Mechanisms driving spatial and temporal changes in soil organic carbon stocks in saline soils in a typical county of the western Songnen Plain, northeast China

The Songnen Plain encompasses a signi�cant grain-producing area and serves as a crucial commercial grain production base in China. Its western region, located within an agro-pastoral transitional zone, is particularly sensitive to environmental changes. Human activities have contributed to the escalating issues of sanding and salinization in this area. In recent years, there has been increasing attention on the in�uence of soil physical and chemical properties, topography, climate, and land use changes on soil organic carbon (SOC). However, there is limited understanding of the interplay between these factors and their combined impact on SOC. To address this gap, this study utilized the second soil census data of Tongyu County in 1982 and the latest �eld survey data conducted in 2022. It examined the spatial and temporal variations of soil organic carbon density (SOCD) and soil organic carbon storage (SOCS) in Tongyu County's surface (0–20 cm), subsurface (20–50 cm), and bottom (50–100 cm) layers using GIS technologies. Additionally, data-driven models, namely random forest regression and structural equation modeling, were employed to identify the environmental factors in�uencing SOCD distribution in different soil layers during two distinct phases. The results revealed several important �ndings: 1) From 1982 to 2022, SOCD in the surface, subsurface, and bottom soil layers of Tongyu County exhibited an average decrease of 0.65 kg·m − 2 , 0.34 kg·m − 2 , and 0.46 kg·m − 2 , respectively, resulting in a total reduction of 15.68 Tg C in carbon storage; 2) In 1982, the vertical distribution of SOCD was high at both ends and low in the middle location, but by 2022, it gradually decreased layer by layer; 3) Topographic factors solely in�uenced surface SOCD, while the in�uence of environmental humidity and land use on SOCD decreased with increasing depth. These �ndings provide valuable scienti�c insights for implementing regional soil carbon sequestration and soil nutrient conservation measures.


Introduction
Global climate change, triggered by greenhouse gas emissions such as carbon dioxide (CO 2 ), presents a signi cant challenge to human society in the 21st century.Throughout the past century, human activities have been the primary driver of climate change, largely due to the release of CO 2 into the atmosphere (Jones et al., 2023).To combat the escalating CO 2 levels, terrestrial carbon sequestration is being explored as a promising solution (Sheikh et al., 2014).Soil, as the largest carbon reservoir in terrestrial ecosystems, stores an estimated 1550 Pg of Soil Organic Carbon (SOC) (1 Pg = 1015 g) (Batjes, 1996).These gures are more than three times the size of the atmospheric carbon pool (Oelkers et al., 2008) and over four times the size of the biogenic carbon pool (Lal, 2004).Research indicates that global soil carbon sequestration has the potential to remove 0.4 ~ 1.2 Pg C yr − 1 from the atmosphere, accounting for 5 ~ 15% of global fossil fuel emissions (Lal, 2004).Being a crucial component of the global carbon cycle, even minor uctuations in the soil carbon pool can have a signi cant impact on greenhouse gas exchange between the soil and the atmosphere.However, the underlying mechanisms that govern soil carbon stock are still not well understood, leading to uncertainty about the role of soils in the global carbon cycle.
Considerable research has been carried out to explore the effects of climate, soil properties, and management practices on soil carbon sequestration (Alidoust et al., 2018;Hu et al., 2018;Lewis et al., 2019).Climatic factors, including precipitation and temperature, have a signi cant impact on vegetation distribution, growth, and the spatial distribution of SOC (Jia et al., 2017).Land use management practices, such as changes in land use (Deng et al., 2016) and afforestation (Korkanç, 2014), can greatly in uence soil organic carbon stocks (SOCS).The variations in vegetation productivity (Wu et al., 2017) and soil physicochemical properties (Wang Y et al., 2016) resulting from different land use management practices are closely associated with changes in SOCS.However, the relative importance of factors in uencing SOCS in different soil layers remains unclear, especially in inland saline soil areas like the western part of the Songnen Plain, China.
The variation in SOC in saline soils is in uenced by mean annual temperature (MAT) and annual precipitation (MAP).However, it remains unclear whether this variation is directly in uenced by climate or indirectly impacted through factors such as soil development and vegetation.Furthermore, the relative importance of these in uences is not well understood.Structural Equation Modeling (SEM) offers a suitable approach to explore the intricate relationships between variables, where variables can serve as both predictors and responses.SEM allows for the incorporation of unobserved variables (latent or constructive variables) into theoretical constructs represented by observed variables (explicit variables or indicators) (Grace et al., 2010).The model assumes a causal relationship between latent variables, enabling the disentanglement of correlations into direct and indirect effects.This facilitates the simultaneous assessment of regression relationships between variables (Prober et al., 2012).As a result, the direct and indirect effects of different combinations of factors on SOC stocks, including signi cant regression weights for potential interaction pathways, can be calculated.
This study aims to comprehensively analyze the spatial and temporal variations of SOC and examine the interrelationships between soil physicochemical properties, climate, topography, land use practices, and SOCD in different soil layers, focusing on a typical case in Tongyu County, located in the western Songnen Plain.The speci c research objectives are as follows: 1) to determine the spatial distribution characteristics of SOC stocks in 1982 and 2022, providing insights into the changes in SOC stocks over the past four decades; 2) to conduct a thorough analysis of the spatial and temporal patterns of SOC variation in the region during the 40-year period, offering a comprehensive understanding of SOC dynamics over time; and 3) to quantitatively assess the impacts of various drivers, such as soil properties, climate, topography, and land use changes, on SOCD using advanced modeling techniques like random forest and structural equation models, which will allow us to elucidate the complex relationships between these factors.By gaining a deeper understanding of these drivers and their in uence on SOC, we can make more accurate and quantitative predictions regarding future soil carbon sequestration, contributing to better-informed strategies for sustainable land management and climate change mitigation.

Study area
The study area covers an expansive land area of 8496 km 2 and is located in Tongyu County, situated between latitudes 44.23° and 45.27°N and longitudes 122.04° and 123.52°E, in the western part of the Songnen Plain in northeast China.The terrain in this region is predominantly at, with a gradual elevation increase towards the northwest and lower elevations towards the southeast.The average altitude ranges from 110 to 180 meters, with the highest point reaching 206 meters.The area experiences a typical temperate semi-humid continental monsoon climate, characterized by well-de ned four seasons.The mean annual temperature (MAT) hovers around 4℃, while the mean annual precipitation (MAP) ranges from 370 to 500 mm, with the majority occurring during July and August.Annual evaporation rates range from 900 to 1000 mm.Spring in this region is marked by inadequate water and heat, often accompanied by frequent wind and sandstorms, earning it the nickname "nine droughts in ten springs".The study area is blessed with three seasonal rivers: the Huolin River, Emmuta River, and Wenniugzhi River.Dominant soil types in the study area include chernozem, aeolian soil, meadow soil, and some solonetz soils (Ren et al., 2010).The primary land use and cover types observed in this region consist of cropland (38%), barren saline-alkali land (27%), and grassland (13%) (as of 2006).The local economy predominantly revolves around farming and livestock grazing, with a population of approximately 0.28 million residents (as of 2020).

Soil data Soil date in 1982
The soil data in1982 used in this study were sourced from the Second National Soil Survey of China.Speci cally, we utilized data from the Second National Soil Survey of Tongyu County, which included detailed information from 66 soil pro les.The locations of these soil pro les are illustrated in Fig. 1.Each soil pro le provided a wealth of information, such as taxonomic classi cation, geographical coordinates, elevation, depth of different soil horizons, organic matter content, gravel content (particle diameter > 2 mm), bulk density, and soil texture.In instances where data were missing, we addressed this by supplementing the values using the average capacity weight for the corresponding soil type.Furthermore, to align with the American soil class cation system, soil texture data were converted from the Kaczynski system using the method described by (Shangguan et al., 2014).

Soil data in 2022
In November 2022, we collected soil samples from 66 sites within the study area, with each site comprising different soil depths (0-20 cm, 20-50 cm, 50-100 cm), as speci ed by the approximate locations provided in the second soil census (Fig. 1).Multiple subsamples were collected and combined to create composite representative soil samples for each site.After removing stones and root material, the soil samples were freeze-dried, ground, and sieved to a particle size of < 2 mm for subsequent analysis.Sample size fractions were determined using a laser particle analyzer (Mastersizer 2000) after eliminating organic matter (with 10% hydrogen peroxide) and carbonate (with 20% hydrochloric acid).The pH of the samples was measured using a pH meter with a 1:2.5 w/v soil-water suspension.SOC content was determined by oxidizing potassium dichromate with external heating at high temperature, while total nitrogen (TN) content was measured using the Kjeldahl method.Cation exchange capacity (CEC) was assessed using the hexaammine-cobalt trichloride-spectrophotometer method.To ensure quality control, duplicate samples were inserted after every ten samples, and the analysis precision remained within ± 5%.

Environmental variables
In this study, we analyzed the relationship between SOCD and eight environmental variables, including MAP, MAT, elevation (ELE), slope, aspect, topographic wetness index (TWI), soil type (ST), and land use type (LUT).For the categorical variables land use type and soil type, we assigned numerical codes to represent the respective categories, as speci ed in Table 1.As the data were sourced from different sectors and platforms, we used ArcGIS 10.2 (ESRI Inc., USA) software to standardize the projection coordinates of the environmental variables.Subsequently, the values of these variables were associated with the sampling points for further modeling and analysis.

Calculation of the SOC densities and stocks
The soil pro les were geotechnically divided into major layers labeled as A, B, C, and D, either at a depth of 100 cm or at the underlying consolidated bedrock, following the methodology described by (Zhong et al., 2001).For this study, data from the Second National Soil Survey of China were converted into different depth increments (0-20 cm, 20-50 cm, and 50-100 cm) based on the soil-forming horizons.The calculation of weighted soil organic carbon content has been previously documented by (Zhong et al., 2001).SOCDi and pH for each soil layer i (i = 1, 2, and 3, corresponding to 0-20 cm, 20-50 cm, and 50-100 cm, respectively) were determined following the method outlined by (Liu et al., 2011).SOCDi,j (kg•m − 2 ) and SOCS (Tg) were computed using the following equations: Where SOCD i,j represents the soil organic carbon density (kg•m − 2 ) in layer i of the jth soil genus, 0.58 is the Bemmelen index for converting organic matter concentration (OM i ) to organic carbon content, T i,j , B i,j and F i,j denote the thickness (cm), average soil bulk weight (g•cm − 3 ), and volume fraction (> 2 mm) of layer i of the jth soil genus.SOCS represents the total soil organic carbon stock in the study area (Tg), S j denotes the soil distribution area of the jth soil genus (km 2 ), and 10 − 3 is the unit conversion factor.The SI units used are speci ed above.

Data analysis
The random forest (RF) algorithm (Liaw et al., 2002)  ).Within the RF algorithm, three key parameters are essential: the number of random regression trees (ntree), which determines the optimal number of nodes per tree splitting variable for the number of splits (mtry), and the minimum number of samples in terminal nodes (nodesize).For our study, we set the parameters ntree, mtry, and nodesize of the RF algorithm to 5000, 3, and 5, respectively.The RF algorithm was implemented using the randomForest package in R. To evaluate the importance of the model and the cross-validated R2, we employed the R package A3 with 5000 permutations of the response variables.Additionally, the importance of each driver was assessed using the R package rfPermute (Archer, 2013).After identifying the main drivers, we further assessed the contribution of these drivers using SEM, with all SEM analyses conducted using R (R Core Team, 2019).Over the course of four decades, the SOCS in the region experienced a decline from 66.29 Tg C to 50.61 Tg C (see Table 2).Among these changes, the transition from Barren to Cropland resulted in a net increase of 0.22 Tg C in SOCS, while the conversion between Cropland types contributed to a reduction of 10.92 Tg C in SOCS.These ndings align with previous studies, such as the estimates by (Tang et al., 2010), albeit with varying magnitudes of reduction.The extent of reduction in SOCS also corresponds to recent ecological restoration efforts, including initiatives like "returning farmland to grass" in Tongyu County.

Main Driving Factors of SOCD in Tongyu County
To identify the key driving factors of soil carbon content in Tongyu County, RF model was employed using all measured variables.The RF model explained 61.5%, 64.3%, and 55.8% of the variation in SOCD for the surface, subsurface, and bottom layers in 1982, and 82.2%, 77.9%, and 76.1% in 2022.The RF model results revealed that TN, pH, Silt, Sand, Clay, Slope, Aspect, TWI, MAP, and LUT were the most in uential drivers of SOCD in Tongyu County in 1982.Similarly, in 2022, TN, pH, Silt, Clay, CEC, Elevation, Aspect, Slope, TWI, MAP, MAT, and LUT were identi ed as the most important factors affecting soil carbon content (Fig. 5).The identi ed driving factors were utilized as inputs to construct SEM to examine the effects of these variables on SOCD in different soil layers and years (Fig. 6).In the 0-20 cm, 20-50 cm, and 50-100 cm depths, the measured factors accounted for 38%, 66%, and 76% of the variation in SOCD in 1982, and 46%, 60%, and 62% of the variation in SOCD in 2022 (Fig. 6).Table 3 presents the results of the SEM models, including the χ2, P-value, Goodness of Fit Index (GFI), and Root Mean Square Error of Approximation (RMSEA) values used to assess the suitability of the SEM model.

Discussion
In uences of Soil Properties on SOCD Soil properties play a critical role in determining the variation of SOC content.Soil texture exerts a signi cant in uence on the spatial distribution of SOC, as the clay and silt content in soil texture contribute to the formation of aggregates that protect SOC from decomposition, thereby promoting SOC sequestration (Šimanský et al., 2019).In Tongyu County, the surface is primarily composed of Quaternary loose sediments, with predominant loess-like subsand, sand, and chalk (Pang X et al., 2003).These soil textures are crucial factors contributing to the spatial variation of SOC. which is further supported by the ndings of this study (Figs. 5 and 6).On the other hand, the correlation between soil pH and SOC is weaker and signi cant only when grassland sites are excluded (Yu et al., 2014).Studies such as (Solly et al., 2020) have observed signi cant correlations between SOC content in forest soil topsoil (0-30 cm) and subsoil (30-120 cm) with the effects of CEC.(Wang et al., 2019) also reported a signi cant correlation between SOC and CEC under long-term fertilization (R 2 = 0.98, P < 0.01), which is consistent with the ndings of this study.
In summary, soil properties, including soil texture, TN, pH, and CEC, have signi cant in uences on the variation of SOCD.Soil texture affects SOC distribution through its impact on aggregation and protection, while TN and CEC serve as indicators of soil fertility and nutrient availability, thus affecting SOC content.However, the correlation between pH and SOC is less pronounced, particularly when grassland sites are excluded.These ndings align with previous research and contribute to our understanding of the relationships between soil properties and SOCD.
In uences of Topography on SOCD for soil moisture and provides a quantitative tool to consider the spatial variability of soil moisture in uenced by topography (Winzeler et al., 2022).(Li et al., 2018) observed the highest correlation coe cient (0.735, P < 0.001) between TWI and SOCD when studying the distribution of SOC across topography.Furthermore, (Hu et al., 2021) found a signi cant negative correlation between TWI and SOC, which aligns with the ndings of this study.This relationship can be attributed to the in uence of microorganisms on SOC decomposition and nutrient mineralization, ultimately impacting soil fertility and quality.
Soil salinity also affects microbial processes and activities, which can lead to changes in SOC and greenhouse gas emissions (Haj-Amor et al., 2022).In Tongyu County, soil salinity inhibits microbial growth and results in low levels of SOCD due to factors such as "salt follows water" dynamics, low oxygen content, and poor permeability (Fig. 6).It is worth noting that topographic factors signi cantly in uence SOCD primarily in the topsoil in Tongyu County, suggesting that their impact is more pronounced in the upper layers of the soil pro le in plains.

In uences of Climate on SOCD
Climate factors play a pivotal role in in uencing the spatial distribution of SOCD (Wang et al., 2018).Numerous prior studies (Batjes, 1996;Papatheodorou et al., 2004;Poeplau et al., 2023) have demonstrated a signi cant correlation between SOC and climatic variables, which this study further con rms (see Fig. 5 and Fig. 6).
SOCD is particularly impacted by rainfall and temperature.In natural ecosystems, the SOC content tends to decrease exponentially with rising temperatures (Willaarts et al., 2016).Climatic conditions (relatively low percentage of rainfall and high index of bioclimatic drought) in the agricultural area of Lesvos Island promote soil salinization and SOCD is lower in areas with high salinity (Lekka et al., 2023).MAT and MAP were higher in the upland soil region of northern Jiangsu Province, and both warmer temperatures and increased precipitation led to lower SOCD (Zhang et al., 2017).This is consistent with our ndings.Semiarid regions, characterized by relatively higher Mean Annual Precipitation (MAP) and larger cultivated areas compared to arid regions, experience increased soil erosion intensity (Mi et al., 2008).Soil erosion results in the loss of organic matter attached to light clay particles in the topsoil, leading to reduced soil organic matter accumulation (Wynn et al., 2006).
On the other hand, arid environments are less conducive to plant growth, and high temperatures can enhance microbial activity, accelerating the mineralization of organic matter while restricting the synthesis and accumulation of soil humus, ultimately resulting in lower SOCD (Wang et al., 2023).
Organic matter residues from cultivated crops contribute to soil organic matter formation, offsetting the organic matter consumed through mineralization processes (Nyssen et al., 2008;Chang et al., 2012).
However, this has led to a gradual decline in soil fertility, particularly in recent years, due to the reduction of arable land area and increasing population pressure (Wang S et al., 2016).

Limitations the Current Study
There are limitations in the data sources for Soil Organic Carbon (SOC) in 1982.During the early 1980s, GPS technology was not available in China, and as a result, the locations of all sampling sites in 1982 were recorded solely based on textual descriptions.However, it is important to note that the SOC data for 1982 were obtained from the Second National Soil Survey of China, which remains the most comprehensive and detailed historical soil database available to date (Zhao et al., 2018).

Conclusions
The ndings of this study reveal a signi cant decline in Soil Organic Carbon Density (SOCD) in Tongyu County, with a total decrease of 15.68 Tg C in Soil Organic Carbon Stocks (SOCS) between 1982 and 2022.The analysis indicates that soil properties and land use changes are the primary factors in uencing changes in SOCD at a soil depth of 0-20 cm, while topographic and climatic factors have a relatively lesser impact compared to soil properties.However, it's worth noting that the model used in this study explains less than 40% of the variation in soil carbon accumulation at the 0-20 cm soil depth, suggesting the importance of incorporating additional factors such as organic carbon input quality and soil microbial data to further improve the model's accuracy.
Furthermore, the in uence of topographic factors is mainly observed in the topsoil, while the effects of ambient moisture and land use on SOCD decrease with increasing soil depth.These insights obtained from this study can greatly contribute to the development of effective strategies for soil and water conservation, environmental protection, agricultural production planning, and carbon sequestration in inland saline soils.

Supplementary Files
Page 25/25 This is a list of supplementary les associated with this preprint.Click to download.

Figure 2
Figure2and S1 present the descriptive statistics of SOCD across all sampling sites, along with selected environmental variables.In 1982, the mean values of SOCD for the surface (0-20 cm), subsurface(20-

Figure 1 Soil
Figure 1

Figure 2 Mean
Figure 2

Figure 5 Importance
Figure 5

Table 1
Description of the major classes used as variables and the corresponding numeric code.

Table 2
Changes in SOCS under different land-use patterns during 1982-2022.The area of change is less than 1km 2 , and the part is omitted from the table.

Table 3
SEM model results.
(Obu et al., 2017;Zhai et al., 2019) fertility and quality in agroecosystems(Li et al., 2022).Numerous studies have con rmed a strong correlation between SOC and TN(Obu et al., 2017;Zhai et al., 2019), (McBratney et al., 2003) role in the soil formation process(Wang Y et al., 2016).It not only governs the redistribution of water and heat resources but also in uences the material cycling process and the intensity of soil ecosystems (Martin et al., 2011; Wang et al., 2018).Topography is an important factor that affects the ow of ecosystem materials and energy(McBratney et al., 2003).In the western part of Jilin Province, which includes Tongyu County, the region is primarily characterized by plains.Topographic factors such as slope and aspect are favorable for organic matter sequestration.TWI serves as a proxy (Song et al., 2016;Wang et al., 2017)f human activities(Adhikari et al., 2019).Different land uses result in varying degrees of disturbance to surface vegetation and soil, leading to signi cant differences in Soil Organic Carbon Density (SOCD)(Wang S et al., 2016;Yu et al., 2020).Our study further sheds light on the impact of land use change on SOCD.Over the past 40 years, we observed that reclamation of barren land increased carbon stocks by 0.22 Tg C, while the interconversion among different types of cropland resulted in a decrease in soil carbon stocks by 10.92 Tg C (Table2).Changes in land use patterns induce alterations in vegetation types, soil microorganisms, and soil physical and chemical properties, which in turn in uence the quantity and decomposition rate of soil organic matter, thereby impacting SOCD to varying degrees(Song et al., 2016;Wang et al., 2017).From 1982 to 2022, the dominant land use type in the region has been arable land, serving as China's main agricultural production base, and this has contributed to the declining trend in SOCD (Table