Forest aboveground biomass estimation method, device and equipment based on multi-source remote sensing data

By constructing a forest above-ground biomass model with multi-source remote sensing variable combination, importance combination, correlation combination and principal component analysis combination, the problem of unscientific selection of multi-source remote sensing variables in the existing technology is solved, and the rapid forest biomass is achieved , accurate estimation and efficient training of models.

CN118918477APending Publication Date: 2024-11-08CHINA UNIV OF MINING & TECH (BEIJING)

Patent Information

Application Number
CN202410998362.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-24
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

In the existing forest biomass assessment based on remote sensing, the selection of multi-source remote sensing variables lacks systematicity and scientificity, making it difficult to quickly build a forest biomass model under a combination of multiple variables, resulting in insufficient estimation efficiency and accuracy.

Method used

By obtaining forest sample data and its remote sensing images, a forest above-ground biomass model is constructed using multiple machine learning methods to combine multi-source remote sensing variables, importance combinations, correlation combinations and principal component analysis combinations, and the most accurate model is selected. To improve prediction accuracy and generalization capabilities.

Benefits of technology

It realizes rapid and accurate estimation of forest biomass, improves the training speed and scope of application of the model, and is suitable for research scenarios in different regions and tree species.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118918477A_ABST
    Figure CN118918477A_ABST
Patent Text Reader

Abstract

The invention provides a forest above-ground biomass estimation method, device and equipment based on multi-source remote sensing data. The method comprises the following steps: firstly, acquiring forest sample plot data and classifying tree species to determine sample plot biomass of different forest types, and meanwhile, reasonably classifying multi-source remote sensing image data to establish a single remote sensing variable; secondly, according to the sample plot biomass and a single remote sensing variable, constructing a forest above-ground biomass model by using multiple different machine learning methods, training the model, and screening out an optimal machine learning method from the model; then, constructing a multi-source remote sensing variable combination, a multi-source variable importance combination, a multi-source variable correlation combination and a multi-source variable principal component analysis combination according to the sample plot biomass and a single remote sensing variable; then, according to the sample plot biomass, the multi-source remote sensing variable combination, the multi-source variable importance combination, the multi-source variable correlation combination and the multi-source variable principal component analysis combination, the selected optimal machine learning method is used for constructing corresponding forest above-ground biomass models, and the models are subjected to precision comparison and model parameter adjustment; and finally, screening out the forest above-ground biomass model with the highest precision, and estimating the forest biomass by using the model. Through the method, various forest aboveground biomass models can be quickly compared and screened to screen out the model with the highest precision, and the method can be applied to different research areas and forest lands with different tree species.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of forest remote sensing monitoring, and in particular to a method, device and equipment for estimating forest aboveground biomass based on multi-source remote sensing data. Background Art

[0002] Using Google Earth Engine (GEE), an advanced remote sensing cloud computing platform, to estimate forest aboveground biomass over a large area using machine learning technology has become a common practice in the field of remote sensing. The GEE cloud platform brings together a large number of multi-source remote sensing images. At the same time, its powerful cloud computing capabilities can quickly call different machine learning methods and test forest aboveground biomass models under different variable combinations. Compared with traditional local stand-alone model testing, there is no need to download images and the model testing time can be greatly reduced. In the traditional construction process of forest aboveground biomass models based on remote sensing images, there is a lack of systematic and scientific multi-source remote sensing variable selection and model construction process.

[0003] In the existing remote sensing-based forest biomass assessment research, the multi-source remote sensing variables involved in the construction of forest biomass models are all selected according to personal wishes or previous experience, such as specific vegetation indexes or texture features, to participate in the construction of forest aboveground biomass models. There is no clear regulation on how to select the optimal multi-source remote sensing variables for forests, and there is no reasonable explanation for the number and combination of variables selected for the importance and correlation of multi-source variables. Most of them are modified according to the technical solutions given in previous literature to construct forest aboveground biomass models, and the selection of forest aboveground biomass model variables lacks rationality. In addition, the construction of forest aboveground biomass models based on local multi-source remote sensing variables mostly relies on local stand-alone workstations, which is not conducive to testing forest aboveground biomass models under a large number of multi-source remote sensing variable combinations. It is difficult to quickly achieve large-scale estimation of forest biomass under too many variables and multiple variable combinations. Summary of the invention

[0004] The present invention describes a method, device and equipment for estimating forest aboveground biomass based on multi-source remote sensing data, which can solve the above technical problems.

[0005] According to a first aspect, a method for estimating forest aboveground biomass based on multi-source remote sensing data is provided, the method comprising:

[0006] Obtain forest plot data and its corresponding remote sensing image data; classify tree species according to forest plot data to determine the biomass of the plot; select corresponding multi-source remote sensing image data to establish single remote sensing variables according to the forest plot data;

[0007] Using the biomass of the sample plot as the basic data and a single remote sensing variable as the feature, a variety of different machine learning methods were used to build a forest aboveground biomass model. Different machine learning methods were evaluated based on the model test results, and the machine learning method with the highest fit and the smallest error was selected as the optimal machine learning method.

[0008] Based on the biomass of the sample plot as the basic data and a single remote sensing variable as the feature, we constructed a combination of multi-source remote sensing variables, a combination of multi-source variable importance, a combination of multi-source variable correlation, and a combination of multi-source variable principal component analysis.

[0009] Taking the biomass of the sample plot as the basic data, the forest aboveground biomass model corresponding to each feature combination was constructed respectively by using the optimal machine learning method with the combination of multi-source remote sensing variables, the importance combination of multi-source variables, the correlation combination of multi-source variables and the principal component analysis combination of multi-source variables. By comparing the accuracy of the models and adjusting the model parameters, the forest aboveground biomass model with the highest accuracy was finally screened out and used to estimate the forest biomass.

[0010] In some embodiments, the indicators for evaluating different machine learning methods include, but are not limited to: model fit, root mean square error, mean absolute error, and relative error.

[0011] In some embodiments, the single remote sensing variable includes but is not limited to: image spectral bands, spectral and vegetation indexes, SAR, terrain, texture features, and tasseled cap transformation.

[0012] In some embodiments, a method for constructing a multi-source remote sensing variable combination includes:

[0013] Taking the sample plot biomass as the basic data and a single remote sensing variable as the feature, the optimal machine learning method is used to construct the forest aboveground biomass model; the forest aboveground biomass model corresponding to each single variable is tested separately, and the fitting degree and error value of the forest aboveground biomass model corresponding to each single variable are calculated. The models are sorted according to the calculation results of the fitting degree and error value, and the first two or more single variables with small error values ​​are selected to construct a multi-source remote sensing variable combination.

[0014] In some embodiments, the method for constructing a combination of importance of multi-source variables specifically includes: taking the sample plot biomass as the basic data and a single remote sensing variable as the feature, using the selected optimal machine learning method to perform importance analysis on the sample plot biomass, and sorting the importance of the single remote sensing variable in order from high to low, thereby constructing different combinations of importance of multi-source variables.

[0015] In some embodiments, a method for constructing a correlation combination of multiple source variables specifically includes: taking the biomass of the sample plot as the basic data and a single remote sensing variable as the feature, performing correlation analysis on the biomass of the sample plot respectively, and according to the correlation analysis results, sorting the correlation of the single remote sensing variable in descending order, thereby constructing different correlation combinations of multiple source variables.

[0016] In some embodiments, the correlation analysis method includes but is not limited to: Pearson product-moment correlation coefficient, Kendall rank correlation coefficient, Spearman rank correlation coefficient.

[0017] In some embodiments, a method for constructing a principal component analysis combination of multiple source variables specifically includes: taking the biomass of the sample plot as the basic data and a single remote sensing variable as the feature, performing principal component analysis on the biomass of the sample plots respectively, and taking the variables whose cumulative eigenvalues ​​of the principal component analysis exceed a certain threshold as the principal component analysis combination of multiple source variables.

[0018] According to a second aspect, a device for estimating forest aboveground biomass based on multi-source remote sensing data comprises:

[0019] Data acquisition module, used to obtain forest plot data and its corresponding multi-source remote sensing image data;

[0020] Data classification module, used to classify tree species based on forest plot data;

[0021] Model building module, used to build forest aboveground biomass models;

[0022] Model training module, used to train forest aboveground biomass model;

[0023] Model testing module, used to test forest floor biological models;

[0024] The model screening module is used to screen out the optimal model based on the test results of different models.

[0025] According to a third aspect, an electronic device is provided, comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, and wherein the processor implements any method described in the first aspect when executing the computer program.

[0026] The present invention provides a method, device and equipment for estimating forest aboveground biomass based on multi-source remote sensing data. Through scientifically constructing a forest aboveground biomass model based on multi-source remote sensing variables, respectively constructing forest aboveground biomass models according to multi-source variable combination, multi-source variable importance combination, multi-source variable correlation combination, and multi-source variable principal component analysis combination, the models are trained and tested, and the accuracy of the models is compared according to the training and test results of the models, and finally the forest aboveground biomass model with the highest accuracy is screened out. The construction of the multi-source variable combination can improve the accuracy and generalization ability of the model prediction, and improve the training speed of the model. The multi-error evaluation system of the forest aboveground biomass model constructed according to the present invention can quickly screen out the optimal forest aboveground biomass model for the research plot, and can be applied to research scenarios with a wider area and a variety of different tree species. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0028] Figure 1 A schematic flow chart of a method for estimating forest aboveground biomass based on multi-source remote sensing data provided in an embodiment of this specification is shown;

[0029] Figure 2 A schematic diagram of a process framework of a forest aboveground biomass estimation method based on multi-source remote sensing data provided in an embodiment of this specification is shown;

[0030] Figure 3 A distribution diagram of sample points in a research area provided in an embodiment of this specification is shown;

[0031] Figure 4 The following is a fitting result of a multi-source variable importance model provided by an embodiment of this specification;

[0032] Figure 5 A multi-source variable correlation model fitting result provided by an embodiment of this specification is shown;

[0033] Figure 6 The fitting result of a multi-source variable principal component analysis model provided in the embodiment of this specification is shown;

[0034] Figure 7 The fitting result of an optimal forest ground organism model provided by the embodiment of this specification is shown;

[0035] Figure 8The estimation result of forest biomass by an optimal forest aboveground organism model provided in the embodiments of this specification is shown;

[0036] Fig. 9 A schematic diagram of a forest aboveground biomass estimation device based on multi-source remote sensing data provided by an embodiment of this specification is shown;

[0037] Fig.10 A schematic diagram of a forest aboveground biomass estimation device based on multi-source remote sensing data provided in an embodiment of this specification is shown. DETAILED DESCRIPTION

[0038] The solution provided in this specification is described below in conjunction with the accompanying drawings.

[0039] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0040] In the description of the embodiments of the present application, words such as "exemplary", "for example" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary", "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary", "for example" or "for example" is intended to present related concepts in a concrete way.

[0041] In the description of the embodiments of the present application, the term "and / or" is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, B exists alone, and A and B exist at the same time. In addition, unless otherwise specified, the term "plurality" means two or more.

[0042] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. The terms "include", "comprises", "has" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0043] Figure 1 A schematic diagram showing the steps of the method for estimating forest aboveground biomass based on multi-source remote sensing data provided in the embodiment of this specification (refer to the flowchart diagram at the same time) Figure 2 ),like Figure 1 As shown, the method comprises the following steps:

[0044] Step S101, obtaining forest plot data and its corresponding remote sensing image data; classifying tree species according to the forest plot data to determine the biomass of the plot; and selecting corresponding multi-source remote sensing image data according to the forest plot data to establish single remote sensing variables respectively.

[0045] Step S1011, obtaining forest plot data and its corresponding remote sensing image data.

[0046] First, the target sample plot for study needs to be determined. Secondly, detailed tree information, such as tree species, diameter, height, crown width, etc., needs to be recorded in the sample plot. In addition, soil samples, vegetation coverage and other ecological parameters can also be collected. Finally, the exact geographic coordinates of each sample plot need to be recorded for later matching with remote sensing data.

[0047] According to the scope of the target plot, multi-source remote sensing image data such as Landsat, Sentinel, and MODIS are obtained in the Google Earth Engine cloud platform (Google Earth Engine (GEE) is a cloud-based platform for storing, processing, and analyzing earth science data sets, especially large-scale remote sensing images). If the target plot is a non-standard plot (10m*10m, 30m*30m), its exact size and shape can be recorded, and then the remote sensing image and the target plot are reprojected to unify the coordinates and resolution of the remote sensing image and the target plot.

[0048] Step S1012, classifying tree species according to the forest plot data to determine the biomass of the plot;

[0049] Tree species are classified according to forest plot data. For example, tree species can be divided into two levels. The first level classification is coniferous forest, broad-leaved forest and mixed forest; the second level classification is coniferous forest (Pinus tabulaeformis, Larix principis-rupprechtii, etc.) and broad-leaved forest (Poplar, willow, oak, etc.). After the tree species are classified, the biomass of the plot is determined based on the forest plot data after the tree species classification.

[0050] Exemplarily, this scheme selects the Huodong mining area within the Taiyue Mountain State Forest Farm in Shanxi Province as the target plot, and records detailed tree information, soil samples, vegetation coverage and other ecological parameters in the target plot; obtains multi-source remote sensing images such as Landsat in the GEE cloud platform (Google Earth Engine, an online planetary-level geospatial analysis platform provided by Google) according to the scope of the target plot; classifies the tree species in the target plot according to the forest plot data. In this example, the tree species are divided into coniferous forests, broad-leaved forests and mixed forests, and then the biomass of the plot is determined based on the forest plot data after tree species classification. Figure 3 This is the distribution map of sample points for this target sample plot.

[0051] Step S1013, according to the forest plot data, corresponding multi-source remote sensing image data are selected to respectively establish single remote sensing variables.

[0052] In one embodiment, the single remote sensing variable includes but is not limited to: image spectral bands, spectrum and vegetation index, SAR, terrain, texture features, and tasseled cap transformation.

[0053] Before obtaining a single remote sensing variable, the remote sensing image data needs to be preprocessed, including radiation correction, atmospheric correction, geometric correction and image registration, etc. Geographic Information System (GIS) software or programming languages ​​(such as Python's GDAL library or R language) can be used to extract the above variables. Among them, SAR variables refer to SAR (Synthetic Aperture Radar, a type of active microwave remote sensing technology). Specifically, SAR variables include: horizontal polarization echo (HH) and horizontal-vertical polarization echo (HV). Horizontal polarization echo refers to the direction of the electric field of the signal emitted and received by the radar is parallel to the ground, and horizontal-vertical polarization echo refers to the direction of the electric field of the signal emitted and received by the radar is perpendicular to each other.

[0054] Step S102, using the biomass of the sample plot as the basic data and a single remote sensing variable as the feature, using a variety of different machine learning methods to build a forest aboveground biomass model, evaluating different machine learning methods based on the model test results, and selecting the machine learning method with the highest fit and the smallest error as the optimal machine learning method.

[0055] The biomass of the sample plot is used as a data set, and the single remote sensing variable established in step S101 is used as a feature. Then, a plurality of different machine learning methods are used to construct a forest aboveground biomass model, a part of the data set is used to train the model, and the hyperparameters are adjusted to optimize the model performance, and a part of the data set is used to test the trained model, and the fit and error values ​​of different machine learning methods are calculated according to the test results of the model, and finally the machine learning method with the highest fit and the smallest error is selected as the optimal machine learning method. The machine learning methods include but are not limited to: random forest, gradient boosting decision tree, CART, and shortest distance method.

[0056] It is worth mentioning that the model construction process can be carried out on the GEE cloud platform (Google Earth Engine, an online planetary-level geospatial analysis platform provided by Google). The GEE cloud platform brings together a large number of multi-source remote sensing images. At the same time, its powerful cloud computing capabilities can quickly call different machine learning methods and test forest aboveground biomass models under different variable combinations. Compared with traditional local stand-alone model testing, there is no need to download images and the model testing time can be greatly reduced.

[0057] In one embodiment, the indicators for evaluating different machine learning methods include, but are not limited to: model fit, root mean square error, mean absolute error, and relative error.

[0058] For example, the Huodong mining area within the Taiyueshan State Forest Farm in Shanxi Province was selected as the target plot. Based on the biomass of the plot and the established single remote sensing variables (i.e., image spectral bands, spectral and vegetation indices, SAR, terrain, texture features, and tasseled cap transform), a variety of machine learning methods (i.e., random forest (RF), gradient boosting decision tree (Gradient tree boots), CART, and minimum distance method (Minimum Distance)) were used in the GEE cloud platform to construct a forest aboveground biomass model. The model fit (R 2 ), root mean square error (RMSE), mean absolute error (RAE) and relative error (RE) are used as evaluation indicators. Different machine learning methods are evaluated according to the test results of the model. The machine learning method with the highest fit and the smallest error is selected as the optimal machine learning method. The results show that the gradient boosting decision tree (Gradient treeboots) has the highest fit and the smallest error. The gradient boosting decision tree is used as the optimal machine learning method. The calculation results are shown in Table 1:

[0059]

[0060] Table 1

[0061] Step S103, taking the biomass of the sample plot as the basic data and a single remote sensing variable as the feature, constructing a multi-source remote sensing variable combination, a multi-source variable importance combination, a multi-source variable correlation combination, and a multi-source variable principal component analysis combination.

[0062] In one embodiment, a method for constructing a multi-source remote sensing variable combination includes:

[0063] Based on the biomass of the sample plots as the basic data and a single remote sensing variable as the feature, the forest aboveground biomass model was constructed using the selected optimal machine learning method.

[0064] The forest aboveground biomass model corresponding to each single variable was tested separately, and the model fitting degree and error value corresponding to each single variable were calculated. The models were sorted according to the calculation results of model fitting degree and error value, and the first two or more single variables with high fitting degree and small error value were selected to construct a multi-source remote sensing variable combination.

[0065] It should be noted that in the process of selecting the optimal machine learning method in step S102, this scheme uses multiple machine learning methods to construct a forest aboveground biomass model based on the sample plot biomass and a single remote sensing variable, including the optimal machine learning method, but the optimal machine learning method has not yet been screened out through comparison. In other words, the process of constructing a forest aboveground biomass model based on the sample plot biomass and a single remote sensing variable using the optimal machine learning method has been carried out in step S102. Therefore, in order to save computing resources, this step S103 can directly select the fitness and error value calculation results corresponding to the optimal machine learning method based on the calculation results in step S102.

[0066] Exemplarily, with the biomass of the sample plot as the basic data, and each single remote sensing variable, namely, the image spectral band (Band), spectrum and vegetation index (Index), SAR, terrain (Terrain), texture feature (GLCM), and tasseled cap transform (KT) as the feature, the optimal machine learning method selected in step S102, namely, the gradient boosting decision tree method, is used to construct a forest aboveground biomass model for each single remote sensing variable, and then the basic data is used to train the model. After training, the forest aboveground biomass model corresponding to each single variable is tested respectively, and the fitting degree and error value of the model corresponding to each single variable are calculated according to the test results (the index can be: model fitting degree (R 2 ), root mean square error (RMSE), mean absolute error (RAE) and relative error (RE), the model's fit and error calculation results are shown in Table 2), the models are sorted according to the model's fit and error value calculation results, and the first two or more single variables with high fit and small error values ​​are selected to construct a multi-source remote sensing variable combination. For example, the model fit (R2) and root mean square error (RMSE) can be used as indicators, and terrain (Terrain), texture features (GLCM) and SAR with high fit and small error values ​​are selected to construct a multi-source remote sensing variable combination. The above process can be performed on the GEE cloud platform. The final results of different multi-source remote sensing variable combinations are shown in Table 3.

[0067]

[0068] Table 2

[0069]

[0070]

[0071] Table 3

[0072] In one embodiment, a method for constructing a combination of importance of multiple source variables specifically includes:

[0073] Taking the sample plot biomass as the basic data and a single remote sensing variable as the feature, the optimal machine learning method is used to conduct an importance analysis on the sample plot biomass. According to the results of the importance analysis, the importance of the single remote sensing variable is ranked from high to low, thereby constructing different importance combinations of multi-source variables.

[0074] Taking the biomass of the sample plot as the basic data and all the single remote sensing variables as the features, the optimal machine learning method is used to construct the forest aboveground biomass model. The constructed forest aboveground biomass model is trained. After the training is completed, the importance score of the single variable in the feature is obtained, and the single variables are ranked according to the size of the importance score, so as to construct different importance combinations of multi-source variables.

[0075] It is easy to understand that this embodiment of ranking the importance of variables belongs to the category of feature engineering, and its purpose is to understand the importance distribution of a single variable among all features. It can not only improve the predictive ability of the model, but also provide the model with stronger interpretability and practicality.

[0076] Exemplarily, with the biomass of the sample plot as the basic data and all the single remote sensing variables, namely, the image spectral band (Band), spectrum and vegetation index (Index), SAR, terrain (Terrain), texture features (GLCM), and tasseled cap transform (KT) as features, the optimal machine learning method selected in step S102, namely, the gradient boosting decision tree method, is used to construct a forest aboveground biomass model, and then the model is trained using the basic data. After the training is completed, an importance analysis is performed on the single variables of the forest aboveground biomass model. The analysis results show that the order of importance of the single variables from large to small is: terrain (Terrain)>texture features (GLCM)>spectrum and vegetation index (Index)>spectral band (Band)>SAR>tasseled cap transform (KT). The above process can be carried out on the GEE cloud platform.

[0077] In some embodiments, the method for constructing a multi-source variable correlation combination specifically includes:

[0078] With the biomass of the sample plots as the basic data and a single remote sensing variable as the feature, correlation analysis was conducted on the biomass of the sample plots. According to the results of the correlation analysis, the correlation of the single remote sensing variable was ranked from high to low, thereby constructing different correlation combinations of multi-source variables.

[0079] In some embodiments, the correlation analysis method includes but is not limited to: Pearson product-moment correlation coefficient, Kendall rank correlation coefficient, Spearman rank correlation coefficient.

[0080] For each single forest aboveground biological variable, use statistical software or programming language (such as Python, R) to calculate the Pearson correlation coefficient, Kendall rank correlation coefficient or Spearman rank correlation coefficient between it and the biomass of the sample plot. According to the absolute value of the calculated correlation coefficient, all variables are sorted according to the strength of their correlation with the biomass of the sample plot. The larger the absolute value, the stronger the correlation. The Pearson correlation coefficient is used to calculate the correlation between all variables to obtain a correlation matrix between variables. The variance inflation factor (VIF) can be used or combined to detect multicollinearity. For variables with high VIF values, consider removing those that are highly correlated with other variables to reduce multicollinearity problems in the model, and the remaining single variables that have been sorted are constructed as different multi-source variable correlation combinations.

[0081] The above method can be carried out in the GEE cloud platform. Through the above method, the correlation between a single remote sensing variable and the biomass of the sample plot can be systematically analyzed, and different combinations of multi-source variable correlations can be constructed, so as to construct a model that is both explanatory and avoids multicollinearity.

[0082] In some embodiments, a method for constructing a principal component analysis combination of multiple source variables specifically includes:

[0083] Taking the biomass of the sample plots as the basic data and a single remote sensing variable as the characteristic, principal component analysis was performed on the biomass of the sample plots respectively, and the variables whose cumulative eigenvalues ​​of the principal component analysis exceeded a certain threshold were used as the principal component analysis combination of multi-source variables.

[0084] Exemplarily, a principal component analysis (PCA) is performed on a single remote sensing variable using statistical software or a programming language (such as Python's scikit-learn library or R language), and finally the variables with cumulative eigenvalues ​​of more than 97% obtained from the principal component analysis are constructed into a multi-source variable principal component analysis combination, and the eigenvalue calculation results corresponding to the single remote sensing variable are shown in Table 4.

[0085]

[0086] Table 4

[0087] Through the above method, PCA can be used to effectively extract key information from a single remote sensing variable and construct a principal component analysis combination of multiple source variables to build a biomass prediction model that can maintain data integrity and reduce computational complexity.

[0088] Step S104, taking the biomass of the sample plot as the basic data, and taking the multi-source remote sensing variable combination, the multi-source variable importance combination, the multi-source variable correlation combination, and the multi-source variable principal component analysis combination as the features, use the selected optimal machine learning method to respectively construct the forest aboveground biomass model corresponding to each feature combination, and finally screen out the forest aboveground biomass model with the highest accuracy by comparing the model accuracy and adjusting the model parameters, and use the model to estimate the forest biomass.

[0089] Exemplarily, the optimal machine learning method selected in step S102, i.e., the gradient boosting decision tree method, is used, and the multi-source remote sensing variable combination, the multi-source variable importance combination, the multi-source variable correlation combination, and the multi-source variable principal component analysis combination are respectively characterized, and the sample plot biomass is used as the basic data to construct a forest aboveground biomass model, and the sample plot biomass data is used to train the forest aboveground biomass model, and the model parameters are appropriately adjusted according to the training status of the model. According to the training data results, four different evaluation indicators, i.e., model fit, root mean square error, mean absolute error, and relative error are used to evaluate each model. Among them, the fit and error calculation results of the trained multi-source remote sensing variable combination are shown in Table 5 (the specific meaning of the combination number is shown in Table 3), and the model fitting results corresponding to the trained multi-source variable importance combination, multi-source variable correlation combination, and multi-source variable principal component analysis combination are shown in Table 5. Figure 4 , 5 , 6; by comparing the fit and error size of each model, the forest aboveground biomass model with the highest accuracy was finally screened out. The screening process can be carried out on the GEE platform. The error calculation results of each model are shown in Table 6. The fitting results of the optimal forest aboveground biological model finally screened out are shown in Figure 7 The estimated forest biomass of the study plots by the optimal forest aboveground organisms model is shown in Figure 8 shown.

[0090] The multi-error evaluation system of forest aboveground biomass model constructed according to the present invention can realize the accuracy comparison of forest aboveground biomass models under the combination of multi-source remote sensing variables, and can be used for selecting the optimal method for forest aboveground biomass of different tree species in different study areas; in addition, the GEE cloud computing platform can be used to quickly realize the comparison and optimization of forest aboveground biomass models using multiple machine learning methods, and can perform importance and correlation analysis of hundreds or thousands of remote sensing variables in parallel, quickly identify the correlation between variables and potential multicollinearity problems, and can perform rapid iterative testing of large-area multi-source remote sensing variables.

[0091]

[0092]

[0093] Table 5

[0094]

[0095] Table 6

[0096] Corresponding to the above method provided by the present invention, the present invention also provides a device. Figure 8 FIG. 2 is a schematic diagram showing a structure of a device for estimating forest aboveground biomass based on multi-source remote sensing data provided by an embodiment of this specification. Fig. 9 As shown, the device comprises:

[0097] The data acquisition module S901 is used to acquire forest plot data and its corresponding multi-source remote sensing image data;

[0098] The data classification module S902 is used to classify tree species according to the forest plot data;

[0099] Model building module S903, used to build a forest aboveground biomass model;

[0100] Model training module S904, used for training forest aboveground biomass model;

[0101] Model testing module S905, used to test forest ground organism models;

[0102] The model screening module S906 is used to screen out the optimal model according to the test results of different models.

[0103] It is worth mentioning that Fig. 9 The device shown is Figure 1 The method embodiments shown correspond to and can be applied to Figure 1 The acquisition of forest plot data and its corresponding multi-source remote sensing image data in the illustrated method embodiment, the classification of tree species based on forest plot data, the construction of forest aboveground biomass model, the training of forest aboveground biological model, the testing of forest aboveground biological model, and the screening of optimal model will not be repeated here. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of each example described in the embodiments disclosed herein, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0104] According to another embodiment, a computing device is also provided, such as Fig.10 As shown, it includes a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, the combination Figure 1 The method described.

[0105] According to another embodiment, there is also provided a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute a combination of Figure 1 The method described.

[0106] Those skilled in the art should be aware that in one or more of the above examples, the functions described in the present invention can be implemented by hardware, software, firmware or any combination thereof. When implemented by software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium.

[0107] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solution of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for estimating forest aboveground biomass based on multi-source remote sensing data, characterized in that: The method comprises: Obtain forest plot data and its corresponding remote sensing image data; classify tree species according to forest plot data to determine the biomass of the plot; select corresponding multi-source remote sensing image data to establish single remote sensing variables according to the forest plot data; Using the biomass of the sample plot as the basic data and a single remote sensing variable as the feature, a variety of different machine learning methods were used to build a forest aboveground biomass model. Different machine learning methods were evaluated based on the model test results, and the machine learning method with the highest fit and the smallest error was selected as the optimal machine learning method. Based on the biomass of the sample plot as the basic data and a single remote sensing variable as the feature, we constructed a combination of multi-source remote sensing variables, a combination of multi-source variable importance, a combination of multi-source variable correlation, and a combination of multi-source variable principal component analysis. Taking the sample plot biomass as the basic data, and taking the multi-source remote sensing variable combination, the multi-source variable importance combination, the multi-source variable correlation combination, and the multi-source variable principal component analysis combination as the features, the optimal machine learning method is used to construct the forest aboveground biomass model corresponding to each feature combination. By comparing the model accuracy and adjusting the model parameters, the forest aboveground biomass model with the highest accuracy is finally screened out, and this model is used to estimate the forest biomass.

2. The forest biomass estimation method based on multi-source remote sensing data according to claim 1, characterized in that: Metrics for evaluating different machine learning methods include, but are not limited to: model fit, root mean square error, mean absolute error, and relative error.

3. The forest biomass estimation method based on multi-source remote sensing data according to claim 1, characterized in that: Its single remote sensing variables include but are not limited to: image spectral bands, spectral and vegetation indices, SAR, terrain, texture features, and tasseled cap transformation.

4. The forest biomass estimation method based on multi-source remote sensing data according to claim 1, characterized in that: The method of constructing a combination of multi-source remote sensing variables includes: Based on the biomass of the sample plots as the basic data and a single remote sensing variable as the feature, the forest aboveground biomass model was constructed using the selected optimal machine learning method. The forest aboveground biomass model corresponding to each single variable was tested separately, and the model fitting and error value corresponding to each single variable were calculated. The models were sorted according to the calculation results of model fitting and error value, and the first two or more single variables with small error values ​​were selected to construct a multi-source remote sensing variable combination.

5. The forest biomass estimation method based on multi-source remote sensing data according to claim 1, characterized in that: The method of constructing the importance combination of multiple source variables includes: Taking the sample plot biomass as the basic data and a single remote sensing variable as the feature, the optimal machine learning method is used to conduct an importance analysis on the sample plot biomass. According to the results of the importance analysis, the importance of the single remote sensing variable is ranked from high to low, thereby constructing different importance combinations of multi-source variables.

6. The forest biomass estimation method based on multi-source remote sensing data according to claim 1, characterized in that: The method of constructing a correlation combination of multiple source variables specifically includes: With the biomass of the sample plots as the basic data and a single remote sensing variable as the feature, correlation analysis was conducted on the biomass of the sample plots. According to the results of the correlation analysis, the correlation of the single remote sensing variable was ranked from high to low, thereby constructing different correlation combinations of multi-source variables.

7. The method for constructing a multi-source variable correlation combination according to claim 6, characterized in that: The correlation analysis method includes but is not limited to: Pearson product-moment correlation coefficient, Kendall rank correlation coefficient, and Spearman rank correlation coefficient.

8. The forest biomass estimation method based on multi-source remote sensing data according to claim 1, characterized in that: The method of constructing a principal component analysis combination of multiple source variables specifically includes: Taking the biomass of the sample plots as the basic data and a single remote sensing variable as the characteristic, principal component analysis was performed on the biomass of the sample plots respectively, and the variables whose cumulative eigenvalues ​​of the principal component analysis exceeded a certain threshold were used as the principal component analysis combination of multi-source variables.

9. A device for estimating forest aboveground biomass based on multi-source remote sensing data, characterized in that: The optimal forest aboveground biomass model is screened out by the method according to any one of claims 1 to 8, wherein the device comprises: Data acquisition module, used to obtain forest plot data and its corresponding multi-source remote sensing image data; Data classification module, used to classify tree species based on forest plot data; Model building module, used to build forest aboveground biomass models; Model training module, used to train forest aboveground biomass model; Model testing module, used to test forest floor biological models; The model screening module is used to screen out the optimal model based on the test results of different models.

10. An electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 8 is implemented.

Citation Information

Patent Citations

  • Soil medium carbon steel corrosion rate prediction method based on machine learning

    CN117894411A

  • Systems and methods for automated machine learning

    US20210150412A1

Cited By

  • High-resolution large-scale forest biomass remote sensing prediction method and system

    CN119323736A

  • A high-resolution and large-scale forest biomass remote sensing prediction method and system

    CN119323736B

  • Forest aboveground biomass estimation method and system based on open shared data

    CN119578703A

  • Forest resource quantity estimation method and device based on artificial intelligence

    CN120525812A

  • Machine learning-based airborne laser radar forest stock estimation method and device

    CN121170568A