Uncertainty Analysis of Remote Sensing Estimation of Chinese Fir (Cunninghamia lanceolata) Aboveground Biomass in Southern China
, Dongbo Xie 1, Zheyuan Wu 1, Yuancai Lei 1, Jinsheng Ye 3 and Qiulai Wang 3
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe aim of the paper is quantify uncertainties at the plot and model scale on AGB estimations using estimation models: k-nearest neighbor regression (KNN), Gradient boosted regression (GBRT), and random Forest (RF). I think that the paper can be accept after revision. My decision is major revision. You can find my comments to paper below:
1-Line 31. Please use different key words from the title.
2-Lİne 42. please remove one of the points from text.
3-Lines 45-71. Please add a few reference or the results of studies to showing the superiority of remote sensing based methods over traditional methods to obtain AGB. This part is too narrow.
4-Equation 1. Is W show dry biomass or fresh biomass please specify.
5-Equation 1. Please use same abbreviation for the terms in the text for diameter at breast height (D or DBH) and tree height (H).
6- Line 224. Please add "approximately" before seventy percent.
7-Line 227. please add explanation for RMSE and rRMSE to text.
8- Line 145. "Data Sources and Processing" Section. Please give us extra information about relationship between field data and remote sensing data (from per-tree biomass to remote sensing model). This part is too narrow.
9-Line 195. Why did you select these methods? As you known, there are different methods as Support Vector Machine and XGBoost Regression.
10-Line 122 "Study area" section. please give us information about distribution of sample plots three area. How many sample plot are there each one?
11- "Analysis of Plot-scale Uncertainty" Section. What is the reason of these differences among estimation methods. Please add extra information about it. This section should be expanded by including additional components.
12-Line 239. Please use "observed values" instead of "observed measurement".
13-Line 244. please add an explanation for this equation. to text.
14-Equation 9 what is the meaning of "y" in the equation.
15-Lines 409-413. What is your recommendation minimal plot numbers for accuracy estimation of AGB using remote sensing model scale.
16-Line 133-135. You said that study have three different vegetation type: tropical rainforest, evergreen broad-leaved forests, and coniferous forest. But, you did not say anything about effects vegetation types on AGB estimation in the your results. Please add something about it to text.
17-Why did you not compare your results with those from traditional field survey methods?
18-Line 448. "Influence of Plot Size and Positioning Errors" section. What is your results about plot size and positioning errors. Please add somethings about it to text.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authorsfunctions are easily limited by terrain, which makes it unsuitable for large-scale and complex terrain areas? I stunned by this statement which makes whatsoever what authors try to convey the message. Thus far LiDAR has been effectively used for AGB modeling way batter than other remote sensing techniques - I agree cost is an issue, yet LiDAR is a gold standard in remote sensing and AGB modeling including all platforms ground based TLS, MLS, aerial, ALS, UAS-LS and space borne - GEDI.
Optical remote sensing suffer from saturation problems and light cannot penetrate through dense canopy. What about 3-D stratification of dense forest ecosystems - LiDAR can do that yet optical remote sensing cannot.
No need to mention free download, most of the medium resolution satellites are providing data free of cost. I would say thus far LiDAR is best and can be fused with optical data for wall-to-wall mapping.
Uncertainty encompasses a broad range of concepts, including inaccuracy, fuzziness, 73 and ambiguity. In the context of forest ecosystem productivity, which represents a critical 74 indicator, three primary categories of uncertainty are pertinent to calculations of forest 75 biomass: model uncertainty, measurement uncertainty, and sampling uncertainty [21,22]. this recent paper also discusses and quantify such uncertainties and should be part of literature review. Evaluating the impact of field-measured tree height errors correction on aboveground biomass modeling using airborne laser scanning and GEDI datasets in Brazilian Amazonia - ScienceDirect
A large portion of your literature seems to be originating from above article e.g., "Shettles et al. [23] posited that model uncertainty constitutes the primary source of 77 uncertainty, representing approximately 70% of the overall uncertainty. Their study 78 focused exclusively on examining model uncertainty and did not consider measurement 79 and sampling uncertainties. Model uncertainties associated can be categorized into four 80 primary dimensions: inherent uncertainty of the input variables, inaccuracies arising 81 from the specification of the model structure, residual variance error of the model, and 82 errors related to the model parameters. Input variable uncertainty primarily pertains to 83 the measurement errors of these variables, specifically, the diameter at breast height 84 (DBH) and tree height (H). Such errors may be influenced by the measurement 85 instruments and techniques and the methodologies applied by the individuals 86 conducting the measurements. The inaccuracies in establishing model frameworks 87 primarily stem from the insufficient application of advanced modeling techniques [24] or 88 validation data [25]" yet the above paper is not inlcuded in the citations.
As approaches 1, model can be overfitting as well.
in AGB estimation wood density is also an important input parameters. But equation 1 does not have wood density. Authors should justify why the wood density is not inclusive of AGB modeling.
Study area does not have total area coverage in km.sq or m.sq.
Results,
why KNN performance was poor. Being non-parametric models, the GBRT and RF should have similar performance both models validating each other with similar performance to ensure that models are not over or under fitted. nBias is one evaluation parameter that ensures the model over and under estimations.
Authors did not provide 30 m by 30 m AGB map of the study area.
the following article did way batter than proposed approach and it also explains everything about research starting from field plot data, AGB estimates, model training and evaluation. AGB maps, evaluation of over and underestimates. Aboveground biomass modeling using simulated Global Ecosystem Dynamics Investigation (GEDI) waveform LiDAR and forest inventories in Amazonian rainforests - ScienceDirect.
Authors should use it a guide line to reconstruct their paper. Also, authors did not performed the importance of features used in this study. models parameters are not provided as well. overall research is not well formulated and presented.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsAll my comments are responded by the authors. I think this form of the paper can be accept to publish.
Reviewer 2 Report
Comments and Suggestions for AuthorsDear authors,
Thank you for providing the revise version of the original submission. I can see authors have made changes which in result improve the manuscript to publication level.
