Next Article in Journal
Assessing Forest Succession Along Environment, Trait, and Composition Gradients in the Brazilian Atlantic Forest
Previous Article in Journal
Moisture Seasonality Dominates the Plant Community Differentiation in Monsoon Evergreen Broad-Leaved Forests of Yunnan, China
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Accurate Evaluation of Urban Mangrove Forest Health Considering Stand Structure Indicators Based on UAVs

1
School of Environment and Energy, Peking University Shenzhen Graduate School, Shenzhen 518055, China
2
Guangdong Mangrove Engineering Technology Research Center, Peking University Shenzhen Graduate School, Shenzhen 518055, China
*
Author to whom correspondence should be addressed.
Forests 2025, 16(7), 1168; https://doi.org/10.3390/f16071168
Submission received: 29 April 2025 / Revised: 17 June 2025 / Accepted: 22 June 2025 / Published: 16 July 2025
(This article belongs to the Section Forest Inventory, Modeling and Remote Sensing)

Abstract

Stand structural configuration dictates ecosystem functional performance. Mangrove ecosystems, located in ecologically sensitive coastal ecotones, require efficient acquisition of stand structure parameters and health assessments based on these parameters for practical applications. Effective assessment of mangrove ecosystem health, crucial for their functional performance in ecologically sensitive coastal ecotones, relies on efficient acquisition of stand structure parameters. This study developed a UAV (Unmanned Aerial Vehicle)-based framework for mangrove health evaluation integrating stand structure parameters, utilizing UAV visible-light imagery, field plot surveys, and computer vision techniques, and applied it to the assessment of a national nature reserve. We obtained the following results: (1) A deep neural network, combining UAV visible-light data with tree height constraints, achieved 88.29% overall accuracy in simultaneously identifying six dominant mangrove species; (2) Stand structure parameters were derived based on individual tree extraction results in seedling zones along forest edges (with canopy individual tree segmentation accuracy ≥ 78.57%), and a stand health evaluation model was constructed; (3) Health assessment revealed that the core zone exhibited significantly superior stand health compared to non-core zones. This method demonstrates high efficiency, significantly reducing the time and effort for monitoring, and offers robust support for future mangrove forest health assessments and adaptive conservation strategies.

1. Introduction

Mangrove forests are intertidal woody plant communities distributed along tropical and subtropical coastlines, uniquely adapted to saline, waterlogged, and low-oxygen environments [1]. Supported by these structural traits, mangroves provide a wide range of ecological services such as coastal stabilization, tsunami mitigation, water purification, carbon sequestration, and biodiversity conservation, while also supporting livelihoods through fisheries and tourism [2,3,4]. Despite these essential services, mangrove ecosystems have experienced widespread degradation, with global cover declining by 35% from 1980 to 2000, and a further 2% (3363 km2) lost between 2000 and 2016 [5,6]. In contrast, China has made substantial progress in mangrove restoration. Although southern mangrove cover declined by more than 50% since the 1950s due to rapid urbanization [7], government-led conservation and blue carbon initiatives since the 1980s have helped expand the national area to over 22,000 ha [8,9,10]. Nonetheless, major challenges remain in current mangrove restoration and management efforts: (1) Overreliance on monocultures (e.g., exotic fast-growing species of Sonneratia apetala) suppresses biodiversity and impairs multifunctionality due to homogenized stand structures [11]; (2) Restoration assessments predominantly evaluate temporal afforestation outcomes rather than structural indicators critical for ecological health [9,12,13,14,15], leaving stand optimization strategies underdeveloped; (3) Most assessments for mangrove ecological restoration are small-scale, taxon-specific evaluations [16,17], lacking systematic frameworks to guide long-term management [18,19]. Structure determines function, and rational stand structures are a prerequisite for fully realizing the multiple ecological functions of forests [20,21]. Therefore, it is imperative to quantify and optimize mangrove stand structures using spatial indicators to enhance ecological performance and align artificial restoration with the resilience of natural systems.
The current research on terrestrial forest stand structures primarily relies on plot-based field surveys, focusing on fundamental indicators such as tree height, diameter at breast height (DBH), species composition, and crown width [22,23]. These parameters are further analyzed through diameter class distributions, height structures, and species diversity metrics to characterize stand structural attributes. In China, preliminary studies on mangrove stand structure have been conducted using traditional ground-based plot surveys. For instance, Liang et al. [24] investigated the Sonneratia caseolaris and Sonneratia apetala community in the Futian Mangrove Reserve of Shenzhen Bay, while Fu and Liu [25,26] examined mangrove structures in the Leizhou Peninsula and Techeng Island. Comparative studies on spatial structures across different tidal zones of mangrove stands were conducted by Li et al. [27], and subsequent simulations for spatial structure optimization and health management were developed for mangroves in Zhanjiang [28] and Zhangjiang Estuary [29]. Internationally, Azman et al. [30] evaluated the recovery duration of secondary mangroves to natural forest structural levels using conventional parameters, biodiversity indices, and biomass metrics. Additionally, Huang et al. [31] applied spatial structure indices like mingling degree and health index to analyze Rhizophora stylosa communities on Techeng Island, emphasizing the critical role of spatial structure adjustment in ecosystem restoration. These studies on natural mangrove stands provide valuable insights for optimizing artificial mangrove plantations. However, current methodologies remain constrained by labor-intensive ground surveys with limited spatial coverage [32], highlighting the need for scalable, cost-effective monitoring approaches.
At present, satellite remote sensing data have been employed in mangrove stand structure surveys; however, such data are limited to the landscape scale [33,34]. UAV (Unmanned Aerial Vehicle) remote sensing offers significant advantages by overcoming some limitations, cost, and timeliness. Preliminary attempts to use UAVs for large-scale spatial structure analysis have been made in other terrestrial forest ecosystems, such as in Northeast China, where encouraging results have been achieved [35]. The studies utilizing long-term Landsat remote sensing data and machine learning methods to monitor mangrove spatial dynamics and structure more systematically [36]. Furthermore, advances have demonstrated that mangrove biomass estimation can be achieved based on high-resolution remote sensing technologies [37], while canopy segmentation can be optimized through fusion models combining radar and optical data [38]. The application of UAV remote sensing, especially visible-light imagery, to the rapid assessment and structure-oriented optimization of mangrove forests has shown promising potential for operational deployment [39,40,41].
In summary, to advance towards near-natural mangrove restoration guided by ecological principles and to address the limitations of existing monitoring methods, there is an urgent need to establish a comprehensive, cost-effective, and systematic evaluation framework that integrates detailed stand structure parameters for rapid and accurate assessment of mangrove forest health. This direction is also aligned with global sustainability agendas such as the United Nations Sustainable Development Goals (e.g., SDG 13 and SDG 15), where vegetation monitoring using multi-source remote sensing and machine learning has been recognized as a vital strategy for land degradation mitigation and ecological resilience [42]. Addressing this critical research gap, this study proposes a novel assessment methodology that utilizes high-resolution UAV visible-light imagery and incorporates quantifiable stand structure parameters derived from individual tree analysis to evaluate mangrove health. The proposed methodology was applied to different ecological functional zones of seedlings within the Futian Mangrove Nature Reserve in Shenzhen Bay to verify its applicability, demonstrating an innovative approach for scientifically informed mangrove conservation and management.

2. Method and Materials

2.1. Study Site

The study was conducted within the Futian Mangrove National Nature Reserve (113°45′ E, 22°32′ N), situated in northeastern Shenzhen Bay, China (Figure 1). As the only national mangrove ecological wetland in the central area of the city [43], the reserve experienced an alarming contraction in mangrove coverage from the 1970s to the 1980s, coinciding with the rapid urbanization of the Greater Bay Area.
The site experiences a subtropical monsoon climate, with a mean annual temperature of 22.4 °C, average precipitation of 1700–1900 mm, and relative humidity of approximately 80% [44]. Shenzhen Bay exhibits semidiurnal tidal patterns, with spring tide amplitudes averaging 1.9 m [45]. Dominant mangrove species include Avicennia marina, Acanthus ilicifolius, S. caseolaris, S. apetala, Aegiceras corniculatum, and Kandelia obovata [39,46], which constituted the target species for this investigation.

2.2. Data Acquisition

2.2.1. UAV Data

The UAV flights were conducted between 19 July and 21 July 2020 (Figure S1 and Table S1). A four rotor-wing UAV system (Phantom 4 RTK, DJI, Shenzhen, China) was used to collect the images. The UAV platform, which has a maximum flight time of 18 min under optimal weather conditions, was equipped with a camera (FC6310R, DJI, China) and GPS as the payload. The camera has a focal length of 8.8 mm with 20 million effective pixels and produces images in three bands, namely, red (R, 625 nm), green (G, 550 nm), and blue (B, 485 nm). The GPS can provide a centimeter-level positioning accuracy through differential positioning technology. The UAV flight strategy was designed by DJI GO (DJI, China), which is a flight planning and monitoring software. The flying height of the UAV was set at 80 m above ground level. The longitudinal overlap was 80%, and the lateral overlap was 70%, which were established according to the size of the sample site. The lens was set vertically to capture an orthophoto of the sample site. To collect mangrove information at low tide levels according to tide time, the study area was divided into two parts for drone flight.

2.2.2. Field Measurement

Fieldwork was carried out between 19 July and 21 July 2020, synchronously with the UAV data collection. The main methods used for research were the quadrat method and the line transect method [47]. The quadrat method was designed according to the survey sample site. The types, quantity, height, crown diameter, and diameter at breast height of mangrove plants in a 10 m × 10 m plot were recorded. For the inaccessible mangrove areas, the line transect method was adopted for the survey. A 200-m sample line was designed on the sample site, with the species and quantity of mangrove plants along this line recorded. The results of the field measurements were used to verify the UAV survey results.

2.3. Method

2.3.1. Reconstruction of the Mangrove Forests

To evaluate mangrove forest health, the workflow in this study is structured into two major stages. The first stage involves UAV image acquisition (CGCS2000/Gauss-Kruger) and structural indicator extraction, including image reconstruction, surface modeling, and structure-related parameter derivation. The second stage focuses on species identification and health score assessment, encompassing DSM-based preclassification, pixel-level deep learning segmentation, fine species classification, and composite scoring.
The entire process integrated multiple software platforms and programming environments. Python 3.9 was used as the primary environment for image processing and classification tasks. ArcGIS Pro 3.4.1 and ENVI 5.6 were employed for geo-referencing, spatial analysis, and image preprocessing. In addition, supporting software tools associated with the DJI UAV platform were used for flight planning and data management, ensuring consistency between data acquisition and subsequent analysis.
Specifically, Context Capture was used to reconstruct the study area using the images collected. The processing procedure consisted of three steps: initial processing, point cloud and mesh, and Digital Surface Model (DSM) orthomosaic and index. In the initial processing step, the fisheye distortion parameters of the camera were calculated according to the parameters embedded in images collected by the drone. Thereafter, the fisheye distortion parameters were used to correct image distortion. Structure from motion dense point cloud data were produced after the point cloud and mesh step. Finally, two UAV raster products were derived from the images: an RGB orthoimage and the DSM. The ground sampling distance was 0.0217061 m for the RGB orthoimage and DSM.

2.3.2. Classification of the Mangrove Forests

The height information of mangrove plants was introduced to preclassify mangrove species, which can effectively distinguish trees with similar colors and different heights. Based on the remote sensing images captured, the proposed two-step recognition method mainly contains two parts (Figure 2). First, preclassification is conducted on the DSM. The height information represented by DSM is used to classify plants of different tree heights. Thereafter, a pixel-level image segmentation algorithm based on deep learning is used for the fine classification of mangrove plants [48,49,50].
(1) Preclassification of the mangrove forests. According to previous studies, mangrove forests have obvious stratification effects, which means different tree species usually have different heights. Therefore, classifying mangrove species based on tree height is a feasible solution. As the digital surface model represents the height information of the mangrove, according to the obtained DSM, mangrove species preclassification was carried out to identify and classify species with obvious tree height distinctions. Based on the existing mangrove survey data, S. caseolaris and S. apetala are higher than 10 m. The height of A. marina, A. corniculatum, and K. obovata is usually 3 to 6 m. A. ilicifolius is lower than 2 m. Hence, according to the height information reflected by DSM, mangrove plants were divided into three major types: higher plants, middle plants, and lower plants. The three types of mangrove plants are represented by three different gray levels in the recognition results (Figure 3). Furthermore, the fine classification of tree species can be conducted based on this step.
(2) Pixel-level classification of the mangrove forests. According to the collected UAV visible light remote sensing data of the mangrove ecosystem, data annotation was performed, and the full image sample set (approximately 12,290 labeled samples) was split into training (70%), validation (20%), and test (10%) sets. A pixel-level image segmentation algorithm based on deep learning, specifically the SegNet neural network, was employed (Figure 4). Deep learning was chosen over traditional machine learning algorithms due to its superior ability to automatically learn complex, hierarchical features from high-resolution imagery, which is critical for accurate classification in the intricate mangrove environment [51]. The model was trained for 100 epochs using the Adam optimizer with an initial learning rate of 0.001, and categorical cross-entropy as the loss function. SegNet’s architecture effectively utilizes “pooling indices” from its max-pooling layers during encoding to perform non-linear upsampling in the decoding phase, which is crucial for precise boundary reconstruction and high-resolution segmentation results. The trained model was then used to classify orthoimages in the test dataset to obtain species recognition results.
(3) Fine classification of the mangrove forests. Preclassification results and pixel-level species recognition outputs can be combined for fine classification (Figure 5). In particular, in the type of higher plants, the two species S. caseolaris and S. apetala are further classified based on the pixel-level species recognition results. Similarly, in the type of middle plants, A. marina, A. corniculatum, and K. obovata three species are further classified. In the type of lower plants, only A. ilicifolius is further classified. For pixel-level classification results that do not meet the preclassification results, the species of plants were determined according to the species type of the surrounding pixels and the preclassification results. In addition, for pixels with low classification confidence, a majority voting mechanism based on the surrounding neighborhood was applied to revise the species assignment and enhance classification robustness. To ensure the spatial consistency of species labels and support further structural analysis, a fixed-size sliding window was used to identify local maxima during crown center estimation, with the window size selected to balance positional precision and noise reduction. Furthermore, for the preclassification results near the banks and fish ponds, the non-mangrove plants are eliminated in combination with the pixel-level classification results. To eliminate the abnormal points, the classification results were filtered.

2.3.3. Crown Segmentation Method

The purpose of single tree crown detection is to find the location of each tree crown, which is the basis of single tree crown extraction [52]. In the forest, the crown shape of various tree crowns is generally high in the middle and low in the surroundings. Therefore, the local maximum method is used to detect the center point of the crown. The local maximum method uses a local maxima filter to detect the local maximum value. Specifically, a moving window was used to scan the image for maximum point of the local spectrum as the center point of the canopy. If the local maximum points are detected, these points can be used as references to further find the edge of the canopy and describe the outline of the canopy.
Single tree crown segmentation is the next step of the local maximum detection process. Based on the center points of tree crowns detected by the local maximum method, the watershed algorithm is used to identify the tree crown area [53]. The main process is to first generate a gradient image from the original image, then process the edge refinement of the gradient image, and finally search for the local maximum of the gradient image. The local maximum is the canopy boundary.
Finally, according to the corresponding relationship between the tree height and the crown width, the crown width was adjusted. Untrustworthy crown segmentation results were deleted and incomplete crown segmentation results were merged.

2.3.4. Stand Structure Parameters of Mangrove Forests

According to the result of crown segmentation, the stand structure parameters of the mangrove forests were calculated [30]. The stand structure indicators used in this paper mainly include size ratio, mixed degree, angular scale, and species diversity [20,21,54].
Size ratio (U) refers to the proportion of adjacent trees whose crown width is larger than the reference tree in the four nearest adjacent trees. There are five cases for the value of the size ratio: 0 (absolute disadvantage), 0.25 (disadvantage), 0.5 (moderate), 0.75 (sub-advantage), and 1 (advantage). The size ratio can be calculated as:
U i = 1 4 j = 1 4 K i j
U = 1 n i = 1 n U i
where U i is the size ratio of the i-th tree, U is the average size ratio of the community, and n is the number of trees in the community. In the above formula, if the reference tree i is larger than the adjacent tree j, K i j is 1; otherwise, K i j is 0.
Mixed degree (M) refers to the proportion of the four nearest adjacent trees that belong to different species from the reference tree species, which reflects the probability of whether a tree and surrounding trees are the same species or different species. The mixed degree of a single tree can be calculated as:
M i = 1 4 j = 1 4 v i j
M = 1 n i = 1 n M i
where M i is the mixing degree of the i-th tree and v i j = 1 if the species of neighboring tree j is different from that of the reference tree i; otherwise v i j = 0. If M i = 0, the four nearest trees and the reference tree are of the same species (i.e., zero-degree mixed). If M i = 0.25, one of the four nearest trees is not the same species as the reference tree (i.e., weakly mixed). If M i = 0.5, two of the four nearest trees are not the same species as the reference tree (i.e., moderately mixed). If M i = 0.75, three of the four nearest trees are not the same species as the reference tree (i.e., intensity mixed). If M i = 1, all four adjacent trees differ in species from the reference tree (i.e., extremely intensive mixed). The community-level mixing degree is given by the average of all M i values in the stand.
Angle scale (W) refers to the proportion of angle α less than the calibrated standard angle α 0 . The angle α is formed by the four nearest trees and the reference tree. The angle scale of a single tree is calculated as:
W i = 1 4 j = 1 4 z i j
W = 1 n i = 1 n W i
where z i j = 1 if the angle α i j < α 0 (usually 90°), and z i j = 0 otherwise. The angle α i j is defined as the angle between two adjacent neighboring trees j and j + 1, with the reference tree i as the vertex, measured in clockwise order. If W i = 0, the stand structure is very uniform; if W i = 0.25, it is uniform; W i = 0.5 indicates randomness; W i = 0.75 indicates unevenness; and W i = 1 means very uneven. The angle scale fully indicates the spatial distribution pattern of individual trees in the stand.
Shannon–Wiener index and Simpson index were used to quantify species diversity, which can be calculated as:
H i = i = 1 S P i ln P i
D i = 1 i = 1 S P i 2
where H i is the Shannon–Wiener index of species i, D i is the Simpson index of species i, P i is the proportion of species i, which represent the relative importance of species i, and S is the total number of mangrove species.
All structural indicators—size ratio, mixing degree, and angle scale—are first calculated at the individual tree level, denoted as U i , M i , and W i , respectively. These values are then averaged across all trees within a local stand to derive the corresponding community-level indices U , M , and W . The composite structural function F ( A ) is defined at the individual tree level, integrating U i , M i , W i and other species-specific attributes to comprehensively describe the structural condition of each tree within the mangrove stand. This allows for detailed evaluation of spatial heterogeneity and structural complexity across the entire forest community.

2.3.5. Mangrove Forest Health Evaluation Model

The multiplication and division method was used to construct the health evaluation model, which facilitates multi-objective planning based on selected indices. A foundational principle for this model is that indicators contributing positively to stand health are placed in the numerator, while those negatively correlated are placed in the denominator. This approach implicitly assumes equal weight across all selected indicators. In future research, we plan to incorporate more sophisticated weighting strategies and conduct sensitivity analyses to further validate the robustness of the model. Specifically, parameters such as size ratio (U), mixed degree (M), Shannon–Wiener index (H), and Simpson index (D) are placed in the numerator, while angle scale (W) is placed in the denominator. To ensure numerical stability and avoid division by zero, each parameter is augmented by 1 during computation. Thus, the stand structure health score of a single tree is calculated as:
F ( A ) = ( 1 + F ( U ) ) ( 1 + F ( M ) ) ( 1 + F ( H ) ) ( 1 + F ( D ) ) ( 1 + F ( W ) )
where F ( A ) is the health evaluation score of an individual tree, and F ( U ) , F ( M ) , F ( H ) , F ( D ) , and F ( W ) , respectively, denote the value of size ratio, mixed degree, Shannon–Wiener index, Simpson index, and angle scale at the individual tree level. The overall mangrove forest health score is calculated by averaging across all trees:
A = 1 N N = 1 N F A
where N is the total number of trees in the mangrove forest and A is the aggregated stand structure evaluation score.

3. Results

3.1. Classification Result of the Mangrove Forests

A total of 3116 orthoimages were obtained and used to reconstruct the mangrove forests, resulting in the generation of both the orthoimage and DSM of the Futian mangrove forests (Figure 6a,b). Using these datasets, along with tree height data from ground surveys, mangrove trees were classified into three categories based on DSM-derived heights and the preclassification method proposed in this study. The preclassification results (Figure 6c) showed that S. caseolaris and S. apetala are predominantly located on the outermost periphery of the mangroves, specifically in the western and eastern parts of the plots. In contrast, A. marina, A. corniculatum, and K. obovata are widely distributed throughout the study site, while A. ilicifolius is mainly found near the river. These findings provide a basis for further refinement of the classification in subsequent steps.
Based on the preclassification results, the proposed two-step classification method was performed on the data of Futian Mangroves Forests. Compared to the preclassification results, the mangrove plants classified into three major types were further accurately identified (Figure 6d). In fact, the misidentification of non-mangrove plants near the fishpond was accurately eliminated. According to the overall classification results, the dominant species of the Futian mangroves was K. obovata. A. ilicifolius exists in areas, such as rivers and the edge of the mangroves. At the edge of Futian mangrove forests, exotic species such as S. caseolaris and S. apetala were found. A. marina was scattered throughout the mangrove forests and mainly distributed in the non-core area. In the strip-shaped non-core area, A. marina was distributed in the middle of the forests, surrounded by K. obovata. Because the community of A. corniculatum was small, fewer mangrove plants of this species were identified.
To validate the accuracy of mangrove species classification, the orthoimages were cropped to ensure the cropped images only contain a single species. The size of the cropped images was 256 × 256 pixels. For the six mangrove plants assessed herein, the number of correctly identified pixels on the recognition results corresponding to the cropped image was counted, and further calculated as the classification accuracy (Table 1). The classification accuracy was the highest for K. obovata (96.33%), the dominant species of Futian mangroves. Species with obvious height differences had higher classification accuracy. A. marina (62.00%) and A. corniculatum (2.07%) were similar in height to the dominant species, and the two species were less distributed in the study site, resulting in relatively low classification accuracy. For the whole study site, the total classification accuracy was 88.29%. While this study primarily presents the overall accuracy, more comprehensive quantitative evaluations using metrics such as Kappa Coefficient, User’s Accuracy (UA), Producer’s Accuracy (PA), and F1-score are widely recognized in remote sensing and ecological research for assessing classification performance, and their detailed analysis will be conducted in future research to further optimize the classification framework and improve species recognition accuracy.
As the mangrove plant seedlings represent the future development trend of mangroves, an analysis of the species identification results of the mangrove seedlings at the edge of the forest was performed. The mangrove seedlings from Futian mangroves that expanded toward the sea were mainly K. obovata. Besides, as shown in Figure 7, the structure characteristics of seedlings differed in different regions.

3.2. Crown Segmentation Results

To analyze the structural characteristics of seedlings outside the mangrove forests, crown segmentation was performed to extract individual canopies. The segmentation was conducted on images of the same size used for mangrove classification, and each individual crown was outlined with a white circle in the image, with the pixel area of each crown recorded. Based on the pixel resolution corresponding to the drone’s flight altitude and combined with ground survey data, the actual crown size of each individual seedling was calculated. In evaluating the segmentation accuracy, multiple metrics such as F1-score, Intersection over Union (IoU), and Dice coefficient are available for selection; however, Overall Accuracy (OA) was selected in this study as the global evaluation metric to comprehensively reflect the overall segmentation performance. The crown segmentation results of typical samples are shown in Figure 8.
As demonstrated by the crown segmentation results of typical samples, all crowns of individual plants were accurately divided, except for overlapping plants or plants that were too small. In the above set of pictures, a total of 16 canopies were segmented, and 2 were not correctly detected due to overlap. The accuracy of this set of crown segmentation was 88.89%. In another set of pictures, a total of 11 canopies were segmented. Among them, two plants were not correctly detected because they were too small, and one plant was not correctly detected because of overlap. The accuracy of this set of crown segmentation was 78.57%.

3.3. Mangrove Forest Health Evaluation

Using the mangrove species and single tree crown data obtained previously, along with the stand structure health evaluation model presented in Section 2.3.5, we calculated the structural parameters for various sample plots. To analyze the stand structure of seedlings outside the Futian mangrove forests more comprehensively, we selected six typical plots from both core and non-core areas for comparative analysis, as shown in Table 2. The stand structure scores of the core area plots were consistently higher than those of the non-core area plots. The average score in the core area was 0.935766, which was 0.047427 points higher than the average score in the non-core area (0.888339). This indicates a clear trend of superior stand health within the core area plots (F = 10.32, p = 0.0325, ANOVA). While this finding demonstrates a clear observed difference, it is based on a limited number of sample plots (n = 3 for each area), suggesting a need for broader validation in future studies.

4. Discussion

4.1. Accuracy of Mangrove Classification and Optimization Strategies

Accurate species classification is the basis for obtaining mangrove ecological information; thus, the classification accuracy of mangrove plants is a key indicator. By using the method proposed herein, the overall classification accuracy of mangrove species was high; however, the classification accuracy of some species was very limited, especially for A. marina and A. corniculatum. Combined with the ground survey data, the heights of the two species with lower recognition accuracy were similar to those of the dominant species. In addition, the three plants, K. obovata, A. marina, and A. corniculatum, were intertwined in mature mangrove forests, and were similar in color, leaf size, and crown structure. Therefore, when fine classification was performed on the preclassified image, the three species failed to achieve a better classification. In contrast, taller and shorter plants have been classified with higher accuracy. In particular, although the heights of S. caseolaris and S. apetala were similar, their colors were quite different, leading to a more accurate classification. The recognition accuracy of A. ilicifolius was relatively high; however, due to the low height of this species, it is covered by other species, and thus, the whole distribution ranges of this species could not be fully identified.
Based on the analysis of the classification results, to improve the classification accuracy of mangrove species, three approaches have been suggested by Sur et al. [55]: the integration of red edge/near-infrared data, high-resolution imagery, and deep learning algorithms follows a clear priority sequence for enhancing species-level classification of vegetation. In this study, first, the flying height of the drone can be lowered to ensure that the resolution of the acquired remote sensing images is improved. Higher-resolution photos allow for the extraction of more detailed information about mangrove plants. The color and structural characteristics of mangrove species can be further captured for refined classification, thereby enhancing the recognition accuracy of species with similar heights. Second, more sensors can be used to collect mangrove remote sensing information, such as hyperspectral and multispectral sensors. As near-infrared and red edge bands better reflect vegetation characteristics, multi-sensor data fusion can provide a more comprehensive spectral profile of mangrove species. This aligns with the current study’s findings by demonstrating the limitations of visible-light imagery alone for certain species and proposing that incorporating broader spectral information would overcome these challenges. Third, as the research continues, orthophotos of mangroves in different seasons are collected and the mangrove species identification dataset is expanded, and thus, the impact of the external environment will be reduced and the accuracy of species recognition will be improved. In addition, expanding the diversity of training samples and exploring more robust feature extraction methods will further improve classification performance in complex mangrove environments. Fourth, the integration of UAV remote sensing with advanced classification algorithms can significantly enhance the accuracy of mangrove species identification. The combination of multispectral/hyperspectral imagery and LiDAR structural data effectively distinguishes species with similar spectral characteristics, thereby improving classification accuracy [56]. LiDAR-derived three-dimensional structural information facilitates the stratification of upper and lower vegetation layers, while Structure-from-Motion (SfM) photogrammetry further refines species differentiation through canopy height modeling [32]. Moreover, deep learning methods, such as convolutional neural networks (CNNs), demonstrate superior performance in mangrove classification when supported by high-resolution UAV imagery and multi-source data fusion, which enable more automated and scalable species identification.

4.2. Stand Structure Parameters of Mangrove Forest in Different Habitats

The stand structure parameters reflect the health information of the mangroves, and the seedlings reflect the future shape of the mangroves. Therefore, the stand structure of the seedlings at the edge of the mangroves must be analyzed. Based on the stand structure health evaluation model presented in Section 2.3.5, the average stand structure score of the Futian mangrove forests core area was 0.935778 and the average stand structure score of the Futian mangrove forests non-core area was 0.888351. Thus, the stand structure score of the core area is better than that of the non-core area. For the stand structure index, the higher the score, the more reasonable the stand structure of the sample plot and the higher the health of the ecosystem. Thus, the mangroves in the core area were healthier, which is consistent with the results of many previous studies on the health of mangroves.
The core area of the mangrove forest is the key protected area of the Futian mangrove forests. Mangrove seedlings outside the core area are naturally expanded. As a result, the seedlings in this area are more suitable for survival in the current habitat after natural selection. Different from the expansion pattern of mangrove seedlings outside the core area, several exotic plants are cultivated outside the non-core area, thereby affecting the expansion of mangrove seedlings. In addition, the mangrove seedlings in this area will be manually removed by the management of the reserve. Therefore, the stand structure of the mangrove in this area is destroyed. Although the exotic species outside the non-core area have been felled and cleared in recent years, the stand structure of the mangrove seedlings in this area has not been restored to their natural state. However, the differences in stand structure scores between the core and non-core areas may not be solely attributed to human intervention. Environmental factors such as tidal regime and salinity also play a significant role in influencing mangrove stand structure. Higher tidal flooding can promote root development and increase seedling survival, but extreme flooding events are likely to cause stress and damage to the root system. Similarly, salinity gradients affect mangrove species composition and growth patterns directly. The high-salinity area tends to support mangroves with slower growth rates and lower canopy density, but these species often develop greater tolerance to osmotic stress, which reflect an adaptive advantage in extreme environments [57].
This suggests that the relatively lower stand structure score in non-core area could be associated with both environmental stressors and human activities. The impact of tidal patterns and salinity levels may limit seedling establishment and structural development, which, combined with manual removal and exotic species invasion, further reduces the structural integrity of non-core area. Future management should, therefore, consider not only controlling exotic species but also regulating tidal patterns and improving soil salinity conditions to enhance mangrove stand structure and ecosystem health.

4.3. Significance of Mangrove Forest Health Evaluation

The current methods for obtaining mangrove stand structure are mainly based on traditional surveys, which are inefficient, partial, and lack overall guidance. At present, satellite remote sensing data have been employed in mangrove stand structure surveys; however, their landscape scale is limited. UAV remote sensing can account for the lack of satellite remote sensing in terms of spatial resolution, cost, and timeliness. It can identify subtle structural variations, thereby improving the accuracy of mangrove stand structure assessment [58]. The application of UAV remote sensing for the rapid assessment and optimization guidance of mangrove stand structure is of practical significance. For example, based on different mangrove stand structure indexes in different habitats, the development pattern of mangrove seedlings under natural conditions is accurately interpreted. Furthermore, the health of the stand structure in mangrove areas under manual intervention can be assessed. Comparing the growth pattern of mangroves under natural conditions, the stand structure of restored mangroves can be optimized. Besides, according to the mangrove stand structure under natural conditions, guidance can be provided for artificial planting and restoration of mangroves in the future.
Under the guidance of scientific theory, a comprehensive and systematic method was established for the rapid evaluation of mangrove stand structure, which enabled in-depth comprehensive research on the structure, composition, and change of mangrove stand. The method proposed in this paper involved the use of drones to efficiently obtain mangrove stand structure, which has important application value and guiding significance for mangrove protection and restoration.
Compared with existing studies, our proposed method offers a more fine-grained and operational approach to evaluating mangrove health. For instance, Giri et al. [59] primarily focused on global mangrove distribution using satellite-derived remote sensing data, while our study emphasizes quantitative health assessment based on stand structure indicators. Pham et al. [60] adopted vegetation indices for classifying mangrove health status, whereas we incorporated canopy height, crown diameter, and density—extracted from UAV imagery—into a multi-indicator scoring model, achieving higher resolution and classification accuracy at the local scale. Moreover, unlike Alongi [61] and Kuenzer et al. [62], who relied on general degradation narratives, our approach uses measurable stand parameters to evaluate ecological status, thereby improving interpretability and practical applicability.
Nevertheless, several limitations should be acknowledged in this study. First, the UAV imagery was acquired in a single season, which restricted the ability to capture seasonal dynamics of mangrove structural changes. Future studies could incorporate multi-temporal observations to improve temporal generalization. Second, due to high visual similarity among certain species—especially in overlapping zones or ecotones—misclassification remains a challenge, and subtle boundaries between species were occasionally misinterpreted. Third, the study was limited to the Futian Mangrove Reserve, and although the findings provide valuable insights, the generalizability of the proposed method requires further validation across broader spatial and ecological contexts.

5. Conclusions

This paper presents a novel and effective UAV-based method for interpreting the stand structure of urban mangrove forests using remote sensing images. The key novelty of this work lies in its integrated approach, which combines deep neural networks with physical constraints (tree height) for accurate species and distribution mapping, followed by the derivation of stand structure parameters for a comprehensive health evaluation model. This provides a streamlined, rapid, and high-resolution assessment capability for complex mangrove ecosystems, addressing limitations of traditional survey methods and broader-scale satellite remote sensing. The experimental application in the Futian Mangrove National Nature Reserve at Shenzhen Bay, China, demonstrated the method’s efficacy. Specifically, the proposed approach accurately identified mangrove species information and extracted single tree crown details for mangrove seedlings. Furthermore, the stand structure health of mangroves was comprehensively evaluated and analyzed to identify the different expansion patterns of mangroves. While not quantitatively assessed in this study, the method considerably reduces manual effort and field time, as observed during initial deployment phases, highlighting its practical efficiency in large-scale mangrove monitoring.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/f16071168/s1, Table S1: Flight parameters and sensor specifications of the UAV platforms; Figure S1: Flight trajectory of the UAV in the Futian mangrove forest.

Author Contributions

Conceptualization, C.Z.; methodology, C.Z.; software, C.Z.; formal analysis, C.Z.; investigation, C.Z. and X.S.; resources, X.S.; data curation, Y.Z. and Y.W.; writing—original draft preparation, C.Z.; writing—review and editing, Y.Z., Y.W., and X.S.; visualization, C.Z. and Y.Z.; supervision, X.S.; project administration, X.S.; funding acquisition, X.S. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the National Natural Science Foundation of China (No. 42201361), the Program of Science and Technology of Shenzhen (No. KCXFZ20230731094059009), and the Innovation Team Project for General Universities in Guangdong Province, China (No. 2023KCXTD050).

Data Availability Statement

The data presented in this study are available on request from the corresponding author. The data are not publicly available due to institutional confidentiality.

Acknowledgments

The authors would like to thank Zhi Zhang, Yueqi Zhang, and Biqian Jiang for their assistance in the annotation and identification of mangrove species during the species classification process. The authors also acknowledge the support provided by the Futian National Mangrove National Nature Reserve Administration for facilitating fieldwork and providing valuable site access.

Conflicts of Interest

The authors declare no conflict of interest.

Abbreviations

The following abbreviations are used in this manuscript:
UAVUnmanned aerial vehicle
DSMDigital surface model
RGBRed, green, blue
GPSGlobal positioning system
CNNConvolutional neural network

References

  1. Tomlinson, P.B. The Botany of Mangroves; Cambridge University Press: Cambridge, UK, 1986; 413p. [Google Scholar]
  2. Aburto-Oropeza, O.; Ezcurra, E.; Danemann, G.; Valdez, V.; Murray, J.; Sala, E. Mangroves in the Gulf of California increase fishery yields. Proc. Natl. Acad. Sci. USA 2008, 105, 10456–10459. [Google Scholar] [CrossRef]
  3. Del Valle, A.; Liu, J.; Farfan, L.M.; Ellison, J.C. Mangroves protect coastal economic activity from hurricanes. Proc. Natl. Acad. Sci. USA 2020, 117, 265–270. [Google Scholar] [CrossRef] [PubMed]
  4. Hu, Z.; Temmerman, S.; Zhu, Q.; Wang, X.; Chen, L.; Wang, C. Predicting nature-based coastal protection by mangroves under extreme waves. Proc. Natl. Acad. Sci. USA 2025, 122, e2410883122. [Google Scholar] [CrossRef] [PubMed]
  5. Valiela, I.; Bowen, J.L.; York, J.K. Mangrove forests: One of the world’s threatened major tropical environments. BioScience 2001, 51, 807–815. [Google Scholar] [CrossRef]
  6. Goldberg, L.; Lagomasino, D.; Thomas, N.; Fatoyinbo, L. Global declines in human-driven mangrove loss. Glob. Change Biol. 2020, 26, 5844–5855. [Google Scholar] [CrossRef]
  7. Liu, H.; Ren, H.; Hui, D.; Wang, W.; Liao, B.; Cao, Q. Carbon stocks and potential carbon storage in the mangrove forests of China. J. Environ. Manag. 2014, 133, 86–93. [Google Scholar] [CrossRef]
  8. Li, M.S.; Lee, S.Y. Mangroves of China: A brief review. For. Ecol. Manag. 1997, 96, 241–259. [Google Scholar] [CrossRef]
  9. Jia, M.; Liu, M.; Wang, Z.; Mao, D.; Ren, C.; Cui, H. Evaluating the effectiveness of conservation on mangroves: A remote sensing-based comparison for two adjacent protected areas in Shenzhen and Hong Kong, China. Remote Sens. 2016, 8, 627. [Google Scholar] [CrossRef]
  10. Fan, B.; Li, Y. China’s conservation and restoration of coastal wetlands offset much of the reclamation-induced blue carbon losses. Glob. Change Biol. 2024, 30, e17039. [Google Scholar] [CrossRef]
  11. Tang, Y.; Fang, Z.; Chen, K.; Zhang, Z.; Zhong, Y.; An, D.; Yang, X.; Liao, B. Ecological influence of exotic plants of Sonneratia apetala on understory macrofauna. Acta Oceanol. Sin. 2012, 31, 115–125. [Google Scholar] [CrossRef]
  12. Bosire, J.O.; Dahdouh-Guebas, F.; Walton, M.; Crona, B.I.; Lewis, R.R.; Field, C.; Kairo, J.G.; Koedam, N. Functionality of restored mangroves: A review. Aquat. Bot. 2008, 89, 251–259. [Google Scholar] [CrossRef]
  13. Iftekhar, M.S. Functions and development of reforested mangrove areas: A review. Int. J. Biodivers. Sci. Manag. 2008, 4, 1–14. [Google Scholar] [CrossRef]
  14. Kamali, B.; Hashim, R. Mangrove restoration without planting. Ecol. Eng. 2011, 37, 387–391. [Google Scholar] [CrossRef]
  15. Ren, H.; Wu, X.; Ning, T.; Huang, G.; Wang, J.; Jian, S.; Lu, H. Wetland changes and mangrove restoration planning in Shenzhen Bay, Southern China. Landsc. Ecol. Eng. 2011, 7, 241–250. [Google Scholar]
  16. Ellison, A.M. Mangrove restoration: Do we know enough? Restor. Ecol. 2000, 8, 219–229. [Google Scholar] [CrossRef]
  17. Su, J.; Friess, D.A.; Gasparatos, A. A meta-analysis of the ecological and economic outcomes of mangrove restoration. Nat. Commun. 2021, 12, 5050. [Google Scholar] [CrossRef] [PubMed]
  18. Hai, N.T.; Dell, B.; Phuong, V.T.; Harper, R.J. Towards a more robust approach for the restoration of mangroves in Vietnam. Ann. For. Sci. 2020, 77, 18. [Google Scholar] [CrossRef]
  19. Lee, S.Y.; Hamilton, S.; Barbier, E.B.; Primavera, J.; Lewis, R.R. Better restoration policies are needed to conserve mangrove ecosystems. Nat. Ecol. Evol. 2019, 3, 870–872. [Google Scholar] [CrossRef]
  20. Hui, G. Studies on the application of stand spatial structure parameters based on the relationship of neighborhood trees. J. Beijing For. Univ. 2013, 35, 1–8. [Google Scholar]
  21. Hui, G.; Zhang, G.; Zhao, Z.; Yang, A. Methods of forest structure research: A review. Curr. For. Rep. 2019, 5, 142–154. [Google Scholar] [CrossRef]
  22. Hui, G.; Zhao, Z.; Chen, M. Important variables for describing forest structure. Temp. For. Res. 2020, 3, 14–19. [Google Scholar]
  23. Hui, G.; Zhao, Z.; Hu, Y.; Liu, S.; Liu, J. Measuring stand spatial structure diversity based on the four-nearest-neighbor relationship. J. Beijing For. Univ. 2023, 45, 18–26. [Google Scholar]
  24. Liang, S.; Liang, M.; Wu, Y.; Zan, Q.; Wang, Y.; Xie, Q. Spatial structure analysis of Sonneratia caseolaris and Sonneratia apetala natural forests in Futian, Shenzhen. Guihaia 2005, 25, 393–398. [Google Scholar]
  25. Fu, C.; Liu, S. Study on spatial structure of mangrove forests in Leizhou Peninsula based on ecosystem management. For. Resour. Manag. 2009, 2, 55–59. [Google Scholar]
  26. Fu, C.; Liu, S. Analysis of the stand spatial structure of mangrove forest in Techeng Isle. Guangdong Agric. Sci. 2010, 2, 4–6, 27. [Google Scholar]
  27. Li, J.J.; Li, J.P.; Liu, S.Q.; Zhang, H.W.; Feng, X.L. Homogeneity index of mangrove stand spatial structure. For. Sci. 2010, 46, 6–14. [Google Scholar]
  28. Li, J. Study on Spatial Structure Optimization of Mangrove Ecosystem in Zhanjiang, Guangdong. Master’s Thesis, Central South University of Forestry and Technology, Changsha, China, 2010. [Google Scholar]
  29. Yue, T. Research on Landscape Dynamics, Structural Characteristics, and Health Management Techniques of Mangroves in the Zhangjiang Estuary. Master’s Thesis, Chinese Academy of Forestry, Beijing, China, 2014. [Google Scholar]
  30. Azman, M.S.; Sharma, S.; Shaharudin, M.A.M.; Hamzah, M.L.; Adibah, S.N.; Zakaria, R.M.; MacKenzie, R.A. Stand structure, biomass and dynamics of naturally regenerated and restored mangroves in Malaysia. For. Ecol. Manag. 2021, 482, 118852. [Google Scholar] [CrossRef]
  31. Huang, J.; Li, J.; Liu, S.; Han, W.; Zeng, Q. Comparative analysis of spatial structures of different Kandelia obovata communities in Techeng Isle. For. Resour. Manag. 2013, 6, 92–95, 101. [Google Scholar]
  32. Navarro, A.; Young, M.; Allan, B.; Carnell, P.; Macreadie, P.; Ierodiaconou, D. The application of unmanned aerial vehicles (UAVs) to estimate above-ground biomass of mangrove ecosystems. Remote Sens. Environ. 2020, 242, 111747. [Google Scholar] [CrossRef]
  33. Pham, T.; Yokoya, N.; Bui, D.; Yoshino, K.; Friess, D. Remote sensing approaches for monitoring mangrove species, structure, and biomass: Opportunities and challenges. Remote Sens. 2019, 11, 230. [Google Scholar] [CrossRef]
  34. Wang, L.; Jia, M.; Yin, D.; Tian, J. A review of remote sensing for mangrove forests: 1956–2018. Remote Sens. Environ. 2019, 231, 111223. [Google Scholar] [CrossRef]
  35. Li, Y.; Zhang, T. Application of centimeter-scale remote sensing imagery in investigating the spatial structure of secondary forests in Yichun. J. Northeast For. Univ. 2013, 41, 139–143. [Google Scholar]
  36. Xiong, Y.; Dai, Z.; Long, C.; Liang, X.; Lou, Y.; Mei, X.; Nguyen, B.A.; Cheng, J. Machine Learning-Based Examination of Recent Mangrove Forest Changes in the Western Irrawaddy River Delta, Southeast Asia. Catena 2024, 234, 107601. [Google Scholar] [CrossRef]
  37. Wen, X.; Liu, K.; Cao, J.; Zhu, Y.; Wang, Z. Estimation of aboveground biomass of mangroves in China based on forest canopy height and allometric equations. Trop. Geogr. 2023, 43, 1–11. [Google Scholar]
  38. Li, X.; Yeh, A.G.O.; Wang, S.; Liu, K.; Liu, X.; Qian, J.; Chen, X.; He, Z.; Qin, C. Radar remote sensing estimation of vegetation biomass in mangrove wetlands. J. Remote Sens. 2006, 10, 387–396. [Google Scholar]
  39. Li, Z.; Zan, Q.; Yang, Q.; Zhu, D.; Chen, Y.; Yu, S. Remote estimation of mangrove aboveground carbon stock at the species level using a low-cost unmanned aerial vehicle system. Remote Sens. 2019, 11, 1018. [Google Scholar] [CrossRef]
  40. Lucas, R.; Van De Kerchove, R.; Otero, V.; Lagomasino, D.; Fatoyinbo, L.; Omar, H.; Satyanarayana, B.; Dahdouh-Guebas, F. Structural characterisation of mangrove forests achieved through combining multiple sources of remote sensing data. Remote Sens. Environ. 2020, 237, 111543. [Google Scholar] [CrossRef]
  41. Yin, D.; Wang, L. Individual mangrove tree measurement using UAV-based LiDAR data: Possibilities and challenges. Remote Sens. Environ. 2019, 223, 34–49. [Google Scholar] [CrossRef]
  42. Kharol, S.K.; Prapavessis, C.; Shephard, M.W.; McLinden, C.A.; Griffin, D. Cloud-based data mapper (CDM): Application for monitoring dry deposition of reactive nitrogen. Front. Environ. Sci. 2023, 11, 1172977. [Google Scholar] [CrossRef]
  43. Chen, L.; Wang, W.; Zhang, Y.; Lin, G. Recent progresses in mangrove conservation, restoration and research in China. J. Plant Ecol. 2009, 2, 45–54. [Google Scholar] [CrossRef]
  44. Ren, H.; Chen, H.; Li, Z.A.; Han, W. Biomass accumulation and carbon storage of four different aged Sonneratia apetala plantations in Southern China. Plant Soil 2010, 327, 279–291. [Google Scholar] [CrossRef]
  45. Li, R.; Chai, M.; Li, R.; Xu, H.; He, B.; Qiu, G.Y. Influence of introduced Sonneratia apetala on nutrients and heavy metals in intertidal sediments, South China. Environ. Sci. Pollut. Res. 2017, 24, 2914–2927. [Google Scholar] [CrossRef] [PubMed]
  46. Zan, Q.; Wang, B.; Wang, Y.; Li, M. Ecological assessment on the introduced Sonneratia caseolaris and S. apetala at the mangrove forest of Shenzhen Bay, China. Acta Bot. Sin. 2003, 45, 544–551. [Google Scholar]
  47. Kauffman, J.B.; Donato, D. Protocols for the Measurement, Monitoring and Reporting of Structure, Biomass and Carbon Stocks in Mangrove Forests; CIFOR: Bogor, Indonesia, 2012. [Google Scholar]
  48. Dyrmann, M.; Karstoft, H.; Midtiby, H.S. Plant species classification using deep convolutional neural network. Biosyst. Eng. 2016, 151, 72–80. [Google Scholar] [CrossRef]
  49. Christin, S.; Hervet, É.; Lecomte, N. Applications for deep learning in ecology. Methods Ecol. Evol. 2019, 10, 1632–1644. [Google Scholar] [CrossRef]
  50. Mohanty, S.P.; Hughes, D.P.; Salathe, M. Using deep learning for image-based plant disease detection. Front. Plant Sci. 2016, 7, 1419. [Google Scholar] [CrossRef]
  51. Badrinarayanan, V.; Kendall, A.; Cipolla, R. SegNet: A deep convolutional encoder-decoder architecture for image segmentation. IEEE Trans. Pattern Anal. Mach. Intell. 2017, 39, 2481–2495. [Google Scholar] [CrossRef]
  52. Yang, J.; Kang, Z.; Cheng, S.; Yang, Z.; Akwensi, P.H. An individual tree segmentation method based on watershed algorithm and three-dimensional spatial distribution analysis from airborne LiDAR point clouds. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2020, 13, 1055–1067. [Google Scholar] [CrossRef]
  53. Kolarik, N.E.; Gaughan, A.E.; Stevens, F.R.; Pricope, N.G.; Woodward, K.; Cassidy, L.; Salerno, J.; Hartter, J. A multi-plot assessment of vegetation structure using a micro-unmanned aerial system (UAS) in a semi-arid savanna environment. ISPRS J. Photogramm. Remote Sens. 2020, 164, 84–96. [Google Scholar] [CrossRef]
  54. Zhao, Z.; Hui, G. Forest management method originated by China since the 21st century. J. Beijing For. Univ. 2019, 41, 50–57. [Google Scholar]
  55. Sur, K.; Verma, V.K.; Panwar, P.; Shukla, G.; Chakravarty, S.; Nath, A.J. Monitoring vegetation degradation using remote sensing and machine learning over India—A multi-sensor, multi-temporal and multi-scale approach. Environ. Res. Commun. 2021, 3, 125007. [Google Scholar] [CrossRef]
  56. Li, Q.; Wong, F.K.K.; Fung, T. Mapping multi-layered mangroves from multispectral, hyperspectral, and LiDAR data. Remote Sens. Environ. 2021, 258, 112403. [Google Scholar] [CrossRef]
  57. Osland, M.J.; Feher, L.C.; López-Portillo, J.; Day, R.H.; Suman, D.O.; Guzmán, D.A.; Cahoon, D.R. The impacts of mangrove range expansion on wetland ecosystem services in the southeastern United States: Current understanding, knowledge gaps, and emerging research needs. Glob. Change Biol. 2022, 28, 504–523. [Google Scholar] [CrossRef]
  58. Ying, L.; Wang, L. How to automate timely large-scale mangrove mapping with remote sensing. Remote Sens. Environ. 2021, 264, 112584. [Google Scholar]
  59. Giri, C.; Ochieng, E.; Tieszen, L.L.; Zhu, Z.; Singh, A.; Loveland, T.; Masek, J.; Duke, N. Status and distribution of mangrove forests of the world using earth observation satellite data. Glob. Ecol. Biogeogr. 2011, 20, 154–159. [Google Scholar] [CrossRef]
  60. Pham, T.D.; Yokoya, N.; Miyamoto, M.; Yoshino, K. UAV-Based Mangrove Forest Health Monitoring Using Spectral-Spatial Deep Features and Machine Learning. Remote Sens. 2022, 14, 2725. [Google Scholar]
  61. Alongi, D.M. Present state and future of the world’s mangrove forests. Environ. Conserv. 2002, 29, 331–349. [Google Scholar] [CrossRef]
  62. Kuenzer, C.; Bluemel, A.; Gebhardt, S.; Quoc, T.V.; Dech, S. Remote Sensing of Mangrove Ecosystems: A Review. Remote Sens. 2011, 3, 878–928. [Google Scholar] [CrossRef]
Figure 1. Study area.
Figure 1. Study area.
Forests 16 01168 g001
Figure 2. General workflow of the process revealing accurate evaluation of mangrove forest health considering stand structure indicators. UAV, Unmanned Aerial Vehicle; DSM, Digital Surface Model. Note: Solid arrows indicate the main workflow; dotted arrows indicate auxiliary or feedback processes; boxes represent processing modules.
Figure 2. General workflow of the process revealing accurate evaluation of mangrove forest health considering stand structure indicators. UAV, Unmanned Aerial Vehicle; DSM, Digital Surface Model. Note: Solid arrows indicate the main workflow; dotted arrows indicate auxiliary or feedback processes; boxes represent processing modules.
Forests 16 01168 g002
Figure 3. Preclassification of mangrove plants. Image on the left is an orthophoto, image in the middle is the DSM, and image on the right is the preclassification results of mangrove plants.
Figure 3. Preclassification of mangrove plants. Image on the left is an orthophoto, image in the middle is the DSM, and image on the right is the preclassification results of mangrove plants.
Forests 16 01168 g003
Figure 4. Schematic of the deep learning network.
Figure 4. Schematic of the deep learning network.
Forests 16 01168 g004
Figure 5. Fine classification results of mangrove plants. Image on the left is the preclassification results of mangrove plants, image in the middle is the pixel-level classification results of mangrove plants, and image on the right is the fine classification results of mangrove plants.
Figure 5. Fine classification results of mangrove plants. Image on the left is the preclassification results of mangrove plants, image in the middle is the pixel-level classification results of mangrove plants, and image on the right is the fine classification results of mangrove plants.
Forests 16 01168 g005
Figure 6. Orthoimage (a), DSM (b), preclassification result (c), and classification result (d) of Futian mangrove forests.
Figure 6. Orthoimage (a), DSM (b), preclassification result (c), and classification result (d) of Futian mangrove forests.
Forests 16 01168 g006
Figure 7. Recognition results of mangrove plant species at forest edge. Image on the left is an orthophoto and image on the right is the result of species recognition. White is the background and green is Kandelia obovata.
Figure 7. Recognition results of mangrove plant species at forest edge. Image on the left is an orthophoto and image on the right is the result of species recognition. White is the background and green is Kandelia obovata.
Forests 16 01168 g007
Figure 8. Crown segmentation results. Image on the left is the original image, and image on the right is the crown segmentation results.
Figure 8. Crown segmentation results. Image on the left is the original image, and image on the right is the crown segmentation results.
Forests 16 01168 g008
Table 1. Classification accuracy of different mangrove species in Futian mangrove forests.
Table 1. Classification accuracy of different mangrove species in Futian mangrove forests.
SpeciesNumber of Test ImagesAccuracy
Kandelia obovata50096.33%
Avicennia marina15962.00%
Acanthus ilicifolius10484.49%
Sonneratia caseolaris23493.66%
Sonneratia apetala21791.04%
Aegiceras corniculatum152.07%
Total122988.29%
Table 2. Stand structure index of the Futian mangrove forests in different areas.
Table 2. Stand structure index of the Futian mangrove forests in different areas.
Study SiteSample NumberStand Structure Index
Core Area10.939063
20.936688
30.931548
Non-core Area40.912698
50.862245
60.890110
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Zhai, C.; Zhang, Y.; Wu, Y.; Shen, X. Accurate Evaluation of Urban Mangrove Forest Health Considering Stand Structure Indicators Based on UAVs. Forests 2025, 16, 1168. https://doi.org/10.3390/f16071168

AMA Style

Zhai C, Zhang Y, Wu Y, Shen X. Accurate Evaluation of Urban Mangrove Forest Health Considering Stand Structure Indicators Based on UAVs. Forests. 2025; 16(7):1168. https://doi.org/10.3390/f16071168

Chicago/Turabian Style

Zhai, Chaoyang, Yiteng Zhang, Yifan Wu, and Xiaoxue Shen. 2025. "Accurate Evaluation of Urban Mangrove Forest Health Considering Stand Structure Indicators Based on UAVs" Forests 16, no. 7: 1168. https://doi.org/10.3390/f16071168

APA Style

Zhai, C., Zhang, Y., Wu, Y., & Shen, X. (2025). Accurate Evaluation of Urban Mangrove Forest Health Considering Stand Structure Indicators Based on UAVs. Forests, 16(7), 1168. https://doi.org/10.3390/f16071168

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop