Decoding spatial consistency of multi-Source land cover products in China: Insights from heterogeneous landscapes
High-resolution land cover (LC) data are essential for ecological monitoring and resource management, especially in heterogeneous landscapes containing diverse LC types. With the growing of available LC products, a comprehensive evaluation of their classification accuracy and spatial consistency is...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-05-01
|
| Series: | International Journal of Applied Earth Observations and Geoinformation |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1569843225001761 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | High-resolution land cover (LC) data are essential for ecological monitoring and resource management, especially in heterogeneous landscapes containing diverse LC types. With the growing of available LC products, a comprehensive evaluation of their classification accuracy and spatial consistency is important for users’ selection and application. In this study, we compared eight widely used LC products in China, including ESA World Cover (ESA20), ESRI GLC10 (ESRI17, ESRI20), FROM-GLC10 (FROM-GLC17), CLCD (CLCD20), GlobeLand30 (GLB20), GLC_FCS30 (GLC_FCS20), and GLC_FCSD30 (GLC_FCSD20), to examine their performances at both national and regional scales. We employed pixel-wise overlay analysis, visually interpreted validation samples, and classical landscape metrics to assess overall consistency and classification accuracy. The results show that the 30m_combination (CLCD20, GLB20, GLC_FCS20, and GLC_FCSD20) exhibits higher overall consistency at the national scale, with perfect consistency exceeding 60 %. In contrast, the 10m_combination (ESA20, ESRI17, ESRI20, and FROM_GLC17) captures finer regional details but displays greater inconsistencies in central and western regions. ESA20 achieves the highest overall accuracy (OA) at 88.5 % (CI: 88.44 %–88.56 %), while FROM_GLC17 records the lowest at 82.79 % (CI: 82.73 %–82.85 %). Cropland, forest, water, and snow/ice demonstrate higher consistency and classification accuracy (F1-scores > 80 %), whereas wetland, grassland, impervious surfaces, and bare land underperform in fragmented regions. Furthermore, spatial consistency is strongly associated with landscape metrics such as the aggregation index (AI) and contagion (CONTAG), which enhance consistency in large, contiguous patches (e.g., Northeast China Plain). Conversely, edge density (ED) and patch density (PD) show negative associations with consistency, highlighting persistent mapping challenges in fragmented regions (e.g., Yunnan-Guizhou Plateau and Qinghai-Tibet Plateau). These findings offer actionable insights for improving LC mapping in complex terrains and underscore the critical role of landscape metrics in advancing ecological monitoring and resource management. |
|---|---|
| ISSN: | 1569-8432 |