Digitalization’s Role in Shaping Sustainable Agriculture—Evidence from Chinese Provincial Panel Data Using the Baidu Index

Abstract

The impact of digital transformation on agricultural sustainability has attracted significant attention, and empirical methods are widely being used to provide a scientific framework for research in this field. However, commonly used digitalization indicators based on the entropy method are prone to distortion due to outliers and the influence of selected evaluation factors. Yet, empirical studies often overlook the heterogeneity in the measurement of explanatory variables, which potentially produces biased estimates. This study addresses these gaps by constructing a digitalization index based on text recognition named the Baidu Index and by employing a dynamic panel model to systematically analyze the intertemporal effects of digitalization on agricultural sustainability across 31 Chinese provinces. The key findings reveal that digitalization not only directly enhances agricultural sustainability but also exerts positive moderating effects through agricultural production, industrial structure, and technological progress. Critically, the results are slightly different when the choices are between absolute and relative units for agricultural carbon emissions and green total factor productivity, highlighting the necessity for standardized measurement frameworks in future research. Practically, policymakers should prioritize rural digital infrastructure investment and narrow the digital divide caused by institutional and technological factors. This study provides both a novel analytical framework and actionable insights for advancing sustainable agriculture in the digital era.

1. Introduction

Sustainable agriculture is the core foundation for ensuring food security, protecting the ecological environment, and achieving long-term economic and social stability and coordination [1,2]. Two critical indicators of agricultural sustainability are agricultural carbon emissions (ACEs), which include carbon emissions generated from resource utilization during agricultural production, and agricultural green total factor productivity (AGTFP), which quantifies the relationship between agricultural economic output and environmental impact while assessing the efficiency of sustainable agricultural development [3,4,5,6]. China’s agricultural sector has experienced significant growth in production capacity, but this growth has created some environmental problems, such as an increase in agricultural carbon emissions [7,8]. This is mainly due to the widespread use of agricultural machinery, fertilizers, and fossil fuels [9]; thus, the key players in the agricultural sector must make an effort to mitigate the environmental challenges of productivity growth. Carbon emissions from agriculture have become one of the most critical issues facing governments and researchers around the world [8]. Controlling and reducing such emissions are important tasks for the Chinese government. In addition to controlling agricultural carbon emissions, the green development of agriculture has become the focus of the policy issued by the Chinese government. AGTFP is a holistic measure of the impact of agricultural green development, incorporating four key dimensions: resource conservation, production efficiency, environmental protection, and economic growth. It can also comprehensively measure agricultural production behavior [10]. Enhancing agricultural green total factor productivity is an essential approach to foster green agricultural development and a vital prerequisite for achieving sustainable agricultural growth [11,12].

The deep integration of digital technology and agriculture is becoming a key path to improving agricultural sustainability, and traditional agriculture is being restructured through digital resource elements. Agricultural digital transformation has become an effective way to improve agricultural sustainable development, improve agricultural production efficiency, and reduce agricultural pollution [13,14]. Relying on digital technology can optimize the agricultural production mode through intelligent means, play a role in reducing industrial carbon footprint intensity, promote ecological resource adaptation, provide a good practice path for green and low-carbon transformation, etc. [15,16]. With the improvement in rural network coverage and the advancement of information technology, digital means are being integrated into the whole process of agricultural production, management, and marketing, and the utilization rate of agricultural resources is also improving [17,18]. At the same time, digitalization can also help mitigate the negative impact of agriculture on the environment, reduce the ecological damage caused by agricultural production, and make agricultural development more sustainable [19].

The impact of digitalization on agricultural sustainability has become a focus of academic attention, among which empirical analysis is one of the key research methods in this field. From an econometric perspective, if the relative unit is a scalar multiple of the cardinal unit, the choice of measurement units typically does not influence the econometric analysis of a single site. As a result, researchers usually choose different units of measurement based on research feasibility or research needs. In the existing literature studying digital transformation for agricultural sustainability, relatively little attention has been paid to the unit of measure, which, coupled with the choice of different measurement units, reflects the neglect of the choice of metrics in this area of research. By reviewing the currently available literature, the impact of the unit of measure on the validity of carbon emission measurement studies is not recognized. Yet the lack of research on such issues in the literature does not mean that the seemingly inconsequential difference in the unit of measure does not make a difference in coefficients and significance.

This study begins with a theoretical analysis of the direct and indirect impacts of digitalization on agricultural sustainability and discusses the results arising from different units of measurement. Then, using panel data from 31 provinces in China from 2013 to 2021, we empirically test whether digitization affects agricultural sustainability and explain the heterogeneity brought about by variable measures. Finally, we further examine the mechanism and heterogeneity of digitalization’s impact on agricultural sustainability. This study has the following marginal academic contributions: First, this study simultaneously investigates the dual dimensions of carbon emission and green production efficiency for the first time and constructs a framework to analyze the influence pathways of digitalization on agricultural sustainability, examining their parallel impacts and addressing the research gap in single-perspective studies. Secondly, it refines the definition of digitalization, the core explanatory variable, moving beyond the frequent use of the entropy method in prior studies. Although the entropy method is an objective index selection method, different types of index combinations bias the research conclusions. In this study, the world’s largest Chinese search engine, the Baidu Index, is used as a digital index to define digitalization more intuitively and dynamically. It increases the validity and scientificity of the use of core explanatory variables and contributes to the empirical research field of agricultural sustainability. Finally, for the first time in this research field, this study compares the research differences produced by different units of measurement, which improves the accuracy and reliability of the research. By systematically studying the impact of digitization on agricultural sustainability, the heterogeneity of measurement units is found, which not only provides micro-evidence support for the optimization of agricultural sustainability policy but also pioneers the analysis paradigm of empirical research in this field.

The rest of this paper is organized as follows: Section 2 presents the relevant literature, and Section 3 provides a theoretical analysis and proposes the research hypotheses. Section 4 contains the explanation of the methods and data, including the model setting, variable selection, and data description. Section 5 provides the empirical results and conclusions, further describing the quantitative analysis of the impact of digitization on agricultural sustainability, and Section 6 provides a summary of the research conclusions and recommendations. Figure 1 below shows the research roadmap of this paper. It illustrates the use of the OLS model to analyze how digitization directly and indirectly affects agricultural sustainability, with ACEs and AGTFP as core variables to describe the impact results. It further proves that the results remain reliable after analyzing measurement heterogeneity.

2. Literature Review

Agricultural activities account for approximately 14% of global greenhouse gas emissions and are thus a significant contributor to global greenhouse gas (GHG) emissions [3,20,21]. The path of traditional agriculture to reduce the environmental impact mainly focuses on the adjustment of the planting structure or the improvement in biotechnology, while its environmental benefits are decreasing under the constraint of land resources. In this context, the penetration of digital technology provides a new paradigm for reducing agricultural carbon emissions and improving agricultural green total factor productivity, and it drives the green low-carbon transformation of agriculture [7,22,23,24].

Existing studies confirm that agricultural digitalization reduces carbon emissions by driving technological innovation, optimizing production modes, and improving capital efficiency, thereby clarifying the nexus between digital technologies and sustainable agriculture. Yue Yuan et al. [18], using data from listed agricultural enterprises, showed that agricultural digitalization indirectly promotes green transformation by enhancing scale economies, technological innovation, and structural adjustments through mechanisms like factor reorganization. Further analysis of China’s 30 provinces’ panel data reveals that the digital economy significantly reduces agricultural carbon intensity by correcting capital misallocation and accelerating green technology adoption [25]. Qiang Zhou et al. [26], using rural digitalization and spatial econometrics, reported that concentrated tourism raises agricultural carbon intensity while digital adoption mitigates tourism-related emissions. The analysis of 30 Chinese provinces (2011–2021) shows that the digital economy reduces agricultural carbon emissions, with regional disparities and an inverted U-shaped emission trajectory [24]. On the basis of empirical analysis, studies show that digital agriculture reduces emissions through farmers scaling up operations, focusing on agri-services, and advancing technological innovation [7].

Agricultural digital transformation demonstrates dual efficacy in reducing carbon emissions and enhancing production efficiency, serving as dual pillars of sustainable agriculture that highlight the critical research imperative for digital innovation in agroecological systems. Haitao Wu et al. [4] used a DEA to measure agricultural energy efficiency and showed that the Internet enhances the efficiency of energy saving and emission reduction, with a dual-threshold effect on China’s agricultural green total factor energy efficiency (ATFEE) in threshold models. The analysis of 2825 farmers surveyed in 2020 shows that digital economy participation boosts EAT adoption, with digital marketing and finance indirectly improving production efficiency via the increased demand for machinery services, enhanced information access, and heightened food safety awareness [27]. It is verified using the World Input-Output Database (WIOD) that, under carbon emission constraints, the promotion of digitalization significantly improves low-carbon green total factor productivity (GTFP). At the same time, digitization promotes low-carbon GTFP through energy efficiency and labor productivity [23]. Junxia Liu et al.’s spatial econometric analysis revealed that digital finance (DIFI) drives low-carbon agriculture, achieving synergy between food security and production system decarbonization [28]. Research establishing a theoretical framework for agricultural digitalization’s impact on agricultural carbon productivity (ACP) reveals a positive effect driven by industrial structure upgrades and large-scale management. Moreover, urbanization and rural human capital improvements create a U-shaped effect, where digital transformation initially inhibits but ultimately promotes ACP [29].

Existing studies predominantly focus on the single-dimensional impact of digital technologies on agricultural carbon emission intensity and production efficiency, yet they lack a systematic synthesis of digital-driven sustainable agricultural transformation. A critical gap lies in the neglect of heterogeneity arising from measurement differences in empirical models. At the same time, most empirical studies quantify digitalization—the core explanatory variable—via index systems, which may distort objectivity due to varying weight distributions, thereby compromising model accuracy. While the existing literature has explored agricultural sustainability through dimensions like digital factor endowment, digital processes, and digital economy development and has conducted quantitative tests on key parameters (e.g., carbon emission estimation and factor productivity), the systematic analysis of agricultural sustainable development remains inadequate, and the robustness of econometric model construction requires further scrutiny. This study improves this analytical framework by explicitly linking agricultural carbon emissions and total factor productivity through digital technology while enhancing model precision and addressing the heterogeneity of measurement units in empirical research.

3. Theoretical Analysis and Research Hypothesis

Digital technological innovation has significantly increased the marginal benefits of agricultural sustainability by reconfiguring production factors. Traditional agricultural practices often suffer from mismatches in resource allocation due to fragmented technical information across various stages of production. The excessive reliance on chemical fertilizers, pesticides, and mechanization has contributed to environmental degradation—most notably, the heightened emissions of agricultural carbon and nitrogen [7,8,30]—which, in turn, impedes improvements in green total factor productivity. In contrast, digital technologies introduce information-driven elements into agriculture, optimizing the marginal productivity of technology, labor, and capital, thereby improving green total factor efficiency.

Aligned with the United Nations’ Sustainable Development Goals (SDGs), digital technologies enable holistic integration across agricultural production, management, and marketing. By enhancing resource allocation and reducing environmental impacts, these innovations support the transition toward resource-efficient, environmentally sustainable agricultural systems. First, digital innovation accelerates both the structural and technical transformation of traditional farming. For example, in crop cultivation, Unmanned Aerial Vehicle (UAV) monitoring enables precision fertilization and targeted disease management through high-resolution field data [27,31], while sensor networks measure soil fertility to reduce costs related to drip irrigation, pesticide application, and soil testing, allowing for fine-tuned control over production processes. Second, digital platforms diversify sales channels for agricultural products. Live-streaming e-commerce, in particular, reduces the number of intermediaries in supply chains, mitigates information asymmetry, and increases producer surplus by facilitating direct connections between farmers and consumers, thereby promoting sustainable agricultural development [32,33,34]. Third, the digital economy lowers transaction costs for agricultural operators by streamlining financial services such as digital payments, credit, and investment. Precision agriculture tools reduce resource waste and carbon emissions, while digital financial instruments foster equipment upgrades and the green restructuring of production factors, further advancing agricultural sustainability [6,35].

Unlike traditional farming, which often depends on chemical inputs, water-intensive irrigation, and extensive management, digital technologies such as UAVs and sensor networks enable the precise application of water, fertilizers, and pest control measures for minimizing resource waste. Concurrently, online marketplaces (e.g., live-streaming e-commerce) decrease the number of intermediaries in the supply chain, thereby enhancing efficiency. By integrating data from multiple sources, digital agricultural systems facilitate dynamic resource allocation and real-time carbon footprint tracking. The comparison of key agricultural digital technology performance is shown in the Appendix A. As a result, they outperform conventional approaches in scalability and environmental impact mitigation.

Hypothesis 1 is therefore proposed.

H1: 

Digitalization promotes agricultural sustainability and has a positive impact on agricultural carbon emission reduction and agricultural green total factor productivity.

Agriculture’s sustainable transition lags digital penetration, partly because technological deployment evolves progressively. On the one hand, the technology application itself has the characteristics of stages, and it is difficult for the initial investment and capacity building to be immediately transformed into the process of productivity improvement, data accumulation, and subject adaptation. On the other hand, the response of agro-ecosystems is naturally retarded, and the optimization of resource utilization efficiency by digital management requires multiple production cycles. This intertemporal effect essentially reflects that technological innovation and agricultural systems need time to match, and a balance between short-term costs and long-term benefits needs to be established in the process.

Digital technology has fundamentally transformed agricultural production by reconstructing the interactive logic of production factors, establishing a release mechanism for environmental pressures, and fostering a virtuous cycle within the agroecological economic system. This transformation is driven by technological progress under the digitalization wave, which addresses critical inefficiencies in agricultural capital allocation while amplifying the marginal returns of green production factors [22,23,35]. As a novel production factor, digitalization synergizes with technological advancements to optimize pathways for sustainable agricultural development [15,36].

The optimization of production scale achieved through land transfer and intensive management can enhance the efficiency of digitalization in agricultural systems. By improving transparency in land transfer processes, transaction costs are reduced, while large-scale operations lower the per-unit costs of advanced equipment such as intelligent irrigation systems and large agricultural machinery [37,38,39]. These structural advantages create an environment where digital technologies can more effectively drive resource optimization and operational efficiency. Furthermore, digitalization empowers business entities to adopt green technologies by enhancing their capacity to absorb innovations and tolerate risks, leading to more pronounced reductions in agricultural carbon emissions [24]. At the same time, digital tools reduce information exchange costs, supporting the transformation of the agricultural industry toward low-carbon structures [10,14].

Digital technologies at the production end, such as the Internet of Things (IoT) and remote sensing monitoring, play a critical role in reducing resource waste inherent in traditional extensive farming models [18]. For instance, the deployment of soil moisture sensors conserves water resources by enabling precision irrigation caused by excessive fertilization through data-driven nutrient management. Supply-side e-commerce platforms streamline production marketing linkages, significantly lowering logistics costs while minimizing energy consumption and post-harvest losses by shortening storage cycles [34]. Real-time sales data obtained via smart analytics help identify market trends, enabling flexible adjustments to crop yields, thus boosting efficiency. Digital transformation reshapes agricultural value chains and industrial structures. Structural shifts are evident in the tertiary sector, which is highly responsive to change, and structural shifts influence digitalization’s effects as technology adoption interacts with evolving frameworks to shape sustainability outcomes.

Given these mechanisms, it is plausible that the effects of digitalization on agricultural sustainability affect the current period of agriculture. At the same time, they are not completely independent but contingent on connected factors such as technological progress and the optimization of agricultural production scale and industrial structure.

Therefore, Hypothesis 2 is proposed.

H2: 

Digitalization affects the sustainable development of agriculture, exhibiting a trans-time effect. Moreover, technological progress and the optimization of the agricultural production scale and industrial structure have a moderating effect on the digitalization affect.

Ordinary least squares (OLS) is a linear unbiased estimation of minimum variance under the classical assumption. Under the classical linear regression assumption, ordinary least squares ensures the unbiased and effective parameter estimation and provides a robust and explanatory model framework for econometric analysis [40]. Therefore, we set up the OLS model to evaluate the impact of digitalization on agricultural sustainability, and we assume the model to be as follows:

$$Y={\beta }_0+{\beta }_1{X}_1+{\beta }_2{X}_2+\sum _{i=3}^k {\beta }_i{X}_i+\xi$$

(1)

where X₁ denotes the core explanatory variable digitalization, X₂ denotes the control variable disposable income of rural residents, X_i denotes the other control variables, and ξ denotes the residual term. Y represents the proxy variables of agricultural sustainability carbon emission and agricultural green total factor productivity. In this OLS model, the explanatory variable coefficient β₂ can be easily obtained as follows:

$${\beta }_2={\left(X^{\prime }X\right)}^{-1}{X}^{\prime }\mathrm{Y}$$

(2)

If the per capita disposable income of rural residents is converted to the per capita disposable income of rural residents divided by the per capita disposable income of all residents, the model is as follows:

$$Y={\beta }_0^{\prime }+{\beta }_1{X}_1+{\beta }_2^{\prime }{X}_2^{\prime }+{\sum }_{i=3}^k {\beta }_i{X}_i+\xi$$

(3)

In the OLS model, when the independent variable X₂ changes from the income of rural residents to the per capita disposable income of all residents, other coefficients do not change, but β₂ changes, and other coefficients such as intercept β₀ may be fine-tuned due to dimensional adjustment. In unit conversion, attention should be paid to the multicollinearity problem. If X₂ is related to other variables, the significance may change due to collinearity. When VIF exceeds 10, significant multicollinearity arises, propagating endogenous distortions in β₂’s estimates so that it is impossible to determine whether the change in β₂ is caused by the change in unit of measure. Although the estimates of β₂ remain unbiased, increases in variance and standard errors can lead to a decrease in T-value and a loss of significance.

In empirical studies on the impact of digitalization on agricultural carbon emissions and green total factor productivity, although the classical OLS theory reveals the mathematical influence of variable quantization on coefficients, more complex challenges are associated with the real data. Even if the credibility of the model is established through variable screening, the variance inflation factor test, and other means, differences in explanatory variables can result in significant variations in estimated outcomes, particularly because of disparities in their dimensions. Therefore, the selection of different variable measures can enhance the explanatory power of the model for real problems, which is also an important embodiment of the scientificity of empirical analysis.

H3: 

The unit heterogeneity analysis of explanatory variables can enrich the conclusions of agricultural sustainability research and obtain more accurate analysis results.

Due to differences in digital inequality between regions and the dependence on technical infrastructure, there are different degrees of digitalization in the four regions of eastern, Central, Western, and northeastern China [41,42]. The eastern region, relying on its economic advantages and technological innovation capability, has the most significant effect of digital emission reduction [7]. However, the layout of data centers also produces additional carbon footprints and partially offsets the reduction in carbon emissions. Relying on the advantages of large-scale production, major agricultural producing areas in Central China show carbon emission reductions due to the application of digital technology [43]. The Western region is constrained by infrastructure and ecological fragility. Although the state promotes the westward relocation of data centers, the digitalization of agriculture still faces challenging constraints of low network coverage. The northeast suffers from institutional inertia and an aging agricultural population, which leads to lagging technology adoption. Due to the different user groups of computer terminal and mobile terminal, the application of technology is different. In areas with weak infrastructure but a widespread mobile network, mobile devices, especially mobile phones, become key for digital applications [44]. In areas where traditional agricultural infrastructure is lagging, the high coverage of mobile networks and the popularity of smart phones form a unique technology empowerment path. As an integrated information terminal, mobile phones, with a built-in instant communication function have effectively eliminated the obstacles of knowledge acquisition and information exchange due to their portability and low threshold of operation. Accordingly, as a medium for collecting Baidu Index data, different hardware may yield different results.

Therefore, Hypothesis 4 is proposed.

H4: 

There are regional differences in the application effect of digitalization in sustainable agricultural development, and there is heterogeneity in the research results estimated by different Internet hardware facilities.

4. Methods and Data

4.1. Model Setting

To test the hypothesis of this study, we built the following OLS panel regression model for baseline regression:

$${ACE}_{it}={\beta }_0+{\beta }_1{Digit}_{it}+{\beta }_2{ECO}_{it}+{\sum }_{k=3}^{\mathrm{h}} {\beta }_{\mathrm{k}}{Control}_{\mathrm{i}\mathrm{t}}+{\mu }_i+{\delta }_t+{\epsilon }_{it}$$

(4)

$${AGTFP}_{it}={\beta }_0+{\beta }_1{Digit}_{it}+{\beta }_2{ECO}_{it}+\sum _{k=3}^{\mathrm{h}} {\beta }_{\mathrm{k}}{Control}_{\mathrm{i}\mathrm{t}}+{\mu }_i+{\delta }_t+{\epsilon }_{it}$$

(5)

Due to regional development, policy adjustment, the characteristics of agricultural products, and industrial characteristics, different regions in China differ in terms of technological progress, agricultural production scale, and industrial structure, and digital agriculture transformation and its characteristics influence agricultural carbon emissions and green total factor productivity through specific mechanisms. Based on this, the following moderating effect model is constructed in this study, considering the agricultural production scale, industrial structure, and technological progress as the mediating variables:

$${ACE}_{it}={\beta }_0+{\beta }_1{Digit}_{it}+{\theta }_M\times M_{it}+{\beta }_2{ECO}_{it}+{\sum }_{k=3}^{\mathrm{h}} {\beta }_{\mathrm{k}}{Control}_{\mathrm{i}\mathrm{t}}+{\mu }_i+{\delta }_t+{\epsilon }_{it}$$

(6)

$${AGTFP}_{it}={\beta }_0+{\beta }_1{Digit}_{it}+{\theta }_M\times M_{it}+{\beta }_2{ECO}_{it}+{\sum }_{k=3}^{\mathrm{h}} {\beta }_{\mathrm{k}}{Control}_{\mathrm{i}\mathrm{t}}+{\mu }_i+{\delta }_t+{\epsilon }_{it}$$

(7)

Previous studies in the field of agriculture have explored the impact of measurement unit on measurement results. For example, a top journal of agricultural economics [45] found that the result of benefit transfer may depend on whether the commodity is measured in the base unit or relative unit. Based on research on the impact of agricultural digitization on agricultural sustainability, this study quantifies the metric heterogeneity of the base unit and the relative unit of the agricultural economic development level.

4.2. Explained Variables

To interpret the scientific term agricultural sustainability, agricultural carbon emission (ACE) and agricultural green total factor productivity (AGTFP) were selected as two explanatory variables according to the previous literature. ACE refers to the research conducted by Chen and Li [7], and the ACE estimate constructed in this study is as follows:

$${ACE}_{it}={\sum }_{j=3}^{\mathrm{n}} S_{\mathrm{i}\mathrm{j}\mathrm{t}}={\sum }_{j=3}^{\mathrm{n}} P_{\mathrm{i}\mathrm{j}\mathrm{t}}{Q}_{\mathrm{j}}$$

(8)

where S_ijt represents the carbon emissions of the j specific carbon source in year t for the i province. In addition, P_ijt represents the amount of the j specific carbon source in year t of the i province, and Q denotes the carbon emission coefficient for the j-th specific carbon source. These carbon sources encompass diesel used in agricultural production, fertilizers, pesticides, agricultural film, irrigation, and tillage.

To estimate agricultural green total factor productivity, we used the SBM-BML method. Slacks-Based Measure (SBM) is a data envelopment analysis (DEA) model based on relaxation variables, and it is suitable for complex systems with multi-input and multi-outputs. The Bootstrap Malmquist–Luenberger (BML) method, which integrates the Malmquist–Luenberger index with Bootstrap techniques, is used for analyzing dynamic efficiency. It corrects the estimation errors of the traditional data envelopment analysis (DEA) model through statistical simulation. In contrast, it is necessary for the DEA method, commonly used in earlier studies, and requires specifying whether it is input-oriented or output-oriented; additionally, it assumes that input and output variables move in the same direction. Tone [46] proposed an SBM model based on relaxation measure and solved the problem of input–output relaxation and conducted an efficiency evaluation of the undesired output. The SBM model better captures efficiency losses under environmental constraints compared to traditional data envelopment analysis (DEA). Notably, input variables consist of carbon sinks, labor, land, fertilizers, pesticides, total agricultural machinery power, and plastic film. The expected output is the total agricultural output value after price adjustment, and the non-expected output is the estimated agricultural carbon emissions.

4.3. Core Explanatory Variables

This study uses the number of occurrences of digital-related terms such as “digitalization” and “digital economy” in the Baidu Index of each province from 2013 to 2021 as a proxy variable for digitalization at the provincial level. As the largest Chinese search engine, the Baidu Index can objectively quantify the intensity of public attention on digital issues based on the real search behaviors of users. Its characteristics of wide data coverage, a fast updating speed, and fine granularity provide a scientific empirical basis for research. In the past literature on digitalization, the entropy method was used to build a comprehensive evaluation system through multi-dimensional indicators, systematically integrating structured data from economic, social, and technical fields. In contrast, the Baidu Index directly captures the public’s active attention on digital-related issues, thereby reflecting the dynamic change in social cognition. Compared with the static statistical data that the entropy method relies on, it shows more immediacy and sensitivity of behavioral data. At the same time, the Baidu Index can effectively reduce the subjectivity of index selection and the influence of data lag in the traditional method by searching behavioral data with high frequency and detail. Since the entropy method is efficient in multidimensional systematic assessments, and the Baidu Index reveals cognitive change from the perspective of behavior, this study focuses on the latter and uses the Baidu Index as a proxy variable.

4.4. Mechanism Variable and Control Variable

According to previous studies by Chen and Li [7] and Li and Gao [25], we use the level of agricultural economic development (ECO) measured by the per capita disposable income of rural residents, and ECO2 is measured by the per capita disposable income of rural residents divided by the per capita disposable income of all residents. Agricultural disaster scope (DIS) is measured by the proportion of agricultural disaster area to agricultural planting area; agricultural financial expenditure (FIS) is characterized by the ratio of agriculture, forestry, and water conservancy expenditure to the total expenditure of the public budget; and the level of agricultural industry agglomeration is expressed by the location quotient (LQ). The LQ is the ratio of the regional employment share of the industry to the national employment share of the industry. The mechanism variable is selected to be the agricultural production scale (Scale) defined by agricultural land area, industrial structure (Stru) is expressed by the ratio of the output value of the tertiary industry to the output value of the secondary industry, and technological progress (Tech) measured by the total power of agricultural machinery as a proxy variable.

5. Descriptive Statistics

We use panel data covering 31 provinces of China from 2013 to 2021 in this study. The core explanatory variable, the Baidu Index, was derived from the official website of Baidu, and the original data of agricultural carbon emission measurement and agricultural green total factor productivity and other data were obtained from the China Rural Statistical Yearbook, China Environmental Statistical Yearbook, and Provincial and Municipal Statistical Yearbook. The descriptive statistics of the data are shown in

Table 1. Descriptive statistics.

Variable	N	Mean	SD	Min	Max
AGTFP	270	1.06	0.11	0.96	1.79
ACE	279	328.5	232.77	14.35	995.75
digit1	279	67.36	48.17	0.62	231.22
digit2	279	35.18	23.86	0.31	121.02
digit3	279	32.18	25.34	0	116.01
URB	279	0.6	0.12	0.24	0.9
ECO	279	1.43	0.58	0.56	3.85
ECO2	279	0.56	0.05	0.42	0.69
FIS	279	0.12	0.04	0.04	0.2
LQ	279	2.39	2.13	0.04	13.89
DIS	279	0.15	0.12	0.01	0.73

. In this table, the mean value of agricultural carbon emissions is 328.5, with maximum, minimum, and standard deviation values of 995.75, 14.35, and 232.77, respectively. This indicates that there are significant differences in agricultural carbon emissions among provinces at the national level. Similar patterns are observed for the digital core explanatory variables, while the descriptive statistical characteristics of the remaining variables show no significant fluctuations.

6. Empirical Results and Conclusions

6.1. Baseline Regression

Our focus is on using baseline regression models to determine the impact of digitization on agricultural sustainability and the heterogeneity arising from measures of different levels of agricultural economic development.

Table 2. Benchmark regression of ACE and AGTFP.

	ACE	ACE	AGTFP	AGTFP
digit1	−0.5380 ***	−0.4573 ***	0.0005	0.0006
	(0.1097)	(0.1014)	(0.0007)	(0.0006)
control varibles	YES	YES	YES	YES
time fixed effect	YES	YES	YES	YES
individual fixed effect	YES	YES	YES	YES
N	279.00	279.00	270.00	270.00
R2	0.9949	0.9951	0.3935	0.4115

Standard errors in parentheses: *** p < 0.01.

shows the main outcomes of digitization regarding agricultural sustainability in different units of measurement for different levels of agricultural economic development, all of which use the bidirectional fixed effects of years and individuals. Columns (1) and (2) show the effect of digitization on the reduction in carbon emissions from agriculture, which is estimated to be negative and statistically significant. Regarding the measure unit of per capita disposable income (ECO) of rural residents, for every 1 unit increase in digitization level, agricultural carbon emissions decrease by 5380 tons. On the other hand, when the per capita disposable income of rural residents is divided by the overall per capita disposable income measure (ECO2), agricultural carbon emissions decrease by 4570 tons. There is no significant difference between the two, but there is a deviation of about 15% in the coefficient. Columns (3) and (4) show that digitalization has a positive effect on agricultural total factor productivity but no statistically significant impact, and there is no heterogeneity under different measures of agricultural economic development levels.

6.2. Variance Inflation Factor Test

Table 3. Variance inflation factor checks.

Panel A: ACE
Variable	VIF
digit1	2.82			1.89
L1		2.72			1.93
L2			2.74			2.07
ECO	4.9	4.69	4.33
ECO2				1.44	1.44	1.39
URB	3.38	3.44	3.4	2.74	2.68	2.66
FIS	2.03	2.21	2.6	2.01	2.2	2.59
LQ	1.48	1.42	1.35	1.45	1.38	1.29
DIS	1.13	1.14	1.11	1.09	1.09	1.05
Mean VIF	2.62	2.6	2.59	1.77	1.79	1.84
Panel B: BTFPCH
Variable	VIF
digit1	3.19			1.85
L1		2.98			1.86
L2			2.84			1.96
ECO	6.11	5.84	5.34
ECO2				1.27	1.26	1.23
URB	4.14	4.15	4	2.49	2.4	2.36
FIS	1.96	2.14	2.5	1.94	2.12	2.5
LQ	1.62	1.52	1.4	1.63	1.52	1.39
DIS	1.12	1.12	1.09	1.12	1.1	1.07
Mean VIF	3.02	2.96	2.86	1.72	1.71	1.75

shows the results of verifying the inflation factors for all variables in the ACE and AGTFP empirical models. In OLS regression, the variance inflation factor must be less than 10 to avoid multicollinearity. In Table 2, the inflation factor of all variables is less than 10, less than five values of the inflation factor are more than 95.8%, and the average value of the overall inflation factor is 2.27. Therefore, multicollinearity is excluded.

6.3. Robustness Test

To test the reliability of the baseline regression results, we used three methods of testing. The first method is the robustness test that entails removing samples from remote areas. Specifically, the samples in the original analysis covered 31 provinces across the country. Considering that there may be differences in data collection or heterogeneity in policy implementation in remote areas, we excluded the sample data from Xinjiang and Tibet, two remote areas. After re-estimation, we found that the direction of coefficients, significance level, and economic significance of core variables did not change.

Furthermore, the panel correction standard error was added to re-estimate the model. In view of the possible problems such as inter-group heteroscedasticity, sequence correlation, and cross-sectional dependence in panel data, this method alleviates the result bias caused by the structural bias of error terms by adjusting the estimation method of standard errors. Based on the original data, the coefficient direction and significance level of the core explanatory variables did not change under the premise of keeping the core variables and control variables unchanged.

Finally, a robustness analysis was carried out by removing samples of extreme values. The specific method involves removing outliers larger than 99% of the explained variables and then re-estimating the model. After re-estimation, the coefficient direction, significance level, and economic significance of the core explanatory variables were consistent with the benchmark results.

In addition, the R² value of the above key hypothesis test hardly changed, indicating that the conclusion is highly resistant to extreme value interference. A series of robustness tests show that the results are robust. Therefore, Hypothesis 1 is confirmed.

The robustness check results are shown in

Table 4. Robustness check.

Panel A: Excluding Xinjiang and Tibet samples
	ACE	ACE	AGTFP	AGTFP
digit1	−0.5467 ***	−0.4075 ***	0.0000	0.0001
	(0.1261)	(0.1076)	(0.0006)	(0.0005)
control varibles	YES	YES	YES	YES
time fixed effect	YES	YES	YES	YES
individual fixed effect	YES	YES	YES	YES
N	261.00	261.00	261.00	261.00
R2	0.9951	0.9955	0.4937	0.5026
Panel B: Add to panel correction standard error
	ACE	ACE	AGTFP	AGTFP
digit1	−0.5380 ***	−0.4573 ***	0.0005	0.0006
	(0.0824)	(0.0803)	(0.0009)	(0.0007)
control varibles	YES	YES	YES	YES
time fixed effect	YES	YES	YES	YES
individual fixed effect	YES	YES	YES	YES
N	279.00	279.00	270.00	270.00
R2	0.9949	0.9951	0.3935	0.4115
Panel C: Excluding extreme values
	ACE	ACE	AGTFP	AGTFP
digit1	−0.5298 ***	−0.4486 ***	0.0001	0.0001
	(0.1095)	(0.1012)	(0.0006)	(0.0005)
control varibles	YES	YES	YES	YES
time fixed effect	YES	YES	YES	YES
individual fixed effect	YES	YES	YES	YES
N	277.00	277.00	268.00	268.00
R2	0.9947	0.9949	0.4524	0.4595

Standard errors in parentheses: *** p < 0.01.

.

6.4. Mechanism Analysis

Mechanism 1: Digitization has an intertemporal impact on sustainable agriculture.

We further examined the intertemporal dynamic effects of digitalization. The regression results show that the first and second phases of digitalization lag have a significant negative effect on agricultural carbon emissions and promote the agricultural green total factor productivity. Each 1 unit increase in digitization can result in 6860 tons and 9170 tons of carbon emission reductions in the case of ECO measurement and 56,100 tons and 65,800 tons of carbon emission reductions in the case of ECO2 measurement. As for the impact of total factor productivity, the different measures of ECO and ECO2 still have heterogeneity. Under the ECO unit, the digitization lag term has a significant positive impact on the green total factor productivity. Each unit of increase in the digitalization level increases the level of the second-year and the third-year total factor green productivity by 0.0014 and 0.0028, respectively. However, the lag in the first phase has no significant positive impact on ECO2’s unit of measurement, and each unit of increase in the second phase of the lag will still significantly increases it by 0.0019 units.

The Mechanism 1 results are shown in

Table 5. Impact of digitization lag 1 and lag 2.

Panel A: Impact of digitization lag 1 and lag 2 on ACE
Variables	ACE
L.digit1	−0.6861 ***		−0.5606 ***
	(0.1181)		(0.1098)
L2.digit1		−0.9169 ***		−0.6582 ***
		(0.1464)		(0.1359)
control varibles	YES	YES	YES	YES
time fixed effect	YES	YES	YES	YES
individual fixed effect	YES	YES	YES	YES
N	248	217	248	217
R2	0.9957	0.9962	0.9958	0.9961
Panel B: Impact of digitization lag 1 and lag 2 on AGTFP
Variables	AGTFP
L.digit1	0.0014 *		0.0012
	(0.0008)		(0.0007)
L2.digit1		0.0028 ***		0.0019 **
		(0.001)		(0.0009)
control varibles	YES	YES	YES	YES
time fixed effect	YES	YES	YES	YES
individual fixed effect	YES	YES	YES	YES
N	240	210	240	210
R2	0.4331	0.3795	0.4596	0.4024

Standard errors in parentheses: * p < 0.1, ** p < 0.05, *** p < 0.01.

.

Mechanism 2: The agricultural production scale, industrial structure, and technological progress have positive moderating effects on agricultural sustainability.

Interaction terms such as digit1 * Tech, digit1 * Sca, and digit1 * Indu were introduced into the regression model to test whether the variables of agricultural production scale, industrial structure, and technological progress play a moderating role in the relationship between digitalization, agricultural carbon emissions, and green total factor productivity. For the analysis of agricultural carbon emissions, both the ECO and ECO2 cases have a significant regulatory effect on technological progress for reducing agricultural carbon emissions, and there is no heterogeneity of measurement units. The agricultural production scale has a significant moderating effect on reducing agricultural carbon emissions, but there is a certain degree of metric heterogeneity. In terms of industrial structure, under different measures, the ECO measure had a negative effect, but no statistical significance was found. For green total factor productivity, the agricultural production scale has a negative moderating effect, but the coefficient is very small, and there is no metric heterogeneity for ECO and ECO2. So far, Hypothesis 2 and Hypothesis 3 are proven.

The Mechanism 2 results are shown in

Table 6. Analysis of moderating effects.

Panel A: Under the ECO unit of measure
Varibles	ACE			AGTFP
digit1	−0.0997	−0.4523 ***	−0.5179 ***	0.0003	0.0015 **	0.0007
	(0.1109)	(0.1195)	(0.1228)	(0.0008)	(0.0007)	(0.0007)
c.digit1~Tech	−0.0001 ***			0.0000
	(0.0000)			(0.0000)
c.digit1~Sca		−0.0002 *			−0.000002 ***
		(0.0001)			(0.000001)
c.digit1~Indu			−0.0141			−0.0001
			(0.0386)			(0.0002)
control varibles	YES	YES	YES	YES	YES	YES
time fixed effect	YES	YES	YES	YES	YES	YES
individual fixed effect	YES	YES	YES	YES	YES	YES
N	279.00	279.00	279.00	270.00	270.00	270.00
R2	0.9960	0.9950	0.9949	0.3920	0.4222	0.3919
Panel B: Under the ECO2 unit of measure
Varibles	ACE			AGTFP
digit1	−0.0686	−0.3026 ***	−0.5098 ***	0.0004	0.0010	0.0009
	(0.0984)	(0.1023)	(0.1189)	(0.0007)	(0.0006)	(0.0007)
c.digit1~h	−0.0001 ***			0.0000
	(0.0000)			(0.0000)
c.digit1~a		−0.0005 ***			0.0000 **
		(0.0001)			(0.0000)
c.digit1~u			0.0317			−0.0002
			(0.0374)			(0.0002)
control varibles	YES	YES	YES	YES	YES	YES
time fixed effect	YES	YES	YES	YES	YES	YES
individual fixed effect	YES	YES	YES	YES	YES	YES
N	279.00	279.00	279.00	270.00	270.00	270.00
R2	0.9963	0.9955	0.9951	0.4099	0.4244	0.4118

Standard errors in parentheses: * p < 0.1, ** p < 0.05, *** p < 0.01.

.

6.5. Heterogeneity Analysis

A regional heterogeneity test and search media heterogeneity test were carried out.

In our study, through the heterogeneity test, significant regional heterogeneity was found in the impact of digitalization on the study objects, which was specifically manifested as systematic differences in the carbon reduction effect and green total factor productivity improvement in the eastern, Central, Western, and northeastern regions of China, and measurement heterogeneity existed in some results. The most obvious coefficients of agricultural carbon emission reduction effect for ECO and ECO2 for the eastern region are 0.597 and 0.396. Digitalization in Central and northeast China has brought about the improvement in agricultural green total factor productivity, which has statistical significance, and there is a certain measure of heterogeneity.

Information search is usually performed using computer and mobile terminals, and the impact of digitization on agricultural sustainability is significantly different in different search media. Specifically, the digitalization index effect on PCs is smaller than that on mobiles. In terms of green total factor productivity, both of them have positive effects, but there is no statistical difference. Thus, Hypothesis 4 is proven.

The heterogeneity results are shown in

Table 7. Tests for regional heterogeneity.

Panel A: East
	ACE	ACE	AGTFP	AGTFP
digit1	−0.5971 ***	−0.3957 **	0.0003	−0.0001
	(0.2229)	(0.1896)	(0.0007)	(0.0006)
control varibles	YES	YES	YES	YES
time fixed effect	YES	YES	YES	YES
individual fixed effect	YES	YES	YES	YES
N	90.00	90.00	90.00	90.00
R2	0.9960	0.9962	0.0864	0.0713
Panel B: Middle
	ACE	ACE	AGTFP	AGTFP
digit1	0.5769	0.0818	0.0004	0.0016 **
	(0.3677)	(0.2667)	(0.0008)	(0.0008)
control varibles	YES	YES	YES	YES
time fixed effect	YES	YES	YES	YES
individual fixed effect	YES	YES	YES	YES
N	45.00	45.00	45.00	45.00
R2	0.9972	0.9981	0.8397	0.8183
Panel C: West
	ACE	ACE	AGTFP	AGTFP
digit1	−0.4108 ***	−0.4644 ***	0.0008	0.0012
	(0.1207)	(0.1301)	(0.0014)	(0.0012)
control varibles	YES	YES	YES	YES
time fixed effect	YES	YES	YES	YES
individual fixed effect	YES	YES	YES	YES
N	108.00	108.00	99.00	99.00
R2	0.9940	0.9935	0.1741	0.2012
Panel D: Northeast
	ACE	ACE	AGTFP	AGTFP
digit1	−0.1843	−0.4647	0.0203 *	0.0121
	(0.3824)	(0.3524)	(0.0107)	(0.0129)
control varibles	YES	YES	YES	YES
time fixed effect	YES	YES	YES	YES
individual fixed effect	YES	YES	YES	YES
N	27.00	27.00	27.00	27.00
R2	0.9968	0.9977	0.4455	0.3266

Standard errors in parentheses: * p < 0.1, ** p < 0.05, *** p < 0.01.

and

Table 8. Heterogeneity of search tools, PC and mobile.

Panel A: PC
	ACE	ACE	AGTFP	AGTFP
	−0.7998 ***	−0.6176 ***	0.0019	0.0017
	(0.2321)	(0.2128)	(0.0014)	(0.0012)
control varibles	YES	YES	YES	YES
time fixed effect	YES	YES	YES	YES
individual fixed effect	YES	YES	YES	YES
N	279.00	279.00	270.00	270.00
R2	0.9947	0.9949	0.3970	0.4143
Panel B: Mobile
	ACE	ACE	AGTFP	AGTFP
	−0.9035 ***	−0.8401 ***	0.0001	0.0004
	(0.1753)	(0.1652)	(0.0011)	(0.0010)
control varibles	YES	YES	YES	YES
time fixed effect	YES	YES	YES	YES
individual fixed effect	YES	YES	YES	YES
N	279.00	279.00	270.00	270.00
R2	0.9950	0.9952	0.3919	0.4097

Standard errors in parentheses: *** p < 0.01.

.

7. Discussion

7.1. Tracing Impact Pathways of Digitalization for Sustainable Agriculture

The digital revolution is considered to be one of the most important driving forces for agricultural development, and its deep transformation of agricultural production and distribution has brought economic and environmental benefits to society. At present, agricultural digitalization is in a rapid development stage, with developed countries developing earlier and making larger investments in the field, while developing countries lag behind developed countries [47,48]. The digitalization process has been actively promoted globally in recent years, which includes various aspects such as digital economy, digital governance, digital infrastructure, and digital services. These efforts not only provide new ideas for solving the problems faced by traditional agriculture but also contribute to improving the environmental efficiency of agriculture [13,28,36]. Despite the potential risks associated with the digitization process, such as information security, digital divide, and technological inequality, it is still recognized as an effective way to promote low-carbon transformation in agriculture. This study focuses on two core indicators, agricultural carbon emissions and agricultural green total factor productivity, and reveals the causal link between digitalization and agricultural sustainability by comparing data changes across overall time as well as regional time. Both the environmental impacts of agricultural production activities and the efficiency of resource utilization are assessed, making agricultural sustainability a two-way study of pollution control and green development in the environmental dimension. Existing studies mostly use the indicator methods to construct digitized proxy variables in which the entropy value method is more common. In this study, unlike previous studies, we adopt the text recognition method to construct digitized indicators and innovatively integrate the dynamic search data of the Baidu Index by capturing the changes in keyword heat. Compared with the traditional entropy method, it not only innovates the research means but also avoids the limitations of the latter due to the complexity of quantification and susceptibility to outliers.

We found that digital transformation has a positive effect on the sustainable development of agriculture. On the one hand, digitalization effectively reduces agricultural carbon emissions, a finding that is consistent with previous research results, including digitalization significantly reducing agricultural carbon emissions [7]; agricultural carbon emissions show a downward trend after the enhancement of rural digital spatial agglomeration [49], the development of the digital economy has a significant inhibiting effect on the intensity of carbon emissions from agricultural energy [25], and so on. On the other hand, agricultural green total factor productivity shows positive changes, and this finding effectively supports past research conclusions that digitalization significantly increases low-carbon green total factor productivity [23]; digital finance has a positive impact on agricultural green total factor productivity [22], farmers’ use of the Internet significantly increases agricultural green total factor productivity [11], etc.

The impact of digitalization on the sustainable development of agriculture does not exist in isolation but forms a synergistic effect with factors such as technological progress, production scale, and industrial structure. In recent years, the Chinese government has attached great importance to the development of agricultural digitalization and has promoted the transformation of governments at all levels, enterprises, and farmers to digital agriculture by vigorously issuing policy documents. These policies not only provide supporting funds but also accelerate the implementation of technology through the tilting of resources. Against this backdrop, digital technologies have penetrated into the production and distribution ends of agriculture, enhancing the economic benefits of agriculture while improving the environmental benefits at the same time. The regulation of agricultural sustainability by digitalization is not only reflected in technological progress but also in the marginal benefits of resource utilization amplified by large-scale land management. In addition, the optimization of industrial structure further promotes the green transformation of agriculture by extending the agricultural industry chain and developing high-value-added products. In other studies, it has also been found that elements such as technological progress and the optimization of the business scale and industrial structure promote agricultural sustainability through mechanisms such as mediating effects. For example, digital rural development plays a mediating role in digital finance on agricultural green total factor productivity [22], digital economic development mitigates agricultural energy carbon emission intensity by mitigating agricultural capital mismatch and promoting green technological innovation [25], digitalization has a mediating effect through technological progress [7], digitalization produces a regional linkage effect on agricultural decarbonization through technological diffusion [24], farmers’ use of Internet through land transfer enhances green total factor productivity in agriculture [11], etc.

The effect of the unit of measure on econometric model estimation in the field of agriculture is still in the early stage of research, except for the descriptions of base unit and relative unit (e.g., policy-affected marsh area), which brings about different econometric estimation results, and it has been investigated in an article published in a top journal of agricultural economics [45]. For the time being, it has not been found in other studies. In this study, an attempt was made in this area, and although individual results show differences due to relative and base units, they do not affect the conclusions.

A group regression test reveals significant group differences in the impact of digitization on agricultural sustainability. The significant facilitation effect in the eastern and Western regions, and the relatively weak performance in the center and northeastern regions, result from a number of factors. Firstly, the high level of infrastructure development in the eastern region and the more efficient functioning of the coastal economy reinforce the benefits of technology diffusion. In the Western region, despite weak digital infrastructure, the environmental benefits of digitization are amplified by the east–west collaboration and a relatively underdeveloped green productivity base. In contrast, the digital divide and over-reliance on traditional factors of production in the Central and northeastern parts of the country due to insufficient infrastructure coverage make the effects of digitization less significant. Second, the eastern region has the human capacity to bridge the skills gap between urban and rural areas, while the northeastern region faces obstacles pertaining to the implementation of digital tools due to limited technology acceptance caused by population loss and other reasons. Finally, the policy transmission mechanism in the eastern region, such as subsidies going directly to farmers, reinforces the impact of digitization, while in the Western region, national policy favoritism and the return of young laborers are combined to form a driving advantage, transforming resource endowments into green transformation momentum. In the Central and northeastern regions, the inefficiency of policy implementation has weakened institutional incentives, and this institutional difference affects the effect of digital empowerment.

7.2. Research Related Practical Value

The application of digital solutions in the production and distribution of agricultural products contributes to agricultural sustainability, but it also faces challenges in implementation.

First, as farmers generally have lower levels of formal education, their digital literacy becomes a critical factor in adopting digital tools. Studies have shown that training programs can help farmers acquire the necessary digital skills, and the development of agricultural socialized services can effectively reduce their learning costs. However, these approaches require additional resource investment [50,51]. Second, in terms of agricultural product distribution, e-commerce has emerged as an important driver of agricultural digitalization, enhancing both farmers’ incomes and overall agricultural sustainability [32]. Nonetheless, farmers’ willingness to engage in e-commerce remains a concern. For instance, in Wuchang, one of the more agriculturally developed regions in northeastern China, some farmers still show low levels of enthusiasm toward participating in agricultural e-commerce initiatives [33]. Third, although policies actively promote the digital transformation of agriculture, funding remains one of the key limiting factors. As a digital tool, inclusive finance can effectively alleviate financing constraints and support sustainable agriculture, but regional disparities have intensified unequal access to such resources [6,28,35].

Moreover, digital investment may carry the risk of negative returns. Advancing digitalization without considering local resource conditions could result in resource waste. Therefore, the promotion of agricultural digitalization should be tailored to local realities to ensure that it truly enhances both economic and environmental outcomes in agriculture.

7.3. Limitations and Future Directions

This study exhibits two limitations. First, this study addresses the overall impact of digitization on agricultural sustainability but only provides a preliminary analysis of representative digital tools or smart facilities, such as drone technology and the Internet of Things (IoT). Since it is difficult to obtain panel data on the variables in the existing database, further research will be conducted in the future after investigating the data. Second, the negative effects of digitization still need to be further explored. Although the existence of regional digital divide and digital facility dependence is initially observed, its mechanism has not been systematically tested, and the potential association can be further explored through case studies in the future.

8. Conclusions and Suggestions

Based on panel data from 31 provinces in China between 2013 and 2021, this study employs a novel digital definition and incorporates the Baidu Index to analyze factors to calculate their impact on agricultural sustainability. It also empirically examines the impact of digitalization on two representative proxy variables, agricultural carbon emissions and agricultural green total factor productivity, as well as the mechanism and the heterogeneity of measures of the agricultural economic development level. The conclusions drawn from the OLS model are as follows.

Firstly, digitalization enhances agricultural sustainability by lowering carbon emissions and boosting green total factor productivity. Furthermore, the robustness test confirms that the findings align with the baseline regression results. Secondly, digitization has intertemporal influence. For agricultural carbon emissions, the effect of interphase and interphase reduction is more obvious. The improvement effect of agricultural green total factor productivity became significant after the interperiod of digitization, while in the current period, although digitization had a positive impact, it did not show statistical significance. Thirdly, there is a mechanism of digitalization to positively moderate agricultural sustainability through the agricultural production scale, industrial structure, and technological progress. The reduction in agricultural carbon emissions is significant, but the improvement in agricultural green total factor productivity is relatively limited. Additionally, the heterogeneity analysis reveals that the digital economy has varied impacts on agricultural energy-related carbon emissions. Specifically, digitalization has a notable positive effect on agricultural sustainability in the eastern and Western regions of China, while its impact is less pronounced in the Central and northeastern regions. From the perspective of search media, the digitalization index effect on PCs is smaller than that on mobiles, but there is no significant difference in green total factor productivity. Last but not the least, among the research results, there are relatively obvious differences in the coefficient of agricultural carbon emission between the two interperiod studies and the heterogeneity analysis in the eastern region, but this does not affect the significance of the results. The heterogeneity of agricultural carbon emissions and agricultural green total factor productivity measures in other parts of the study is not large, which proves that the research results are credible and enrich the research conclusions.

Based on the above conclusions, the following suggestions are put forward in this study. First, the government should speed up the promotion of digitalization and improve the application level of digital technology in the agricultural field. To advance agricultural sustainability, it is essential to enhance the widespread adoption of digital infrastructure and Internet accessibility. Additionally, fostering the implementation of technologies like the Internet of Things and big data across the entire agricultural supply chain is crucial. Moreover, providing farmers with robust technical training can facilitate the integration of scientific and technological innovations with agricultural practices. Secondly, in the process of deepening the digital transformation of agriculture, it is necessary to simultaneously promote the intensification of production scale, the upgradation of industrial structure, and the intelligent development of technological systems. Through the popularization of intelligent equipment such as intelligent agricultural machinery to achieve scale management, a food traceability system should be built with the help of digital platform technology, and data should be used to empower the agricultural industry to achieve industrial chain extension and interaction. Digital elements and traditional resources should be implemented to improve green total factor productivity while reducing the carbon footprint, a positive feedback mechanism should be developed for the entire agricultural industry chain under digitalization, and the momentum of sustainable agricultural development should be increased. Then, digital construction should be promoted according to local conditions. While retaining the digital advantages of the Central and eastern regions, market-oriented digital reform should be promoted in the northeast and Western regions. The digital farm model contracted by enterprises should be promoted, and large-scale agricultural machinery networking should be used to improve operation accuracy. Western mountainous areas focus on training local e-commerce experts to expand the market for featured agricultural products through live streaming. At the same time, a cross-provincial technical assistance mechanism has been established, and a team of experts from the east has been organized to provide targeted support to farmers in the Western region and impart advanced agricultural technologies and other practical skills. As a large developing nation with a strong aggregate but low per capita agricultural output, China’s digitalization offers valuable insights for less-developed agrarian economies. Furthermore, its regionally varied digitalization has successfully reduced agricultural emissions and the observed cross-temporal effects of digitalization, providing strategies for countries with uneven development to improve environmental outcomes. Finally, it is suggested that in the empirical research of agricultural sustainability and digital transformation, attention should be paid to the influence of variable measurement methods on research conclusions. In a specific operation, the key explanatory variables (such as economy-related variables) can be compared and analyzed using different measurement standards. If it is found that after changing the dimension, the significance of the variable does not fluctuate significantly and the direction of the coefficient is stable, this indicates that the conclusion has strong reliability. On the contrary, if the significance disappears or the coefficient changes significantly, it is necessary to comprehensively analyze the experimental results to avoid misjudgment caused by a single measurement method and ensure the reliability of the research results.

Due to data limitations, this study was restricted to the provincial level and did not delve into the municipal or county levels, resulting in a lack of more precise research findings and insights. Future research could utilize computer technology to further collect data and improve research accuracy. Additionally, representative digital tools such as big data and drones have significant research value. However, existing studies primarily rely on qualitative analysis and lack large-scale empirical studies, which is another area worthy of in-depth exploration.