Croplayer is a 2020 China global high-precision (2-meter resolution) crop distribution dataset. It aims to solve the problems of large statistical deviations in existing mainstream data (such as CLCD, WorldCover, SinoLC, etc.) in provincial area, missing details of complex terrain farmland (such as southern terraces and dam fields), and insufficient identification of fragmented farmland (such as northern non-agricultural interference areas).

The data papers are now published in the discussion section of Earth System Science Data(ESSD) in pre-printed form and are under the CC BY 4.0 license agreement.

By fusing high-resolution images from Mapbox and Google, the dataset first uses the ResNet model to conduct block-level image quality assessment (IQA) to screen high-quality data and correct missing metadata; and then builds an active learning framework based on Mask2Former semantic segmentation and XGBoost error assessment to improve the accuracy of farmland boundary segmentation by iteratively optimizing samples; finally innovatively integrates the four types of characteristics: geographical features, image quality, regional attributes and consistency, and uses a weighted fusion strategy to generate a unified farmland layer.

Croplayer has achieved a mapping accuracy of 88.73%, and has reached a strict standard of error of ≤±10% with official data in farmland area statistics of 30 provincial administrative regions across the country (significantly better than the performance of only 1-9 provinces in the existing dataset), providing high-precision and high-statistical consistency spatial data support for crop yield estimation, agricultural structure optimization and food security early warning.