Mapping cropland use changes in the Guanzhong Plain of China based on a two-level classification framework

This is the authors' abstract. We don't add key points for this paper.

Cropland use changes significantly impact food security and have been extensively studied worldwide, including in China. Accurately mapping crop-planting changes in regions with complex cropping patterns remains a challenge due to the limitations of existing classification methods. In this study, we propose a two-stage classification framework to generate 30-m crop maps for monitoring cropland use changes in the Guanzhong Plain, China. This framework integrates sample migration and cropland mask extraction, with initial grain and non-grain crop masks generated from the Random Forest classification. These masks are then used in subsequent Random Forest and Decision Tree classifications to delineate three types of grain crops and five types of non-grain crops. Standard deviation ellipse and gravity centre analysis is used to study the spatio-temporal changes of cropland use. The results showed high classification performance, with overall accuracy (OA) and kappa coefficients exceeding 0.8 and 0.7, respectively. A complex and mixed agricultural system has been formed in Guanzhong plain with traditional agriculture along the central line of the region and an expanding agricultural diversification of non-grain crop cultivation on the marginal cropland. The analysis revealed significant shifts in the spatial distribution of crops, with grain crops expanding from the central and western parts to the entire Guanzhong Plain and clustering in the central part, while non-grain crops have expanded from the east to the central and western parts. This two-stage classification framework offers valuable insights into cropland use changes in the Guanzhong Plain and provides essential information for policy -making and land use planning.