Abstract Analyzing consistency of different land-cover data is significant to reasonably select land-cover data for regional development and resource survey. Existing consistency analysis of different datasets mainly focused on the phenomena of spatial consistency regional distribution or accuracy comparison to provide guidelines for choosing the land-cover data. However, few studies focused on the hidden inconsistency distribution rules of different datasets, which can provide guidelines not only for users to properly choose them but also for producers to improve their mapping strategies. Here, we zoned the Sindh province of Pakistan by the Terrestrial Ecoregions of the World as a case to analyze the inconsistency patterns of the following three datasets: GlobeLand30, FROM-GLC, and regional land cover (RLC). We found that the inconsistency of the three datasets was relatively low in areas having a dominant type and also showing homogeneity characteristics in remote sensing images. For example, cropland of the three datasets in the ecological zoning of Northwestern thorn scrub forests showed high consistency. In contrast, the inconsistency was high in areas with strong heterogeneity. For example, in the southeast of the Thar desert ecological zone where cropland, grassland, shrubland, and bareland were interleaved and the surface cover complexity was relatively high, the inconsistency of the three datasets was relatively high. We also found that definitions of some types in different classification systems are different, which also increased the inconsistency. For example, the definitions of grassland and bareland in GlobeLand30 and RLC were different, which seriously affects the consistency of these datasets. Hence, producers can use the existing land-cover products as reference in ecological zones with dominant types and strong homogeneity. It is necessary to pay more attention on ecological zoning with complex land types and strong heterogeneity. An effective way is standardizing the definitions of complex land types, such as forest, shrubland, and grassland in these areas.

中文翻译：

---

生态区划视角下不同遥感土地覆盖数据的不一致分布格局

摘要 分析不同土地覆盖数据的一致性对于合理选择土地覆盖数据进行区域开发和资源调查具有重要意义。现有不同数据集的一致性分析主要侧重于空间一致性区域分布或精度比较的现象，为选择土地覆盖数据提供指导。然而，很少有研究关注不同数据集的隐藏不一致分布规则，这不仅可以为用户正确选择它们，也可以为生产者改进其映射策略提供指导。在这里，我们以世界陆地生态区对巴基斯坦信德省进行分区，以分析以下三个数据集的不一致模式：GlobeLand30、FROM-GLC 和区域土地覆盖 (RLC)。我们发现三个数据集的不一致性在具有主导类型的区域相对较低，并且在遥感图像中也表现出同质性特征。例如，西北荆棘灌丛林生态区划中三个数据集的农田表现出高度的一致性。相比之下，异质性强的地区不一致性较高。例如，在农田、草地、灌丛、裸地交错且地表覆盖复杂度较高的塔尔沙漠生态区东南部，三个数据集的不一致性较高。我们还发现，不同分类体系中某些类型的定义不同，这也增加了不一致性。例如GlobeLand30和RLC中草地和裸地的定义不同，这严重影响了这些数据集的一致性。因此，生产者可以在类型优势和同质性强的生态区中使用现有的土地覆盖产品作为参考。对土地类型复杂、异质性强的生态区划要给予更多关注。一个有效的方法是规范复杂土地类型的定义，如这些地区的森林、灌丛和草地。