How does the digital transformation of agriculture affect carbon emissions? Evidence from China’s provincial panel data

The study investigates whether and how the digital transformation of agriculture (DTA) mitigates agricultural carbon emissions (ACEs) in China. It situates the research within the global and national context where agriculture contributes substantially to greenhouse gas emissions, and China faces pressing decarbonization commitments. With rapid diffusion of digital technologies (IoT, big data, AI) into agriculture, the DTA is posited to improve resource efficiency, enable cleaner production, and strengthen environmental governance. The paper highlights gaps: limited agriculture-specific measures of digitalization, insufficient understanding of mediating mechanisms (scale, structure, technology), and potential nonlinear/heterogeneous effects across regions and emission sources. The study aims to quantify DTA’s effect on ACEs, identify mediating channels, and test for regional, industrial, and source-based heterogeneity and threshold nonlinearity using provincial panel data (2015–2021).

The literature review covers three strands. (1) Measurement of digitalization: Numerous index systems measure digitalization at interprovincial and city levels in China and in Europe; however, agriculture-specific digitalization measures are scarce. Some studies use rural internet penetration, Taobao Villages, and digital industrialization indicators to proxy agricultural digitalization. (2) Digital economy and carbon emissions: Many studies find the digital economy reduces emissions via environmental governance, technological innovation, shifts in energy structure, and institutional improvements, while others note non-monotonic or rebound patterns. (3) Drivers of agricultural carbon emissions: Prior work attributes ACEs to factors such as industrial structure, population, development level, energy intensity, consumption patterns, and land endowment using LMDI, STIRPAT extensions, and econometric analyses. Few studies focus specifically on DTA’s impact on ACEs, often overlooking heterogeneity and mechanisms, especially the roles of scale, structure, and technology and potential nonlinear effects. This study addresses these gaps by constructing a multidimensional DTA index, testing mediating mechanisms, and probing heterogeneity and threshold effects.

Research design employs panel econometric models on 30 Chinese provinces (excluding Xizang, Hong Kong, Macao, Taiwan) from 2015 to 2021. Key components: (1) Benchmark panel regression with province and year effects: ln(ACE_it) = α0 + α1 ln(DTA_it) + α2 ln(C_it) + μ_i + ν_t + ε_it, where ACE denotes agricultural carbon emissions and C is a vector of controls. (2) Mediation analysis (Wen & Ye, 2014) combining stepwise regressions and Bootstrap to test three mediators: agricultural production scale (SCALE), agricultural industrial structure (STRU), and agricultural technological progress (TECH). Equations include ln(M_it) = β0 + β1 ln(DTA_it) + β2 ln(C_it) + μ_i + ν_t + ε_it, and ln(ACE_it) regressed on DTA, mediators, and controls; significance of β1, η2, and changes in α1 determine mediation. (3) Threshold model (Hansen-type) with DTA as the threshold variable to detect nonlinear effects: ln(ACE_it) = α0 + Σ λ_k ln(DTA_it)·I(·) + α0 ln(C_it) + ν_t + ε_it, testing single vs multiple thresholds via bootstrap. Variables: ACE constructed as the sum of emissions from six sources (fertilizers, pesticides, agricultural film, diesel oil, cultivation, irrigation), using published emission coefficients (e.g., fertilizers 0.895 kg/kg; pesticides 4.9341 kg/kg; film 5.18 kg/kg; diesel 0.5297 kg/kg; cultivation 312.6 kg/km²; irrigation 25 kg/hm²). DTA (core explanatory variable) is a composite index of (a) digital infrastructure (rural internet and mobile penetration, agromet station coverage, cable TV coverage), (b) digitalization of agricultural industry (Taobao Villages, e-commerce demonstration counties), and (c) digital finance (digital inclusive finance index). The index is computed via the entropy weight method: normalization, contribution shares p_ij, entropy e_j, redundancy d_j, weighting, and aggregation. Mediators: SCALE = average contracted arable land area per household; STRU = share of value added by agricultural support services; TECH = cumulative DEA-Malmquist technological progress index. Controls: urbanization rate (URB), rural per capita income (ECO), agricultural disaster extent (DIS), agricultural fiscal expenditure share (FIS), and agricultural industrial agglomeration (AGG). Estimation strategy includes OLS, random effects (RE), fixed effects (FE), with Hausman tests favoring FE; instrumental variable (IV) approach to address endogeneity (with Kleibergen-Paap statistics reported); multiple robustness checks (dynamic panel GMM with AR(1)/AR(2)/Hansen tests, alternative variable specifications, etc.). Heterogeneity analyses split samples by region (eastern, central, western, northeastern), by major grain-producing vs non-grain regions, and by emission source components. Threshold analysis identifies and estimates single-threshold nonlinearity in DTA’s effect on ACEs.

- Benchmark regressions: DTA significantly reduces ACEs. In FE models, ln(DTA) coefficient ≈ −0.041 (10% significance). Hausman tests indicate FE is preferred; time fixed effects are jointly significant. - Control variables: Urbanization tends to increase ACEs; higher agricultural fiscal expenditure is associated with lower ACEs; disaster extent negatively associated with ACEs; effects of ECO and AGG align with prior studies (some mixed significance). - Endogeneity: IV estimation supports a negative causal effect of DTA on ACEs (second stage ln(DTA) ≈ −0.504*, Kleibergen-Paap rk Wald F ≈ 5.385, indicating marginal instrument strength). - Robustness: Across specifications (including dynamic GMM), DTA’s coefficient remains significantly negative (e.g., −0.105**, −0.085**, −0.039* depending on model), with appropriate AR(1)/AR(2) and Hansen diagnostics. - Mediation: DTA increases SCALE (β>0), STRU (β>0), and TECH (β>0); each mediator significantly reduces ACEs when included, indicating partial mediation through all three channels. Magnitudes: Agricultural industrial structure exhibits the largest mediating effect; production scale the smallest. Sobel tests show significant indirect effects (e.g., z ≈ −2.12 for SCALE, −1.77 for STRU). - Heterogeneity by region: Significant negative effect only in the eastern region (ln(DTA) ≈ −0.057**, t≈−2.56); effects in central, western, northeastern regions are statistically insignificant. - Heterogeneity by industrial type: Non-grain production areas show significant reductions (ln(DTA) ≈ −0.026**), whereas major grain-producing areas show no significant effect. - Heterogeneity by emission source: DTA significantly reduces ACEs from fertilizers (≈ −0.046*, 10%) and diesel oil (≈ −0.194**, 5%); effects on pesticides, agricultural film, cultivation, and irrigation are not significant. - Nonlinear threshold effect: A single threshold in DTA is detected (threshold ≈ −1.9495; significant at 1–5%). Below the threshold, the DTA effect on ACEs is negative but insignificant; above the threshold, the effect becomes significantly negative (coefficient ≈ −0.028**, 5%). By 2021, 24 provinces exceeded the threshold, implying widespread but uneven realization of DTA’s emission-reducing impact.

The findings confirm that the digital transformation of agriculture is an effective pathway to reduce agricultural carbon emissions, addressing the core research question. Mechanistically, DTA lowers ACEs by enabling moderate scaling-up of farm operations (economies of scale and more efficient mechanization), optimizing the agricultural industrial structure (expanding service-support activities and integrating with other sectors), and fostering technological progress (precision inputs, smart irrigation, data-driven management). However, these benefits materialize unevenly: regions with stronger economic bases, better infrastructure, and innovation resources (eastern China) experience significant reductions, while less-developed regions lag, underscoring the importance of complementary investments in digital infrastructure, skills, and institutional support. Source-specific results suggest DTA’s near-term impact is strongest where precision and operational efficiencies are most tractable (fertilizer management and machinery fuel use). The threshold effect reveals that partial or fragmented digitalization may be insufficient; emission reductions become statistically robust only beyond a certain DTA level, highlighting nonlinearity and the need for critical mass in digital capabilities. Overall, the results underscore DTA’s strategic role in China’s agricultural decarbonization agenda and inform targeted policy design sensitive to regional, industrial, and source-specific heterogeneity.

The study contributes by (1) constructing a multidimensional, entropy-weighted index to measure DTA across provinces; (2) empirically demonstrating that DTA significantly suppresses ACEs with robust evidence; (3) revealing three partial mediating channels—production scale, industrial structure (largest mediator), and technological progress; (4) documenting pronounced heterogeneity across regions, industrial types, and emission sources; and (5) identifying a single-threshold nonlinearity in DTA’s impact on ACEs, with effects becoming significantly negative only above the threshold. Policy recommendations include: accelerate DTA—especially in central, western, and northeastern regions—via digital infrastructure, integration of digital tech with production, and digital literacy; encourage moderate farm-scale expansion and mechanized, intensive operations; support restructuring toward service-intensive, integrated agricultural systems and promote green inputs and renewable energy; and expand R&D and diffusion of green agricultural technologies through government–industry–university collaboration. Future work should refine DTA measurement, explore additional mediators/moderators, and extend analyses to finer spatial scales and longer periods, including cross-country comparisons.

- Measurement of DTA: Lack of unified definitions and comprehensive indicators for agricultural digital transformation; future work should develop more complete evaluation systems. - Mechanisms: The study focuses on scale, structure, and technology; other potential mediators/moderators (e.g., governance quality, market access, human capital) may shape the DTA–ACEs link and warrant investigation. - Spatial scope and granularity: Analysis is at the provincial level within China; cross-country comparisons and municipal/county-level data could improve precision and external validity. - Time span and data constraints: The 2015–2021 window reflects data availability and statistical adjustments; longer horizons and diverse data sources could enhance robustness and capture dynamic effects.