Subtropical eucalyptus man-made forest parameter extraction method based on unmanned aerial vehicle laser radar

Abstract

The invention discloses a method for extracting parameters of a (subtropical) eucalyptus man-made forest based on an unmanned aerial vehicle laser radar. The method comprises the following steps: investigating a sample plot, and measuring and calculating actual forest parameters such as average diameter, average height, sectional area, stand volume and the like; acquiring and preprocessing high-density unmanned aerial vehicle laser radar point cloud data, and extracting three types of feature variables of height, density and vertical structure; deducing a model structural formula on the basis of a stock different-speed growth equation, and screening variables according to rules; constructing a power model by using a regular exhaustion method; the optimal model is screened out through model parameter estimation, inspection, significance analysis and the like. According to the method, the variable synergistic effect is mined through the exhaustion method, the multi-dimensional test system is constructed, and the exclusive model is customized, so that the problems of one-sided variable screening, insufficient model reliability and poor specific forest stand adaptability in the prior art are solved, high-precision estimation of core forest parameters is realized, and the method is suitable for large-scale popularization and application. And scientific and effective technical support is provided for refined management and sustainable operation of eucalyptus man-made forest resources.

Description

Technical Field

[0001] This invention belongs to the technical field of forestry remote sensing and forest resource survey, specifically involving a method for extracting parameters of subtropical eucalyptus plantations based on UAV lidar. It is applicable to the fields of accurate forest resource survey, growth status monitoring, structural parameter analysis, and efficient management and sustainable operation of forestry resources in subtropical eucalyptus plantations. Background Technology

[0002] Accurate forest parameter estimation provides crucial data support for the sustainable management of forest resources and ecological protection. UAV lidar technology, with its efficient and flexible operation modes, high-density lidar point cloud data acquisition capabilities, and high-precision, large-scale acquisition of forest three-dimensional structure data, demonstrates significant application value in the quantitative inversion of forest parameters. Although existing research has constructed various estimation models for different forest types and parameters and achieved preliminary results, the following problems still exist: 1. Limitations of variable selection methods: The selection of variables and the accuracy of forest parameter estimation models are significantly affected by specific factors such as forest stand type and environmental conditions in the study area. The synergistic effect of multi-dimensional variables is often ignored, resulting in insufficient capture of the three-dimensional structural features of the forest and limited estimation accuracy. 2. Incomplete model validation: Existing model validation relies on simple fitting indices and lacks multi-dimensional quantitative evaluation of systematic bias, generalization ability and statistical significance, making it difficult to verify reliability under complex forest conditions and resulting in uncontrollable errors when applied across regions. 3. Insufficient adaptation to specific forest stands: There is a lack of dedicated models for the unique structure of subtropical eucalyptus plantations. Directly applying general models can easily overlook their specific variables, and the accuracy of core parameter estimation cannot meet the needs of refined management. Summary of the Invention

[0003] 1. Purpose of the invention This invention proposes a method for extracting parameters of (subtropical) eucalyptus plantations based on UAV lidar. It extracts feature variables from high-density UAV lidar point cloud data, then uses an exhaustive method to select variables to construct a power-law model. The method then fits, evaluates, and verifies the parameters of each model, selecting the optimal estimation model structure suitable for subtropical eucalyptus plantations. This achieves accurate estimation of core forest parameters such as tree height, diameter at breast height (DBH), sectional area, and volume. It solves the problems of insufficient exploration of synergistic effects in variable selection, lack of multi-dimensional model analysis systems, and scarcity of specialized models for estimating stand parameters in subtropical eucalyptus plantations in existing technologies. This provides scientific and effective technical support for the refined management and sustainable operation of eucalyptus plantation resources.

[0004] 2. Technical Solution A method for extracting parameters of (subtropical) eucalyptus plantations based on UAV lidar includes the following steps: Step 1: Plot survey and data calculation (1) Measure the diameter at breast height, height of dominant trees, average age, and canopy closure of trees in each sample plot; (2) Based on the measured data, calculate the actual forest parameters such as average diameter (DBH), average height (H), basal area (BA), and stand volume (VOL) of trees in each sample plot. The stand volume is calculated according to the corresponding allometric growth equation, and the formula is as follows: In the formula: Forest stand volume (m³) 3 / ha); Forest stand area (m²) 2 / ha); The average height of the forest stand (m); , , These are model parameters; This is the error term.

[0005] Step 2: Acquisition and Preprocessing of High-Density UAV LiDAR Point Cloud Data After steps including denoising filtering, point cloud classification, DEM generation, and elevation normalization, normalized point cloud data for each sample plot were obtained. Using all echoes, corresponding point cloud feature variables were extracted for three variable groups: height, density, and vertical structure. Specifically, the calculation formulas for the canopy leaf area density-related variables and the vertical branch-leaf profile-related variables in the vertical result variable group are as follows: ; ; ; ; ; ; ; ;

[0006] In the formula: For the first Leaf area index at each height level For gap ratio, Extinction coefficient, For height intervals. For a height range, For the first The gap ratio of each height layer.

[0007] Step 3: Derivation of Forest Parameter Estimation Model and Variable Selection (1) Based on the allometric growth equation model of stock volume, the structural formula of the forest parameter estimation model is derived. The specific process is as follows: a. Replacing the basal area (BA) in the allometric growth equation of forest stand volume with density indices (P) such as canopy closure and forest density, the equation can be expressed as:

[0008] In the formula: This represents the stand density variable. It can be derived and used... Alternative .

[0009] b. The other three forest parameters have a certain relationship with the volume of timber. After derivation, the above formula can be transformed into:

[0010] In the formula: It can be the volume, or it can be the cross-sectional area, average diameter, or average height. For forest stand height variable.

[0011] c. Introducing another relevant variable reflecting the vertical structure information of the forest stand into the above formula, namely the vertical structure variable, we can derive the following structural formula for the forest parameter estimation model:

[0012] In the formula: Forest parameters such as DBH, H, BA, and VOL can be used. For stand density variable, For stand height variable, For forest stand vertical structure variables, , , These are model parameters; This is the error term.

[0013] (2) The point cloud feature variables were filtered, and the filtering results are as follows: Height variable group: point cloud mean height (hmean), 95th percentile height (hp95), standard deviation of point cloud height distribution (hstdv), and coefficient of variation (hcv); Density variable groups: canopy coverage (cc), dp50, dp75; Vertical structure variable group: mean leaf area density (LADmean) and its coefficient of variation (LADcv), mean of branch and leaf vertical profile (VFPmean), standard deviation (VFPstdv), and coefficient of variation (VFPcv).

[0014] Step 4: Building a Forest Parameter Estimation Model Based on the actual parameters of the sample plots and the characteristic variables of the sample plot lidar point clouds, a power-law model was constructed using a regular exhaustive method. The rules for variable selection and combination are as follows: the forest parameter estimation model structure consists of 3-5 variables. Each model should include any one of the primary height variables hmean and hp95, 1 to 2 density variables (if two density variables are selected, they should be composed of the primary density variable cc and a secondary density variable dp50 or dp75), and 1 vertical structure variable.

[0015] Step 5: Select the optimal model (1) SPSS was used to perform Pearson correlation analysis on the lidar point cloud variables and parameters, and significance analysis (F-test) was performed on the model. The Gauss-Newton iterative method was used to estimate the model parameters. The optimal solution of the parameters was found through iteration. Then, the accuracy of each forest parameter was estimated by the ten-fold cross-validation method. The three indicators of the model were calculated: coefficient of determination (R²), relative root mean square error (rRMSE), and mean prediction error (MPE) to evaluate the model fitting effect. Thus, the optimal estimation model of each forest parameter of subtropical eucalyptus plantation based on UAV lidar technology was obtained. Among them, the calculation formula of the evaluation index used in the model accuracy evaluation is as follows: ; ; ; *100 In the formula: For the observed values, For predicted values, The average of the observed values. For the sample size, Let t be the t-value at a confidence level of α, and SEE be the standard error of the predicted value.

[0016] (2) The model's adaptability was tested using the ten-fold cross-validation method. In addition to the three indicators used in the structural test for model fit evaluation, mean absolute error (MAE) and mean square error (MSE) were added to evaluate the model. If the model still maintains high accuracy and low error on the new dataset, then the model has good adaptability and can be extended to a wider range of forestry practices.

[0017] 3. Advantages of the present invention (1) Use of high-density UAV lidar point cloud data: Compared with the limitations of traditional airborne lidar point cloud data, which has a large range but low density and is difficult to realize three-dimensional reconstruction of forest stand structure, UAV lidar can flexibly and efficiently acquire high-precision, high-density, and large-range point cloud data, significantly improving the accuracy of forest parameter estimation models.

[0018] (2) Innovative application of exhaustive combination optimization method to explore synergistic effect and improve estimation accuracy: For the first time, the exhaustive combination optimization method is systematically applied to the forest parameter estimation of high-density UAV lidar data. By systematically exhaustively combining point cloud variables, the contribution of single variables is quantified and the synergistic gain effect between variables is captured. This breaks through the limitations of single variable analysis, effectively screens the optimal variable combination, significantly improves the model estimation accuracy, and provides an innovative strategy for parameter estimation in complex forest environments.

[0019] (3) Full-process multi-dimensional analysis ensures model reliability and generalization ability: A full-process analysis framework covering parameter estimation, model fitting, and multi-index testing is constructed, integrating statistical significance testing (F test) and accuracy evaluation indicators (R², rRMSE, MPE, MAE, MSE) to scientifically quantify the reliability of the model under complex forest conditions. The ten-fold cross-validation method is used to deeply verify the model's adaptability, ensuring the model's cross-scenario extrapolation ability and providing a reusable evaluation system for practical applications and similar studies.

[0020] (4) Construction of a dedicated estimation model to enable precise extraction and management of core parameters: In response to the unique vegetation structure of subtropical eucalyptus plantations, an innovative power model system is constructed to accurately invert core forest parameters such as average diameter (DBH), average height (H), basal area (BA), and volume (VOL) of the stand, providing scientific and effective technical support for the refined management, growth monitoring and sustainable operation of eucalyptus plantation resources. Attached Figure Description

[0021] Figure 1 Flowchart of a method for extracting parameters of subtropical eucalyptus plantations based on UAV lidar; Figure 2 Scatter plot of forest parameters from UAV lidar data; Figure 3 This is a residual distribution map of various forest parameters from UAV lidar data; Figure 4 Correlation diagram of forest parameters and lidar variables in eucalyptus plantation data from UAV lidar; Figure 5 Adaptability test diagram for the optimal estimation model of eucalyptus forest data by UAV lidar; Detailed Implementation

[0022] The features of the present invention will be further illustrated below through examples, but the claims of the present invention are not limited in any way.

[0023] The study area was selected from Gaofeng Forest Farm and its surrounding areas in Guangxi Zhuang Autonomous Region, and the sample plots were laid out following the principle of relatively uniform distribution. Laser rangefinders were used to determine the boundaries of the sample plots, and RTK technology was simultaneously employed to obtain their geographic coordinates. Each sample plot was set at 30 m × 20 m, arranged in a north-south direction, resulting in a total of 35 subtropical eucalyptus plantation sample plots established within the study area.

[0024] 1. Conduct the sample plot survey according to the method in step 1. Within each survey sample plot, measure the diameter at breast height (DBH) of trees with a DBH of not less than 5.0 cm. Simultaneously, use an ultrasonic height gauge to measure the average tree height and the height of the dominant tree. Record stand characteristic parameters such as average stand age and canopy closure at the same time. Based on the measured data, calculate the key parameters of the sample plot, including average diameter (DBH), average height (H), basal area (BA), and stand volume (VOL). The sample plot survey data are shown in Table 1.

[0025]

[0026] 2. Following the method in step 2, the UAV is controlled to perform terrain-following flight at a relative altitude of 200 meters and a speed of 32 meters per second to acquire high-density lidar point cloud data, with an average point cloud density of 250 points / square meter. Based on the sample land vector boundary determined in step 1, the corresponding lidar point cloud is cropped, followed by noise reduction filtering, point cloud classification, DEM generation, and elevation normalization processing to finally obtain normalized point cloud data for each sample area. Based on this data, using Lidar360 software or Python programming tools, target point cloud parameters are extracted from three types of feature variables—height, density, and vertical structure—based on the full echo signal.

[0027] 3. An estimation study was conducted on four core forest parameters of subtropical eucalyptus plantations: mean diameter (DBH), mean height (H), basal area (BA), and stand volume (VOL). Following the method described in step 3, a suitable forest parameter estimation model structure was derived, and 12 variables were selected from three groups of lidar point cloud feature variables with different characteristics for subsequent steps. In the height variable group, hmean and hp95 were identified as core variables to characterize the height characteristics of the vegetation community, while hstdv and hcv were used as auxiliary variables to reflect the discrete characteristics of height distribution. In the density variable group, cc was used as a core indicator to quantify canopy cover, while dp50 and dp75 were used as supplementary variables to further describe the spatial distribution characteristics of density parameters. All five parameters in the vertical structure variable group were defined as key variables for systematically analyzing the detailed characteristics of the vertical stratification structure of the vegetation.

[0028] 4. Based on the parameter adaptation variable system and model structure framework constructed in step 3, and following the exhaustive rules set in step 4, 70 sets of forest parameter estimation model expressions are generated through systematic variable combination operations. The main findings are summarized as follows: (1) ; (2) ; (3) ; (4) ; (5) ; (6) In the formula: It can be DBH, H, BA, or VOL. For the average height of the point cloud, It is at the 95th percentile. Canopy coverage, For hstdv or hcv, For dp50 or dp75, For vertical structure variables, , , , , , All of these are model parameters.

[0029] 5. Implement the framework using Python, based on the method in step 5, including index calculation, cross-validation, significance analysis, and correlation analysis, and finally output the optimal model of the four parameters and the three index values ​​for evaluating the accuracy of the model (Table 2).

[0030]

[0031] The coefficient of determination (R²) reflects the model's ability to fit the observed data; the closer the value is to 1, the stronger the explanatory power. The relative root mean square error (rRMSE) measures the difference between the predicted and observed values ​​as a percentage, facilitating comparison of the accuracy of different models. The mean prediction error (MPE) measures the prediction bias; a value close to zero indicates that the model has no systematic bias. Although the optimal model structures and R² values ​​for the four forest parameters of eucalyptus differ, they all have high R² values, and their rRMSE and MPE are also in similar good ranges, indicating that the models' generalization ability and systematic error control are both good and similar. A scatter plot is generated using the measured forest parameter values ​​from step 1 and the estimated values ​​from the optimal model. Figure 2The study observed that in the optimal model for each parameter of the eucalyptus forest, data points were densely distributed around a 1:1 line, indicating high consistency between estimated and measured values, demonstrating model reliability and high estimation accuracy. Furthermore, residual distribution plots were used to further reveal the differences between model estimates and measured values. Figure 3 In this study, most residuals were concentrated near zero, with a reasonable and random distribution, indicating no significant systematic bias; only a few large residuals suggested that the model might be overfitting. SPSS was used to evaluate the Pearson correlation between 12 lidar point cloud variables and various forest parameters. The correlation plot (…) Figure 4 The results show that each parameter is highly correlated with the height variable, with tree height H having a correlation of 0.955 with hp95, while the correlation with density variables (such as cc) is low. The parameters are negatively correlated with vertical structure variables LADmean, VFPmean, and VFPstd, and positively correlated with LADcv and VFPcv, all with high correlations. Although cc is not highly correlated with any parameter, the optimal model still includes this variable, indicating that the Pearson correlation coefficient is not the sole basis for variable selection; the synergistic effect between variables is equally crucial for model construction. Then, significance analysis (F-test, see Table 3) was conducted on each model. Generally, a p-value less than 0.05 is considered statistically significant. The p-values ​​of the optimal estimation models for each forest parameter are all less than 0.05, indicating that the selected point cloud feature variables are significantly correlated with the corresponding forest parameters, proving the effectiveness of the regression-power model established in this study.

[0032]

[0033] Finally, the model was tested for adaptability, and the optimal model adaptability test results for each forest parameter were obtained. Figure 5 The results show that the R-squared value of the fitness test is... 2 The value is lower than the R-value of previous evaluations of the structural accuracy of each model. 2 Value, eucalyptus forests have the highest VOL (volume level) R. 2 The value (0.85) indicates that the corresponding optimal estimation model performs well in terms of extrapolation and generalization of accumulation; next are DBH and H(R). 2 (0.81 and 0.78 respectively); R 2The lowest was BA (0.66). The rRMSE values ​​of all models varied within a certain range, indicating that the accuracy of the models may differ when estimating different data. All models performed relatively well in terms of average prediction error, with low MPE values, indicating the absence of significant systematic bias. In summary, these detailed analyses comprehensively demonstrate that the proposed forest parameter extraction method has high accuracy, low bias, and strong generalization ability in estimating forest parameters in eucalyptus plantations. The constructed optimal model achieves reliable estimation of core parameters through variable synergistic effects, and both statistical significance and error indices verify the scientific validity and practicality of the method.

[0034] It should be noted that the embodiments described above are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art and related fields based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention. Modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and should all be covered within the scope of the claims of the present invention.

Claims

1. Claim 1: A method for extracting parameters of subtropical eucalyptus plantations based on UAV lidar, characterized in that, Includes the following steps: (1) Sample plot survey and data calculation: a. Measure the diameter at breast height (DBH), dominant tree height, average age, and canopy closure of trees in each sample plot; b. Calculate actual forest parameters such as average diameter (DBH), average height (H), basal area (BA), and stand volume (VOL) of trees in each sample plot based on measured data; (2) Acquisition and preprocessing of high-density UAV lidar point cloud data: a. After steps such as noise reduction filtering, point cloud classification, DEM generation, and elevation normalization, normalized point cloud data of various plots are obtained; b. Extract the corresponding point cloud feature variables from the normalized lidar point cloud data for each location; (3) Based on the allometric growth equation model of stock volume, derive the structural formula of the forest parameter estimation model and screen the point cloud feature variables; (4) Based on the actual parameters of the sample plot and the characteristic variables of the lidar point cloud of the sample plot, a power model is constructed using a regular exhaustive method; (5) Conduct correlation analysis and model significance analysis on the model variables, and perform parameter estimation and accuracy evaluation on the model. Based on the selected evaluation indicators, select the optimal model for estimating forest parameters of eucalyptus plantations using UAV lidar. Finally, conduct an adaptability test on the optimal model for each forest parameter of eucalyptus plantations.

2. Claim 2: The method according to claim 1, characterized in that, The formula for calculating the stand volume (VOL) mentioned in step (1) is as follows: In the formula: Forest stand volume (m³) 3 / ha); Forest stand area (m²) 2 / ha); The average height of the forest stand (m); , , These are model parameters; This is the error term.

3. Claim 3: The method according to claim 1, characterized in that, The corresponding point cloud feature variables mentioned in step (2) are lidar point cloud variables extracted using all echoes, and the relevant calculation formulas for the extracted point cloud feature variables related to canopy leaf area density and vertical branch and leaf profile are as follows: ; ; ; ; ; ; ; ; In the formula: For the first Leaf area index at each height level For gap ratio, Extinction coefficient, For height intervals. For a height range, For the first The gap ratio of each height layer.

4. Claim 4: The method according to claim 1, characterized in that, The specific process of deriving the structural formula of the forest parameter estimation model in step (3) is as follows: (1) Based on the allometric growth equation of forest stand volume as described in claim 2, the basal area (BA) is replaced by density indices (P) such as canopy closure and forest density. The equation can be expressed as: In the formula: This represents the stand density variable. It can be derived and used... Alternative (2) The other three forest parameters have a certain relationship with the stock volume. After derivation, the above formula can be transformed into: In the formula: It can be the volume, or it can be the cross-sectional area, average diameter, or average height. For the stand height variable. (3) In the above formula, another relevant variable reflecting the vertical structure information of the stand is introduced, namely the vertical structure variable. Thus, the structural formula of the forest parameter estimation model is derived as follows: In the formula: Forest parameters such as DBH, H, BA, and VOL can be used. For stand density variable, For stand height variable, For forest stand vertical structure variables, , , These are model parameters; This is the error term.

5. Claim 5: The method according to claim 1, characterized in that, The point cloud variables selected in step (3) include: (1) height variable group: point cloud mean height (hmean), 95th percentile height (hp95), standard deviation of point cloud height distribution (hstdv) and coefficient of variation (hcv); (2) density variable group: canopy coverage (cc), dp50, dp75; (3) vertical structure variable group: leaf area density mean (LADmean) and its coefficient of variation (LADcv), branch and leaf vertical profile mean (VFPmean), standard deviation (VFPstdv) and coefficient of variation (VFPcv).

6. Claim 6: The method according to claim 1, characterized in that, The variable selection and combination rules used in step (4) are as follows: the forest parameter estimation model structure consists of 3-5 variables. Each model should include any one of the main height variables hmean and hp95, 1 to 2 density variables (if two density variables are selected, they should be composed of the main density variable cc and a secondary density variable dp50 or dp75) and 1 vertical structure variable.

7. Claim 7: The method according to claim 1, characterized in that, In step (5), the model variable analysis used SPSS to perform corresponding Pearson correlation analysis to quantitatively assess the correlation between the 12 lidar point cloud variables and each parameter. The model significance analysis mainly used the F-test to verify the effectiveness of the selected lidar variables in the estimation model, determine which variables contribute most to the model's prediction, and evaluate the model's predictive ability. Model parameter estimation used the Gauss-Newton iterative method to find the optimal solution for the parameters through iteration. In the model accuracy evaluation, the ten-fold cross-validation method was used to accurately assess the model accuracy. To evaluate the model's fit, three indicators were mainly calculated: the coefficient of determination (R²), the relative root mean square error (rRMSE), and the mean prediction error (MPE). The coefficient of determination (R²) reflects the model's ability to interpret the observed data, with a value range of [0,1]. A higher value indicates a stronger explanatory power. The relative root mean square error (rRMSE) is a relative measure of the difference between the observed value and the model's predicted value. It analyzes the percentage of estimation accuracy and helps in comparing different models and studies. The Mean Prediction Error (MPE) measures the degree of bias in predicted values; an MPE close to zero indicates that the model has no systematic bias. The model's fitness test verifies its generalization ability. A ten-fold cross-validation method is used, adding Mean Absolute Error (MAE) and Mean Squared Error (MSE) to the three indicators used in the structure test for model fit evaluation. The model's generalization ability can be assessed by testing its performance on independent datasets. If the model maintains high accuracy and low error on new datasets, it can be concluded that the model has good fitness and can be extended to a wider range of forestry practices. The calculation formulas for all evaluation indicators involved are as follows: ; ; ; *100 ； In the formula: For the observed values, For predicted values, The average of the observed values. For the sample size, Let t be the t-value at a confidence level of α, and SEE be the standard error of the predicted value.

Images