Integrating Remote Sensing and Field Data to Quantify Mangrove Biomass Carbon Hotspots and Protection Gaps in Guangdong, China

Abstract

Mangroves are important blue carbon coastal ecosystems and play a crucial role in mitigating global climate change. However, fine spatial patterns of mangrove biomass carbon hotspots and protection gaps in Guangdong have not been quantified. In this study, we mapped mangrove biomass carbon by integrating Sentinel-2 satellite imagery and field survey data from Guangdong’s coastlines acquired in 2023 for the first time. Using the Getis-Ord Gi* spatial statistic method, we identified the mangrove biomass carbon hotspots and highlighted protection gaps in mangrove conservation. The total mangrove biomass carbon of Guangdong was estimated to be 1,209,305.68 Mg C (with a mean density of 80.56 Mg C/ha), with Zhanjiang containing the highest carbon stock, accounting for over half of the total. Nature reserves supported higher mean biomass carbon (83.03 Mg C/ha), compared with areas outside nature reserves (77.99 Mg C/ha), underscoring their important role in enhancing mangrove carbon storage. The overlapping area between the mangrove biomass carbon stock hotspot areas and the nature reserves is 71.62 km², accounting for 51.13% of the total hotspot area. In terms of mangrove biomass carbon stocks, the main protection gaps in Guangdong are distributed in Anpu Gang, the region south of Zhanjiang, Shuidong Harbor, Pearl River Estuary, Kaozhou Yang, and Yifengxi Port. Our findings reveal the spatial heterogeneity of mangrove carbon stocks in Guangdong and provide novel insights for optimizing mangrove management and spatial planning of nature reserves for conservation and restoration.

1. Introduction

Mangroves growing on tropical and subtropical coastal intertidal coasts play critical roles in safeguarding marine ecological security, conserving biodiversity, improving and maintaining local water quality, sequestrating carbon, and mitigating climate change [1,2,3,4,5,6,7,8]. Despite multiple ecosystem services and functions, global mangroves have faced problems of deforestation and degradation at a high rate [9,10,11]. China is among the few countries worldwide that has experienced a net expansion of mangrove coverage [12]. China attached great importance to the conservation and restoration of mangroves, through multi-dimensional measures including top-level design, legislative safeguard, and restoration projects, among other conservation initiatives. The enforcement of key laws and regulations, such as the “Marine Environmental Protection Law”, “Regulations of the People’s Republic of China on Nature Reserves”, “Ecological Conservation Red Line”, and the “Wetland Protection Law”, is critical for effective mangrove conservation [12,13]. The Chinese government has established nature reserves and marine ecological red lines to counteract the negative impacts of anthropogenic activities on mangroves [14,15]. Furthermore, China has strengthened its marine conservation and restoration planning framework through the enactment of several strategic plans, notably the Special Action Plan for Mangrove Conservation and Restoration (2020–2025).

Researchers have undertaken to assess the effectiveness of the protection work for mangrove conservation. Jia et al. (2016) [16] compared the conservation effectiveness on mangroves in the Futian reserve and the Mai Po reserve using Landsat images from 1973 to 2015 and landscape metrics. Lu et al. (2022) [14] assessed the effectiveness of conservation within China’s national mangrove nature reserves, using Landsat images from 1987 to 2019, as well as landscape metrics and an entropy weight model. Liu et al. (2022) [17] developed an index framework and a landscape health composite index (LHCI) to assess the landscape health status of mangroves within mangrove nature reserves in China, with mangrove distribution datasets from 1978 to 2018. Sun et al. (2025) [12] assessed the landscape health of mangroves within 37 nature reserves and non-reserve areas, based on Sentinel-2 imagery from 2016 to 2023, deep learning models, and LHCI. Liang et al. (2025) [15] assessed the efficacy of marine ecological red lines in protecting mangroves, focusing especially on carbon storage and mangrove coverage. Utilizing the InVEST (version 3.13.0) Carbon model, Landsat imagery from 2013 to 2017, and field survey sampling plot data, they quantified mangrove carbon storage specifically within the ecological red lines, without considering nature reserves in the assessment [15]. Thus, researchers commonly employed landscape metrics or integrated landscape evaluation systems to assess the effectiveness of mangrove conservation within nature reserves [12], rarely considering carbon stocks to quantify the protection gaps, despite their significance to the fulfillment of China’s Dual Carbon (carbon peaking and neutrality) strategy. In addition, few studies have examined the relationship between nature reserves and mangrove carbon stocks.

Traditional mangrove field surveys are point-specific, time-consuming, and laborious [18,19,20,21,22], while remote sensing enables comprehensive wall-to-wall monitoring of mangroves [23,24]. Researchers have explored multi-sensor, multi-spatial-resolution, multi-spectral satellite imagery, various classification algorithms, and new remote sensing platforms like UAVs for accurate mangrove monitoring; please refer to Giri et al. (2016) [25], Wang et al. (2019) [26], Lu et al. (2021) [27], and Tran et al. (2022) [28] for details. Roy et al. (2024) [29] recently summarized the mangrove blue carbon estimation research findings based on remote sensing coupled with modeling, and pointed out that belowground mangrove blue carbon was less studied. Meng et al. (2022) [30] estimated the mangrove ecosystem carbon stocks in Hainan by modeling the relationship between field survey aboveground carbon stock and Sentinel-2 image features and correlating the aboveground and belowground carbon stocks. Dong et al. (2025) [13] quantified the spatiotemporal evolution of mangrove and salt marsh carbon stocks from 1986 to 2020 in Guangdong Province, integrating a novel mangrove and salt marsh mapping approach using Landsat time-series imagery (30 m) and the InVEST model. Pirasteh et al. (2024) [31] developed a regression model between mangrove biomass and NDVI, and mapped mangrove biomass in Persian Gulf coasts. Ali and Rahman (2025) [32] quantified the long-term changes in mangrove canopy height, aboveground biomass (AGB), and carbon stocks, by leveraging Shuttle Radar Topography Mission, GEDI LiDAR datasets, and a widely used model which corelates mangrove AGB with mean canopy height. Dey et al. (2025) [33] investigated the influence of human activities on mangrove carbon storage potential along the Gulf of Kachchh by combining the field data of sediment organic carbon and remote sensing NDVI values. However, fine-scale spatial patterns of Guangdong mangrove biomass carbon remain poorly quantified.

Although strictly protected, mangroves in China are still suffering from degradation owing to widespread human disturbances [34]; and ecosystem services, such as carbon sequestration, should be considered for future nature reserve management plans [35]. Guangdong Province in China is a highly significant and representative region for assessing how nature reserves affect mangrove ecosystem quality due to its extensive and diverse mangrove habitats, presence of multiple nature reserves, and varying levels of anthropogenic pressure [15]. Hence, the specific objectives of this study are (1) to map the mangrove biomass carbon by integrating remote sensing and field data; and (2) to quantify the mangrove biomass carbon hotspots and protection gaps in Guangdong, China.

2. Materials

2.1. Study Area

This study focused on the coastal regions of Guangdong Province, located in southern China (Figure 1). The coastal region is characterized by numerous bays, extensive tidal flats, and a broad continental shelf, which create a stable and diverse geological environment that facilitates wetland sediment accumulation and provides habitats for marine organisms and ecosystems. This region falls within the southern subtropical monsoon zone, characterized by abundant solar radiation and heat, with precipitation and temperature largely coinciding in season. The mean annual rainfall is approximately 1798.8 mm, with about 80% occurring between April and September. The combination of high rainfall and elevated temperatures provides critical hydrothermal conditions for mangrove growth and the maintenance of ecosystem functions. Guangdong has the longest coastline in China, extending for about 4100 km, with an annual river runoff of 311.33 billion m³ entering the sea. The complex interplay of freshwater input from rivers and saline water from the South China Sea creates diverse ecological conditions ideal for various mangrove species. Given the strategic ecological importance of these mangroves in carbon sequestration and coastal protection, coupled with significant anthropogenic pressures from rapid urbanization and economic activities, Guangdong provides a critical and representative area for assessing mangrove carbon stocks.

2.2. Datasets

2.2.1. Satellite Data and Preprocessing

All Sentinel-2 Surface Reflectance (SR) images with less than 10% cloud cover along Guangdong coastlines from June to November in 2023, obtained from the Copernicus Data Space Ecosystem (https://dataspace.copernicus.eu/, accessed on 1 February 2024), were utilized to monitor mangrove biomass carbon stocks. For every Sentinel-2 image, we used eight spectral bands, which were blue (B2), green (B3), red (B4), vegetation red edge (B5, B6, and B7), near-infrared (B8), and short-wave infrared (B11). Since the spatial resolution of B2, B3, B4, and B8 is 10 m, we resampled the images of B5, B6, B7 and B11 (20 m) to a resolution of 10 m. We employed a mangrove distribution map of Guangdong in 2023 which was based on high spatial resolution domestic satellite data (spatial resolution better than 2.5 m), UAV photos using DJI Phantom 4 RTK (DJI, Shenzhen, China), and the visual interpretation method [36], in order to obtain the detailed information of the mangrove distribution boundaries with a detection accuracy higher than 90%. Then, we generated the Sentinel-2 image composite by deriving the median value of each band from the gathered Sentinel-2 satellite imagery [13], and applied the mangrove distribution map to crop the median composite with Arcmap software (version 9.3). Generating the median composite could reduce the influence of transient tidal conditions across different acquisition dates of Sentinel-2 satellite images.

2.2.2. Field Data

For the mangroves of Guangdong Province, we selected representative mangrove distribution areas for sampling and set up 50 field survey plots from June to November in 2023 (Figure 1). The field survey plot sampling strategy, although constrained by accessibility and conservation regulations, intentionally covered the main mangrove regions in both western Guangdong and the Pearl River Estuary to capture the key mangrove structural and compositional variability (

Table 1. Information for the study sites in Guangdong with direct measurements.

Site Name	Latitude	Longitude	Number of Field Survey Plots
Yingluo Harbor (YLH)	$21°31\prime$–$21°35\prime$ N	$109°45\prime$–$109°47\prime$ E	15
Techeng Island (TCI)	$21°9\prime$–$21°10\prime$ N	$110°25\prime$–$110°{27}^{\prime }$ E	5
Tongming (TM)	$20°59\prime$–$21°$ N	$110°10\prime$–$110°11\prime$ E	9
Shenzhen Bay (SZB)	$22°31\prime$–$22°32\prime$ N	$114°$–$114°2\prime$ E	8
Qi’ao Island (QAI)	$22°25\prime$–$22°27\prime$ N	$113°37\prime$–$113°38\prime$ E	7
Kaozhou Yang (KZY)	$22°43\prime$–$22°45\prime$ N	$114°55\prime$–$114°56\prime$ E	6

). Within each 10 m × 10 m field survey plot, we quantified mangrove species composition and individual counts and measured canopy height and diameter at breast height for distinct mangrove species. We used mangrove allometric equations (

Table 2. Allometric equations and wood density of mangrove species sampled in this study [15].

Species	Allometric Equation *	Wooden Density a	References
Aegiceras corniculatum	$AGB=0.021\times D_0^{2.019}$	0.597	[40]
Avicennia marina	$AGB=0.308\times D^{2.11}$	0.732	[41]
	$BGB=1.28\times D^{1.17}$		[42]
	$TB=1.507\times D^{1.595}$		[41]
Excoecaria agallocha	$AGB=0.139\times D^{2.1992}$	0.429	[43]
Laguncularia racemosa	$AGB=0.102\times D^{2.50}$	0.610	[44]
Sonneratia caseolaris	$AGB=0.000318\times D_{0.3}^{4.19917}$	0.534	[45]
	$BGB=0.000431\times D_{0.3}^{5.56175}$
Sonneratia apetala	$AGB=0.213\times D^{2.187}$	0.478 b	[46]
	$BGB=0.667\times D^{1.982}$
	$TB=0.228\times D^{2.218}$
Kandelia obovata	$AGB=651.628\times ({D_m^2\times H)}^{1.053}$	0.523	[40]
	$BGB=271.019\times ({D_m^2\times H)}^{0.99}$
Bruguiera gymnorhiza	$TB=105.196\times ({D_{0.6}^2\times H)}^{0.889}$	0.868	[46]
Rhizophora stylosa	$AGB=0.105\times D^{2.68}$	0.940	[47]
	$BGB=0.134\times D^{2.40}$
Common c	$AGB=0.251\times \rho \times D_c^{2.46}$	-	[41]
	$BGB=0.199\times {\rho }^{0.899}\times D_c^{2.22}$

^a Wooden density data is from World Agroforestry Center (https://apps.worldagroforestry.org/sea/Products/AFDbases/WD/Index.htm, accessed on 15 February 2024). ^b Wooden density is from [46]. ^c Other mangrove species used common equation. * D₀ refers to the diameter of basal stem, D_0.3 is diameter at 0.3 m above the ground, D_0.6 is diameter at 0.6 m above the ground, D_m is diameter at breast height of Kandelia obovata, D_c is DBH for other species.

) to calculate AGB and belowground biomass (BGB) for distinct mangrove species. For the mixed-species field survey plot, we calculated the biomass of each tree according to the species from Table 2 and obtained the total mangrove biomass value for this plot. The mangrove biomass carbon stock was calculated by multiplying the vegetation biomass and carbon concentration ratio [30,37]. According to the previous literature, the carbon concentration ratio for AGB was 0.48 [38], while the carbon concentration ratio for BGB was 0.39 [37]. The mangrove biomass carbon stock (BC) contains the aboveground biomass carbon (AGBC) and belowground biomass carbon (BGBC) (i.e., BC = AGBC + BGBC) [39].

2.2.3. Nature Reserve Data

We collected the planning maps of nature reserves from departments of natural resources and manually delineated the polygon vectors of each nature reserve in Guangdong [12] using Arcmap software. We employed two experienced GIS digitization experts and ensured the accuracy and reliability of the nature reserve area digitization data through mutual verification.

3. Methods

3.1. Mangrove Carbon Stock Estimation with Satellite and Field Data

3.1.1. Modeling the Relationship Between AGB and Sentinel-2-Based Features

In this study, we selected 13 Sentinel-2-based features to estimate the mangrove AGB by applying the random forest (RF) algorithm to the Sentinel-2 image composite (Figure 2). RF is used here because it can model non-linear relationships between spectral/structural predictors and biomass, and generally yields higher predictive accuracy compared to traditional regression for heterogeneous ecosystems like mangroves. This approach has been widely used for mangrove biomass estimation [30,48,49,50].

The selected features include B2, B3, B4, B5, B6, B7, B8, B11, Normalized Difference Vegetation Index (NDVI) [51], Enhanced Vegetation Index (EVI) [52], Difference Vegetation Index (DVI), Simple Ratio Index (SRI) [53], and one texture feature. The texture feature is the standard deviation of NDVI computed over a neighborhood with a 5-pixel radius [54]. The RF model was built with a number of trees of 100, and the number of variables per split was set as the square root of the number of variables [13,54].

$$NDVI={\displaystyle \frac{{\rho }_{nir}-{\rho }_{red}}{{\rho }_{nir}+{\rho }_{red}}}$$

(1)

$$EVI=2.5\times {\displaystyle \frac{{\rho }_{nir}-{\rho }_{red}}{{\rho }_{nir}+6\times {\rho }_{red}-7.5\times {\rho }_{blue}+1}}$$

(2)

$$DVI={\rho }_{nir}-{\rho }_{red}$$

(3)

$$SRI={\displaystyle \frac{{\rho }_{nir}}{{\rho }_{red}}}$$

(4)

where ${\rho }_{nir}$, ${\rho }_{red}$, and ${\rho }_{blue}$ are the SR values of near-infrared, red, and blue bands in Sentinel-2 satellite images.

All field survey plot datasets (50 plots) were divided into a 70% training dataset (35 plots) and a 30% verification dataset (15 plots). To assess the model performance, we used the root mean square error (RMSE) and coefficient of determination (R²) between the observed and predicted data.

3.1.2. Determination of AGBC and BGBC Correlation and BC Estimation

We applied two common mathematical models (linear and logarithmic) to determine the relationship between the AGBC and BGBC for mangroves in Guangdong Province. The model with the highest R² was used to apply the mangrove BGBC estimation based on the predicted mangrove AGBC data. And the three mangrove BC, AGBC, and BGBC maps were generated using Arcmap software.

3.2. Hotspot Analysis and Relationship with the Nature Reserve Distribution Data

The Getis-Ord $G_i^{\mathrm{*}}$ spatial statistic in the Arcmap software was applied to identify statistically significant spatial trends (degree of clustering of high or low mapped values) of BC in mangroves, creating clusters of mangroves with high/low vegetation carbon stock values, described as “hotspots”/“cold spots” [55,56]. The Getis-Ord $G_i^{\mathrm{*}}$ statistic is given by the formula:

$$G_i^{\mathrm{*}}={\displaystyle \frac{\sum _{j=1}^n{w}_{i,j}{x}_j-\overline{X}\sum _{j=1}^n{w}_{i,j}}{S\sqrt{\frac{\left[n\sum _{j=1}^n{w}_{i,j}^2-{\left(\sum _{j=1}^n{w}_{i,j}\right)}^2\right]}{n-1}}}}$$

where $\overline{X}={\displaystyle \frac{\sum _{j=1}^n{x}_j}{n}}$; $S=\sqrt{({\displaystyle \frac{\sum _{j=1}^n{x}_j^2}{n}}-{\left(\overline{X}\right)}^2)}$; $x_j$ represents the value for feature $j$; $w_{i,j}$ is the spatial weight between feature $i$ and $j$; and $n$ is the number of features.

We constructed $w_{i,j}$ using the fixed distance method, and the threshold distance was set as 1000 m. As Clarke (1993) [57] has shown that most mangrove propagules are within 1 km of the origin, and mangroves are generally distributed along Guangdong coastlines in patchy patterns, a 1000 m threshold is suitable to describe the autocorrelation of mangrove ecosystems in Guangdong. The False Discovery Rate correction was also applied to ensure the accuracy of hotspot identification.

The $G_i^{\mathrm{*}}$ statistic is a z-score, and we used a 99% confidence interval as the criterion and generated the hotspot zones of mangrove biomass carbon stocks. A spatial overlay analysis was conducted between the identified hotspots and the nature reserve distribution data of Guangdong Province. Further, this analysis quantified the areal extent and percentage of hotspots overlapping with protected areas, clarifying the spatial relationship between mangrove biomass carbon storage and the province’s nature reserve network [58].

4. Results

4.1. Field Survey

The mangrove biomass carbon stocks in Guangdong varied in different regions. Mangroves with the highest average AGBC were at Tongming (TM; 103.87 Mg C/ha) in Zhanjiang, followed by Shenzhen Bay (SZB; 91.73 Mg C/ha) in Shenzhen, Qi’ao Island (QAI; 82.60 Mg C/ha) in Zhuhai, Techeng Island (TCI; 79 Mg C/ha) east of Zhanjiang, Yingluo Harbor (TLH; 56.10 Mg C/ha) west of Zhanjiang, and Kaozhou Yang (KZY; 35.88 Mg C/ha) in Huizhou. Mangrove BGBC presented the highest value in YLH (38.69 Mg C/ha) and the lowest value in KZY (14.72 Mg C/ha).

With all field survey plots, the mangrove BGBC increased significantly (p < 0.001) with mangrove AGBC (Figure 3). The trend was more evident with the linear regression model (R² = 0.88) compared to the logarithmic regression model (R² = 0.35). Thus, we used the linear regression model to represent the relationship between AGBC and BGBC, and calculated mangrove BGBC using the mangrove AGBC map in Guangdong. The goodness of fit of the linear relationship between AGBC and BGBC was evaluated using the standard error of the estimate (SE); the SE of the linear model here was 66.07 Mg C/ha.

4.2. Biomass Carbon Stock Estimation

The mangrove AGB estimation model in this study shows better results (R² = 0.74, RMSE = 59.42 $\pm$ 12.47), compared with previous studies from Meng et al. (2022) [30] (R² = 0.63), Pham et al. (2018) [59] (R² = 0.596), Aslan et al. (2016) [60] (R² = 0.46), and Zhao et al. (2016) [61] (R² = 0.28~0.44). Our results indicate that integrating field data and Sentinel-2 satellite imagery yields reasonable estimates in terms of R² and RMSE for mangrove biomass prediction in Guangdong, China. To quantify model robustness further, we implemented a 10-fold cross-validation, and the random forest model yielded a cross-validated R² of 0.41 and an RMSE of 84.75, which were more conservative than the previous random-split results.

The BC map was displayed using seven numerical intervals, ranging from 32.31 to 158.94 Mg C/ha (Figure 4). The study estimated that Guangdong Province had a total mangrove biomass carbon storage of 1,209,305.68 Mg C (Figure 5), with a mean value of 80.56 Mg C/ha (

Table 3. Estimations of mangrove biomass carbon stocks for Guangdong Province.

Location	Mangrove Area (ha)	Mean AGBC (Mg C/ha)	Mean BGBC (Mg C/ha)	Mean BC (Mg C/ha)	Total BC (Mg C)	Contribution to Provincial Total BC
Zhanjiang	7009.77	62.29	17.48	79.77	685,190.41	56.66%
Maoming	402.17	65.11	19.04	84.15	41,284.54	3.41%
Yangjiang	1014.02	59.65	16.02	75.67	95,888.93	7.93%
Jiangmen	1514.62	60.21	16.33	76.54	139,592.09	11.54%
Zhuhai	708.08	71.46	22.55	94.01	71,339.16	5.90%
Zhongshan	148.04	66.02	19.54	85.56	23,953.75	1.98%
Guangzhou	385.23	62.8	17.76	80.56	41,641.56	3.44%
Dongguan	74.62	56.97	14.53	71.5	7317.58	0.61%
Shenzhen	229.49	66.18	19.63	85.81	24,609.89	2.04%
Huizhou	374.3	65.91	19.48	85.39	39,462.12	3.26%
Shanwei	65.38	71.29	22.46	93.75	7725.21	0.64%
Jieyang	2.81	60.54	16.51	77.05	319.79	0.03%
Shantou	263.8	68.71	21.03	89.74	29,390.50	2.43%
Chaozhou	15.23	66.48	19.8	86.28	1590.15	0.13%
Guangdong	12,207.56	62.80	17.76	80.56	1,209,305.68	100%

). The average AGBC and BGBC densities were 62.80 and 17.76 Mg C/ha, respectively. The mean AGBC was greatest in Zhuhai (71.46 Mg C/ha), followed by Shanwei (71.29 Mg C/ha). The highest mean BGBC was also observed in Zhuhai (22.55 Mg C/ha), followed by Shanwei (22.46 Mg C/ha). Mangrove BC was highest in Zhuhai (94.01 Mg C/ha) and lowest in Dongguan (71.50 Mg C/ha). Moreover, mangroves in Zhanjiang had the highest biomass carbon storage (685,190.41 Mg C), accounting for more than half of the total mangrove biomass carbon storage in Guangdong, whereas those in Jieyang were the lowest (319.79 Mg C).

4.3. Biomass Carbon Stocks Inside and Outside Nature Reserves

To analyze mangrove biomass carbon stocks inside and outside nature reserves, we conducted overlay analysis and statistical analysis using nature reserve data and the generated mangrove biomass carbon stock map in Guangdong. Mangroves in nature reserves demonstrated higher mean BC (83.03 Mg C/ha) compared with those outside nature reserves (77.99 Mg C/ha) (

Table 4. Estimations of mangrove BC inside and outside nature reserves.

Location	Mangrove Area (ha)	Mean AGBC (Mg C/ha)	Mean BGBC (Mg C/ha)	Mean BC (Mg C/ha)	BC (Mg C)	Contribution to Provincial Total BC
Inside nature reserves	6531.15	64.39	18.64	83.03	636,514.14	52.63%
Outside nature reserves	5676.41	61.14	16.84	77.99	572,791.54	47.37%

). Mangroves in nature reserves stored 636,514.14 Mg C of biomass carbon stock, whiles those outside nature reserves stored biomass carbon stock of 572,791.54 Mg C in Guangdong Province.

4.4. Mangrove Biomass Carbon Hotspots and the Protection Gaps

The Getis-Ord $G_i^{\mathrm{*}}$ spatial statistic results derived from the mangrove biomass carbon stock map are displayed in Figure 6. Using a confidence interval greater than 99% as the criterion, hotspots with high mangrove biomass carbon stock values are mainly distributed in Anpu Gang, Leizhou, and Mazhang in Zhanjiang, Shuidong Harbor, Yangjiang Harbor, Zhenhai Harbor, east of the Huangmao River, the Pearl River Estuary, Kaozhou Yang, Changsha Gang, and Yifengxi Port. The overlapping area between the mangrove biomass carbon stock hotspot areas and the nature reserves is 71.62 km², accounting for 51.13% of the total hotspot area and 1.34% of the total area of nature reserves in Guangdong.

To analyze the mangrove protection gaps, we obtained a map of mangrove hotspots outside nature reserves (Figure 7). In terms of mangrove biomass carbon stocks, the main protection gaps in Guangdong are distributed in Anpu Gang, the region south of Zhanjiang, Shuidong Harbor, the Pearl River Estuary, Kaozhou Yang, and Yifengxi Port.

5. Discussion

5.1. Mangrove Biomass Carbon in Guangdong

By integrating multi-spectral Sentinel-2 satellite imagery and mangrove field data, we mapped the province-scale mangrove biomass carbon in 2023 in Guangdong for the first time. The total mangrove biomass carbon in Guangdong was 1,209,305.68 Mg C, in line with the results from Su et al. (2024) [62] (1,094,465 Mg C). Su et al. (2024) [62] calculated mangrove carbon stocks, including biomass carbon stock and soil carbon stock, based on mangrove species distribution data in 2018 and published carbon density data. But Su et al. (2024) [62] assumed uniform tree heights and breast height diameters for each mangrove species across Guangdong, overlooking variations across different growth stages. Similarly, Li et al. (2023) [63] estimated the carbon storage of coastal wetlands in China, combing remote sensing land cover data and carbon density data. Li et al. (2023) [63] assumed the carbon density of mangrove wetlands in Guangdong was a constant value derived from the literature, and estimated the total carbon storage of mangrove wetlands (biomass carbon stock and soil carbon stock) in Guangdong to be 2.91 × 10⁶ Mg C. In comparison, our study first determined the correlation between AGBC and BGBC with field survey data acquired in 2023; then, we combined field data, satellite imagery, and a random forest model to predict a mangrove AGB map, obtained a BGBC map through regression; and finally, generated the mangrove biomass carbon stock map in Guangdong. The differences between our results and those from Su et al. (2024) [62] and Li et al. (2023) [63] are mainly due to the different datasets derived from different years, as well as the different calculation methods. Since mangroves of different ages and species exhibit distinct variations in spectral characteristics, our study can more effectively capture mangrove spatial heterogeneity by integrating field data with spectral features from satellite imagery using a machine learning method.

Meng et al. (2022) [30] estimated the total mangrove carbon stock in Hainan Island to be 703,181 Mg C (including the AGBC, BGBC, and mangrove soil carbon), smaller than the value in Guangdong here, mainly due to the larger mangrove distribution area in Guangdong. The mangrove carbon stock in North America was 386.39 × 10⁶ Mg C, considerably larger than the mangrove biomass carbon stock in Guangdong in this study. In comparison with terrestrial forests in Guangdong, the mangrove biomass carbon stock value in 2023 was comparable to the carbon storage value of economic forests in 1979 [64].

Mangrove biomass carbon stocks in Guangdong show pronounced spatial heterogeneity, with higher values primarily concentrated in western Guangdong and the Pearl River Estuary, while lower values mainly distributed in eastern Guangdong (Figure 5). Previous studies indicate that spatial differences in mangrove biomass carbon are often influenced by a combination of factors, including mangrove tree species [65], environmental conditions [31,65], anthropogenic activities [66,67,68], and hydrological changes [31]. For example, Wang et al. (2024) [69] proposed that mangrove species, water pH, salinity, dissolved oxygen, elevation, soil organic matter, soil total phosphorus, soil total nitrogen, and soil total potassium were important factors influencing the spatial patterns of mangrove AGB in Qinglan Harbor Mangrove Nature Reserve. Li et al. (2018) [65] argued that environmental factors, such as air temperature, humidity, fertility, tidal currents, and geomorphology, could lead to spatial variability of the mangrove carbon in Taiwan, China. Pirasteh et al. (2024) [31] pointed out that various factors, such as local geomorphological features, hydrological processes, geological conditions, erosion and sedimentation rates, climatic conditions, salinity, pollutants, and sea level rise, could contribute to the mangrove biomass spatial differences along the Persian Gulf coasts. The spatial heterogeneity of mangrove biomass carbon observed in Guangdong is likely a product of the factors discussed above. With more field measurement data collected, we plan to investigate the primary drivers of mangrove biomass carbon spatial patterns in Guangdong in the future.

5.2. The Linkage Between Mangrove Biomass Carbon Stocks and Nature Reserves

Using spatial statistics and remote sensing data, we identified the clustering of higher biomass carbon stocks outside nature reserves as mangrove protection gaps (Figure 7). The mangrove total BC, mean ABGC, mean BGBC, and mean BC inside nature reserves are much larger than those outside nature reserves in Guangdong, demonstrating the significant role of nature reserves in enhancing mangrove carbon storage. Mangroves within nature reserves contribute to 52.63% of the total mangrove BC in Guangdong Province, with Zhanjiang being the dominant contributor. In this study, we found that nature reserves in Guangdong still suffer from inadequate protection coverage for high-biomass-carbon mangroves, with hotspot analysis revealing the spatial mismatch between high-carbon-density zones and nature reserve boundaries. From the perspective of mangrove biomass carbon, protection gaps in Guangdong’s mangroves include areas such as Anpu Gang, the region south of Zhanjiang, Shuidong Harbor, the Pearl River Estuary, Kaozhou Yang, and Yifengxi Port. And 48.87% of the mangrove biomass carbon stock hotspot areas are distributed outside nature reserves in Guangdong. Therefore, it is crucial to strengthen mangrove carbon storage monitoring and implement adjustments to the boundaries of nature reserves, ensuring that more mangroves with high biomass carbon storage are included within protected areas. The mangrove hotspot maps allow managers to prioritize mangrove conservation in high-carbon-density areas, and target mangrove restoration or expansion in hotspots where carbon gains would be maximized in Guangdong Province. At the national level, the mangrove BC results provide spatially explicit data that could be integrated into China’s Dual Carbon goals, ensuring that mangrove blue carbon is factored into nature-based solutions. At the global level, our approach contributes to ongoing efforts to incorporate mangrove carbon dynamics into climate mitigation frameworks, supporting the Global Mangrove Alliance’s restoration targets.

As researchers noted that “expanding the area does not equate to enhancing the quality of mangroves” [15], the ecosystem functions and services are also significant [70]. Hence, it is imperative to optimize the spatial planning of nature reserves, incorporating ecological service assessment results, such as carbon storage values, and providing a more integrated plan for mangrove ecosystem protection [71].

5.3. Uncertainty and Future Improvement

In this study, we used random forest to estimate mangrove AGB using 13 features (B2, B3, B4, B5, B6, B7, B8, B11, NDVI, EVI, DVI, SRI, and one texture feature). We further compared random forest models using only subsets of the variables and found that the model including all variables achieved the highest accuracy. The vegetation indices here are among the most widely applied indices for mangrove and forest biophysical estimation [30,72]. While these variables provided satisfactory model performance, the feature space could be further enriched in the future. For example, integrating LiDAR-derived canopy structure metrics and synthetic aperture radar (SAR) backscatter features [73,74,75] may improve the robustness of the ABG inversion model across heterogeneous mangrove and coastal wetland environments. However, the differences in spatial resolution, acquisition geometry, and radiometric characteristics among these multi-source remote sensing datasets pose challenges for data fusion and joint application in biomass inversion. Addressing issues of scale mismatch, co-registration accuracy, and feature harmonization will therefore be essential for realizing the full potential of multi-sensor approaches to improve the accuracy and robustness of mangrove AGB estimation.

The available field survey datasets (n = 50 plots) limit the feasibility of more detailed uncertainty modeling. By collecting denser and more spatially representative field survey plots in the future, we plan to generate pixel-level mangrove BC uncertainty maps and investigate the main sources of uncertainty in BC estimates at the provincial scale in Guangdong. Since mangrove biomass carbon storage is highly heterogeneous across tree species, ages, and environmental conditions, we plan to further gather additional field survey data and multi-source remote sensing imagery and apply deep learning algorithms, in order to enhance the estimation of mangrove carbon stocks. This approach can ensure the generation of more accurate and refined carbon stock maps, thereby providing critical baseline data for the conservation and restoration of mangrove ecosystems.

Regional variation in wood density and carbon fraction is significant, driven by factors such as climate, altitude, and floristic composition [76,77], contributing to uncertainties in biomass estimation. And different allometric equations may yield divergent estimates for the same mangrove field survey plot. To reduce the biomass uncertainty, we consulted mangrove ecology experts, and employed equations (Table 2) suitable for mangrove biomass estimation in Guangdong Province [15]. The issue of sensitivity to equation choice and the uncertainty of plot-level mangrove carbon estimates deserves more in-depth investigation in the future.

We applied the Getis-Ord $G_i^{\mathrm{*}}$ spatial statistic to perform hotspot analysis, in which the spatial weight matrix was constructed using the fixed-distance method. The distance threshold is a key parameter influencing the identification of hot and cold spots. As the objective of this study was to detect the protection gaps at a broad spatial scale, we initially used a 1000 m threshold and then tested an 800 m threshold for comparison. The results showed that the spatial patterns of significant clusters remained consistent across the two thresholds, indicating the robustness of the observed spatial clustering patterns here.

This study focused on above- and belowground biomass carbon due to the availability of field measurements and remote sensing predictors. However, soil carbon represents the dominant carbon reservoir in mangrove ecosystems [78,79,80] and accounts for 49%–98% of carbon in mangroves [81,82]. It is significant and urgent to assess the spatial variability of mangrove soil carbon so as to fully understand carbon storage in mangrove ecosystems. Researchers have usually relied on the literature [13] or collected a very limited amount of in situ soil sampling data [30] to study the distribution of mangrove soil carbon, which are insufficient to reveal the spatial heterogeneity of soil carbon stocks. Researchers have noted that environmental properties such as salinity zonation, nitrogen to carbon ratio, as well as vegetation characteristics including forest type and tree diameter, have important effects on mangrove soil carbon [83]. Given the diversity of mangrove species and the numerous fragmented small mangrove patches across the entire coastline, it is challenging to conduct a comprehensive and accurate assessment of mangrove soil carbon in Guangdong. A comprehensive dataset integrating environmental and remote sensing data should be incorporated, using AI tools to improve the accuracy of mangrove soil carbon mapping and the completeness of mangrove carbon stock assessments in the future [84,85,86,87].

Furthermore, by integrating socio-economic and remote sensing data, we can simulate scenario-based responses of mangrove conservation policies and project the carbon storage of mangroves in 2030 and 2060 [88,89,90]. With the assessment results, we could further evaluate the contribution of mangrove protection and restoration initiatives to the realization of China’s Dual Carbon goals and provide insights to guide the improvement of future mangrove management strategies.

6. Conclusions

In this study, we estimate the mangrove biomass carbon stocks by combing remote sensing and field data and quantify the mangrove biomass carbon hotspots and protection gaps using the Getis-Ord $G_i^{\mathrm{*}}$ spatial statistic method. The main conclusions are as follows:

(1)

The spatial pattern of mangrove biomass carbon shows notable variations along the Guangdong coastlines.

(2)

The proposed mangrove AGB estimation model here demonstrates satisfactory results, confirming that the method integrating field survey data and Sentinel-2 satellite imagery can effectively and accurately estimate mangrove biomass carbon stocks.

(3)

Guangdong Province had a total mangrove biomass carbon storage of 1,209,305.68 Mg C, with Zhanjiang having the highest biomass carbon storage and accounting for more than half of the total mangrove biomass carbon storage in Guangdong. Zhuhai has the highest mean AGBC and BGBC values in mangroves, followed by Shanwei.

(4)

Mangroves in nature reserves demonstrated higher mean total carbon stock (83.03 Mg C/ha) compared with those outside nature reserves (77.99 Mg C/ha), demonstrating the significant role of nature reserves in enhancing mangrove carbon storage.

(5)

The overlapping area between the mangrove biomass carbon stock hotspot areas and the nature reserves is 71.62 km², accounting for 51.13% of the total hotspot area. In terms of mangrove biomass carbon stocks, the main protection gaps in Guangdong are distributed in Anpu Gang, the region south of Zhanjiang, Shuidong Harbor, the Pearl River Estuary, Kaozhou Yang, and Yifengxi Port.

Beyond these findings, our results provide forward-looking implications for conservation and policy. Mangrove hotspot maps and protection gaps can guide the refinement of nature reserve planning in Guangdong, inform targeted restoration and expansion programs, and contribute to China’s Dual Carbon goals by quantifying the role of mangroves in climate mitigation. At the global scale, this approach offers a transferable framework for blue carbon monitoring and management, while underscoring the need for future research to incorporate soil carbon pools, species- and age-specific biomass dynamics, and multi-sensor remote sensing to improve accuracy and completeness.