Reducing farmland abandonment is essential for protecting farmland resources and ensuring national food security. The driving mechanisms of farmland abandonment in China at the 1-km grid scale remain insufficiently understood. In this study, a composite theoretical framework integrating natural factors, socioeconomic factors, and cultivation conditions was developed to explain farmland abandonment at the grid scale. Fifteen driving variables, together with a 1-km gridded dataset of farmland abandonment ratio (the proportion of abandoned farmland relative to the total farmland area), were modelled using random forest models. The SHAPley additive explanations algorithm was applied to quantify both the individual effects and interaction effects. The results revealed that in 2019, the overall farmland abandonment ratio in China was 4.49%, with higher rates observed in mountainous areas (6.26%) than in the plains (2.22%). Stable hotspots of farmland abandonment emerged in the agro-pastoral transition zone between the northern arid and semiarid regions and the Loess Plateau, in the southeastern coastal region, in the southwestern mountainous region, and around the Beijing-Tianjin-Hebei urban agglomeration. Distinct driving mechanisms of farmland abandonment were observed between plains and mountainous areas. Increased farmland density mitigated abandonment risk across both terrains by strengthening large-scale farming and organizational structure. In plains, farmland abandonment was governed mainly by labour, location, and organizational structure. Engineering improvements and institutional support weakened the effects of climatic and topographic factors. The influence of ageing was both stage-specific and bidirectional. In mountainous areas, farmland abandonment was driven mainly by strong coupling among climatic conditions, topography, and service accessibility. The effects of climatic conditions were amplified by their interactions with slope and altitude, resulting in the formation of a chain that extends from natural stress to mechanization constraints and ultimately to cost accumulation. The dominant driving factors of farmland abandonment varied regionally. Natural factors dominated in western high-altitude, cold, and arid areas. Socioeconomic factors dominated in the plains and rapidly urbanizing regions. Cultivation conditions were dominant in hilly regions and areas characterized by fragmented farmland. This study enhances the understanding of farmland abandonment from a grid-scale perspective and provides guidance for regionalized governance strategies.