Effects of Multi-Growth Periods UAV Images on Classifying Karst Wetland Vegetation Communities Using Object-Based Optimization Stacking Algorithm

Round 1

Reviewer 1 Report

Combining machine learning algorithms with multi-temporal remote sensing data to finely classify wetland vegetation is a current research hotspot, and the authors developed an ensemble learning model of stacked random forest (RF), CatBoost and XGBoost algorithms to complete the fine mapping of karst wetland vegetation communities and evaluate the performance of their classification algorithms. The classification accuracy is obtained as follows: Stacking > CatBoost > RF > XGBoost. Finally, based on the SHAP (SHapley Additive exPlanations) method, the influence and contribution of different characteristic bands to the classification of vegetation communities are clarified from both local and global perspectives. The contribution rate of vegetation index and texture features to the classification of vegetation communities in karst wetlands was higher than that of the original spectral band, geometric features and location characteristics. However, there are still some issues that need to be solved and explained：

1.       Line 117 Can you tell us about the advantages of the OBIA algorithm compared to other algorithms?

2.       Line 305 Describe in detail the principles of the RFE method。

3.       Line 307 What is the basis for setting the correlation coefficient to 0.95 and 0.8？

4.       Line 361 What is the author's rationale for using the McNemar chi-square test method？

5.       Line 402 The author selected different vegetation community distribution images in 6 scenes for classification by different algorithms, and would like to ask whether the vegetation communities in these six scenes have certain typicality or are randomly selected？

6.       Line 555 Scenario 6 can more accurately identify what is the cause of BO and WH？

7.       From the perspective of growth period, the authors concluded that the classification accuracy in July was the highest, the classification accuracy in October was the lowest, and the autumn caused the misdivision of vegetation in the spectral characteristics because of the withering of leaves, and if the autumn 10 was selected as September to classify again, whether the accuracy would be improved？

Author Response

请参阅附件。

Author Response File: Author Response.docx

Reviewer 2 Report

Review: Effects of Multi-Growth Periods UAV Images on Classifying Karst Wetland Vegetation Communities Using Object-Based Optimization Stacking Algorithm

In this paper, the authors propose to use multi-growth period UAV data and optimized Stacking algorithm to classify the vegetation communities in karst wetlands, and elucidate the influence and contribution of different feature bands to the classification of vegetation communities using the SHAP method. This paper has a novel idea and a clear scientific question, but the research program is designed to be complete and the workload is valuable. Although this study is interesting, there are some minor issues,, and the authors are advised to make appropriate revisions.

All concerns, major and minor, are noted below.

1. The introduction needs to be further condensed, especially paragraphs 2 and 3.

2.  (a)-(e) are labeled in figure 1 but do not appear in the text.

3. How do the authors take into account the large differences in sample sizes for each category in Table 2, which may affect the accuracy of the overall precision?

4. Whether the meta model in the methodology is a fixed model that is not described or illustrated in the results or discussion section.

5. For the number of decimal places in table 5, 3 places should be retained, too many are useless.

6. Can the optimal parameters of each model be given in Table 6.

7. Line 466-473. McNemar's test only compares the variability between variables, how do the authors conclude that the Stacking algorithm is more reliable？

8. Line 681-690 can be placed in the Methods section.

9. The labeling of the statistical graphs in the article is generally not clear, and the authors are advised to revise them.

Author Response

Please see the attachment.

Author Response File: Author Response.docx