Ecological farm typology and comparison in China: An unsupervised machine learning approach

Promoting ecological farms is a critical initiative for greening agriculture and advancing eco-agricultural theory into practice. Although China integrated ecological farm construction into national plans post-2020, its development remains nascent, necessitating objective and quantitative methods to classify farm models and evaluate their performance. This study addresses this gap by applying unsupervised machine learning—specifically Partitioning Around Medoids (PAM) clustering—to 2022 declaration data from 678 national ecological farms. Our analysis identified three distinct typologies: National Varieties Fine Management Type (NVFMT, 40.27 %), characterized by small-scale, specialty-crop operations with minimal synthetic inputs; Diversified Business Type (DBT, 33.78 %), integrating vegetable production, agritourism, and high adoption of eco-measures; and Large-scale Traditional Type (LTT, 25.96 %), focused on grain cultivation with extensive land use and flood irrigation. Performance evaluation using life cycle assessment, total factor productivity (TFP), and revenue analysis revealed no significant economic differences across types. However, NVFMT and DBT exhibited 48–52 % lower greenhouse gas (GHG) emission intensity than LTT, attributable to reduced fertilizer use and diversified practices. DBT also achieved significantly higher TFP (1.18) compared to NVFMT (0.68) and LTT (0.72), linked to operational diversity and technological integration. As the first national-scale application of machine learning to ecological farm typology in China, this research provides an evidence-based framework to guide targeted policies for sustainable agricultural transformation. • Classify ecological farms using unsupervised machine learning methods. • Use 678 selection data of national-level ecological farms in China. • Ecological farms were divided into 3 clusters according to 38 feature indicators. • A comparative analysis was conducted on GHG emissions, productivity, and revenue.