the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 21 Aug 2025
National-Scale Sub-meter Mapping of Spartina alterniflora in Mainland China 2020
Abstract. Current large-scale maps of Spartina alterniflora (S. alterniflora) with 10 m resolution hinder accurate delineation of community boundaries, detection of internal features such as creeks, and identification of small patches. These limitations further compromise the accuracy of spatial distribution extraction and subsequent analyzes. To this end, this study produced the first 2020 national-scale Sub-meter S. alterniflora Map of Mainland China (CM-SSM), using an Object- and Sub-meter-enhanced Pixel-based Phenological Feature (OSPPF) composite method. The method integrates phenological features from Sentinel-2 with spatial and textural details from Google Earth imagery, improving the spectral separability and mitigating mixed-pixel effects. Compared to the 10 m S. alterniflora product of Mainland China (CMSA), CM-SSM improved overall accuracy by 14.60 % and the F1 score by 0.21. Although the total mapped areas of CM-SSM (59,371 ha) and CMSA (58,006 ha) differ by only 1,365 ha, their spatial distributions diverge substantially. When benchmarked against CM-SSM, CMSA exhibited commission and omission errors totaling 34,273 ha (57.73 %). Moreover, the number of patches identified by CM-SSM (148,072) was over 17 times greater than that of CMSA, reflecting its superior capability in detecting fragmented distributions. In addition, Soil Organic Carbon (SOC) estimates derived from CM-SSM were 706.69 Gg (23.09 %) higher than those reported by the latest national SOC product, emphasizing the essential contribution of high-resolution mapping to accurate carbon accounting for S. alterniflora. These advances enhance understanding of S. alterniflora invasion dynamics, support carbon accounting, and inform evidence-based coastal wetland management and restoration. The map is available at https://doi.org/10.5281/zenodo.16296823 (Xu et al., 2025).
- Preprint
(8960 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on essd-2025-436', Xue Liu, 19 Sep 2025
-
RC2: 'Comment on essd-2025-436', Anonymous Referee #2, 21 Sep 2025
The manuscript presents an OSPPF method that effectively integrates multi-source data to generate a sub-meter resolution distribution map of Spartina alterniflora. The work is thorough, the methodology is reliable and rigorous, and the resulting dataset holds significant value for the scientific community. However, several issues require clarification or improvement:
-
The authors determine phenological transitions based on NDVI and propose thresholds of 0.3 and 0.5. However, in Figure 3, the Y-axis lacks tick marks corresponding to these threshold values. The authors are advised to add tick marks or include horizontal reference lines to improve readability.
-
How were the thresholds of NDVI < 0.3 for the senescence period and NDVI > 0.5 for the green period determined? Were they derived from statistical distributions, field observations, or a specific phenological model? Clarification on the rationale behind these thresholds is needed.
-
Table 5 shows that CM-SSM achieves significant improvements over CMSA in terms of F1 score and overall accuracy (OA). However, the current evaluation of classification accuracy relies on a limited set of metrics. It is recommended that the authors construct confusion matrices to compare the composition of error types (e.g., omission and commission errors) between the two methods, thereby providing deeper insight into the specific aspects in which CM-SSM outperforms CMSA.
-
Given that the study area spans a considerable latitudinal range, there may be substantial heterogeneity in phenological characteristics across regions. Consequently, spatial variability in classification performance should be considered. Currently, the evaluation appears to rely solely on CN-SSM as a reference for calculating omission and commission errors for CMSA. The authors are encouraged to establish a validation dataset spanning multiple latitudinal zones and use it to comparatively assess the performance of CN-SSM and CMSA, demonstrating whether CN-SSM exhibits generalizability across diverse geographical regions.
-
While the authors employ a variety of features for classification, they do not discuss the relative contribution of each feature to classification performance. It is recommended to analyze and report feature importance using the built-in measures from the Random Forest classifier. Such an analysis would enhance understanding of the key factors driving Spartina alterniflora identification and provide valuable insights for future method development.
Citation: https://doi.org/10.5194/essd-2025-436-RC2 -
Data sets
National-Scale Sub-meter Mapping of Spartina alterniflora in Mainland China 2020 Meng Xu, Jinyan Tian, and Bingfeng Zhou https://doi.org/10.5281/zenodo.16296823
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 1,154 | 43 | 12 | 1,209 | 15 | 18 |
- HTML: 1,154
- PDF: 43
- XML: 12
- Total: 1,209
- BibTeX: 15
- EndNote: 18
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The manuscript presents a significant and novel contribution by producing the first sub-meter resolution map of Spartina alterniflora for mainland China. The proposed OSPPF method effectively integrates multi-source data and object-based analysis to address critical limitations of existing products. The study is well-structured, the methodology is sound, and the dataset is highly valuable for the community. However, several major and minor points require clarification and improvement before the manuscript can be considered for publication.
Â
1. The use of Google Earth imagery from 2019 and 2021 alongside 2020 Sentinel-2 data introduces a potential source of error. S. alterniflora dynamics can be rapid. Please quantify the extent (e.g., percentage of area) where non-2020 imagery was used and discuss the potential impact of this temporal mismatch on classification accuracy and the final area estimate. A sensitivity analysis in these areas would significantly strengthen the manuscript.
Â
2. The final manual refinement, while understandable for a first-of-its-kind map, introduces subjectivity. Please describe the protocol followed for this manual correction (e.g., number of interpreters, ruleset used, process for resolving disagreements) to ensure consistency. Discussing the potential magnitude of error or bias introduced by this step is crucial for assessing the dataset's reliability.
Â
3. The method relies on existing products (CMSA) for defining the study area and segmentation buffers. This limits its application to regions or time periods where such prior maps are unavailable or inaccurate. Please discuss the transferability of the OSPPF method to other regions without relying on existing S. alterniflora products. Could the method be adapted to be more automated and independent?
Â
4. The authors rightly identify DL as a promising future direction and even note that CM-SSM could serve as training data. Given that DL models (e.g., U-Net, Transformers) are now state-of-the-art for many fine-scale land cover mapping tasks, a discussion on why an object-based RF was chosen over a DL approach is necessary. A direct comparison, even on a subset, would greatly strengthen the methodological justification, or the limitations of not using DL should be explicitly acknowledged.
Â
5. The reported improvements in OA and F1-score are substantial. However, please support these claims with a statistical test (e.g., McNemar's test) to confirm that the difference in accuracy between CMSA and CM-SSM is statistically significant and not due to chance.
Â
6. The SOC comparison is a compelling application but is briefly described. Please provide more detail on the methodology: How was the "unified provincial-level SOC unit storage coefficient" derived? Was it based on field measurements? A table of these coefficients and a reference to the method (Zhang et al., 2024) should be included in the main text or supplement. This is critical for readers to assess the validity of the 706.69 Gg difference.
Â
7. The Random Forest classifier provides the valuable ability to rank feature importance. An analysis showing which features (e.g., Sentinel-2 phenological bands, GE texture features, RGB indices) were most important for the classification would provide deeper insight into the ecology of S. alterniflora and validate the design choices of the OSPPF method.
Â
8. The manuscript mentions masking water pixels (SCL=6) to reduce tidal effects. However, tidal state can significantly influence the appearance and detectability of S. alterniflora. Please clarify if the Sentinel-2 compositing process considered tidal height information to select images from a consistent low-tide period, or discuss the potential residual impact of tidal variability on the phenological feature compositing.
Â
9. The text describing the workflow (Figure 2) could be more precise. Please explicitly state the final number of bands in the PPF, SPPF, and OSPPF feature sets. A clear listing of all input features for the RF model would improve reproducibility.
Â
10. Figures 8, 9, and 10 are critical but lack clarity. The y-axis labels in Fig. 8 are cut off. Fig. 9's Venn diagram is simple but effective; ensure the values are clearly visible.
Â
11. The captions for Figures 5, 6, and 7 should explicitly state that the OSPPF result is the final, manually refined CM-SSM product for clarity.
Â
12. The term "sub-meter" is used throughout, but the actual resolution of the final CM-SSM product should be explicitly stated early in the abstract and method (it is 0.9m, as mentioned later). Briefly justify why this specific resolution from GE was chosen.