Land Management Scale and Net Carbon Effect of Farming in China: Spatial Spillover Effects and Threshold Characteristics

Abstract

The net carbon effect of farming is crucial for climate change mitigation, yet there is insufficient research on the impact of land management scale on it in China. This study aims to explore the magnitude and role of land management scale on the net carbon effect of farming at the spatial level, including threshold characteristics. Unlike previous studies focused on the domestic agricultural economy, this study employs ecological findings to calculate carbon sinks and certain carbon emissions. The carbon-balance ratio is used to characterise the net carbon effect of farming. The spatial Durbin model and threshold regression model were utilised with a sample of 30 provincial-level regions in China from 2004 to 2019. The results indicate that national farming generally exhibits a net sink effect, with significant interannual fluctuations. After applying robust standard errors, the expansion of the land management scale significantly increases sinks and reduces emissions, and it has a positive spatial spillover effect on the carbon-balance ratio, demonstrating significant spatial heterogeneity. Furthermore, as the land management scale expands, the influence of rural residents’ income and education level on the carbon-balance ratio changes direction, showing significant non-linear relationship characteristics.

1. Introduction

Human activities, particularly greenhouse gas (GHG) emissions, are widely recognised as significant contributors to global warming [1]. Achieving the goal of limiting temperature rise to 1.5 °C and meeting global annual GHG emission targets necessitates that agriculture, forestry, and other land use sectors achieve carbon neutrality by approximately 2030, two decades earlier than the global target [2]. Reducing GHG emissions from the global food system is crucial for reaching the 1.5 °C target [3]. Additionally, farmland ecosystems are observed to function as modest carbon sinks [4], contributing to climate stability [5]. Given that agriculture both emits and sequesters carbon [6], understanding its net carbon effect is crucial.

Tian Yun introduced the concept of the net carbon effect in agricultural economics research in China, defining it as the difference between carbon sinks and emissions caused by agricultural activities [7]. This integrated concept of carbon sinks (CS) and emissions (CE) has garnered significant attention among Chinese agricultural economists for its innovative approach. Consequently, numerous studies have been conducted to explore the net carbon effect in agriculture. Current research primarily focuses on evaluating the net carbon sink (NCS) [7], its spatial and temporal variations [8], and the factors influencing it [9]. While different studies show varying results in calculating the carbon sink effect of Chinese agriculture [6], there is a consensus that Chinese agriculture overall acts as a net carbon sink, with indications that its NCS is increasing [10]. Geographically, Chinese agriculture exhibits significant spatial imbalances in its NCS [11]. Research by Chen et al. (2015) identifies a distinct spatial pattern, with a positive NCS in the southeast contrasting with predominantly negative values in the northwest [8]. Additionally, alternative approaches, such as using the ecological support coefficient (ESC) to assess NCS in agriculture, have been explored [12].

However, when it comes to accounting for CS, the majority of studies on agricultural economics have placed undue emphasis on the biogenic carbon contained in crops, in contrast to the prevailing consensus within the ecological community. For instance, Li et al. (2023) considered it appropriate to define the agricultural carbon sink as the amount of solid carbon converted by crops through photosynthesis, so an estimate of the CS was obtained by converting crop yields through the parameters of economic coefficient, water content coefficient and carbon sequestration rate coefficient guided by this idea [13]. In fact, except for a few studies that have noted the importance of soil organic carbon (SOC) in farming for their net carbon effect [14], the vast majority of existing studies have calculated CS in this way, but this approach still suffers from certain shortcomings. The uptake of carbon by crops through photosynthesis is not the same as CS, nor is the net uptake of carbon after excluding the CO₂ produced by the crop’s own respiration [15], even though they are so closely linked to CS. The former is often referred to in ecology as GPP (Gross Primary Productivity) and the latter as NPP (Net Primary Productivity). Moreover, such a method of calculation makes the results different and even incomparable due to the vagueness of the system boundaries, incomplete accounting indicators, and parameters that do not correspond to reality [6].

To comprehend the interplay between biological productivity and environmental change at the ecosystem or larger scales, traditional productivity concepts like Gross Primary Productivity (GPP) and Net Primary Productivity (NPP), initially focused on individual or group levels, are now deemed inadequate [16]. Woodwell et al. (1978) pioneered the concept of NEP (Net Ecosystem Productivity) to evaluate whether terrestrial ecosystems act as sources or sinks of atmospheric CO₂, calculated as the fraction of NPP remaining after deducting the carbon consumed by heterotrophic respiration (soil respiration) [17]. Since the 1990s, NEP has been pivotal in carbon sink studies. Schulze et al. (2000) introduced NBP (Net Biome Productivity) [18], however reducing NBP’s uncertainty remains a significant challenge in global carbon cycle research [19]. Presently, NEP is predominantly used to quantify carbon sinks in agriculture due to limited large-scale NBP data availability [20].

In contrast, ecological measures of CS in farmland ecosystems have concentrated on soil carbon fluxes in addition to crop carbon sinks. The net carbon effect of farming on atmospheric CO₂ concentrations ultimately depends upon changes in their soil carbon pools [21]. It is therefore evident that specific agricultural practices that are closely linked to land farming practices are pivotal in influencing the net carbon effect of farming. Agricultural measures such as conservation tillage, crop rotation, cover crops, and biochar application have been shown to effectively reduce CE from farming and increasing CS from farming [22,23]. Conversely, traditional agricultural practices such as deep ploughing, long-term monocropping, opening up forests and meadows to farmland, and burning agricultural residues are recognised as emitting large amounts of GHGs [24,25].

Farm size is a significant factor in determining how land is utilised in farming. Moreover, an alternative measure of farm size is provided by land management scale, as a proxy variable that reflects the size of farms in the area. Existing studies collectively indicate that there may be a significant correlation between the scale of land management and the CE of farming. For instance, Li et al. (2022) posit that the continuous expansion of land management scale is conducive to reducing agricultural carbon emissions [26]. Similarly, Cheng et al. (2023) also emphasised the pivotal role of an appropriate land management scale in fostering green agricultural development [27]. Nevertheless, further investigation is required to ascertain the relationship between land management scale and the net carbon effect of farming.

Summarising the existing research findings, it is evident that there is room for improvement in the NCS accounting methods used in current studies. Additionally, there is a notable gap in research concerning the relationship between land management scale and the net carbon effect of farming. Addressing these gaps, this study aims to enhance the literature by employing improved NCS accounting methods and conducting a detailed analysis of how land management scale impacts the net carbon effect of farming. This research aims to contribute to more accurate carbon management strategies in farming.

To achieve this objective, the study integrates data from the National Ecological Centre of China’s NEP dataset and utilises the carbon-balance ratio (CBR) to assess the net carbon effect of farming. Focusing on 30 regions in China from 2004 to 2019, the study employs spatial and threshold models to explore how land management scale influences the CBR. The analysis will delve into the influencing mechanisms and their underlying causes.

The following outlines the remaining sections of the paper: Section 2 outlines the theoretical mechanisms by which land management scale and income level impact the CBR, proposing corresponding theoretical hypotheses. In Section 3, variables and data are identified, and econometric models are constructed in baseline, spatial panel, and threshold panel forms. Section 4 briefly describes the evolution of the dependent variables and analyses the results of the different econometric models, with a focus on the possible causes of spatial spillovers and the threshold characteristics of rural residents’ income and education levels. Section 5 discusses the results of this paper in the context of other existing studies. Lastly, Section 6 concludes the paper and offers recommendations based on the findings.

2. Theoretical Mechanisms

2.1. Analysis of the Mechanisms Influencing Land Management Scale

To grasp the concept of land management scale, critical in Chinese agricultural economics research, it is important to distinguish it from the globally used term “farm size”. Land management scale refers to the area of land contracted by peasants, distinct in that the land is not owned individually but collectively by the village. Typically, peasants manage multiple plots rather than a single unified farm [28]. China introduced the household contract responsibility system in the 1980s, which revitalized the agricultural family as the fundamental unit of production and operation. While this system significantly boosted productivity, it did not necessarily lead to an expansion in land management scale. Unlike historical patterns where farm sizes grew with economic development [29], China’s current reality diverges due to factors like land transfer restrictions and incomplete migration of surplus labour from rural areas. According to the latest data from the Third National Agricultural Census, the national average land management scale in 2016 was approximately 0.43 hectare·person⁻¹. Confronting this reality, the Chinese government has implemented various measures to address these challenges. In the Northeastern region, the number of large-scale agricultural households is approaching 7% of the total number of local agricultural business entities, while in the Central region, the proportion has only reached 1.3%. There is a significant disparity in land management scales between regions, and the development process of appropriately scaled agricultural operations is uneven.

Changes in the scale of land management significantly influence factor inputs and technologies, directly impacting the net carbon footprint of farming. The persistence of small-farm advantages can constrain overall agricultural growth and competitiveness [30], particularly in developing countries. Efforts to consolidate small farms into larger ones have demonstrated that new technologies and institutional arrangements can yield substantial economies of scale [31]. This consolidation effectively reduces the misallocation of agricultural factors and enhances the efficiency of resource utilization. For instance, as farm size increases, the use of chemical inputs such as fertilizers, pesticides, and insecticides per unit area tends to decrease [32], thereby positively affecting the reduction of indirect carbon emissions from farming. Moreover, larger-scale land management tends to increase soil organic carbon (SOC) stocks compared to smaller scales, primarily due to better incorporation of crop residues into the soil and increased application of organic fertilizers [33]. On the adoption of technologies, larger-scale peasants are more inclined to adopt and invest in new agricultural technologies and knowledge [34], whereas smaller-scale peasants may face limitations in resource availability, hindering their adoption rate of Green Revolution technologies at a comparable pace [35]. However, the net impact of these changes on the carbon footprint of farming can vary depending on factors such as the inherent carbon sequestration capacity of new high-yield, stable crops and their effects on the demand for carbon-emitting agricultural inputs. It is speculated that the Green Revolution in farming could potentially lead to a seasonal cyclical effect on CO₂, possibly increasing global NPP by 5–10% [36]. If a significant portion of this increase could be sequestered as biological or soil carbon, it would substantially benefit the net carbon effect of farming. Thus, technological advancements in farming resulting from the expansion of land management scale are poised to have a beneficial impact on the overall net carbon effect of farming.

Spatial correlations between agricultural management practices across different regions can significantly influence agricultural production dynamics. Research indicates that factors of production, including agricultural materials and technologies, tend to diffuse across regions, with efficiency inversely related to the distance between them [37]. Consequently, neighbouring regions often exhibit similar agricultural production characteristics. This proximity facilitates the exchange and adoption of agricultural management practices, potentially fostering a diffusion effect and promoting homogenization in agricultural science and technology development [38]. Expanding the regional land management scale can amplify these spatial impacts by facilitating the movement of production factors, thereby influencing the overall net carbon effect of farming within the region. On the other hand, regarding the net carbon effect of farming, similar natural conditions can exist across administrative boundaries, which may also lead to possible spatial clustering effects.

Therefore, this paper will investigate the spatial effect of expanding land management scale on the net carbon effect of farming, which is of great importance for coordinating the carbon function of the national agricultural industry and achieving the national carbon peak. In summary, this paper obtains the first research hypothesis:

Hypothesis 1.

Land management scale directly influences the net carbon effect of farming and exhibits significant spatial spillover effects.

2.2. Analysis of Threshold Mechanisms

The concept of the environmental Kuznets curve (EKC) illustrates an “inverted U-shaped” relationship between environmental pollution and per capita income [39]. Initially, during early economic development stages, environmental pollution tends to increase alongside per capita income. However, as economic development progresses, there comes a point where environmental pollution starts to decrease despite continued growth in per capita income. Similarly, the net carbon effect of farming, as a measure of ecological impact, may exhibit a “U-shaped” relationship with rural per capita income under conditions aligning with the EKC. This relationship implies that initially, as rural per capita income increases, the net carbon effect of farming might decline. Yet, beyond a certain stage of development, further increases in rural per capita income could lead to a rise in the net carbon effect of farming. (See Figure 1).

The expansion of the land management scale may be a crucial factor in establishing this “U-shaped” curve. The EKC highlights that economic development and rising incomes alone are insufficient to naturally produce this pattern. The inflection point in the curve does not occur spontaneously with economic growth; rather, it emerges through transformations in industrial structure, technological advancements, or regulatory measures concerning the environment [40]. These factors collectively demonstrate the economic heterogeneity across different developmental stages. Considering this perspective, it can be inferred that the scale of land management, which reflects variations in agricultural production methods, is likely to significantly influence the relationship between economic development and the net carbon effect of farming. This hypothesis suggests that at a specific scale of land management, increases in per capita income cease to have a detrimental environmental impact and may even contribute positively.

One possible explanation for this transformation is that as peasants attain a certain scale of land management, they may transition from traditional survival-focused peasants to modern rational peasants. Here, we refrain from delving into the merits or criticisms of the substantive views on peasant behaviour represented by Chayanov and Scott versus the formalist perspectives of Schultz and Popkin. Instead, we argue that such a shift in peasant behaviour logic better aligns with observed realities. In times when market economies were less developed, the behavioural logic of Chinese peasants likely leaned more towards subsistence. Peasant household decision-making revolved around survival rather than profit maximization [41]. Under this framework, peasants prioritized crop quantity over factors like price or quality. Concerns over crop yields often deterred them from reducing pesticide usage or adopting non-chemical pest control methods [42]. Consequently, small-scale peasants tended to adopt unsustainable land management practices to meet subsistence needs. This approach may have boosted rural incomes to some extent but exacerbated the net carbon effect of farming. With increasing marketization, land transfers in rural China have become more frequent, and surplus rural labour increasingly migrates to urban centres, though challenges persist. Concurrently, there has been a trend towards consolidating land-management rights among full-time peasants rather than part-time peasants. Improvements in market price mechanisms have compelled them to consider maximizing economic returns from cultivation by balancing benefits against costs. This shift marks a significant change in peasant behaviour logic. Their economic decisions are not irrational; indeed, they exhibit economic rationality akin to capitalist entrepreneurs [43]. This economic rationality translates into increased farming income, which serves as the capital base for subsequent agricultural production cycles [44]. Higher-income incentivizes peasants to adopt more advanced technologies and sustainable farming practices [45], mirroring strategies employed by savvy capitalists.

The potential impact of education levels on the net carbon effect of farming, and how this impact varies with the scale of land management, is another focal point of this paper. While most studies indicate that higher education levels positively influence ecological improvements and reduce CE [46], Guan et al. (2023) identified a non-linear relationship between rural residents’ education levels and the NCS in farming (for specific theoretical analysis, refer to their study) [10]. Despite differences in the methodological approach to dependent variables between our study and theirs, we include this non-linear relationship in our analysis. Thus, this paper proposes the following research hypothesis:

Hypothesis 2.

The influence of income level and education level on the net carbon effect of farming exhibits a significant threshold effect. Only when the scale of land management reaches a certain threshold does the effect of income level transform from negative to positive. Similarly, only beyond this threshold does the effect of education level transition from positive to negative.

3. Materials and Methods

3.1. Variables

(1)

Carbon-balance ratio (CBR): This paper initially conducts the accounting of net carbon sink (NCS) and carbon-balance ratio (CBR), integrating findings from relevant research in NCS and ESC fields. NCS is derived by subtracting CE from CS, while CBR is computed by dividing CS by CE, which characterises the situation of carbon sinks and carbon sources of farming from different perspectives. A CBR greater than 1 signifies a net carbon sink effect in farming, where CS exceeds CE, benefiting other industries by offsetting part of their emissions. Conversely, a CBR less than 1 indicates a net carbon source effect, where farming fails to absorb its generated carbon, potentially increasing net carbon emissions.

In this study, CS is calculated using a unique approach: matching global NEP data with georaster data of Chinese farmland, ensuring robust data quality. This method addresses limitations in traditional crop yield-based calculations, which often lack detailed carbon sequestration rates and economic coefficients. CE from farming includes three main components: Energy Emissions (EE), Direct Emissions (DE), and Indirect Emissions (IDE), detailed in Figure 2. IDE is computed by multiplying input usage by carbon coefficients for agricultural inputs, as integrated by Hu et al. (2023) [47], referencing Table 1. Finally, CS and CE are converted into CO₂ equivalents using relative molecular mass (One issue to consider is that the NEP dataset that accounts for CS subtracts organic carbon losses from farmland, which overlaps with some of the GHG emissions from farmland soils. To address this issue, this paper assumes that the C elements of CO₂ and CH₄ in the Farmland Soil GHG Emissions dataset are derived solely from the deducted organic carbon losses in the NEP dataset, allowing for the elimination of duplicates through appropriate processing. The method is as follows: One unit of mass of CH₄ released from agricultural soils can be converted into 27 units of CO₂ equivalent according to the GWP. This conversion contains 0.75 units of C, based on the relative molecular mass. Assuming that the 0.75 units of C coincide with the C removed in the accounting of CS, these units can be converted into 2.75 units of CO₂ equivalent based on their relative molecular mass. As a result, 1 unit of mass of CH₄ released from agricultural soils should be converted into 24.25 units of CO₂ equivalent.) and Global Warming Potential (GWP) (refer to Table 2) [48], yielding values for NCS and CBR. NCS values may be negative, requiring careful handling and logarithmic transformation. CBR as a ratio offers a more symmetric distribution and straightforward economic interpretation. So CBR serves as the dependent variable in this paper’s model.

(2)

Land management scale (SCALE): This indicator indicates the average area of cultivated land per agricultural labourer in each region, calculated by dividing the total cultivated land area by the number of labourers in the farming industry [46]. In China, there are discrepancies in sampling statistics concerning cultivated land area and the labour force, stemming from variations in statistical precision and methodology. Typically, China’s periodic sampling of arable land area and labour force data may not align temporally with decadal census data, resulting in data fluctuations. Following the suggestions of Wang et al. (2008) [49] and the revision principles of the National Bureau of Statistics of China (https://www.aof.org.hk/uploads/conference_detail/420/con_paper_0_608_paper-2-_shi_faqi.pdf, accessed on 5 May 2024.), which advocate using census data to rectify historical data, adjustments have been made to stabilise the data while preserving its underlying trends.

(3)

Income level of rural residents (INCOME): This paper uses per capita net income (disposable income) of rural residents as a measure. To ensure comparability and accuracy, the end of the study period (2019) serves as the base period, and the CPI of rural residents is applied to adjust the INCOME.

(4)

Education level of rural residents (EDU): This paper adopts per capita years of education of rural residents as a measure. Statistical data provides only the number and proportion of people with varying literacy levels. Therefore, China’s per capita years of education are calculated using the formula provided by Chen et al. (2004) [50] to weight the treatment.

(5)

Control variables. Sown area (SA): Changes in the total sown area can influence the intensity of CS and CE per unit area, thereby potentially affecting the results of CBR. Therefore, it is essential to account for the effects of these changes in SA.

Unit yield of cereal crops (UCE) and unit yield of cash crops (UCA): This paper includes UCE and UCA as separate control variables in the regression model. The decision to include UCE as an additional control variable, instead of combining both UCE and UCA into a single measure of crop unit yields, was made to specifically evaluate the traditional approach to accounting for CS.

Land-use intensity (LI): Replanting is an essential aspect of agricultural cultivation in China. Replanting indices are used to measure the frequency of replanting and can be divided into potential replanting indices (PI) and actual replanting indices (AI) [51]. The actual replanting index is calculated based on statistical data availability:

$$\mathrm{AI}=\mathrm{SA}/\mathrm{CA}$$

(1)

The equation above utilises the variables SA to represent the sown area and CA to represent the cultivated area. According to Fan et al. (2004) [52], it may be possible to quote the PI for each region directly:

$$\mathrm{LI}={\displaystyle \frac{\mathrm{AI}}{\mathrm{PI}}}={\displaystyle \frac{\mathrm{SA}/\mathrm{CA}}{\mathrm{PI}}}$$

(2)

It has been seen that the higher the LI, the greater the pressure on the land from agricultural production, which can be used as an indicator to control the impact of different levels of afforestation in different regions on the net carbon effect of farming.

Mechanized returning stalk into soil-area ratio (SR): The study emphasizes that conservation tillage practices, such as straw returning to soil and reduced tillage methods like no-till, have a positive impact on soil carbon fixation and reduce (SOC) loss [15]. Calculating SR by dividing the mechanized straw return area by the total cultivated area enhances model control.

Annual precipitation (PRE), annual temperature (TEM) and annual sunshine duration (SUN): the net carbon effect of farming, particularly the carbon sequestration potential of farming, is significantly influenced by various natural factors. The selection of three key natural factors PRE, TEM and SUN in the agricultural production process can effectively control the inter-annual variation of CS.

Table 1. Carbon emission factors of agricultural inputs.

Carbon Emissions Source	Emission Factor	Data Source
Nitrogen fertilizer production, transportation and use	1.53 kg (CO2)·kg−1	[53]
Phosphate fertilizer production, transportation and use	1.63 kg (CO2)·kg−1	[53]
Potassium fertilizer production, transportation and use	0.65 kg (CO2)·kg−1	[53]
Compound fertilizer production, transportation and use	1.77 kg (CO2)·kg−1	[54]
Pesticide production, transportation and use	4.9341 kg (C)·kg−1	[54]
Agricultural plastic film production, transportation and use	5.18 kg (C)·kg−1	[55]

Table 1. Carbon emission factors of agricultural inputs.

Carbon Emissions Source	Emission Factor	Data Source
Nitrogen fertilizer production, transportation and use	1.53 kg (CO2)·kg−1	[53]
Phosphate fertilizer production, transportation and use	1.63 kg (CO2)·kg−1	[53]
Potassium fertilizer production, transportation and use	0.65 kg (CO2)·kg−1	[53]
Compound fertilizer production, transportation and use	1.77 kg (CO2)·kg−1	[54]
Pesticide production, transportation and use	4.9341 kg (C)·kg−1	[54]
Agricultural plastic film production, transportation and use	5.18 kg (C)·kg−1	[55]

Table 2. GWP100 values.

Gas	AR6-GWP100	Lifetime
CO2	1	N/A (Not Applicable)
CH4	27	11.8
N2O	273	109

Table 2. GWP100 values.

Gas	AR6-GWP100	Lifetime
CO2	1	N/A (Not Applicable)
CH4	27	11.8
N2O	273	109

3.2. Data Sources

The study covers a sample period from 2004 to 2019, including 30 provincial administrative regions in China. Tibet, Hong Kong, Macao, and Taiwan were excluded from the sample area due to a large amount of missing data. The CS data was obtained from research conducted by [56] at the National Ecological Data Centre. The data on DE were provided by EDGAR [57]. The integration of geographic information data was carried out using ArcMap 10.8. The data used for EE was partially sourced from CEADs [58]. PRE, TEM, and SUN were obtained from the China National Meteorological Science Data Sharing Service Platform. All other data were sourced from the China Rural Statistical Yearbook, China Statistical Yearbook, China Agricultural Statistical Yearbook, and China Population and Employment Statistical Yearbook. Before conducting the regression analysis, the non-proportional variables (SCALE, INCOME, EDU, SA, UCE, UCA, PRE, TEM, SUN) were transformed using natural logarithms. Variables with extreme values were adjusted using 1% quantile shrinkage on one side of the extreme values. All variables underwent a unit root test, and LI and SR were treated differently as they did not pass the test. The subsequent statistical regression process was performed using Stata17. A descriptive statistical analysis of the variables is presented in

Table 3. Descriptive statistics.

Variables	Unit	Observations	Mean	Standard Deviation	Min	Max
CBR		480	1.727	1.543	−2.395	7.932
SCALE	hectare·person−1	480	0.701	0.578	0.192	3.524
INCOME	yuan	480	9975	5450	2638	33,195
EDU	year	480	8.421	0.877	5.368	11.50
SA	1000 hectares	480	5371	3643	88.60	14,783
UCE	kg·hectare−1	480	5130	1016	2870	8169
UCA	kg·hectare−1	480	7702	5899	1159	33,608
LI		480	0.646	0.144	0.343	0.982
SR		480	0.191	0.164	0	0.680
TEM	°C	480	13.93	5.328	2.549	25.43
SUN	hour	480	2047	485.9	933.0	2960
PRE	mm	480	953.1	447.3	200.8	2232

.

3.3. Modelling

This paper explores the relationship between SCALE and CBR, and the net carbon effect of farming can also be influenced by various economic, social, and natural factors. As a result, we have established the following baseline econometric model:

$${\mathrm{CBR}}_{\mathrm{it}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{lnSCALE}}_{\mathrm{it}}+{\mathsf{\beta }}_2{\mathrm{lnX}}_{\mathrm{it}}+{\mathsf{\beta }}_3{\mathrm{lnCONTROL}}_{\mathrm{it}}+{\mathsf{\epsilon }}_{\mathrm{it}}$$

(3)

$${\mathrm{CBR}}_{\mathrm{it}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{lnSCALE}}_{\mathrm{it}}+{\mathsf{\beta }}_2{\mathrm{lnX}}_{\mathrm{it}}+{\mathsf{\beta }}_3\ln {\mathrm{CONTROL}}_{\mathrm{it}}+{\mathsf{\mu }}_{\mathrm{i}}+{\mathsf{\lambda }}_{\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{it}}$$

(4)

Equations (3) and (4) introduce the variables used in the analysis. The explained variable is $\mathrm{CBR}$, while the core explanatory variable is $\mathrm{lnSCALE}$. $\mathrm{X}$ represents the variables INCOME and EDU, and $\mathrm{CONTROL}$ is the control variable representing the variables SA, UCE, UCA, LI, SR, PRE, TEM, and SUN. Additionally, $\mathsf{\epsilon }$ is the random perturbation term, and $\mathsf{\mu }$ and $\mathsf{\lambda }$ are the individual fixed effects and time fixed effects, respectively.

In order to estimate the potential spatial effects of SCALE expansion on emission reductions and sink enhancement, it may be appropriate to use a spatial panel model. The initial step involves constructing the spatial adjacency matrix:

$${\mathrm{W}}_{\mathrm{ij}}=\left\{\begin{array}{c}0,\mathrm{when}\mathrm{region}\mathrm{i}\mathrm{and}\mathrm{j}\mathrm{are}\mathrm{adjacent}.\\ 1,\mathrm{when}\mathrm{region}\mathrm{i}\mathrm{and}\mathrm{j}\mathrm{are}\mathrm{not}\mathrm{adjacent}.\end{array}\right.$$

(5)

Upon successful passing of at least one LM test, the following three classical spatial panel models are estimated separately [59]:

$${\mathrm{CBR}}_{\mathrm{it}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{lnSCALE}}_{\mathrm{it}}+{\mathsf{\beta }}_2{\mathrm{lnX}}_{\mathrm{it}}+{\mathsf{\beta }}_3{\mathrm{lnCONTROL}}_{\mathrm{it}}+\mathsf{\rho }\sum _{\mathrm{j}=1}^{\mathrm{n}}{\mathrm{W}}_{\mathrm{ij}}{\mathrm{CBR}}_{\mathrm{jt}}+{\mathsf{\mu }}_{\mathrm{i}}+{\mathsf{\lambda }}_{\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{it}}$$

(6)

$${\mathrm{CBR}}_{\mathrm{it}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{lnSCALE}}_{\mathrm{it}}+{\mathsf{\beta }}_2{\mathrm{lnX}}_{\mathrm{it}}+{\mathsf{\beta }}_3\ln {\mathrm{CONTROL}}_{\mathrm{it}}+\mathsf{\sigma }\sum _{\mathrm{j}=1}^{\mathrm{n}}{\mathrm{W}}_{\mathrm{ij}}{\mathsf{\epsilon }}_{\mathrm{jt}}+{\mathsf{\mu }}_{\mathrm{i}}+{\mathsf{\lambda }}_{\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{it}}$$

(7)

$$\begin{array}{c}{\mathrm{CBR}}_{\mathrm{it}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{lnSCALE}}_{\mathrm{it}}+{\mathsf{\beta }}_2{\mathrm{lnX}}_{\mathrm{it}}+{\mathsf{\beta }}_3\mathrm{l}\mathrm{n}{\mathrm{CONTROL}}_{\mathrm{it}}+\mathsf{\rho }{\displaystyle \sum _{\mathrm{j}=1}^{\mathrm{n}}}{\mathrm{W}}_{\mathrm{ij}}{\mathrm{CBR}}_{\mathrm{jt}}\\ +{\mathsf{\theta }}_1{\displaystyle \sum _{\mathrm{j}=1}^{\mathrm{n}}}{\mathrm{W}}_{\mathrm{ij}}{\mathrm{lnSCALE}}_{\mathrm{jt}}+{\mathsf{\theta }}_2{\displaystyle \sum _{\mathrm{j}=1}^{\mathrm{n}}}{\mathrm{W}}_{\mathrm{ij}}{\mathrm{lnX}}_{\mathrm{jt}}+{\mathsf{\theta }}_3{\displaystyle \sum _{\mathrm{j}=1}^{\mathrm{n}}}{\mathrm{W}}_{\mathrm{ij}}{\mathrm{lnCONTROL}}_{\mathrm{jt}}+{\mathsf{\mu }}_{\mathrm{i}}+{\mathsf{\lambda }}_{\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{it}}\end{array}$$

(8)

Equations (6)–(8) are based on the Spatial Autoregressive Model (SAR), the Spatial Error Model (SEM), and the Spatial Durbin Model (SDM), respectively. In Equations (6)–(8), ${\mathrm{W}}_{\mathrm{ij}}{\mathrm{CBR}}_{\mathrm{jt}}$ represents the spatial lagged variable, while $\mathsf{\rho }$ is the spatial autoregressive coefficient. The spatial lagged error term is denoted by ${\mathrm{W}}_{\mathrm{ij}}{\mathsf{\epsilon }}_{\mathrm{jt}}$, and $\mathsf{\sigma }$ is the spatial autocorrelation coefficient of the errors. Additionally, ${\mathrm{W}}_{\mathrm{ij}}{\mathrm{lnSCALE}}_{\mathrm{jt}}$, ${\mathrm{W}}_{\mathrm{ij}}{\mathrm{lnX}}_{\mathrm{jt}}$, and ${\mathrm{W}}_{\mathrm{ij}}{\mathrm{lnCONTROL}}_{\mathrm{jt}}$ denote spatial lagged variables of explanatory or control variables, and $\mathsf{\theta }$ is the estimated coefficient of their spatial lagged terms.

In contrast, the SDM model not only measures the spatial correlation effect of the explained variable but also portrays the spatial effect of the explanatory variables, which can provide greater explanatory power. With the support of the results of such aids as statistical tests, this paper will focus on analysing the estimation results of the SDM model.

Classical spatial econometric models assume the existence of only one stable equilibrium in space, which may not account for other potential modes of action. This paper seeks to construct a threshold panel regression model to explore the potential non-linear relationship between some important economic variables and the net carbon effect of farming, taking into account the condition of SCALE change. The threshold model is constructed as follows [60]:

$${\mathrm{CBR}}_{\mathrm{it}}=\left\{\begin{array}{c}{\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{lnSCALE}}_{\mathrm{it}}+{\mathsf{\beta }}_{21}{\mathrm{lnX}}_{\mathrm{it}}+{\mathsf{\beta }}_3{\mathrm{lnCONTROL}}_{\mathrm{it}}+{\mathsf{\epsilon }}_{\mathrm{it}},{\mathrm{lnSCALE}}_{\mathrm{it}}\le {\mathsf{\delta }}_1\\ {\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{lnSCALE}}_{\mathrm{it}}+{\mathsf{\beta }}_{22}{\mathrm{lnX}}_{\mathrm{it}}+{\mathsf{\beta }}_3{\mathrm{lnCONTROL}}_{\mathrm{it}}+{\mathsf{\epsilon }}_{\mathrm{it}},{\mathsf{\delta }}_1<{\mathrm{lnSCALE}}_{\mathrm{it}}\le {\mathsf{\delta }}_2\\ \cdots \\ {\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{lnSCALE}}_{\mathrm{it}}+{\mathsf{\beta }}_{2(\mathrm{n}-1)}{\mathrm{lnX}}_{\mathrm{it}}+{\mathsf{\beta }}_3{\mathrm{lnCONTROL}}_{\mathrm{it}}+{\mathsf{\epsilon }}_{\mathrm{it}},{\mathsf{\delta }}_{\mathrm{n}-1}<{\mathrm{lnSCALE}}_{\mathrm{it}}\le {\mathsf{\delta }}_{\mathrm{n}}\\ {\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{lnSCALE}}_{\mathrm{it}}+{\mathsf{\beta }}_{2\mathrm{n}}{\mathrm{lnX}}_{\mathrm{it}}+{\mathsf{\beta }}_3{\mathrm{lnCONTROL}}_{\mathrm{it}}+{\mathsf{\epsilon }}_{\mathrm{it}},{\mathrm{lnSCALE}}_{\mathrm{it}}\ge {\mathsf{\delta }}_{\mathrm{n}}\end{array}\right.$$

(9)

where $\mathsf{\delta }$ is the threshold value to be estimated. To ensure the reliability of the results, the above econometric models were estimated using clustered robust standard errors at the provincial level.

4. Results

4.1. CBR Measure

Figure 3 illustrates the discrepancy in Total CS across 30 provincial districts, comparing values derived from the NEP dataset with those calculated using traditional crop yields. The latter consistently shows significantly higher values, with its mean from 2004 to 2019 approximately 30.7% greater than that of the NEP dataset over the same period. This discrepancy can be attributed to traditional methods neglecting soil heterotrophic respiration, thus leading to an overestimation. Moreover, the Total CS trend derived from crop yields appears overly smooth, failing to accurately capture actual fluctuations in farming CS. In contrast, the NEP dataset-based Total CS shows greater sensitivity in reflecting these changes. This suggests that China’s agricultural carbon sinks have demonstrated a fluctuating upward trend from 2004 to 2019, influenced largely by temperature and precipitation variations during each respective period [61].

Figure 4 illustrates the trends of CS, CE, NCS and CBR for the 30 provinces in total from 2004 to 2019. During the sample period, NCS and CBR show three very distinct peaks in 2008, 2012, and 2015, with both reaching their maximum in 2015. In general, the trends of NCS and CBR are primarily influenced by the fluctuations in CS, with relatively stable changes in CE. The large inter-annual variations in CS result in a fluctuating upward trend in NCS and CBR. In 2015, CE reached a maximum value of 1075.088 million tonnes, after which it began to decline slowly. Among these, DE accounted for the largest proportion of CE, and the main reason for the decreasing trend of CE was also the gradual decrease of DE since 2013. The net carbon effect of China’s farming industry has become increasingly significant over time, suggesting that it has the potential to absorb a greater quantity of CE generated by other industries.

4.2. Spatial Empirical Results

The Moran’s I test was conducted to determine the spatial correlation of $\mathit{CBR}$ and $\mathit{lnSCALE}$ from 2004 to 2019. The results indicate that the explained variable $\mathit{CBR}$ was found to be significant at a 5% level in all years except for 2009 which did not pass the test. Additionally, the core explanatory variable, $\mathit{CBR}$, passed the significance test of 1% in all years. These findings suggest that both $\mathit{CBR}$ and $\mathit{lnSCALE}$ have a significant spatial correlation. Upon observing Figure 5, it becomes apparent that $\mathit{CBR}$ displays a distinct pattern of “high–high” and “low–low” aggregation. The regions with significant net carbon effects shifted from being dominated by the Central and Southwest regions at the beginning of the study period to being dominated by the Northeast and Southwest regions at the end of the period. As a result, it would be advisable to employ a spatial econometric model to examine the impact of land management scale expansion on emission reduction and sink enhancement.

4.2.1. Spatial Spillover Effects

The results of OLS estimation for Equation (3) and FE estimation for Equation (4) are illustrated in

Table 4. Spatial econometric results.

Variables	(1) OLS	(2) FE	(3) SAR	(4) SEM	(5) SDM
	CBR	CBR	CBR	CBR	CBR
lnSCALE	0.773 **	1.495 **	1.447 **	1.213 *	1.345 **
	(2.23)	(2.22)	(2.06)	(1.76)	(2.05)
lnINCOME	0.436	−2.370 *	−1.115 ***	−0.873 **	−3.108 **
	(1.18)	(−1.76)	(−3.71)	(−2.46)	(−2.39)
lnEDU	−2.002	4.581	4.140 *	4.820 *	4.996 **
	(−1.08)	(1.62)	(1.92)	(1.94)	(2.06)
lnSA	0.399 ***	2.462 **	1.423 *	1.248	1.308
	(4.43)	(2.62)	(1.70)	(1.38)	(1.57)
lnUCE	−3.118 ***	5.075 ***	4.575 ***	4.358 ***	4.051 ***
	(−3.03)	(3.81)	(3.58)	(2.85)	(3.29)
lnUCA	0.596 **	−0.734	−0.438	−0.281	−0.249
	(2.53)	(−1.61)	(−1.14)	(−0.68)	(−0.70)
LI	−3.426 **	−3.924 **	−3.366 ***	−3.748 ***	−2.785 *
	(−2.32)	(−2.29)	(−2.83)	(−2.72)	(−1.69)
SR	1.834	1.456	0.290	0.441	1.129
	(1.20)	(1.29)	(0.35)	(0.49)	(1.52)
lnPRE	−0.232	2.287 ***	1.277 ***	1.980 ***	2.279 ***
	(−0.67)	(4.26)	(3.90)	(4.10)	(3.53)
lnTEM	−0.599	−2.572 ***	−1.545 **	−1.801 **	−2.333 ***
	(−1.18)	(−2.82)	(−2.37)	(−2.27)	(−2.99)
lnSUN	−2.628 **	−1.228	−1.355 ***	−2.427 **	−2.036 *
	(−2.28)	(−1.44)	(−2.58)	(−2.46)	(−1.80)
Constant	43.618 ***	−42.929 **
	(2.85)	(−2.15)
ρ			0.469 ***		0.467 ***
			(8.03)		(8.56)
σ				0.540 ***
				(9.83)
Observations	480	480	480	480	480
R2	0.433	0.470	0.406	0.389	0.497
ID FE		YES	YES	YES	YES
YEAR FE		YES

Notes: Robust t-statistics/z-statistics in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

. Following the application of Hausman’s test, it was recommended that a fixed effect model be employed. The outcomes of the LM test, LR test and Wald test indicated that the SDM model should be utilised, and the LR test indicated that the individual fixed effect should be employed. Consequently, it was deemed appropriate to select the individual fixed SDM model as the final estimation of the spatial model.

Table 4 also shows the final estimates of the corresponding coefficients for each spatial model (Includes SAR, SEM and SDM.). Upon analysing the estimation results of the SDM model, it was found that the estimated value of the spatial autoregressive coefficient $\mathsf{\rho }$ is 0.467 and is significant at the 1% level, which suggests that the use of the spatial model in this paper is appropriate. There is a significant positive spatial correlation effect of $\mathit{CBR}$, indicating that the carbon balance status of China’s farming industry shows obvious spatial clustering characteristics, meaning the carbon balance status of the region is closely related to that of the neighbouring regions, and the neighbouring regions may have a positive impact on the region.

The regression coefficients suggest that most explanatory and control variables are significant, except of $\mathrm{lnSA}$, $\mathrm{lnUCA}$, and $\mathrm{SR}$. As for the core explanatory variable, $\mathrm{lnSACLE}$, its directions of coefficients align with the theoretical expectations presented in the previous section. Among the control variables, the coefficient of $\mathrm{lnUCE}$ is significantly positive at the 1% level, while the coefficient of $\mathrm{lnUCA}$ is insignificantly negative. This suggests that the yield of cash crops may not have a statistically significant effect on the carbon balance of farming. Even if it does, it may be negative. Additionally, the coefficient of $\mathrm{LI}$ is significantly negative. This indicates that while an increase in food production can be beneficial to carbon fixation and accumulation, the negative effects of increasing land use intensity may offset a part of the positive effects of increasing production. It has been observed that the coefficient of $\mathrm{SR}$ is positive, which is in line with previous studies that have suggested that returning straw to the field may have a positive impact on the NCS of farming. Additionally, it is worth noting that all three natural factors’ control variables are significant, which means that $\mathrm{CBR}$ is not solely an economic variable.

The spatial lag variable of the explanatory variables is an indicator that can effectively analyse whether there is a spatial relationship between the explained variable and the explanatory variables. According to

Table 5. The direct and indirect effect decomposition of SDM.

Variables	Wx	Direct Effect	Indirect Effect	Total Effect
lnSCALE	1.676 *	1.693 **	4.058 **	5.750 ***
	(1.76)	(2.35)	(2.44)	(2.77)
lnINCOME	2.156	−3.062 ***	1.253	−1.810 ***
	(1.50)	(−2.65)	(0.90)	(−2.92)
lnEDU	−4.924	4.918 **	−4.562	0.356
	(−1.47)	(2.20)	(−0.94)	(0.07)
lnSA	1.213	1.575 *	3.139 **	4.714 ***
	(1.18)	(1.95)	(2.05)	(2.58)
lnUCE	0.057	4.317 ***	3.332	7.649 ***
	(0.03)	(3.85)	(1.43)	(3.21)
lnUCA	−0.791	−0.368	−1.587	−1.955
	(−0.90)	(−1.15)	(−1.24)	(−1.54)
LI	0.634	−2.947 *	−1.185	−4.131
	(0.32)	(−1.86)	(−0.43)	(−1.43)
SR	1.911	1.426 *	4.171	5.598 *
	(1.29)	(1.85)	(1.63)	(1.91)
lnPRE	−1.811 **	2.228 ***	−1.363 *	0.865 *
	(−2.48)	(3.87)	(−1.70)	(1.79)
lnTEM	0.859	−2.375 ***	−0.477	−2.852 *
	(0.77)	(−3.18)	(−0.31)	(−1.96)
lnSUN	2.049	−1.901 *	1.917	0.016
	(1.42)	(−1.92)	(1.18)	(0.02)

Notes: Robust z-statistics in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

, it can be observed that the spatial coefficient of $\mathrm{lnSCALE}$ is significantly positive, which implies that expanding the land management scale could have a positive spatial effect on the carbon balance of farming in neighbouring areas. Thus, further analysis based on the effect’s decomposition is necessary. Table 5 illustrates that $\mathrm{lnSCALE}$ has a noteworthy spatial spillover effect and total effect on $\mathrm{CBR}$, which confirms Hypothesis 1 (H1). The coefficients for the spatial spillover effect and the total effect are 4.058 and 5.750 respectively. Notably, the coefficient for the indirect effect (4.058) is larger than the coefficient for the direct effect (1.693), implying that the spatial spillover intensity of $\mathrm{lnSCALE}$ on $\mathrm{CBR}$ has an increasing trend.

To ensure the robustness of the spatial measurements, the same regression of Equation (8) was performed after replacing the spatial adjacency matrix with a spatial distance matrix. Comparison of the results indicates that, with the exception of a slight modification in the size of the impact coefficients, the direction and significance of the impacts remain consistent with Model (5) in Table 4. This highlights the robustness of the empirical findings.

4.2.2. Analysis of Spatial Heterogeneity

Further sub-regional analysis is discussed in this paper, given the wide variation across China’s geographical sub-regions. The 30 provinces have been categorized into four major economic regions, namely Northeast, East, Central and West, as developed by the National Bureau of Statistics of China. To represent land management scale in different regions, three interaction term variables have been generated:

$${\mathrm{nlnSCALE}}_{\mathrm{it}}=\left({\mathrm{lnSCALE}}_{\mathrm{it}}-\overline{{\mathrm{lnSCALE}}_{\mathrm{it}}}\right)\times \mathrm{northeast}$$

(10)

$${\mathrm{elnSCALE}}_{\mathrm{it}}=\left({\mathrm{lnSCALE}}_{\mathrm{it}}-\overline{{\mathrm{lnSCALE}}_{\mathrm{it}}}\right)\times \mathrm{east}$$

(11)

$${\mathrm{wlnSCALE}}_{\mathrm{it}}=\left({\mathrm{lnSCALE}}_{\mathrm{it}}-\overline{{\mathrm{lnSCALE}}_{\mathrm{it}}}\right)\times \mathrm{west}$$

(12)

In the above equation, $\mathrm{northeast}$, $\mathrm{east}$ and $\mathrm{west}$ are dummy variables representing the corresponding regions, taking the value 1 if the province in which the observation is located is in the corresponding region and 0 otherwise. $\overline{\mathrm{lnSCALE}}$ is the mean of $\mathrm{lnSCALE}$, for which centring avoids multicollinearity between variables.

The results presented in

Table 6. Regional heterogeneity analysis.

Variables	(1) SDM	(2) SDM_GA
	CBR	CBR
lnSCALE	1.345 **	1.409
	(2.05)	(1.52)
nlnSCALE		10.833 ***
		(4.35)
elnSCALE		−1.938 *
		(−1.90)
wlnSCALE		1.822
		(1.05)
ρ	0.467 ***	0.444 ***
	(8.56)	(8.22)
Observations	480	480
R2	0.497	0.560
ID FE	YES	YES

Notes: z-statistics in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

were derived by performing a regression analysis on the three interaction variables in Equation (8), with the coefficient of $\mathrm{nlnSCALE}$ being 10.833. At a 1% significance level, the result indicates that expanding the land management scale in the Northeast region is more effective in reducing CE and increasing CS from the farming industry compared to the non-Northeast region. Moreover, the coefficient of $\mathrm{elnSCALE}$ is significantly negative, indicating that expanding the land management scale in the East region has a limited effect on the carbon balance of the farming industry. In contrast, the insignificant result of $\mathrm{wlnSCALE}$ suggests that the Western region may have a little spatial heterogeneity compared to other regions.

There appears to be a significant spatial heterogeneity between the Northeast and non-Northeast regions, as indicated by the coefficients. Furthermore, as shown in Figure 6, the spatial heterogeneity between Northeast and non-Northeast regions is reflected in the fact that the provinces with an average annual variation of CBR > 0.2 are Heilongjiang (0.51), Jilin (0.26) in the Northeast, and Inner Mongolia (0.31) (Inner Mongolia is economically divided into Western regions, but the eastern part of the province is usually considered part of the northeastern region, both naturally and socially). Their mean SCALE in the sample period is greater than 1.4 hectare·person⁻¹, which is much larger than the mean SCALE in China. This difference can be attributed to the distinct agricultural production practices in the Northeast and non-Northeast regions, which are influenced by a multitude of interrelated factors.

From the perspective of natural science, The Northeast Plain of China boasts a vast black soil belt, compared to the plains of Eastern Europe and the Great Plains of the United States. By expanding the scale of land management and promoting the scientific rationalisation of farming systems, the carbon sequestration capacity of this fertile land can be greatly stimulated [62]. The Northeast Blackland Protected Tillage Action Plan (2020–2025), implemented by the Chinese government, stipulates that each county-level high-standard protected tillage application base must have a concentrated area of no less than 1000 mou (approximately 66.7 ha). In principle, the minimum area required at the county level is 1000 mou (approximately 66.7 ha), while at the township level, a minimum of 200 mou (approximately 13.3 ha) is required. These figures demonstrate the importance of large-scale land management for protected tillage, as they allow for the full implementation of the techniques’ potential to enhance soil fertility and CS capacity. Although the mean of SCALE in Ningxia and Xinjiang, located in the Northwest region, is also above 1 hectare·person⁻¹, the average annual variation of CBR did not increase significantly during the sample period due to agricultural constraints imposed by natural endowments.

The Northeast region stands out in agricultural economic development compared to other regions. It boasts a highly mechanised agricultural production system that prioritises technological advancements, a departure from the smallholder economy prevalent in China’s traditional core regions [63]. With its sparse population, favourable climatic conditions, and less fragmented land, the Northeast favours large-scale mechanised farming. By expanding land management scales and reducing fragmentation, the cost per unit area of adopting conservation tillage techniques decreases [64]. This in turn makes peasants more aware of the use of agricultural technologies to support wider adoption of sustainable agricultural practices [65], encourages them to adopt this type of farming and improves SOC fixation. This approach in the Northeast aligns strategically with modern agricultural trends aimed at boosting productivity and ensuring sustainability through scaled-up land management.

4.3. Threshold Characteristics

To explore the threshold characteristics, the existence of the threshold effect was first tested. The F-values corresponding to the single-threshold model rejected the original hypothesis at the 5% level, but the double-threshold model failed to reject the original hypothesis. According to the results of the significance level test, it can be concluded that there is a single threshold effect on the impact of rural residents’ income level and education level on the carbon balance of farming in the context of changes in land management scale.

The estimation results of model (1) in

Table 7. The estimation results of panel threshold regression model.

Variables	(1)	(2)	(3)	(4)
	CBR	CE	UCS	lnSA
lnSCALE	1.780 ***	−1.136	3.536 ***	0.101 **
	(4.38)	(−1.20)	(5.21)	(2.41)
lnINCOME (lnSCALE ≤ δ1)	−1.116 ***	4.733 ***	−1.171 ***	−0.017
	(−5.03)	(6.59)	(−3.32)	(−0.75)
lnINCOME (δ1 < lnSCALE ≤ δ2)	1.259 **	2.414 ***	−1.513 ***	0.334 ***
	(2.36)	(4.84)	(−3.19)	(8.91)
lnEDU (lnSCALE ≤ δ1)	5.665 ***	−13.309 ***	5.953 **	0.056
	(3.64)	(−3.29)	(2.42)	(0.34)
lnEDU (δ1 < lnSCALE ≤ δ2)	−3.927 *	−2.853	6.676 **	−1.526 ***
	(−1.66)	(−0.85)	(2.51)	(−8.09)
lnSA	1.534 ***	7.821 ***	0.089
	(3.56)	(8.72)	(0.13)
lnUCE	4.632 ***	7.658 ***	8.562 ***	−0.360 ***
	(6.75)	(5.60)	(8.53)	(−5.15)
lnUCA	−0.647 ***	0.007	−0.877 ***	0.066 ***
	(−2.95)	(0.02)	(−2.61)	(2.91)
LI	−3.775 **	0.664	−2.722	1.323 ***
	(−2.54)	(0.21)	(−1.18)	(9.31)
SR	1.210	1.109	3.952 **	0.139
	(1.15)	(0.49)	(2.42)	(1.28)
lnPRE	1.550 ***	−0.754	2.580 ***	0.024
	(4.47)	(−1.00)	(4.79)	(0.67)
lnTEM	−3.534 ***	−2.233	−4.363 ***	0.037
	(−5.12)	(−1.49)	(−4.05)	(0.52)
lnSUN	−1.742 ***	−4.378 ***	−3.654 ***	0.060
	(−2.60)	(−3.01)	(−3.51)	(0.87)
Constant	−33.918 ***	−68.352 ***	−41.320 ***	10.080 ***
	(−3.69)	(−3.55)	(−2.89)	(11.99)
Observations	480	480	480	480
R2	0.469	0.599	0.455	0.369
Threshold evaluation	0.594 **	−1.033	−0.564 ***	0.007 **
	(37.90)	(41.59)	(42.64)	(121.24)

Notes: t-statistics/F-statistics in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

indicate that when $\mathrm{lnSCALE}$ is below the threshold value of 0.594, the coefficient estimate of $\mathrm{lnINCOME}$ is −1.116, which implies that the income level of rural residents has a negative effect on the carbon balance of farming at this stage; when $\mathrm{lnSCALE}$ is greater than 0.594, the coefficient estimate of $\mathrm{lnINCOME}$ is 1.259, indicating that with the expansion of the scale of land management, the effect of income level on the carbon balance turns from negative to positive. The regression results prove that the “U-shaped” curve of income and carbon balance is valid. It can be postulated that after a certain size of area per labourer, the agricultural production and management mode of peasants may begin to change from the traditional factor-driven production mode to the environmentally friendly green production mode.

The findings of Guan et al. (2023) [10] are supported by the threshold model outlined in this paper. It is observed that as land management scale increases, the effect of $\mathrm{lnEDU}$ on $\mathrm{CBR}$ shifts from positive to negative once the $\mathrm{lnSCALE}$ surpasses the threshold, resulting in an “inverted U-shaped” pattern of the association. To further explore the factors contributing to the formation of the “inverted U-shaped” curve, we conducted a thorough analysis of the CBR by replacing the explained variables with CE and CS, and decomposed CS into carbon sinks per unit of sown area (UCS) and SA, applying a regression of the threshold panel model. The findings of this analysis are presented in Table 7, Model (2) (3) (4). According to the estimation results, it appears that a considerable proportion of the threshold characteristics of $\mathrm{lnEDU}$ on $\mathrm{CBR}$ are attributed to $\mathrm{lnSA}$. It seems that the impact of $\mathrm{lnEDU}$ on $\mathrm{lnSA}$ is not significant until $\mathrm{lnSCALE}$ reaches the threshold; after the land management scale reaches a certain level, it has been observed that the effect of $\mathrm{lnEDU}$ on $\mathrm{lnSA}$ becomes significantly negative, which suggests that an increase in the average level of education of rural residents by 1% will lead to a reduction of 1.526% in the total sown area of crops, an inverse relationship consistent with the findings of Zheng (2024) [66]. As a result, Hypothesis 2 is confirmed.

In summary, before the threshold of the land management scale is reached, income growth has a negative effect on CBR, while education improvement has a positive effect on CBR; after the threshold of the land management scale is reached, the effect of income becomes positive, while the effect of education becomes negative.

In practical terms, the threshold value of approximately 1.81 hectare·person⁻¹ signifies that each agricultural labourer manages about 27 mou of arable land on average. This threshold is economically significant, particularly for farm households with two to three labourers, whose operations typically range between 54 and 81 mou, aligning with China’s encouraged scale of 50 to 100 mou. To optimise farming’s net carbon effect, increasing incomes may be more crucial for peasants in appropriately scaled operations, while enhancing education levels could benefit average smallholders more. However, with 210 million farm households operating less than 10 mou of land, most provinces are likely to remain below the threshold for the foreseeable future. Data from 2004 to 2019 indicate an average annual growth rate of 7.47‰ for $\mathrm{lnINCOME}$ and 3.02‰ for $\mathrm{lnEDU}$ across China’s provinces. Based on estimates, the negative effect of INCOME at this stage can be fully offset by the positive effect of EDU. Overall, the development of China’s rural economic and social levels is unlikely to worsen the net carbon effect of farming in the short term.

It is worth noting that only 6.25% of the sample exceeds this threshold, and these observations belong to only 1/10th of the total number of regions, primarily concentrated in the provinces of Heilongjiang, Jilin, and Inner Mongolia. In northeast China, land management in farming has surpassed the threshold value. The factors driving this expansion exhibit similarities with other regions, as well as distinct reasons unique to the northeast. Since the 1990s, the population of the Northeast region has been declining due to various factors, including economic system rigidity and resource depletion. From 2010 to 2020, the population decreased by 10%, equivalent to the population of Belgium. In response, rural residents in the northeast, particularly those with higher human capital, are increasingly migrating out of the region. Higher education levels among rural populations correlate with a greater willingness to migrate to other sectors or regions, potentially negatively impacting the carbon balance of farming in the short term. Local peasants remaining in farming may consequently increase their use of high-carbon agricultural materials and adopt less efficient production methods as the agricultural labour force diminishes rapidly in the short term [67].

5. Discussion

This paper concludes, based on data from China, that expanding the land management scale can enhance the net carbon effect of farming, yet this conclusion warrants further discussion. One crucial question is whether such findings apply universally worldwide. While there is consensus that adopting conservation agriculture (CA) generally benefits the net carbon effect of farming, opinions diverge on how scaling up land management impacts CA adoption, potentially blurring its effect on the net carbon outcome. For instance, a study by Ntshangase et al. (2018) in South Africa found that larger cultivated farm sizes significantly hindered the adoption of no-till CA, reducing adoption likelihood by approximately 33.8% per hectare increase [68]. Conversely, research by Nazu et al. (2022) among wheat peasants in Bangladesh suggests that larger farms are more likely to adopt conservation tillage techniques [69]. These contrasting findings reflect varied socio-economic conditions across studies.

In similar smallholder-dominated research contexts, differing conclusions across studies likely stem from variations in socio-economic conditions. According to this empirical study from South Africa, management practices, particularly in herbicide-free environments, may be unsustainable on larger farms, whereas smallholders can effectively leverage labour-intensive methods under government subsidies to implement CA. In contrast, China benefits from more advanced Green Revolution technologies, though subsidies for sustainable farming practices remain modest relative to peasant incomes. This disparity influences peasant behaviour based on cost-benefit analyses. Such socio-economic distinctions likely contribute to varying net farming benefits as land management scales up, influencing agricultural practices with potentially divergent impacts on the net carbon effect. Peasants in regions with robust agricultural policies and economic support systems face fewer barriers to adopting sustainable practices, potentially enhancing the net carbon effect as they utilize modern technologies more efficiently. Conversely, in low-income countries dominated by smallholders, scaling up land management may not yield equivalent environmental benefits due to inadequate infrastructure and technical support. Thus, the environmental outcomes of farm size expansion vary by region, underscoring the need for tailored policy approaches aligned with local conditions. As Knowler et al. (2007) emphasize, promoting CA requires customized strategies for each region [70], reflecting the nuanced impact of regional farming upgrades on the net carbon effect.

6. Conclusions

This study employs a novel accounting method to assess the net carbon effect of the farming industry, expanding on previous academic research. Analysing data from 30 provinces in China spanning 2004 to 2019, the study utilizes spatial panel regression models and threshold panel regression models. The primary objective is to uncover spatial correlations between land management scale and carbon-balance ratios, alongside exploring non-linear relationships between carbon-balance ratios and the income and education levels of rural residents:

(1)

During the study period, it was found that China has generally demonstrated a net sink effect of the farming industry. The national carbon-balance ratio of the farming industry showed fluctuating growth.

(2)

The results of the spatial panel model show that the scale of land management has a positive spatial spillover effect on the carbon balance. This effect varies spatially, with a particularly significant effect in the northeast.

(3)

Based on the analysis of the threshold panel model, it can be inferred that in most regions, an increase in the level of income leads to a decrease in the net sink effect of farming, while an increase in the level of education enhances this effect.

This study proposes several policy recommendations based on its research findings. Firstly, there is a suggestion to expedite the development of a unified carbon accounting system and establish a scientifically and reasonably defined scope and methodology for carbon accounting. Secondly, it is recommended to continue promoting appropriately scaled agricultural operations and to incentivize and support large-scale agricultural operations. Thirdly, there is a proposal to deploy agricultural production factors in a rational manner and gradually eliminate institutional mechanisms that hinder land transfer. Additionally, besides improving peasants’ income, it is emphasized to recognize the intangible value of human capital and encourage the development of new types of agribusinesses. Lastly, the study suggests establishing pilot carbon functional regions where areas with high carbon sinks receive material and economic compensation. This approach aims to maximize carbon and ecological advantages while ensuring national food security and ecological safety.