A parcel-level cropland dataset with field shape indicators for Xizang (Tibet) from VHR remote sensing

High-resolution cropland data are essential for regional agricultural research, particularly in Xizang, China, where fragmented landscapes and complex cropping systems require accurate monitoring of crop distribution, yield estimation, and land management planning. This study presents the first parcel-level cropland dataset for Xizang, generated by a deep learning-based field extraction model and high-resolution Google Earth imagery. Field validation of the dataset with 155 UAV samples showed 90% area matching accuracy, and cross-comparison with two existing cropland products of Xizang also showed strong linear correlations (R² = 0.87 and 0.90). The dataset contains 1,072,110 individual parcels, totaling approximately 290,721.35 ha, mainly concentrated in the Three River Region (28°-31°N, 87°-92°E), with larger parcels (>0.5 ha) constituting the majority of cultivated area. Additionally, the field shape indicators are further calculated and provided, including main direction, regularity, and contiguity. This dataset offers critical insights into Xizang's fine-scale cropland patterns and can serve as a foundation for parcel-level agricultural monitoring, land-use planning, and policy formulation.