the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
GLC_FCS10: a global 10-m land-cover dataset with a fine classification system from Sentinel-1 and Sentinel-2 time-series data in Google Earth Engine
Abstract. The continuous development of remote sensing techniques provides ample opportunities for high-resolution land-cover mapping. Although global 10-m land-cover products have made considerable progress over past few years, their simple classification system makes it difficult to meet the needs of diverse applications. In this work, we propose a hierarchical land-cover mapping framework to produce a novel global 10-m land-cover dataset with a fine classification system (called GLC_FCS10) using Sentinel-1 and Sentinel-2 time-series observations from 2023. First, the globally distributed training samples are hierarchically obtained from multisource prior products after applying a series of refinements. Then, a combination of hierarchical land-cover mapping, local adaptive modeling, and multisource features is used to produce land-cover maps for each 5 × 5 geographical tile. Next, using 56121 globally distributed validation samples and a third-party validation dataset (LCMAP_Val), the GLC_FCS10 is assessed. The GLC_FCS10 achieves an overall accuracy of 83.16 % and a kappa coefficient of 0.789 globally and an overall accuracy of 85.09 % in the United States. Meanwhile, comparisons with five released 10- or 30-m land-cover products also demonstrate that GLC_FCS10 has higher accuracy and captures more diverse land-cover information than three of the released global 10-m land-cover products. In summary, the novel GLC_FCS10 land-cover maps can provide important support for high-resolution land-cover related research and applications. The GLC_FCS10 can be freely access via https://doi.org/10.5281/zenodo.14729665 (Liu and Zhang, 2025).
- Preprint
(3154 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on essd-2025-73', Anonymous Referee #1, 15 Apr 2025
This work is of great importance in multiple fields, such as earth science, geography, terrestrial ecosystem. These dataset could be very useful in many places. But some shortcomings in methods, figures, and tables need careful clarification and revisement, here are comments:
- Fig 1 is too colorful to demonstrate the core idea. Also, there is no legend to show what are different colors indicating. The relationships between different boxes are confusing too. Suggesting simplifying the figure by summarizing the main and key steps using icon and/or key words, do not put every step and all the datasets in this single figure; be consistent in color, size, and font. More detailed techniques can be in new figure.
- Table 2. how was the newly added forest- and wetland-related subcategories defined in the new system? What are the quantitative standards for closed vs open forest? Need more justification. These subclasses will of great importance in understanding the diversity of forest and wetland ecosystem, but only if the definition and classification of these subclasses are reasonable and practical.
- 2.3 How was the GLC_FCS30D dataset used as the training dataset for the 10m global cover mapping in this study? They are in different spatial resolutions, also, there are uncertainties in the GLC_FCS30D, not to mention the GLC_FCS30D does not cover 2023. How were all the uncertainties in the training dataset evaluated? Without solid evaluation, these training dataset cannot be high-confidence.
- Line 277, how as the percentiles be quantified, by date? by quality or what? So was the VV VH percentiles in Line 280.
- 4 Line 297-300 add a figure to show how was the hierarchical land cover constructed.
- Line 307 what are the 5 × 5 geographical tiles indicating? how large is the tile, and why choosing 5X 5?
- Fig 3 is a map not maps.
- Table 3. why only 10 types evaluated, what about the other 20 types? what are the bottom line of OA. Kappa mean, why they are different from others?
Citation: https://doi.org/10.5194/essd-2025-73-RC1 -
RC2: 'Comment on essd-2025-73', Anonymous Referee #2, 25 Apr 2025
This study developed a novel global 10-m land-cover dataset with a fine classification system. The data performs well and is of great value to high-resolution land-cover applications. Here are some of my concerns:
1. As highlighted in previous studies (Wang et al, 2023; Xu et al, 2024), Imp-ESRI_LC exhibits extensive patches of the impervious surface and lacks spatial details, which can also be found in Figure 6. This raises concerns about the MaxBound_imp (the union of several impervious surface products), which appears to include substantial areas of inner-city vegetation. If all training samples for natural land cover types are collected outside the MaxBound_imp, could this lead to the omission of inner-city vegetation types like grass and avenue trees?
See: Wang, Y., Xu, Y., Xu, X., Jiang, X., Mo, Y., Cui, H., Zhu, S., and Wu, H.: Evaluation of six global high- resolution global land cover products over China, International Journal of Digital Earth.
Xu, P., Tsendbazar, N.-E., Herold, M., de Bruin, S., Koopmans, M., Birch, T., Carter, S., Fritz, S., Lesiv, M., Mazur, E., Pickens, A., Potapov, P., Stolle, F., Tyukavina, A., Van De Kerchove, R., and Zanaga, D.: Comparative validation of recent 10 m-resolution global land cover maps, Remote Sensing of Environment.
2. Line 307, how many the samples for urban, rural and natural surfaces? The author just mentioned the ratio of these three land cover types.3.Line 330, why divide the non-wetlands into water body, forest, grassland, bare land, and others but not the remaining 8 basic land cover types?
4. Line 356 and Line 357, the manuscript references both "LCMAP_V" and "LCMAP_AL", are these two distinct datasets?
5. Line 436, P.A. is complementary to the omission error, not the commission error.
6. Why was the land cover type “Impervious surface” written as “Developed” in Table 5? These two have different definitions.
7. The confusion matrix (Table 5) shows that GLC_FCS10 misclassifies a large proportion of actual vegetation types as developed land (44 cropland, 69 forest and 164 grass/shrub reference samples are misclassified as developed land by GLC_FCS10), resulting in low UA for developed land, while only 4 developed land reference samples are misclassified elsewhere. This contradicts the statement in Line 433-435.
Citation: https://doi.org/10.5194/essd-2025-73-RC2
Data sets
GLC_FCS10: global 10 m land-cover dataset with fine classification system from Sentinel-1 and 2 time-series data Liangyun Liu and Xiao Zhang https://doi.org/10.5281/zenodo.14729665
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
1,117 | 183 | 14 | 1,314 | 21 | 14 |
- HTML: 1,117
- PDF: 183
- XML: 14
- Total: 1,314
- BibTeX: 21
- EndNote: 14
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1