Mapping and Estimating Blue Carbon in Mangrove Forests Using Drone and Field-Based Tree Height Data: A Cost-Effective Tool for Conservation and Management

Abstract

Mangrove forests are vital blue carbon (BC) ecosystems that significantly contribute to climate change mitigation through carbon sequestration. Accurate, scalable, and cost-effective methods for estimating carbon stocks in these environments are essential for conservation planning. In this study, we assessed the potential of drones, also known as unmanned aerial vehicles (UAVs), for estimating above-ground biomass (AGB) and BC in Avicennia marina stands by integrating drone-based canopy measurements with field-measured tree heights. Using structure-from-motion (SfM) photogrammetry and a consumer-grade drone, we generated a canopy height model and extracted structural parameters from individual trees in the Melgonze mangrove patch, southern Iran. Field-measured tree heights served to validate drone-derived estimates and calibrate an allometric model tailored for A. marina. While drone-based heights differed significantly from field measurements (p < 0.001), the resulting AGB and BC estimates showed no significant difference (p > 0.05), demonstrating that crown area (CA) and model formulation effectively compensate for height inaccuracies. This study confirms that drones can provide reliable estimates of BC through non-invasive means—eliminating the need to harvest, cut, or physically disturb individual trees—supporting their application in mangrove monitoring and ecosystem service assessments, even under challenging field conditions.

1. Introduction

Mangrove forests are critical ecosystems that contribute to the health of the environment by playing crucial roles in carbon sequestration, protection of the coasts, and supporting biodiversity [1]. Mangroves cover only 0.5% of the global coastal area but store 10%–15% of the total carbon found in coastal sediments, termed “blue carbon” [2,3,4]. This high carbon sequestration capacity highlights the key role of mangrove ecosystems in supporting climate change mitigation, particularly within coastal zones [5,6]. The growing recognition of the economic value of BC has led to the emergence of projects that leverage carbon markets, capitalizing on the ability of coastal ecosystems to capture and store carbon [7,8]. As a nature-based solution, the reforestation of mangroves is a climate change mitigation strategy [9]. Despite such important roles, human activities are putting mangroves on a decline [10,11]. Therefore, accurate estimation of carbon sinks in BC ecosystems is crucial for effective management and climate mitigation [12].

Remote sensing technologies have emerged as indispensable tools for large-scale environmental monitoring, enabling the efficient, non-destructive, and scalable analysis of ecosystem structure and function. Recent advancements in remote spectral imaging, GIS, and machine learning have particularly improved the ability to assess vegetation dynamics and physiological traits with high accuracy [13,14]. While traditional methods for estimating AGB and BC, such as ground-based measurements, remain accurate, they are often expensive, labor-intensive, and spatially limited. In contrast, newer approaches such as drone-based remote sensing offer rapid, cost-effective alternatives for biomass and carbon estimation across challenging terrains [15,16].

1.1. Drone-Based Estimation of Biomass and Carbon in Coastal Ecosystems

Recent studies have increasingly employed drone-based methods combined with field surveys to estimate above-ground biomass (AGB) and blue carbon (BC) content in coastal environments. For example, a study conducted in the Seto Inland Sea, Japan, estimated eelgrass AGB and carbon content in shallow waters by integrating drone imagery, quadrat surveys, and biomass sampling. The approach demonstrated strong correlations between leaf area, coverage, and biomass (R² > 0.97), while proving robust against water interference and well suited to intertidal conditions [17]. Similarly, in southeastern Australia, drone-based structure-from-motion (SfM) photogrammetry was used to estimate tree height, canopy diameter, and AGB in both natural and restored mangrove forests. The results closely matched the field-based measurements and highlighted the method’s cost-effectiveness and scalability [15].

1.2. Structural Trait Extraction and Limitations

Several studies have focused on the extraction of biometric traits such as tree height and canopy area from drone-derived data. For instance, drone-acquired RGB imagery has been used to estimate structural traits of Avicennia marina (gray mangrove) trees. While tree height showed a strong correlation with field measurements (R² = 0.98), canopy area and height were found to be weaker predictors of biomass compared to trunk diameter, which remains difficult to determine remotely. To overcome this limitation, model-based estimations were applied, though they introduced added uncertainty to biomass predictions. Nevertheless, the study emphasized the operational advantages and spatial efficiency of drone-based methods, while calling for improved calibration strategies [18].

1.3. Species Classification and Health Assessment

Recent developments in remote sensing and artificial intelligence have shown great potential for integrating drone-based hyperspectral and LiDAR data in species-level classification and forest health assessment. For example, Ou et al. (2023) demonstrated that fusing drone hyperspectral imagery with LiDAR-derived canopy height models significantly enhanced species classification accuracy, achieving over 96% overall accuracy using machine learning algorithms such as XGBoost [19]. Furthermore, vegetation health indicators like canopy chlorophyll content (CCC) have been estimated using drones and GF-6 satellite data. Deng et al. (2023) used ensemble machine learning methods (e.g., XGBoost and random forest) to estimate CCC in the Beibu Gulf region, successfully capturing species-specific health variability and supporting mangrove conservation efforts [20]. Also, in a mixed broadleaf–conifer forest in the Czech Republic, Abdollahnejad and Panagiotidis (2020) [21] demonstrated the effectiveness of unmanned aerial systems (UASs) equipped with five-band multispectral sensors in classifying tree species and evaluating forest health. Their method employed bi-temporal imagery (spring and summer), a combination of spectral vegetation indices (e.g., CI, NDRE, PSRI) and texture analysis (e.g., entropy, contrast, GLDV metrics), and support vector machine (SVM) modeling to distinguish between healthy, infested, and dead trees—primarily impacted by bark beetle infestation. The study achieved high classification accuracies, with 81.18% overall accuracy (OA) for species classification and 84.71% OA for health status detection. Importantly, the inclusion of red-edge bands and texture metrics improved the model’s sensitivity to subtle changes in canopy condition, highlighting the utility of fused multispectral and structural data for forest monitoring applications in temperate ecosystems [21].

1.4. Individual Tree-Based Approaches

Moreover, individual tree-based models have also been developed. Qiu et al. (2019) [22], for instance, applied drone LiDAR and WorldView-2 satellite imagery to infer tree-level AGB. Although their method showed slightly lower accuracy than grid-based random forest models (R² = 0.49 vs. 0.67), it offered detailed spatial resolution for biomass distribution—a key advantage for precise mangrove management [22]. In a similar comparison, Yu et al. (2010) [23] evaluated the performance of both area-based and individual tree-based approaches using airborne laser scanning data in Finnish boreal forests. They found that while both methods yielded promising results for predicting forest attributes such as mean height, diameter, and volume, the individual tree-based method achieved slightly better accuracy for height and volume (e.g., an RMSE of 5.69% for mean height vs. 6.42% in the area-based approach), highlighting its advantage in high-resolution forest monitoring and the importance of point density in enhancing model performance [23].

In this study, we integrated drone-based geometry analysis with ground-based measurements to estimate AGB and BC in individual Avicennia marina trees located in a mangrove patch within the Melgonze at Mond protected area, southeastern Iran. This combined approach facilitates a scalable estimation of carbon stocks in A. marina, offering a practical alternative to traditional field-based methods, particularly in regions where accessibility and fieldwork are constrained. Given that many mangrove regions lie within intertidal zones with muddy substrates, where conventional carbon estimation techniques are often constrained, the adopted method provides a more efficient, non-invasive, and feasible solution for assessing biomass and carbon stocks in such environments.

This study has the following aims: (i) to estimate the AGB and BC of A. marina using an integrated approach combining drone-derived data with field measurements, and (ii) to evaluate the reliability of drone-based estimates compared to ground-truth data.

2. Materials and Methods

2.1. Study Site

This study was conducted in the Melgonze mangrove forest along the southern section of the Mond Protected Area, located at 27°50′48″ N latitude and 51°34′59″ E longitude, in southwestern Iran along the northern coast of the Persian Gulf (Figure 1). The site is primarily characterized by A. marina, forming a unique ecotonal habitat influenced by both terrestrial and marine environments. These mangroves thrive in a harsh environment with an arid climate and hypersaline coastal bed (37.9–41.3 ppt) [24,25,26,27,28]. The satellite imagery from Google Earth indicates a decrease by nearly half in this mangrove region between 2015 (~3.2 ha) and June 2022 (~1.7 ha) [11]. However, considering the variable acquisition dates, image sources, and georeferencing limitations inherent to Google Earth, this estimate should be interpreted with caution. The site contains trees of varying heights and crown sizes, offering structural diversity that facilitates the evaluation of drone image quality and accuracy across different tree sizes. Another suitable characteristic of the site is the vegetation coverage, which we estimated to range between 50% and 75% based on visual inspection of the orthomosaic and field observations. Additional satellite-based archives, such as Earth Explorer, could further support retrospective assessments of land cover dynamics, though they were not central to this UAV-focused study. Since in some areas, such as those adjacent to airports, drone flight was not allowed, the area was also checked to make sure drone flight was not prohibited within the region. The site was also selected for its accessibility and protection status within the Mond Protected Area.

Figure 1. Location of the study site in the Melgonze mangrove forest (Mond Protected Area, southern Iran). Regional context map showing the study area along the Persian Gulf coast. Drone-derived true-color (RGB) orthomosaic generated from images captured on 8 November 2021. Red triangles (a–c) are the sample filed measurements of mangrove tree heights (see Figure 2).

Figure 1. Location of the study site in the Melgonze mangrove forest (Mond Protected Area, southern Iran). Regional context map showing the study area along the Persian Gulf coast. Drone-derived true-color (RGB) orthomosaic generated from images captured on 8 November 2021. Red triangles (a–c) are the sample filed measurements of mangrove tree heights (see Figure 2).

Figure 2. Ground-based measurement of Avicennia marina tree height using a leveling staff placed adjacent to the trunk. These photos were taken at an angle ensuring the visibility of both the crown apex and the measurement scale to reduce parallax errors. The photos were taken on 22 November 2024 (see Figure 1 for the location of the numbered photos). Panels (a–c) show trees with different sizes, highlighting the structural variability among the sampled individuals.

Figure 2. Ground-based measurement of Avicennia marina tree height using a leveling staff placed adjacent to the trunk. These photos were taken at an angle ensuring the visibility of both the crown apex and the measurement scale to reduce parallax errors. The photos were taken on 22 November 2024 (see Figure 1 for the location of the numbered photos). Panels (a–c) show trees with different sizes, highlighting the structural variability among the sampled individuals.

The total area surveyed by drone, as delineated in Figure 1, was approximately 2.65 hectares, based on GIS analysis of the orthomosaic in ArcMap 10.8.2

The process began with the selection of the Melgonze mangrove patch as the study area, followed by in situ data collection, including tree height measurements and GPS-based tree localization. Subsequently, drone-based imagery was acquired using a consumer-grade drone flown at a 150 m altitude. Photogrammetric processing of the drone imagery generated key spatial products such as DSM, DTM, and CHM, from which canopy metrics were extracted. These data were integrated with field measurements to estimate AGB using a species-specific allometric model. A carbon conversion factor of 0.48 was then applied to estimate blue carbon stocks. Finally, statistical analyses including regression modeling, height class grouping, and paired t-tests were conducted to assess the accuracy and variability of the drone-derived estimates.

2.2. Field Data Collection

Field measurements were necessary to verify the accuracy of the drone-derived data, particularly tree height estimations based on 3D image models. To validate these measurements, the actual height (AH) of individual trees was recorded as the primary ground-truth dataset. A leveling staff (also called a leveling rod or Mire) was used to measure tree height. For each selected tree, the staff was placed vertically adjacent to the trunk, and a photograph was taken (Figure 2). The photographs were captured from an angle that allowed both the highest point of the tree crown and the corresponding scale on the staff to be clearly visible, minimizing visual distortion. Then, the highest visible point of the crown was considered the total tree height. Moreover, the coordinates of each tree were recorded using a handheld GPS Waypoints (Garmin 78s) device, enabling tree identification within the drone-derived orthomosaic. A total of 36 accessible trees were surveyed; however, due to errors (mostly coming from GPS-related positional), six of them could not be reliably located in the orthomosaic. Ultimately, height data from 30 individual trees (hereafter referred to as the field-measured samples) were successfully used for analysis.

To provide a clear overview of the research approach, a methodological workflow is illustrated in Figure 3.

Standard forestry instruments such as clinometers, hypsometers, and laser rangefinders are commonly used for tree height measurements in terrestrial forests [30]. However, in our field site, the highly muddy and waterlogged conditions limited the use of such instruments, as they require firm ground for stability and often line-of-sight calibration from a fixed distance. Instead, we employed a leveling staff and photographic method, which, although less common, has been adopted in previous mangrove studies for measuring trees of a small to moderate height (typically <6 m) [16,30]. This method allowed us to visually record the full tree profile and reference height against the scale on the staff. While we acknowledge this approach may introduce minor errors due to perspective or angle, we minimized distortion by adjusting the camera height and angle to ensure the visibility of both the crown apex and the base reference. Moreover, since our goal was to assess the relative performance of drone-derived data against field reference measurements, our method remained internally consistent and suitable for comparative analysis.

2.3. Drone-Based Data Acquisition

DJI^TM Phantom 4 Pro, a commercially available, low-cost, and consumer-grade drone, was utilized for aerial data acquisition in this study. Equipped with a standard RGB camera (non-multispectral), the drone’s flight planning and execution were managed using the Pix4Dcapture mobile application. The drone used was equipped with a 1-inch 20-megapixel CMOS sensor camera (DJI Phantom 4 Pro standard RGB payload), with an 84° field of view and a 24 mm equivalent focal length. This sensor provides high-resolution imagery suitable for photogrammetric applications in vegetation structure analysis. The drone operated at an altitude of 150 m above ground level, capturing a total of 168 nadir images with approximately 80% forward and side overlap, ensuring optimal photogrammetric coverage and geometric precision. The imagery obtained achieved a Ground Sampling Distance (GSD) of 4.29 cm, providing high spatial resolution for detailed canopy analysis. The survey was conducted on November 8, 2021, under clear weather conditions with low wind speed, ensuring maximum image quality and minimizing atmospheric distortion.

The GSD was calculated using the following formula:

$$GSD={\displaystyle \frac{H}{f}}\times a$$

(1)

where H represents the flight altitude (in meters), f is the focal length of the camera (in millimeters), and a denotes the pixel size (in millimeters). Once these parameters are defined, the corresponding applications can estimate the expected flight time and battery consumption, considering battery capacity [31].

2.4. Image Processing

The acquired aerial images were processed following a photogrammetric workflow to generate geospatial products required for tree height estimation. The process began with image alignment using feature matching techniques, wherein keypoints were detected across overlapping images and matched to form a sparse point cloud. Subsequently, a dense point cloud was generated using multi-view stereo (MVS) reconstruction with high-resolution settings and mild depth filtering, enabling detailed three-dimensional modeling of the scene.

From the dense point cloud, a digital surface model (DSM) was generated, representing elevation data for all above-ground features, including vegetation and structures. Ground point classification was then applied using slope-based and elevation filtering criteria to isolate terrain points. These ground points were interpolated to create a digital terrain model (DTM), representing the bare-Earth surface.

A digital elevation model (DEM) was also extracted prior to classification, representing a raw elevation surface that includes both terrain and canopy features. In this study, DEM refers to this unclassified model. Individual tree heights were measured manually by calculating the elevation difference between the highest point of each tree crown and the nearest ground point within the DEM, corresponding to the first location where the canopy visibly ends.

The canopy height model (CHM) was calculated by subtracting the DTM from the DSM using pixel-wise raster subtraction (map algebra), isolating the vertical structure of vegetation. To reduce artifacts caused by reflective water surfaces, a polygon mask was used to manually exclude water-covered areas. Additional noise elements such as shadows and image distortions were also reviewed and corrected manually.

Despite CHM generation, we chose to use the unclassified DSM (referred to as the DEM here) for the manual height extraction of supplementary trees. This approach allowed us to visually identify the highest crown point and subtract the nearest bare ground value manually for each tree, minimizing the effects of canopy overlap or misclassified pixels that were occasionally present in the CHM due to the proximity of water surfaces or shadows. While CHMs are generally more suitable for automated tree height estimation, this manual method ensured precise crown-to-ground height retrieval in our small sample.

A semi-automatic approach was applied in ArcMap, combining Marker-Controlled Watershed Segmentation with manual refinement to accurately delineate individual tree crowns, following the methodology described by Panagiotidis et al. (2017) [32].

2.5. Biomass and Blue Carbon Estimation

The estimation of AGB was performed using an allometric equation developed specifically for Avicennia marina by Owers et al. (2018) [29]:

$$AGB={1.088e}^{-5.392+1.185\times log \left(H\right)+1.142\times log \left(CA\right)}$$

(2)

where H is the height in centimeters and CA is the crown area in square meters; the resulting AGB is expressed in kilograms.

This model was selected because of its species specificity and compatibility for estimating biomass at the individual tree level in mangrove ecosystems. It should be mentioned that the CA of trees was extracted from the orthomosaic and was not measured in the field. The CA was extracted from the orthomosaic using the Measure tool in ArcMap, which allows for accurate two-dimensional measurements. For each individual tree, the major and minor crown diameters were measured based on drone-derived orthomosaic imagery. Assuming an elliptical crown shape, the CA was calculated using the standard ellipse formula (π × a × b), where a and b are the semi-major and semi-minor axes. This method ensured precise estimation of crown dimensions, leveraging the high spatial resolution of the georeferenced imagery.

For the estimation of BC, a factor of 0.48 to convert ABG into BC was applied, representing the proportion of carbon content in the biomass of A. marina. The 0.48 value was derived from Kauffman and Donato’s [33] research, provided the global default factor of 0.48 for AGB. While local assessments in some cases have suggested lower conversion factors (such as 0.42), we used the commonly applied value of 0.48 to align our results with globally comparable BC estimates.

2.6. Statistical Analysis

In this study, linear regression was used to analyze the relationship between tree height and crown diameter, utilizing data from both field measurements and drone-based estimates. Data from each source were processed and analyzed independently. Microsoft Excel facilitated the regression analysis and generated scatter plots to visualize the relationships:

(i)

Height vs. average crown diameter;

(ii)

Height vs. minor crown diameter;

(iii)

Height vs. major crown diameter.

The regression equations, including the intercept and R² values, were extracted directly from the plots. A paired t-test was performed to compare the height and biomass estimates from field measurements and drone data for field-measured samples. Following this, another 30 individual tree (supplementary drone sample) heights were extracted from the DEM. Tree selection in this group was based on their clear visibility within the orthomosaic, avoiding trees with overlapping canopies or significant shadow interference. This approach ensured accurate CA measurements and minimized errors in biomass estimation. And then, in order to explore structural variability and its influence on biomass and BC distribution, all of the 60 individual trees were categorized into three height classes:

(i)

Trees < 150 cm;

(ii)

Trees > 150 cm and < 300 cm;

(iii)

Trees > 300 cm.

Biomass and BC storage were then estimated separately for each height class. Similar breakpoints have been used in previous mangrove studies to distinguish trees by functional and ecological roles [24,25]. Additionally, this categorization enables a more structured comparison of drone estimation performance across height classes, as drone-derived height errors tend to be more pronounced in shorter individuals.

3. Results

3.1. Drone-Derived Geospatial Data Products

Four main spatial datasets were produced through the remote sensing workflow: the DSM, DTM, CHM, and orthomosaic (Figure 4). The DSM indicated the surface features of the study area, such as vegetation; the DTM indicated ground-level features; the CHM showed vegetation and changes in tree heights; the orthomosaic provided a high-resolution map of the study area, and it was used for identifying and referencing individual tree locations during analysis. Notably, the DSM and DTM revealed substantial elevation artifacts over water surfaces, highlighting the challenges of accurately modeling non-solid features in photogrammetric processing.

3.2. Field Measurements Output

To visualize the spatial distribution and height classification of the field-measured trees, 30 accessible Avicennia marina trees were mapped onto the drone-generated orthomosaic (Figure 5). Tree heights were categorized into three distinct classes, <150 cm, 150–300 cm, and >300 cm, allowing for a structured analysis of biomass and BC distribution across height groups. Trees with heights exceeding 300 cm, although relatively few in number, were primarily located in the northern and central sections of the study area and played a disproportionately large role in carbon accumulation. The mapped tree distribution also provided a visual validation of the spatial heterogeneity of the mangrove stand, supporting the integration of field data with remote sensing outputs.

Figure 4. Four main remote sensing outputs: (a) orthomosaic: a high-resolution true-color map generated from drone images, providing a detailed view of the study area; (b) digital surface model (DSM): a 3D representation of surface elevations, including vegetation, built structures, and other above-ground features; (c) digital terrain model (DTM): a representation of the bare ground surface, created by removing above-ground objects from the DSM; (d) canopy height Model (CHM): a model of vegetation height, calculated by subtracting the DTM from the DSM, providing a clear visualization of canopy structure across the landscape.

Figure 4. Four main remote sensing outputs: (a) orthomosaic: a high-resolution true-color map generated from drone images, providing a detailed view of the study area; (b) digital surface model (DSM): a 3D representation of surface elevations, including vegetation, built structures, and other above-ground features; (c) digital terrain model (DTM): a representation of the bare ground surface, created by removing above-ground objects from the DSM; (d) canopy height Model (CHM): a model of vegetation height, calculated by subtracting the DTM from the DSM, providing a clear visualization of canopy structure across the landscape.

Figure 5. Spatial distribution of field-measured individual trees overlaid on the drone-derived orthomosaic. Trees are categorized into three height classes: trees shorter than 150 cm (small red dots), trees between 150 and 300 cm (medium red dots), and trees taller than 300 cm (large red dots). Tree height values (in centimeters) are labeled next to each tree location.

Figure 5. Spatial distribution of field-measured individual trees overlaid on the drone-derived orthomosaic. Trees are categorized into three height classes: trees shorter than 150 cm (small red dots), trees between 150 and 300 cm (medium red dots), and trees taller than 300 cm (large red dots). Tree height values (in centimeters) are labeled next to each tree location.

Table 1. Descriptive statistics of tree height measurements across three defined height classes.

Height Class	Count (n)	Min (cm)	Max (cm)	Mean (cm)	Median (cm)	SD (cm)
All	60	55	515	215	209	90
<150 cm	17	55	148	112	115	25
150–300 cm	31	154	295	220	220	35
>300 cm	12	309	515	350	335.5	56

summarizes the descriptive statistics for each class, including the minimum, maximum, mean, median, and standard deviation of tree heights. This classification allows for a clearer understanding of the role of different tree sizes in biomass accumulation and carbon storage potential.

3.3. Field and Drone Data Comparison

Field-measured and drone-derived tree heights were compared. The relationship between drone-derived and field-measured tree heights was assessed using regression analysis. The comparison results are presented in Figure 6. The result of the paired t-test between field-measured and drone-derived tree heights showed a statistically significant difference (p-value ≈ 1.03 × 10⁻⁵), indicating that the two sets of measurements are not identical. This difference may reflect minor discrepancies introduced by drone-based estimation methods.

In addition, Figure 7 presents a comparison of AGB and BC estimates using actual and drone-derived heights for the same set of trees. The paired t-test comparing AGB and BC estimates based on actual and drone-derived heights yielded a p-value of approximately 0.116. This suggests that there is no statistically significant difference between the two estimation methods at the 0.05 level, indicating that drone-derived heights can provide comparable carbon estimates to field-based measurements for the same set of trees.

Linear regression models were utilized to examine the relationship between tree height and crown diameters (average, minor, and major), providing insights into the structural variability within the mangrove stand (Figure 8). This analysis highlights the consistency and reliability of drone-based data, and a strong correlation between tree height and diameter.

One of the observed differences between the two groups was the higher R² values among the supplementary drone samples. Since taller trees were selected for supplementary drone samples, crown diameters in this group had generally higher values, as shown in

Table 2. Supplementary drone sample.

Tree No.	DH (cm)	CA (m2)	AGB (kg)	BC (kg)
1	180	26.64	90.38	43.38
2	140	9.89	22.25	10.68
3	92	3.75	4.59	2.2
4	235	31.24	148.06	71.07
5	213	20.59	82.8	39.74
6	205	10.83	38.68	18.56
7	220	15.57	63.02	30.25
8	183	14.81	47.91	22.99
9	335	58.89	456.84	219.28
10	230	29.37	134.74	64.67
11	180	10.26	31.2	14.97
12	154	6.14	14.63	7.02
13	242	25.65	123.09	59.08
14	171	15.13	45.28	21.73
15	186	8.29	25.58	12.28
16	248	29.36	147.28	70.69
17	350	77.4	652.58	313.24
18	329	67.68	522.22	250.66
19	199	10.19	34.87	16.74
20	188	16.32	55.14	26.46
21	196	14.52	50.83	24.4
22	336	43.07	323.53	155.29
23	372	94.1	872.17	418.64
24	201	26.12	100.79	48.38
25	309	40.1	270.49	129.83
26	378	100.36	954.94	458.37
27	325	65.18	493.57	236.91
28	191	9.81	31.86	15.29
29	338	43.17	326.65	156.79
30	131	7.41	14.9	7.15

.

3.4. Allometric Modeling Results

Allometric modeling was used to estimate the AGB and BC content of individual trees using the measured tree height and CA as input variables. The CA was calculated using both the minor and major crown diameters assuming an elliptical shape, but the diameters themselves were not directly included in the model. The equation proposed by Owers et al. (2018) [29] was used to calculate AGB, and a carbon conversion factor of 0.48 was applied. Results are presented for three tree groups: (

Table 3. Field-measured sample with AH.

Tree No.	AH (cm)	CA (m2)	AGB (kg)	BC (kg)
1	130	1.02	1.62	0.77
2	100	0.38	0.4	0.19
3	102	0.63	0.71	0.34
4	305	23.07	143.82	69.03
5	240	18.80	86.18	41.36
6	130	1.32	2.17	1.041
7	275	24.62	136.78	65.65
8	105	0.44	0.49	0.23
9	196	1.88	5.22	2.5
10	95	0.37	0.36	0.17
11	55	0.12	0.05	0.02
12	237	22.54	103.97	49.9
13	267	12.17	60.23	28.91
14	118	0.9	1.26	0.6
15	221	3.8	13.15	6.31
16	228	6.53	24.96	11.98
17	115	0.37	0.46	0.22
18	70	0.28	0.18	0.08
19	107	2.37	3.3	1.58
20	130	13.19	28.08	13.48
21	220	30.67	134.17	64.4
22	295	19.87	117.04	56.18
23	227	10.36	41.54	19.94
24	140	5.89	12.47	5.99
25	148	6.86	15.79	7.58
26	250	9.62	42.86	20.57
27	283	17.15	94.57	45.39
28	260	7.06	31.84	15.28
29	316	23.84	155.59	74.68
30	515	53.01	676.3	324.62

) the field-measured sample with AHs, (

Table 4. Field-measured sample with drone height.

Tree No.	DH (cm)	CA (m2)	AGB (kg)	BC (kg)
1	62	1.02	0.67	0.32
2	6	0.38	0.01	0.006
3	23	0.63	0.12	0.05
4	294	23.07	137.69	66.09
5	259	18.80	94.32	45.27
6	81	1.32	1.23	0.59
7	252	24.62	123.33	59.2
8	46	0.44	0.18	0.08
9	102	1.88	2.41	1.15
10	29	0.37	0.09	0.04
11	12	0.12	0.009	0.004
12	207	22.54	88.57	42.51
13	213	12.17	46.09	22.12
14	51	0.9	0.46	0.22
15	96	3.8	4.89	2.35
16	147	6.53	14.84	7.12
17	16	0.37	0.04	0.02
18	20	0.28	0.04	0.02
19	73	2.37	2.09	1
20	185	13.19	42.66	20.47
21	285	30.67	182.33	87.51
22	170	19.87	60.91	29.24
23	195	10.36	34.7	16.65
24	139	5.89	12.37	5.93
25	157	6.86	16.94	8.13
26	203	9.62	33.48	16.07
27	279	17.15	92.99	44.63
28	196	7.06	22.78	10.93
29	289	23.84	139.97	67.18
30	463	53.01	596.17	286.16

) the same sample using drone-derived heights (DHs), and (Table 2) the supplementary drone sample. These estimates provide a basis for analyzing biomass patterns and carbon storage capacity in the study area.

3.5. Tree Height Grouping and Carbon Distribution

To explore structural variability in the mangrove stand, a total of 60 trees (30 field-measured sample trees + 30 supplementary drone sample trees) were grouped into three height classes: <150 cm, 150–300 cm, and >300 cm. The associated AGB and BC were summed up for each class. Figure 9 illustrates the proportion of trees and carbon stocks in each category. Particularly, trees taller than 300 cm, despite representing only a minority of the sample, accounted for 72% of the total BC. Medium-height trees (150–300 cm) contributed 27%, while short trees (<150 cm) contributed only 1%. This result emphasizes the disproportionate contribution of tall mangroves to overall carbon storage.

3.6. Total Estimated Biomass and Blue Carbon

Ultimately, the total AGB and BC storage values for the field-measured sample group, when using actual tree heights, were 1936 kg and 929 kg, respectively. When drone-derived heights were used for the same group, the totals decreased to 1753 kg for biomass and 841 kg for BC. In comparison, the supplementary drone sample group, which relied solely on drone-derived data, yielded totals of 6181 kg for AGB and 2966 kg for BC. These values were considerably higher than those of the field-measured sample (1753 kg AGB and 841 kg BC), primarily due to the greater proportion of tall trees (>300 cm) and the presence of fewer short trees (<150 cm) in the drone dataset. Many of the taller individuals were located in extremely muddy and inaccessible areas, making them unsuitable for field measurement. Conversely, the field-measured group consisted mostly of shorter trees, selected based on ease of access. This contrast highlights both the advantages of drone-based sampling in capturing spatial variability and its potential for improving large-scale biomass estimation in challenging mangrove environments.

4. Discussion

4.1. Overview of Key Findings

This study demonstrated the potential of drones as reliable tools for estimating biomass and BC in A. marina, even under challenging coastal conditions. Our results revealed a strong correlation between tree height and crown diameter, including average, minor, and major diameters, for A. marina in both field-measured and drone-derived datasets. Although the paired t-test indicated a statistically significant difference between actual and drone-derived tree heights, the estimated AGB and BC values derived from both datasets showed no significant difference. This outcome suggests that the CA, in combination with the coefficients in the allometric model, effectively buffered the influence of minor height discrepancies, thereby supporting the reliability of the applied equation.

In the initial version of this study, a local conversion factor of 0.42 was used to estimate BC from AGB. However, for broader applicability and alignment with global standards, we adopted the widely used conversion factor of 0.48, as recommended by Kauffman and Donato [33], which has been applied in numerous similar studies worldwide.

These findings are important for advancing cost-effective mangrove monitoring strategies, especially in areas where ground access is limited or unsafe. Our results are consistent with recent studies emphasizing the potential of drones for coastal ecosystem monitoring (e.g., [10,15]), and they strengthen the case for wider operational use of drone technologies in BC assessments. The findings of this study can be applied to other mangrove ecosystems dominated by A. marina, providing a scalable approach for BC estimation and contributing to more accurate carbon stock assessments in similar coastal environments.

4.2. Evaluation of Drone Accuracy and Sources of Error

Several previous studies have highlighted the potential of remote sensing tools—including UAVs, LiDAR, and satellite stereo imagery—for estimating mangrove tree heights, yet have also acknowledged various sources of error. For instance, Yin et al. reported average estimation deviations within 0.1 m using UAV-LiDAR CHMs [34]; Lagomasino et al. found height errors ranging from 1.33 to 1.88 m using WorldView-1 DSMs and CHMs [35]; and Fu et al. observed that more than 75% of UAV- and SAR-based height predictions fell within 10% of measured values [36]. These studies underscore that even advanced systems are not immune to estimation challenges, particularly in structurally complex or heterogeneous environments.

In our study, the greatest discrepancies between drone-derived and field-measured tree heights were observed in individuals shorter than 150 cm, which represented 28% of the total sample (17 trees). Shorter trees are more susceptible to height estimation errors due to their smaller size and the influence of surrounding vegetation in drone images. However, this group accounted for only 1% of the total BC, indicating that despite greater measurement uncertainty, their impact on overall carbon estimates was minimal. This finding supports the reliability of drone-based estimations, especially given the dominant contribution of taller trees to total biomass. To further minimize such discrepancies, lower flight altitudes are recommended for future data collection. It should be mentioned that several factors likely contributed to the observed discrepancies, including flight altitude (150 m), image resolution, and the timing mismatch between drone flights (2021) and field surveys (2024). The temporal mismatch between drone flights and field surveys is a key factor contributing to the observed differences. Although A. marina grows slowly, even minor growth over three years may affect height measurements. Nonetheless, the strong height–crown diameter correlations in both datasets (R² > 0.7) suggest that drone data retained a high degree of internal consistency. Future studies could further minimize errors by optimizing flight altitude, using multispectral or LiDAR-equipped drones, employing higher-resolution sensors, and synchronizing drone and field data collection periods.

While CHMs are commonly used for canopy height estimation, we found that in some locations—particularly near reflective water or near an area of dense canopy overlap—the CHM exhibited local artifacts. Thus, we used the DSM and terrain references for controlled manual height measurement in our supplementary drone sample.

4.3. Comparison with Previous Studies

As the application of drones in coastal ecosystem monitoring has expanded [37], several recent studies have demonstrated their effectiveness in estimating biomass and carbon stocks in mangrove forests, aligning with the methodology adopted in this study. For example, Navarro et al. (2020) [15] utilized drone-SfM techniques to derive accurate tree-level height, crown dimensions, and AGB values, comparable to our structure-based estimations. Similarly to our findings, Owers et al. (2018) [29] applied an allometric equation using height and CA to estimate AGB in A. marina, demonstrating the effectiveness of these two variables for biomass estimation without requiring direct trunk diameter measurements. The accuracy of tree height and crown diameter estimates derived from UAV imagery in this study is consistent with the results reported by Panagiotidis et al. (2017) [32], who demonstrated that high-resolution drone-based photogrammetry can effectively capture individual tree metrics in dense forest environments. However, our workflow included additional manual corrections and canopy refinement to address specific challenges associated with mangrove environments, such as water-induced artifacts and overlapping crowns. Larekeng et al. [38] employed the multispectral drone-derived NDVI to estimate carbon in Indonesian mangroves, highlighting drones’ potential in BC assessment despite using a spectral rather than structural approach. Shaltout et al. [39] used field measurements of height and crown size to estimate AGB in A. marina stands in arid coastal zones, adopting a 0.48 conversion factor to derive carbon from biomass, matching our conversion coefficient. Most notably, Blanco-Sacristán et al. [40] combined drone-derived Leaf Area Index (LAI) maps with leaf-level photosynthesis rates to map carbon assimilation in A. marina along the Red Sea. Their findings showed that tall trees (>3 m) contributed disproportionately to carbon assimilation, reinforcing the core result of our study, where only 12 trees taller than 3 m accounted for 72% of the total carbon stored across all 60 trees sampled. Previous studies reveal that mangrove reforestation contributes to small emission reduction, so mangrove conservation is still the priority [41]. Our results show the important role of old trees (>3 m) in carbon sequestration, which is consistent with the results of previous studies. Moreover, our study supports the conclusion that structural metrics such as canopy area and height, extracted through photogrammetry, can serve as viable proxies for biomass estimation even without direct trunk diameter measurements—an observation echoed by Jones et al. (2020) [18]. Unlike most previous studies that used multispectral or LiDAR-equipped drones, our approach, involving a consumer-grade RGB drone, demonstrated comparable accuracy in carbon estimation, offering a cost-effective alternative for large-scale assessments.

4.4. Practical Implications for Conservation and Research

As climate change accelerates, the importance of BC ecosystems, such as mangrove forests, as major carbon reservoirs is becoming increasingly evident. These ecosystems not only sequester significant amounts of atmospheric carbon but also provide critical ecosystem services, including coastal protection, habitats for marine species, and support for local livelihoods. Effective management and conservation of these ecosystems depend on accurate carbon stock assessments, which can inform climate mitigation strategies and coastal management policies.

Drones offer a transformative approach to carbon estimation in such environments, providing scalable, efficient, and non-invasive data collection. Our study demonstrated that drone-based methods could accurately estimate AGB and BC, even in challenging conditions such as the muddy shoreline of the Mond. Notably, many of the supplementary drone sample trees in our study were inaccessible through ground-based methods, highlighting the significant advantage of drones in such terrains.

Beyond direct data collection, drones also present cost-saving benefits by reducing field time and minimizing on-site risks for researchers [15]. The ability to archive and revisit drone data provides a valuable resource for long-term monitoring, enabling the detection of changes in canopy structure, tree health, and overall carbon storage over time. Such historical datasets are critical for assessing the impacts of climate change and human activities on mangrove ecosystems.

4.5. Limitations and Future Work

The application of drones in BC monitoring is not without challenges. Image processing and data analysis require technical expertise, which may limit the method’s immediate applicability for conservation managers without specialized training. Additionally, the effectiveness of drone-based biomass estimation may vary depending on forest density, as noted by Navarro et al. [15]. In dense canopies, individual tree segmentation becomes difficult, potentially affecting measurement accuracy.

Despite the numerous advantages of drone-based data collection, several challenges emerged during this study. The most significant limitation involved the legal and logistical restrictions associated with drone operations. Acquiring flight permits was particularly difficult, and in some locations, such as near airports, military zones, or correctional facilities, permission was actually unattainable. These constraints resulted in delayed drone deployment, which, in our case, led to a temporal mismatch between drone imagery and fieldwork. Specifically, aerial images were captured on 8 November 2021, while ground-based measurements were conducted on 22 November 2024.

This temporal gap may partially explain why the regression coefficients between drone-derived height and crown diameter (R² = 0.88, 0.84, 0.90) were noticeably higher than those between field-measured height and crown diameter (R² = 0.77, 0.76, 0.76). Since the crown diameter in both datasets was derived from the same set of drone images, the consistency of temporal reference likely improved the performance of drone-based models. Although Avicennia marina is a slow-growing species, we believe this time discrepancy remains the most plausible explanation for the observed difference in model accuracy.

Additionally, drone-derived height estimates for shorter trees introduced considerable uncertainty, suggesting that a flight altitude of 150 m may be suboptimal for monitoring low-stature vegetation. Environmental factors can also influence drone performance. While we did not encounter major issues during flights, windy conditions can reduce battery efficiency by increasing the need for stabilization, which may limit flight duration and reduce spatial coverage in similar studies. High-resolution image acquisition—necessary for reliable modeling—requires substantial data storage capacity, which can become a constraint in large-scale studies. Furthermore, 3D model generation over water surfaces resulted in elevation artifacts. Sunlight reflections on the water also interfered with image quality, reducing the effectiveness of surface reconstructions. Ideally, drone flights should be conducted on bright, overcast days to balance lighting and minimize glare, yet such conditions are rare along the southern coast of Iran.

While this study successfully estimates AGB and BC at the individual tree level, no extrapolation was performed to estimate total carbon storage across the entire mangrove area. This is a recognized limitation, as upscaling would require detailed information on tree density, spatial heterogeneity, and the full extent of the mangrove forest, which were beyond the scope of this pilot study. Future research will aim to combine plot-based sampling with drone-derived canopy metrics and spatial mapping to generate area-wide estimates of carbon stocks. Such efforts can benefit from advanced approaches already demonstrated in recent studies. For instance, Huang et al. [42] have shown that integrating UAV-LiDAR data with machine learning models, such as support vector machines, can yield highly accurate predictions of mangrove AGB, achieving coefficients of determination up to 0.89 and RMSE values around 0.48 kg/m². These results highlight the potential of remote sensing and AI-based methods for scaling up carbon assessments in similar ecosystems [42].

To enhance the reliability of drone-based carbon assessments, future studies will focus on the following: exploring drone applications in different mangrove ecosystems with varying canopy densities, investigating the potential of integrating multispectral or LiDAR-equipped drones for improved canopy structure analysis, developing user-friendly software solutions that can streamline image processing for non-experts, lowering the flight altitude (e.g., 50–100 m) for studies targeting shorter vegetation, and implementing standardized protocols for drone data collection, ensuring consistency across studies. A previous study by Sadeghi and Sohrabi (2019) found that a flight altitude of 100 m provided the highest accuracy in tree height estimation (R² = 0.83, RMSE = 0.31 m), further supporting the importance of selecting an optimal altitude for UAV-based forest structure analysis [43].

5. Conclusions

In this study, we evaluated the capabilities and reliability of drones in estimating AGB and BC in Avicennia marina. Our combined approach involving field measurements and drone imagery allowed for the accurate assessment of individual trees. The strong correlations between tree height and crown diameter support the use of drones for mangrove monitoring. Although drone-derived tree heights significantly differed from field measurements, the final estimations of AGB and BC based on both height sources showed no significant difference.

It is important to note that our estimates were limited to individual trees and did not aim to provide area-wide biomass or carbon extrapolations. The study included 30 field-measured trees and 30 drone-measured trees, which, while informative, represent a small sample size and should be interpreted accordingly. Nevertheless, drone-based estimates closely matched field-based ones at the individual level. As shown in Figure 7, there was no significant difference in BC values per tree when using drone-derived versus field-measured heights. The total BC derived from drone data (841 kg) differed by only 88 kg from the field-based estimate (929 kg), a relatively small margin considering the scale and sample size of the study. These results highlight the viability of drone data for accurate carbon estimation in similar ecological contexts.

These findings support the use of drones as practical tools for monitoring mangrove forests. Drones offer several advantages, including cost and time efficiency, as well as access to otherwise impassable environments—such as muddy, tidal zones—thus reinforcing their value in coastal management programs.

Nevertheless, several challenges remain. Legal restrictions, weather conditions, and surface reflections can limit drone applications. In our case, regulatory constraints caused a three-year gap between drone imaging and field measurements. While such a time gap can be advantageous in long-term monitoring, it introduced inconsistencies in our comparative analysis.

Furthermore, drone-derived height errors were most prominent among smaller trees (<150 cm); however, since these individuals contributed minimally to total AGB and carbon, their impact was negligible. To improve accuracy in such cases, we will investigate lower flight altitudes in future research. In addition, we intend to explore machine learning methods such as random forest as part of future studies with larger datasets, to evaluate their performance relative to allometric models used here.

In conclusion, we believe drones are valuable and increasingly accessible tools for coastal conservation initiatives. As drone technology continues to advance, their integration into ecological research and environmental management is expected to expand.