Can digital economy truly improve agricultural ecological transformation? New insights from China

The paper addresses how digitalization intersects with green development in agriculture under China's dual-carbon goals. It highlights that traditional agricultural practices (e.g., heavy pesticide and fertilizer use, livestock waste) strain ecosystems and public health, prompting China to issue policies for green agricultural development and rural revitalization. The rapid expansion of China's digital economy (50.2 trillion yuan in 2022; 41.5% of GDP) offers tools—digital infrastructure, data elements, and digital technologies—to enhance resource allocation, reduce transaction costs, and modernize agriculture. The study focuses on low-carbon technological innovation as a key mechanism enabling sustainable agricultural transformation. Research questions: (1) Does the digital economy promote agricultural ecological transformation (AEE)? (2) Through what mechanism does low-carbon technological innovation mediate or condition this effect? (3) Does the impact of the digital economy on AEE differ at varying levels of low-carbon technological innovation? Hypotheses: (I) Digitalization affects green development in a nonlinear way. (II) The impact of the digital economy on AEE is constrained by threshold effects related to the level of low-carbon innovation.

The literature indicates the digital economy is a driver of rural revitalization and green transformation, reshaping agricultural production and management, improving efficiency, and balancing economic and environmental outcomes. Digital tools optimize inputs (e.g., water, fertilizer), reduce emissions, and foster modern ecological agriculture. Spillovers from digital technologies can strengthen industry chains and promote structural upgrading, enhancing land output and resource utilization. However, developing countries face low informatization and limited technological content in agriculture, making low-carbon innovation crucial. Studies show mixed effects: digitalization can raise green TFP in forestry and promote sustainability via operational optimization; green nanotechnology benefits agro-ecosystems; yet green technology's impact can be heterogeneous and ICTs can have both positive and negative environmental effects. Gaps identified: prior work focuses more on industry/manufacturing than agriculture; many models overlook endogeneity and dynamics; and few examine nonlinear threshold effects of low-carbon innovation on the digital economy–AEE nexus. This study contributes by focusing on agriculture, employing a dynamic threshold approach to address endogeneity/dynamics, and incorporating low-carbon innovation heterogeneity to reveal nonlinear impacts.

Measurement of agricultural ecological transformation (AEE): The authors define agricultural ecological transformation through shifts from extensive to intensive growth and reductions in high-carbon pollution. They construct agricultural green total factor productivity (GTFP) using a Super-SBM (slacks-based measure) model with undesirable outputs. Inputs: labor (primary industry practitioners scaled by sector output), land (crop sown area), irrigation (effective irrigation area), fertilizer use, pesticide use, agricultural film use, energy (agricultural diesel), and machinery (total power). Expected output: agricultural total output value. Undesirable output: composite index of agricultural non-point source pollution (fertilizer, pesticide, film). AEE is computed for 30 Chinese provinces, 2013–2021. Data sources: National Bureau of Statistics of China, CSMAR, CNRDS. Dynamic panel threshold model: To capture heterogeneity by low-carbon technological innovation (LCI) and address dynamics/endogeneity, the study applies a dynamic panel threshold regression, extending Hansen (2006) with System GMM. Dependent variable: AEE. Core regressor: digital economy (DE). Threshold variable: LCI. Controls: disaster rate (DR, disaster area/crop sown area), urbanization rate (URB, urban pop share), agricultural internal structure (AIS, cereal sown area share), rural human capital (HC, per capita education & entertainment expenditure of rural residents). Model includes lagged AEE terms, individual and time effects, and partitions DE’s effect across LCI regimes via indicator functions. DE index construction: A composite digital economy index is built via the entropy method from 14 indicators across three subsystems—digital infrastructure, digital industrialization, and industry digitalization (e.g., mobile switching capacity per capita, broadband users per capita, base stations per area, telecom traffic per capita, software revenue per capita, e-commerce volumes, digital inclusive finance index, domain names, enterprise webpages, IPv4 addresses). LCI measurement: number of green patent applications (CNRDS) proxies regional low-carbon innovation. Estimation and inference: Threshold significance is tested (bootstrap), thresholds estimated with confidence intervals, and coefficients obtained via System GMM. Robustness includes Hansen test for instrument validity and AR(1)/AR(2) serial correlation tests.

- Level and trend of AEE: The national average AEE (2013–2021) is 0.607, indicating a relatively low level but a steady upward trend. The average annual growth rate is 6.99% over 2013–2021. - Regional disparities: Significant heterogeneity exists. Guangdong exhibits the highest average level (0.866), while Jilin is the lowest (0.320); Guangdong’s level is about 2.71 times Jilin’s. Provinces like Guangdong, Shaanxi, and Henan lead; Qinghai, Shanxi, and Jilin lag. - Threshold effects of LCI on the DE→AEE relationship: Threshold tests reject triple-threshold but support single and double thresholds (5% significance). The study proceeds with a double-threshold specification. Reported threshold estimates partition LCI into three regimes: LCI ≤ 4.283 (weak), 4.283 < LCI ≤ 4.642 (moderate), and LCI > 4.642 (strong). - Regime-specific DE effects on AEE (Table 7): • LCI ≤ 4.283: DE has a significant negative effect on AEE (coef −0.536, p < 0.01). • 4.283 < LCI ≤ 4.642: DE has a significant positive effect (coef 0.866, p < 0.01). • LCI > 4.642: DE remains significantly positive (coef 0.368, p < 0.01). This pattern indicates a U-shaped nonlinear relationship—DE hinders AEE at low LCI but promotes it once LCI surpasses critical thresholds. - Effects of controls (Table 7): URB positive and significant (coef 1.700, p < 0.001); DR positive and significant (coef 0.069, p < 0.01); AIS positive but not significant; HC negative, not significant. Lagged AEE terms are positive and significant, indicating persistence. - Model diagnostics: Hansen test Prob > chi2 = 0.439 (supports instrument validity). AR(1) p = 0.007 and AR(2) p = 0.888 indicate appropriate error structure under System GMM.

Findings confirm that the digital economy’s impact on agricultural ecological transformation depends critically on the local level of low-carbon technological innovation, yielding a U-shaped relationship. At low LCI, technological readiness and data inputs are insufficient to complement digitalization, so DE investments can crowd out resources and fail to enhance green outcomes, impeding AEE. As LCI rises, knowledge spillovers and innovation-driven efficiency gains improve resource allocation, upgrade technologies and energy structures, and enable the digital economy to support green production, intensive management, and extended value chains. This fosters higher-quality growth, employment, income gains, and alignment of consumer green preferences with producer practices. The positive roles of urbanization and proactive disaster response suggest that market demand shifts and public investment in resilience can further facilitate AEE. Policy implications include: strengthening digital infrastructure and platforms; promoting collaborative digital and low-carbon innovation; improving interregional connectivity to diffuse digital benefits; and tailoring strategies to regional LCI levels so that low-LCI regions prioritize building innovation capacity while high-LCI regions deepen digital-agriculture integration.

- AEE in China remains at a relatively low level but has risen steadily from 2013 to 2021 (average 0.607; 6.99% average annual growth), with pronounced regional disparities. - The digital economy’s effect on AEE is nonlinear and conditioned by low-carbon technological innovation. Below critical LCI thresholds, DE inhibits AEE; beyond them, DE significantly promotes AEE, evidencing a U-shaped relationship. - Other factors: disaster incidence and urbanization rate positively influence AEE; agricultural internal structure is positive but not significant; rural human capital shows a mild inhibitory, non-significant effect. Policy recommendations: invest in rural digital infrastructure; foster digital innovation ecosystems and talent; enhance interregional digital connectivity; deeply integrate digital technologies with agricultural production to improve efficiency and reduce emissions; and calibrate strategies to regional LCI—build innovation capacity (e.g., green R&D, talent cultivation, innovation collaborations) in low-LCI regions while scaling digital applications in high-LCI regions. Future research: expand to micro/enterprise-level data as they become available; incorporate additional heterogeneity factors (e.g., environmental regulation, industrial integration) to refine the understanding of the DE–AEE mechanism.

- Data scope: analysis is based on provincial-level panel data due to current data availability; micro-level (enterprise or farm) dynamics are not captured. - Omitted heterogeneity: beyond low-carbon technological innovation, other regional factors (e.g., environmental regulation intensity, industrial integration, governance quality) may condition the DE–AEE relationship and warrant future inclusion. - Measurement constraints: proxies such as green patent applications for LCI and composite indices (DE, AEE) may not fully capture qualitative dimensions of innovation and ecological outcomes.