Answers are grounded in this paper and cite their sources.
Demonstrates improved cropland mapping accuracy in Northeast China, suggesting a new evaluate method for agricultural monitoring.
Cropland parcels (CPs) constitute the fundamental physical units of agricultural production and management. However, their regional-scale mapping from meter-level very-high-resolution (VHR) imagery remains challenging due to computational constraints, tile-induced boundary fragmentation, and semantic confusion associated with single-date observations. To address these challenges, we develop a cloud-based method for large-scale cropland parcel mapping from meter-level VHR imagery. This method consists of geospatial tiling, HRNet-based boundary extraction from VHR imagery, and parcel-level post-processing, including morphology-based parcel merging and Sentinel-2 phenology-based filtering. This framework enables efficient large-area processing while preserving parcel continuity and boundary integrity. Using this framework, we generate NECCPB-1, a high-resolution cropland parcel boundary dataset for the black soil region of Northeast China (NEC), delineating 32.4 million cropland parcels (≥900 m²) across 40 prefecture-level cities. Independent parcel-based validation indicates a mean extraction precision (EP) of 0.85, a mean area deviation CompD of 0.0042, and a mean intersection-over-union (IoU) of 0.90. In tile-overlap regions, the proposed parcel merging algorithm effectively improves parcel completeness and boundary continuity, achieving an IoU of 0.98. Phenology-based filtering further removes approximately 4.4 million non-cropland parcels (about 12%), reducing confusion with spectrally similar land covers. Cross-comparative analyses show that NECCPB-1 maintains stable accuracy across mountainous and non-mountainous regions, while its mapped cropland area exhibits a mean relative difference of −3.6% compared with the Third National Land Survey. Relative to seven widely used cropland products, NECCPB-1 achieves higher parcel completeness (IR = 0.94) and classification accuracy (precision = 0.93, recall = 0.91, F1-score = 0.92). Beyond mapping accuracy, NECCPB-1 reveals a clear south-to-north gradient in cropland parcel number and area across NEC. Overall, the proposed framework provides a transferable solution for large-scale parcel-level cropland mapping from VHR imagery and offers a stable spatial baseline for boundary-constrained agricultural monitoring and analysis.
No takes yet. Share an insight, caveat, or question.
Zheng et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: