Revealing coastal vegetation structural diversity through LiDAR-derived relative entropy

Qi, Guanpu; Wei, Jiawei; Xin, Zaizhou; Yao, Longcheng; Chen, Xiong; Ma, Jun

doi:10.5194/essd-2026-68

Preprints

https://doi.org/10.5194/essd-2026-68

Preprints

23 Mar 2026

| 23 Mar 2026

Status: this preprint is currently under review for the journal ESSD.

Revealing coastal vegetation structural diversity through LiDAR-derived relative entropy

Guanpu Qi, Jiawei Wei, Zaizhou Xin, Longcheng Yao, Xiong Chen, and Jun Ma

Abstract. Coastal wetlands are among the most valuable ecosystems globally, due to the high ecological function of their structurally complex vegetation communities. However, there remains a lack of vegetation structural complexity (VSC) indicators tailored for coastal wetland applications. Here, we developed a new VSC index, vegetation structure relative entropy (VSRE), based on a measure of the asymmetry of the difference between two probability distributions. While the currently used VSC index all fail to capture the discrete and continuous complexity gradients of coastal vegetation communities, VSRE demonstrated ideal performance in these applications, exhibiting strong robustness across varying point cloud densities. By applying this indicator to 1,337 LiDAR samples of natural coastal vegetation, we used Alpha Earth Foundation data and a deep learning model to create a seamless VSRE spatial map of coastal wetlands in China, with high spatial resolution (10 m) and accuracy (R² = 0.96). VSRE mapping provides crucial ecosystem structural information beyond vegetation classification data and conventional optical indices, highlighting the high spatial heterogeneity of VSC in coastal wetlands. This study offers a valuable foundation for prioritizing conservation areas and enhancing the resolution and accuracy of coastal zone ecological modelling.

Received: 26 Jan 2026 – Discussion started: 23 Mar 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Preprint (PDF, 2414 KB)

Supplement (1263 KB)

Download & links

Guanpu Qi, Jiawei Wei, Zaizhou Xin, Longcheng Yao, Xiong Chen, and Jun Ma

Status: open (until 14 May 2026)

Post a comment Subscribe to comment alert

RC1:
'Comment on essd-2026-68', Anonymous Referee #1, 05 Apr 2026 reply
Comments on “Revealing coastal vegetation structural diversity through LiDAR derived relative entropy” by Qi et al.

General comments
Vegetation structural complexity (VSC) is increasingly recognized as a key attribute influencing ecosystem processes and functions. The quantification of VSC from 3D LiDAR data has therefore become an important area of research to support such studies. In this work, the authors investigate methodologies for quantifying VSC specifically in coastal ecosystems, propose a new index, namely VSRE, and demonstrate that this metric outperforms existing approaches. They further combine this metric with the latest Google AEF dataset to produce a nationwide VSC product. Overall, this is a timely study for advancing VSC research in coastal ecosystems, and the resulting dataset is valuable. However, I have several major concerns that prevent me from recommending its publication in its current form.
First, I have difficulty understanding the rationale for using relative entropy to quantify VSC. Fundamentally, relative entropy measures the dissimilarity of LiDAR point distributions between different voxels. It is unclear how this directly reflects vegetation structural complexity. By definition, structural complexity relates to both the randomness of canopy element distribution and the extent of space occupation. In this sense, VSC should capture at least two components: the abundance of structural elements and the variability in their spatial arrangement. However, strong dissimilarity between layers does not necessarily indicate high structural complexity. For instance, in tropical forests, the canopy is often densely filled due to complementary layering, resulting in relatively similar element distributions across layers. Despite this similarity, such forests exhibit high structural complexity. Therefore, I remain fundamentally uncertain about the appropriateness of using relative entropy as a metric for quantifying VSC.
Second, the comparison among metrics appears methodologically unclear and potentially unfair. The authors used three groups of plots—categorized as low, medium, and high VSC—to evaluate the performance of the proposed metric against existing ones. However, it is not clear how these groups were defined in the field. Were they based on visual assessment or some quantitative criteria? Given that VSC is inherently difficult to measure directly, establishing reliable ground-truth data is challenging without the use of simulation or well-defined structural proxies. Furthermore, I do not consider the comparison between VSC and species richness to be a robust or appropriate way to demonstrate the superiority of the proposed metric. The relationship between VSC and species richness remains debated in the literature and is not necessarily positive or consistent. For example, in northern temperate forests, single-layer, monocultural stands can still exhibit relatively high VSC due to selection effects and high space occupation.
Third, the overall writing of the manuscript requires improvement. The current version attempts to integrate a wide range of content, which reduces its focus, while several important components lack sufficient detail. For instance, the classification of coastal vegetation types is itself a substantial task, yet it is only briefly described. A similar issue arises with the nationwide VSC mapping, where methodological and implementation details are limited. I therefore suggest that the authors consider dividing the work into two separate papers: one focusing on the technical development and validation of the proposed VSC metric, and another dedicated to its application in national-scale mapping.
Below are some specific comments for the authors to consider.

Specific comments
Title: The current title doesn’t seem like a data paper.

Line 38: “the their”. Typo.

Line 58-59: Numerous studies have already addressed this question in terrestrial ecosystems. Therefore, the discussion should be more focused on the functional and structural characteristics of coastal ecosystems.

Line 68-71: What are the unique challenges associated with quantifying canopy structural complexity in coastal ecosystems? It would be helpful to include one or two sentences to clarify this point.

Line 72-90: This content belongs in the Methods section. In the Introduction, you may briefly summarize the data and overall approach of the study, but detailed procedural steps should not be included.

Line 92: To be honest, this section is somewhat disorganized. The descriptions of fixed experimental plot data and drone-based LiDAR data should be clearly separated. At present, key information regarding the drone LiDAR data used for national mapping is missing. For example, the manuscript does not specify where these plots are located or what their spatial extent is.

Line 93: This sentence is misleading, as it suggests that the entire study was conducted at a single site. In fact, the site was only used for method validation.

Line 104-105: Beyond point density, the vertical penetration capability of the LiDAR sensor is critical for accurately measuring canopy structural complexity. To my knowledge, the DJI L1 sensor has limited ability to penetrate dense canopies. It would therefore be helpful to include a vertical profile visualization to illustrate this limitation, allowing readers to better understand the characteristics and potential constraints of the data

Line114: “numerous” It would be better to provide a specific number.

Line 119: This represents a very wide range of point densities. Additional information is needed to justify the experimental design and the selection of point density levels.

Line 141: Did you measure tree height manually in the field? If so, what is the purpose of using LiDAR-segmented individual trees?

Line 155: I am not convinced that this index effectively quantifies canopy structural complexity. Rather, it appears to primarily capture the dissimilarity in vertical canopy distribution. Please see my general comment.

Fig. 2: This figure is not an effective illustration. It does not clearly convey how the metric is designed or calculated, and instead only shows a set of voxels without sufficient explanation.

Line 233: Five? But you only mentioned four here.

Line 239-306: This is an interesting study; however, the manuscript does not allocate sufficient space to adequately present the results. Key outcomes, such as classification and mapping accuracy, are only mentioned in the Methods section, which is inappropriate. Please refer to my general comments for further details.

Line 305: This level of accuracy suggests potential overfitting, which is a known issue for MLP models when trained on a limited number of samples.

Line 332: you have defined the abbreviation of VSC before.

Fig. 3: How were the three VSC groups defined? I could not find sufficient details on this. Notably, most existing metrics indicate that the “medium” group corresponds to relatively low VSC, whereas only your metric suggests otherwise. This raises questions about the criteria used to establish these groupings, particularly considering the difficulty to quantify VSC in real world.

Line 362: Point density is not the only factor that should be considered (see my previous comments). It would be helpful to include a vertical profile of the LiDAR data to better illustrate its structural characteristics. To my knowledge, the DJI L1 sensor primarily captures returns from the upper canopy and has limited penetration into lower layers. As a result, even with increasing point density, the additional points may still be disproportionately concentrated in the upper canopy. This could artificially increase relative entropy by enhancing apparent dissimilarity between layers, without necessarily reflecting changes in VSC.

Line 386: High species diversity does not necessarily mean high canopy structural complexity. Please see my general comment.

Line 471-484 and Fig. 9: I may be misunderstanding something here, but I am not clear how these conclusions are derived. Wouldn’t a mature forest with strong canopy complementarity exhibit a near-random distribution of canopy elements, and therefore correspond to high entropy?

Line 493-494: I am lost again here. What do you mean by structural reproducibility?

Line 507: I did not put too much focus on the discussion on the national mapping here, because I think this should be put in a separate paper.

Reply
Citation: https://doi.org/10.5194/essd-2026-68-RC1

Guanpu Qi, Jiawei Wei, Zaizhou Xin, Longcheng Yao, Xiong Chen, and Jun Ma

Supplement

https://doi.org/10.5194/essd-2026-68-supplement

Data sets

VSRE mapping for “Revealing coastal vegetation structural diversity through LiDAR-derived relative entropy” Guanpu Qi et al. https://doi.org/10.6084/m9.figshare.30588722

Model code and software

VSRE Calculator for LiDAR Point Clouds Guanpu Qi et al. https://github.com/EmpTyset-phi/VSRE

Guanpu Qi, Jiawei Wei, Zaizhou Xin, Longcheng Yao, Xiong Chen, and Jun Ma

Viewed

Total article views: 128 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
94	28	6	128	20	8	7

HTML: 94
PDF: 28
XML: 6
Total: 128
Supplement: 20
BibTeX: 8
EndNote: 7

Views and downloads (calculated since 23 Mar 2026)

Month	HTML	PDF	XML	Total
Mar 2026	59	16	2	77
Apr 2026	35	12	4	51

Cumulative views and downloads (calculated since 23 Mar 2026)

Month	HTML	PDF	XML	Total
Mar 2026	59	16	2	77
Apr 2026	35	12	4	51

Viewed (geographical distribution)

Total article views: 129 (including HTML, PDF, and XML) Thereof 129 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 13 Apr 2026

Download

Preprint (2414 KB)
Metadata XML

Short summary

Using LiDAR scanning data and a deep learning model, we developed a new algorithm and produced a high-resolution dataset mapping vegetation structural complexity across China's coastal wetlands. Unlike existing metrics, this approach reliably captures subtle structural differences in coastal vegetation. The resulting publicly available maps reveal strong spatial heterogeneity and provide a valuable data foundation for wetland conservation, ecosystem assessment, and coastal ecological modelling.


Total:	0
HTML:	0
PDF:	0
XML:	0