Multi-scale forest aboveground biomass estimation method based on single tree detection technology

Abstract

The invention discloses a multi-scale forest above-ground biomass estimation method based on an individual tree detection technology, relates to the technical field of individual tree detection, and solves the problems that in an existing AGB estimation method, the overall accuracy of area AGB estimation is influenced by uncertainty of an individual tree method, and the overall accuracy of area AGB estimation is influenced due to the fact that a model adopts a complex mechanism. And therefore, the accuracy of the prediction is influenced by the random residual error in the actual predicted value, and the precision of the estimation result is reduced, and the like. According to the method, AGB estimation from the individual tree scale to the sample plot scale is carried out by adopting an individual tree method; based on the EIV model, constructing an AGB upscaling estimation model from the sample plot scale to the forest stand scale; amplifying a forest stand scale reference AGB sample by adopting an EIV model and canopy height extracted by ICESat-2 / ATL08; estimating a high-resolution forest AGB by using a random forest model, and the like. According to the method, multi-scale AGB high-precision estimation from a single tree to a region can be realized, and the estimation efficiency and precision of the forest biomass are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of single tree detection, and in particular to a multi-scale forest aboveground biomass estimation method based on single tree detection technology. Background Art

[0002] As an important indicator for measuring forest carbon sinks and productivity, forest aboveground biomass (AGB) accounts for the largest proportion of forest biomass and is a key parameter for assessing forest carbon balance. At present, remote sensing technology (RS) is mainly used to estimate regional AGB. From the perspective of estimation, there are currently two main methods to calculate AGB: one is the surface method, which extracts features related to AGB from remote sensing data to establish a statistical model of AGB; the other is the single tree method, which calculates the biomass of a single tree and then accumulates the biomass of the single tree to obtain the biomass of the sample plot or stand.

[0003] Although the biomass estimation based on the Individual Tree-based Approach (ITA) is relatively precise, it conflicts with the demand for large-scale AGB estimation. At present, the application of the individual tree-based approach to AGB estimation is mainly concentrated in small areas, and large-scale AGB estimation mainly uses the surface method. Based on remote sensing data and the development trend from rough to fine, the realization of precision forestry is the main research trend in the future, that is, the application of the individual tree method in large-scale AGB estimation, but there are currently some problems such as algorithm accuracy and computer efficiency. It is quite difficult to apply the individual tree method to large-scale AGB estimation.

[0004] Compared with ordinary optical remote sensing, lidar remote sensing (Unmanned aerial vehicle laser scanning, ULS, handheld laser scanning, HLS) and satellite-borne laser radar (such as ICESat-2)) can obtain horizontal and vertical information of forests, and has obvious advantages in forest parameter extraction. In regional-scale forest AGB estimation, the ALT08 product of the satellite-borne laser radar ICESat-2 (Ice, Cloud, and Land Elevation Satellite-2) can provide massive forest canopy height data with high temporal resolution. As an efficient remote sensing tool, it is one of the new forces in forestry remote sensing in recent years and can provide an effective technical means for forest structure research and AGB estimation. Since the canopy height provided by ICESat-2 / ATL08 is discrete footprint point data, the existing studies mostly use the surface method (i.e., in units of sample plots), and use a relatively small number of samples to extrapolate it through machine learning models, and the estimation accuracy of large scales is difficult to guarantee.

[0005] There are a lot of uncertainties in the process of estimating AGB from the individual tree scale and plot scale to the regional scale. These uncertainties come from many factors, such as sampling errors, individual tree parameter errors, conversion coefficient errors, estimation model errors, and remote sensing data errors. Among them, the uncertainty based on the individual tree method mainly comes from the uncertainty of predicting individual tree biomass using the allometric growth model (i.e., prediction uncertainty). This uncertainty gradually accumulates with the expansion of spatial scale, and may eventually significantly affect the overall accuracy of regional biomass estimation.

[0006] In the process of regional-scale AGB estimation based on the surface method, the constructed regression model strives to fit the target variable as accurately as possible through complex mechanisms. However, the actual predicted value not only contains trend components, but also comes with certain random residuals. The existence of residuals often affects the accuracy of the prediction, thereby reducing the accuracy of the estimation results to a certain extent. Therefore, in the process of regional biomass estimation, how to effectively reduce and control uncertainty, especially the impact of residual errors, has become one of the key challenges to improve the accuracy of biomass estimation.

[0007] In summary, the present invention provides a multi-scale forest aboveground biomass estimation method based on single tree detection technology. Summary of the invention

[0008] The present invention aims to solve the problems in existing AGB estimation methods, such as the uncertainty of the single tree method affecting the overall accuracy of regional AGB estimation, and the random residuals in the actual predicted values ​​of the surface method affecting the accuracy of the prediction due to the complex mechanism adopted in the model, thereby reducing the accuracy of the estimation result, and the like. A multi-scale forest aboveground biomass estimation method based on single tree detection technology is provided.

[0009] A multi-scale forest aboveground biomass estimation method based on single tree detection technology is implemented by the following steps:

[0010] Step 1: Use the single-tree method to estimate AGB from the single-tree scale to the sample plot scale;

[0011] Step 2: Based on the EIV model, construct an AGB upscaling estimation model from the plot scale to the forest scale;

[0012] Step 3: Use the EIV model and the canopy height H extracted by ICESat-2 / ATL08 ICE Amplifying the stand scale reference AGB samples; dividing the amplified stand scale reference AGB samples into a training set and a validation set;

[0013] Step 4: Use the RF-EBK model to estimate high-resolution forest AGB.

[0014] Furthermore, the specific process of step one is:

[0015] Step 1: Obtain ULS data and HLS data and perform data preprocessing; at the same time, extract the canopy height H of the sample plot in the ULS data. ULS ;

[0016] Step 1 and 2: Perform ICP point cloud registration on the preprocessed data to achieve single tree segmentation of the fused point cloud data and obtain single tree parameter estimation (H, DBH); DBH and H are the single tree breast diameter and tree height respectively;

[0017] Step 13: Estimate the AGB at the individual tree scale based on the individual tree parameter estimation (H, DBH), and summarize the AGB at the individual tree scale in the plot to obtain the estimated value of the AGB at the plot scale.

[0018] Furthermore, in step 11, the data preprocessing includes data cropping, noise point removal, ground point cloud classification, normalization and point cloud fusion processing.

[0019] Furthermore, in steps one and two, the tree height H is determined by identifying the local maximum of the canopy height in the 1:1 matching of individual trees, and DBH is estimated by applying a nonlinear least squares circle fitting algorithm to the 1:1 matched trees at 1.3 m from the trunk.

[0020] Furthermore, in steps 1 and 3, the additive biomass model was used to estimate the AGB at the individual tree scale, and the AGB at the individual tree scale in the plot was summarized to obtain the AGB estimate at the plot scale.

[0021] Furthermore, in step 2, the AGB upscaling estimation model from the plot scale to the forest scale is expressed as follows:

[0022]

[0023] Where D and H are the average DBH and tree height of the plots measured by field surveys, respectively, and are considered as error-free variables. st,ITA is the stand-scale AGB estimated based on ITA, a1, b1, c1, a2, b2 are the model parameters to be fitted; ε1, ε2 are the model residuals.

[0024] Furthermore, in step 3, the canopy height H of ICESat-2 / ATL08 is used. ICE Substitute the canopy height H extracted from the ULS data in the EIV model ULS , to estimate the stand-scale aboveground biomass B in the ICESat-2 / ATL08 footprint st,ICE ; It is expressed as follows:

[0025]

[0026] The stand-scale aboveground biomass B st,ICE The AGB samples were used as the stand scale reference to train the RF model.

[0027] Furthermore, in step 4, the regional continuous forest AGB is estimated using the RF model to calculate the prediction deviation of the validation set; the prediction deviation of the validation set is interpolated using the EBK method to generate a continuous residual map;

[0028] The residual map is added to the prediction deviation result of the RF model to correct the prediction deviation, and finally the corrected regional continuous forest AGB is obtained.

[0029] Beneficial effects of the present invention:

[0030] In the method of the present invention, the pre-processed ULS and HLS fused point cloud data (U-HLS) is segmented into individual trees, and the aboveground biomass of individual trees is estimated using an additive biomass model based on the extracted individual tree parameters. The AGB of the detected individual trees is accumulated to calculate the AGB at the plot scale.

[0031] The method of the present invention introduces the Errors-in-Variables (EIV) model and uses the canopy height extracted by ULS as a medium to construct an AGB upscaling estimation model from the plot scale to the forest scale. In view of the good consistency between the canopy height extracted by ICESat-2 / ATL08 and the canopy height data extracted by ULS, the constructed EIV model is used to realize the estimation of the ICESat-2 / ATL08 footprint scale reference AGB. This process significantly increases the sample size in the AGB estimation, thereby laying a more solid foundation for the regional scale biomass estimation. In order to further improve the accuracy of regional scale AGB estimation, the Random Forest model (RF) is finally combined with the Empirical Bayesian Kriging interpolation method (EBK) to optimize the prediction residuals and ensure the accuracy of regional scale AGB. Through the above steps, the present invention can achieve high-precision estimation of multi-scale AGB from single trees to regions, greatly improving the estimation efficiency and accuracy of forest biomass. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a flow chart of a multi-scale forest aboveground biomass estimation method based on single tree detection technology according to the present invention;

[0033] Figure 2 Flow chart of AGB estimation from single tree scale to plot scale based on ITA;

[0034] Figure 3 is a schematic diagram of the sample amplification method;

[0035] Figure 4 The schematic diagram of the high-precision estimation of large-scale AGB using the RF-EBK model. DETAILED DESCRIPTION

[0036] Combination Figures 1 to 4 The present embodiment is described as follows: a multi-scale forest aboveground biomass estimation method based on single tree detection technology is implemented by the following steps:

[0037] Step 1: Estimation of AGB from single tree scale to plot scale based on ITA; the specific process is as follows:

[0038] Step 11: Obtain ULS data through the UAV laser radar ULS and obtain HLS data through the handheld laser radar HLS, and perform data preprocessing on the ULS data and the HLS data; at the same time, extract the canopy height of the sample plot from the ULS data to obtain the canopy height H of the sample plot ULS ;

[0039] In this embodiment, the data preprocessing includes data cropping, noise point removal, ground point cloud classification, normalization and point cloud fusion processing.

[0040] Step 1 and 2: Use the point cloud segmentation (PCS) algorithm to perform ICP point cloud registration on the preprocessed data to achieve single tree segmentation and single tree parameter estimation (H, DBH) of the fused point cloud data; wherein DBH and H represent the single tree breast diameter and tree height, respectively.

[0041] In this implementation, tree height H is determined by identifying the local maximum of canopy height in 1:1 matching of individual trees, and DBH is estimated by applying a nonlinear least squares circle fitting algorithm at 1.3 m on the trunk of the 1:1 matched trees.

[0042] Step 1: Based on the estimated single tree parameters (H, DBH), the additive biomass model was used to estimate the AGB at the single tree scale, and the AGB at the scale of all single trees in the plot was summarized to obtain the estimated value of AGB at the plot scale.

[0043] In this embodiment, it also includes a single-tree scale AGB reference value obtained based on the sample plot survey data and a sample plot scale AGB reference value obtained by summarizing the AGB of all single-tree scales in the sample plot, and the sample plot scale AGB reference value is used to verify the accuracy of the sample plot scale AGB estimate obtained based on the ITA method.

[0044] Step 2: Introduce the EIV model and calculate the canopy height H of the sample plot extracted in step 1. ULS , construct an AGB upscaling estimation model from plot scale to forest scale;

[0045]

[0046] Where D and H represent the average DBH and tree height of the plots measured by field surveys, respectively, and are considered as error-free variables. ULS is the canopy height extracted from the ULS data and is considered as a variable with error. st,ITA is the stand-scale AGB estimated based on ITA and is also considered as a variable with error. i ,b i ,c i (i=1,2,3) are the model parameters that need to be fitted, and ε j (j=1,2) is the model residual.

[0047] Step 3: Use the EIV model and the canopy height H extracted by ICESat-2 / ATL08 ICE Amplify the stand scale reference AGB samples; divide the amplified stand scale reference AGB samples into a training set and a validation set;

[0048] like Figure 3 As shown in Figure 2, the canopy height extracted by ICESat-2 / ATL08 is defined as the 98th percentile canopy height. ICE Canopy height H extracted from ULS ULS With good consistency, the canopy height H of ICESat-2 / ATL08 ICE Replace the canopy height H of ULS ULS , to estimate the stand-scale aboveground biomass B in the ICESat-2 / ATL08 footprint st,ICE ; It is expressed as follows:

[0049]

[0050] This stand-scale aboveground biomass (B st,ICE ) as stand-scale reference AGB samples can be used to construct and train a random forest model (RF), thereby achieving spatially continuous AGB estimation. Through this method, the AGB sample size has been significantly increased, providing a large amount of training data for large-scale AGB estimation, further supporting the construction and optimization of the random forest model.

[0051] Step 4: Use the RF-EBK model that combines the RF model with the empirical Bayesian Kriging (EBK) interpolation method to estimate high-resolution forest AGB.

[0052] like Figure 4 As shown in the figure, first, feature variables are extracted from satellite remote sensing image data (Landsat 8 OLI), SRTM data (ShuttleRadar Topography Mission), forest cover (Tree cover) and forest type (Forest type) data, and importance analysis is performed: that is, the key features that have the greatest influence on model prediction are screened out through the recursive feature elimination method, and the RF-EBK model is constructed.

[0053] In this implementation, the RF model is used to estimate the regional continuous forest AGB to calculate the prediction deviation of the validation set; the empirical Bayesian Kriging (EBK) method is used for interpolation processing to generate a continuous residual map of the prediction deviation of the validation set.

[0054] Subsequently, the prediction deviation of the interpolated validation set was added to the prediction result of the RF model to correct the prediction deviation, and finally the corrected regional continuous forest AGB was obtained, that is, a more accurate AGB estimate.

[0055] The method described in this embodiment, by introducing the EIV model, constructs an AGB upscaling estimation model from the sample plot scale to the forest stand scale, taking into account the sample plot biomass estimated based on the single tree detection technology and the error. At the same time, the canopy height data extracted by ICESat-2 / ATL08 is converted into a reference AGB sample at the footprint level scale. This process can better capture the characteristics of forest biomass under different environmental conditions and reduce the deviation caused by insufficient sample size. By introducing a large number of samples, the model can reflect a wider range of ecological changes, thereby improving its representativeness and accuracy. In addition, this method makes full use of the spatial autocorrelation of the residuals and uses the empirical Bayesian Kriging interpolation method to model the prediction residuals of the random forest model. Through this method, the spatial information contained in the remote sensing data is fully excavated, so that the residuals are refined. Specifically, the interpolation results are combined with the prediction results of the random forest, which not only improves the accuracy of the prediction, but also makes the model more adaptable.

[0056] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0057] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A multi-scale forest aboveground biomass estimation method based on single tree detection technology is characterized by: The method is implemented by the following steps: Step 1: Use the single-tree method to estimate AGB from the single-tree scale to the sample plot scale; Step 2: Based on the EIV model, construct an AGB upscaling estimation model from the plot scale to the forest scale; Step 3: Use the EIV model and the canopy height H extracted by ICESat-2 / ATL08 ICE Amplifying the stand scale reference AGB samples; dividing the amplified stand scale reference AGB samples into a training set and a validation set; Step 4: Use the RF-EBK model to estimate high-resolution forest AGB.

2. The multi-scale forest aboveground biomass estimation method based on single tree detection technology according to claim 1 is characterized by: The specific process of step one is: Step 1: Obtain ULS data and HLS data and perform data preprocessing; at the same time, extract the canopy height H of the sample plot in the ULS data. ULS ; Step 1 and 2: Perform ICP point cloud registration on the preprocessed data to achieve single tree segmentation of the fused point cloud data and obtain single tree parameter estimation (H, DBH); where DBH and H represent the breast diameter and tree height of a single tree, respectively; Step 13: Estimate the AGB at the individual tree scale based on the individual tree parameter estimation (H, DBH), and summarize the AGB at the individual tree scale in the plot to obtain the estimated value of the AGB at the plot scale.

3. The multi-scale forest aboveground biomass estimation method based on single tree detection technology according to claim 2 is characterized by: In step 11, the data preprocessing includes data cropping, noise point removal, ground point cloud classification, normalization and point cloud fusion processing.

4. The multi-scale forest aboveground biomass estimation method based on single tree detection technology according to claim 2 is characterized by: In steps one and two, the tree height H is determined by identifying the local maximum of canopy height in 1:1 matching of individual trees, and DBH is estimated by applying a nonlinear least squares circle fitting algorithm at 1.3 m on the trunk of the 1:1 matched trees.

5. The multi-scale forest aboveground biomass estimation method based on single tree detection technology according to claim 2 is characterized by: In steps 1 and 3, the additive biomass model was used to estimate the AGB at the individual tree scale, and the AGB at the individual tree scale in the plot was summarized to obtain the AGB estimate at the plot scale.

6. The multi-scale forest aboveground biomass estimation method based on single tree detection technology according to claim 1 is characterized by: In step 2, the AGB upscaling estimation model from the plot scale to the forest scale is expressed as follows: In the formula, D and H are the average DBH and tree height of the plots measured by field surveys, respectively, and are considered as error-free variables. st,ITA is the stand-scale AGB estimated based on ITA; a1, b1, c1, a2, b2 are the model parameters to be fitted; ε1, ε2 are the model residuals.

7. The multi-scale forest aboveground biomass estimation method based on single tree detection technology according to claim 6 is characterized by: In step 3, the canopy height H of ICESat-2 / ATL08 is used. ICE Substitute the canopy height H extracted from the ULS data in the EIV model ULS , to estimate the stand-scale aboveground biomass B in the ICESat-2 / ATL08 footprint st,ICE ; It is expressed as follows: The stand-scale aboveground biomass B st,ICE The AGB samples were used as the stand scale reference to train the RF model.

8. The multi-scale forest aboveground biomass estimation method based on single tree detection technology according to claim 6 is characterized by: In step 4, the regional continuous forest AGB is estimated using the RF model to calculate the prediction deviation of the validation set; the prediction deviation of the validation set is interpolated using the EBK method to generate a continuous residual map; The residual map is added to the prediction deviation result of the RF model to correct the prediction deviation, and finally the corrected regional continuous forest AGB is obtained.

Images