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

Abstract

The invention discloses a forest above-ground biomass estimation method and system based on open shared data, and relates to the technical field of forest carbon reserve measurement, and the method comprises the steps: obtaining the open shared data, carrying out the preprocessing, extracting a variable set from the open shared data, and taking the variable set as model input, then, variable roughing is carried out through a Kendall correlation coefficient analysis method, variable selection is carried out through a random forest change replacement method, and stable and reliable model variables with low redundancy are obtained; and respectively training and establishing above-ground biomass nonlinear estimation models of the coniferous forest, the broad-leaved forest and the coniferous and broad-leaved mixed forest by utilizing the optimized variables to obtain biomass estimation results of the coniferous forest, the broad-leaved forest and the mixed forest, and comprehensively calculating to obtain the above-ground total biomass of the forest. According to the method, nonlinear estimation models are innovatively established for a single tree species type and a mixed tree species type respectively, the precision of overground biomass estimation is improved, and the method can serve accurate estimation of carbon sources and carbon sinks of'region-country-global '.

Description

Technical Field

[0001] The present application relates to the technical field of forest carbon stock measurement, and in particular to a method and system for estimating forest aboveground biomass based on open shared data. Background Art

[0002] Traditional methods for forest carbon stock measurement mainly include forest plot inventory and nonlinear estimation model of forest aboveground biomass based on remote sensing technology, and the research scale involves provincial and national levels. These traditional forest plot inventories rely on field sampling surveys, which are labor-intensive, require a lot of manpower and financial resources, and are inefficient. They have poor timeliness in reflecting the dynamic changes of forest ecosystems, and it is difficult to achieve accurate measurement of large-scale forest structural parameters and forest carbon stocks. Although remote sensing technology makes up for the shortcomings of traditional plot inventory methods, and can quickly, accurately and non-destructively determine forest structural parameters and spatial distribution patterns, providing effective technical means for large-scale forest resource surveys and investigations, its main problems are: optical remote sensing images can provide rich spectral information, but are easily affected by weather, making it difficult to obtain three-dimensional vertical structural information of the canopy; synthetic aperture radar images are less affected by clouds and rain, but are easily affected by terrain and soil conditions, and are prone to saturation; three-dimensional laser scanning point cloud data acquisition costs are high, making it difficult to conduct research in large areas; in addition, in the coordinated inversion of forest biomass using multiple types of remote sensing data, there are still problems with the spatial and temporal scale of data and how to mine different data features for estimation model inversion. Therefore, in the study of remote sensing mechanism estimation models, it is necessary to rely on open and shared data from forestry-related research to improve the accuracy of ecological research and break through related theories such as forest biophysical parameters.

[0003] The open and shared data of forestry-related research mainly include satellite remote sensing data, ground survey data, global or regional forest resource assessment data, etc. These open and shared data can effectively improve the accuracy of ecological research, promote the development of ecological models, and reveal ecological processes in ecological applications. However, in the ecological applications of these open and shared data (especially satellite remote sensing data), there are problems such as complex data processing technology, low product accuracy, and high cost of large-scale applications, which greatly affect the continuous production and application of digital products of ecological environment. Therefore, it is of great significance to use these open and shared data to accurately and efficiently monitor the forest carbon sink capacity, and the breakthrough of its key technologies has become a national innovation strategy development demand. Summary of the invention

[0004] The purpose of this application is to provide a method and system for estimating forest aboveground biomass based on open and shared data, which can achieve high-precision estimation of aboveground biomass of coniferous forests, broad-leaved forests and mixed forests.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a method for estimating forest aboveground biomass based on open shared data, comprising the following steps:

[0007] The open and shared data are acquired and preprocessed to obtain the preprocessed open and shared data; the open and shared data include satellite remote sensing data, digital elevation model data, global forest height data, and land use and land cover data.

[0008] Based on the preprocessed open shared data, a set of model input variables is extracted; the model input variable set includes a spectral vegetation index set, a biophysical variable set, a terrain variable set and a forest canopy height variable.

[0009] For any tree species in the forest, the variables in the model input variable set are screened based on correlation analysis to obtain the model input variable set after the tree species are roughly selected.

[0010] Based on variable replacement analysis, the variables in the model input variable set after the rough selection of tree species are screened twice to obtain the model input variable set after the fine selection of tree species.

[0011] The nonlinear estimation models of aboveground biomass of different tree species are established and trained respectively by using the selected set of model input variables to obtain the trained nonlinear estimation models of aboveground biomass of different tree species; the nonlinear estimation models of aboveground biomass of different tree species at least include the nonlinear estimation model of aboveground biomass of coniferous forest, the nonlinear estimation model of aboveground biomass of broad-leaved forest and the nonlinear estimation model of aboveground biomass of mixed forest.

[0012] Based on the trained nonlinear estimation model of aboveground biomass of different tree species and the optimized set of input variables of the aboveground biomass model, the estimation results of the aboveground biomass of each tree species in the forest are obtained.

[0013] Based on the estimated results of the above-ground biomass of each tree species in the forest, the estimated results of the above-ground biomass of the forest are determined.

[0014] In the second aspect, the present application provides a forest aboveground biomass estimation system based on open shared data, including the following modules:

[0015] The open shared data acquisition module is used to acquire open shared data and preprocess it to obtain preprocessed open shared data; the open shared data includes satellite remote sensing data, digital elevation model data, global forest height data and land use and land cover data.

[0016] The model input variable extraction module is used to extract the model input variable set based on the preprocessed open shared data; the model input variable set includes a spectral vegetation index set, a biophysical variable set, a terrain variable set and a forest canopy height variable.

[0017] The model input variable screening module is used to screen the variables in the model input variable set for any tree species in the forest based on correlation analysis to obtain the model input variable set after the rough selection of tree species; it is also used to perform secondary screening on the variables in the model input variable set after the rough selection of tree species based on variable replacement analysis to obtain the model input variable set after the fine selection of tree species.

[0018] The single tree species biomass estimation module is used to establish and train nonlinear estimation models of aboveground biomass of different tree species using a selected set of model input variables to obtain trained nonlinear estimation models of aboveground biomass of different tree species; it is also used to obtain the aboveground biomass estimation results of each tree species in the forest based on the trained nonlinear estimation models of aboveground biomass of different tree species and the optimized set of aboveground biomass model input variables; the nonlinear estimation models of aboveground biomass of different tree species at least include a nonlinear estimation model of aboveground biomass of coniferous forests, a nonlinear estimation model of aboveground biomass of broad-leaved forests, and a nonlinear estimation model of aboveground biomass of mixed forests.

[0019] The forest total biomass estimation module is used to determine the forest aboveground biomass estimation result based on the aboveground biomass estimation result of each tree species in the forest.

[0020] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0021] The present application provides a forest aboveground biomass estimation method and system based on open and shared data, which obtains satellite remote sensing data, digital elevation model data, global forest height data and land use and land cover data and preprocesses them respectively, and then extracts a set of model input variables based on the preprocessed open and shared data, including a spectral vegetation index set, a biophysical variable set, a terrain variable set and a forest canopy height variable; then, the Kendall correlation coefficient analysis method is used to perform rough variable selection, and the random forest change replacement method is used to perform fine variable selection, so as to obtain model variables with low redundancy and stable and reliable; then, these optimized variables are used to train and establish nonlinear aboveground biomass estimation models for coniferous forests, broad-leaved forests and mixed coniferous and broad-leaved forests, respectively, to obtain aboveground biomass estimation results of coniferous forests, broad-leaved forests and mixed forests, and finally, the aboveground biomass estimation results of forests can be obtained by synthesis. The solution of this application is based on open shared data. By preprocessing the open shared data and extracting the model input variable set, it can reflect the forest physiological and ecological structure characteristics and spatiotemporal remote sensing characteristics. Then, a nonlinear estimation model of biomass of different tree species is constructed using the hierarchically screened variable set. After training, it is used for high-precision estimation of aboveground biomass of coniferous forests, broad-leaved forests and mixed forests. The solution of this application innovatively establishes estimation models for single tree species and mixed types respectively. Compared with the traditional solution of ignoring tree species differences and estimating the biomass of the entire forest, this application improves the accuracy of aboveground biomass estimation and provides key technical support and scientific data for carbon budget. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0023] Figure 1 A flowchart of a method for estimating forest aboveground biomass based on open shared data provided in one embodiment of the present application.

[0024] Figure 2 A flowchart of the model construction and training process in a method for estimating forest aboveground biomass based on open shared data provided in one embodiment of the present application.

[0025] Figure 3 An overall technical concept diagram of a method for estimating forest aboveground biomass based on open shared data provided in another embodiment of the present application.

[0026] Figure 4A schematic diagram of selected variables and their importance for each model in a method for estimating forest aboveground biomass based on open shared data provided in another embodiment of the present application.

[0027] Figure 5 A schematic diagram of the results of nonlinear biomass estimation models for each model in a method for estimating aboveground biomass of a forest based on open shared data provided in another embodiment of the present application.

[0028] Figure 6 A spatially continuous forest aboveground biomass spatial distribution map produced in a forest aboveground biomass estimation method based on open shared data is provided in another embodiment of the present application.

[0029] Figure 7 A schematic diagram of the functional modules of a forest aboveground biomass estimation system based on open shared data provided in one embodiment of the present application.

[0030] Figure 8 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0032] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0033] In an exemplary embodiment, Figure 1 As shown in the figure, a method for estimating forest aboveground biomass based on open shared data is provided, which includes the following steps:

[0034] A1. Obtain open shared data and preprocess it to obtain preprocessed open shared data; the open shared data includes satellite remote sensing data, digital elevation model data, global forest height data, and land use and land cover data. Step A1 specifically includes the following steps:

[0035] A11. Obtain satellite remote sensing data, digital elevation model data, global forest height data, and land use and land cover data of forest areas.

[0036] A12. For satellite remote sensing data and land use and land cover data, data correction, atmospheric correction, noise removal, data enhancement, resolution adjustment and data cropping are performed in sequence to obtain preprocessed satellite remote sensing data and preprocessed land use and land cover data.

[0037] A13. For the digital elevation model data and the global forest height data, noise removal, data enhancement, resolution adjustment and data cropping are performed in sequence to obtain the preprocessed digital elevation model data and the preprocessed global forest height data.

[0038] A2. Based on the pre-processed open shared data, extract the model input variable set; the model input variable set includes a spectral vegetation index set, a biophysical variable set, a terrain variable set and a forest canopy height variable. Step A2 specifically includes the following steps:

[0039] A21. Based on the preprocessed satellite remote sensing data, a set of spectral vegetation indices and a set of biophysical variables are extracted.

[0040] A22. Based on the preprocessed digital elevation model data, a terrain variable set is extracted.

[0041] A23, extracting forest canopy height variables based on the preprocessed global forest height data and the preprocessed land use and land cover data. The specific step A23 includes the following steps:

[0042] Using the preprocessed land use and land cover data, pixels with field attributes of broad-leaved forest, coniferous forest or mixed broad-leaved and coniferous forest in the forest area are extracted and the vector boundary range is constructed to obtain the forest vector range.

[0043] The forest vector extent was overlaid and clipped with the preprocessed global forest height data to obtain the forest canopy height variable of the forest area.

[0044] In this embodiment, the spectral vegetation index set includes the ratio vegetation index, the enhanced vegetation index, the difference vegetation index and the red-edge normalized vegetation index established using different red-edge bands; the red-edge normalized vegetation index established using different red-edge bands includes NDVI48, NDVI56, NDVI57, NDVI58a, NDVI67, NDVI68a and NDVI78a; the biophysical variable set includes leaf area index, chlorophyll content, forest coverage, absorbable photosynthetically active radiation ratio and canopy moisture content; the terrain variable set includes slope, aspect, ground elevation, slope position index and terrain humidity index.

[0045] A3: For any tree species in the forest, the variables in the model input variable set are screened based on correlation analysis to obtain a model input variable set after the tree species are roughly selected. Step A3 specifically includes the following steps:

[0046] A31. The Kendall rank correlation coefficient method is used to calculate the Kendall rank correlation coefficient between the variables in the model input variable set.

[0047] A32. In the model input variable set, the variables whose pairwise correlation coefficients are greater than the Kendall rank correlation coefficient are eliminated to obtain the model input variable set after rough selection of tree species.

[0048] In this embodiment, the model input variable set after rough selection of coniferous forest includes forest canopy height variable, difference vegetation index, NDVI56, forest coverage, ground elevation, slope aspect and slope; the model input variable set after rough selection of broad-leaved forest includes forest canopy height variable, enhanced vegetation index, NDVI57, absorbable photosynthetically active radiation ratio, ground elevation, slope and broken position index; the model input variable set after rough selection of mixed forest includes forest canopy height variable, enhanced vegetation index, NDVI78a, NDVI68a, NDVI48, slope aspect, slope and broken position index.

[0049] A4. Perform a secondary screening of the variables in the model input variable set after the rough selection of tree species based on variable replacement analysis to obtain the model input variable set after the fine selection of tree species. Figure 2 In the flowchart shown, step A4 specifically includes the following steps:

[0050] A41. Based on all the variables in the model input variable set after the rough selection of tree species, a random forest model of the tree species is trained.

[0051] A42. For any variable, randomly change the value of the variable and keep the values ​​of other variables unchanged to obtain a new random forest model of the tree species.

[0052] A43. Determine the importance of variables based on the benchmark performance of the new random forest model of tree species.

[0053] A44. In the model input variable set after the rough selection of tree species, the variables whose importance is lower than the importance threshold are eliminated to obtain the model input variable set after the fine selection of tree species.

[0054] In this embodiment, the selected model input variable set for coniferous forests includes forest coverage rate, forest canopy height variable, slope aspect and difference vegetation index; the selected model input variable set for broad-leaved forests includes forest canopy height variable, absorbable photosynthetically active radiation ratio and NDVI57; the selected model input variable set for mixed coniferous and broad-leaved forests includes forest canopy height variable, slope position index, slope and NDVI48.

[0055] A5. Use the selected model input variable set to establish and train nonlinear estimation models of aboveground biomass of different tree species, and obtain trained nonlinear estimation models of aboveground biomass of different tree species; the nonlinear estimation models of aboveground biomass of different tree species at least include a nonlinear estimation model of aboveground biomass of coniferous forest, a nonlinear estimation model of aboveground biomass of broad-leaved forest, and a nonlinear estimation model of aboveground biomass of mixed forest. Step A5 specifically includes the following steps:

[0056] A51. According to the different tree species in the forest, the nonlinear estimation models of aboveground biomass of different tree species are established by using the selected model input variable set; the tree species in the forest include coniferous forest, broad-leaved forest and mixed forest;

[0057] A52. For any tree species in the forest, a nonlinear estimation model for the aboveground biomass of the tree species is trained using the selected model input variable set and the measured value of the aboveground biomass of the tree species to obtain a trained nonlinear estimation model for the aboveground biomass of the tree species.

[0058] A6. Based on the trained nonlinear estimation models of aboveground biomass of different tree species and the optimized set of input variables of the aboveground biomass model, the aboveground biomass estimation results of each tree species in the forest are obtained; the trained nonlinear estimation models of aboveground biomass of different tree species at least include the trained nonlinear estimation model of aboveground biomass of coniferous forest, the trained nonlinear estimation model of aboveground biomass of broad-leaved forest and the trained nonlinear estimation model of aboveground biomass of mixed forest.

[0059] A7. Based on the estimated results of the aboveground biomass of each tree species in the forest, the estimated results of the aboveground biomass of the forest are determined.

[0060] The above embodiments introduce the forest aboveground biomass estimation method based on open shared data provided by the present application from the perspective of actual use of the scheme. In another exemplary embodiment, the process of the method provided by the present application is described from the perspective of research and development. The overall technical concept is as follows Figure 3 As shown, the following steps are included:

[0061] Step 1: Open and shared data acquisition and processing.

[0062] Open and shared data mainly refers to remote sensing images and geographic data that can be freely accessed and used. Commonly used open and shared data platforms at home and abroad include Google EarthEngine (GEE), Sentinel Hub, USGS EarthExplorer (Landsat, DEM, etc.), Copernicus Open Access Hub, National Remote Sensing Center (NRSCC), National Earth Science System Science Sharing Service Platform, Geospatial Information Cloud, etc.

[0063] The satellite remote sensing data in this application uses open Sentinel satellite images, which can be downloaded from https: / / www.sentinel-hub.com / ; the global forest height data uses the global 10-meter resolution tree height data released in recent years, which can be downloaded from https: / / earthengine.google.com / ; the digital elevation model (Digital Elevation Model, DEM) uses the global 30-meter resolution elevation data provided by SRTM, which can be downloaded from http: / / gdex.cr.usgs.gov / gdex / ; the land use and land cover data (Land Use and Land Cover, LCLU) uses the global 10-meter resolution land cover product developed by Tsinghua University, which can be obtained through https: / / data-starcloud.pcl.ac.cn / . It should be noted that these open and shared data can also be processed and obtained through self-developed algorithms.

[0064] The acquired open data is generally raw data, which needs to be processed to ensure its quality and accuracy, laying a foundation for subsequent analysis and application. In this embodiment, the processing of satellite remote sensing data and land use and land cover data includes data correction, atmospheric correction, noise removal, data enhancement, resolution adjustment, data clipping, etc., and the processing of global forest height data and digital elevation model data includes noise removal, data enhancement, resolution adjustment, data clipping, etc.

[0065] Data correction mainly carries out radiation correction (adjusting the image to correct the radiation error caused by sensors and environmental factors) and geometric correction (correcting the geometric distortion in the image to align it with the actual position on the earth's surface); atmospheric correction mainly removes atmospheric effects, such as aerosols and water vapor, to ensure that the spectral information of the image is true and reliable; noise removal refers to cleaning the noise in the data and improving the data quality; data enhancement is to improve the visualization effect of image-based data (de-clouding, de-noising, etc.) to make the features more obvious; data cropping is to use the study area to extract the area of ​​interest to reduce the scale of processed data; resolution adjustment is to resample these data of different resolutions to keep the resolution of these data consistent. It should be noted that these preprocessing operations can be implemented using professional software (such as GEE, SNAP, ENVI, PIESAT) or self-compiled algorithm code to generate data that meets application requirements. In this embodiment, ENVI and SNAP are used to perform the above operations.

[0066] In the above step 1 of this embodiment, it is proposed to use open and shared data to conduct estimation research on biomass models. Open and shared data has brought many conveniences to scientific research, public services, business innovation, etc., facilitates multidisciplinary collaboration, can promote social progress and development, and achieve broader social benefits and economic value. Compared with the optical remote sensing images, synthetic aperture radar images, three-dimensional laser scanning point clouds, etc. used in traditional technologies to obtain more accurate biomass nonlinear estimation models, the current biomass estimation research uses high-resolution data, such as optical remote sensing images, synthetic aperture radar images, and three-dimensional laser scanning point clouds. The cheapest price is tens to hundreds of dollars per square kilometer. In the inversion of biomass models, a single optical remote sensing image can provide rich spectral information, but it is easily affected by weather and it is difficult to obtain vertical structure information of the canopy; synthetic aperture radar images are less affected by clouds and rain, but they are greatly affected by terrain and soil conditions, and saturation phenomena are prone to occur; three-dimensional laser scanning point cloud data is mainly due to its high acquisition cost, and it is difficult to carry out research in large areas. This embodiment proposes a biomass collaborative inversion method that combines multiple open data. First, select suitable open shared data according to the research area, and then perform preprocessing such as data correction, atmospheric correction, noise removal, data enhancement, resolution adjustment, and data cropping on the image format data. For paper document data, key fields need to be cleaned and verified to ensure the spatial connectivity of these shared data.

[0067] Step 2: Extract model input variables.

[0068] In this embodiment, a nonlinear estimation model of aboveground biomass is constructed according to different forest types, and the variables input to the model include: spectral vegetation index, biophysical variables, terrain variables, and forest canopy height variables, etc. Among them, the spectral vegetation index and biophysical variables are obtained through Sentinel-2A multispectral image data, that is, satellite remote sensing data; terrain variables (slope, aspect, and ground elevation) are calculated through digital elevation model data (SRTM DEM); forest canopy height variables are calculated through land use and land cover data and global forest canopy height data; the specific description of these variable extraction and calculation processes is as follows:

[0069] Step 2.1: Extraction of spectral vegetation index set.

[0070] In the present invention, different bands of Sentinel-2A images are used for correlation calculation to obtain spectral vegetation index information that can effectively estimate forest biomass information, mainly including the Normalized Difference Vegetation Index (NDVI), the Ratio Vegetation Index (RVI), the Enhanced Vegetation Index (EVI), the Difference Vegetation Index (DVI) and the red-edge normalized vegetation index established using different red-edge bands. The names of the spectral vegetation variables and the calculation models are shown in Table 1.

[0071] Table 1 Spectral vegetation index set variable names and calculation models

[0072]

[0073] In Table 1, the image bands are represented by the following symbols: blue band (B), green band (G), red band (R), near infrared band (NIR1), narrow near infrared band (NIR2), red edge band 1 (RE1), red edge band 2 (RE2), and red edge band 3 (RE3).

[0074] Step 2.2: Extraction of biophysical variable sets.

[0075] The set of biophysical variables of forest vegetation canopy mainly includes: leaf area index, chlorophyll content, forest coverage rate, ratio of absorbable photosynthetically active radiation and canopy water content. These variables are closely related to forest vegetation photosynthesis, respiration, transpiration, intercepted precipitation and carbon cycle process; they can reflect the growth status and biomass size of forest vegetation, thereby enhancing the prediction ability of forest aboveground biomass.

[0076] In this embodiment, the forest canopy physical variables used in the estimation of forest aboveground biomass are mainly derived from Sentinel-2A image data, which are extracted through the biophysical processing module of SNAP software (Menu→Optical→ThematicLand Processing→Biophysical Processor). The names and descriptions of the biophysical variables are shown in Table 2:

[0077] Table 2 Biophysical variable names and variable descriptions

[0078]

[0079] Step 2.3: Terrain variable set extraction.

[0080] Topographic variables affect the distribution and growth of forest vegetation. They can obtain the topographic attribute information of the forest area and provide a meaningful forest stand site attribute map for the model to accurately capture the local changes in the forest canopy structure. Therefore, the use of topographic variables to estimate the aboveground biomass of the forest can enhance the prediction ability of the estimation model. In the present invention, the topographic variables used for estimating the aboveground biomass of the forest mainly include slope, slope aspect, ground elevation, slope index and terrain moisture index, etc. These topographic variables can be obtained from SRTM DEM data. The names of the topographic variables and variable descriptions are shown in Table 3:

[0081] Table 3 Topographic variable names and variable descriptions

[0082]

[0083] Step 2.4: Extraction of forest canopy height variables.

[0084] The forest canopy height is an important indicator for measuring the structure and health of the forest. It can be obtained from data such as lidar, remote sensing images, ground measurements and model fitting by using self-compiled algorithms or software, but this requires a large amount of field measured data, and field surveys in large forest areas are difficult to access. In the present invention, the image data of land cover / utilization are used to extract pixels with field attributes of broad-leaved forests, coniferous forests or mixed coniferous and broad-leaved forests and construct their vector boundary ranges. Based on this vector area, the forest canopy height data of different forest areas are obtained by superimposing and clipping with the global forest height data, thereby obtaining the forest canopy height data of different forest areas, and then obtaining the forest canopy height variable Mheight.

[0085] Step 3: Variable hierarchical screening and optimization.

[0086] The optimization of model input variables can delete unimportant variables in the model, avoid the problem of "dimensionality disaster" of the model, and reduce the difficulty of the biomass nonlinear estimation model. In order to remove redundant variable information, reduce the variable dimension, and obtain simple and effective model input variables from the above-extracted model input variable set, this embodiment proposes a variable hierarchy screening method that integrates variable correlation analysis and variable replacement analysis, which mainly includes model input variable rough selection and model input variable refinement.

[0087] Step 3.1: Rough selection of model input variables.

[0088] When selecting model input variables, a subset of variables with strong correlation with the predicted items and weak correlation between model input variables is usually selected. Considering that most of the binary relationships between candidate model input variables (spectral vegetation index, biophysical variables, topographic variables) are difficult to meet the normality or linear relationship assumptions. Therefore, this embodiment uses the Kendall rank correlation coefficient method to evaluate the correlation between candidate model input variables, using the correlation coefficient and Kendall rank correlation coefficient between model input variables, and deleting highly correlated variables (such as variables with correlation coefficients greater than 0.8 or 0.9) to determine the location of redundant variables, and complete the elimination and rough selection of redundant variables.

[0089] Regarding the Kendall rank coefficient method, for the two candidate model input variables X and Y, denoted as (X, Y). If for any two pairs of variables (X i ,Y i ) and (X j ,Y j ), if (X i <X j And Y i <Y j ) or (X i >X j And Y i >Y j ), then the pair of samples is called a consistent pair; otherwise, the pair of samples is called an inconsistent pair. The calculation formula of Kendall rank correlation coefficient ρ is as follows:

[0090]

[0091] Where C represents the number of consistent pairs, D represents the number of inconsistent pairs, and n is the number of candidate model input variables.

[0092] Step 3.2: Select model input variables.

[0093] In order to further condense the model input variables and improve the interpretation ability of the biomass nonlinear estimation model. For the variable set screened by the Kendall correlation coefficient method, this embodiment uses the random forest variable replacement method to perform secondary optimization selection of variables. That is, using the above-mentioned roughly selected model input variables as the initial input, using random forest to iteratively eliminate the variables with the lowest importance, and compare and determine the best prediction variables of the model, thereby completing the variable refinement and selection of the biomass nonlinear estimation model. The basic idea of ​​the replacement method is: if the feature of a variable is more important, then changing the feature value will make its predicted value more likely to be wrong. In this embodiment, the basic process of random forest variable replacement processing is as follows:

[0094] 1) All variables after rough selection are used as features to train the random forest model;

[0095] 2) For each variable feature, randomly change its feature value and keep other features unchanged, and then recalculate the performance of the model with the modified data;

[0096] 3) Calculate the benchmark performance (such as accuracy, mean square error, etc.) of the variable feature before and after the change. If the model performance decreases more, the importance of the variable feature is higher; otherwise, its importance is low;

[0097] 4) Perform the same operation on all variable features to obtain the permutation importance of each feature;

[0098] 5) According to the importance ranking of each feature permutation, the feature importance order is obtained, and certain rules (such as setting the accuracy of feature benchmark performance, etc.) are used to eliminate variables with low importance to complete the optimization screening of model variables.

[0099] The quantitative expression of the characteristic benchmark performance is shown as follows:

[0100]

[0101] The first term represents the error measure between the true value (observed value) and its predicted value before the variable feature replacement; the second term represents the error measure between the true value (observed value) and its predicted value after the variable feature replacement; i represents the number of the sample; k represents the number of the tree (decision tree) constructed by the random forest; m k The total number of variables input to the kth tree built for the random forest model. In addition, A = max|y i |,i∈N, where N represents the total number of variables.

[0102] The model input variables extracted in step 2 of this embodiment can reflect the characteristics of forest physiological and ecological structure and spatiotemporal remote sensing characteristics. In order to effectively solve the reliability problem of large-scale regional biomass nonlinear estimation model, this embodiment proposes a variable level screening strategy that combines Kendall correlation coefficient analysis and variable replacement analysis strategy to achieve robust elimination of redundant variables and optimal selection of effective model input variables.

[0103] In the rough selection of model input variables, the initial variables (spectral vegetation index, biophysical variables, and terrain variables) were analyzed by Kendall correlation according to different forest types (coniferous forest, broad-leaved forest, mixed coniferous and broad-leaved forest, and mixed all forest types). The results of the rough selection of variables are shown in Table 4:

[0104] Table 4 Rough selection results of model input variables

[0105] Model Candidate variables Broadleaf forest Mheight,EVI,NDVI57,FAPAR,DEM,Slope,SPI Coniferous forest Mheight,DVI,NDVI56,FVC,DEM,Aspect,Slope Mixed forest Mheight,EVI,NDVI78a,NDVI68a,NDVI48,Aspect,Slope,SPI Comprehensive forest type Mheight, EVI, DVI, NDVI68a, FVC, Aspect, Slope, SPI

[0106] On the basis of the rough selection of model input variables, this embodiment further optimizes the model input variables through the random forest algorithm variable replacement method. Using the variable importance index and backward selection method in the random forest model, the optimization selection of model input variables was carried out for different biomass nonlinear estimation models. The variables finally selected for the nonlinear estimation model of aboveground biomass of coniferous forests are (FVC, MHeight, Aspect, DVI), the variables finally selected for the nonlinear estimation model of aboveground biomass of broad-leaved forests are (MHeight, FAPAR, NDVI57), the variables finally selected for the nonlinear estimation model of aboveground biomass of mixed forests are (MHeight, SPI, Slope, NDVI48), and the variables finally selected for the nonlinear estimation model of aboveground biomass of comprehensive forest types (coniferous forests, broad-leaved forests and mixed coniferous and broad-leaved forests) are (FVC, MHeight, DVI, SPI). The final selected variables and their importance are as follows: Figure 4 shown.

[0107] Step 4: Construction of nonlinear biomass estimation model.

[0108] The traditional biomass statistical regression model is simple in form and easy to calculate, but it is difficult to effectively express the complex relationship between forest aboveground biomass and multi-source remote sensing variables, and the model has poor expansion ability. In order to fully explore the relationship between the model prediction variables (independent variables) and the interpretation variables (dependent variables) after the above screening, and improve the estimation ability of the forest biomass model, this embodiment proposes a forest biomass nonlinear estimation model construction method based on random forest regression analysis theory and a model comprehensive evaluation method.

[0109] In the construction of the forest biomass nonlinear estimation model, this embodiment constructs three forest biomass nonlinear estimation models for a single forest type (coniferous forest, broad-leaved forest and mixed forest), and simultaneously constructs one biomass nonlinear estimation model for all forest types; in the model construction process, the ratio of field sample data used for training and testing in this embodiment is 7:3, the training set is used for random forest model training, and the test set is used for model performance evaluation. It should be noted that the field forest biomass used for verification in this embodiment is obtained by field survey.

[0110] In the forest biomass model evaluation and verification, the metric indicators in this embodiment use the root mean square error (RMSE) and the coefficient of determination (R2), which evaluate the accuracy of the prediction estimation model by comparing the deviation between the predicted value and the observed value of the forest biomass. The specific calculation formula is as follows:

[0111]

[0112] Among them, y i represents the observed value of forest biomass; represents the corresponding predicted value of forest biomass; y i represents the average value of forest biomass verification data; n represents the number of forest biomass sample plots data.

[0113] Based on the rough and fine selection of model input variables in the above step 3, step 4 of this embodiment uses coniferous forest plots (33), broad-leaved forest plots (28), mixed forest plots (18), and mixed plots (79) to invert the nonlinear estimation model of aboveground biomass of each tree species. The determination coefficient R2 of these biomass nonlinear estimation models is between 0.69 and 0.74, and the root mean square error RMSE is between 23.375 and 24.139 Mg / ha.

[0114] Among the nonlinear estimation models of aboveground biomass constructed based on different tree species, the performance of the nonlinear estimation models of aboveground biomass of coniferous forests (R2=0.7386, RMSE=24.207Mg / ha) and broad-leaved forests (R2=0.7425, RMSE=23.38Mg / ha) is similar and better than the performance of the nonlinear estimation model of aboveground biomass of mixed forests (R2=0.6909, RMSE=24.139Mg / ha). However, the performance of the nonlinear estimation model of aboveground biomass of forests constructed by mixing all forest types is poor, with an R2 value of 0.4197 and a RMSE value of 31.016Mg / ha. Compared with the nonlinear estimation model of aboveground biomass constructed by mixing all forest types, the nonlinear estimation models of aboveground biomass constructed for different forest types are more robust and reliable, so that the constructed nonlinear estimation model of biomass has a higher predictive ability. The results of the nonlinear estimation model of biomass constructed in this embodiment are as follows. Figure 5 Based on the nonlinear estimation model of forest aboveground biomass constructed in this embodiment, a spatial distribution map of spatially continuous forest aboveground biomass was produced for a certain study area, as shown in FIG. Figure 6 shown.

[0115] In summary, this application proposes a biomass collaborative inversion method that combines multiple open data. First, suitable open shared data are selected according to the study area, and then the data in image format are preprocessed by data correction, atmospheric correction, noise removal, data enhancement, resolution adjustment, data cropping, etc. For paper document data, key fields need to be cleaned and verified to ensure the spatial correlation of these shared data. Then, variables related to aboveground biomass are extracted from the preprocessed data, and the Kendall correlation coefficient analysis method and random forest variable replacement method are used to perform hierarchical screening of candidate variables for the biomass nonlinear estimation model. Finally, innovative nonlinear regression models are established for single forest types and mixed forest types to achieve high-precision estimation of forest biomass, serve the dual carbon policy at the "regional-national-global" scale, and promote green and low-carbon transformation.

[0116] Based on the same inventive concept, the embodiment of the present application also provides a forest aboveground biomass estimation system based on open shared data for implementing the above-mentioned forest aboveground biomass estimation method based on open shared data. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more system embodiments provided below can refer to the limitations of the forest aboveground biomass estimation method based on open shared data above, and will not be repeated here.

[0117] In an exemplary embodiment, Figure 7 As shown in the figure, a forest aboveground biomass estimation system based on open shared data is provided, which includes the following modules:

[0118] The open shared data acquisition module is used to acquire open shared data and preprocess it to obtain preprocessed open shared data; the open shared data includes satellite remote sensing data, digital elevation model data, global forest height data and land use and land cover data.

[0119] The model input variable extraction module is used to extract the model input variable set based on the preprocessed open shared data; the model input variable set includes a spectral vegetation index set, a biophysical variable set, a terrain variable set and a forest canopy height variable.

[0120] The model input variable screening module is used to screen the variables in the model input variable set for any tree species in the forest based on correlation analysis to obtain the model input variable set after the rough selection of tree species; it is also used to perform secondary screening on the variables in the model input variable set after the rough selection of tree species based on variable replacement analysis to obtain the model input variable set after the fine selection of tree species.

[0121] The single tree species biomass estimation module is used to establish and train nonlinear estimation models of aboveground biomass of different tree species using a selected set of model input variables to obtain nonlinear estimation models of aboveground biomass of different tree species; it is also used to obtain nonlinear estimation results of aboveground biomass of each tree species in the forest based on the trained nonlinear estimation models of aboveground biomass of different tree species and the selected set of aboveground biomass model input variables; the nonlinear estimation models of aboveground biomass of different tree species at least include a nonlinear estimation model of aboveground biomass of coniferous forests, a nonlinear estimation model of aboveground biomass of broad-leaved forests, and a nonlinear estimation model of aboveground biomass of mixed forests.

[0122] The forest total biomass estimation module is used to determine the forest aboveground biomass estimation result based on the aboveground biomass estimation result of each tree species in the forest.

[0123] certainly, Figure 7 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different functions. Figure 7 One or at least two components of the system shown.

[0124] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 8 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the forest aboveground biomass estimation method based on open shared data mentioned in the above-mentioned embodiment is implemented.

[0125] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0126] The technical features of the above embodiments may be arbitrarily combined. 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.

[0127] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for estimating forest aboveground biomass based on open shared data, characterized in that: include: Acquire open shared data and perform preprocessing to obtain preprocessed open shared data; The open and shared data include satellite remote sensing data, digital elevation model data, global forest height data, and land use and land cover data; Extracting a model input variable set based on the preprocessed open shared data; the model input variable set includes a spectral vegetation index set, a biophysical variable set, a terrain variable set, and a forest canopy height variable; For any tree species in the forest, the variables in the model input variable set are screened based on correlation analysis to obtain the model input variable set after the tree species are roughly selected; Performing secondary screening on the variables in the model input variable set after the rough selection of tree species based on variable replacement analysis to obtain the model input variable set after the fine selection of tree species; Using the selected model input variable set, respectively establish and train nonlinear estimation models of aboveground biomass of different tree species to obtain trained nonlinear estimation models of aboveground biomass of different tree species; the nonlinear estimation models of aboveground biomass of different tree species at least include a nonlinear estimation model of aboveground biomass of coniferous forest, a nonlinear estimation model of aboveground biomass of broad-leaved forest, and a nonlinear estimation model of aboveground biomass of mixed forest; Based on the trained nonlinear estimation model of aboveground biomass of different tree species and the optimized set of input variables of aboveground biomass model, the aboveground biomass estimation results of each tree species in the forest are obtained; Based on the estimated results of the above-ground biomass of each tree species in the forest, the estimated results of the above-ground biomass of the forest are determined.

2. The method for estimating forest aboveground biomass based on open shared data according to claim 1, characterized in that: Using the selected model input variable set, respectively establish and train nonlinear estimation models of aboveground biomass of different tree species to obtain trained nonlinear estimation models of aboveground biomass of different tree species, specifically including: According to different tree species in the forest, nonlinear estimation models of aboveground biomass of different tree species are established respectively using the selected model input variable set; the tree species in the forest include coniferous forest, broad-leaved forest and mixed forest; For any tree species in the forest, a nonlinear estimation model for aboveground biomass of the tree species is trained using a selected model input variable set and the measured value of aboveground biomass of the tree species to obtain a trained nonlinear estimation model for aboveground biomass of the tree species.

3. The method for estimating forest aboveground biomass based on open shared data according to claim 1, characterized in that: Obtain open shared data and preprocess it to obtain preprocessed open shared data, including: Obtain satellite remote sensing data, digital elevation model data, global forest height data, and land use and land cover data of forest areas respectively; For the satellite remote sensing data and the land use and land cover data, data correction, atmospheric correction, noise removal, data enhancement, resolution adjustment and data clipping are performed in sequence to obtain preprocessed satellite remote sensing data and preprocessed land use and land cover data; For the digital elevation model data and the global forest height data, noise removal, data enhancement, resolution adjustment and data clipping are performed in sequence to obtain preprocessed digital elevation model data and preprocessed global forest height data.

4. The method for estimating forest aboveground biomass based on open shared data according to claim 3, characterized in that: Based on the preprocessed open shared data, extract the model input variable set, including: Based on the preprocessed satellite remote sensing data, a set of spectral vegetation indices and a set of biophysical variables are extracted; According to the pre-processed digital elevation model data, a terrain variable set is extracted; Based on the preprocessed global forest height data and preprocessed land use and land cover data, the forest canopy height variable is extracted.

5. The method for estimating forest aboveground biomass based on open shared data according to claim 4, characterized in that: Based on the preprocessed global forest height data and preprocessed land use and land cover data, forest canopy height variables are extracted, including: Using the preprocessed land use and land cover data, extract the pixels with the field attributes of broad-leaved forest, coniferous forest or mixed broad-leaved forest in the forest area and construct the vector boundary range to obtain the forest vector range; The forest vector range is superimposed and clipped with the preprocessed global forest height data to obtain the forest canopy height variable of the forest area.

6. The method for estimating forest aboveground biomass based on open shared data according to claim 1, characterized in that: The variables in the model input variable set are screened based on correlation analysis to obtain the model input variable set after the rough selection of tree species, which specifically includes: The Kendall rank correlation coefficient between the variables in the model input variable set is calculated by the Kendall rank correlation coefficient method; In the model input variable set, variables whose correlation coefficients between each other are greater than the Kendall rank correlation coefficient are eliminated to obtain the model input variable set after the rough selection of tree species.

7. The method for estimating forest aboveground biomass based on open shared data according to claim 1, characterized in that: Based on variable replacement analysis, the variables in the model input variable set after the rough selection of tree species are screened twice to obtain the model input variable set after the fine selection of tree species, specifically including: According to all variables in the model input variable set after the rough selection of the tree species, a random forest model of the tree species is trained; For any variable, randomly change the value of the variable and keep the values ​​of other variables unchanged to obtain a new random forest model of the tree species; determining the importance of the variable based on a benchmark performance of a new random forest model of the tree species; In the model input variable set after the rough selection of tree species, variables whose importance is lower than the importance threshold are eliminated to obtain the model input variable set after the fine selection of tree species.

8. The method for estimating forest aboveground biomass based on open shared data according to claim 1 or 2, characterized in that: The spectral vegetation index set includes a ratio vegetation index, an enhanced vegetation index, a difference vegetation index, and a red edge normalized vegetation index established using different red edge bands; the red edge normalized vegetation index established using different red edge bands includes NDVI48, NDVI56, NDVI57, NDVI58a, NDVI67, NDVI68a, and NDVI78a; The biophysical variable set includes leaf area index, chlorophyll content, forest coverage, absorbable photosynthetically active radiation ratio and canopy water content; the terrain variable set includes slope, aspect, ground elevation, slope position index and terrain humidity index.

9. The method for estimating forest aboveground biomass based on open shared data according to claim 8, characterized in that: For the coniferous forest in the forest, the roughly selected model input variable set of the coniferous forest includes the forest canopy height variable, the difference vegetation index, NDVI56, the forest coverage rate, the ground elevation, the slope aspect and the slope; the finely selected model input variable set of the coniferous forest includes the forest coverage rate, the forest canopy height variable, the slope aspect and the difference vegetation index; for the broad-leaved forest in the forest, the roughly selected model input variable set of the broad-leaved forest includes the forest canopy height variable, the enhanced vegetation index, NDVI57, the ratio of photosynthetically active radiation that can be absorbed, the ground elevation, the slope aspect and the slope. elevation, slope and break index; the selected model input variable set for the broad-leaved forest includes forest canopy height variable, absorbable photosynthetically active radiation ratio and NDVI57; for the mixed forest in the forest, the roughly selected model input variable set for the mixed forest includes forest canopy height variable, enhanced vegetation index, NDVI78a, NDVI68a, NDVI48, slope aspect, slope and break index; the selected model input variable set for the mixed forest includes forest canopy height variable, break index, slope and NDVI48.

10. A forest aboveground biomass estimation system based on open shared data, characterized in that: include: An open shared data acquisition module, used to acquire and preprocess open shared data to obtain preprocessed open shared data; The open and shared data include satellite remote sensing data, digital elevation model data, global forest height data, and land use and land cover data; A model input variable extraction module is used to extract a model input variable set based on the preprocessed open shared data; the model input variable set includes a spectral vegetation index set, a biophysical variable set, a terrain variable set and a forest canopy height variable; A model input variable screening module is used to screen the variables in the model input variable set for any tree species in the forest based on correlation analysis to obtain the model input variable set after the rough selection of the tree species; and is also used to perform secondary screening on the variables in the model input variable set after the rough selection of the tree species based on variable replacement analysis to obtain the model input variable set after the fine selection of the tree species; A single tree species biomass estimation module is used to establish and train nonlinear estimation models of aboveground biomass of different tree species using the selected model input variable set to obtain trained nonlinear estimation models of aboveground biomass of different tree species; it is also used to obtain the aboveground biomass estimation results of each tree species in the forest based on the trained nonlinear estimation models of aboveground biomass of different tree species and the aboveground biomass model input variable set after optimization and selection; the aboveground biomass nonlinear estimation models of different tree species at least include a nonlinear estimation model of aboveground biomass of coniferous forest, a nonlinear estimation model of aboveground biomass of broad-leaved forest and a nonlinear estimation model of aboveground biomass of mixed forest; The forest total biomass estimation module is used to determine the forest aboveground biomass estimation result based on the aboveground biomass estimation result of each tree species in the forest.

Images