Improving the Accuracy of Estimating Forest Carbon Density Using the Tree Species Classification Method

Round 1

Reviewer 1 Report

This study takes the tree forest in Yueyang City, Hunan Province as the research object, and use the random forest classification algorithm to classify the dominant tree species in the the study area. Based on the classification, MLR, SVM and RF were constructed for different dominant tree species by combining Forest Resource Inventory data and remote sensing data.  Results showed that method used in this study eliminates the problems of severe overfitting and significantly underestimation of peak values when estimating under unclassified conditions in some extent. But there are some problems need to be considere.

1) In the introduction section, it is suggested to add the latest research status of forest species classification and forest carbon density inversion using remote sensing technology.

2) The acquisition time of both ground data and remote sensing data is relatively early, and it is recommended to use the latest data for the experiment.

3) In the parts of tree species classification and forest carbon density modeling, the latest deep learning models can be appropriately considered to incorporate, in order to further improve the accuracy.

4) It is recommended to remove the outer border of Figure 1.

5) Suggest adding a latitude and longitude grid to Figures 2, 3, 4 and 5 respectively.

6) It is recommended that the corresponding RMSE values be added separately for each subgraph legend of Figure 6; In addition, it is recommended that all subplots of Figure 6 have uniform maximum axis values.

7) In Table 1, suggest changing "Forest type" to "tree species group".

8) In Table 2, it is proposed to change the formula of NDVI to (NIR-red)/(NIR+red).

9) Some indicators in Tables 3, 4 and 5 do not have units and it is proposed to add them.

Author Response

请参阅附件。

Author Response File: Author Response.docx

Reviewer 2 Report

General recommendations

1.      Provide graphical abstract for the paper which represents the whole scheme of study.

Introduction

1.      Introduction is very weak. The author should take start from the general and then conclude the introduction towards the specific. And at the end the write one paragraph on the introduction of plant, including its taxonomic characteristics, distribution and ecology etc.

RESULTS

Classification results for dominant species

1.       Why you have taken the forest cover of Yueyang City 2013? It is 8-9 years old.

2.       The distribution range of some areas of LUCC_2013 is more obviously different from that of Forest Resource Inventory data. What is the reason?

3.       The dominant tree species in the forest range of the study area were classified using RF and SVM machine learning algorithms based on primary classification. What is the reason for selecting this method?

4.       Comparing the validation samples with the classification results of the 2 classification algorithms, the results based on the accuracy calculations showed that the OA for classification using RF reached 0.8730, the Kappa coefficient was 0.7747, where the PA for both Poplar and Soft broadleaf classes was higher than 0.9, and the classification was better. What is the reason?

5.       Because of the complex topography of the Yueyang forest area, the classification units of the dominant tree species are fragmented, resulting in a "salt and pepper noise" in the final classification results. What is the reason?

Forest carbon density inversion mapping

1.       the 331 inversion results after classification by dominant tree species show that the forest carbon stock in Yueyang City in 2013 was 10.2975 Tg, and the spatial distribution of forest carbon density values ranged from 3.06 to 62.80 t·hm-2; the inversion results without classification showed that the forest carbon stock was 10.7405 Tg, and the spatial distribution of forest 335 carbon density values ranged from 4.64 to 31.96 t·hm-2, the carbon density values were 336 severely over-fitted. What is the reason?

 

2.       In terms of geospatial distribution patterns (Figure 8), forest carbon density in the city is higher in Huarong Xian (20.23 t·hm-2) and Xiangyin Xian (20.04 t·hm-2) and lowest in Yueyang Xian (16.10 t·hm-2). What is the reason?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

1）The introduction is still a little weak, it is suggested to give more introduction of the main methods of the current forest carbon density inversion, such as advantages, disadvantages and the latest research progess.

2) The acquisition time of both ground data and remote sensing data is relatively old, please give detailed reasons in the revised draft.

3) In parts of tree species classification and forest carbon density modeling, why the author choose the models of MLR, SVM and RF, please give sufficient reasons in the revised draft.

4) In Table 1, Tble 5 and Table 6, suggest changing "Forest type" to "tree species and group".

5) In Table 6, suggest adding units for RMSE and MAE.

Author Response

Please see the attachment.

Author Response File: Author Response.docx