various types of land use, and have more realistic spatial conditions. Secondly, such differences may be related to our different depths. Although our results showed a slight increase in SICs in the 0–20 cm layer, a significant decrease in SICs was detected in the 20–40 cm layer, which could counteract the slight increase, thereby showing an SIC decrement of 0.14 Gt in the 0–30 cm layer (Figure 14b). Thirdly, that is mechanistically linked to recurrent aeolian dust deposition (particularly from frequent dust storms in the northern arid and semi-arid regions). The large amount of exogenous carbonate flux introduced by atmospheric dustfall and dust storms complements SIC [51].

4.4. Soil Inorganic Carbon Stock Losses Dominated Deeper Soil Carbon Sources

As shown in Figure 15, the β_ji in partial least-squares path modeling (PLS-PM) revealed that SICs had stronger contributions over C sources in the 20–100 cm deeper soil (β_ji = 0.81) and the one-meter depth soil (β_ji = 0.77), while SOCs demonstrated stronger contributions on C sinks in the one-meter depth soil (β_ji = 0.72) compared to other soil C dynamics. Therefore, soil C stock losses were not constrained by SOC losses in the 0–20 cm topsoil but by SIC losses in the 20–100 cm deeper soil, consistent with hypotheses (ii) and (iii). The variations in SOC and SIC exhibit distinct patterns along with the soil profile, characterized by an increase in SOC and a significant decrease in SIC within deeper soil.

As expected, the launch of ecological restoration programs from 1999, such as the Grain for Green Project, the Returning Grazing Land to Grassland Program, and the Grassland Ecological Compensation Policy Program, by China has achieved great success in SOC sequestration and sinks [8,9,10,11,12]. Higher SOC accumulation of grassland and forestland has been approved in the arid and semi-arid areas by researchers for its susceptible to survival and higher underground biomass than above ground [52,53]. Therefore, vegetation restoration-induced C accumulation by root organisms forms larger C sinks in the 60–80 cm layer, reflecting the positive effects of the vegetation restoration policy for China’s Green Business. The results of SOCs in this study also found that significant increases in SOCs from the 1980s to the 2010s were observed in subregions of vegetation restoration projects in North China, including the Loess Plateau and most of the northern arid and semi-arid regions. As of 2015, about 88.2% of the Loess Plateau region has achieved effective improvement in vegetation cover [54]. The growth rate of the normalized difference vegetation index (NDVI) in northern Xinjiang ranks first in Xinjiang [55]. The vegetation restoration process in the above areas is highly consistent with the spatial distribution of the areas with significantly increased SOCs in this study, which further supports our conclusions.

The significant losses of SICs in the 20–100 cm deeper soil should be analyzed from the following two aspects: consumption and formation. For the consumption of SIC, soil acidification poses a major threat to SIC storage, with future scenarios identifying pH as the dominant predictor (29%), dwarfing other climate factors (10.2%) [3,56]. Huang et al. [3] found that at least 1.13 ± 0.33 billion tonnes of SICs are lost to inland waters through soil every year, with China and India being the most strongly threatened. Worsening nitrogen-induced soil acidification may cause losses of up to 23 billion tons of SIC globally over the next three decades. From a micro perspective for SIC consumption, the carbonate dissolution process in the soil and atmospheric CO₂ environment during natural evolution should be as follows [57]:

Ca(1 − x)Mg_xCO₃ + H₂O + CO₂ → (1 − x)Ca²⁺ + xMg²⁺ + 2HCO₃^–,

(16)

Over the past few decades, soil acidification has been caused by additional anthropogenic and atmospheric nitrogen supplement, and the balance shown by the above formula has also been broken. The nitrification process after extra nitrogen input will produce protons. If the nitric acid produced by these processes is used to replace the carbonic acid in the following formula, the dissolution of carbonate becomes [57,58]

NH₄⁺ + 2O₂→ NO₃^– + H₂O + 2H⁺,

(17)

Ca(1 − x)Mg_xCO₃ + NH₄⁺ + 2O₂→ (1 − x)Ca²⁺ + xMg²⁺ + NO₃^– + CO₂ + 2H₂O,

(18)

In the above process, C is released into the atmosphere primarily as the main C sources in the form of CO₂, making it a key C source in the agricultural production process. In the regions with higher pH such as the northern arid and semi-arid regions, more CO₂ will continue to react with carbonates to form HCO₃^– [57,58], where

2Ca(1 − x)Mg_xCO₃+ NH₄⁺ + 2O₂ → 2(1 − x)Ca²⁺ + xMg²⁺ + NO₃^– + 2HCO₃^– + H₂O.

(19)

According to the study by Gandois et al. [58], this process is still a potential C source. This is the mechanism by which the most significant SIC losses were detected in the northern arid and semi-arid regions. Meanwhile, increased precipitation in North China may bring leaching of deeper Ca²⁺ and Mg²⁺ into the soil, resulting in SIC losses in deeper soil [16,59].

For the formation of SIC, Ca²⁺, and Mg²⁺, which rise from shallow water through capillary motion and form carbonates driven by seasonal evapotranspiration [10,11,60]. However, with the decrease of seasonal evapotranspiration [61,62], the formation of secondary carbonates in soil is limited. The limitation of SIC formation will further increase the losses of SIC. The “CaCO₃ equivalent” used in this paper contains soil carbonates such as secondary dolomite. Although the origin of primary dolomite is still controversial, the soil secondary processes (Reactions (16)–(19)) concerned in this study do not involve this deep theoretical issue. In addition to the two natural ways above, huge nitrogen inputs can also lead to soil acidification, which greatly consumes SICs [6,17,63]. When significant SIC losses exceed the accumulation of SOCs in deeper soil, the SC losses and C sources will be dominated by SICs losses. The C sources in deeper soil caused by SIC losses are likely to greatly hinder the global efforts on C sequestration in terrestrial ecosystems.

5. Conclusions

Our results revealed that soil inorganic carbon stocks are 1~1.15 times greater than organic carbon stocks in the one-meter soil profile in North China; however, soil carbon stocks reduced by 5.86 Gt; organic carbon stocks increased by 1.17 Gt, and inorganic carbon stocks decreased by 7.03 Gt. The soil organic carbon stock losses counteracted inorganic carbon stocks, causing soil carbon stock losses in the 0–20 cm topsoil. Significant soil inorganic carbon stock losses counteracted organic carbon stock increases, causing huge soil carbon stock losses in the 20–100 cm deeper soil. The results indicated that the carbon sources in deeper soil caused by inorganic carbon stock losses are likely to greatly hinder the global efforts on carbon sequestration in terrestrial ecosystems. Our findings highlight the need to consider both top and deeper dynamics of soil carbon stocks, especially inorganic carbon stock losses in deeper soil, and will serve as useful reference information for global soil carbon sequestration initiatives. Future research should focus on the use of isotope tracing technology to quantify the deep soil organic carbon loss flux and explore the carbonate dissolution–precipitation feedback under global climate and hydrology changes so as to analyze the specific formation mechanism of soil carbon sinks and sources.