Unpacking spatially explicit cropland non-agriculturalization in China based on satellite remote sensing data: A 30-year analysis of spatial patterns and conversion types

Cropland non-agriculturalization, encompassing permanent loss and transitions to non-grain uses, drives global food insecurity. However, accurately mapping these diverse pathways across large spatiotemporal scales remains methodologically elusive. To address this, we developed a novel approach integrating multi-index temporal-spectral segmentation (LandTrendr) with Random Forest (RF) classification to map three primary cropland conversion types in China (1991–2020): conversion to non-grain uses, waterbody, and built-up land. Crucially, to explicitly overcome phenological confounding effects, we innovatively utilized the Coefficient of Variation (CV) of vegetation indices to suppress seasonal noise and extract robust inter-annual cultivation features. By employing the LandTrendr algorithm to segment these CV metrics and other multi-spectral indices, we detected the features of distinct disturbances, which subsequently served as predictors within an RF framework to map the different cropland non-agriculturalization types. This methodology yielded an 87.5 ± 1.0% overall accuracy. China's non-agriculturalized cropland reached 426,388 km2, comprising conversions to non-grain (312,771 km2), built-up (92,462 km2), and waterbody (21,155 km2). Spatially, non-grain conversions concentrated in the Northeast China Plain and northern arid/semi-arid regions. Waterbody transitions aggregated in the Middle and Lower Yangtze Plain, while built-up land expansion dominated major coastal and urban agglomerations (Beijing-Tianjin-Hebei, Yangtze/Pearl River Deltas). Temporally, non-grain transitions exhibited early fluctuating upward trajectories, whereas waterbody and built-up conversions peaked in 2001 and 2013, respectively. Ultimately, this research establishes a robust multi-index framework capable of untangling concurrent cropland non-agriculturalization types over vast extents. By accurately attributing spatially explicit transitions, our methodology offers a reliable mechanism for continuous monitoring and a strong empirical basis for global sustainable agricultural policies.