The Impact of the Digital Capability of Farmers on Low-Carbon Agriculture Adoption: A Machine Learning Perspective

Based on microscopic survey data of 10,057 kiwifruit growers from the Shaanxi, Guizhou, and Sichuan provinces in China, this study takes the initial and sustained adoption of low-carbon agricultural technology by farmers as dependent variables. A comprehensive evaluation system is constructed based on the digital capability of farmers. The system covers four dimensions, namely, digital access, digital acquisition, digital processing, and digital sharing. Multiple empirical models, including Lasso regression, extreme gradient boosting, random forest, and back propagation neural network are adopted to compare the fitting and predictive performance of different approaches. The model comparison results demonstrate that the random forest model presents the optimal fitting performance among all alternative specifications. To identify the core influencing factors and explore the underlying mechanisms, this study employs the SHAP method to quantitatively evaluate the marginal contribution of each variable. The empirical findings reveal that the digital capability of farmers significantly and positively promote both initial and sustained adoption of low-carbon agricultural technology with distinct dimensional heterogeneity. Specifically, digital access and digital acquisition capability act as core driving factors for the initial adoption of technologies by farmers, while digital processing and digital sharing capability play a decisive role in facilitating sustained low-carbon technology application. In addition, farmers’ education level, health status, household income, cooperative participation, and household labour scale are crucial characteristic variables affecting the behaviours of farmers in the adoption of low-carbon technology across different stages. Accordingly, this study proposes targeted policy implications for low-carbon agricultural development. On the basis of promoting digital rural construction and improving the systems of digital skill training and digital agricultural technical services, it is essential to strengthen dynamic tracking surveys of farmers, accurately identify their technology adoption stages, and implement differentiated supporting policies. These targeted measures can effectively consolidate and assess the low-carbon production behaviours of farmers and promote the long-term and sustainable development of low-carbon agriculture.