China’s digital villages boost grain and cut emissions but spare carbon sinks

Rural digitalization in China has delivered a lopsided payoff in its first years: counties that pushed hardest on the national Digital Village Development strategy grew more grain and cut agricultural carbon emissions, yet showed no measurable improvement in crop-biomass carbon uptake or the net carbon footprint of farming. That is the central finding of a new study published in Agricultural and Food Economics, which links county-level policy data with household survey records to test whether going digital can serve food security and climate goals at the same time. The answer, at least in the short run, is only partially.

The research team, led by Weilong Gao with Qian Zhang and Ying Peng, assembled an unusually layered dataset covering 989 counties and 1,835 farm households. At the county level they used the Digital Village Development index, jointly released by the Institute of New Rural Development at Peking University and the Ali Research Institute, which scores rural digital infrastructure, the digital economy, digital governance and digital life on a single composite scale. At the household level they built a digital adoption index from the 2019 China Rural Revitalization Survey, a national panel covering ten provinces, fifty counties, three hundred villages and more than 3,800 households, capturing how deeply farmers actually use digital tools rather than simply whether they have internet access. The two-level design was intended to bridge a common blind spot in digitalization research: the gap between what governments implement and what farmers adopt.

Food security was measured as county-level annual total grain output, covering cereals, legumes and tubers, drawn from the National Bureau of Statistics and local yearbooks. The environmental side was more elaborate. Carbon emissions were estimated in tons of CO2-equivalent from six sources, including fertilizer and pesticide production, agricultural machinery use, rice cultivation, direct and indirect farmland N2O emissions, and open-field burning of crop residue, using a random forest downscaling approach that combines remote sensing with statistical inputs to produce high-resolution county-level estimates. Carbon uptake was derived from the net primary productivity of crops, converting economic yields into total biomass carbon through harvest indices and carbon assimilation coefficients. The net carbon footprint then combined the two, expressing uptake relative to emissions. Crucially, the carbon uptake measure reflects crop biomass only, not long-term soil carbon storage, a distinction the authors stress repeatedly.

Using two-way fixed effects models that compare changes within counties between 2018 and 2019, the baseline results were clear in one direction and null in another. A one standard deviation increase in the county digitalization index was associated with a 0.050 standard deviation reduction in agricultural carbon emissions and a 0.009 standard deviation increase in grain output. The production gain is statistically significant but modest in size. The emission reduction is more substantial. Neither crop-biomass carbon uptake nor the net carbon footprint moved significantly, meaning the emission cuts were not matched by proportional gains in the carbon that crops pull out of the atmosphere.

Because the baseline model cannot rule out reverse causality, the possibility that better-performing agricultural regions simply digitalize faster, the team turned to instrumental variable estimation. They built two Bartik-style instruments by interacting fixed geographic traits with a time-varying national digital finance index: one based on county topographic relief, since rugged terrain raises the cost of laying fiber and building signal towers, and another on the distance from each county to Hangzhou, headquarters of Alibaba Group and a hub of Chinese digital commerce. The two-stage least squares results echoed the baseline patterns, with digitalization still negatively associated with emissions and positively associated with grain output. But the instrumented coefficients were substantially larger, and a Hansen J test on an overidentified model rejected instrument exogeneity in some specifications, so the authors present the IV estimates as evidence that mitigates endogeneity concerns rather than as definitive causal estimates.

The search for mechanisms produced the study’s most revealing results. Changes in cultivated land structure, such as total farmland area, irrigated land or facility agriculture, showed no significant response to digitalization, confirming that these physical endowments are sticky in the short run. Cropping systems, by contrast, moved decisively. Counties with higher digitalization scores expanded double-cropping area significantly, while single- and triple-cropping areas were unaffected. Among the sub-dimensions, digital governance showed a significant negative effect on single-cropping area and a significant positive effect on double-cropping area, suggesting the shift reflects enhanced governance capacity, through e-government platforms and land-use monitoring, rather than spontaneous market forces. The authors note that this intensification supports grain output but may carry a carbon cost, since more frequent cropping shortens fallow periods and increases soil disturbance, likely offsetting any gains in biomass uptake.

The household-level analysis added a behavioral layer. Using the average digital adoption among other households in the same village as an instrument, the two-stage estimates linked household digital adoption to higher yield per mu, lower household labor input per mu, and a marginally significant reduction in pesticide cost per mu. The OLS regressions showed only a weak positive association with yield and nothing else. Because the village-level instrument may capture peer effects, shared infrastructure or local market conditions that directly affect farm outcomes, the authors treat these estimates cautiously as complementary evidence. Still, the pattern is consistent with digital tools helping farmers time operations, identify pests and diseases, and trim redundant inputs, which aligns with the county-level emission reductions.

Disaggregating the composite index exposed sharp functional differences among its components. Digital infrastructure, the hard layer of network access, digital financial coverage and commercial facility reach, was significantly and positively associated with both grain output and carbon uptake, likely reflecting productivity gains from better market connectivity and factor acquisition. Digital governance and digital life, covering online government services and everyday information access, were strongly associated with emission reductions, plausibly through improved environmental policy transmission and broader access to green production knowledge. The digital economy dimension showed no statistically significant effect on any outcome. This heterogeneity suggests that different components of the digital village strategy operate through distinct pathways, and that treating digitalization as a single lever obscures where its real leverage lies.

Robustness checks reinforced the core pattern. When outcomes were converted to per-unit-area indicators, the positive association with grain yield per unit area held, while per-unit effects on carbon uptake, emissions and the net carbon footprint were insignificant, indicating that the emission reduction stems more from changes in cropping structure or sown area than from lower emission intensity per hectare. Adding total power of agricultural machinery as a control left the emission and output results intact, and 1 percent winsorization of all variables changed nothing of substance. The authors also acknowledge limitations: the two-year window captures only short-run dynamics, the carbon uptake measure excludes soil carbon, and county-level data cannot reveal whether intensification came with conservation tillage or deeper ploughing, practices that would determine the ultimate carbon balance.

The broader implication is that digitalization is an intermediary force, not a sufficient condition for agricultural sustainability. It can sharpen information, coordination and input management quickly, supporting production and trimming emissions, but whether it improves the overall carbon balance depends on land-use patterns, cropping arrangements and long-term ecological management that digital tools alone do not supply. The authors propose coupling intensification with conservation incentives, such as directing digital subsidies in double-cropping counties to farmers who also practice no-till and straw retention, with soil carbon changes monitored through remote sensing. For now, the digital dividend in Chinese agriculture is real but asymmetric: it feeds people and trims emissions, while the deeper carbon accounting remains unresolved.

Subject of Research: Short-run effects of China's Digital Village Development strategy on grain output, agricultural carbon emissions, crop-biomass carbon uptake and net carbon footprint across 989 counties and 1,835 farm households.

Article Title: Digital transformation at a cost? Asymmetric effects on food security and the agricultural carbon cycle: evidence from China’s digital village strategy

Article References: Gao, W., Zhang, Q., & Peng, Y. (2026). Digital transformation at a cost? Asymmetric effects on food security and the agricultural carbon cycle: evidence from China’s digital village strategy. Agricultural and Food Economics, 14(1), Article 61. https://doi.org/10.1186/s40100-026-00503-9

Image Credits: AI Generated

DOI: 10.1186/s40100-026-00503-9

Keywords: digital villages, food security, carbon emissions, carbon uptake, net carbon footprint, double cropping, China, rural digitalization, grain output, pesticide use, land intensification, agricultural economics