Under carbon neutrality targets, understanding the spatial heterogeneity of agricultural carbon mitigation and its linkage with resource endowments and economic constraints is of critical importance. This study constructs baseline and mitigation scenarios and applies global and local spatial autocorrelation methods, including Moran's I, Local Indicators of Spatial Association (LISA), and spatial difference Moran's I, to investigate the spatial patterns and dynamic evolution of cropland distribution, mitigation cost, and emission reduction potential in China's agricultural system. The results show that: (1) Key variables exhibit significant spatial clustering, with global Moran's I ranging from 0.218 to 0.363, indicating a stable positive spatial dependence overall. This positive spatial correlation implies that high−value regions tend to be adjacent to other high−value regions, reflecting the spatial aggregation of areas with similar resource endowments; (2) Local spatial patterns show pronounced regional heterogeneity, with high–high clusters of cropland decreasing from about 8.8% to 2.9%, while low−cost clusters shrink from approximately 14.7% under the baseline scenario to about 5.9% under the mitigation scenario, indicating an overall optimization of the spatial structure; (3) Significant spatial coupling exists between variables, as the bivariate Moran's I between cropland and net cost decreases from approximately 0.431 to 0.192, suggesting that policy interventions weaken the spatial association between resource endowment and cost; (4) Spatial difference analysis indicates that the spatial dependence of cropland change becomes insignificant (from 0.385), whereas that of net cost change remains significantly positive. Overall, the mitigation scenario improves the spatial configuration of the agricultural system to some extent, although regional disparities and structural imbalances persist.