Enhancing Forest Canopy Height Retrieval: Insights from Integrated GEDI and Landsat Data Analysis

Round 1

Reviewer 1 Report

I want to thank the authors for the interesting paper. I have provided my comments below to better prepare the article for potential publication in Sustainability.

Line 42: This is a general comment. If possible, try to incorporate examples that use optical sensors in canopy height detection.

Line 44 and 45: Landsat is a moderate-resolution satellite. In addition, MODIS can provide remote sensing imagery at 0.1 degree which is around 10 km. How 10km images can be considered as high-resolution imagery?

Line 45: Landsat is not a sensor but a satellite. 

Line 46 and 47: "By analyzing the spectral characteristics and spatial patterns of vegetation, optical sensors can infer the height of forest canopy": very general statement. Remote sensing and spectral characteristics and spatial patterns of vegetation can be used to identify many other different phenomenon as well. Please be concise by integrating examples:

Example: changes in the visible spectrum (red and near-infrared) is adopted in xx studies to identify the height of black spruce patches in southern ......

Line 68 to 70: please read the entire article and consider shortening and eliminating unnecessary lengthy sentences:

Instead of "the diversity of available ........machine learning algorithms" you might want to say: A variety of data has led to diverse methods for mapping forest canopy height, which can be grouped into parametric algorithms, physical models, and machine learning algorithms.

Line 85 to 91: "The BP neural network, as a machine learning... novel perspectives for canopy height retrieval": how about other machine learning algorithm? what you didn't include other studies that used other type of ML models? For example, I can talks about the benefits of Random Forest model as a machine learning tools, easy adoption, selection of regression trees, non-overfitting model, and several other benefits.

Line 92 to 94: What is your previous discussion? are your readers responsible to recall what was your previous discussion? or where you published the earlier work? please consider removing this statement or providing sufficient information and details for audiences.

Line 95: what do you mean by "improved image quality and accuracy"? spatial resolution improvement? (from 30 meter to ???) image retrieval method improvement? how accuracy is improved? if you do not want to focus on these details, please consider removing redundant details.

Line 104 to 117: As a reader of your work, I would like to first see your study area. I recommend you to follow a simple yet organized method for your "materials and method" section as below:

A)study area: please explain where is your area of interest and why: for example: southern yy as is undergoes dramatic changes of vegetation towards the recent years (....2019).

B) collected materials: you can highlight all satellites, retrieval time and date, the spatial, spectral, and temporal resolution of your sensors and satellites (preferably in a table).

C) methods: any used models, attributes of models, training and testing sets, errors in data, data cleaning, data wrangling, manipulation, and evaluation.

D) your expected outcomes: comparison between canopy heights in status quo and modeled data ......

With this in mind, you started your "materials and method" with collected data in a scattered way.

Lines 120 to 132: should be placed in the beginning of your "M&M" section.

Lines 251 to 254: "Insufficient neurons can result in underfitting, where the model fails to compute complex patterns and exhibits poor performance. On the other hand .... poorly on unseen data" : it this coming from your observation or originated in the previous literature, if from previous literature, please cite them, of yours, please move it to the implications, results and discussions, etc..

Line 254  to 259: please see the comment above.

Line 267 to 268: it is common knowledge that the higher the number of spectral bands as inputs, the better the accuracy of your model would be. 

Line 269 to 270: "However, it may lack precise vertical information": this is self-defeating. You incorporate a wide range of spectral information to identify the canopy height, however, they lack precise vertical information. Please justify your statement to better transfer your model. For example, the integration of vegetation indices rooted in the vegetation spectrum along with lidar data can better identify the canopy height which is original in this study.

Line 275 to 283: I am confused about how you integrated vegetation indices instead of bands alone. All your data coming from bands and you can find the relative importance of bands in forming your model, however, you decided to go with indices instead. The issue that may arise in these cases is the use of a single band in multiple indices.

For example, EVI, NDVI, and SAVI all come from B5 and B4. What is the purpose of adding all these indices? better to only find the contribution of bands in estimating the canopy height.

Figure 6: this type of outcome are very well expected. For example, NDVI is pretty much the same as SAVI (see the comment above) and both are considered as unimportant factors. While RHs are relatively important which is directly coming from relative height metrics.

 

 

 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Forest canopy height serves as a vital parameter for quantifying forest biomass and carbon storage. In this manuscript, GEDI L2A version 2 data, Landsat 8/9 and airborne laser scanning data were employed to assess the accuracy of CHM extracted from GEDI data. The results demonstrated that algorithm a2 demonstrated higher accuracy than the other algorithms in detecting ground 16 elevation and canopy height. After combining with Landsat data, integrating GEDI and OLI-2 data exhibits the highest performance. The experimental scheme and process design are scientific and reasonable, but the innovation is not very clear.

 

 

1.  In the part of abstract, describing the methods and results should be further improved, and it is necessary to express the innovation of the manuscript more clearly.

2.  In the part of introduction, Landsat and MODIS images can not be defined as high resolution images. Commonly, they are medium and low-resolution images.

3. “Machine learning algorithms can automatically extract features and handle intricate nonlinear relationships, resulting in higher predictive accuracy”. Normally, feature extraction and feature selection methods should be applied using Machine learning algorithms. Deep learning algorithms can automatically extract features.

4. in your manuscript, CHM was obtained by subtracting the DTM from DSM. Why do you choose DTM, not DEM?

5. Please add coordination for figure 3,

Check the grammar and expression errors in the manuscripts, and further improve the coherence of paragraphs.

Author Response

Please see the attachment.

Author Response File: Author Response.docx