An improved global land cover mapping in 2015 with 30 m resolution (GLC-2015) based on a multisource product-fusion approach
An improved global land cover mapping in 2015 with 30 m resolution (GLC-2015) based on a multisource product-fusion approach
复制标题
DOI：
10.5194/essd-15-2347-2023
复制
发表时间：
2023-06
影响因子：
11.4
通讯作者：
Bingjie Li;Xiaocong Xu;Xiaoping Liu;Q. Shi;Haoming Zhuang;Yaotong Cai;Da He
中科院分区：
文献类型：
--
作者：
Bingjie Li;Xiaocong Xu;Xiaoping Liu;Q. Shi;Haoming Zhuang;Yaotong Cai;Da He
抽象的。高空间分辨率的全球土地覆盖信息是研究地球系统地球化学循环和全球气候变化的基础数据。虽然有几个30 m分辨率的GLC产品，但它们之间存在相当大的不一致性，特别是在碎片化区域和过渡区，这给各种应用任务带来了很大的不确定性。在本文中，我们开发了一个改进的全球土地覆盖地图在2015年与30米分辨率（GLC-2015）通过融合多个现有的土地覆盖（LC）产品的基础上，Dempster-Shafer证据理论（DSET）。首先，我们使用超过160 000个全球点为基础的样本，以本地评估的可靠性输入产品的每个土地覆盖类在每个4 <$× 4 <$地理网格的基本概率分配（BPA）函数的建立。然后，Dempster的组合规则被用于每30米像素，从所有候选地图中获得每个可能的土地覆盖类的组合概率质量。最后，每个像元确定一个土地覆盖类的决策规则的基础上。通过这种融合过程，每个像素预计将被分配的土地覆盖类，有助于实现更高的精度。我们使用34711个全局点样本和201个全局块样本分别评估了我们的产品。结果表明，与现有的30 m GLC地图相比，GLC-2015地图在全球、大陆和生态区域上实现了最高的制图性能，总体准确率为79.5%（83.6%），与基于点（基于块）的验证样本的kappa系数为0.757（0.566）。此外，我们发现GLC-2015地图在不一致的区域表现出显著的优势，在中度不一致的区域准确度提高了19.3%-28.0%，在高度不一致的区域提高了27.5%-29.7%。希望该改进后的GLC-2015产品能够应用于减少全球环境变化研究、生态系统服务评估和灾害损害评估中的不确定性。本研究中开发的GLC-2015图谱可在https://doi.org/10.6084/m9.figshare.22358143.v2获得（Li et al.，2023年）。
Abstract. Global land cover (GLC) information with fine spatial resolution is a fundamental data input for studies on biogeochemical cycles of the Earth system and global climate change. Although there are several public GLC products with 30 m resolution, considerable inconsistencies were found among them, especially in fragmented regions and transition zones, which brings great uncertainties to various application tasks. In this paper, we developed an improved global land cover map in 2015 with 30 m resolution (GLC-2015) by fusing multiple existing land cover (LC) products based on the Dempster–Shafer theory of evidence (DSET). Firstly, we used more than 160 000 global point-based samples to locally evaluate the reliability of the input products for each land cover class within each 4∘ × 4∘ geographical grid for the establishment of the basic probability assignment (BPA) function. Then, Dempster's rule of combination was used for each 30 m pixel to derive the combined probability mass of each possible land cover class from all the candidate maps. Finally, each pixel was determined with a land cover class based on a decision rule. Through this fusing process, each pixel is expected to be assigned the land cover class that contributes to achieving a higher accuracy. We assessed our product separately with 34 711 global point-based samples and 201 global patch-based samples. Results show that the GLC-2015 map achieved the highest mapping performance globally, continentally, and ecoregionally compared with the existing 30 m GLC maps, with an overall accuracy of 79.5 % (83.6 %) and a kappa coefficient of 0.757 (0.566) against the point-based (patch-based) validation samples. Additionally, we found that the GLC-2015 map showed substantial outperformance in the areas of inconsistency, with an accuracy improvement of 19.3 %–28.0 % in areas of moderate inconsistency and 27.5 %–29.7 % in areas of high inconsistency. Hopefully, this improved GLC-2015 product can be applied to reduce uncertainties in the research on global environmental changes, ecosystem service assessments, and hazard damage evaluations. The GLC-2015 map developed in this study is available at https://doi.org/10.6084/m9.figshare.22358143.v2 (Li et al., 2023).