Abstract: Amid escalating climate change threats, improving the assessment of grain yield variability and its determinants is imperative. However, existing literature has paid insufficient attention to the spatial scale effects of grain yield variability and its socioeconomic drivers. This study examines grain yield variability in China from a multiscale perspective and investigates its influencing mechanisms and regional heterogeneity by developing an integrated theoretical framework and applying LightGBM, path analysis, and decision tree models. The results show that: 1) although China’s overall grain yield variability declined between 2000 and 2022, finer spatial scales reveal increasingly pronounced regional heterogeneity and intensified local yield fluctuations. 2) Most high-yield-variability areas overlap with key grain yield growth regions, and a strong positive correlation is observed between yield variability and the grain yield growth rate in regions with positive growth. This indicates that spatial shifts in grain production are detrimental to yield stability. 3) Regional economic development exerts a U-shaped effect on grain yield variability. 4) In key grain yield growth regions such as Northeast China, agricultural labor outmigration and inadequate agricultural infrastructure—both associated with economic underdevelopment—combined with frequent climate disasters, a low multiple cropping index, and low crop diversity, collectively contribute to high grain yield variability. The study proposes a ‘U-shaped spatial continuum’ model to generalize the economic-geographical patterns of the relationship between yield variability and spatial shifts in grain production. The study demonstrates the academic value of assessing grain yield variability from a multiscale perspective and provides theoretical insights for managing yield variability from socioeconomic and regional heterogeneity perspectives.