Forest biomass estimation method combining polarization radar and satellite-borne laser radar

Abstract

The invention relates to a forest biomass estimation method combining a polarization radar and a satellite-borne laser radar. The method comprises the following steps: preprocessing discrete footprint point data and polarimetric SAR data to obtain geographic position information and a polarimetric SAR backscattering coefficient; according to a random forest interpolation algorithm in combination with geographic position information and a polarized SAR backscattering coefficient, a nonlinear mapping relation among space coordinates, polarized SAR features and spaceborne laser radar parameters is automatically established through machine learning, and continuous spaceborne laser radar features are generated; and calculating the relative importance of all the satellite-borne laser radar features and the polarimetric SAR features with the forest biomass, sequentially selecting the features ranked in the top according to a relative importance ranking result to construct a forest biomass prediction model, and realizing forest biomass prediction by using the forest biomass prediction model. By adopting the method, high-precision forest biomass estimation can be realized.

Description

Technical Field

[0001] The present application relates to the field of remote sensing information processing and application technology, and in particular to a forest biomass estimation method combining polarimetric radar and spaceborne lidar. Background Art

[0002] Forest biomass (Above Ground Biomass, AGB) directly assesses forest quality and carbon sequestration capacity. It is the dominant factor in the material and energy cycles of forest ecosystems. Accurately estimating forest biomass can deepen our understanding of the interactions within terrestrial ecosystems. Estimating forest biomass using remote sensing technology combined with machine learning algorithms can significantly improve survey efficiency and reduce costs compared to traditional manual surveys.

[0003] Spaceborne LiDAR (Light Detection and Ranging) systems, such as the Ice, Cloud and Land Elevation Satellite (ICESat-2), utilize satellite platforms with high orbits and wide fields of view. They can provide a wide range of three-dimensional surface structural parameters and have the potential to reveal the distribution and dynamics of forest biomass over large regions. However, due to the spatially discontinuous nature of spaceborne LiDAR data, coupling ICESat-2 data with other spatially continuous remote sensing data to achieve spatially continuous forest biomass distribution mapping is a major limitation in the current application of spaceborne LiDAR data. Therefore, developing a forest biomass estimation method that combines polarimetric radar and spaceborne LiDAR has significant application value. Summary of the Invention

[0004] Based on this, it is necessary to provide a forest biomass estimation method that combines polarimetric radar and spaceborne lidar to achieve high-precision forest biomass estimation in order to address the above technical problems.

[0005] A forest biomass estimation method combining polarimetric radar and spaceborne lidar, the method comprising: Obtain discrete footprint point data and polarimetric SAR data from the ICESat-2 spacecraft-borne lidar; preprocess the discrete footprint point data and polarimetric SAR data to obtain the geographic location information and polarimetric SAR backscatter coefficient corresponding to each spacecraft-borne lidar footprint point; Based on the random forest interpolation algorithm combined with geographic location information and polarimetric SAR backscatter coefficients, machine learning automatically establishes a nonlinear mapping relationship between spatial coordinates, polarimetric SAR features, and spaceborne lidar parameters to generate continuous spaceborne lidar features. The relative importance of all spaceborne lidar features and polarimetric SAR features to forest biomass was calculated. The top-ranked features were selected in order according to the relative importance ranking results to establish a model. The features that contributed more to the model were determined as modeling features based on the change in model error to construct a forest biomass prediction model. The forest biomass prediction model was used to realize forest biomass prediction.

[0006] The aforementioned forest biomass estimation method, combining polarimetric radar and spaceborne lidar, first acquires and preprocesses discrete ICESat-2 footprint data and polarimetric SAR data to obtain geolocation information and polarimetric SAR backscatter coefficients. The random forest interpolation algorithm breaks through the limitations of traditional linear assumptions. Using discrete footprints as the core, it combines geolocation information with polarimetric SAR features. Through machine learning, it establishes a nonlinear mapping relationship to generate continuous spaceborne lidar features. This algorithm addresses the adaptability issues of discrete point interpolation in heterogeneous environments and the spatial discontinuity of spaceborne lidar data, making it adaptable to complex forest environments. It leverages the advantages of both data types. The direct measurement advantage of spaceborne lidar provides three-dimensional structural information for biomass estimation, while the full coverage of polarimetric SAR compensates for the spatial limitations of the former. The fusion of the two provides comprehensive forest information. The lidar interpolation features are then combined with the polarimetric SAR features to calculate the relative importance of the spaceborne lidar and polarimetric SAR features to forest biomass. The top features are sequentially selected for modeling, and key features are identified based on the error, avoiding redundant interference and improving model accuracy. The forest biomass prediction model finally constructed can accurately reflect the relationship between biomass and characteristics, effectively solving the problem of insufficient adaptability of traditional interpolation methods in complex forest environments. At the same time, it fully utilizes the direct measurement advantages of spaceborne lidar and the full coverage characteristics of polarimetric SAR, thereby realizing large-scale, high-precision continuous spatial inversion of forest biomass. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 1 is a flow chart of a forest biomass estimation method combining polarimetric radar and spaceborne lidar in one embodiment; Figure 2 A schematic diagram of a random forest spatial interpolation method flow in one embodiment; Figure 3 A schematic diagram of optimizing different functions in one embodiment; Figure 3 (a) is a schematic diagram of the random forest spatial interpolation results. Figure 3 (b) to (d) are schematic diagrams of inverse distance weighted interpolation (IDW), Kriging interpolation (Kriging) and radial basis function (RBF) interpolation respectively; Figure 4 A comparison diagram of forest biomass estimation results in another embodiment; Figure 4(a) is the forest biomass estimate predicted by the model using only ALOS-2 features. Figure 4 (a) The forest biomass estimation results of the XGBoost model constructed after feature optimization. DETAILED DESCRIPTION

[0008] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0009] In one embodiment, Figure 1 As shown, a forest biomass estimation method combining polarimetric radar and spaceborne lidar is provided, comprising the following steps: Step 102: Obtain discrete footprint point data and polarimetric SAR data of the ICESat-2 satellite-borne lidar; preprocess the discrete footprint point data and polarimetric SAR data to obtain geographic location information and polarimetric SAR backscatter coefficient corresponding to each satellite-borne lidar footprint point.

[0010] Spaceborne lidar (such as ICESat-2) can provide a wide range of three-dimensional surface structural parameters, offering the advantage of direct measurement. Processing these discrete data to generate continuous features fully leverages these advantages, providing a reliable three-dimensional structural basis for forest biomass estimation. Polarimetric SAR data offers comprehensive coverage and provides a rich set of surface features. Acquiring polarimetric SAR data and fusing it with spaceborne lidar features fully leverages the advantages of polarimetric SAR and compensates for the spatial discontinuity of spaceborne lidar data. The combination of these two complements each other, providing more comprehensive forest information for biomass estimation.

[0011] In step 104, a nonlinear mapping relationship between spatial coordinates, polarimetric SAR features, and spaceborne lidar parameters is automatically established through machine learning based on the random forest interpolation algorithm combined with the geographic location information and the polarimetric SAR backscatter coefficient, thereby generating continuous spaceborne lidar features.

[0012] Traditional spatial interpolation methods are limited by linear assumptions and are difficult to adapt to complex forest environments. Using the random forest interpolation algorithm, based on geographic location information and polarimetric SAR backscatter coefficients, this method automatically establishes a nonlinear mapping between spatial coordinates, polarimetric SAR signatures, and spaceborne lidar parameters, thereby generating continuous spaceborne lidar signatures. This approach overcomes linear limitations and better handles the complex and changing nature of forest environments. It effectively addresses the spatial discontinuity of spaceborne lidar data acquisition, transforming discrete footprint data into a continuous signature distribution, paving the way for subsequent spatially continuous forest biomass mapping.

[0013] Step 106, calculate the relative importance of all spaceborne lidar features and polarimetric SAR features to forest biomass, select the top-ranked features in order according to the relative importance ranking results to establish a model, determine the features that contribute more to the model according to the change in model error as modeling features to construct a forest biomass prediction model, and use the forest biomass prediction model to achieve forest biomass prediction.

[0014] The relative importance of spaceborne lidar and polarimetric SAR features to forest biomass was calculated, and the top-ranked features were selected in order of importance to establish a model. The features that contributed most to the model were then determined as modeling features based on changes in model error. This approach avoids interference from irrelevant or redundant features, ensuring that the model focuses on key features and improving model accuracy. Using the selected features to construct a forest biomass prediction model can more accurately reflect the relationship between forest biomass and related features than traditional methods, thereby achieving large-scale, high-precision, continuous spatial inversion of forest biomass. This effectively addresses the lack of adaptability of traditional methods in complex forest environments and meets the need for accurate forest biomass estimation.

[0015] The aforementioned forest biomass estimation method, combining polarimetric radar and spaceborne lidar, first acquires and preprocesses discrete ICESat-2 footprint data and polarimetric SAR data to obtain geolocation information and polarimetric SAR backscatter coefficients. The random forest interpolation algorithm breaks through the limitations of traditional linear assumptions. Using discrete footprints as the core, it combines geolocation information with polarimetric SAR features. Through machine learning, it establishes a nonlinear mapping relationship to generate continuous spaceborne lidar features. This algorithm addresses the adaptability issues of discrete point interpolation in heterogeneous environments and the spatial discontinuity of spaceborne lidar data, making it adaptable to complex forest environments. It leverages the advantages of both data types. The direct measurement advantage of spaceborne lidar provides three-dimensional structural information for biomass estimation, while the full coverage of polarimetric SAR compensates for the spatial limitations of the former. The fusion of the two provides comprehensive forest information. The lidar interpolation features are then combined with the polarimetric SAR features to calculate the relative importance of the spaceborne lidar and polarimetric SAR features to forest biomass. The top features are sequentially selected for modeling, and key features are identified based on the error, avoiding redundant interference and improving model accuracy. The forest biomass prediction model finally constructed can accurately reflect the relationship between biomass and characteristics, effectively solving the problem of insufficient adaptability of traditional interpolation methods in complex forest environments. At the same time, it fully utilizes the direct measurement advantages of spaceborne lidar and the full coverage characteristics of polarimetric SAR, thereby realizing large-scale, high-precision continuous spatial inversion of forest biomass.

[0016] In one embodiment, pre-processing the discrete footprint data and polarimetric SAR data includes: Perform data quality checks on discrete footprint point data, remove outliers, and extract spaceborne lidar footprint points; Perform radiometric calibration and terrain correction on polarimetric SAR data and calculate polarimetric SAR backscatter coefficients; The coordinate systems of all data are unified, the spaceborne lidar footprint points are spatially matched with the polarimetric SAR data, and the latitude and longitude and polarimetric SAR feature information corresponding to each spaceborne lidar footprint point are extracted.

[0017] In one embodiment, a nonlinear mapping relationship between spatial coordinates, polarimetric SAR features, and spaceborne lidar parameters is automatically established through machine learning based on a random forest interpolation algorithm combined with geographic location information and polarimetric SAR backscatter coefficients to generate continuous spaceborne lidar features, including: The target variable is set to the extracted spaceborne lidar features, and the input variables are geographic location information and polarimetric SAR backscatter coefficient. The trained random forest model is used to integrate multiple decision trees for spatial interpolation to predict the spaceborne lidar features of each grid point and output a continuous spaceborne lidar parameter distribution map.

[0018] In one embodiment, spatial interpolation is performed by integrating multiple decision trees using a trained random forest model, including: Using the trained random forest model to integrate multiple decision trees for spatial interpolation, the interpolation result of a single tree is:

[0019] in, For the The predicted spaceborne lidar features for each grid point, For the The prediction function of a decision tree, is the input feature vector, which contains the geographic location information and polarimetric SAR backscatter coefficient.

[0020] In one embodiment, the method further comprises: The overall prediction result is expressed as

[0021] in, is the total number of decision trees.

[0022] In one embodiment, the method further comprises: During the interpolation process, each tree recursively splits the feature space and selects the optimal split feature and threshold To minimize the mean square error, thereby reducing the error of the prediction results; Minimize the mean square error

[0023] in, and is the mean of the target variables of the left and right nodes after splitting, is the true value of the training sample.

[0024] In one embodiment, the method further comprises: The variance of the interpolation result is calculated to measure the confidence of the prediction and evaluate the prediction result. The reliability of the output space-borne lidar feature can be evaluated according to the confidence. The prediction confidence can be expressed as .

[0025] In one embodiment, the relative importance is represented by the average gain, and the expression of the average gain is defined as:

[0026] in, Indicates the characteristics of space-borne lidar or polarimetric SAR, Indicates in Samples in the feature The first derivative (gradient) on , Indicates in Samples in the feature The second derivative (gradient) on ; It is a regularization parameter used to control the size of node weights to prevent overfitting and is a complexity penalty term.

[0027] In one embodiment, a forest biomass prediction model is constructed based on the optimally obtained polarization rotation domain features as modeling features, including: The forest biomass prediction model is constructed based on the optimal polarization rotation domain characteristics as the modeling features:

[0028]

[0029]

[0030] in, For the The modeling features of samples, For the The true value of the sample, for The prediction results, For the The function of a decision tree, is the number of trees to be constructed, is the number of training samples; is the regularization term, is the number of leaf nodes, is the number of leaf node splits; and are the control coefficients to prevent overfitting.

[0031] In one embodiment, the predicted forest biomass value and the actual measured forest biomass are determined by the coefficient of determination. , root mean square error and mean absolute error Conduct model accuracy evaluation and use the evaluation results to guide the optimization of forest biomass prediction models; The evaluation index is calculated as follows:

[0032]

[0033]

[0034] in, and are the model prediction values, is the mean of the measured values, is the sample size.

[0035] In a specific embodiment, Figure 2 The following is a flowchart of the random forest spatial interpolation method. First, ICESat-2 is used to extract lidar features and footprint latitude and longitude coordinates, and polarimetric SAR data is used to extract backscatter coefficients. A feature matrix is ​​constructed based on the parameters extracted from the data source. The spaceborne lidar feature to be predicted is used as the target variable, and other features are used as input features. A training set and a test set are randomly selected to construct a random forest model for interpolation training. The spaceborne lidar feature results are predicted for each grid point, and a continuous distribution map of the spaceborne lidar parameters is output and evaluated.

[0036] Figure 3 This is a comparison chart of interpolation results of spaceborne lidar features. The remote sensing data used are ALOS-2 polarimetric SAR data (including HH and HV backscatter coefficients) and ICESat-2 spaceborne lidar data (including 7060 footprints). Taking the 98% forest canopy height feature from the spaceborne lidar feature as an example, Figure 3 (a) is the random forest spatial interpolation result, Figure 3 (b)–(d) show the inverse distance weighted (IDW) interpolation, kriging interpolation, and radial basis function (RBF) interpolation, respectively. The random forest spatial interpolation results achieve a more reasonable range and spatial distribution of spaceborne lidar feature values ​​compared to other typical interpolation methods. The random forest spatial interpolation error is 3.17 m, while the errors of other methods range from 6.01 to 7.09 m, significantly improving the interpolation performance. This method effectively addresses the issue of discontinuous spatial coverage of spaceborne lidar data and fully leverages the synergistic advantages of multi-source remote sensing data.

[0037] Figure 4 This chart compares forest biomass estimates. The measured forest survey data were collected in 2019 in natural forest reserves in the Tibet Autonomous Region. The data include 841 forest biomass measurements. The measured forest biomass ranged from 100 mg / ha to 250 mg / ha, with an average of 143.06 mg / ha. Figure 4 (a) The polarimetric SAR features used are the backscatter coefficients (HH and HV) and their texture features, as well as the ICESat-2 satellite-borne lidar data after random forest spatial interpolation. After feature optimization, an XGBoost model is constructed to estimate forest biomass. After adding ICESat-2 features, Figure 4(b) Compared with the model using only ALOS-2 features, the accuracy is significantly improved, and the root mean square error is reduced by 38.6%. The prediction results of the present invention are consistent with the actual forest biomass, verifying the effectiveness of the present invention in forest biomass estimation applications.

[0038] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0039] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0040] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above 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.

[0041] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and such modifications and improvements are intended to fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A forest biomass estimation method combining polarimetric radar and spaceborne lidar, characterized in that: The method comprises: Obtaining discrete footprint point data and polarimetric SAR data of the ICESat-2 satellite-borne lidar; preprocessing the discrete footprint point data and polarimetric SAR data to obtain geographic location information and polarimetric SAR backscatter coefficient corresponding to each satellite-borne lidar footprint point; Based on the random forest interpolation algorithm combined with geographic location information and polarimetric SAR backscatter coefficients, machine learning automatically establishes a nonlinear mapping relationship between spatial coordinates, polarimetric SAR features, and spaceborne lidar parameters to generate continuous spaceborne lidar features. The relative importance of all spaceborne lidar features and polarimetric SAR features to forest biomass is calculated. The top-ranked features are selected in order according to the relative importance ranking results to establish a model. The features that contribute more to the model are determined as modeling features based on the change in model error to construct a forest biomass prediction model. The forest biomass prediction model is used to achieve forest biomass prediction.

2. The method according to claim 1, characterized in that Preprocessing the discrete footprint point data and polarimetric SAR data includes: Performing data quality check on the discrete footprint point data, removing outliers, and extracting spaceborne lidar footprint points; performing radiation calibration and terrain correction on the polarimetric SAR data and calculating the polarimetric SAR backscatter coefficient; The coordinate systems of all data are unified, the spaceborne lidar footprint points are spatially matched with the polarimetric SAR data, and the latitude and longitude and polarimetric SAR feature information corresponding to each spaceborne lidar footprint point are extracted.

3. The method according to claim 1, characterized in that Based on the random forest interpolation algorithm combined with geographic location information and polarimetric SAR backscatter coefficients, machine learning automatically establishes a nonlinear mapping relationship between spatial coordinates, polarimetric SAR features, and spaceborne lidar parameters, generating continuous spaceborne lidar features, including: The target variable is set to the extracted spaceborne lidar features, and the input variables are geographic location information and polarimetric SAR backscatter coefficient. The trained random forest model is used to integrate multiple decision trees for spatial interpolation to predict the spaceborne lidar features of each grid point and output a continuous spaceborne lidar parameter distribution map.

4. The method according to claim 3, characterized in that Use the trained random forest model to perform spatial interpolation by integrating multiple decision trees, including: Using the trained random forest model to integrate multiple decision trees for spatial interpolation, the interpolation result of a single tree is: in, For the The predicted spaceborne lidar features for each grid point, For the The prediction function of a decision tree, is the input feature vector, which contains the geographic location information and polarimetric SAR backscatter coefficient.

5. The method according to claim 4, wherein The method further comprises: The overall prediction result is expressed as in, is the total number of decision trees.

6. The method according to claim 4, characterized in that The method further comprises: During the interpolation process, each tree recursively splits the feature space and selects the optimal split feature and threshold Minimize the mean square error to allow the decision tree to learn the characteristics of the input feature vector; The minimized mean square error is in, and is the mean of the target variables of the left and right nodes after splitting, is the true value of the training sample.

7. The method according to claim 4, wherein The method further comprises: The variance of the interpolation result is calculated to measure the prediction confidence and evaluate the prediction result. The satellite-borne lidar feature is output according to the evaluation result. The prediction confidence is 。 8. The method according to claim 1, characterized in that The relative importance is characterized by the average gain, and the expression of the average gain is defined as: in, Indicates the characteristics of space-borne lidar or polarimetric SAR, Indicates in Samples in the feature The first derivative on , Indicates in Samples in the feature The second derivative on ; is the regularization parameter.

9. The method according to claim 1, wherein A forest biomass prediction model is constructed based on the optimally selected polarization rotation domain features as modeling features, including: The forest biomass prediction model is constructed based on the optimal polarization rotation domain characteristics as the modeling features: in, For the The modeling features of samples, for The true value of for The prediction results, For the The function of a decision tree, is the number of trees to be constructed, is the number of training samples; is the regularization term, is the number of leaf nodes, is the number of leaf node splits; and are the control coefficients to prevent overfitting.

10. The method according to claim 1, characterized in that The method further comprises: The predicted forest biomass value and the actual measured forest biomass were compared by the coefficient of determination. , root mean square error and mean absolute error Conduct model accuracy evaluation and use the evaluation results to guide the optimization of forest biomass prediction models; The evaluation index is calculated as follows: in, and are the model prediction values, is the mean of the measured values, is the sample size.

Images