中国高分辨率耕地融合产品（CCropLand30）

准确了解耕地的范围和动态变化对于解决不断增长的粮食需求与减少环境足迹这一双重挑战至关重要。尽管目前已经存在许多全球和区域耕地图，但仍存在显著的不确定性和差异。我们通过将目前较新的遥感土地利用和覆盖产品（包括 GlobeLand30、GLAD、CLUD、CLCD 和 CACD）与最新的国土调查成果（"三调“）相融合，开发了中国的高分辨率（30m）耕地产品（CCropLand30）。我们提出了一种经济高效的数据融合方法，即多数投票-模糊一致得分（MV-FAS）方法。使用目视解释和第三方样本以及县级土地调查数据，我们对CCropLand30和输入的耕地图进行了评估。结果显示：CCropLand30的总体精度、Kappa系数和F1分数分别为0.88、0.76和0.86。与输入的耕地产品相比，CCropLand30与参考点更为一致，使总体精度提高了10-30%。此外，CCropLand30与土地调查数据的空间一致性更强（平均提高了25%）。

Accurately understanding the extent and dynamic changes of cropland is critical to addressing the dual challenge of meeting growing food demand and reducing environmental footprints. Although numerous global and regional cropland maps currently exist, significant uncertainties and discrepancies remain. We developed a high-resolution (30 m) cropland product for China, named CCropLand30, by integrating state-of-the-art remote sensing land use and cover products (including GlobeLand30, GLAD, CLUD, CLCD, and CACD) with the latest national land survey results, specifically the Third National Land Survey of China, commonly referred to as the "Three Surveys". We propose a cost-effective data fusion method, namely the Majority Voting-Fuzzy Agreement Score (MV-FAS) approach. We evaluated CCropLand30 and the input cropland products using visual interpretation, third-party samples, and county-level land survey data. The results demonstrate that CCropLand30 achieves an overall accuracy of 0.88, a Kappa coefficient of 0.76, and an F1-score of 0.86, respectively. Compared with the input cropland products, CCropLand30 shows greater consistency with reference points, leading to an improvement of 10–30% in overall accuracy. Furthermore, CCropLand30 exhibits stronger spatial consistency with land survey data, with an average improvement of 25%.

搜集汇总

背景与挑战

背景概述

中国高分辨率耕地融合产品（CCropLand30）是一个30米空间分辨率的中国耕地地图数据集，覆盖2000年至2020年间的五个年份。它通过融合多种遥感产品和国土调查数据，采用多数投票-模糊一致得分方法生成，显著提高了耕地识别的准确性，总体精度达0.88，比输入产品提升10-30%。该数据集以开放获取方式提供，适用于土地利用、粮食安全和环境研究。