the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 04 Jan 2026
CCAV-10m: An Annual Spatiotemporal China's Coastal Wetland Vegetation Dataset Integrating Sentinel-1/2 Observations via Deep Learning
Abstract. Coastal wetland vegetation plays a vital role in shoreline protection and ecosystem management, highlighting the need for accurate and high-resolution mapping of these unique and vulnerable habitats. Here, we present CCAV-10m, the first publicly available annual species-level coastal wetland dataset for China at 10 m resolution (2016–2023). This dataset was generated using a novel phenology-guided coastal wetland vegetation classification network (P_SVCN), which integrates Sentinel-1/2 satellite imagery with extensive in situ observations. Validation based on 4,668 in situ samples confirms that P_SVCN delivers strong classification performance, achieving an overall accuracy of 0.916 and a Kappa coefficient of 0.898. Spatiotemporal analysis of CCAV-10m reveals that Suaeda spp. is the dominant vegetation type, followed by S. alterniflora, whose coverage nearly equals the combined extent of P. australis, mangroves, S. mariqueter, and T. chinensis. Notably, this work fills critical gaps in both spatial detail and temporal consistency across existing coastal wetland datasets, demonstrating the effectiveness of deep-learning-based fusion of optical and SAR data for high-resolution vegetation mapping. Regular updates to CCAV-10m will support long-term coastal wetland research, enhance invasive species monitoring, and inform wetland restoration and precision management efforts. The CCAV-10m dataset is openly accessible at https://doi.org/10.57760/sciencedb.31077 (Li et al.,2025).
- Preprint
(12738 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on essd-2025-741', Anonymous Referee #1, 13 Feb 2026
-
RC2: 'Comment on essd-2025-741', Anonymous Referee #1, 13 Feb 2026
Publisher’s note: the content of this comment was removed on 13 February 2026 since the comment was posted by mistake.
Citation: https://doi.org/10.5194/essd-2025-741-RC2 -
RC3: 'Comment on essd-2025-741', Anonymous Referee #1, 13 Feb 2026
Publisher’s note: the content of this comment was removed on 13 February 2026 since the comment was posted by mistake.
Citation: https://doi.org/10.5194/essd-2025-741-RC3 -
RC4: 'Comment on essd-2025-741', Anonymous Referee #1, 13 Feb 2026
This study presents the CCAV-10m dataset, the first annual, species-level coastal wetland vegetation map of China at a 10-meter resolution, spanning from 2016 to 2023. By developing a novel phenology-guided classification network (P_SVCN) that integrates Sentinel-1 SAR data with Sentinel-2 optical imagery, the authors have generated a highly consistent dataset that captures the spatiotemporal dynamics of six key coastal wetland vegetation types. The manuscript is well-written and presents a technically sound methodology, effectively filling a critical data gap in coastal ecosystem monitoring.
However, I have some minor comments before this manuscript is published.
(1) Please ensure that all botanical taxonomic names follow standard nomenclature throughout the manuscript. Genus names should be capitalized, specific epithets should be lowercase, and the entire binomial must be italicized (e.g., Spartina alterniflora). After the first mention of the full name, the genus may be abbreviated (e.g., S. alterniflora)
(2) While Figure 7 provides a clear visual trend of provincial stacked areas, it is difficult for readers to extract precise annual values for specific provinces and species . To enhance the data’s utility and citability for future research, I recommend including a detailed statistical table in the Appendix/Supplementary Materials containing these specific values.
(3) The manuscript states that Sentinel-2 imagery was selected as bi-temporal data based on phenological information. However, the selection criteria or compositing strategy for Sentinel-1 imagery remains unclear. Please supplement the Methodology section with a description of how the Sentinel-1 data were selected.
(4) High-quality vegetation samples are critical for model training. In Section 3.3, it is mentioned that "samples underwent systematic quality control," but the specific procedures are not detailed. Please provide a clear explanation of the quality control criteria and methods used to ensure the reliability and representativeness of the training data.
(5) The public availability of the CCAV-10m dataset via a DOI is highly commendable, yet the long-term maintenance of the data remains a point of interest. Since the title emphasizes "Annual Spatiotemporal" monitoring while the series currently ends in 2023, could the authors clarify whether there are concrete plans to extend the dataset to include 2024, 2025, and subsequent years?
Citation: https://doi.org/10.5194/essd-2025-741-RC4 -
AC1: 'Reply on RC4', Min Liu, 22 Feb 2026
On behalf of all co-authors, we sincerely appreciate the reviewers' positive evaluation of our CCAV-10m dataset and the P_SVCN framework. We have addressed all specific comments in the revised manuscript, starting with the correction of botanical nomenclature throughout the text to ensure scientific rigor. To improve the data's utility, we have added Appendix A containing a precise annual statistical table (2016–2023) by province and species. We have also supplemented the methodology to clarify our annual mean compositing strategy for Sentinel-1 imagery and introduced Appendix B to detail our two-stage quality control process, featuring a Kappa-based benchmarking (k = 0.95) and double-blind cross-validation. Finally, we have formalized our commitment in the Data Availability section to provide annual version-controlled updates for 2024, 2025, and beyond. We believe these revisions, detailed in the attached point-by-point response PDF, significantly enhance the reliability and long-term value of our study.
-
AC1: 'Reply on RC4', Min Liu, 22 Feb 2026
-
RC5: 'Comment on essd-2025-741', Naiqing Pan, 17 Mar 2026
This study produces CCAV-10m, an annual 10 m species-level coastal wetland vegetation dataset from 2016 to 2023, using a phenology-guided deep learning framework. The topic is well suited to Earth System Science Data, and the dataset is valuable for long-term monitoring of coastal wetland vegetation, invasive species expansion, and ecosystem management. The manuscript is generally well organized, and the reported classification performance is impressive. My comments mainly concern clarity, reproducibility, and several points of interpretation and presentation.
- The dataset is valuable and timely, but the methodological description could be more reproducible.
The manuscript provides a reasonably clear description of the P_SVCN framework, including the dual-branch design, the SAR–optical fusion strategy, and the attention-based architecture. However, the actual training procedure remains insufficiently described. I therefore suggest that the authors add a concise but explicit description of the model training procedure. This could include, where applicable, the loss function, optimizer, learning rate, stopping criterion, and software framework used. These details are important because the novelty of the manuscript partly lies in the deep learning framework itself.
- The manuscript would benefit from clearer interpretation of dataset characteristics and limitations.
The manuscript reports good classification performance for the CCAV-10m dataset, and the proposed P_SVCN model shows higher accuracy than the previous SVCN model. I suggest that the authors add more discussion of the mechanisms or characteristics of the P_SVCN model that may explain its better performance. At the same time, since coastal wetland vegetation is strongly affected by tides, salinity gradients, and phenological variation, the discussion should more explicitly address how these factors may still influence uncertainty or local misclassification. In Figure 7, the results suggest large interannual variations in the area of some species, for example, P. australis and Suaeda spp. in Liaoning from 2022 to 2023, and Suaeda spp. in Shandong before and after 2020. Do these large interannual variations reflect real-world changes, or might they be influenced by uncertainties or local misclassifications? The manuscript already acknowledges these issues to some extent, but the discussion could be strengthened by linking them more directly to dataset quality and potential limitations in application.
Specific comments
1 Since the study area only covers eight coastal provinces and does not include Guangdong, Guangxi, or Hainan. I suggest revising the title to make the spatial coverage more precise, for example by using “eastern coastal China” rather than “China”.
2 Line 94: The manuscript states that 320 Sentinel-2 scenes were selected, which is much smaller than the number of Sentinel-1 scenes. Please clarify the criteria used for scene selection, such as acquisition timing and cloud coverage thresholds.
3 Line 111-113: The description of the manually interpreted and field-based samples is useful, but I suggest adding a clearer visual summary of their spatial distribution in the supplementary material. A map would improve transparency.
4 Figure2: Does each line represent the NDVI of one representative plot, or the average NDVI of all wetlands within that region? Also, why is the NDVI time series of S. alterniflora in Figure 2(c) discontinuous?
5 It would be helpful to clarify whether the reported validation design is entirely sample-based or whether there was also any spatial separation to reduce possible spatial autocorrelation between training and validation samples.
6 Figure 4: The left panel appears to show mangroves in Hangzhou Harbor. Do mangroves actually exist there?
Overall, I find the dataset contribution promising and suitable for publication in Earth System Science Data after minor revision.
Citation: https://doi.org/10.5194/essd-2025-741-RC5 -
AC2: 'Reply on RC5', Min Liu, 21 Mar 2026
On behalf of all co-authors, we sincerely appreciate the reviewer’s positive assessment of the CCAV-10m dataset and the P_SVCN framework. We have addressed all specific comments in the revised manuscript to enhance its technical transparency and geographic accuracy, starting with the addition of a new section (Section 3.4.4) to explicitly detail the P-SVCN model training procedure, including the optimizer, loss function, and learning rate strategy. To ensure geographic precision, the title and all related descriptions have been revised to "Eastern Coastal China" to accurately reflect the study's actual spatial scope. We have introduced a new section (Section 5.4) to discuss the inherent limitations of the CCAV-10m dataset. This includes a detailed attribution analysis of environmental factors, focusing on how hydrometeorological anomalies and observational constraints influence classification uncertainty. Furthermore, we clarified the Sentinel-2 scene selection criteria (10% cloud threshold), added Appendix C to visualize the spatial distribution of our 15,558 in-situ samples, and acknowledged localized misclassifications (e.g., mangroves in Hangzhou Bay) while proposing a knowledge-guided post-processing strategy for future iterations. We believe these revisions, detailed in the attached point-by-point response, significantly strengthen the reliability and scientific rigor of our study for Earth System Science Data.
Data sets
CCAV-10m: An Annual Spatiotemporal China Coastal Wetland Vegetation Dataset Integrating Sentinel-1/2 Observations via Deep Learning Yuying Li et al. https://doi.org/10.57760/sciencedb.31077
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 606 | 287 | 49 | 942 | 25 | 26 |
- HTML: 606
- PDF: 287
- XML: 49
- Total: 942
- BibTeX: 25
- EndNote: 26
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Publisher’s note: the content of this comment was removed on 13 February 2026 since the comment was posted by mistake.