Exploring the scale-specific effects of nature and human factors on perennial rice yield using a multi-scale modeling framework

Rice is consumed as a staple food by approximately 50% of the global population. As the world’s largest producer of rice, China’s consistent and reliable production levels have a substantial impact on global food security. Yield formation mechanisms exhibit significant spatial heterogeneity due to diverse environments, requiring analysis of multi-factor interactions and scale effects. This study employs an integrated Optimal Parameters-based Geodetector (OPGD) and Multi-scale Geographically Weighted Regression (MGWR) framework to identify key drivers and quantify their spatial scales across China’s perennial rice systems. Results show that hydrothermal conditions, distance to geological hazards (DistDisas), potassium fertilizer application (M2), and soil pH are dominant factors, with all pairs exhibiting nonlinear synergies. MGWR further revealed distinct operational scales: M2 and distance from residential areas acted as large-scale, universally consistent factors. Soil pH, GDP, and slope length and steepness operated at medium scales with regional variation. Soil bulk density, pesticide usage, DistDisas, fertilizer usage, and climate conditions were small-scale factors with high spatial heterogeneity. These findings underscore that yield formation emerges from multi-scale interactions between natural and human factors, affirming the necessity of scale-explicit approaches for accurate yield prediction and targeted agricultural management.