Coniferous Forests Aboveground Biomass Inversion in Typical Regions of China with Joint Sentinel-1 and Sentinel-2 Remote Sensing Data Supported by Different Feature Optimizing Algorithms
Round 1
Reviewer 1 Report
Comments and Suggestions for Authors
The study is noteworthy for its application and comparative examination of various models for estimating forest AGB. Nevertheless, the overall influence is somewhat constrained due to the absence of substantial progress in methodology or original discoveries.
One way to improve the title's impact and specificity would be to make a mention of the particular forest types or geographic regions that were examined. This update would give the reader immediate background information regarding the particular application area of the study.
The abstract is excessively lengthy. As stated in the journal, "The abstract section should not exceed 200 words."
The introduction provides a sufficient background on the significance of forest AGB and the general use of remote sensing for its estimation; however, it falls short of outlining specific hypotheses, detailing the study's objectives, conducting a thorough literature review, discussing divergent viewpoints, and clearly stating the study's aim and anticipated conclusions. Addressing these areas would greatly enhance the introduction.
In the introduction, it would be helpful to state the precise hypotheses that are being tested more explicitly.
1. Literature Review: It is advisable to conduct a comprehensive examination of the existing body of research in the field, encompassing contentious and contradictory hypotheses.
Materials and Methods:
Include more details about why these specific places were chosen, how representative they are for the study, and how their unique traits may influence the study's outcomes.
The selection of Sentinel-1 and Sentinel-2 data could have benefited from a more comprehensive elucidation, specifically regarding the rationale behind their suitability for the study's objectives. The reader might also benefit from an explanation of the thought process that went into selecting the dates for data collection.
1. Feature Selection and Optimization (Lines 341-390): Discuss possible feature intercorrelations in the RF and KNN-FIFS models and provide an explanation for the selection of particular features, including window sizes.
2. Parameters of the Model (Lines 356-366): In the RF model, the number of decision trees is fixed at one thousand. Please provide a more rigorous reason for this decision tree count.
Section 2.3.1: Elaborate on the methodology employed to determine the particular texture features to be utilized and their correlation with the estimation of forest AGB. It would also be helpful to talk about these features' possible drawbacks.
Results and Discussion:
While the results are reported in Section 3.1, a more thorough explanation of the reasons for the superior performance of some models in various scenarios will enhance the section. It could be easier for readers to understand if there are charts or graphs showing how the models' performances differ from one another.
1. Analysis Depth (Lines 515–556): To a greater extent, investigate the factors that are responsible for the different degrees of performance that are exhibited by the various models. Talk about the potential causes of the performance variations. Considerations such as these can include the forest's overall make-up, the impact of local weather patterns, or the relative merits of different models when applied to the characteristics of the regions under investigation.
2. Comparison with Other Studies (Lines 515–556): Provide a thorough comparison with other studies, emphasizing the special benefits and constraints of this investigation. The text might benefit from a more thorough comparative analysis that explains how these results compare to and differ from those of other comparable studies. This could entail contrasting the outcomes with those attained in comparable settings or with comparable techniques in other studies.
3. Future Research Directions: Provide evidence for the recommended courses of study in light of the study's conclusions.
Conclusion:
1. Synopsis of Results : Improve the conclusion by making a stronger connection between the particular results of this investigation and more general field implications.
Clarity and Language in English:
1. Language Usage (Throughout the Manuscript): To improve readability, simplify difficult sentences and make grammatical corrections.
Novelty and Technical Highlights:
1. Novelty (Lines 48–88): Describe the study's novelty in relation to previous research.
2. Concerns about Technology (Lines 341-390): Talk about the feature selection procedure used in the RF and KNN-FIFS models as well as the reasoning behind the model parameters, especially in the RF model.
This research presents a knowledge of forest AGB estimate using machine learning algorithms and remote sensing data. The comparison between different forest types and various models and data sources is a subject of great fascination. Addressing issues with the methodology, the introduction, and the clarity of the conclusions could significantly improve the overall quality of the study. It is advisable that the authors implement these modifications to ensure that their contribution to the field is more comprehensive and lucid.
Comments on the Quality of English LanguageLanguage Usage (Throughout the Manuscript): To improve readability, simplify difficult sentences and make grammatical corrections.
Author Response
Response to Reviewers’ Comments on the manuscript (No. 2755235) with:
This manuscript was submitted to Forests (No. 2755235). As per the reviewers’ suggestions, it is now revised and submitted to the journal along with the responses to comments from the editor and reviewers.
The authors are indebted to the anonymous reviewers for their constructive criticism, which has been instrumental in improving the quality of the manuscript.
In the following we transcribe the comments in gray italics, and our responses in blue in a response section. The changes are highlighted in red for which changed in the revised manuscript.
General Comments to the Author
The study is noteworthy for its application and comparative examination of various models for estimating forest AGB. Nevertheless, the overall influence is somewhat constrained due to the absence of substantial progress in methodology or original discoveries.
Detailed Comment 1:
One way to improve the title's impact and specificity would be to make a mention of the particular forest types or geographic regions that were examined. This update would give the reader immediate background information regarding the particular application area of the study.
Detailed Response 1:
The title of this manuscript has been revised to: Coniferous Forests Aboveground Biomass Inversion in typical regions of China with joint Sentinel-1 and Sentinel-2 remote sensing data sup-ported by different feature optimizing algorithms.
Detailed Comment 2:
The abstract is excessively lengthy. As stated in the journal, "The abstract section should not exceed 200 words."
Detailed Response 2:
Thank you for the reviewer's comments. The abstract of this manuscript has been reduced to the maximum extent possible according to your requirements.
Abstract: Multispectral remote sensing data and synthetic aperture radar (SAR) data can provide horizontally and vertically information of forest AGB under different stand conditions. With the abundance of remote sensing (RS) features extracted from multispectral and SAR datasets, a key point for accurate forest AGB estimation is to use suitable feature optimization inversion algorithms. In this study, feature optimization inversion algorithms including Multiple linear stepwise regression (MLSR), K-nearest neighbor with fast iterative features selection (KNN-FIFS) and random forest (RF) were explored with a total of 93 RS features working as inversion model input for forest AGB inversion. The results showed that KNN-FIFS and the combination of Sentinel-1 and Sentinel-2 performed best at both test sites (R2 = 0.568 and RMSE =15.05 t/hm2 for Puer and R2 = 0.511 and RMSE =32.29 t/hm2 for Genhe). Among three feature optimization inversion algorithms, RF performed worst for forest AGB estimation with R2 = 0.348 and RMSE =18.06 t/ hm2 for Puer and R2 = 0.345 and RMSE =35.98 t/hm2 for Genhe using feature combination of Sentinel-1 and Sentinel-2. The results indicates that the combination of multispectral data and C-band data SAR features improve the inversion accuracy of forest AGB and KNN-FIFS algorithm has robustness transferability in the inversion of forests AGB.
Detailed Comment 3:
The introduction provides a sufficient background on the significance of forest AGB and the general use of remote sensing for its estimation; however, it falls short of outlining specific hypotheses, detailing the study's objectives, conducting a thorough literature review, discussing divergent viewpoints, and clearly stating the study's aim and anticipated conclusions. Addressing these areas would greatly enhance the introduction.
Detailed Response 3:
Thank you for the reviewer's comments. I have made the following comprehensive and thorough supplements and modifications in the introduction section as per your request.
Relevant studies have pointed out that there are some limitations in the estimation of forest AGB using a single SAR data or a single optical data, and the inversion accuracy and efficiency can be improved when both are used jointly [9-11]. Shao et al. estimated coniferous forest AGB using Radarsat-2 and Landsat-8 OLI in Genhe, Inner Mongolia, and showed that combined optical and microwave information can estimate forest AGB ore accurately [12]. Li et al. found poor performance of forest AGB estimation when only SAR backscatter coefficients were used while the combination of Radarsat-2 and Landsat-5 TM improved the estimation accuracy with R2 = 0.768, RMSE = 19.14t/ha [13]. Pan et al. estimated forest AGB using combination GF-3 PolSAR and Landsat-8 OLI and KNN-SFS in Guangxi and the results showed that the accuracy of inversion results was higher than that of the estimation using a single data source with R2= 0.75 and RMSE = 21.05t/ha [14].
Although the combination of optical and SAR datasets can improve forest biomass, there are still significant limitations in data access, and some of the data are not available to the public and result in certain costs. Sentinel -1 SAR data and Sentinel -2 multispectral data are not only free of charge but also timely revisit in global coverage and provide new opportunities for inversion and monitoring of forest AGBs at regional or global scales. Pan et al. used backscatter coefficients, texture features extracted from Sentinel-1 and multiple linear stepwise regression (MLSR) algorithm for forest AGB estimation at regional scale [15]; The study of Guo et al. showed that Sentinel-2 data can be better used for regional forest AGB inversion [16]. David et al. combined Sentinel-1 and Sentinel-2 to estimate forest AGB and achieved a good result with R2 = 0.95 and RMSE is 0.25t/ha [17]. However, only few studies explored the potential of using combination of Sentinel-1 and Sentinel-2 to estimate forest AGBs.
Another key issue in forest AGB estimation is to select suitable inversion models especially with capability of feature optimization algorithms. Tian et al. classified the estimation methods of forest AGB into traditional model estimation methods and machine learning methods [7]; Traditional model estimation methods have been widely used for forest AGB estimation, among which MLSR algorithm has become a more commonly used algorithm due to its simple principle and easy operation, and is often used for forest AGB estimation in tropical, subtropical, and temperate zones. Machine learning algorithms can solve the problems of data nonlinearity and high dimensionality, significantly improving the estimation accuracy of forest AGB, but since the "black-box" operation, machine learning algorithms is difficult to reflect the mechanistic process between remote sensing features and forest AGB [15]. Exploring how to select an appropriate forest AGB inversion model under the support of optimal RS features selection algorithms in different test sites, which aims to improve the estimation accuracy of forest AGB and the generalization capability of the inversion model in different forest scenes, has become one of the most hot research topics.
Detailed Comment 4:
In the introduction, it would be helpful to state the precise hypotheses that are being tested more explicitly.
Detailed Response 4:
Thank you for the reviewer's comments. I have added precise hypotheses in the introduction section as your request.
The north-eastern and south-western regions of China, as the main sources of timber supply and forest products in China, have made an important contribution to China's timber industry and development [18-19]. Larch is one of the major afforestation species in the Northeast, while Simao Pine is one of the major high-yielding resin harvesting species in the Southwest, and turpentine is an important industrial raw material, while both species are the main species in the National Natural Forest Protection Project [20]. Therefore, accurate estimation of biomass of larch and Simao pine is not only beneficial to forest resource management and conservation but also can accurately calculate carbon stock. Currently, fewer studies are using Sentinel-1 SAR data and Sentinel-2 multispectral data for the inversion of AGB of larch and Simao pine, and there are uncertainties in the inversion results. In this study, two types of coniferous forests, Xing'an larch in Inner Mongolia, and Simao pine in Yunnan Province were selected to explore the suitable RS feature optimization inversion algorithms for their forest AGB estimation. RS features extracted from Single Sentinel-1, single Sentinel-2 and combination of Sentinel-1 and Sentinel-2 were input for the forest AGB inversion. The objective of this study is to explore the advantages of coniferous forest AGB estimation in typical regions in China under different forms of combinations of the two RS data supported by different feature optimization inversion algorithms. Meanwhile, explore the robustness and transferability of these inversion algorithms to enhance the monitoring and decision-making capabilities for the forest resources management.
Detailed Comment 5:
Literature Review: It is advisable to conduct a comprehensive examination of the existing body of research in the field, encompassing contentious and contradictory hypotheses.
Detailed Response 5:
Thank you for the reviewer's comments. I have made the following comprehensive and thorough supplements and modifications in the introduction section as per your request.
Relevant studies have pointed out that there are some limitations in the estimation of forest AGB using a single SAR data or a single optical data, and the inversion accuracy and efficiency can be improved when both are used jointly [9-11]. Shao et al. estimated coniferous forest AGB using Radarsat-2 and Landsat-8 OLI in Genhe, Inner Mongolia, and showed that combined optical and microwave information can estimate forest AGB ore accurately [12]. Li et al. found poor performance of forest AGB estimation when only SAR backscatter coefficients were used while the combination of Radarsat-2 and Landsat-5 TM improved the estimation accuracy with R2 = 0.768, RMSE = 19.14t/ha [13]. Pan et al. estimated forest AGB using combination GF-3 PolSAR and Landsat-8 OLI and KNN-SFS in Guangxi and the results showed that the accuracy of inversion results was higher than that of the estimation using a single data source with R2= 0.75 and RMSE = 21.05t/ha [14].
Although the combination of optical and SAR datasets can improve forest biomass, there are still significant limitations in data access, and some of the data are not available to the public and result in certain costs. Sentinel -1 SAR data and Sentinel -2 multispectral data are not only free of charge but also timely revisit in global coverage and provide new opportunities for inversion and monitoring of forest AGBs at regional or global scales. Pan et al. used backscatter coefficients, texture features extracted from Sentinel-1 and multiple linear stepwise regression (MLSR) algorithm for forest AGB estimation at regional scale [15]; The study of Guo et al. showed that Sentinel-2 data can be better used for regional forest AGB inversion [16]. David et al. combined Sentinel-1 and Sentinel-2 to estimate forest AGB and achieved a good result with R2 = 0.95 and RMSE is 0.25t/ha [17]. However, only few studies explored the potential of using combination of Sentinel-1 and Sentinel-2 to estimate forest AGBs.
Another key issue in forest AGB estimation is to select suitable inversion models especially with capability of feature optimization algorithms. Tian et al. classified the estimation methods of forest AGB into traditional model estimation methods and machine learning methods [7]; Traditional model estimation methods have been widely used for forest AGB estimation, among which MLSR algorithm has become a more commonly used algorithm due to its simple principle and easy operation, and is often used for forest AGB estimation in tropical, subtropical, and temperate zones. Machine learning algorithms can solve the problems of data nonlinearity and high dimensionality, significantly improving the estimation accuracy of forest AGB, but since the "black-box" operation, machine learning algorithms is difficult to reflect the mechanistic process between remote sensing features and forest AGB [15]. Exploring how to select an appropriate forest AGB inversion model under the support of optimal RS features selection algorithms in different test sites, which aims to improve the estimation accuracy of forest AGB and the generalization capability of the inversion model in different forest scenes, has become one of the most hot research topics.
Detailed Comment 6:
Include more details about why these specific places were chosen, how representative they are for the study, and how their unique traits may influence the study's outcomes.
Detailed Response 6:
Thank you for the reviewer's comments. I have made the following modifications as your request.
The north-eastern and south-western regions of China, as the main sources of timber supply and forest products in China, have made an important contribution to China's timber industry and development [18-19]. Larch is one of the major afforestation species in the Northeast, while Simao Pine is one of the major high-yielding resin harvesting species in the Southwest, and turpentine is an important industrial raw material, while both species are the main species in the National Natural Forest Protection Project [20]. Therefore, accurate estimation of biomass of larch and Simao pine is not only beneficial to forest resource management and conservation but also can accurately calculate carbon stock.
Detailed Comment 7:
The selection of Sentinel-1 and Sentinel-2 data could have benefited from a more comprehensive elucidation, specifically regarding the rationale behind their suitability for the study's objectives. The reader might also benefit from an explanation of the thought process that went into selecting the dates for data collection.
Detailed Response 7:
Thank you for the reviewer's comments. Regarding the reasons for collecting remote sensing data, I have also made modifications and additions in the manuscript.
The Sentinel-1A data selected for the study were obtained from European Space Agency (ESA, http://www.esa.int/ESA). The Sentinel-1A is equipped with a C-band SAR sensor operating at 5.4 GHz. SAR data are not significantly affected by cloudiness, so a considerable number of complete images are available each month. However, SAR signal can be affected by recent surface precipitation or wind, so images from a good weather period close to the field collection date were selected for analysis [17,23]. Considering the consistency with ground survey data from various test sites and the impact of soil moisture, the Sentinel-1A image for Pu'er was acquired on November 10, 2020, and the image for Genhe was acquired on August 17, 2016. Referring to the existing studies, the data processing was carried out using SNAP software with the aid of resampled 10 m DEM data to perform pre-processing steps such as radiometric correction, filtering, and terrain correction.
Detailed Comment 8:
Feature Selection and Optimization (Lines 341-390): Discuss possible feature intercorrelations in the RF and KNN-FIFS models and provide an explanation for the selection of particular features, including window sizes.
Detailed Response 8:
Thank you for your comment. In Section 3, the results and analysis of feature selection for each feature selection algorithm were analyzed and explained in detail, and window size analysis was also conducted. As shown below:
3.1.2. Construction of KNN-FIFS
The optimal model parameters for the KNN-FIFS algorithm are shown in Table 4. In Pu'er, B6 and S2REP parameters were selected from single Sentinel-2 and combinations of Sentinel-1 and Sentinel-2. It indicated that the red-edge bands and short-wave infrared bands of Sentinel-2 are more sensitive to forest AGB and more significant than other RS features in forest AGB estimation. Among the many vegetation factors, the factors involved in the calculation of the red-edge band have the strongest correlation with the forest AGBs, the reason is that they can effectively relieve the signal saturation problem corresponding to the high biomass value and the dense canopy. In addition, the selected features in the Pu'er and Genhe are mostly divided into features with windows of 7 and 9, indicating that the size of the window has a significant impact on the correlations between RS features and forest AGB.
3.1.3 Construction of RF
The results of feature optimizing for RF algorithms were summarized in Table 5. For using single Sentinel-1 data source, entropy, non-similarity, and contrast are selected as the optimized features. Among them entropy indicate the degree of complexity of reflected RS characterization, the bigger the entropy value, the higher the complexity, and the larger the amount of information it contains. The non-similarity reflects the fact that the greater the grey scale difference within the image element, the clearer the visual effect of the image is. The feature indicated that the vegetation cover portion of the image element has a high degree of fractional anisotropy, which can better improve the model inversion accuracy. As for the single Sentinel-2 data, S2REP, B5, and NDI45 were selected to participate in the modeling in the combination of features at the two test sites. Red-edge band information is included in S2REP, B5, and NDI45, which respond to the small changes in the structure of vegetation canopies and the content of chlorophyll and are more sensitive to the growth status of vegetation. For the combination of Sentinel-1 and Sentinel-2, the entropy and S2REP were selected as optimized RS features in both of the two test sites. From the perspective of window size, the texture features extracted at window size of 9×9 and 3×3 were ranked relatively high in the single data and the combination of Sentinel-1 and Sentinel-2. The results indicated that the small window texture has a strong ability to explain the changes of the forest AGBs and has certain advantages in their inversion.
Detailed Comment 9
Parameters of the Model (Lines 356-366): In the RF model, the number of decision trees is fixed at one thousand. Please provide a more rigorous reason for this decision tree count.
Detailed Response 9:
Thank you for the reviewer's comments. I have made the following modifications as your request.
There are two important parameters like mtry and ntree in the RF. ntree is the number of decision trees, mtry is the number of random features, and its size is usually set to one-third of the total number of variables by default in regression problems [35]. In order to determine the size of the parameter ntree, the plot function in the random forest package is used to draw the trend graph of the regression error with the change of the number of decision trees, and it can be seen from Fig. 5 that the regression error tends to stable when the number of decision trees reaches 1,000, and in order to ensure the credibility of the results and the efficiency of the operation, the value of 1,000 for ntree was selected.
RS features are determined by two metrics in the RF parameters, named as the incremental model MSE (%IncMSE) and the node purity of the model tree (IncNodePurity). when the independent variable is used as out-of-bag data, larger parameter values of these two metrics represent the more important the feature. The importance of the features was calculated by the important function in the Random Forest package, and each RS feature was ranked in descending order according to the model MSE increment. According to the ordering of RF feature importance, variables with variable ordering in the top 10 were selected for inversion model training and validation [31,36]. The importance ranks of RS variables used in the two test sites are shown in Fig 5.
Detailed Comment 10:
Section 2.3.1: Elaborate on the methodology employed to determine the particular texture features to be utilized and their correlation with the estimation of forest AGB. It would also be helpful to talk about these features' possible drawbacks.
Detailed Response 10:
Thank you for the reviewer's comments. I have made the following modifications as your request.
(2) Sentinel-1 texture features. Texture features are an important source of information for high-resolution SAR data, and image texture has certain advantages in the identification of stand structure such as stand age, stand density and leaf area index [24]. Texture, the fine structure of an image, refers to the frequency of tonal changes on an image. Visually finer textures indicate less spatial variation in image luminance values over regions [23, 25]; visually coarser textures have more drastic variations in pixel values over regions [25]. Although texture parameters are used to classify land-use types and vegetation, the relationship between image texture and forest AGB has not been fully explored [15].
Detailed Comment 11:
While the results are reported in Section 3.1, a more thorough explanation of the reasons for the superior performance of some models in various scenarios will enhance the section. It could be easier for readers to understand if there are charts or graphs showing how the models' performances differ from one another.
Detailed Response 11:
Thank you for the reviewer's comments. I have made the following modifications as your request.
3.3 Comparative analysis of the results of three data sources under different features optimization
Using the four indicators of R2、RMSE、rRMSE and MAPE, the performance of MLSR, KNN-FIFS, and RF models were compared and analyzed in Fig. 8. According to figure 8 we found that under the data combination forms in both study areas, R2 values for the 3 models ranges from 0.116 to 0.568. Among them, R value for KNN-FIFS is the highest and for RF is the lowest. In Pu’er, under the performance of combination of Sentinel-1 and Sentinel-2, the RMSE from KNN-FIFS is the lowest (15.04 t/hm2). The estimation resutls from KNN-FIFS has the lowest RMSE value and estimation results from RF has the highest RMSE value. But in the case of using single Sentinel-1 data in Genhe, the RMSE value from MLSR is the highest (35.38 t/hm2).The values of rRMSE also confirmed the best performance of KNN-FIFS with the lowest value of 12.4%. The analysis from MAPE showed the lowest value acquried from KNN-FIFS and highest accuracy in Pu'er. However, the performance of KNN-FIFS in Genhe is not stable with large errors, although the error of MLSR is relatively small. Overall, in two study areas and three data combinations, KNN-FIFS generally showed the highest accuracy, indicating the robustness of feature optimization of this algorithm.
|
(a) |
(b) |
|
(c) |
(d) |
|
(e) |
(f) |
|
(g) |
(h) |
|
(s)、(t)、(u) 、(y)represent the Pu'er test site; (v)、(w)、(x)、(z) represent the Genhe test site. |
|
|
Figuer 8. Accuracy evaluation and comparsion |
|
Detailed Comment 12:
Analysis Depth (Lines 515–556): To a greater extent, investigate the factors that are responsible for the different degrees of performance that are exhibited by the various models. Talk about the potential causes of the performance variations. Considerations such as these can include the forest's overall make-up, the impact of local weather patterns, or the relative merits of different models when applied to the characteristics of the regions under investigation.
Detailed Response 12:
Thank you for the reviewer's comments. I have made the following modifications as your request.
(1) Estimation of forest AGB with single Sentinel-1 and Sentinel-2 data under different feature optimizing methods.
Single Sentinel-1 and single Sentinel-2 data have different estimated potentials in co niferous forest AGB estimation under different feature optimizing approaches. The R2 of the inversion results of the MLSR, RF, and KNN-FIFS models using single Sentinel-1 data were 0.414, 0.116, and 0.512 in Pu'er, whereas those of the inversion results from of Genhe, were 0.380, 0.274 and 0.470. Pan et al. used Sentinel-1 as the RS data source and took the fir forest in Jiangle County as the test sample, and the highest precision R2 of the inversion results was 0.636, which was slightly higher than results of this study, probably due to differences in texture factors across different study areas, the accuracy of this study area is slightly lower [15]. In the study of Shi et al., GF-1 and GF-3 were selected as the data sources for the inversion study of the total biomass of forest AGB and component AGB, in which the results of the Yunnan pine forest inversion study of the total AGB of a single GF-3 data were slightly lower than that of in Pu'er in this study, and at the same time were slightly higher than that of Genhe in this study, with an R2 of 0.500. The better performance may result from the quad-polarization information from GF-3. Quad-polarization information can better characterize forest structure than the dual-polarization sentinel data in this study [38]. In addition, Xu et al. used Sentinel-1, Landsat-8 OLI, and continuous forest inventory data as the main data source, and took subtropical evergreen broadleaf forests in Guidong to explore the performance of forest AGB inversion using single RS data sourse. In this study, the R2 for the inversion result using single Sentinel-1 data was 0.49, which was slightly lower than that our results, the reason maybe that SAR data has stronger peroration capability in coniferous forests than in broadleaf forest [39].
The R2 of the inversion results of the MLSR, RF, and KNN-FIFS models using single Sentinel-2 data in Pu'er were 0.345, 0.255, and 0.480, while those values of the inversion results in Genhe were 0.320, 0.188 and 0.390. Wang et al. selected Sentinel-2, Landsat-8 OLI and GF-1 as data sources to estimate forest AGB and found Sentinel-2 data had the best results with an R2 = 0.60 and an RMSE = 21.40 t/ha. The results also indicated that the red-edge band has a strong sensitivity to forest AGB, and has great potential for quantitative inversion studies [40]. Guo et al. used single Sentinel-2 data as data source, constructed MLSR model to invert coniferous forest AGB in Inner Mongolia. The R2 of the inversion results was 0.765 and RMSE was 39.49t/ha, which were slightly higher than that in our study. It may due to the influence of topographic factors, which affect forest diversity to a large extent, distribution and local heterogeneity, thus directly affected forest productivity and structure [41]. Lopez-Serrano et al. used Landsat-8 to study temperate coniferous forests in Mexico. Band indices, vegetation indices, texture indices, topographic and climatic variables were extracted and used for forest AGB estimation through RF and support vector regression (SVR) algorithm. They found that SVR performed better in coniferous forest AGB inversion with R2 = 0.8 and RMSE = 8.20t/ha. The results revealed that considering climatic variables helps to improve the inversion accuracy of forest AGB with RS features since climatic characteristics are one of the most important factors affecting the distribution and growth of forest species [42].
(2) Estimation of forest AGB for combination of Sentinel-1 and Sentinel-2 data under different feature optimizing methods.
The R2 of the inversion results of the MLSR, RF, and KNN-FIFS models using combination of Sentinel-1 and Sentinel-2 data in Pu'er were 0.461, 0.348, and 0.568, while those values in Genhe were 0.446, 0.345, and 0.511. Pan et al. used Sentinel-1 and Sentinel-2 estimate coniferous forests AGB in subtropical regions, and the inversion result of R2 was 0.575, which was similar to that of the subtropical coniferous forests in Pu'er in this study. However, the values were slightly higher than that obtained in the cold-temperate coniferous forests in Genhe in our study [14]. In the study of Liu et al. the combination of Sentinel-1 and Sentinel-2 were used as the data source for forest AGB estimation as well, but in this study the effects of topographic factors on forest AGB were considered. In this study, with the comparison of performance of MLSR, BP neural network, and SVR models the best performance for forest AGB estimation was acquired by BP neural network model with R2 of 0.821, RMSE is 32.39t/ha. The results of the study are higher than the results of our study. The results of the study showed that the potential of combination of optical data and SAR data for improving the inversion accuracy of forest AGB [43]. Forkuor et al. used Sentinel-1 and Sentinel-2 and RF for forest AGB estimation in West African dryland forests and showed that the joint Sentinel-1 and Sentinel data could be better for inversion of forest AGB with R2 = 0.90 and RMSE = 54.5t/hm2[44].
(3) The advantages of Forest AGB estimation using KNN-FIFS algorithm.
KNN-FIFS shows most robust in forest AGB inversion in both test sites, moreover, KNN-FIFS achieved superior inversion results in forest AGB inversion both using different data sources and at different test sites. Shi et al. used KNN-FIFS model and GF-3 data for forest AGB estimation in Yiliang and also confirmed its better performance with R2=0.500 and RMSE=14.11t/ha, which is slightly lower than the inversion results of our study accuracy with R2=0.512 and RMSE=15.34 t/hm2 [38]. Han et al. also used KNN-FIFS and Landsat-8 OLI and airborne SAR P-band for forest AGB inversion. The results of from them were slightly higher than the results of our study with R2 = 0.770 and RMSE = 22.74t/ha. The higher R2 value and lower RMSE value may resulted from the involved the P-band SAR data, which can penetrate forest deeper and interact with the large branches and trunks accounted for a larger proportion of forest AGB [30]. Ji et al. used ALOS-1 PALSAR-1 and ALOS-2 PALSAR as data sources and KNN-FIFS as inversion model to estimate forest AGB in Genhe. The inversion accuracy of KNN-FIFS was outperformed than RF and KNN and its R2 is 0.37, which was lower than the value obtained in this study. The lower R2 values in their study may due to the incomplete coverage of RS images to the study area [45]. The study confirmed the robustness of KNN-FIFS and best performance of combined Sentinel-1and Sentinel-2 in forest AGB estimations.
Detailed Comment 13:
Comparison with Other Studies (Lines 515–556): Provide a thorough comparison with other studies, emphasizing the special benefits and constraints of this investigation. The text might benefit from a more thorough comparative analysis that explains how these results compare to and differ from those of other comparable studies. This could entail contrasting the outcomes with those attained in comparable settings or with comparable techniques in other studies.
Detailed Response 13:
Thank you for the reviewer's comments. I have made the following modifications as your request.
(1) Estimation of forest AGB with single Sentinel-1 and Sentinel-2 data under different feature optimizing methods.
Single Sentinel-1 and single Sentinel-2 data have different estimated potentials in co niferous forest AGB estimation under different feature optimizing approaches. The R2 of the inversion results of the MLSR, RF, and KNN-FIFS models using single Sentinel-1 data were 0.414, 0.116, and 0.512 in Pu'er, whereas those of the inversion results from of Genhe, were 0.380, 0.274 and 0.470. Pan et al. used Sentinel-1 as the RS data source and took the fir forest in Jiangle County as the test sample, and the highest precision R2 of the inversion results was 0.636, which was slightly higher than results of this study, probably due to differences in texture factors across different study areas, the accuracy of this study area is slightly lower [15]. In the study of Shi et al., GF-1 and GF-3 were selected as the data sources for the inversion study of the total biomass of forest AGB and component AGB, in which the results of the Yunnan pine forest inversion study of the total AGB of a single GF-3 data were slightly lower than that of in Pu'er in this study, and at the same time were slightly higher than that of Genhe in this study, with an R2 of 0.500. The better performance may result from the quad-polarization information from GF-3. Quad-polarization information can better characterize forest structure than the dual-polarization sentinel data in this study [38]. In addition, Xu et al. used Sentinel-1, Landsat-8 OLI, and continuous forest inventory data as the main data source, and took subtropical evergreen broadleaf forests in Guidong to explore the performance of forest AGB inversion using single RS data sourse. In this study, the R2 for the inversion result using single Sentinel-1 data was 0.49, which was slightly lower than that our results, the reason maybe that SAR data has stronger peroration capability in coniferous forests than in broadleaf forest [39].
The R2 of the inversion results of the MLSR, RF, and KNN-FIFS models using single Sentinel-2 data in Pu'er were 0.345, 0.255, and 0.480, while those values of the inversion results in Genhe were 0.320, 0.188 and 0.390. Wang et al. selected Sentinel-2, Landsat-8 OLI and GF-1 as data sources to estimate forest AGB and found Sentinel-2 data had the best results with an R2 = 0.60 and an RMSE = 21.40 t/ha. The results also indicated that the red-edge band has a strong sensitivity to forest AGB, and has great potential for quantitative inversion studies [40]. Guo et al. used single Sentinel-2 data as data source, constructed MLSR model to invert coniferous forest AGB in Inner Mongolia. The R2 of the inversion results was 0.765 and RMSE was 39.49t/ha, which were slightly higher than that in our study. It may due to the influence of topographic factors, which affect forest diversity to a large extent, distribution and local heterogeneity, thus directly affected forest productivity and structure [41]. Lopez-Serrano et al. used Landsat-8 to study temperate coniferous forests in Mexico. Band indices, vegetation indices, texture indices, topographic and climatic variables were extracted and used for forest AGB estimation through RF and support vector regression (SVR) algorithm. They found that SVR performed better in coniferous forest AGB inversion with R2 = 0.8 and RMSE = 8.20t/ha. The results revealed that considering climatic variables helps to improve the inversion accuracy of forest AGB with RS features since climatic characteristics are one of the most important factors affecting the distribution and growth of forest species [42].
(2) Estimation of forest AGB for combination of Sentinel-1 and Sentinel-2 data under different feature optimizing methods.
The R2 of the inversion results of the MLSR, RF, and KNN-FIFS models using combination of Sentinel-1 and Sentinel-2 data in Pu'er were 0.461, 0.348, and 0.568, while those values in Genhe were 0.446, 0.345, and 0.511. Pan et al. used Sentinel-1 and Sentinel-2 estimate coniferous forests AGB in subtropical regions, and the inversion result of R2 was 0.575, which was similar to that of the subtropical coniferous forests in Pu'er in this study. However, the values were slightly higher than that obtained in the cold-temperate coniferous forests in Genhe in our study [14]. In the study of Liu et al. the combination of Sentinel-1 and Sentinel-2 were used as the data source for forest AGB estimation as well, but in this study the effects of topographic factors on forest AGB were considered. In this study, with the comparison of performance of MLSR, BP neural network, and SVR models the best performance for forest AGB estimation was acquired by BP neural network model with R2 of 0.821, RMSE is 32.39t/ha. The results of the study are higher than the results of our study. The results of the study showed that the potential of combination of optical data and SAR data for improving the inversion accuracy of forest AGB [43]. Forkuor et al. used Sentinel-1 and Sentinel-2 and RF for forest AGB estimation in West African dryland forests and showed that the joint Sentinel-1 and Sentinel data could be better for inversion of forest AGB with R2 = 0.90 and RMSE = 54.5t/hm2[44].
(3) The advantages of Forest AGB estimation using KNN-FIFS algorithm.
KNN-FIFS shows most robust in forest AGB inversion in both test sites, moreover, KNN-FIFS achieved superior inversion results in forest AGB inversion both using different data sources and at different test sites. Shi et al. used KNN-FIFS model and GF-3 data for forest AGB estimation in Yiliang and also confirmed its better performance with R2=0.500 and RMSE=14.11t/ha, which is slightly lower than the inversion results of our study accuracy with R2=0.512 and RMSE=15.34 t/hm2 [38]. Han et al. also used KNN-FIFS and Landsat-8 OLI and airborne SAR P-band for forest AGB inversion. The results of from them were slightly higher than the results of our study with R2 = 0.770 and RMSE = 22.74t/ha. The higher R2 value and lower RMSE value may resulted from the involved the P-band SAR data, which can penetrate forest deeper and interact with the large branches and trunks accounted for a larger proportion of forest AGB [30]. Ji et al. used ALOS-1 PALSAR-1 and ALOS-2 PALSAR as data sources and KNN-FIFS as inversion model to estimate forest AGB in Genhe. The inversion accuracy of KNN-FIFS was outperformed than RF and KNN and its R2 is 0.37, which was lower than the value obtained in this study. The lower R2 values in their study may due to the incomplete coverage of RS images to the study area [45]. The study confirmed the robustness of KNN-FIFS and best performance of combined Sentinel-1and Sentinel-2 in forest AGB estimations.
Detailed Comment 14:
Future Research Directions: Provide evidence for the recommended courses of study in light of the study's conclusions.
Detailed Response 14:
Thank you for the reviewer's comments. I have made the following modifications as your request.
In this study, we investigated the potential of feature optimization inversion model for forest AGB estimation using Sentinel-1 SAR data, Sentinel-2 multispectral data and the combination of them. Larch pure forests in northeastern China and Simao pine pure forests in southwestern China are investigated in this study. Through the study we concluded that: (1) Combining Sentinel-1 and Sentinel-2 data allows a certain degree of information complementarity, and a higher estimation accuracy than only using Sentinel-1 and Sentinel-2 alone. (2) Comparative analyses of the two test sites in two typical coniferous forests with three combinations of two data sources showed that tree species or AGB levels may result in the effect of forest AGB estimation accuracy. (3) All of MLSR, RF and KNN-FIFS were subjected to feature optimization during the forest AGB modelling, among them KNN-FIFS show best suitability and promotability in different forest scenes and different test sites. However, due to the limitation of field data collection, only 26 larch pure forest sample plots and 27 Simao pine pure forest sample plots were used for modelling analysis in this study. Although the results are convincing, more different observations such as LiDAR data, UAV data, and L- and P-band SAR data with longer wavelengths are needed to be explored and validated in the future [7,46-47]. In addition, only one forest type like pure coniferous forest was used in this study, and other different forest types such as broadleaf forest, mixed coniferous forest need to be further studied in the future [7,48-49].
Detailed Comment 15
Synopsis of Results : Improve the conclusion by making a stronger connection between the particular results of this investigation and more general field implications.
Detailed Response 15:
Thank you for the reviewer's comments. I have made the following modifications as your request.
In this study, we investigated the potential of feature optimization inversion model for forest AGB estimation using Sentinel-1 SAR data, Sentinel-2 multispectral data and the combination of them. Larch pure forests in northeastern China and Simao pine pure forests in southwestern China are investigated in this study. Through the study we concluded that: (1) Combining Sentinel-1 and Sentinel-2 data allows a certain degree of information complementarity, and a higher estimation accuracy than only using Sentinel-1 and Sentinel-2 alone. (2) Comparative analyses of the two test sites in two typical coniferous forests with three combinations of two data sources showed that tree species or AGB levels may result in the effect of forest AGB estimation accuracy. (3) All of MLSR, RF and KNN-FIFS were subjected to feature optimization during the forest AGB modelling, among them KNN-FIFS show best suitability and promotability in different forest scenes and different test sites. However, due to the limitation of field data collection, only 26 larch pure forest sample plots and 27 Simao pine pure forest sample plots were used for modelling analysis in this study. Although the results are convincing, more different observations such as LiDAR data, UAV data, and L- and P-band SAR data with longer wavelengths are needed to be explored and validated in the future [7,46-47]. In addition, only one forest type like pure coniferous forest was used in this study, and other different forest types such as broadleaf forest, mixed coniferous forest need to be further studied in the future [7,48-49].
Detailed Comment 16 :
Language Usage (Throughout the Manuscript): To improve readability, simplify difficult sentences and make grammatical corrections.
Detailed Response 16:
Thank you for the reviewer's comments. I have made language modifications to the entire manuscript as per your request.
Detailed Comment 17 :
Novelty (Lines 48–88): Describe the study's novelty in relation to previous research.
Detailed Response 17:
Thank you for the reviewer's comments. I have made comprehensive revisions to the entire text in the introduction section (line 31-112) as per your request.
Detailed Comment 18 :
Concerns about Technology (Lines 341-390): Talk about the feature selection procedure used in the RF and KNN-FIFS models as well as the reasoning behind the model parameters, especially in the RF model.
Detailed Response 18:
Thank you for the reviewer's comments. I have made the following modifications as your request.
There are two important parameters like mtry and ntree in the RF. ntree is the number of decision trees, mtry is the number of random features, and its size is usually set to one-third of the total number of variables by default in regression problems [35]. In order to determine the size of the parameter ntree, the plot function in the random forest package is used to draw the trend graph of the regression error with the change of the number of decision trees, and it can be seen from Fig. 5 that the regression error tends to stable when the number of decision trees reaches 1,000, and in order to ensure the credibility of the results and the efficiency of the operation, the value of 1,000 for ntree was selected.
RS features are determined by two metrics in the RF parameters, named as the incremental model MSE (%IncMSE) and the node purity of the model tree (IncNodePurity). when the independent variable is used as out-of-bag data, larger parameter values of these two metrics represent the more important the feature. The importance of the features was calculated by the important function in the Random Forest package, and each RS feature was ranked in descending order according to the model MSE increment. According to the ordering of RF feature importance, variables with variable ordering in the top 10 were selected for inversion model training and validation [31,36]. The importance ranks of RS variables used in the two test sites are shown in Fig 5.
Detailed Comment 19 :
This research presents a knowledge of forest AGB estimate using machine learning algorithms and remote sensing data. The comparison between different forest types and various models and data sources is a subject of great fascination. Addressing issues with the methodology, the introduction, and the clarity of the conclusions could significantly improve the overall quality of the study. It is advisable that the authors implement these modifications to ensure that their contribution to the field is more comprehensive and lucid.
Detailed Response 19:
Thank you for the reviewer's comments. I have made modifications to the introduction, methodology, and conclusion sections as per your request.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsReview of Inversion study of forest above-ground biomass with joint Sentinel-1 and Sentinel-2 remote sensing data supported by different feature optimizing algorithms
This paper requires substantial modification and revision in terms of introducing the used parameters and, particularly, presenting the explanatory statistics of the data used.
Also, the results and discussion sections should be rewritten for better clarification and understanding for the readers.
The authors need to clearly explain the main contribution of the study.
The authors need to re-check the provided results. For instance, I could see that the difference between the R2 coefficient for Sentinel-1 using the KNN-FIFS and RF is considerable. In my experience, the authors can get similar results by tuning the adjustable parameters of the models.
The abstract is too long for a research paper.
The literature review is not up-to-date. I could not find any new studies (e.g., 2023).
The representation of the results is rather weak. Using simple scatter plots does not suffice for a proper diagnostic analysis.
Figures need more details in their captions. For example, Figure 3 has a two-word caption.
Add more statistical parameters, such as MAPE, NSE, and BIAS, to the outcomes of the study.
Comments on the Quality of English Language
Needs moderate revision.
Author Response
Response to Reviewers’ Comments on the manuscript (No. 2755235) with:
This manuscript was submitted to Forests (No. 2755235). As per the reviewers’ suggestions, it is now revised and submitted to the journal along with the responses to comments from the editor and reviewers.
The authors are indebted to the anonymous reviewers for their constructive criticism, which has been instrumental in improving the quality of the manuscript.
In the following we transcribe the comments in gray italics, and our responses in blue in a response section. The changes are highlighted in red for which changed in the revised manuscript.
Detailed Comment 1
This paper requires substantial modification and revision in terms of introducing the used parameters and, particularly, presenting the explanatory statistics of the data used. Also, the results and discussion sections should be rewritten for better clarification and understanding for the readers.
Detailed Response 1:
Thank you for the reviewer's comments. I have made the modification to the results and discussion section according to your request.
Detailed Comment 2
The authors need to clearly explain the main contribution of the study.
Detailed Response 2:
Thank you for the reviewer's comments.I have made the modifications as per your request.
The north-eastern and south-western regions of China, as the main sources of timber supply and forest products in China, have made an important contribution to China's timber industry and development [18-19]. Larch is one of the major afforestation species in the Northeast, while Simao Pine is one of the major high-yielding resin harvesting species in the Southwest, and turpentine is an important industrial raw material, while both species are the main species in the National Natural Forest Protection Project [20]. Therefore, accurate estimation of biomass of larch and Simao pine is not only beneficial to forest resource management and conservation but also can accurately calculate carbon stock.Currently, fewer studies are using Sentinel-1 SAR data and Sentinel-2 multispectral data for the inversion of AGB of larch and Simao pine, and there are uncertainties in the inversion results. In this study, two types of coniferous forests, Xing'an larch in Inner Mongolia, and Simao pine in Yunnan Province were selected to explore the suitable RS feature optimization inversion algorithms for their forest AGB estimation. RS features extracted from Single Sentinel-1, single Sentinel-2 and combination of Sentinel-1 and Sentinel-2 were input for the forest AGB inversion. The objective of this study is to explore the advantages of coniferous forest AGB estimation in typical regions in China under different forms of combinations of the two RS data supported by different feature optimization inversion algorithms. Meanwhile, explore the robustness and transferability of these inversion algorithms to enhance the monitoring and decision-making capabilities for the forest resources management.
In this study, we investigated the potential of feature optimization inversion model for forest AGB estimation using Sentinel-1 SAR data, Sentinel-2 multispectral data and the combination of them. Larch pure forests in northeastern China and Simao pine pure forests in southwestern China are investigated in this study. Through the study we concluded that: (1) Combining Sentinel-1 and Sentinel-2 data allows a certain degree of information complementarity, and a higher estimation accuracy than only using Sentinel-1 and Sentinel-2 alone. (2) Comparative analyses of the two test sites in two typical coniferous forests with three combinations of two data sources showed that tree species or AGB levels may result in the effect of forest AGB estimation accuracy. (3) All of MLSR, RF and KNN-FIFS were subjected to feature optimization during the forest AGB modelling, among them KNN-FIFS show best suitability and promotability in different forest scenes and different test sites. However, due to the limitation of field data collection, only 26 larch pure forest sample plots and 27 Simao pine pure forest sample plots were used for modelling analysis in this study. Although the results are convincing, more different observations such as LiDAR data, UAV data, and L- and P-band SAR data with longer wavelengths are needed to be explored and validated in the future [7,46-47]. In addition, only one forest type like pure coniferous forest was used in this study, and other different forest types such as broadleaf forest, mixed coniferous forest need to be further studied in the future [7,48-49].
Detailed Comment 3
The authors need to re-check the provided results. For instance, I could see that the difference between the R2 coefficient for Sentinel-1 using the KNN-FIFS and RF is considerable. In my experience, the authors can get similar results by tuning the adjustable parameters of the models.
Detailed Response 3:
Thank you for the reviewer's comments. I have checked all the results and added statistical indicators to the results section as you requested.
3.3 Comparative analysis of three data sources under different features optimization algorithms
Using the four indicators of R2、RMSE、rRMSE and MAPE, the performance of MLSR, KNN-FIFS, and RF models were compared and analyzed in Fig. 8. According to figure 8 we found that under the data combination forms in both study areas, R2 values for the 3 models ranges from 0.116 to 0.568. Among them, R value for KNN-FIFS is the highest and for RF is the lowest. In Pu’er, under the performance of combination of Sentinel-1 and Sentinel-2, the RMSE from KNN-FIFS is the lowest (15.04 t/hm2). The estimation resutls from KNN-FIFS has the lowest RMSE value and estimation results from RF has the highest RMSE value. But in the case of using single Sentinel-1 data in Genhe, the RMSE value from MLSR is the highest (35.38 t/hm2).The values of rRMSE also confirmed the best performance of KNN-FIFS with the lowest value of 12.4%. The analysis from MAPE showed the lowest value acquried from KNN-FIFS and highest accuracy in Pu'er. However, the performance of KNN-FIFS in Genhe is not stable with large errors, although the error of MLSR is relatively small. Overall, in two study areas and three data combinations, KNN-FIFS generally showed the highest accuracy, indicating the robustness of feature optimization of this algorithm.
|
(a) |
(b) |
|
(c) |
(d) |
|
(e) |
(f) |
|
(g) |
(h) |
|
|
|
|
(s)、(t)、(u) 、(y)represent the Pu'er test site; (v)、(w)、(x)、(z) represent the Genhe test site. |
|
|
Figuer 8. Accuracy evaluation and comparsion |
|
Detailed Comment 4
The abstract is too long for a research paper.
Detailed Response 4:
Thank you for the reviewer's comments. The abstract of this manuscript has been reduced to the maximum extent possible according to your requirements.
Abstract: Multispectral remote sensing data and synthetic aperture radar (SAR) data can provide horizontally and vertically information of forest AGB under different stand conditions. With the abundance of remote sensing (RS) features extracted from multispectral and SAR datasets, a key point for accurate forest AGB estimation is to use suitable feature optimization inversion algorithms. In this study, feature optimization inversion algorithms including Multiple linear stepwise regression (MLSR), K-nearest neighbor with fast iterative features selection (KNN-FIFS) and random forest (RF) were explored with a total of 93 RS features working as inversion model input for forest AGB inversion. The results showed that KNN-FIFS and the combination of Sentinel-1 and Sentinel-2 performed best at both test sites (R2 = 0.568 and RMSE =15.05 t/hm2 for Puer and R2 = 0.511 and RMSE =32.29 t/hm2 for Genhe). Among three feature optimization inversion algorithms, RF performed worst for forest AGB estimation with R2 = 0.348 and RMSE =18.06 t/ hm2 for Puer and R2 = 0.345 and RMSE =35.98 t/hm2 for Genhe using feature combination of Sentinel-1 and Sentinel-2. The results indicates that the combination of multispectral data and C-band data SAR features improve the inversion accuracy of forest AGB and KNN-FIFS algorithm has robustness transferability in the inversion of forests AGB.
Detailed Comment 5
The literature review is not up-to-date. I could not find any new studies (e.g., 2023).
Detailed Response 5:
Thank you for the reviewer's comments. I have made the modifications as per your request.
Detailed Comment 6
The representation of the results is rather weak. Using simple scatter plots does not suffice for a proper diagnostic analysis.
Detailed Response 6:
Thank you for the reviewer's comments. I have made the modifications as per your request.
3.3 Comparative analysis of the results of three data sources under different features optimization
Using the four indicators of R2、RMSE、rRMSE and MAPE, the performance of MLSR, KNN-FIFS, and RF models were compared and analyzed in Fig. 8. According to figure 8 we found that under the data combination forms in both study areas, R2 values for the 3 models ranges from 0.116 to 0.568. Among them, R value for KNN-FIFS is the highest and for RF is the lowest. In Pu’er, under the performance of combination of Sentinel-1 and Sentinel-2, the RMSE from KNN-FIFS is the lowest (15.04 t/hm2). The estimation resutls from KNN-FIFS has the lowest RMSE value and estimation results from RF has the highest RMSE value. But in the case of using single Sentinel-1 data in Genhe, the RMSE value from MLSR is the highest (35.38 t/hm2).The values of rRMSE also confirmed the best performance of KNN-FIFS with the lowest value of 12.4%. The analysis from MAPE showed the lowest value acquried from KNN-FIFS and highest accuracy in Pu'er. However, the performance of KNN-FIFS in Genhe is not stable with large errors, although the error of MLSR is relatively small. Overall, in two study areas and three data combinations, KNN-FIFS generally showed the highest accuracy, indicating the robustness of feature optimization of this algorithm.
Detailed Comment 7
Figures need more details in their captions. For example, Figure 3 has a two-word caption.
Detailed Response 7:
Thank you for the reviewer's comments. In addition to modifying the title of Figure 3, I also made comprehensive modifications to the table and image titles throughout the entire text.
Figure 3. AGB details for the two test sites. The changes of each bar mean the increase in AGB values, the number on each bar of circle represents the label of the plots, and the numbers out each bar of the circle represent the measured AGB (t/ha) of each plot.
Detailed Comment 8
Add more statistical parameters, such as MAPE, NSE, and BIAS, to the outcomes of the study.
Detailed Response 8:
Thank you for the reviewer's comments. I have made the following modifications as your request.
The validation indice used here include coefficient of determination (R2, eq.(5)), root mean square error (RMSE, eq.(6)), relative root mean square error (rRMSE, eq.(7)), and mean absolute percentage error (MAPE, eq.(8)). The value of R2 ranges from 0 to 1, and the closer it is to 1, the higher the accuracy of the inversion results, and vice versa, the lower the accuracy[34]. RMSE and rRMSE indicate the difference between the estimated and measured values, the smaller values indicate higher inversion accuracy.
3.3 Comparative analysis of three data sources under different features optimization algorithms
Using the four indicators of R2、RMSE、rRMSE and MAPE, the performance of MLSR, KNN-FIFS, and RF models were compared and analyzed in Fig. 8. According to figure 8 we found that under the data combination forms in both study areas, R2 values for the 3 models ranges from 0.116 to 0.568. Among them, R value for KNN-FIFS is the highest and for RF is the lowest. In Pu’er, under the performance of combination of Sentinel-1 and Sentinel-2, the RMSE from KNN-FIFS is the lowest (15.04 t/hm2). The estimation resutls from KNN-FIFS has the lowest RMSE value and estimation results from RF has the highest RMSE value. But in the case of using single Sentinel-1 data in Genhe, the RMSE value from MLSR is the highest (35.38 t/hm2).The values of rRMSE also confirmed the best performance of KNN-FIFS with the lowest value of 12.4%. The analysis from MAPE showed the lowest value acquried from KNN-FIFS and highest accuracy in Pu'er. However, the performance of KNN-FIFS in Genhe is not stable with large errors, although the error of MLSR is relatively small. Overall, in two study areas and three data combinations, KNN-FIFS generally showed the highest accuracy, indicating the robustness of feature optimization of this algorithm.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe manuscript uses the Sentinel-1 and Sentinel-2 data and employs MLSR, RF and KNN-FIFS methods to analyze the inversion of forests’ above-ground biomass. The authors need to address the following issues:
1) All abbreviations should be clearly expressed in the text when they are mentioned for the first time (e.g., SAR in line 10).
2) Abstract is too long, and it should be shortened. I suggest the authors to focus on the findings of the study. The numerical values for R2 and RMSE presented in lines 24-40 can be removed from the Abstract. Also, the use of R2 and RMSE as performance indicators for the algorithms is repeated in lines 20-23, and repeated statements should be removed.
3) Language editing is required. The manuscript contains many punctuation errors. Especially, the level of English for the Abstract should be improved.
4) The scope of the study is presented in the final paragraph of Section 1. However, the novelty of the study and its contribution to the literature are not explained. Regarding the research hotspots mentioned in lines 110-114 and the discussion presented in Section 4, the novelty of the work should be presented in Section 1.
5) A reference should be provided for Eq. (1).
6) Sentine-2 in lines 207 and 209 should be corrected.
7) Figure 4 is not referred to in the text.
8) Figure number in line 344 should be checked.
9) “… my research results …” in line 534 should be revised as “results of this study”.
Comments on the Quality of English LanguageLanguage editing is required. Please see the Comments and Suggestions for Authors.
Author Response
Response to Reviewers’ Comments on the manuscript ((No. 2755235) with:
This manuscript was submitted to Forests (No. 2755235). As per the reviewers’ suggestions, it is now revised and submitted to the journal along with the responses to comments from the editor and reviewers.
The authors are indebted to the anonymous reviewers for their constructive criticism, which has been instrumental in improving the quality of the manuscript.
In the following we transcribe the comments in gray italics, and our responses in blue in a response section. The changes are highlighted in red for which changed in the revised manuscript.
Detailed Comment 1
All abbreviations should be clearly expressed in the text when they are mentioned for the first time (e.g., SAR in line 10).
Detailed Response 1:
Thank you for the reviewer's comments. I have corrected the abbreviations of professional vocabulary throughout the entire text.
Detailed Comment 2
Abstract is too long, and it should be shortened. I suggest the authors to focus on the findings of the study. The numerical values for R2 and RMSE presented in lines 24-40 can be removed from the Abstract. Also, the use of R2 and RMSE as performance indicators for the algorithms is repeated in lines 20-23, and repeated statements should be removed.
Detailed Response 2:
Thank you for the reviewer's comments. I have made the modifications to the abstract section according to your request.
Abstract: Multispectral remote sensing data and synthetic aperture radar (SAR) data can provide horizontally and vertically information of forest AGB under different stand conditions. With the abundance of remote sensing (RS) features extracted from multispectral and SAR datasets, a key point for accurate forest AGB estimation is to use suitable feature optimization inversion algorithms. In this study, feature optimization inversion algorithms including Multiple linear stepwise regression (MLSR), K-nearest neighbor with fast iterative features selection (KNN-FIFS) and random forest (RF) were explored with a total of 93 RS features working as inversion model input for forest AGB inversion. The results showed that KNN-FIFS and the combination of Sentinel-1 and Sentinel-2 performed best at both test sites (R2 = 0.568 and RMSE =15.05 t/hm2 for Puer and R2 = 0.511 and RMSE =32.29 t/hm2 for Genhe). Among three feature optimization inversion algorithms, RF performed worst for forest AGB estimation with R2 = 0.348 and RMSE =18.06 t/ hm2 for Puer and R2 = 0.345 and RMSE =35.98 t/hm2 for Genhe using feature combination of Sentinel-1 and Sentinel-2. The results indicates that the combination of multispectral data and C-band data SAR features improve the inversion accuracy of forest AGB and KNN-FIFS algorithm has robustness transferability in the inversion of forests AGB.
Detailed Comment 3
Language editing is required. The manuscript contains many punctuation errors. Especially, the level of English for the Abstract should be improved.
Detailed Response 3:
Thank you for the reviewer's comments. I have made language modifications to the entire text as per your request.
Detailed Comment 4
The scope of the study is presented in the final paragraph of Section 1. However, the novelty of the study and its contribution to the literature are not explained. Regarding the research hotspots mentioned in lines 110-114 and the discussion presented in Section 4, the novelty of the work should be presented in Section 1.
Detailed Response 4:
Thank you for the reviewer's comments. I have made the modifications in the final paragraph of the intruoduction as per your request.
The north-eastern and south-western regions of China, as the main sources of timber supply and forest products in China, have made an important contribution to China's timber industry and development [18-19]. Larch is one of the major afforestation species in the Northeast, while Simao Pine is one of the major high-yielding resin harvesting species in the Southwest, and turpentine is an important industrial raw material, while both species are the main species in the National Natural Forest Protection Project [20]. Therefore, accurate estimation of biomass of larch and Simao pine is not only beneficial to forest resource management and conservation but also can accurately calculate carbon stock. Currently, fewer studies are using Sentinel-1 SAR data and Sentinel-2 multispectral data for the inversion of AGB of larch and Simao pine, and there are uncertainties in the inversion results. In this study, two types of coniferous forests, Xing'an larch in Inner Mongolia, and Simao pine in Yunnan Province were selected to explore the suitable RS feature optimization inversion algorithms for their forest AGB estimation. RS features extracted from Single Sentinel-1, single Sentinel-2 and combination of Sentinel-1 and Sentinel-2 were input for the forest AGB inversion. The objective of this study is to explore the advantages of coniferous forest AGB estimation in typical regions in China under different forms of combinations of the two RS data supported by different feature optimization inversion algorithms. Meanwhile, explore the robustness and transferability of these inversion algorithms to enhance the monitoring and decision-making capabilities for the forest resources management.
Detailed Comment 5
A reference should be provided for Eq. (1).
Detailed Response 5:
Thank you for the reviewer's comments. I have made the modifications as per your request.
The AGB of each tree in the sample plots was calculated with equation (1) [21], and the AGB value of each tree in each plot was totaled to obtain the AGB of each plot, then the area of each sample plot (0.04 ha) was divided to obtain the AGB of each plot.
- Xu, H.; Zhang, Z.Y.; Ou, G.L. Estimation and distribution of forest biomass and carbon stocks in Yunnan province. Yunnan Science and Technology Press, 2019.
Detailed Comment 6
Sentine-2 in lines 207 and 209 should be corrected.
Detailed Response 6:
Thank you for the reviewer's comments. I have made the modifications as per your request.
Detailed Comment 7
Figure 4 is not referred to in the text.
Detailed Response 7:
Thank you for the reviewer's comments. I have made the modifications as per your request.
The basic principle of KNN-FIFS is shown in Fig 4:
Detailed Comment 8
Figure number in line 344 should be checked.
Detailed Response 8:
Thank you for the reviewer's comments. I have made the modifications as per your request.
The optimized RS feature selection and related constructed models are shown in Table 3.
Detailed Comment 9
“… my research results …” in line 534 should be revised as “results of this study”.
Detailed Response 9:
Thank you for the reviewer's comments. I have made the modifications as per your request.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe legends in the figures are not legible and need improvement for better clarity.
Comments on the Quality of English LanguageOverall, the language is quite comprehensible, but a thorough proofreading focusing on sentence structure and grammar could enhance its readability.
Author Response
Response to Reviewers’ Comments on the manuscript (No. 2755235) with:
This manuscript was submitted to Forests (No. 2755235). As per the reviewers’ suggestions, it is now revised and submitted to the journal along with the responses to comments from the editor and reviewers.
The authors are indebted to the anonymous reviewers for their constructive criticism, which has been instrumental in improving the quality of the manuscript.
In the following we transcribe the comments in gray italics, and our responses in blue in a response section. The changes are highlighted in red for which changed in the revised manuscript.
General Comments to the Author
Detailed Comment 1:
The legends in the figures are not legible and need improvement for better clarity.
Detailed Response 1:
Thank you for your comments. We have improved all the figures according to your advices.
Detailed Comment 2:
Overall, the language is quite comprehensible, but a thorough proofreading focusing on sentence structure and grammar could enhance its readability.
Detailed Response 2:
Thank you for your comments. We read thorough all the manuscript, corrected some language errors, and modified some sentence structure. We changed the modified contents into red colors in the revised manuscript.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe authors have addressed my concerns properly. The only matter is the resolution of the figures that is low and should be checked by the editor.
Comments on the Quality of English Language-
Author Response
Response to Reviewers’ Comments on the manuscript (No. 2755235) with:
This manuscript was submitted to Forests (No. 2755235). As per the reviewers’ suggestions, it is now revised and submitted to the journal along with the responses to comments from the editor and reviewers.
The authors are indebted to the anonymous reviewers for their constructive criticism, which has been instrumental in improving the quality of the manuscript.
In the following we transcribe the comments in gray italics, and our responses in blue in a response section. The changes are highlighted in red for which changed in the revised manuscript.
General Comments to the Author
Detailed Comment 1:
The authors have addressed my concerns properly. The only matter is the resolution of the figures that is low and should be checked by the editor.
Detailed Response 1:
Thank you for your comments. We have improved the resolution of all the figures according to your advices.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe manuscript can be published with the following minor corrections:
1) Language correction should be made on lines 24, 458, 460, 461, 540-542.
2) The sentence on lines 501-502 needs to be cited.
Comments on the Quality of English LanguageMinor editing is required.
Author Response
Response to Reviewers’ Comments on the manuscript ((No. 2755235) with:
This manuscript was submitted to Forests (No. 2755235). As per the reviewers’ suggestions, it is now revised and submitted to the journal along with the responses to comments from the editor and reviewers.
The authors are indebted to the anonymous reviewers for their constructive criticism, which has been instrumental in improving the quality of the manuscript.
In the following we transcribe the comments in gray italics, and our responses in blue in a response section. The changes are highlighted in red for which changed in the revised manuscript.
General Comments to the Author
Detailed Comment 1:
Language correction should be made on lines 24, 458, 460, 461, 540-542.
Detailed Response 1:
Thank you for your comments. We read thorough all the manuscript, corrected some language errors, and modified some sentence structure. We changed the modified contents into red colors in the revised manuscript.
Lines 24 changed as:
The results indicated that combination of features extracted from Sentinel-1 and Sentinel-2 can improve the inversion accuracy of forest AGB and KNN-FIFS algorithm had robustness and transferability in forests AGB inversions.
lines 458, 460, 461 changed as:
The estimation results from KNN-FIFS have the lowest RMSE values and estimation results from RF has the highest RMSE values. But in the case of using single Sentinel-1 data in Genhe, the RMSE value from MLSR is the highest (35.38 t/hm2). The values of rRMSE also confirmed the best performance of KNN-FIFS with the lowest value of 12.4%.
lines 540-542 changed as:
The results of from them were slightly higher than the results of our study with R2 = 0.770 and RMSE = 22.74 t/hm2. The higher R2 value and lower RMSE value may result from the involved P-band SAR features, which have better penetration capability in forest and can interact with the large branches and trunks that accounted for main AGB in forest [30].
Detailed Comment 2
The sentence on lines 501-502 needs to be cited.
Detailed Response 2:
Thank you for your comments. We have made the modifications as your advices.
Guo et al. used single Sentinel-2 data as data source, constructed MLSR model to invert coniferous forest AGB in Inner Mongolia, the R2 of the inversion results was 0.765 and RMSE was 39.49 t/hm2[41].
[41] Guo, Z.Q.; Zhang, X.L.; Wang, Y.T. Ability evaluation of coniferous forest aboveground biomass inversion using Sentinel-2A multiple characteristic variables. Journal of Beijing Forestry University, 2020, 42(11): 27−38.
Author Response File: Author Response.pdf
