Forest carbon reserve estimation method fusing airborne hyperspectrum and LiDAR data
By fusing airborne hyperspectral and LiDAR data, a forest carbon storage estimation model was constructed, which solved the problems of time-consuming, labor-intensive and accuracy-limited traditional methods. This model achieves high-precision forest carbon storage estimation, is adaptable to different forest types and terrains, and supports carbon cycle research and management.
Patent Information
- Application Number
- CN202511682126.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-17
- Publication Date
- 2026-03-06
AI Technical Summary
Traditional methods for estimating forest carbon storage are time-consuming, labor-intensive, ecologically damaging, and difficult to promote on a large scale. Furthermore, existing remote sensing technologies rely on a single data source, resulting in limited accuracy, especially with significant edge effects in different forest types and complex terrains.
By integrating airborne hyperspectral and LiDAR data, we can determine forest types, obtain corresponding carbon storage estimation models, extract key features, construct plot shapes and sizes, and perform carbon storage estimation by combining hyperspectral and LiDAR data. We also employ machine learning algorithms to optimize feature selection and model accuracy.
It significantly improves the accuracy and practicality of forest carbon storage estimation, can adapt to different forest types, provides more accurate carbon storage estimation results, and serves carbon cycle research and forest management decision-making.
Smart Images
Figure CN121614822A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of forest carbon storage prediction and data screening technology, and in particular to a method for estimating forest carbon storage by integrating airborne hyperspectral and LiDAR data. Background Technology
[0002] Forest ecosystems are a crucial component of the global carbon cycle, and accurately estimating their carbon storage is essential for addressing climate change and managing forest resources. While traditional field survey methods offer high accuracy, they are time-consuming, labor-intensive, ecologically damaging, and difficult to implement on a large scale. Remote sensing technologies, such as UAV LiDAR, can efficiently acquire three-dimensional forest structure information, overcoming the limitations of traditional optical remote sensing, such as saturation and susceptibility to weather conditions, and have been widely applied to forest parameter inversion.
[0003] Currently, forest resource monitoring primarily utilizes fixed-size square plots, which are ill-suited to diverse forest stand types and complex terrains, leading to significant edge effects and limited estimation accuracy. Studies have shown that plot shape and size significantly impact the accuracy of carbon storage models. Circular plots, due to their uniform boundaries and small perimeter-to-area ratio, effectively mitigate edge effects; however, the optimal plot configuration for different forest types (such as coniferous and broadleaf forests) remains unclear. Furthermore, the relationship between LiDAR-extracted feature variables (such as height, density, intensity, canopy structure, and topography) and carbon storage is complex, and traditional linear models struggle to capture nonlinear relationships. Machine learning algorithms (such as Random Forest (RF) and Extreme Gradient Boosting (XGBoost)) demonstrate advantages in optimizing feature selection and improving model accuracy. Summary of the Invention
[0004] To overcome the shortcomings of existing technologies that rely on a single LiDAR data source and have complex terrain relationships when estimating forest carbon storage using remote sensing technology, this invention proposes a method for estimating forest carbon storage that integrates airborne hyperspectral and LiDAR data. This method can adapt to different forests and automatically adjust to different sample plots.
[0005] This invention proposes a method for estimating forest carbon storage by fusing airborne hyperspectral and LiDAR data, characterized in that: First, determine the forest type and obtain the corresponding carbon storage estimation model for each type; For various types of forests, the shape, size, and input data of the corresponding sample plots for carbon storage estimation models are extracted. The input data includes hyperspectral data and LiDAR data. The carbon storage estimation model extracts key features from the input data of the sample plots and estimates the carbon storage of the corresponding sample plots based on the key features. In each forest type, sample plots of the corresponding shape and size for the carbon storage estimation model are set up. Hyperspectral and LiDAR data of the sample plots are obtained and input into the corresponding carbon storage estimation model to obtain the carbon storage estimation value of the sample plot. Based on the area ratio of forest stands to sample plots, the estimated carbon storage values for each forest type are calculated and summed to obtain the forest carbon storage estimation result.
[0006] Preferably, the method for obtaining the carbon storage estimation model corresponding to each forest type is as follows: First, sample plots of different sizes and shapes were set up in the forest stand. Hyperspectral and LiDAR data of each sample plot were obtained, and feature factors were extracted and screened to obtain the optimal feature set of each sample plot. The plots were classified based on forest type, plot size, and shape. Among similar sample plots, the intersection of the optimal feature sets is taken to obtain the key features, and a dataset {key features, measured carbon storage values} is constructed for each type of sample plot. The datasets of various sample plots are divided into training sets and test sets. The carbon storage estimation module is trained on the training set as a candidate model, and then the performance of the candidate model is tested on the test set. By comparing the performance of candidate models for various sample plots under the same forest type, the best candidate model is selected as the carbon storage estimation module for this type. The carbon storage estimation module is combined with the feature extraction module to form the carbon storage estimation model for this forest type. The feature extraction module extracts key features based on hyperspectral data and LiDAR data, and the carbon storage estimation module estimates carbon storage based on key features. The carbon storage estimation model is associated with forest type, sample plot size, area and key features.
[0007] The preferred method for obtaining the key features of each forest type is as follows: First, feature factors are extracted from the input data of the sample plots to form a feature set; Calculate the weights of all feature factors in the feature set; delete the feature factor with the smallest weight in the feature set to update the feature set; Repeat the above steps to obtain a feature set with successively decreasing sizes; Calculate the weights of feature sets of different sizes, and select the feature factor from the feature set with the largest weight as the key feature.
[0008] Preferably, when calculating the weights of feature sets of different sizes, the random forest algorithm is used to calculate the out-of-bag error of each feature set, so that the weights of the feature sets and the out-of-bag error are inversely proportional, thereby selecting the feature set with the smallest out-of-bag error as the set of key features.
[0009] Preferably, the weights of each feature factor in the feature set are calculated using the random forest algorithm or the Pearson correlation algorithm.
[0010] Preferably, the carbon storage estimation module uses a random forest algorithm or an extreme gradient boosting machine learning algorithm.
[0011] Preferably, circular plots with a radius of 12.91 meters are selected for both coniferous and broad-leaved forests.
[0012] Preferably, all forest plots are square plots with a side length of 25.82 meters.
[0013] The present invention proposes a forest carbon storage estimation system that integrates airborne hyperspectral and LiDAR data, comprising a memory and a processor. The memory stores a computer program, and the processor is connected to the memory. The processor is used to execute the computer program to realize the forest carbon storage estimation method that integrates airborne hyperspectral and LiDAR data.
[0014] The present invention proposes a storage medium storing a computer program, which, when executed, is used to implement the forest carbon storage estimation method that integrates airborne hyperspectral and LiDAR data.
[0015] The advantages of this invention are: This invention proposes a forest carbon storage estimation method that integrates airborne hyperspectral and LiDAR data. It fuses and filters features extracted from airborne LiDAR and hyperspectral data, analyzing the correlation and influence between specific features and forest aboveground carbon storage to identify the most effective feature subset. This process not only lays a solid data foundation for subsequent forest aboveground carbon storage model construction but also significantly improves the model's prediction accuracy and practicality through meticulous data screening, thus better serving carbon cycle research and forest management decisions. The feature fusion strategy integrates the two data sources to comprehensively evaluate and analyze the contribution of each feature variable to predicting forest aboveground carbon storage, thereby identifying key feature parameters strongly correlated with forest aboveground biomass. Based on this, the method achieves precise selection and optimization of model independent variables. Attached Figure Description
[0016] Figure 1 This is a flowchart of the forest carbon storage estimation method that integrates airborne hyperspectral and LiDAR data proposed in this invention. Figure 2(a) shows the characteristic factor analysis of plot L12.91m_CF; Figure 2(b) shows the characteristic factor analysis of plot L6.445m_CF; Figure 2(c) shows the characteristic factor analysis of plot R6.445m_CF; Figure 2(d) shows the characteristic factor analysis of plot R3.2275m_CF; Figure 2(e) shows the characteristic factor analysis of plot L12.91m_BF; Figure 2(f) shows the characteristic factor analysis of plot L6.445m_BF; Figure 2(g) shows the characteristic factor analysis of plot R6.445m_BF; Figure 2(h) shows the characteristic factor analysis of R3.2275m_BF for the sample plot; Figure 2(i) shows the characteristic factor analysis of plot L25.82m_all; Figure 2(j) shows the characteristic factor analysis of plot L12.91m_all; Figure 2(k) shows the characteristic factor analysis of plot L6.445m_all; Figure 2(l) shows the characteristic factor analysis of plot R6.445m_all; Figure 2(m) shows the characteristic factor analysis of plot R3.2275m_all; Figure 3(a) shows the model performance of plot R12.91m_CF; Figure 3(b) shows the model performance of plot R12.91m_BF; Figure 3(c) shows the model performance of plot L25.82m_all. Detailed Implementation
[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0018] like Figure 1 As shown in this embodiment, the forest carbon storage estimation method that integrates airborne hyperspectral and LiDAR data is proposed. First, a carbon storage estimation model that associates forest type, plot shape and size is constructed. The carbon storage estimation model predicts the carbon storage of the plot based on the key features of the corresponding plot's hyperspectral and LiDAR data.
[0019] In this embodiment, sample plots of different sizes and shapes were first constructed for coniferous and broadleaf forests, respectively. Then, LiDAR and hyperspectral data were collected from these sample plots, and characteristic variables were extracted. The hyperspectral data was acquired using an airborne hyperspectral camera. The characteristic variables of the LiDAR data are shown in Tables 1-1 and 1-2; the characteristic variables of the hyperspectral data are shown in Table 1-3.
[0020] Table 1-1 Characteristic Variables of Carbon Storage in Airborne LiDAR
[0021] Table 1-2 Explanation of Some Feature Variables
[0022] Table 1-3 Characteristic Variables of Hyperspectral Data
[0023] In this embodiment, different bands, windows, and texture features are selected and combined for hyperspectral data to obtain a variety of feature factors; for example, this embodiment sets: The three bands are: Band 32 (482 nm), Band 56 (550 nm), and Band 91 (645 nm). The three window types are: 3 × 3, 5 × 5, and 7 × 7; The eight second-order texture features are: second moment of angle, contrast, correlation, entropy, homogeneity, mean, heterogeneity, and variance. Second-order texture features were calculated using different windows at different wavelengths, resulting in a total of 72 texture feature factors, for example: b56_Variance3×3: Gray-scale variance in a Band 56 3×3 window. b56_Variance5×5: Gray-scale variance under a Band 56 5×5 window. b56_Contrast3×3: Contrast level in a Band 56 3×3 window. b32_Contrast7x7: Contrast in a 7x7 window with a Band 56 setting. b56_Homogeneity5×5: Homogeneity in a Band 56 5×5 window b56_Contrast5×5 indicates the contrast in a 5x5 window with a band size of 56. b56_Correlation7×7 represents the correlation under a Band 56 7×7 window.
[0024] In this embodiment, different preprocessing methods are used for hyperspectral data to obtain various feature factors; the preprocessing methods include: Standard normal variate transformation (SNV), also known as variable standardization, is identified by the suffix .snv. The first-order derivative reflectance spectra (FDR), often abbreviated as first-order derivative, is identified by the suffix D1. The second-order derivative reflectance spectra (SDR), often abbreviated as second derivative, is identified by the suffix D2. Multiplicative scatter correction (MSC), also known as spectral multiplicative scatter correction, is identified by the suffix MSC. Discrete wavelet transform (DWT), or simply wavelet transform, is identified by the suffix "wave". Savitzky-Golaysmoothing (SG); The first derivative after polynomial smoothing is identified by the suffix sgfd. The second derivative after polynomial smoothing is identified by the suffix sgsd.
[0025] Vegetation indices were calculated from the spectral curve feature values after hyperspectral data preprocessing. A total of 147 feature factors were obtained from 21 vegetation indices across 7 preprocessing methods. For example: SIPI_D1: Structure Insensitive Pigment Index (SIPI) under first derivative processing. ARVI_D1: Atmospheric impedance vegetation index (ARVI) under first derivative treatment. SAVI_D2: Soil-modified vegetation index under second derivative treatment; EVI_sgsd: Enhanced vegetation index (EVI) after polynomial smoothing and second derivative processing. SIPI_sgsd: Structure Insensitive Pigment Index (SIPI) after polynomial smoothing and second derivative processing. CI_msc: Chlorophyll Index (CI) under spectral multivariate scattering correction. SR_sgfd: Simple structure insensitive pigment index (SR) after polynomial smoothing and first derivative processing. VOG1_sgfd: Red-edge exponent 1 (VOG1) after polynomial smoothing and first derivative processing. VOG2_snv: Red-edge index 2 (VOG2) under normalization; GI_snv: Greenness index (GI) under normalization; WBI_snv: Water Band Index (WBI) under normalized processing; SAVI_wave: Soil-modified vegetation index (SAVI) under wavelet transform.
[0026] In this embodiment, an initial set of feature factors (hereinafter referred to as the initial feature set) is first set up. For each sample plot, feature factors are extracted from airborne hyperspectral and LiDAR data to form the initial feature set. Based on the initial feature set, optimization and screening are performed to obtain the optimal feature set for each sample plot. Optimizing the filtering method includes the following steps: S1. The feature factors in the feature set are sorted by weight using the random forest algorithm. The initial value of the feature set is the initial feature set. S2. Delete the feature factor with the smallest weight in the feature set; S3. Determine whether the number of feature factors in the feature set is greater than or equal to 2; If yes, return to step S1; otherwise, proceed to step S4. S4. Use the random forest algorithm to calculate the out-of-bag error of each size feature set, and select the feature set with the smallest out-of-bag error as the optimal feature set of the sample plot.
[0027] For all optimal feature sets of plots of the same type, the intersection is taken as the key feature of that type of plot. Plots of the same type have the same forest type, plot size, and shape.
[0028] In this embodiment, 18 combinations were constructed for three forest types (coniferous forest CF, broadleaf forest BF, and all forests), as shown in Table 4 below. All forests refer to the sum of coniferous and broadleaf forests.
[0029] Table 4: Key characteristics of various forest types
[0030]
[0031] In Tables 1-4 above, Lnum_CF represents a square coniferous forest plot with a side length of num, and Rnum_CF represents a circular coniferous forest plot with a radius of num; Lnum_BF represents a square broadleaf forest plot with a side length of num, and Rnum_BF represents a circular broadleaf forest plot with a radius of num; Lnum_all represents a square mixed forest plot with a side length of num, and Rnum_all represents a circular mixed forest plot with a radius of num; that is, L25.82m_CF is a coniferous forest plot with a side length of 25.82m, R12.91m_BF is a broadleaf forest plot with a radius of 12.91m, R12.91m_all is all forest plots with a radius of 12.91m, and so on.
[0032] As shown in Tables 1-4, the key features of plots L25.82m_CF, R12.91m_CF, L25.82m_BF, R12.91m_BF, and R12.91m_all all originated from LiDAR data. For the remaining 13 plots, during the iterative phase to reach the optimal feature set, the key features were ranked by importance, and their correlation coefficients and significance levels were labeled. The results are shown in Tables 1-5 and... Figure 2(a)-Figure 2(m) As shown, among which, "Right now P Significant correlation at <0.05, "Right now P The correlation is highly significant at <0.01. "Right now P <0.001 indicates a highly significant correlation, where P represents the probability.
[0033] Although the selected features for the 12.91 m coniferous forest plot included texture features, these texture features did not show a significant relationship with carbon storage. In the 6.445 m coniferous forest plot, the density variable [0] (density_metrics[0]) showed a negative correlation, but the significance was very significant. The correlation under the Band 56 7×7 window (b56_Correlation7×7) and PSRI_sgfd showed a negative correlation, but not a significant relationship with PSRI_sgfd. The 6.445 m coniferous forest plots and the 3.2275 m coniferous forest plots each had a selected hyperspectral feature, ARVI_D1 and SIPI_D1, respectively, both of which showed a negative correlation with carbon storage and no significant relationship.
[0034] In different broadleaf forest plots, more hyperspectral features were selected overall than in coniferous forest plots. The hyperspectral features of the plot with a side length of 12.91 m were highly significant, with a correlation coefficient of 0.38. In contrast, the circular plot with a radius of 6.445 m had three selected hyperspectral features: b56_Variance5×5, b56_Variance3×3, and SIPI_sgsd. The texture feature had correlation coefficients all above 0.3, and its significance and importance were also good; however, the vegetation index feature showed poor importance, correlation, and significance. Plots with a side length of 6.445 m and a radius of 3.2275 m only had vegetation index features selected. These features had extremely low correlation with carbon storage and were slightly less important. The significance of the VOG1_sgfd feature differed significantly between the two plots of different sizes, showing no significant relationship and being highly significant, respectively. This indicates that different feature variables behave differently in different plot shapes, requiring the selection of appropriate features based on different circumstances.
[0035] Based on the classification and composition analysis of forest ecosystems, typical plots in coniferous forests are usually dominated by Masson pine and Chinese fir. These species are large trees growing at high altitudes, with distinctive trunk morphological characteristics, exhibiting straight and rounded trunks, and branch distribution tending towards the apical region. Regarding carbon storage mechanisms, the main carbon storage is concentrated within the trunk, indicating a close correlation between forest carbon storage and tree height variation parameters, reflecting important characteristics of carbon cycling processes and biomass allocation patterns in the ecosystem. In contrast, broadleaf forests are mainly represented by Liquidambar formosana. While the height variation within these forests is not significant, the canopy morphology, branch structure, and leaf outlines among different species exhibit significant diversity. This diversity not only affects the structure and function of the forest ecosystem but is also closely related to the distribution of forest carbon storage, particularly the shape and texture characteristics of the trees.
[0036] Therefore, it can be seen that the key features of the coniferous forest plots in this embodiment involve fewer hyperspectral characteristic variables, while the key features of the broad-leaved forest plots involve more hyperspectral characteristic variables. This phenomenon is consistent with the theoretical derivation and proves the reliability of the key feature screening proposed in this invention.
[0037] Table 1-5 Importance and Correlation Coefficients of Modeling Factors for Each Forest Type Based on Airborne LiDAR and Hyperspectral Data
[0038] In this embodiment, a dataset {key features, measured carbon storage values} is constructed for each type of sample plot. The dataset is then divided into a training set and a test set. The carbon storage estimation module of the carbon storage estimation model is trained on the training set of each sample plot, and the performance of the carbon storage estimation module is tested on the test set.
[0039] In this embodiment, three algorithms—RF (Random Forest), XGBoost (Gradient Boosting Decision Tree), and SVM (Support Vector Machine)—were used to construct a carbon storage estimation module for performance comparison. This embodiment compared model performance on 13 sample sites where key features simultaneously included hyperspectral and LiDAR data features. The results are shown in Table 2.
[0040] Table 2 shows that the XGBoost model performed best on plot L12.91m_CF. Overall, on coniferous forest plots and all forest plots, the RF model and XGBoost model had similar accuracy, while SVM performed poorly. The situation was similar on broadleaf forest plots, but the differences between the three models were smaller.
[0041] Table 2 Comparison of model accuracy for different forest types based on airborne LiDAR and hyperspectral data.
[0042] In this embodiment, the prediction performance of the XGBoost model is represented by a scatter plot. The scatter plot can intuitively show the correlation between the measured and predicted values of forest aboveground carbon storage, specifically as follows: Figures 3(a)-3(c) As shown in the figure, analysis of the scatter plots illustrating the optimal sample plot feature modeling for different forest types reveals a high degree of agreement between the predicted and measured values, demonstrating the reliability of this invention.
[0043] Of course, those skilled in the art will recognize that the present invention is not limited to the details of the exemplary embodiments described above, but also includes the same or similar structures that can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.
[0044] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.
[0045] The technologies, shapes, and structures not described in detail in this invention are all known technologies.
Claims
1. A forest carbon storage estimation method fusing airborne hyperspectral and LiDAR data, characterized in that: firstly, forest types are determined, and corresponding carbon storage estimation models are obtained; for each type of forest, the shape, size and input data of the corresponding sample plot of the carbon storage estimation model are extracted; the input data includes hyperspectral data and LiDAR data; the carbon storage estimation model extracts key features from the input data of the sample plot and estimates the carbon storage of the corresponding sample plot based on the key features; for each forest type, sample plots of the shape and size corresponding to the carbon storage estimation model are set, and the hyperspectral data and LiDAR data of the sample plots are obtained and input into the corresponding carbon storage estimation model to obtain the carbon storage estimation value of the sample plot; according to the area ratio of the stand to the sample plot, the carbon storage estimation value of each forest type is calculated and summed as the forest carbon storage estimation result.
2. The method of claim 1, wherein the fusion of hyperspectral and LiDAR data for forest carbon stock estimation is performed by using a machine learning algorithm. The method for obtaining the carbon storage estimation model corresponding to each forest type is as follows: firstly, sample plots of different sizes and shapes are set in the stand, and the hyperspectral data and LiDAR data of each sample plot are obtained, and the characteristic factors are extracted, and the characteristic factors are screened to obtain the optimal feature set of each sample plot; the sample plots are classified according to the forest category, size and shape; in the same type of sample plot, the intersection of the optimal feature sets is taken as the key features, and the data set {key features, measured carbon storage value} of each type of sample plot is constructed; the data set of each type of sample plot is divided into a training set and a test set, the carbon storage estimation module is trained on the training set as a candidate model, and then the performance of the candidate model is tested on the test set; the performances of the candidate models of each type of sample plot under the same type of forest are compared, and the candidate model with the best performance is selected as the carbon storage estimation module of the type; the carbon storage estimation module and the feature extraction module are combined to form the carbon storage estimation model of the forest type, the feature extraction module extracts the key features based on the hyperspectral data and LiDAR data, and the carbon storage estimation module estimates the carbon storage based on the key features; and the carbon storage estimation model is associated with the forest type, sample plot size, area and key features.
3. The method of claim 2, wherein the fusion of hyperspectral and LiDAR data for forest carbon stock estimation is performed by: The method for obtaining the key features of each forest type is as follows: firstly, the characteristic factors are extracted from the input data of the sample plot to form a feature set; the weights of all characteristic factors in the feature set are calculated; the characteristic factor with the smallest weight in the feature set is deleted to update the feature set; the above steps are repeated to obtain feature sets with decreasing sizes; the weights of the feature sets of different sizes are calculated, and the characteristic factor in the feature set with the largest weight is selected as the key feature.
4. The method of claim 3, wherein the fusion of hyperspectral and LiDAR data for forest carbon stock estimation is performed by using a machine learning algorithm. When calculating the weights of the feature sets of different sizes, the out-of-bag error of each feature set is calculated using the random forest algorithm, and the weight of the feature set is inversely proportional to the out-of-bag error, so that the feature set with the smallest out-of-bag error is selected as the key feature set.
5. The method of claim 4, wherein the fusion of hyperspectral and LiDAR data for forest carbon stock estimation is performed by, The weights of each characteristic factor in the feature set are calculated using the random forest algorithm or the Pearson correlation algorithm.
6. The method of claim 2, wherein the fusion of hyperspectral and LiDAR data for forest carbon stock estimation is performed by using a machine learning algorithm. The carbon storage estimation module uses the random forest algorithm or the extreme gradient boosting machine learning algorithm.
7. The fusion of hyperspectral and LiDAR data onboard forest carbon stock estimation method according to claim 1, characterized in that, Coniferous forests and broad-leaved forests select circular sample plots with a radius of 12.91 meters.
8. The method of claim 1, wherein the fusion of hyperspectral and LiDAR data for forest carbon stock estimation is performed by: All forests use square sample plots with a side length of 25.82 meters.
9. A forest carbon stock estimation system fusing airborne hyperspectral and LiDAR data, characterized by, The application discloses a forest carbon storage estimation method fusing airborne hyperspectral and LiDAR data.
10. A storage medium, characterized by The application discloses a forest carbon storage estimation method fusing airborne hyperspectral and LiDAR data.