Develop a modelling framework to identify and optimize the dominant factors that limit cropland productivity

Context: Analysing the influence mechanism of farming conditions (soil properties and agricultural infrastructure) on cropland productivity is a key prerequisite for increasing yields in low- to medium-quality land. Objective: We proposed a modelling framework to identify key farming condition factors that limit cropland productivity and analyse the numerical ranges within which they exert a dominant influence. The responses of cropland productivity to changes in dominant farming conditions were simulated. Methods: The framework consisted of processed long-term sequence earth observation data and random forest model. By filtering high-density cropland samples and increasing crop identification accuracy, gross primary production (GPP) was proven to be an appropriate indicator of cropland productivity in scenarios lacking high-precision crop yield data. Results and conclusions: Farming conditions explained >60% of the spatial differences in the rice GPP and > 65% of those in the wheat GPP. Soil texture and pH were key factors limiting rice and wheat GPP. A decrease in sand content and a corresponding increase in clay content increased rice GPP. Soil nitrogen supply rapidly decreased when clay content approached 20%, decreasing rice GPP. Climate conditions influenced the preference of wheat for soil water retention and drainage-permeability, resulting in an increase wheat GPP in northern and decrease in southern regions with raising clay content. The annual total GPP of rice and wheat increased by up to 6.8% through adjusting clay and sand contents and increasing mean field size. In the northwestern and southeastern regions, small adjustments (−5% … +5%) to clay and sand contents led to annual GPP increases of >600 kg·C·ha−1 for rice and > 800 kg·C·ha−1 for paddy-wheat rotations. Significance: The framework can provide support to optimize farming conditions in low- to medium-yield cropland renovation projects. © 2026 Elsevier Ltd