Spatial Differentiation of Mangrove Aboveground Biomass and Identification of Its Main Environmental Drivers in Qinglan Harbor Mangrove Nature Reserve

Abstract

In the Bamen Bay area of the Qinglan Harbor Mangrove Provincial Nature Reserve in Wenchang, Hainan Province, China, mangrove aboveground biomass (AGB) was estimated using high-resolution UAV ortho-imagery and UAV LiDAR data. The spatial distribution characteristics of AGB were studied using global Moran’s I index and hotspot analysis. Optimal geographic detectors and regression models were employed to analyze the relationship between AGB and key environmental factors. The results indicate that (1) the average AGB in the study area was 141.22 Mg/ha, with significant spatial variation. High AGB values were concentrated in the southwestern and northeastern regions, while low values were mainly found in the central and southeastern regions. (2) Plant species, water pH, soil total potassium, salinity, dissolved oxygen, elevation, soil organic matter, soil total phosphorus, and soil total nitrogen were identified as major factors influencing the spatial distribution of AGB. The interaction results indicate either bifactor enhancement or nonlinear enhancement, showing a significantly higher impact compared with single factors. (3) Comprehensive regression model results reveal that soil total nitrogen was the primary factor affecting AGB, followed by soil total potassium, with water pH having the least impact. Factors positively correlated with AGB promoted biomass growth, while elevation negatively affected AGB, inhibiting biomass accumulation. The findings provide critical insights that can guide targeted conservation efforts and management strategies aimed at enhancing mangrove ecosystem health and resilience, particularly by focusing on key areas identified for potential improvement and by addressing the complex interactions among environmental factors.

1. Introduction

Mangroves are considered one of the most reliable blue carbon ecosystems [1,2]. The biomass of mangrove plants is closely linked to their carbon sink function, reflecting not only the growth status of mangroves but also the strength of their blue carbon capabilities [3,4,5]. The mangrove carbon pool is primarily divided into aboveground live biomass (AGB), aboveground dead biomass or litter layer mass (LLM), belowground biomass (BGB), and soil organic carbon (SOC) [6]. This study focuses solely on analyzing the AGB of mangroves.

Although mangrove AGB has been estimated by various methods, such as the harvest method [7], allometric growth method [8,9,10], point structure model [11,12,13], outer hull models (OHMs) [14], and remote sensing inversion [15,16], each has its advantages and limitations. The harvest method was the most accurate but caused irreversible damage. The allometric growth method, based on the correlation between forest biomass and tree-measuring factors such as diameter at breast height (DBH), estimated the allometric growth equation between mangrove AGB and tree-measuring factors [10,17]. Due to its nondestructive nature, this method has been widely used; however, it is difficult to construct a universal allometry equation for various tree species in different regions due to species diversity, environmental complexity, and spatial differentiation of mangrove trees, resulting in significant differences in mangrove AGB estimation results [18,19]. With the advancement of remote sensing technology, optical sensor data, synthetic aperture radar (SAR) data, and LiDAR data have been increasingly applied in mangrove AGB estimation. Currently, there is a growing body of research utilizing multisource remote sensing, particularly LiDAR technology, to obtain mangrove physical parameters (such as height) for AGB estimation, and the accuracy of these estimations continues to improve [20,21,22,23]. The feasibility of using a remote sensing method to estimate mangrove AGB is illustrated in [24].

Analyzing the spatial distribution characteristics of mangrove AGB and its influencing factors is a crucial approach to understanding forest vegetation growth patterns and the evolution of forest ecosystems. Compared with inland forests, the spatial distribution characteristics of mangroves exhibit a distinct zonation pattern with mangrove plants aligned perpendicularly to the coastline [25]. Under different successional sequences, mangrove AGB also shows a gradient variation from the tidal boundary inland. The growth of forest trees depended on environmental factors and the characteristics of the plants [26]. Mangrove AGB not only reflected mangrove characteristics but also predicted the environmental conditions of mangrove growth. On a large scale, mangrove AGB was mainly affected by natural factors such as temperature, solar radiation, precipitation, storm surges, sea level rise, and changes in coastal freshwater injection [27,28]. On a smaller scale, hydrology, soil, and topography significantly affected the accumulation of mangrove AGB [29,30,31,32]. Soil carbon, nitrogen, and phosphorus contents reflect the utilization efficiency and effectiveness of soil nutrients in the plant growth process [33,34]; elevation and tidal energy indicate the degree of seawater immersion [35]; and hydrological indexes such as salinity and dissolved oxygen affect the AGB size of mangroves by influencing mangrove species [36]. In addition, mangrove AGB is also affected by mangrove plant characteristics such as tree species, tree age, interspecific competition, and human factors [37,38,39].

Most current studies on factors influencing mangrove AGB focus on qualitative or semiquantitative analysis [40,41,42]. Quantitative research is mainly concentrated on single environmental factors [43], and there is a lack of coupling studies between mangrove AGB and various environmental factors. This paper aims to address the following questions: (1) How can mangrove AGB be accurately estimated? (2) How can the spatial differentiation characteristics of mangrove AGB be identified? (3) Is there a coupling relationship between environmental factors, and which factors predominantly influence the spatial differentiation of mangrove AGB? This research aims to provide decision support for the management, conservation, and restoration of mangrove ecosystems.

2. Study Site

The Qinglan Mangrove Provincial Nature Reserve in Wenchang, Hainan, China is among China’s significant mangrove wetlands, boasting some of the country’s highest, oldest, and most extensive mangrove communities. The reserve comprises three distinct areas, with the core area located in Bamwan, the principal distribution zone (Figure 1). Positioned between 110°42′ E and 110°57′ E longitude and 19°15′ N to 19°43′ N latitude, the study area lies on the northern fringe of the tropics, characterized by a tropical oceanic monsoon climate. The average annual temperature is 23.9 °C, with the coldest month averaging 14.8 °C and an annual precipitation of approximately 1650 mm. Situated within the western branch of an irregular, inverted eight-shaped lagoon, the area features rich silt, minimal wind and wave action, and an average tidal range of about 0.9 m, creating favorable conditions for mangrove growth and reproduction.

3. Materials and Methods

3.1. Data Collection and Processing

This study focused on the carbon estimation of mangrove AGB. Since soil carbon storage data of mangroves could not be directly obtained through remote sensing methods, we were unable to estimate underground carbon storage. Typically, soil carbon estimation required field sampling and laboratory analysis, which was beyond the scope of this study. Consequently, the results of this paper primarily reflected the carbon storage characteristics of aboveground biomass.

There were six types of data in this study:

(1)

Soil and water environment data. Soil factors included organic matter, total nitrogen, total phosphorus, total potassium, and particle size, while water environment factors included pH, salinity, and dissolved oxygen. Field sampling of 67 soil sites and 79 water sites was completed in the Bamwan Area from 2021 to 2022 (Figure 1). The pH, salinity, and dissolved oxygen of water were determined on-site, and the soil organic matter, total nitrogen, total phosphorus, total potassium, and particle size were measured using the potassium dichromate volumetric method, Kjeldahl method, alkali melt-molybdenum-antimony resistance spectrophotometry, atomic absorption spectrophotometry, and laser particle size analyzer, respectively.

(2)

Tidal data. Based on the average tidal data of Qinglan Port from 2000 to 2019, the neap low water line, spring low water line, neap high water line, and spring high water line were determined. The intertidal zones in the study area were then extracted: low water zone (−0.15 m to 0.55 m), mid-tide zone (0.55 m to 1.33 m), and high water zone (1.33 m to 2.12 m).

(3)

Mangrove plant species. The mangrove plant species in the study area were primarily divided into five species and one genera: B. sexangula (BS), E. agallocha (EA), R. apiculata (RA), H. tiliaceus (HT), and L. racemosa (LR), Sonneratia spp. (SS, including S. alba, S. apetala, and S. ovata). UAV ortho imagery with 0.06 m spatial resolution was used, and supervised classification was performed using the support vector machine, minimum distance method, and maximum likelihood method. The results from the maximum likelihood method that had the highest classification accuracy and Kappa coefficient were selected. These results were then combined with field investigation data, UAV aerial photography, and historical data to manually refine the supervised classification results, resulting in the spatial distribution map of mangrove plants in the study area (Figure 2), with a modified classification accuracy of about 87% (

Table 1. Evaluation error matrix of mangrove plant classification results.

Species	Classification Result
	BS	EA	HT	LR	RA	SS	Total	Producer’s Accuracy
BS	162	1	5	9	2	6	185	87.57%
EA	2	103	3	5	6	4	123	83.74%
HT	1	2	92	2	4	5	106	86.79%
LR	9	5	6	225	4	6	255	88.24%
RA	3	6	4	2	92	1	108	85.19%
SS	5	4	3	2	1	105	120	87.50%
Total	182	121	113	245	109	127	897
User’s Accuracy	89.01%	85.12%	81.42%	91.84%	84.40%	82.68%
Overall Accuracy: 86.85%; Cohen’s Kappa coefficient: 0.84

).

(4)

UAV Orthophoto Image Acquisition and Preprocessing. The orthophoto images were acquired in the field using a DJI Phantom 4 quadcopter UAV (SZ DJI Technology Co., Ltd., Shenzhen, Guangdong Province, China) equipped with a camera system. The camera system included six 1/2.9-inch CMOS (Complementary Metal-Oxide-Semiconductor) sensors, with one color sensor for visible light imaging and five monochromatic sensors for multispectral imaging (2.08 million effective pixels). The monochromatic sensors covered five multispectral imaging bands: blue (450 nm), green (560 nm), red (650 nm), red-edge (730 nm), and near-infrared (840 nm). To ensure the quality of the images and to successfully capture comprehensive mangrove vegetation data, aerial photography was conducted during low tide and under sufficient sunlight. The DJI DJIGO software system (DJI Go v3.1.72(799_go3_official)) was used for flight path planning, setting the sensor’s shooting angle to 90° vertical to the ground, with a 70% lateral overlap and a 60% longitudinal overlap, flying at a speed of 3 m/s and an altitude of 100 m. A 90% forward overlap and 70% side overlap were used. Before each flight mission, a whiteboard was used for radiometric calibration of the sensors, resulting in images with a resolution of 0.1 m. The images were processed by loading the original UAV POS (position and orientation system) data and image data using free network matching; feature point coordinates were extracted from the point cloud; the extracted feature points were imported and underwent bundle adjustment in the software; and after satisfactory adjustment, DSM/DEM production was carried out, and the final orthomosaic image was produced.

(5)

UAV-LiDAR Data Acquisition and Preprocessing. The LiDAR data in the study area were collected using a Hornet quadcopter UAV equipped with a Huace AS900 multiplatform LiDAR scanning system (sensor: RIEGL-VUX-1UVA) (Shanghai Huace Navigation Technology Ltd., Shanghai, China). The AS-900HL LiDAR system’s multiple return wave technology allowed it to penetrate vegetation, quickly obtaining high-precision laser point clouds under complex terrain conditions. The point cloud data collected had a density of ≥100 points/m², with a median error in plane accuracy of ≤0.1 m and a median error in elevation accuracy of ≤0.15 m. Before the UAV flight, tidal data for the study area were collected, and actual data collection was conducted during low tide. The flight was conducted at an altitude of 150 m, at a speed of 7 m/s, with a 70% lateral overlap and an 80% forward overlap, following a snake-shaped flight path.

The collected data were preprocessed using CoPre2 software(CoPre-2.5.1-20221017), which included POS and point cloud solution steps. The CoPre2 software utilized a self-developed POS calculation module for POST solution. The original high-density point cloud, original UAV POS data, and base station data were acquired through UAV flights. After the POS solution, point cloud solution was performed, and the output was in Las point cloud format. The Quick Terrain Modeler software (v8.0.7) was then used to check for any noticeable stratification in the point cloud data. If stratification was observed, the data were reprocessed or adjusted during the data processing phase. The final point cloud adjustment accuracy was controlled within 3 cm (both vertically and horizontally). The preprocessed point cloud was further denoised using LiDAR360 5.2.2 software, removing noise points between the UAV flight altitude and the mangrove canopy. Then, based on an improved progressive triangulated irregular network (TIN) filtering algorithm, the point clouds were classified into ground and nonground points. The automatically extracted ground points were further manually checked and edited to obtain the final ground points, which were used to generate the digital elevation model (DEM) of the study area. Nonground points were used to generate the digital surface model (DSM). Finally, the point clouds were clipped to the mangrove area, resulting in a pure mangrove point cloud for the study area.

(6)

Biomass data. The AGB and the density of each quadrat were estimated based on 116 quadrat data collected from June to August in 2022 and 2023 and known mangrove AGB allometry equations. The data were split into a 7:3 ratio for the training set and validation set. Using a two-order model, mangrove height and canopy were first extracted by LiDAR. Then, the tree height and canopy of the full coverage area of the study were retrieved using spectral characteristic variables, polarization characteristic variables, and texture information from Sentinel-2 and Landsat. Finally, tree height, canopy, and image spectral characteristics were combined. The inverse model of mangrove AGB was constructed using a random forest regression algorithm. The optimal inversion results have an R² of 0.68 and an RMSE of 43.375, indicating good accuracy (Figure 3). Applying the optimal model to the entire study area, the mangrove AGB map for the whole area was obtained.

3.2. Research Method

3.2.1. Spatial Statistical Analysis

In this study, the application of spatial autocorrelation techniques, including Moran’s I and Getis-Ord Gi* statistics, is critical for a comprehensive analysis of the spatial distribution patterns and clustering of mangrove AGB. These methods allow us to assess the extent to which AGB is spatially correlated and identify significant hotspots and coldspots within the study area.

(1)

Moran’s I: The global Moran’s I coefficient was used to describe the overall correlation degree of regions and to reflect the overall spatial agglomeration characteristics. The calculation formula is as follows [44]:

$$\mathrm{I}={\displaystyle \frac{\mathrm{n}{\displaystyle {\sum }_{i=1}^n}{\displaystyle {\sum }_{j=1}^n}{w}_{ij}(x_i-\overline{x})(x_j-\overline{x})}{{\displaystyle {\sum }_{i=1}^n}{\displaystyle {\sum }_{j=1}^n}{w}_{ij}{\displaystyle {\sum }_{i=1}^n}{(x_i-\overline{x})}^2}}$$

(1)

where I is the global autocorrelation index; $x_i$ and $x_j$ are the index values of attribute unit and are the mean values; and $w_{ij}$ is a spatial weight matrix, which reveals the spatial differentiation of mangrove AGB. Moran’s I ∈ [−1, 1], where I < 0 indicates that there is a spatial negative correlation between the observed values, and I > 0 indicates a spatial positive correlation.

(2)

Getis-Ord Gi*: The ArcGIS Getis-ORD Gi* analysis tool was used to identify the hotspots and coldspots of mangrove AGB in the study area. The hotspots or coldspots represent the spatial agglomeration of high or low AGB values. The calculation formula is as follows [45]:

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

(2)

where G_i^∗ denotes the Z value (z-score). A positive Z value indicates a hotspot (high-value aggregation), with larger values signifying stronger hotspots. Conversely, a negative Z value indicates a coldspot (low-value aggregation), with smaller values indicating stronger coldspots.

3.2.2. Geographic Detector

As a statistical tool for detecting spatial differentiation and revealing the driving forces behind it, the geographic detector can analyze the influence of single factors and the interaction of two factors on dependent variables [46,47,48]. In this study, we utilized the GD package optimal geographic detector in R to grade and detect various factors. The optimal discretization method for independent variable data was automatically selected and implemented [49]. We measured the explanatory power (q value) of each factor on mangrove AGB and evaluated the explanatory power (q value) of the interaction of two factors on mangrove AGB. The q value is calculated as follows:

$$q=1-{\displaystyle \frac{{\displaystyle {\sum }_{h=1}^L}{N}_h{\delta }_h^2}{N{\delta }^2}}$$

(3)

where q represents the explanatory power of each factor on mangrove AGB; h denotes the classification number of factors; N and N_h signify the number of units in the population and in class h, respectively; and ${\delta }^2$ and ${\delta }_h^2$ are the variance of the biomass of the population and class h, respectively. The q value ranges from 0 to 1, where a large q indicates a stronger explanatory power of the factor for spatial differentiation.

By comparing the explanatory power of individual factors with the explanatory power of their interactions, we can explore the degree to which mangrove AGB is influenced when two different factors interact. There are five types of interactions between factors (

Table 2. Type of interaction between factors.

Criterion	Interaction
$q\left(X_1\bigcap X_2\right)<M\mathrm{i}\mathrm{n}[q\left(X_1\right),q(X_2\left)\right]$	Nonlinear attenuation
$M\mathrm{i}\mathrm{n}[q\left(X_1\right),q(X_2\left)\right]<q\left(X_1\bigcap X_2\right)<M\mathrm{a}\mathrm{x}[q\left(X_1\right),q(X_2\left)\right]$	Single-factor nonlinear enhancement
$q\left(X_1\bigcap X_2\right)>M\mathrm{a}\mathrm{x}[q\left(X_1\right),q(X_2\left)\right]$	Two-factor enhancement
$q\left(X_1\bigcap X_2\right)=q\left(X_1\right)+q\left(X_2\right)$	Independent
$q\left(X_1\bigcap X_2\right)>q\left(X_1\right),q\left(X_2\right)$	Nonlinear enhancement

) [48].

3.2.3. Regression Analysis

In this study, stepwise regression and OLS regression are employed to identify the key environmental factors influencing mangrove AGB and to select variables for subsequent geographic weighted regression (GWR) analysis. The primary role of GWR is to further explore how these factors affect the spatial distribution of AGB when spatial location is taken into account.

According to the results of the correlation analysis, independent variables with a significance level lower than 0.05 were eliminated. A stepwise regression model was then constructed to determine the significance of the preliminary screening variables in explaining the influencing factors of mangrove AGB. The regression model progressively incorporated the variable with the largest R², continuing this process until all independent variables were evaluated and the calculation was complete [50].

Ordinary least squares (OLS), the simplest regression model, is often used to estimate parameters in linear regression models. OLS fits the best function to the data by minimizing the sum of squared residuals. It has a closed-form solution, making it simple, efficient, and free from the need for iterative calculations. Its calculation formula is as follows [51]:

$$f\left(x\right)={\alpha }_1{\phi }_1\left(x\right)+{\alpha }_2{\phi }_2\left(x\right)+\cdots +{\alpha }_m{\phi }_m\left(x\right)$$

(4)

where $f\left(x_i\right)$ is the explanatory variable function, representing mangrove AGB in this paper. $\phi k\left(x\right)$ is a matrix of linealy independent functions representing the influencing factors of mangrove AGB. ${\alpha }_k$ is the vector of undetermined coefficients $(\mathrm{k}=\mathrm{1,2},\cdots ,m,m<n)$. The fitting criterion is $y_i(i=1,2,\cdots ,n)$ to $f\left(x_i\right)$ minimize the sum of squared residuals.

Geographically weighted regression (GWR) is a spatial statistical analysis method that extends ordinary least squares (OLS). GWR not only accounts for the spatial autocorrelation of data but also effectively handles nonstationarity and spatial heterogeneity, thereby enhancing model fit and prediction accuracy in regression models. It provides comprehensive spatial insights to address ecological spatial heterogeneity issues [52]. Its formula is as follows [53]:

$$y_i=\sum _r{x}_{ir}{\beta }_r(u_i,v_i)+{\epsilon }_i$$

(5)

where $y_i$ represents the mangroves AGB in the $i$th grid, $x_{ir}$ represents the $r$th factor in the $i$th grid, ${\beta }_r$ represents the regression coefficient of the $r$th independent variable in the $i$th grid, $\left(u_i,v_i\right)$ represents the centroid coordinate of the $i$th grid, and ${\epsilon }_i$ represents the random error term.

3.3. Technical Overview

Focusing on the spatial differentiation coupling between mangrove AGB and major environmental factors in the study area, this paper discussed the measurement, characterization, and identification of the main driving factors influencing this interaction relationship. The specific technical approach of this study is illustrated in Figure 4.

4. Results

4.1. Descriptive Statistical Analysis of Mangrove AGB and Various Environmental Factors

In this study, we selected a 10 m × 10 m grid as our research unit, generating a total of 22,288 grids within the study area. The mangrove AGB in individual grids ranged from 48 to 320 Mg/hm², with an average of 141.22 Mg/hm². Mangrove AGB was categorized into five grades: <100, 100–150, 150–200, 200–250, and >250 Mg/hm², denoted as Q1, Q2, Q3, Q4, and Q5, respectively. These categories were then analyzed based on their distribution across tidal zones. According to the biomass distribution map (Figure 5), mangroves in the study area predominantly occupied the low-tide and mid-tide zones, with fewer occurrences in the high-tide zones. The highest AGB values were observed in the 150–200 Mg/hm² range, followed by 100–150 Mg/hm² and <100 Mg/hm². Analyzing by tidal zone, mangrove biomass frequencies in the low-tide zone were ordered Q1 > Q2 > Q3 > Q4 > Q5, while in the midtide zone, the order was Q1 > Q3 > Q2 > Q4 > Q5. Frequencies of Q1, Q2, and Q3 were similar and significantly higher than those of Q4 and Q5.

Based on previous studies on mangrove AGB influencing factors, and considering the study area’s characteristics and data collected, we selected 10 factors: plant species, soil organic matter, soil total nitrogen, soil total phosphorus, soil total potassium, soil particle size, water pH, water salinity, dissolved oxygen, and elevation. These factors were chosen from the perspectives of plant biology, soil composition, hydrology, and topography to explore the spatial differentiation of mangrove AGB. Upon testing, all original data were found to adhere to a normal distribution (

Table 3. Descriptive statistics and correlation tests for explanatory variables.

Independent Variable	Abbreviation	Minimum Value	Maximum Value	Mean	Standard Deviation	Skewness	Kurtosis	Correlation Coefficient
Elevation	DEM	−0.28	3.12	0.93	0.34	−0.68	0.82	−0.229 **
Salinity	SA	9.40	18.26	12.21	2.01	0.88	−0.16	−0.001
Water pH	PH	7.02	7.68	7.32	0.13	0.58	0.59	−0.150 **
Dissolved oxygen	DO	5.09	7.07	6.22	0.30	0.74	−0.29	−0.271 **
Soil total nitrogen	TN	1.62	2.91	2.25	0.30	0.09	−0.87	−0.230 **
Soil total potassium	TK	6.28	8.82	7.94	0.52	−0.27	−0.81	0.322 **
Soil total phosphorus	SP	4.25	26.78	18.37	5.00	−0.68	−0.35	−0.151 **
Soil particle size	SIZE	0.04	0.13	0.07	0.02	0.79	−0.65	−0.150 **
Soil organic matter	OM	41.28	90.91	66.61	10.44	0.18	−0.03	0.138 **

** indicates significance at the 0.01 level.

). For spatial interpolation, the simple Kriging method was applied to eight factors, excluding plant species and elevation. Using the spatial connection tool in ArcGIS 10.8, values were assigned to the nearest mangrove AGB and corresponding grid for each factor attribute, producing maps illustrated in Figure 6.

The descriptive statistics and correlation analysis of environmental factor data are presented in Table 3. The data were found to adhere to a normal distribution based on their kurtosis and skewness values. The correlation coefficients represented the results of Spearman’s correlation tests between each explanatory variable and mangrove AGB.

Significantly, all variables except salinity were found to be correlated with mangrove AGB. Soil total potassium (TK) was positively correlated with soil organic matter. Conversely, elevation, pH, dissolved oxygen, soil total nitrogen, soil total phosphorus, and soil particle size were negatively correlated with mangrove AGB.

4.2. Spatial Distribution Characteristics of Mangrove AGB

Upon conducting grid-based statistics of mangrove AGB in 10 m × 10 m cells and confirming its normal distribution using QQ plots, spatial autocorrelation and hotspot analysis were employed to explore the spatial distribution characteristics of mangrove AGB.

4.2.1. AGB Spatial Autocorrelation

The study area exhibited significant spatial autocorrelation of mangrove AGB, as evidenced by a global Moran’s I index of 0.749, a Z value of 224.2944, and a significant p-value of 0.001. Moran’s I scatter plot (Figure 7) displays that grid sites are predominantly clustered in the first and third quadrants, representing high–high and low–low-value clusters, respectively. Conversely, the occurrence of high–low-value clusters was relatively scarce in the second and fourth quadrants. Overall, mangrove AGB in the study area displayed pronounced spatial positive correlation characteristics.

4.2.2. AGB Hotspot Analysis

Based on global spatial autocorrelation analysis, the hotspot analysis of mangrove AGB spatial distribution in the study area (Figure 7) revealed that hotspots were predominantly clustered in the southwest and northeast, while coldspots were concentrated in the central and southeastern regions. Specifically, there were 9934 hotspots and 8590 coldspots identified across the study area. Hotspots and coldspots with 99% confidence levels constituted 88.87% and 90.55% of the total respective spots, encompassing 44.57% and 38.54% of all grids in the study area. This underscored strong spatial clustering of mangrove AGB in both high and low values, particularly in low values. Both hot- and coldspots exhibited a gradual decrease and dispersion from the center to the periphery.

4.3. Coupling Analysis of Spatial Differentiation of AGB and Influence Factors

To delve deeper into the factors influencing mangrove AGB spatial differentiation, geographic detectors were employed to assess the explanatory power (q) of each factor, as detailed in

Table 4. Single-factor-driven detection of mangrove AGB. CLA stands for mangrove plant species.

Factor	q	Factor	q
CLA	0.2547 **	DEM	0.1298 **
pH	0.2020 **	OM	0.1274 **
TK	0.1869 **	SP	0.1122 **
SA	0.1719 **	TN	0.1076 **
DO	0.1551 **	SIZE	0.0828 **

** indicates significance at the 0.01 level.

. All 10 factors demonstrated significant influence with p-values below 0.01. Leading the pack were plant species (q = 0.2547), water pH (q = 0.2020), and soil total potassium (q = 0.1869), which exhibited the highest explanatory power. Conversely, soil total phosphorus (q = 0.1122), soil total nitrogen (q = 0.1076), and soil particle size (q = 0.0828) had the least impact on mangrove AGB spatial differentiation. Plant species, water pH, soil total potassium, salinity, and dissolved oxygen each yielded q values exceeding 0.15, while elevation, soil organic matter, soil total phosphorus, and soil total nitrogen exceeded 0.1. These findings underscore the substantial influence of these factors on mangrove AGB spatial patterns while highlighting the comparatively minor impact of soil particle size in the study area.

The interaction of 10 influencing factors was investigated to understand the explanatory power of the interaction of each factor on the spatial differentiation of mangrove AGB. As can be seen from Figure 8, the q value after the interaction of all factors is significantly higher than that of the single factor independent effect, and its explanatory power is enhanced. The explanatory power of SIZE∩TK, SA∩TN, SA∩DEM, DO∩SP, DEM∩SIZE, OM∩SP, OM∩TN, TN∩SIZE is greater than the sum of the two factors. It shows nonlinear enhancement. The enhancement effect of SA∩CLA is the largest, and the q value reaches 0.3579. The explanatory power of TP∩SIZE is the smallest, and the q value is 0.1971. The explanatory power of the interaction between plant species and each factor is greater than 0.25, and the q value of the interaction between the other two factors is greater than 0.2. In general, the two-factor interaction is mainly divided into two types: two-factor enhancement and nonlinear enhancement. The explanatory power of two-factor interaction on mangrove AGB spatial differentiation is significantly improved, which indicates that the mangrove AGB spatial differentiation is driven by the synergistic effect of multiple factors.

4.4. Screening of Driving Factors for Mangrove AGB Spatial Differentiation

Based on the stepwise linear regression model using species SS as the reference term for plant classification (

Table 5. Stepwise linear regression results.

Model	Unstandardized Coefficients		Standardized Coefficients	R-Squared	t	Sig.	Collinearity Statistics
	B	Standard Error					Tolerance	VIF
Constant	195.606	4.980		0.330	39.278	0.000
BS	28.821	1.217	0.246		23.675	0.000	0.296	30.380
pH	−298.565	7.918	−0.317		−37.707	0.000	0.452	20.213
DO	−249.863	12.277	−0.184		−20.352	0.000	0.390	20.564
TN	−41.621	6.033	−0.051		−6.899	0.000	0.588	10.699
LR	−27.594	1.493	−0.147		−18.478	0.000	0.502	10.991
DEM	−4.612	3.933	−0.008		−1.173	0.041	0.629	10.589
TK	141.802	9.044	0.142		15.679	0.000	0.389	20.572
OM	59.991	5.616	0.065		10.682	0.000	0.848	10.180
RA	4.020	1.391	0.024		2.890	0.004	0.475	20.105
SA	43.404	7.102	0.055		6.112	0.000	0.391	20.555
EA	−14.737	1.401	−0.103		−10.521	0.000	0.331	30.026
HT	−15.424	1.514	−0.087		−10.190	0.000	0.435	20.297

), the key explanatory variables identified were water pH, dissolved oxygen (DO), total nitrogen (TN), total potassium (TK), organic matter (OM), salinity (SA), digital elevation model (DEM), and five additional plant types. According to Table 4, water pH emerges as the most influential factor affecting mangrove AGB, indicating that lower pH levels correlate with higher biomass within the appropriate range. Consistently, the effects of DO, TK, EA, and LR on AGB demonstrate that higher values within reasonable ranges correspond to larger AGB.

The explanatory variables mentioned were used to construct an OLS regression model, and the results are presented in

Table 6. OLS estimation result.

Variable	Regression Coefficient	Standard Error	p	Variable	Regression Coefficient	Standard Error	p
BS	28.930	1.227	0.000	TK	32.138	2.869	0.000
pH	−85.948	2.567	0.000	OM	15.486	2.010	0.000
DO	−61.333	3.375	0.000	RA	4.400	1.384	0.001
TN	−13.423	1.788	0.000	SA	23.219	3.342	0.000
LR	−27.640	1.509	0.000	EA	−14.793	1.423	0.000
DEM	−8.790	3.939	0.025	HT	−15.168	1.529	0.000
R2	0.337
F	753
K value	0.000 *

* indicates significance at the 0.1 level.

. The model’s R² increased from 0.330 to 0.337. Additionally, both soil factors and mangrove plant species reached statistical significance (p < 0.01), indicating their significant effects on AGB. Specifically, water pH, dissolved oxygen, soil total nitrogen, and elevation exhibited negative correlations with AGB, while salinity, soil total potassium, and soil organic matter showed positive correlations with AGB. Moreover, various mangrove plant species were positively correlated with AGB. The OLS model’s K value is notably significant, suggesting pronounced driving factors in certain regions but not in others, highlighting spatial heterogeneity in the study area.

According to the results of step linear regression and OLS model, the water pH, dissolved oxygen, soil total nitrogen, soil total potassium, soil organic matter, salinity, elevation, and the mangrove plant species were treated, and the factors were screened according to the tolerance (T), variance expansion factor (VIF), and the analysis results of the geographical detector. The six explanatory variables were water pH, dissolved oxygen, soil total nitrogen, soil total potassium, soil organic matter, and elevation. These factors were further analyzed using the GWR model. The goodness of fit (R²) of GWR model analysis results is 0.592, indicating that the selected factors have a strong comprehensive interpretation of mangrove AGB spatial distribution characteristics. Compared with the OLS model, GWR model R² increased from 0.337 to 0.592, and in the GWR model calculation results, each grid has a specific regression coefficient, and the minimum, maximum, median, and average of the regression coefficients of the six influence factors can be obtained by analyzing the regression coefficients of each influence factor (

Table 7. GWR impact factor regression coefficient statistics.

Factor	Regression Coefficient
	Minimum Value	Maximum Value	Median	Mean
TK	−0.391	0.609	0.218	0.219
TN	−0.084	0.916	0.832	0.832
OM	−0.482	0.518	0.037	0.038
pH	−0.494	0.506	0.011	0.009
DO	−0.466	0.534	0.067	0.065
DEM	−0.543	0.457	−0.083	−0.085

).

As illustrated in Table 7, each influence factor had both positive and negative effects on mangrove AGB in the grid. From the average value of the absolute GWR regression coefficient, the influence intensity of each factor on AGB is ranked from high to low as follows: soil total nitrogen (TN) > soil total potassium (TK) > elevation (DEM) > dissolved oxygen (DO) > organic matter (OM) > water pH. From the spatial distribution of regression coefficients (Figure 9), TK and OM were found to positively influence AGB. Areas with higher OM content in the surface soil generally exhibited larger AGB [54], as indicated in previous studies correlating higher soluble salt ions and TK content with increased biomass and biological diversity [55]. Spatially, TK showed a negative correlation with AGB in the low tide areas in the northwest and southeast of the study area but a positive correlation in the middle tide zone. Meanwhile, the positive correlation between OM and AGB decreased from low tide to inland areas, primarily concentrated in the northwest of the study area, reflecting the potassium demand and adaptability of mangrove plants such as HT and EA.

Environmental factors such as pH, TK, DEM, and DO exhibited negative correlations with AGB. Elevation influences AGB through its impact on tidal levels. Mangrove habitats on tidal flats experience varying depths of tidal flooding, flood duration, and wave intensity, resulting in differing AGB levels. Moreover, since soil TN is typically highest in the 20–40 cm layer rather than near the surface (0–20 cm), and AGB primarily occurs at the surface, TN exerts a negative influence [56]. Spatially, soil TN concentrations are highest in the middle tidal zone, where dense mangrove forests trap litter, facilitating decomposition and nitrogen release. Additionally, higher soil clay content in this zone aids nutrient accumulation. Soil TN negatively correlates with AGB at the northwest and southeast edges of the study area, with soil TN content decreasing from the low tide zone to the high tide zone. Previous research has noted that DO indirectly affects mangrove plant niches primarily through its impact on surrounding animal diversity, with minimal direct impact on AGB [57,58]. Total sulfur elements primarily influence mangrove water pH. AGB distribution in the study area ranks middle tide zone > low tide zone > high tide zone, with mangrove water pH negatively correlated with AGB.

5. Discussion

5.1. Comparative Analysis of AGB Distribution Pattens and Drivers

The spatial distribution patterns of mangrove AGB are often studied at national or island scales using remote sensing techniques [59,60]. However, due to the limitations of spatial resolution in remote sensing imagery, finer-scale studies focusing on specific nature reserves are essential for accurately identifying precise locations that require targeted conservation efforts [61,62]. Analyzing the spatial characteristics of mangrove AGB in the study area revealed distinct patterns of local clustering. It has important reference value for ecological restoration projects and the location of key restoration areas. The global Moran’s I index indicated a significant positive spatial correlation, demonstrating both “high–high” and “low–low” spatial distribution characteristics of mangrove AGB. Hotspot analysis found that hotspots were mainly distributed in the southwest and northeast, with coldspots concentrated in the central and southeastern regions. The concentration of both hot- and coldspots gradually decreased from the center to the periphery. Coldspot areas exhibited lower mangrove AGB and poorer mangrove plant growth, highlighting key areas for potential improvement of mangrove quality. These findings highlight key areas where targeted management efforts could be most effective. For instance, enhancing conservation practices in the “high–high” hotspots could help maintain or improve the high-quality mangrove areas, potentially increasing their resilience and carbon sequestration capacity. Conversely, areas identified as “low–low” coldspots may benefit from restoration initiatives to boost their AGB and overall ecological function.

For similar habitats globally, soil carbon storage was significantly higher where vegetation biomass and stand density were high [63,64]. In this study area, soil total nitrogen was identified as the most important driving factor, whereas in other regions, species diversity, annual average precipitation, or intertidal zonation played a more significant role [65,66,67]. This difference was closely related to climatic conditions, soil properties, vegetation types, and human management practices within the region [68,69]. Therefore, although the spatial differentiation of mangrove AGB across the country was influenced by similar environmental factors, the relative importance and mechanisms of these factors varied significantly across habitats. Small-scale studies of specific nature reserves helped to more precisely identify the key environmental factors that had the greatest impact on the carbon sink capacity of mangroves, thereby providing the basis for more targeted conservation and management measures.

5.2. The Main Driving Factors of Mangrove AGB Spatial Heterogeneity

This study identified several primary factors influencing the spatial differentiation of mangrove AGB, including CLA, pH, TK, SA, DO, DEM, OM, TP, and TK. The interaction effects among these factors significantly enhanced their explanatory power, revealing both synergistic and nonlinear relationships. This finding underscores the importance of a multifaceted approach to mangrove management. For practical application, conservation strategies should consider the interplay between these factors. For example, managing soil conditions (such as pH and TK) and water quality (such as DO) can collectively influence AGB. Additionally, integrated monitoring systems that assess multiple environmental variables simultaneously could provide more accurate data for adaptive management.

The GWR analysis revealed that TN was the predominant driving factor influencing mangrove AGB in the study area, with a positive correlation indicating that higher TN levels corresponded to greater mangrove AGB. This finding aligned with previous studies by Naidoo et al. (2009) [70], Reff et al. (2010) [71], Sardans et al. (2012) [72], Wang et al. (2022) [73], and Sun et al. (2024) [74], which underscored the limitation of mangrove growth by nitrogen or phosphorus levels, or both, depending on the intertidal zone dynamics. Meng. Y et al. (2022) [67] explored significant positive impacts of soil nitrogen content on both aboveground and subsurface carbon storage of mangroves in Hainan Island through a structural equation model. However, excessive nitrogen enrichment could negatively impact mangrove AGB, leading to biomass concentration in leaves and decreased plant survival rates under extreme conditions like drought [75]. This ranking of factors suggests that nitrogen management should be a priority in conservation strategies. Implementing practices to enhance nitrogen availability in mangrove ecosystems, such as reducing nutrient runoff or using nitrogen-fixing plants, could significantly impact AGB levels.

TK emerged as the second most influential factor affecting mangrove AGB, with its interaction with salinity intensifying its impact. This finding was corroborated by Sherman et al. (2003) [76], who found a negative correlation between mangrove AGB in the intertidal zone and salinity, suggesting a synergistic enhancement effect with other soil factors. Salinity influenced mangrove AGB by affecting plant growth and development, with specific salinity conditions necessary for optimal growth [77]. Fatoyinbo et al. (2006) [78] noted that stable freshwater input enhanced mangrove biomass compared with high-salinity areas, where biomass levels were lower for the same species.

Mangrove plant species also significantly influenced AGB, likely due to the higher biomass contribution of productive or dominant species within the community [78,79]. Soil particle size exerted minimal direct influence on mangrove AGB, yet it interacted with organic matter (OM) to impact the mangrove ecosystem [54]. The study area’s soil particle size ranged from 0.04 mm to 0.13 mm, predominantly comprising sand, silty sand, sandy silt, and silty sand textures, which had low water retention capabilities affecting soil OM content and, consequently, aboveground biomass. Chaikaew et al. (2017) [79] observed that OM and soil organic carbon exhibited spatial variability decreasing from land to sea or from high to low water, though OM differentiation in the study area was not pronounced. Water pH in the study area showed minimal impact on mangrove AGB. Acidic water or environments could disrupt the nutrient balance in soil, adversely affecting mangrove biodiversity maintenance [79,80].

5.3. Comparison and Limitation of Analysis Methods of Influencing Factors

GWR was used to accurately detect the contribution values of various factors to mangrove AGB and clarify the correlations among different factor combinations [48]. In this study, plant species exerted the greatest influence on AGB, with the combined effect of two factors greater than their individual effects, highlighting the pronounced spatial heterogeneity of mangrove AGB under the influence of multiple environmental factors. Stepwise regression and OLS models effectively eliminated redundant factors, enhancing the accuracy of the geographically weighted regression model while mitigating issues of multicollinearity. This approach facilitated a detailed quantitative assessment of how different influencing factors impacted mangrove AGB across varied geographical locations and enabled the spatial visualization of the local impacts of major environmental factors.

Comprehensive analysis results from the geographic detector and regression models indicate that while the models passed significance tests, the correlation between independent and dependent variables was not notably high. Thus, future studies might benefit from further consideration or inclusion of additional environmental factors such as temperature, tree age, DBH, biodiversity, MAP (mean annual precipitation), NPP (vegetation primary net productivity), carbon–nitrogen ratio, and nitrogen–phosphorus ratio. Addressing these factors could potentially improve the correlation between independent and dependent variables. Moreover, the modest correlation observed in this study might also have stemmed from using uniform bandwidths for each influencing factor in the GWR analysis. Future research could explore multiscale GWR models to adjust bandwidths based on specific influencing factors, thereby enhancing model fit. Alternatively, employing multiple spatial regression models like the Spatial Error Model, Spatial Lag Model, and Spatial Durbin Model could offer comparative insights to optimize explanatory power and model performance.

6. Conclusions

(1)

The spatial analysis of mangrove AGB revealed significant local clustering, with “high–high” hotspots mainly in the southwest and northeast and “low–low” coldspots in the central and southeastern regions, identifying key areas for potential mangrove quality improvement.

(2)

CLA, pH, TK, SA, DO, DEM, OM, TP, and TK emerged as the primary factors influencing the spatial differentiation of mangrove AGB. Interaction effects significantly enhanced the explanatory power of each factor, revealing both synergistic interactions and nonlinear enhancements among them. This underscores that the impact of various factors on mangrove AGB involves complex interactions beyond simple positive or negative correlations among individual factors.

(3)

The main drivers of mangrove AGB spatial differentiation were identified through comprehensive analysis using geographic detectors and multiple regression models, considering single-factor effects, two-factor interactions, and multiple factors. Factors were ranked by their influence intensity from highest to lowest: TN > TK > DEM > DO > OM > pH. TN exhibited the strongest effect on AGB (0.832), followed by TK, while pH had the least effect. TK, TN, OM, pH, and DO were positively correlated with mangrove AGB, promoting its growth. Conversely, DEM exhibited a negative correlation with AGB, indicating an inhibitory effect.

Nevertheless, this study has several limitations. Human activities and temporal effects on mangrove AGB were not considered. Discrepancies in soil and hydrological data across different years posed challenges in establishing uniform intervals for vertical gradient zones during data collection. Future studies could be extended to other aspects, including human factors or additional environmental variables, improvement of data acquisition methods, increased sample size, multitime change analysis of AGB spatial differentiation factors, and construction of mechanism interpretation models, to improve the accuracy and reliability of driver identification.