Mountain forest leaf area index inversion method based on ground-based laser point cloud data

Abstract

The invention discloses a mountain forest leaf area index inversion method based on ground-based laser point cloud data. The method comprises the following steps: step 1, setting a plurality of stations by using a ground-based laser radar to collect dense laser point cloud data of a mountain forest area; 2, point cloud data registration, cutting, ground point filtering and height normalization preprocessing are carried out; step 3, performing point cloud voxelization, and calculating an effective leaf area index eLAI by counting a collision probability; step 4, correcting an aggregation effect, estimating a canopy aggregation index through a Pielou separation index method PCS, and correcting an effective leaf area index eLAI to obtain a leaf area index LAI; and step 5, verification is carried out in different forest scenes. According to the method, the influence of topographic relief and the aggregation effect on estimation of the mountain forest leaf area index is considered at the same time, high precision can be kept under the scenes of different vegetation structures and different topographic conditions, and compared with a traditional inversion method, the method has better universality and higher inversion precision.

Description

Technical Field

[0001] The invention relates to the field of forest leaf area index inversion, and in particular to a mountain forest leaf area index inversion method based on ground-based laser point cloud data. Background Art

[0002] Leaf Area Index (LAI) is a key parameter for describing the structure and function of forest canopies, and can directly reflect the productivity and health of forests. Accurate measurement and inversion of LAI is of great significance for understanding and predicting changes in forest ecosystems and formulating scientific forest management policies (Tang et al., 2016). Traditional methods for measuring LAI mainly rely on field measurements. Although these methods can provide high accuracy, they face many challenges in practical applications. In recent years, with the rapid development of remote sensing technology, LAI inversion methods based on remote sensing data have gradually become a research hotspot. Remote sensing technology can quickly obtain surface information over a large range, is not restricted by terrain, and can effectively make up for the shortcomings of traditional measurement methods (Zheng & Moskal, 2009). Terrestrial Laser Scanning (TLS) has the ability to obtain fine three-dimensional structural information because of its data collection on the ground. It has unique advantages in monitoring forest ecosystems and provides new possibilities for the inversion of LAI.

[0003] The methods used to invert LAI based on ground-based lidar mainly include the inversion method based on gap rate and the voxel method based on collision probability (Wang & Fang, 2020). At present, many scholars at home and abroad have used ground-based lidar data to invert LAI. For example, Takeda et al. obtained voxels by dividing the forest horizontally and vertically, calculated the gap rate and leaf area for each voxel, and determined the three-dimensional distribution of the canopy leaf area. The vertical distribution results of the leaf area obtained have a good correlation with the true value, but the leaf area density at the top of the canopy is slightly overestimated (Takeda et al., 2008). Chen et al. used the true leaf area index obtained by stratified sampling to verify the leaf area index inverted under different layer thicknesses and different voxel sizes, and determined the optimal combination of layer thickness and voxel size to accurately estimate LAI, but the versatility is low (Chen et al., 2022).

[0004] At present, some studies assume a simple random leaf distribution, and the resulting LAI value is actually the effective LAI. However, the aggregation state of vegetation is very common in forests (Fang, 2021; García et al., 2015), including aggregation within leaf clusters, leaf aggregation within the crown, and aggregation between crowns. When two vegetation canopies have the same LAI, the canopy with a clustered distribution will contain more gaps than the canopy with a random distribution. Therefore, when using indirect optical methods to correctly estimate LAI, it is crucial to consider the clumping effect. Not considering the clumping index (CI) will result in an LAI error of up to 70% (Fang, 2021). Previous methods for calculating the clumping index from ground-based lidar data mainly include the finite length averaging method (LX), the gap size distribution method (CC), the correction method combining the finite length averaging method and the gap size distribution method (CLX), and the path length distribution method (Kuusk et al., 2018). For example, Zhu et al. estimated the leaf angle distribution of deciduous and coniferous forests to obtain the extinction coefficient, calculated the aggregation index by the gap size distribution method, and further improved the classification method of wood components to obtain the wood ratio. The above parameters were used to correct the LAI and VFP (Vertical Foliage Profile), which corrected the underestimation caused by the aggregation effect and the height brought by the wood component, and effectively improved the estimation accuracy of LAI and VFP (Zhu et al., 2018).

[0005] Therefore, existing research methods have made significant progress in LAI inversion. However, due to the undulating terrain and complex and changeable canopy structure of mountain forests, the accuracy and versatility of LAI inversion are still lacking. Summary of the invention

[0006] Purpose of the invention: The present invention provides a mountain forest leaf area index inversion method based on ground-based laser point cloud data, which solves the problem that the traditional LAI inversion method has low accuracy and versatility and does not meet actual needs.

[0007] Technical solution: The method for inverting the leaf area index of mountain forests based on ground-based laser point cloud data of the present invention comprises the following steps:

[0008] Step 1: Use ground-based laser radar to set up multiple stations to collect dense laser point cloud data in mountain forest areas;

[0009] Step 2: Preprocess the laser point cloud data;

[0010] Step 3: voxelize the point cloud and calculate the effective leaf area index (eLAI) by statistically analyzing the collision probability;

[0011] Step 4: Correct the aggregation effect, estimate the canopy aggregation index by using the Pielou separation index method PCS, and correct the eLAI to obtain the LAI;

[0012] Step 5: Verify in different forest scenarios.

[0013] Furthermore, in step 1, the specific steps of setting up multiple stations to collect dense laser point cloud data in mountain forest areas using ground-based LiDAR are as follows: determine the sample area that needs TLS scanning, and set up 9 TLS instruments around the area to minimize occlusion and ensure the integrity of the point cloud data.

[0014] Furthermore, in step 2, the laser point cloud data is preprocessed by registration, cropping, ground point filtering, and height normalization.

[0015] Furthermore, in step 2, point cloud data registration, cropping, ground point filtering, and height normalization preprocessing specifically include the following steps:

[0016] Step 21, accurately register the TLS point cloud data of the 9 stations, align them to the same coordinate system, and crop the point cloud of the sample area;

[0017] Step 22: The registered and cropped point cloud data is filtered using the CSF cloth simulation filtering method to classify the point cloud into vegetation points and ground points.

[0018] Step 23: normalize the height of the filtered ground points and extract the vegetation point cloud of the sample area.

[0019] Furthermore, in step 3, the point cloud is voxelized, and the eLAI is calculated by statistically analyzing the collision probability, which specifically includes the following steps:

[0020] Step 31, voxelizing vegetation point cloud;

[0021] Step 32: Calculate the probability of laser interception based on the voxelized point cloud model;

[0022] Step 33: According to the probability of laser beam being intercepted, the eLAI calculation formula of the k-th horizontal layer is:

[0023] l(k)=α(θ)N(k)

[0024] Where l(k) represents the leaf area volume density of the kth horizontal layer, α(θ) represents the correction factor of the laser beam direction and the leaf inclination angle, assuming that the canopy leaves in the plot are randomly distributed, the correction factor of the leaf inclination angle is taken as 1.1;

[0025] Step 34: Accumulate the eLAI of each horizontal layer to obtain the total eLAI of the plot. The calculation expression is as follows:

[0026]

[0027] Furthermore, in step 31, the vegetation point cloud voxelization is specifically as follows: determining the range of the sample point cloud, taking the minimum value (X min , Y min , Z min ) is used as the starting point of voxel segmentation, and the point cloud data is segmented into i×j×k grids with a voxel size of Δi×Δj×Δk. The voxelization calculation formula is:

[0028]

[0029] Among them, (i, j, j) is the number of voxels in the X, Y, and Z axis directions of the point cloud data, int is the rounding function, (x max ,y max ,z max ) is the maximum value of the sample point cloud bounding box, (x min ,y min ,z min ) is the minimum value of the bounding box of the sample point cloud, and (Δi, Δj, Δk) is the voxel size in the X, Y, and Z axis directions.

[0030] Furthermore, in step 32, the probability of laser interception is calculated based on the voxelized point cloud model as follows: if there is a point cloud distributed in the voxel, it means that the laser beam is intercepted, and it is marked as a point cloud voxel; otherwise, it is a canopy gap voxel, indicating that the laser beam is not intercepted. After the sample point cloud is voxelized, it is divided into k layers in the Z-axis direction. By counting the laser collision probability of each horizontal layer after voxel stratification, the ratio of voxels with point clouds and without point clouds is calculated to obtain the gap rate of the layer:

[0031]

[0032] Among them, N k represents the collision probability of the laser beams at the kth level, n I (k) represents the number of laser interceptions at the kth level, that is, the number of voxels with point clouds, n T (k) represents the total number of voxels in the k-th level layer.

[0033] Further, in step 4, the aggregation effect is corrected, and the canopy aggregation index is estimated by the Pielou separation index method PCS, and the eLAI is corrected to obtain the LAI. Specifically, the aggregation index is calculated by the Pielou separation index method (PCS), which is a coefficient used to estimate the degree of separation between two co-growing plants. This method provides a randomness measure for each species relative to other species, where a and b are the probabilities of encountering species A and species B in two species populations, respectively. Assuming a+b=1, for two randomly distributed populations, there is a 95% probability:

[0034]

[0035] in, and is the maximum likelihood estimate of A and B, μ a and μ b is the average length of occurrences of A and B, and is the variance of a and b; if PCS<1, the species are clustered, if PCS=1, the species are randomly distributed, and if PCS>1, the species are uniformly distributed;

[0036] Correcting eLAI to obtain the LAI of the sample site, the correction expression is:

[0037] LAI=eLAI / λ PCS .

[0038] Furthermore, in step 5, verification was carried out in different forest scenarios. The three-dimensional radiation transfer model and computer simulation technology were used to construct forest scenes with different vegetation types (including broad-leaved forests and coniferous forests) and terrain conditions (including flat land, sloping land, and irregular undulating terrain). The TLS point cloud data were simulated and the LAI of the sample plot was estimated according to the above method. The data was compared with the LAI provided by the scenario to verify its estimation accuracy.

[0039] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: In the estimation of mountain forest leaf area index (LAI), the present invention fully considers the dual effects of terrain undulation and vegetation aggregation effect, thereby being able to significantly improve the accuracy and stability of LAI inversion under complex vegetation structure and changeable terrain conditions; compared with traditional LAI inversion methods, the present invention not only has stronger versatility and adaptability, but also can significantly improve the accuracy of estimation, reduce deviation, and provide more reliable data support for ecological monitoring and forest resource management. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram of the method flow of the present invention.

[0041] Figure 2 It is a scatter plot of the LAI accuracy results of the LAI estimation method of the present invention in all scenarios. DETAILED DESCRIPTION

[0042] like Figure 1 As shown, a mountain forest leaf area index inversion method based on ground-based laser point cloud data includes the following steps:

[0043] Step 1: Use ground-based laser radar to set up multiple stations to collect dense laser point cloud data in mountain forest areas;

[0044] Step 2: Preprocessing of point cloud data such as registration, cropping, ground point filtering, and height normalization;

[0045] Step 3: voxelize the point cloud and calculate the eLAI by counting the collision probability;

[0046] Step 4: Correct the aggregation effect, estimate the canopy aggregation index by the Pielou separation index method (PCS), and correct the eLAI to obtain the LAI;

[0047] Step 5: Verify this method in different forest scenarios.

[0048] Preferably, the specific steps of collecting point cloud data of mountain forest areas by ground-based lidar in step 1 are: determining the sample area that needs TLS scanning, setting up 9 TLS instruments around the area to minimize occlusion and ensure the integrity of the point cloud data.

[0049] In order to eliminate the interference of factors other than terrain and vegetation structure, the present invention uses a three-dimensional radiation transfer model combined with computer simulation technology to verify the above calculation method. First, a virtual forest scene library containing 480 different forest scenes was constructed, and the scene information is shown in Table 1. TLS point cloud data was simulated, and the accuracy of the above TLS inversion method was verified using the scene true value. The results can reflect the accuracy and versatility of LAI inversion of this method in different scenes.

[0050] Table 1 Virtual forest scene information

[0051]

[0052] Preferably, the specific steps of preprocessing such as point cloud data registration, cropping, ground point filtering, height normalization, etc. in step 2 are:

[0053] Step 2.1, accurately register the TLS point cloud data of the 9 stations, align them to the same coordinate system, and crop the point cloud of the sample area;

[0054] Step 2.2, perform CSF ground point filtering on the registered and cropped point cloud data, and classify the point cloud into vegetation points and ground points;

[0055] Step 2.3: Normalize the height of the filtered ground points and extract the vegetation point cloud of the sample area.

[0056] Preferably, the specific steps of voxelizing the point cloud in step 3 and calculating the eLAI by statistically analyzing the collision probability are as follows:

[0057] Step 3.1: Voxelize vegetation point cloud. Determine the size of the sample point cloud, and use the minimum value of the three-dimensional coordinates XYZ of the bounding box of the sample point cloud (X min , Y min , Z min ) is used as the starting point of voxel segmentation, and the point cloud data is segmented into i×j×k grids with a voxel size of Δi×Δj×Δk. The voxelization calculation formula is:

[0058]

[0059] Among them, (i, j, k) is the number of voxels in the X, Y, and Z axis directions of the point cloud data, int is the rounding function, (x max ,y max ,z max ) is the maximum value of the sample point cloud bounding box, (x min ,y min ,z min ) is the minimum value of the bounding box of the sample point cloud, (Δi, Δj, Δk) is the voxel size in the X, Y, and Z axis directions;

[0060] Step 3.2, calculate the probability of laser interception based on the voxelized point cloud model. If there are point clouds distributed in the voxel, it means that the laser beam is intercepted, and it is marked as a point cloud voxel. Otherwise, it is a canopy gap voxel, indicating that the laser beam is not intercepted. After the sample point cloud is voxelized, it is divided into k layers in the Z-axis direction. By counting the laser collision probability of each horizontal layer after voxel stratification, the ratio of voxels with point clouds and without point clouds is calculated to obtain the gap rate of this layer:

[0061]

[0062] Among them, N k represents the collision probability of the laser beams at the kth level, n I (k) represents the number of laser interceptions at the kth level, that is, the number of voxels with point clouds, n T (k) represents the total number of voxels in the k-th level;

[0063] Step 3.3: According to the probability of laser beam being intercepted, the eLAI calculation formula of the k-th horizontal layer is:

[0064] l(k)=α(θ)N(k)

[0065] Where l(k) represents the leaf area volume density of the kth horizontal layer, α(θ) represents the correction factor of the laser beam direction and the leaf inclination angle, assuming that the canopy leaves in the plot are randomly distributed, the correction factor of the leaf inclination angle is taken as 1.1;

[0066] Step 3.4: Accumulate the eLAI of each horizontal layer to obtain the total eLAI of the plot. The calculation expression is as follows:

[0067]

[0068] Preferably, in step 4, the aggregation effect is corrected, the canopy aggregation index is estimated by PCS, and the eLAI is corrected to obtain the LAI. The specific steps are as follows:

[0069] The clustering index was calculated by the gap-step method (PCS), which is a coefficient used to estimate the degree of separation between two co-growing plants. This method provides a measure of randomness for each species relative to other species, where a and b are the probabilities of encountering species A and species B in the two populations, respectively, assuming a + b = 1. For two randomly distributed populations, there is a 95% probability that:

[0070]

[0071] in, and is the maximum likelihood estimate of A and B, μ a and μ B is the average length of occurrences of A and B, and is the variance of a and b. If PCS < 1, the species are clustered, if PCS = 1, the species are randomly distributed, and if PCS > 1, the species are uniformly distributed;

[0072] Correcting eLAI to obtain the LAI of the sample site, the correction expression is:

[0073] LAI=eLAI / λ PCS

[0074] Preferably, the specific steps for verifying this method in different forest scenes in step 5 are: using a three-dimensional radiation transfer model and computer simulation technology to construct forest scenes with different vegetation types (including broad-leaved forests, coniferous forests) and terrain conditions (including flat land, sloping land, and irregular undulating terrain), simulating to obtain TLS point cloud data, estimating the LAI of the sample plot according to the above method, and comparing it with the LAI provided by the scene to verify its estimation accuracy.

[0075] To evaluate the above calculation method, we chose R 2(coefficient of determination), RMSE (root mean square error) and nRMSE (normalized root mean square error) were used as evaluation indicators. 2 The mathematical formulas for , RMSE and nRMSE are as follows:

[0076]

[0077] Among them, i represents the sequence number of the i-th sample point, Represents the predicted value of the i-th sample point, y i represents the true value of the i-th sample point, represents the average value of the true value of all sample points, m represents the number of sample points, and mean(y) represents the average value of the true value of the sample points.

[0078] The PCS clustering index estimation method is used to estimate the clustering index of the plot and perform eLAI correction to obtain LAI. The accuracy of the estimated LAI result is evaluated using the reference true value of the virtual plot. Considering the influence of terrain undulation on the accuracy of TLS inversion LAI, the present invention compares the estimation accuracy of the PCS method under different vegetation types and different slope conditions. The results are shown in Table 2. It can be seen that the PCS index correction of the clustering effect can obtain a better estimation accuracy (R 2 =0.9061 (broad-leaved forest), R 2 =0.8907 (coniferous forest)). Although the LAI estimation method of the present invention is reduced as the terrain becomes more rugged, it can still maintain a good accuracy (R 2 >0.7350). The LAI estimation results under all scenarios are summarized as follows: Figure 2 As shown (R 2 =0.9295), which can meet the actual needs.

[0079] Table 2 Summary of LAI estimation accuracy under different vegetation types and terrain conditions

[0080]

[0081] In the table, Slope0-Slope50 indicates that the terrain is a regular terrain with a slope of 0°-50°, Slopeterrain1 and Slopeterrain2 are two types of irregular terrain, and All indicates the summary of all terrain scenes.

Claims

1. A mountain forest leaf area index inversion method based on ground-based laser point cloud data, characterized in that: The steps include: Step 1: Use ground-based laser radar to set up multiple stations to collect dense laser point cloud data in mountain forest areas; Step 2: Preprocess the laser point cloud data; Step 3: voxelize the point cloud and calculate the eLAI by counting the collision probability; Step 4: Correct the aggregation effect, estimate the canopy aggregation index by using the Pielou separation index method PCS, and correct the eLAI to obtain the LAI; Step 5: Verify in different forest scenarios.

2. The mountain forest leaf area index inversion method based on ground-based laser point cloud data according to claim 1, characterized in that: In step 1, the specific steps of setting up multiple stations to collect dense laser point cloud data in mountain forest areas using ground-based LiDAR are as follows: determine the sample area that needs TLS scanning and set up 9 TLS instruments around the area.

3. The mountain forest leaf area index inversion method based on ground-based laser point cloud data according to claim 1, characterized in that: In step 2, the laser point cloud data is preprocessed by registration, cropping, ground point filtering, and height normalization.

4. The mountain forest leaf area index inversion method based on ground-based laser point cloud data according to claim 1, characterized in that: In step 2, point cloud data registration, cropping, ground point filtering, and height normalization preprocessing specifically include the following steps: Step 21, accurately register the TLS point cloud data of the 9 stations, align them to the same coordinate system, and crop the point cloud of the sample area; Step 22: The registered and cropped point cloud data is filtered using the CSF cloth simulation filtering method to classify the point cloud into vegetation points and ground points. Step 23: normalize the height of the filtered ground points and extract the vegetation point cloud of the sample area.

5. The mountain forest leaf area index inversion method based on ground-based laser point cloud data according to claim 1, characterized in that: In step 3, the point cloud is voxelized and the eLAI is calculated by statistically analyzing the collision probability. The specific steps include: Step 31, voxelizing vegetation point cloud; Step 32, calculating the probability of laser interception based on the voxelized point cloud model; Step 33: According to the probability of laser beam being intercepted, the eLAI calculation formula of the k-th horizontal layer is: l(k)=α(θ)N(k) Where l(k) represents the leaf area volume density of the kth horizontal layer, α(θ) represents the correction factor of the laser beam direction and the leaf inclination angle, assuming that the canopy leaves in the plot are randomly distributed, the correction factor of the leaf inclination angle is taken as 1.1; Step 34: Accumulate the eLAI of each horizontal layer to obtain the total eLAI of the plot. The calculation expression is as follows:

6. The mountain forest leaf area index inversion method based on ground-based laser point cloud data according to claim 5, characterized in that: In step 31, the vegetation point cloud voxelization is specifically as follows: determine the range of the sample point cloud, and use the minimum value (X min , Y min , Z min ) is used as the starting point of voxel segmentation, and the point cloud data is segmented into i×j×k grids with a voxel size of Δi×Δj×Δk. The voxelization calculation formula is: Among them, (i, j, k) is the number of voxels in the X, Y, and Z axis directions of the point cloud data, int is the rounding function, (x max ,y max ,z max ) is the maximum value of the sample point cloud bounding box, (x min ,y min ,z min ) is the minimum value of the bounding box of the sample point cloud, and (Δi, Δj, Δk) is the voxel size in the X, Y, and Z axis directions.

7. The mountain forest leaf area index inversion method based on ground-based laser point cloud data according to claim 5, characterized in that: In step 32, the probability of laser interception is calculated based on the voxelized point cloud model as follows: if there are point clouds distributed in the voxel, it means that the laser beam is intercepted, and it is marked as a point cloud voxel; otherwise, it is a canopy gap voxel, indicating that the laser beam is not intercepted; After the sample point cloud is voxelized, it is divided into k layers in the Z-axis direction. By counting the laser collision probability of each horizontal layer after voxel stratification, the ratio of voxels with point cloud and without point cloud is calculated to obtain the gap rate of the layer: Among them, N k represents the collision probability of the laser beams at the kth level, n I (k) represents the number of laser interceptions at the kth level, that is, the number of voxels with point clouds, n + (k) represents the total number of voxels in the k-th level layer.

8. The mountain forest leaf area index inversion method based on ground-based laser point cloud data according to claim 1, characterized in that: In step 4, the aggregation effect is corrected, and the canopy aggregation index is estimated by the Pielou separation index method PCS. The eLAI is corrected to obtain the LAI. Specifically, the aggregation index is calculated by the Pielou separation index method PCS, which is used to estimate the coefficient of the degree of separation between two co-growing plants. It provides a randomness measure for each species relative to other species, where a and b are the probabilities of encountering species A and species B in the two species populations, respectively. Assuming a+b=1, for two randomly distributed populations, there is a 95% probability: in, and is the maximum likelihood estimate of A and B, μ a and μ7 is the average length of occurrences of A and B, and is the variance of a and b; if PCS<1, the species are clustered, if PCS=1, the species are randomly distributed, and if PCS>1, the species are uniformly distributed; Correct eLAI to get the LAI of the sample site, the correction expression is: LAI=eLAI / λ 123 。 9. The mountain forest leaf area index inversion method based on ground-based laser point cloud data according to claim 1, characterized in that: In step 5, verification is carried out in different forest scenarios. Forest scenarios with different vegetation types and terrain conditions are constructed using a three-dimensional radiation transfer model and computer simulation technology. TLS point cloud data are simulated and the LAI of the sample plot is estimated according to the above method. The data is compared with the LAI provided by the scenario to verify its estimation accuracy.

10. The mountain forest leaf area index inversion method based on ground-based laser point cloud data according to claim 9, characterized in that: Different vegetation types include broad-leaved forests and coniferous forests, and terrain conditions include flat land, sloping land, and irregular undulating terrain.

Images