Next Article in Journal
Seasonal and Long-Term Water Regime Trends of Cheremsky Wetland: Analysis Based on Sentinel-2 Spectral Indices and Composite Indicator Development
Previous Article in Journal
Remotely Sensing Phytoplankton Size Structure in the Mediterranean Sea: Insights from In Situ Data and Temperature-Corrected Abundance-Based Models
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Remote Sensing
Volume 17
Issue 14
10.3390/rs17142358
remotesensing-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Estimation of Forest Aboveground Biomass Using Sentinel-1/2 Synergized with Extrapolated Parameters from LiDAR Data and Analysis of Its Ecological Driving Factors

by
Xu Xu
1,2,
Jingyu Yang
3,
Shanze Qi
1,2,
Yue Ma
1,2,
Wei Liu
1,2,
Luanxin Li
1,2,
Xiaoqiang Lu
1,2,*,† and
Yan Liu
1,2,†
1
Nanjing Institute of Environmental Sciences, Ministry of Ecology and Environment of China, Nanjing 210042, China
2
State Environmental Protection Key Laboratory on Biodiversity and Biosafety, Nanjing 210042, China
3
School of Geomatics, Anhui University of Science and Technology, Huainan 232001, China
*
Author to whom correspondence should be addressed.
†
These authors contributed equally to this work.
Remote Sens. 2025, 17(14), 2358; https://doi.org/10.3390/rs17142358
Submission received: 2 June 2025 / Revised: 3 July 2025 / Accepted: 8 July 2025 / Published: 9 July 2025
Browse Figures
Versions Notes

Abstract

Accurate estimation of forest aboveground biomass (AGB) and understanding its ecological drivers are vital for carbon monitoring and sustainable forest management. However, AGB estimation using remote sensing is hindered by signal saturation in high-biomass areas and insufficient attention to ecological structural factors. Focusing on Guangdong Province, this study proposes a novel approach that spatially extrapolates airborne LiDAR-derived Forest structural parameters and integrates them with Sentinel-1/2 data to construct an AGB prediction model. Results show that incorporating structural parameters significantly reduces saturation effects, improving prediction accuracy and AGB maximum range in high-AGB regions (R2 from 0.724 to 0.811; RMSE = 10.64 Mg/ha; max AGB > 180 Mg/ha). Using multi-scale geographically weighted regression (MGWR), we further examined the spatial influence of forest type, age structure, and species mixture. Forest age showed a strong positive correlation with AGB in over 95% of the area, particularly in mountainous and hilly regions (coefficients up to 1.23). Species mixture had positive effects in 87.7% of the region, especially in the north and parts of the south. Natural forests consistently exhibited higher AGB than plantations, with differences amplifying at later successional stages. Highly mixed natural forests showed faster biomass accumulation and higher steady-state AGB, highlighting the regulatory role of structural complexity and successional maturity. This study not only mitigates remote sensing saturation issues but also deepens understanding of spatial and ecological drivers of AGB, offering theoretical and technical support for targeted carbon stock assessment and forest management strategies.

1. Introduction

Forests are vital to the global carbon balance, with aboveground biomass (AGB) serving as a key indicator of carbon storage and ecosystem productivity [1]. Accurate AGB estimation is crucial for climate change monitoring and terrestrial carbon cycle assessment [2]. Spatially explicit AGB patterns also support biodiversity conservation and carbon neutrality strategies [3,4]. Traditional field-based methods, such as destructive sampling and allometric equations, provide accurate plot-level estimates [5,6], but are limited by terrain complexity and high labor costs, hindering regional scaling and leaving significant spatial and temporal gaps. Remote sensing offers a cost-effective alternative, enabling large-scale AGB estimation through broad spatial coverage and frequent observations [7].
Medium-resolution optical satellite data, such as the Landsat series, have been widely used for AGB estimation due to their suitable spatiotemporal resolution and free access [8]. For instance, Gizachew et al. [9] demonstrated that Landsat 8 spectral variables have a strong linear relationship with AGB, supporting accurate modeling in Tanzanian forests. However, Landsat’s coarse spatial resolution limits its ability to capture fine-scale vegetation structure, and its spectral sensitivity to forest parameters is often insufficient [10,11]. Moreover, the 16-day revisit cycle reduces data availability in cloudy regions [12]. In contrast, Sentinel-2, launched in 2017, significantly improves spatial, temporal, and spectral resolution [13,14], enhancing the detection of structural variations, mitigating spectral mixing, and providing red-edge bands highly sensitive to vegetation structure [15]. Despite these advances, optical data still face spectral saturation under dense canopies (typically at 15–70 Mg/ha AGB), leading to biomass underestimation [16,17,18,19]. For example, in a study on AGB estimation in Zhejiang Province, Zhao et al. [20] observed that the diversity of vegetation types and structural complexity introduced pronounced spectral saturation effects, which ultimately limited the accuracy and reliability of optical-based AGB inversion. Integrating complementary remote sensing data sources has been recognized as an effective strategy to address these limitations [19]. Synthetic Aperture Radar (SAR) data, such as from Sentinel-1, are widely recognized for mitigating AGB saturation due to their ability to penetrate clouds and canopies and their sensitivity to canopy moisture [17]. SAR backscatter carries rich structural and biochemical information, making it particularly effective for AGB prediction in high-biomass regions, as confirmed by Minh et al. [21]. However, SAR does not fully eliminate saturation; Lucas et al. [22] found that while SAR extends the saturation threshold up to ~150 Mg/ha, signal saturation still occurs at very high biomass levels. In contrast, LiDAR provides direct structural metrics (e.g., tree height, DBH, and basal area) highly correlated with AGB [23], effectively avoiding spectral saturation. Yet, due to high costs, complex processing, and sparse sampling, LiDAR is often used as a reference rather than a primary data source for large-scale mapping [24]. And the derived forest structural metrics are spatially discrete and cannot provide seamless, continuous forest information coverage like optical or microwave data [25]. In summary, although optical, SAR, and LiDAR data each offer advantages for AGB estimation, their inherent limitations highlight the need for multi-source data integration [19]. Therefore, our research focuses on developing effective fusion strategies to overcome issues of weather interference, spectral saturation, and spatial discontinuity, aiming for seamless, accurate large-scale AGB mapping.
It is important to recognize that forest AGB is dynamic and heterogeneous, influenced by multiple factors [26]. Most studies emphasize site-scale natural conditions, such as soil nutrients, climate, and topography, in AGB prediction [27]. However, an exclusive focus on environmental variables risks overlooking the crucial influence of intrinsic ecological attributes—such as forest origin, stand age, and species mixture patterns—which directly affect growth rates, carbon stocks, and AGB heterogeneity through mechanisms like interspecific competition and resource allocation [28,29,30]. Indirabai et al. [31] further highlighted that geographic and environmental factors only partially explain AGB variability, with stand structure and management practices playing significant roles. In response, our study shifts focus toward these intrinsic forest properties, emphasizing their importance for improving AGB prediction accuracy [32]. Although similar investigations exist at the individual tree scale—linking intrinsic traits (e.g., sap flow rate and stomatal conductance) to growth and carbon accumulation [33]—they are typically limited by small spatial scales and cannot be directly extrapolated to ecosystems or regions. A key challenge, therefore, lies in whether remote sensing technologies can effectively capture forest origin, stand age, and species diversity at large scales, an issue this study aims to address.
In summary, this study focuses on the forest ecosystems of Guangdong Province, developing a forest AGB prediction model and a forest ecological attribute prediction model based on multi-source remote sensing data and in situ measurements using the Random Forest algorithm. Multi-scale geographically weighted regression (MGWR) and statistical analyses were further employed to explore the relationships between forest origin, stand age, species mixture patterns, and AGB. Specifically, this research addresses two key questions:
(1)
Can the integration of Sentinel-1, Sentinel-2, and airborne LiDAR data achieve high-accuracy forest AGB prediction?
(2)
What are the effects of forest attributes such as forest origin (plantation/natural forest), stand age, and degree of species mixing on the spatial distribution and temporal accumulation patterns of forest AGB?

2. Study Area and Data

2.1. Study Area

Guangdong Province, located in southeastern China, spans approximately 180,000 square kilometers and has a subtropical monsoon climate. The region features diverse landforms, with mountains and hills covering approximately 60% of the area. Elevations range from 4 m to 1989 m, and the forest coverage exceeds 53%, with a forest stock volume of approximately 580 million cubic meters (as shown in Figure 1). Forest types include subtropical/tropical evergreen broadleaf and coniferous forests. Since the 1980s, large-scale conservation efforts, such as natural forest protection, afforestation, and returning farmland to forests, have been implemented to restore forest resources and boost carbon sequestration. Currently, artificial forests cover over 45% of the region, contributing more than 46% to the forest stock volume (https://lyj.gd.gov.cn/ (accessed on 12 May 2025)). These artificial forests mainly consist of fast-growing species like Pinus massoniana, Pinus elliottii, and Eucalyptus, which are young and ecologically unstable. Meanwhile, natural forests are facing issues such as aging and reduced growth vitality. The resulting structural heterogeneity makes this area an ideal case for studying the relationship between forest origin, age, species mix, and forest AGB, offering valuable insights for forest management and carbon sequestration optimization.

2.2. Data

2.2.1. Sentinel-2 Data

To ensure temporal consistency with the 2022 forest management inventory (FMI) data used as ground truth, Sentinel-2 surface reflectance data from the same year (2022) were selected for this study. Additionally, considering the frequent cloudy and rainy conditions in the study area, 2022 data were retrieved from the Google Earth Engine (GEE) platform, with images filtered to retain those with less than 50% cloud cover. Cloud and shadow pixels were masked using the QA60 band, and monthly median composites were generated to further reduce cloud contamination. Due to complex meteorological conditions, the compositing period was extended in 7-day intervals until the null value proportion in the area was below 15%, with less than 5% in forested areas, resulting in nine valid tiles. Band selection included visible bands (B2–B4), red-edge bands (B5–B7), near-infrared bands (B8–B8A), and shortwave infrared bands (B11–B12), resampled to 10 m resolution using cubic convolution interpolation for pixel alignment. Null values were filled using cubic spline interpolation to ensure spatial continuity and completeness.
For accurate AGB prediction, several widely used vegetation indices were calculated, including NDVI, SAVI, RVI, NIRV, REIP, EVI, GVI, and MSAVI. Feature dimensionality reduction was performed using Principal Component Analysis (PCA), retaining the first five components (NDVI, SAVI, NIRV, REIP, and EVI), which explained over 96.4% of the variance, the sources and formulas for the indices are shown in Table 1. These indices were selected for model building. Data acquisition and index calculation were completed on the GEE platform, with null value filling implemented in Python 3.7.0.

2.2.2. Sentinel-1 Data

While Sentinel-1 data are not affected by cloudy or rainy weather, its quality can be influenced by recent precipitation. Therefore, Sentinel-1 images from 2022 with favorable weather conditions were selected, using the compositing time of Sentinel-2 data as a reference. Preprocessing operations, including boundary and thermal noise removal, radiometric calibration, and terrain correction, were performed on each image using the S1 Toolbox on the GEE platform. The Sentinel-1 images were then median-composited into 9 tiles, including both VV and VH bands, based on the same time intervals as the Sentinel-2 data.

2.2.3. Laser Point Cloud Data

This study utilized the DJI Matrice 300 RTK UAV equipped with the RIEGL VUX-1 LiDAR sensor to collect forest point cloud data. The Matrice 300 RTK ensures high-precision positioning for large-area, high-resolution data acquisition, while the RIEGL VUX-1 LiDAR sensor, with multi-echo detection and full waveform recording, effectively penetrates the forest canopy to capture terrain and vegetation structure. The sensor’s sampling width is 20 m by 20 m. Sampling plots were selected based on uniformity and representativeness, covering various forest types (coniferous, broadleaf, pure, and mixed forests) to ensure data applicability. The collected point cloud data were processed with LASTools, registered with RTK-GNSS data, and transformed to the WGS84 coordinate system to maintain a positional accuracy within 5 cm. The Progressive Morphological Filter (PMF) method was applied to classify ground and non-ground points, and quality control measures were set: a signal-to-noise ratio of at least 50, a maximum height of 100 m above ground, and a point cloud density of at least 400 points/m2. These procedures resulted in 349 flight paths and 4420 quality-compliant point cloud footprints (Figure 2).
For each retained full waveform dataset, the point cloud data were interpolated using the TIN method to construct a Canopy Height Model (CHM). A 5 × 5 sliding window was applied to identify local maxima as canopy center points. These maxima served as seed points for the watershed algorithm, which delineated the canopy boundaries. After manual adjustments and vector smoothing, the canopy range for each dataset was determined. Based on this range, the following parameters were extracted: First Return Elevation (FRE), Last Return Elevation (LRE), Number of Ground Return Points (NGRPs), and Total Number of Return Points (TNRPs), which are closely linked to forest structural features such as canopy height and cover.

2.2.4. Forest Reference Data

The forest AGB reference data are sourced from the 2022 forest management inventory data (FMI) by the Guangdong Forestry Bureau, which provides detailed forest information for Guangdong Province, including data on forest origin, dominant tree species, tree height, diameter at breast height (DBH), volume, and AGB. AGB information is derived from destructive sampling, which is determined by constructing an anisotropic growth equation parameterized by the average tree height and diameter at breast height within the felled sample plots. However, the FMI data have some limitations, such as small patch sizes, boundary ambiguities, location errors, and disturbances like deforestation and pests. To address these issues, high-quality patches were retained using filters based on area, canopy cover, and health status: (1) area greater than 30 m × 30 m, (2) canopy cover exceeding 0.2 (as the Chinese Forestry Administration considers areas with canopy cover > 0.2 to be forests), and (3) forest health status better than “good”. A 500 m buffer zone was applied to reduce spatial autocorrelation effects in the AGB predictions, and further filtering was done using a leave-one-out method. The remaining patches were visually inspected and cross-checked with high-resolution imagery, resulting in 10,350 valid patches.
Field measurements were conducted from April to October 2022 across 4347 plots of 10 m x 10 m to explore the relationship between forest mixing patterns and AGB. Each plot had an average tree height greater than 2 m and DBH greater than 5 cm. A 500 m buffer zone was applied between plots to reduce spatial autocorrelation. Species counts and tree numbers were recorded, and plot locations were georeferenced in ArcGIS Pro 3.0. Plots affected by recent disturbances were excluded, leaving 3874 valid field measurements.

3. Method

3.1. Extrapolation of Laser Point Cloud Derivative Parameters

The laser point cloud parameters extracted in this study, namely FRE, LRE, NGRPs, and TNRPs, do not have direct ecological significance and cannot be directly applied to forest mapping tasks. Typically, relationships between these parameters and common forest structural parameters need to be established to convert them into parameters with ecological and biological meaning, such as tree height and canopy cover. Furthermore, due to the cost limitations of drone flights, irregular and substantial gaps exist between the point cloud footprints in this study, which makes it difficult for the derived point cloud parameters to be comparable with seamless spectral data, hindering their direct application in forest mapping tasks. To address these issues, this study calculates the average tree height and canopy cover characteristics corresponding to each canopy range based on the above-derived parameters. These characteristics are considered to be highly correlated with forest aboveground biomass (AGB) height [40]. The calculation method is as follows:
M T H = 1 n ∑ i = 1 n ( F R E i − L R E i )
M C C = 1 n ∑ i = 1 n ( T N P R i − N G P R i ) / T N P R i
where M T H is the abbreviation for average tree height, n is the number of crowns extracted from the point cloud data through vectorization, F R E i is the FRE of the i − t h canopy, L R E i is the LRE of the i − t h canopy, M C C is the abbreviation for average canopy cover, T N P R i is the TNRP of the i − t h canopy, and N G P R i is the NGRP of the   i − t h canopy.
The point cloud data are resampled into a 10 m grid to ensure pixel size matching, with average tree height and canopy cover values assigned to each grid cell based on the canopy range. After establishing the ecological significance of the derived parameters, the study uses the resampled point cloud grid data alongside preprocessed Sentinel-1/2 data in a Random Forest regression model (RF), which is effective for modeling nonlinear relationships in high-dimensional multi-source remote sensing data. To ensure model stability and generalization, a strict sample partitioning strategy and hyperparameter optimization are applied. The K-Means clustering method is used to spatially group grid samples, preventing data leakage by assigning each cluster exclusively to either the training or testing set. Hyperparameters such as the number of decision trees (n_estimators), tree depth (max_depth), and minimum samples for splitting (min_samples_split) are tuned using grid search with 10-fold cross-validation. The final model is evaluated on the independent test set, and predictive accuracy is assessed using R2 and RMSE metrics.

3.2. Forest AGB Prediction

The forest AGB reference data points were randomly divided into training and validation sets based on the sample partition strategy in Section 3.1. The RF regression model was applied using Sentinel-1/2 spectral bands, vegetation indices, and tree height and canopy cover features derived from laser point cloud data to predict forest AGB. Model training and hyperparameter tuning followed the steps outlined in Section 3.1. For accuracy validation, the R2 and RMSE metrics were used, and a control model excluding tree height and canopy cover was constructed to assess the impact of the extrapolated parameters on AGB prediction. Additionally, the predicted forest AGB was compared with the high-resolution dataset by Yang et al. [41] by overlaying the maps and evaluating their spatial consistency through overlap comparison.

3.3. Forest Attribute Variable Acquisition

To explore the relationship between forest AGB and factors such as forest origin (plantation/natural forests), tree age, and degree of mixing, this study performed predictive analysis on forest attributes, including forest type classification, tree age prediction, and forest mixing pattern prediction. For forest origin classification, the study used reference data from Section 2.2.4 and Sentinel-1/2 spectral bands and vegetation indices as features. To mitigate bias from a single classifier, four algorithms were employed: Random Forest (RF), Support Vector Machine (SVM), Extreme Gradient Boosting (XGB), and Maximum Likelihood Classification (MLC). Classification performance was evaluated using Overall Accuracy (OA) and Kappa coefficients. A correction scheme was applied to improve reliability by overlaying the results of the four models, using a sliding window method for consistency testing. For pixels with classification discrepancies, the mode method was used, and in case of a 2:2 conflict, the model with the highest accuracy was chosen as the final classification criterion.
For tree age prediction, due to the lack of detailed age attributes in the FMI data, the study utilized the 2022 China Forest Tree Age Dataset by Shang et al. [42], which offers 30 m resolution forest tree age data in China. This dataset, based on change detection algorithms and machine learning regression techniques, provided reliable forest tree age data at a larger scale, forming the basis for analyzing the relationship between forest AGB and tree age.
The prediction of forest mixing degree quantifies species diversity using the Shannon Diversity Index (H’) as an indicator of alpha diversity (α diversity) (see Formula (3)). This index considers both species richness and evenness, reflecting the species composition of the forest. The H’ value ranges from 0 (a homogeneous forest with one species) to ln(S) (maximum diversity when all species are equally abundant), with higher values indicating a higher degree of species mixing. After calculating H’ for all plots, the study used Sentinel-1/2 remote sensing features to construct a diversity prediction model. The RF algorithm was employed for model training and accuracy validation, following the procedures outlined in Section 3.2 to ensure the model’s reliability and generalization ability.
H ′ = − ∑ i = 1 s P i ln P i
where S is the total number of species in the plot, P i is the proportional abundance of species, and i is the relative to the total abundance of all species in the plot.

3.4. Validation of the Relationship Between Forest AGB and Forest Attribute Variables

This study analyzed the relationship between forest attributes and AGB by spatially overlaying three key forest variables—tree age, mixing pattern, and AGB data. To ensure spatial consistency, the tree age data with a 30 m resolution was resampled to 10 m using cubic convolution interpolation, aligning it with the other variables. This allowed accurate spatial matching of each variable with AGB data for subsequent analysis. We employed multi-scale geographically weighted regression (MGWR), an advanced spatial regression method, to account for spatial heterogeneity by allowing each independent variable to have a unique spatial bandwidth. This approach is more flexible and precise than traditional geographically weighted regression (GWR) as it captures the varying influence of explanatory variables across geographical space. Forest origin data, a binary variable, were excluded from the MGWR model due to its failure to meet the continuous variable assumption. The remaining variables—tree age, mixing pattern, and AGB—were aggregated into 7.5 km × 7.5 km grid cells to meet MGWR requirements, with grid size determined through multiple manual tests. The analysis focused on spatial distribution of local regression coefficients and their significance, with pseudo t-values processed to identify areas where values exceeded 1.96, marking them as significant. This approach helped identify regional variations in the impact of tree age and mixing pattern on AGB.
Further, 1000 random validation points were sampled from different forest types to compare AGB across regions, compensating for the exclusion of forest origin data. A control variable analysis was then performed to examine AGB variations with tree age under different combinations of forest origin and mixing degree, categorized into high, medium, and low levels. Tree age and AGB data were overlaid for each combination, with random sampling performed on the extracted grid units. AGB was then averaged by tree age for the sampled points, and continuous curves were generated to show how AGB evolves with tree age under different forest conditions.

4. Results

4.1. The Extrapolation Results of the Derived Parameters from the Laser Point Cloud

Figure 3 shows the spatial extrapolation of laser point cloud-derived parameters using the RF regression model. Figure 3a displays tree height distribution across Guangdong Province, with values ranging from 2.88 m to 24.67 m. The western and northwestern regions generally have higher tree heights (13.22 m to 24.67 m), while the northern and southern regions show lower heights (2.88 m to 12.35 m). Figure 3b illustrates the extrapolation of canopy cover values, ranging from 0.177 to 0.924, with higher canopy cover in the western and central regions and lower values in the northwest and northern areas. Figure 4 demonstrates the accuracy and stability of the extrapolation model. In Figure 4a, the regression model achieved high fitting accuracy for tree height prediction with an R2 of 0.798 and an RMSE of 1.595. Similarly, canopy cover prediction in Figure 4b showed an R2 of 0.738 and an RMSE of 0.077, confirming the feasibility of using laser point cloud-derived parameters for forest structural mapping.

4.2. Forest AGB Prediction Results

Figure 5 shows the comparison of prediction accuracy before and after including laser point cloud-derived parameters (tree height and canopy cover) in the forest AGB prediction model. The analysis shows that adding extrapolated parameters significantly improved model performance and mitigated the saturation effect in AGB prediction. In panel (a), the model without extrapolated parameters had an R2 of 0.724, RMSE of 15.59, and a regression coefficient of 0.6899, with underfitting in high-AGB areas (>130 Mg/ha). Panel (b) shows improved performance, with R2 rising to 0.811, RMSE decreasing to 10.64, and the regression coefficient increasing to 0.8603. The inclusion of these parameters addressed the saturation issue, expanding the AGB range from 159.8 Mg/ha to over 180 Mg/ha. The regression equation in panel (b) also shows a closer fit to the 1:1 line, confirming the model’s enhanced prediction of high-AGB values.
Figure 6 shows the forest AGB prediction results with the inclusion of laser point cloud extrapolated parameters. High-AGB areas (>120 Mg/ha) are concentrated in the central, northern, and northwestern mountainous regions, characterized by higher elevations, large topographic variations, and a humid climate, supporting natural broadleaf and mixed coniferous-broadleaf forests. These areas have high forest density, greater tree heights, and complex canopy structures, resulting in higher AGB accumulation. Medium-AGB areas (60–120 Mg/ha) are found in the hilly regions of the Pearl River Delta and some central plateaus, exhibiting significant spatial heterogeneity and local AGB hotspots. Low-AGB areas (<60 Mg/ha) are primarily located in the southwestern and core regions of the Pearl River Delta, where intense human disturbance and frequent land-use changes result in lower AGB levels.
To further assess the accuracy and reliability of the forest AGB predictions in this study, the predicted AGB values were compared with publicly available forest AGB products [41]. The relative differences between the two were calculated for corresponding grid cells, with a ±20% threshold used for accuracy evaluation (as shown in Figure 7). The results show that in 80% of the area, the predicted AGB values differ from the existing product within ±20%, while only 20% of the area exhibits differences greater than ±20%. Notably, the areas with larger discrepancies show distinct geographical patterns, with significant differences concentrated in the Pearl River Delta, coastal hilly areas, the northeastern part of the study area, the southwestern coastal areas, and the edges of certain mountainous regions. Overall, these findings confirm that the model used in this study provides reliable simulations of forest AGB, with strong consistency with authoritative products in most regions, demonstrating good spatial consistency and stability. This establishes a solid foundation for further exploring the relationship between forest attributes and AGB.

4.3. Prediction Results of Various Forest Attribute Variables

This study used machine learning algorithms and existing forest variable products to obtain high-precision forest attribute data. Figure 8 presents the spatial distribution of artificial and natural forests in Guangdong Province, derived using a multi-classifier ensemble strategy. It also includes confusion matrix results for four classifiers (RF, SVM, XGB, and MLC). The confusion matrices in Figure 8b–e show strong performance from all classifiers, with dominant diagonal values indicating high reliability in the classification results. Table 2 reports Overall Accuracy and Kappa coefficients, with all algorithms (except SVM) achieving 82% accuracy and Kappa coefficients is 0.64. The XGB model achieved the highest accuracy at 93% and a Kappa of 0.85. Sliding window detection and mode assignment were used to obtain the final artificial/natural forest classification, minimizing classification disagreement (Figure 8a). The results reveal regional differences in forest distribution: natural forests dominate the northern, western, and mountainous areas, especially in the Nanling Mountains, while artificial forests are concentrated in the northern and central Pearl River Delta. Statistically, natural forests cover approximately 54% of the area, while artificial forests account for 46%.
Figure 9 illustrates the spatial distribution of forest age in Guangdong Province, based on the processed tree age product. The forest age exhibits significant spatial heterogeneity, with most forests being middle aged (20–60 years), and only a few regions having forests older than 60 years. The southern coastal and western areas are dominated by younger forests (under 20 years old), including artificial and secondary forests, reflecting intensive reforestation efforts. The central and eastern regions mainly feature forests aged 20–60 years, which were planted or naturally restored in the late 20th century. In contrast, the northern region hosts older forests, with some areas containing forests over 60 years old, which are remnants of old-growth forests. The bar chart in the lower right quantifies the age distribution: forests aged 20–40 years make up 35.25%, 40–60 years account for 33.92%, forests under 20 years comprise 29.28%, and forests older than 60 years represent only 1.55%. This distribution highlights that the forest resources in Guangdong are primarily middle aged, with old-growth forests being rare.
This study uses the Random Forest inversion algorithm to predict tree species diversity, specifically the Shannon index (Figure 10), achieving high accuracy (R2 = 0.791, RMSE = 0.272). The study area shows significant spatial heterogeneity in its mixed forest pattern, with an average Shannon index of approximately 1.17. High-altitude mountainous regions in the north, central, and west, where natural forests dominate, generally exhibit higher diversity, with H’ values ranging from 0.83 to 2.3. In contrast, the northeastern and southwestern regions, dominated by plantations, show lower diversity, with H’ values between 0 and 0.4. The northern–central and southern regions, with a mix of plantations and natural forests, show strong heterogeneity and fragmentation, with H’ values ranging from 0 to 1.22.

4.4. The Relationship Between Various Forest Variables and Forest AGB

In the previous section, high-precision data on forest type, stand age structure, and mixed forest patterns were used to represent key attributes of forest ecosystem structure and succession. These attributes exhibit significant spatial heterogeneity and regional distribution patterns. Specifically, forest type reflects differences in origin and management, tree age structure indicates temporal succession, and mixed forest patterns highlight species composition complexity and ecological stability. These attributes are crucial for understanding regional forest functions and AGB distribution. The MGWR modeling results (Figure 11) show that forest age and mixed forest patterns significantly influence AGB distribution, with notable spatial heterogeneity. Figure 11a indicates a positive correlation between tree age and AGB in most areas (95.4% of forest grids), with higher coefficients in northern, southwestern, and southeastern regions (over 0.75, some exceeding 1.23), suggesting tree age promotes AGB accumulation. Only 4.6% of the area shows minimal or negative effects, confirming tree age’s key role in spatial AGB variation in Guangdong. Figure 11b shows that mixed forest patterns positively impact AGB in northern and some southern regions, with coefficients over 0.88, indicating these patterns enhance biomass accumulation. In contrast, the Pearl River Delta and northeastern regions exhibit lower or negative coefficients. Overall, the effect of mixed forest patterns is weaker than tree age, with positive effects observed in 87.7% of the area.
Figure 12 illustrates the relationship between forest origin type and AGB (Figure 12a) and the cumulative patterns of AGB changes with stand age under different origin and mixing levels (Figure 12b), further addressing the absence of origin type data in the MGWR analysis and deepening the understanding of the relationship between forest structural variables and AGB. Figure 12a shows that natural forests have significantly higher AGB than planted forests. The median AGB of natural forests is 113.68 Mg/ha, compared to only 87.37 Mg/ha for planted forests, representing a statistically significant difference of 26.31 Mg/ha. Additionally, the upper quartile of AGB in natural forests approaches 140 Mg/ha, substantially exceeding that of planted forests at approximately 120 Mg/ha. Figure 12b presents the cumulative AGB-age curves for different mixing classes across origin types. Overall, AGB increases steadily with age across all forest types, but distinct differences are observed in growth patterns and accumulation rates. All curves display a typical growth pattern: rapid accumulation during the early stage (1–30 years), followed by a gradual slowdown in the middle and late stages. In terms of origin type, natural forests (LN, MN, HN) consistently exhibit higher AGB accumulation patterns than planted forests (PP, LP), with the differences becoming more pronounced in middle and older age classes. For example, at 55–60 years of age, natural forests across all mixing levels surpass 100 Mg/ha, whereas planted forests remain below 100 Mg/ha.
Notably, plantation forests exhibit faster early-stage AGB accumulation: by age 20, the AGB of PP and LP forests reaches approximately 80 Mg/ha and 85 Mg/ha, respectively, compared to approximately 60–70 Mg/ha in natural forests. However, the growth rate of plantation forests slows earlier, plateauing around 40 years of age, and stabilizing at approximately 95 Mg/ha. Within the same origin type, forests with higher mixing degrees exhibit faster early-stage AGB accumulation and achieve higher final AGB levels. For example, among natural forests, the high-mixing class (HN) consistently maintains the highest AGB accumulation rate across the age series. Similarly, within plantation forests, low-mixed forests (LP) show higher AGB growth rates and sustained AGB levels compared to pure plantations (PP) throughout the age sequence.

5. Discussion

In this study, we propose a novel synergistic approach that integrates Sentinel-1/2 imagery with discrete airborne laser scanning point cloud data for high-precision mapping of forest AGB distribution in Guangdong Province. Moreover, as another core component of the study, we further investigated the influence of forest ecological attributes on the spatial distribution and accumulation patterns of AGB. Although this study made significant progress in validating AGB prediction accuracy and elucidating influencing relationships, several issues remain that warrant further investigation. Therefore, to comprehensively assess the applicability and reliability of the proposed framework, this section focuses on the following three aspects:

5.1. Uncertainty in the Elimination Process of Saturation Effects

While the integration of extrapolated LiDAR structural parameters significantly improved the AGB prediction accuracy and mitigated saturation effects, several sources of uncertainty remain. The Random Forest model used to spatially extrapolate tree height and canopy cover was trained on UAV-LiDAR samples and satellite spectral features. Although the model achieved high accuracy (R2 = 0.798 for tree height and 0.738 for canopy cover), the extrapolation process is inherently sensitive to landscape heterogeneity.
Specifically, in areas characterized by complex topography, abrupt forest boundaries, or diverse forest compositions, the relationships between spectral responses and canopy structure may become highly nonlinear. As a result, prediction errors in the structural layers may accumulate in specific regions, particularly in mountainous zones or fragmented landscapes, where misclassification or overgeneralization of tree structural attributes can occur [23,40]. These uncertainties in the structural inputs can propagate into the subsequent AGB prediction process, leading to spatially biased biomass estimates in localized areas. Therefore, while our strategy effectively reduces the saturation bias at a broader scale, its performance in heterogeneous or transitional environments still warrants further validation using independent field data.

5.2. Potential Constraints in the Construction of Forest Ecological Attribute Sets

Beyond the technical uncertainty in saturation correction, our analysis of ecological drivers of AGB also involves several limitations related to data sources and variable selection. One notable source of uncertainty is the forest age data, which were derived from the national-scale forest age product developed by Shang et al. [43]. Although this dataset offers comprehensive coverage at a 30 m resolution and fills a critical data gap, it was not specifically validated in our study region. The use of remote sensing-based change detection and regression may lead to inaccuracies in areas with unclear disturbance histories or intensive plantation activities, potentially introducing systematic bias into the age–AGB relationship modeling via MGWR.
Similarly, our approach to characterizing species mixture relied on the Shannon Diversity Index (H′), which, while commonly used to represent α-diversity, does not account for the structural complexity of forest stands. H′ captures species richness and evenness but ignores vertical layering, canopy stratification, and variations in tree size distribution, which are crucial for understanding carbon accumulation patterns in complex forest ecosystems [28,30]. In particular in mature natural forests or structurally rich mixed forests, the oversimplification of diversity may limit our ability to fully capture its influence on AGB. These limitations suggest that while forest ecological attributes such as stand age and species mixture are indeed important drivers of biomass variation, their representation using remote sensing data may suffer from data availability, measurement proxy effects, and ecological simplification [44]. We recommend that future research incorporate more structurally informative diversity metrics (e.g., functional diversity and vertical heterogeneity indices) and field-verified age information to enhance model interpretability and ecological relevance.

5.3. Uncertainty About the Drivers of AGB Spatial Heterogeneity and Accumulation Patterns

The spatial heterogeneity and accumulation patterns of forest AGB are closely related to natural factors such as temperature, precipitation, and site conditions [45]. Numerous studies have examined the individual and synergistic effects of geographic and climatic factors on AGB [46]. For example, Qiuyue et al. [26] found that the combined effects of climate, topographic relief, and soil elemental content led to significant spatial variation in AGB distribution and accumulation. However, beyond natural environmental factors, forest ecological attributes also play a critical role in shaping AGB patterns [32]. These attributes reflect successional stages and determine resource utilization efficiency and growth potential. Under similar climate and site conditions, planted and natural forests often show distinct AGB accumulation trajectories due to differences in silvicultural density, regeneration methods, and management intensity. Neglecting these attributes would result in a biased understanding of AGB drivers.
To reveal these ecological drivers, we selected three representative variables: forest origin, stand age structure, and species mixture degree. Using multi-scale geographically weighted regression (MGWR), we found that stand age exerted a significant positive influence on AGB in over 95% of forested areas, with especially high coefficients observed in the northern mountainous regions and southern hills. Species mixture effects were more spatially variable, but still showed positive contributions in 87.7% of the area, particularly in ecologically stable natural forests. Although forest origin was excluded from the MGWR due to its categorical nature, statistical comparison revealed that natural forests had markedly higher AGB than plantations (113.68 vs. 87.37 Mg/ha).
Nevertheless, this study mainly assessed the independent effects of forest attributes, without fully exploring their interactions. In reality, these attributes often jointly influence biomass accumulation through complex mechanisms. For instance, the effect of mixing structure may vary with stand age [47,48], and plantation and natural forests may respond differently to similar mixing levels [49]. Ignoring such interactions risks misrepresenting the dominant drivers and their marginal effects across regions. Moreover, although the importance of forest ecological attributes was emphasized, their interactions with climate, hydrothermal conditions, and topography were not fully incorporated. Previous studies have shown that the regulatory capacity of ecological attributes varies with climate: in areas with abundant hydrothermal resources, optimizing the mixing structure can significantly enhance AGB [49], whereas in resource-limited areas, topography and soil constraints may weaken these effects [47]. As Rodrigues et al. [50] found, ecological and natural factors synergistically regulate forest processes, with strong spatial heterogeneity. Topographic variation and forest attributes jointly affect soil fertility, thereby influencing forest productivity and AGB distribution. Thus, failing to systematically account for the interactions between ecological attributes and natural factors may obscure key regional variations and affect the overall understanding of AGB formation mechanisms.
Moreover, while the proposed methodology was developed and validated within the subtropical monsoon climate of Guangdong Province, its core components—including the LiDAR-based structural extrapolation, Sentinel-1/2 integration, and MGWR-based spatial modeling—are broadly transferable to other climate zones. However, successful application in different ecological settings will require careful adaptation to account for regional variability in species composition, disturbance regimes, and spectral/structural responses. For example, radar backscatter behavior and vegetation index sensitivity may differ significantly in tropical, boreal, or arid forests. Therefore, region-specific recalibration and validation will be essential to ensuring model robustness in other bioclimatic contexts.

6. Conclusions

In this study, we propose a new strategy for forest AGB modeling that combines point cloud parameter extrapolation with multi-source remote sensing data to mitigate the AGB saturation effect. Forest structural parameters extracted from LiDAR are extrapolated into spatially continuous layers using a Random Forest inversion algorithm, and combined with Sentinel-1/2 imagery to build an AGB prediction model, significantly improving estimation accuracy. To further uncover the sources and accumulation mechanisms of AGB spatial heterogeneity, we systematically analyze the influence of ecological attributes—forest origin, stand age structure, and mixing pattern—on AGB distribution using MGWR and statistical methods. The main conclusions are as follows:
1. Remote sensing models incorporating LiDAR-derived structural parameters significantly mitigate AGB saturation effects. In this study, structural parameters extracted from LiDAR point clouds were spatially extrapolated via the Random Forest model and integrated with Sentinel-1/2 data to construct the AGB prediction model. This approach effectively reduced saturation bias in medium- and high-AGB regions. Results show that introducing extrapolated structural variables improved model performance, raising the R2 from 0.724 to 0.811 and reducing RMSE to 10.64. In high-AGB areas, the fit to measured values increased markedly, with the fit coefficient improving from 0.6899 to 0.8603, and the AGB response range expanding from 159.8 Mg/ha to over 180 Mg/ha. These findings demonstrate that the proposed strategy not only enhanced overall prediction accuracy but also strengthened model fitting and generalization in high-biomass regions, effectively alleviating the saturation effect in spatial estimations.
2. Multi-scale modeling quantitatively reveals the role of forest ecological attributes in AGB spatial distribution. This study systematically analyzed the effects of forest type, age structure, and mixing pattern on AGB accumulation and spatial heterogeneity. The MGWR results showed a significant positive correlation between tree age and AGB, with positive regression coefficients in 95.4% of areas, peaking at 1.23 in the northern mountainous regions and southern hilly areas, indicating that tree age positively contributes to AGB. The mixed pattern had a positive effect in 87.7% of areas, with coefficients greater than 0.88 in the north and part of the south, highlighting the role of structural diversity in biomass enhancement. Additionally, natural forests had significantly higher AGB than planted forests (median 113.68 vs. 87.37 Mg/ha), with the difference most pronounced in middle and upper age stages. AGB increased with tree age in all forest types, and highly mixed forests showed faster growth and higher accumulation rates in both natural and planted forests. These findings confirm that forest structural variables are key drivers of AGB’s spatial and temporal patterns.

Author Contributions

Conceptualization, X.X., J.Y., S.Q., Y.M., W.L., L.L., X.L. and Y.L.; Methodology, X.X., S.Q., Y.M., X.L. and Y.L.; Software, X.X., J.Y., S.Q., L.L., X.L. and Y.L.; Validation, X.X., J.Y., W.L. and L.L.; Formal analysis, X.X., Y.M., X.L. and Y.L.; Investigation, X.L.; Resources, S.Q., X.L. and Y.L.; Data curation, S.Q., Y.M., W.L. and L.L.; Writing—original draft, X.X., J.Y., S.Q., Y.M., W.L., L.L. and Y.L.; Writing—review & editing, X.X., J.Y., X.L. and Y.L.; Visualization, X.X., J.Y. and Y.M.; Supervision, X.L.; Project administration, X.L. and Y.L.; Funding acquisition, X.L. All authors have read and agreed to the published version of the manuscript.

Funding

This work was supported by the National Natural Science Foundation of China, Grant No. 41961144022.

Data Availability Statement

The data are not publicly available due to privacy restrictions.

Acknowledgments

We thank the reviewers for their thoughtful comments and constructive suggestions which substantially improved this manuscript.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Piao, S.; Yue, H.; Wang, X.; Chen, F. Estimation of China’s terrestrial ecosystem carbon sink: Methods, progress and prospects. Sci. China Earth Sci. 2022, 65, 641–651. [Google Scholar] [CrossRef]
  2. Van der Werf, G.R.; Morton, D.C.; DeFries, R.S.; Olivier, J.G.; Kasibhatla, P.S.; Jackson, R.B.; Collatz, G.J.; Randerson, J.T. CO2 emissions from forest loss. Nat. Geosci. 2009, 2, 737–738. [Google Scholar]
  3. Pan, Y.; Birdsey, R.A.; Fang, J.; Houghton, R.; Kauppi, P.E.; Kurz, W.A.; Phillips, O.L.; Shvidenko, A.; Lewis, S.L.; Canadell, J.G. A large and persistent carbon sink in the world’s forests. Science 2011, 333, 988–993. [Google Scholar] [PubMed]
  4. Shendryk, Y. Fusing GEDI with earth observation data for large area aboveground biomass mapping. Int. J. Appl. Earth Obs. Geoinf. 2022, 115, 103108. [Google Scholar]
  5. Lu, D.; Chen, Q.; Wang, G.; Liu, L.; Li, G.; Moran, E. A survey of remote sensing-based aboveground biomass estimation methods in forest ecosystems. Int. J. Digit. Earth 2016, 9, 63–105. [Google Scholar]
  6. Torre-Tojal, L.; Bastarrika, A.; Boyano, A.; Lopez-Guede, J.M.; Grana, M. Above-ground biomass estimation from LiDAR data using random forest algorithms. J. Comput. Sci. 2022, 58, 101517. [Google Scholar]
  7. Powell, S.L.; Cohen, W.B.; Healey, S.P.; Kennedy, R.E.; Moisen, G.G.; Pierce, K.B.; Ohmann, J.L. Quantification of live aboveground forest biomass dynamics with Landsat time-series and field inventory data: A comparison of empirical modeling approaches. Remote Sens. Environ. 2010, 114, 1053–1068. [Google Scholar]
  8. Soenen, S.A.; Peddle, D.R.; Hall, R.J.; Coburn, C.A.; Hall, F.G. Estimating aboveground forest biomass from canopy reflectance model inversion in mountainous terrain. Remote Sens. Environ. 2010, 114, 1325–1337. [Google Scholar]
  9. Gizachew, B.; Solberg, S.; Næsset, E.; Gobakken, T.; Bollandsås, O.M.; Breidenbach, J.; Zahabu, E.; Mauya, E.W. Mapping and estimating the total living biomass and carbon in low-biomass woodlands using Landsat 8 CDR data. Carbon Balance Manag. 2016, 11, 13. [Google Scholar]
  10. Pasquarella, V.J.; Holden, C.E.; Woodcock, C.E. Improved mapping of forest type using spectral-temporal Landsat features. Remote Sens. Environ. 2018, 210, 193–207. [Google Scholar]
  11. Hościło, A.; Lewandowska, A. Mapping forest type and tree species on a regional scale using multi-temporal Sentinel-2 data. Remote Sens. 2019, 11, 929. [Google Scholar]
  12. Sibanda, M.; Mutanga, O.; Rouget, M. Discriminating rangeland management practices using simulated hyspIRI, landsat 8 OLI, sentinel 2 MSI, and VENµs spectral data. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2016, 9, 3957–3969. [Google Scholar]
  13. Wan, L.; Ryu, Y.; Dechant, B.; Hwang, Y.; Feng, H.; Kang, Y.; Jeong, S.; Lee, J.; Choi, C.; Bae, J. Correcting confounding canopy structure, biochemistry and soil background effects improves leaf area index estimates across diverse ecosystems from Sentinel-2 imagery. Remote Sens. Environ. 2024, 309, 114224. [Google Scholar] [CrossRef]
  14. Mandl, L.; Viana-Soto, A.; Seidl, R.; Stritih, A.; Senf, C. Unmixing-based forest recovery indicators for predicting long-term recovery success. Remote Sens. Environ. 2024, 308, 114194. [Google Scholar] [CrossRef]
  15. Zeug, G.; Immitzer, M.; Neuwirth, M.; Atzberger, C. Machbarkeitsstudie zur Nutzung von Satellitenfernerkundungsdaten (Copernicus) für Zwecke der Ableitung Ökologischer Belastungsgrenzen und der Verifizierung von Indikatoren der Deutschen Anpassungsstrategie an den Klimawandel: Feasibility Study on the Use of Earth Observation Data for the Purpose of Determining Ecological Exposure Limits and for the Verification of Indicators of the German Adaptation Strategy to Climate Change; Terranea UG: Geltendorf, Germany, 2018. [Google Scholar]
  16. Majasalmi, T.; Rautiainen, M. The potential of Sentinel-2 data for estimating biophysical variables in a boreal forest: A simulation study. Remote Sens. Lett. 2016, 7, 427–436. [Google Scholar]
  17. Woodhouse, I.H.; Mitchard, E.T.A.; Brolly, M.; Maniatis, D.; Ryan, C.M. Radar backscatter is not a ‘direct measure’ of forest biomass. Nat. Clim. Change 2012, 2, 556–557. [Google Scholar]
  18. Mitchard, E.T.; Saatchi, S.S.; White, L.J.; Abernethy, K.A.; Jeffery, K.J.; Lewis, S.L.; Collins, M.; Lefsky, M.A.; Leal, M.E.; Woodhouse, I.H. Mapping tropical forest biomass with radar and spaceborne LiDAR in Lopé National Park, Gabon: Overcoming problems of high biomass and persistent cloud. Biogeosciences 2012, 9, 179–191. [Google Scholar]
  19. Su, Y.; Guo, Q.; Xue, B.; Hu, T.; Alvarez, O.; Tao, S.; Fang, J. Spatial distribution of forest aboveground biomass in China: Estimation through combination of spaceborne lidar, optical imagery, and forest inventory data. Remote Sens. Environ. 2016, 173, 187–199. [Google Scholar] [CrossRef]
  20. Zhao, P.; Lu, D.; Wang, G.; Wu, C.; Huang, Y.; Yu, S. Examining Spectral Reflectance Saturation in Landsat Imagery and Corresponding Solutions to Improve Forest Aboveground Biomass Estimation. Remote Sens. 2016, 8, 469. [Google Scholar]
  21. Minh, D.H.T.; Le Toan, T.; Rocca, F.; Tebaldini, S.; Villard, L.; Réjou-Méchain, M.; Phillips, O.L.; Feldpausch, T.R.; Dubois-Fernandez, P.; Scipal, K. SAR tomography for the retrieval of forest biomass and height: Cross-validation at two tropical forest sites in French Guiana. Remote Sens. Environ. 2016, 175, 138–147. [Google Scholar]
  22. Lucas, R.M.; Mitchell, A.L.; Armston, J. Measurement of forest above-ground biomass using active and passive remote sensing at large (subnational to global) scales. Curr. For. Rep. 2015, 1, 162–177. [Google Scholar]
  23. Popescu, S.C.; Zhao, K.; Neuenschwander, A.; Lin, C. Satellite lidar vs. small footprint airborne lidar: Comparing the accuracy of aboveground biomass estimates and forest structure metrics at footprint level. Remote Sens. Environ. 2011, 115, 2786–2797. [Google Scholar] [CrossRef]
  24. May, P.B.; Schlund, M.; Armston, J.; Kotowska, M.M.; Brambach, F.; Wenzel, A.; Erasmi, S. Mapping aboveground biomass in Indonesian lowland forests using GEDI and hierarchical models. Remote Sens. Environ. 2024, 313, 114384. [Google Scholar] [CrossRef]
  25. Xi, Y.; Tian, Q.; Zhang, W.; Zhang, Z.; Tong, X.; Brandt, M.; Fensholt, R. Quantifying understory vegetation density using multi-temporal Sentinel-2 and GEDI LiDAR data. GISci. Remote Sens. 2022, 59, 2068–2083. [Google Scholar]
  26. Qiuyue, F.; shan, Z.; Su, H.; Chongyu, Y.; Dongming, F.; Tingting, M.; Guomo, Z. Above-ground biomass of Moso bamboo forests in China influenced by climate, soil and topography: A meta-analysis. Adv. Bamboo Sci. 2024, 8, 100091. [Google Scholar] [CrossRef]
  27. Xu, M.; Ji, H.; Zhuang, S. Carbon stock of Moso bamboo (Phyllostachys pubescens) forests along a latitude gradient in the subtropical region of China. PLoS ONE 2018, 13, e0193024. [Google Scholar]
  28. Bai, J.; Ren, C.; Shi, X.; Xiang, H.; Zhang, W.; Jiang, H.; Ren, Y.; Xi, Y.; Wang, Z.; Mao, D. Tree species diversity impacts on ecosystem services of temperate forests. Ecol. Indic. 2024, 167, 112639. [Google Scholar] [CrossRef]
  29. Laclau, J.-P.; Ranger, J.; de Moraes Gonçalves, J.L.; Maquère, V.; Krusche, A.V.; M’Bou, A.T.; Nouvellon, Y.; Saint-André, L.; Bouillet, J.-P.; de Cassia Piccolo, M. Biogeochemical cycles of nutrients in tropical Eucalyptus plantations: Main features shown by intensive monitoring in Congo and Brazil. For. Ecol. Manag. 2010, 259, 1771–1785. [Google Scholar]
  30. Felipe-Lucia, M.R.; Soliveres, S.; Penone, C.; Manning, P.; van der Plas, F.; Boch, S.; Prati, D.; Ammer, C.; Schall, P.; Gossner, M.M. Multiple forest attributes underpin the supply of multiple ecosystem services. Nat. Commun. 2018, 9, 4839. [Google Scholar]
  31. Indirabai, I.; Nilsson, M. Estimation of above ground biomass in tropical heterogeneous forests in India using GEDI. Ecol. Inform. 2024, 82, 102712. [Google Scholar] [CrossRef]
  32. Sun, H.; Terhonen, E.; Koskinen, K.; Paulin, L.; Kasanen, R.; Asiegbu, F.O. Bacterial diversity and community structure along different peat soils in boreal forest. Appl. Soil Ecol. 2014, 74, 37–45. [Google Scholar]
  33. Qu, Z.; Liu, B.; Ma, Y.; Sun, H. Differences in bacterial community structure and potential functions among Eucalyptus plantations with different ages and species of trees. Appl. Soil Ecol. 2020, 149, 103515. [Google Scholar] [CrossRef]
  34. Tucker, C.J. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sens. Environ. 1979, 8, 127–150. [Google Scholar] [CrossRef]
  35. Bolyn, C.; Michez, A.; Gaucher, P.; Lejeune, P.; Bonnet, S. Forest mapping and species composition using supervised per pixel classification of Sentinel-2 imagery. Biotechnol. Agron. Soc. Environ. 2018, 22, 172–187. [Google Scholar]
  36. Jordan, C.F. Derivation of Leaf-Area Index from Quality of Light on the Forest Floor. Ecology 1969, 50, 663–666. [Google Scholar] [CrossRef]
  37. Badgley, G.; Field, C.B.; Berry, J.A. Canopy near-infrared reflectance and terrestrial photosynthesis. Sci. Adv. 2017, 3, e1602244. [Google Scholar] [CrossRef]
  38. Martimort, P.; Fernandez, V.; Kirschner, V.; Isola, C.; Meygret, A. Sentinel-2 MultiSpectral imager (MSI) and calibration/validation. In Proceedings of the 2012 IEEE International Geoscience and Remote Sensing Symposium, Munich, Germany, 22–27 July 2012; pp. 6999–7002. [Google Scholar]
  39. Huete, A.; Didan, K.; Miura, T.; Rodriguez, E.P.; Gao, X.; Ferreira, L.G. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ. 2002, 83, 195–213. [Google Scholar] [CrossRef]
  40. Qi, Y.; Coops, N.C.; Daniels, L.D.; Butson, C.R. Comparing tree attributes derived from quantitative structure models based on drone and mobile laser scanning point clouds across varying canopy cover conditions. ISPRS J. Photogramm. Remote Sens. 2022, 192, 49–65. [Google Scholar] [CrossRef]
  41. Yang, Q.; Niu, C.; Liu, X.; Feng, Y.; Ma, Q.; Wang, X.; Tang, H.; Guo, Q. Mapping high-resolution forest aboveground biomass of China using multisource remote sensing data. GISci. Remote Sens. 2023, 60, 2203303. [Google Scholar]
  42. Shang, R.; Lin, X.; Chen, J.M.; Liang, Y.; Fang, K.; Xu, M.; Yan, Y.; Ju, W.; Yu, G.; He, N.; et al. China’s annual forest age dataset at 30 m spatial resolution from 1986 to 2022. Earth Syst. Sci. Data Discuss. 2025, 2025, 1–31. [Google Scholar] [CrossRef]
  43. Shang, R.; Chen, J.M.; Xu, M.; Lin, X.; Li, P.; Yu, G.; He, N.; Xu, L.; Gong, P.; Liu, L.; et al. China&#x2019;s current forest age structure will lead to weakened carbon sinks in the near future. Innovation 2023, 4, 100515. [Google Scholar] [CrossRef] [PubMed]
  44. Zolkos, S.G.; Goetz, S.J.; Dubayah, R. A meta-analysis of terrestrial aboveground biomass estimation using lidar remote sensing. Remote Sens. Environ. 2013, 128, 289–298. [Google Scholar] [CrossRef]
  45. Liu, L.; Zeng, F.; Song, T.; Wang, K.; Du, H. Stand structure and abiotic factors modulate karst forest biomass in Southwest China. Forests 2020, 11, 443. [Google Scholar] [CrossRef]
  46. Liu, L.; Chang, S.X.; Huang, C.; Zhi, Y.; Jie, Y.; Yu, X.; Jiang, P. Enhancement of phytolith-occluded carbon accumulation of Moso bamboo response to temperatures elevation and different fertilization. Front. Plant Sci. 2023, 14, 1144961. [Google Scholar]
  47. Waite, O.J.M.; Coops, N.C.; Grubinger, S.; Isaac-Renton, M.; Degner, J.; King, J.; Liu, A. Responses of spectral indices to heat and drought differ by tree size in Douglas-fir. Sci. Remote Sens. 2025, 11, 100193. [Google Scholar] [CrossRef]
  48. Marias, D.E.; Meinzer, F.C.; Woodruff, D.R.; McCulloh, K.A. Thermotolerance and heat stress responses of Douglas-fir and ponderosa pine seedling populations from contrasting climates. Tree Physiol. 2017, 37, 301–315. [Google Scholar]
  49. Wang, S.; Zuo, X.; Zhao, X.; Awada, T.; Luo, Y.; Li, Y.; Qu, H. Dominant plant species shape soil bacterial community in semiarid sandy land of northern China. Ecol. Evol. 2018, 8, 1693–1704. [Google Scholar]
  50. Rodrigues, A.C.; Villa, P.M.; Ferreira-Júnior, W.G.; Schaefer, C.E.R.G.; Neri, A.V. Effects of topographic variability and forest attributes on fine-scale soil fertility in late-secondary succession of Atlantic Forest. Ecol. Process. 2021, 10, 62. [Google Scholar] [CrossRef]
Remotesensing 17 02358 g001
Figure 1. Overview of the study area. (a) Location of the study area in China. (b) Elevation of the study area. (c) Land-use status of the study area. (d) Proportion of each land-use type in the study area.
Figure 1. Overview of the study area. (a) Location of the study area in China. (b) Elevation of the study area. (c) Land-use status of the study area. (d) Proportion of each land-use type in the study area.
Remotesensing 17 02358 g001
Remotesensing 17 02358 g002
Figure 2. Distribution map of LiDAR point cloud data sampling. A-1 and B-1 represent examples of LiDAR point cloud footprints. A-2 and B-2 represent examples of LiDAR point cloud tiles at certain locations, the different colors in the figure correspond to the heights relative to the ground after grading.
Figure 2. Distribution map of LiDAR point cloud data sampling. A-1 and B-1 represent examples of LiDAR point cloud footprints. A-2 and B-2 represent examples of LiDAR point cloud tiles at certain locations, the different colors in the figure correspond to the heights relative to the ground after grading.
Remotesensing 17 02358 g002
Remotesensing 17 02358 g003
Figure 3. Spatial extrapolation results of laser point cloud-derived parameters. (a) Spatial distribution of average tree height after extrapolation. (b) Spatial distribution of canopy cover after extrapolation. Panels A-1 to C-1 display typical examples of average tree height prediction results, while panels A-2 to C-2 show typical examples of canopy cover prediction results.
Figure 3. Spatial extrapolation results of laser point cloud-derived parameters. (a) Spatial distribution of average tree height after extrapolation. (b) Spatial distribution of canopy cover after extrapolation. Panels A-1 to C-1 display typical examples of average tree height prediction results, while panels A-2 to C-2 show typical examples of canopy cover prediction results.
Remotesensing 17 02358 g003
Remotesensing 17 02358 g004
Figure 4. Accuracy validation results of laser point cloud-derived parameters during spatial extrapolation. (a) Accuracy validation results for average tree height. (b) Accuracy validation results for canopy cover. N represents the size of the validation dataset.
Figure 4. Accuracy validation results of laser point cloud-derived parameters during spatial extrapolation. (a) Accuracy validation results for average tree height. (b) Accuracy validation results for canopy cover. N represents the size of the validation dataset.
Remotesensing 17 02358 g004
Remotesensing 17 02358 g005
Figure 5. Forest AGB prediction accuracy before and after the inclusion of laser point cloud-derived parameters. (a) Prediction accuracy without the inclusion of laser point cloud extrapolated parameters. (b) Prediction accuracy with the inclusion of laser point cloud extrapolated parameters.
Figure 5. Forest AGB prediction accuracy before and after the inclusion of laser point cloud-derived parameters. (a) Prediction accuracy without the inclusion of laser point cloud extrapolated parameters. (b) Prediction accuracy with the inclusion of laser point cloud extrapolated parameters.
Remotesensing 17 02358 g005
Remotesensing 17 02358 g006
Figure 6. Spatial distribution of forest AGB prediction results with the inclusion of laser point cloud extrapolated parameters. A–C represent examples of local scales for aboveground biomass inversion results.
Figure 6. Spatial distribution of forest AGB prediction results with the inclusion of laser point cloud extrapolated parameters. A–C represent examples of local scales for aboveground biomass inversion results.
Remotesensing 17 02358 g006
Remotesensing 17 02358 g007
Figure 7. Consistency analysis results of predicted forest AGB and existing AGB products.
Figure 7. Consistency analysis results of predicted forest AGB and existing AGB products.
Remotesensing 17 02358 g007
Remotesensing 17 02358 g008
Figure 8. Forest type variable prediction results. (a) Represents the artificial/natural forest prediction results after mode assignment. (b–e) Represent the confusion matrices corresponding to RF, SVM, XGB, and MLC classification models, respectively.
Figure 8. Forest type variable prediction results. (a) Represents the artificial/natural forest prediction results after mode assignment. (b–e) Represent the confusion matrices corresponding to RF, SVM, XGB, and MLC classification models, respectively.
Remotesensing 17 02358 g008
Remotesensing 17 02358 g009
Figure 9. Tree age distribution results in the study area.
Figure 9. Tree age distribution results in the study area.
Remotesensing 17 02358 g009
Remotesensing 17 02358 g010
Figure 10. Prediction results of the forest tree species diversity index H’ in the study area. (a) Represents the spatial distribution of H’. (b) Represents the prediction accuracy of H’.
Figure 10. Prediction results of the forest tree species diversity index H’ in the study area. (a) Represents the spatial distribution of H’. (b) Represents the prediction accuracy of H’.
Remotesensing 17 02358 g010
Remotesensing 17 02358 g011
Figure 11. Results of MGWR modeling analysis. (a) shows the results of local regression coefficients and significance analysis for tree age. (b) shows the results of local regression coefficients and significance analyses for mixed patterns. The diagonally shaded lines in the figure represent areas with significance, and the white areas in the circular pie chart in the lower right corner represent local regression coefficients less than zero.
Figure 11. Results of MGWR modeling analysis. (a) shows the results of local regression coefficients and significance analysis for tree age. (b) shows the results of local regression coefficients and significance analyses for mixed patterns. The diagonally shaded lines in the figure represent areas with significance, and the white areas in the circular pie chart in the lower right corner represent local regression coefficients less than zero.
Remotesensing 17 02358 g011
Remotesensing 17 02358 g012
Figure 12. Results of the validation of the relationship between forest origin and AGB and the results of the accumulation curves of forest AGB with age corresponding to each mixing degree class under different origins. (a) represents forest AGB differences under different origins. (b) represents differences in accumulation patterns of forest AGB under different forest origins with different levels of mixing. pp: pure plantation forest, LP: plantation forest with low degree of mixing, LN: natural forest with low degree of mixing, MN: natural forest with medium degree of mixing, and HN: natural forest with high degree of mixing.
Figure 12. Results of the validation of the relationship between forest origin and AGB and the results of the accumulation curves of forest AGB with age corresponding to each mixing degree class under different origins. (a) represents forest AGB differences under different origins. (b) represents differences in accumulation patterns of forest AGB under different forest origins with different levels of mixing. pp: pure plantation forest, LP: plantation forest with low degree of mixing, LN: natural forest with low degree of mixing, MN: natural forest with medium degree of mixing, and HN: natural forest with high degree of mixing.
Remotesensing 17 02358 g012
Table 1. Sources and Formulas of Selected Vegetation Indices.
Table 1. Sources and Formulas of Selected Vegetation Indices.
Spectral IndicesDescriptionFormula
NDVI [34] Normalized Difference Vegetation Index (B8 − B4)/(B8 + B4)
SAVI [35]Soil-Adjusted Vegetation Index(1 + 0.2) × float (B8 − B4)/(B8 + B4 + 0.2)
RVI [36]Ratio Vegetation IndexB4/B8
NIRV [37]Near-Infrared Reflection of Vegetation ( ( B 8 − B 4 ) / ( B 8 + B 4 ) ) × B 4
REIP [38]Red-Edge Inflection Point Index 705 + 35   ×   ((B4 + B7)/2 − (B5/B6) − B5)
EVI [39]Enhanced Vegetation Index2.5 × (B8 − B4)/(B8 + 6 × B4 − 7.5 × B2 + 1)
Table 2. Prediction accuracy of each model for forest type classification.
Table 2. Prediction accuracy of each model for forest type classification.
ModelOAKappa Coefficient
RF89%0.77
SVM79%0.57
XGB93%0.85
MLC82%0.64
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Xu, X.; Yang, J.; Qi, S.; Ma, Y.; Liu, W.; Li, L.; Lu, X.; Liu, Y. Estimation of Forest Aboveground Biomass Using Sentinel-1/2 Synergized with Extrapolated Parameters from LiDAR Data and Analysis of Its Ecological Driving Factors. Remote Sens. 2025, 17, 2358. https://doi.org/10.3390/rs17142358

AMA Style

Xu X, Yang J, Qi S, Ma Y, Liu W, Li L, Lu X, Liu Y. Estimation of Forest Aboveground Biomass Using Sentinel-1/2 Synergized with Extrapolated Parameters from LiDAR Data and Analysis of Its Ecological Driving Factors. Remote Sensing. 2025; 17(14):2358. https://doi.org/10.3390/rs17142358

Chicago/Turabian Style

Xu, Xu, Jingyu Yang, Shanze Qi, Yue Ma, Wei Liu, Luanxin Li, Xiaoqiang Lu, and Yan Liu. 2025. "Estimation of Forest Aboveground Biomass Using Sentinel-1/2 Synergized with Extrapolated Parameters from LiDAR Data and Analysis of Its Ecological Driving Factors" Remote Sensing 17, no. 14: 2358. https://doi.org/10.3390/rs17142358

APA Style

Xu, X., Yang, J., Qi, S., Ma, Y., Liu, W., Li, L., Lu, X., & Liu, Y. (2025). Estimation of Forest Aboveground Biomass Using Sentinel-1/2 Synergized with Extrapolated Parameters from LiDAR Data and Analysis of Its Ecological Driving Factors. Remote Sensing, 17(14), 2358. https://doi.org/10.3390/rs17142358

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

Article Metrics

Back to TopTop