Vegetation carbon density and carbon reserve estimation method and device based on machine learning

Abstract

The invention discloses a vegetation carbon density and carbon reserve estimation method and device based on machine learning, and relates to the field of vegetation carbon density and carbon reserve accounting, the vegetation carbon density and carbon reserve estimation method based on machine learning mainly comprises the following steps: obtaining land cover and vegetation type data of a research area; setting a standard sample plot in the research area and investigating to obtain overground, earth surface and underground vegetation carbon density and carbon reserve data of the sample plot; performing preprocessing and information extraction on the land cover data of the research area to obtain remote sensing ecological factors of the sample plot; according to the sample plot remote sensing ecological factors and sample plot carbon density measured data, vegetation carbon density model construction and optimization are carried out, and an optimal estimation model is obtained; and according to the vegetation type and the vegetation area, utilizing the optimal estimation model to obtain the vegetation total carbon reserve in the research area. According to the vegetation carbon density and carbon reserve estimation method and device based on machine learning provided by the invention, rapid, dynamic and accurate estimation of multi-element vegetation carbon density and carbon reserve of regional scale forest grass wet and rough vegetation can be realized.

Description

Technical Field

[0001] The present invention relates to the field of vegetation carbon density and carbon stock accounting, and more specifically, to a vegetation carbon density and carbon stock estimation method and device based on machine learning. Background Art

[0002] Vegetation is an important component of natural ecosystems, the main ecological and environmental resources on Earth and the basis for human survival. It plays a vital role in regulating global climate, habitat quality, soil and water conservation, etc. Vegetation carbon storage is not only a basic parameter for studying the carbon cycle between ecosystems and the atmosphere, but also an important indicator reflecting the structure and service functions of community ecosystems. Accurately assessing vegetation carbon storage is an effective way to study the carbon sink function of vegetation and actively respond to climate change, and is of great significance to the development and trading of vegetation carbon sink projects.

[0003] At present, the main methods for estimating forest vegetation carbon storage are sample plot inventory method, model simulation method, micrometeorology method and remote sensing estimation method. Sample plot inventory method is to obtain biomass or forest stock through field survey by setting up sample plots, so as to calculate vegetation carbon storage. The data accuracy is high, but it is time-consuming and labor-intensive, and it is not suitable for large regional scales. Model simulation mainly simulates the carbon cycle process mechanism of the ecosystem to establish a carbon storage estimation model, which is suitable for large-scale estimation, but most models are developed abroad, the required parameters are difficult to obtain, and the results are difficult to verify. Micrometeorology method directly measures the dynamic changes of CO2 flux to calculate the carbon storage of forest ecosystems. It can realize long-term real-time observation and has high estimation accuracy, but the equipment cost is expensive, the data requirements are strict, the monitoring range is small, and it is not suitable for mountainous areas and areas with drastic landscape changes. Remote sensing estimation method obtains forest vegetation parameters through remote sensing data to obtain forest aboveground biomass and carbon storage. This method is convenient and fast, and can realize dynamic evaluation from sample plot scale to regional scale, but lacks estimation of carbon storage in the understory.

[0004] At present, the estimation of vegetation carbon storage mainly focuses on forest vegetation carbon storage, but terrestrial natural vegetation types also include grassland, wetland and desert vegetation. Secondly, vegetation carbon storage usually includes three parts: aboveground, surface and underground. Based on this, in order to scientifically and accurately evaluate regional vegetation carbon storage, it is very necessary to study a carbon storage estimation method that comprehensively considers vegetation types such as forests, grasslands, wetlands and deserts, has high accuracy and low cost, and can be promoted and applied. Summary of the invention

[0005] The purpose of the present invention is to provide a method and device for estimating vegetation carbon density and carbon storage based on machine learning, which can realize rapid, dynamic and accurate estimation of multi-factor vegetation carbon density and carbon storage in forest, grassland, wetland and wasteland at a regional scale.

[0006] The present invention provides a vegetation carbon density and carbon storage estimation method based on machine learning, comprising the following steps: S1: acquiring land cover data of a study area, the land cover data of the study area including land type, vegetation type and area of ​​the study area; S2: setting a standard sample plot according to the vegetation distribution status of the study area, investigating the standard sample plot, and obtaining the measured data of the carbon density and carbon storage of the aboveground, surface and underground vegetation of the sample plot; S3: preprocessing and extracting information of the land cover data of the study area, and obtaining the remote sensing ecological factor of the sample plot; S4: constructing and optimizing a vegetation carbon density model according to the remote sensing ecological factor of the sample plot and the measured data of the vegetation carbon density of the sample plot, and obtaining an optimal estimation model; S5: obtaining the total carbon storage of the vegetation in the study area by using the optimal estimation model according to the vegetation type and area.

[0007] Furthermore, step S2 specifically includes: S21: setting up a standard sample plot according to the vegetation distribution conditions in the study area, investigating the standard sample plot, taking above-ground, surface and underground plant samples, and obtaining the biomass of the standard sample plot according to the above-ground, surface and underground plant samples and the vegetation type and area; S22: taking the above-ground, surface and underground plant samples collected according to the standard sample plot investigation back to the indoor constant temperature drying to determine the carbon content, and obtaining the carbon content coefficient of the above-ground, surface and underground plants in the standard sample plot; S23: according to the above-ground, surface and underground plant carbon content coefficient of the standard sample plot and the biomass of the standard sample plot, combined with the sampling area, obtaining the measured data of the carbon density and carbon storage of the above-ground, surface and underground vegetation in the sample plot.

[0008] Furthermore, step S3 specifically includes: S31: preprocessing the land cover data of the study area to obtain processed land cover data of the study area; S32: extracting information from the processed land cover data of the study area to obtain remote sensing ecological factors of the sample site.

[0009] Further, step S31 specifically includes: S311: performing radiation calibration on the land cover data of the study area to obtain full-band and thermal infrared band radiation correction data; S312: performing atmospheric correction on the land cover data of the study area to obtain full-band atmospheric correction data; S313: performing image mosaicking on the land cover data of the study area to obtain mosaicked image data; S314: performing geometric correction on the land cover data of the study area to obtain geometrically corrected image data; S315: performing image cropping on the land cover data of the study area to obtain cropped image data; S316: performing image enhancement on the land cover data of the study area to obtain enhanced image data; S317: performing supervised classification on the land cover data of the study area to obtain classified image data; S318: obtaining the processed land cover data of the study area based on the full-band and thermal infrared band radiation correction data, full-band atmospheric correction data, mosaicked image data, geometrically corrected image data, cropped image data, enhanced image data, and classified image data.

[0010] Furthermore, step S32 specifically includes: S321: extracting band information of the processed land cover data of the study area to obtain the band information of the study area; S322: extracting vegetation index of the processed land cover data of the study area to obtain the vegetation index of the study area; S323: extracting texture features of the processed land cover data of the study area to obtain texture features of the study area.

[0011] Furthermore, step S322 specifically includes: extracting vegetation index from the processed land cover data of the study area to obtain vegetation index of the study area, such as formula: , , , , , , in, represents the normalized difference vegetation index; represents the ratio vegetation index; represents the difference vegetation index; represents the soil adjusted vegetation index; represents the enhanced vegetation index; Represents the normalized difference vegetation index of the green band; It is near infrared band; is the red band; is the soil adjustment coefficient. As an exemplary embodiment, the soil adjustment coefficient is 0.5; It is the blue band; It is the green band. Further, step S323 specifically includes: using the gray level co-occurrence matrix method to extract texture features from the processed land cover data of the study area to obtain texture features of the study area, such as formula: , , , , , , , , in, represents variance; Indicates homogeneity; Indicates contrast; Indicates heterogeneity; represents entropy; represents the angular second moment; Indicates relevance: represents the mean; , is the gray level co-occurrence matrix Line Gray value of the column; The gray value is probability; is the number of rows and columns of the remote sensing image; and for The mean of and is the standard deviation.

[0012] Furthermore, step S4 specifically includes: S41: performing correlation analysis on the remote sensing ecological factors of the sample plots and the measured data of vegetation carbon density of the sample plots to obtain remote sensing ecological factors with correlation and statistical significance; S42: based on the remote sensing ecological factors with correlation and statistical significance, using four machine learning models including multivariate linear regression, random forest, support vector machine and extreme gradient boosting, combined with vegetation carbon density, to establish multiple machine learning models; S43: based on the measured data of vegetation carbon density of the sample plots, obtaining training sets and validation sets, using the training sets to train multiple machine learning models, and obtaining multiple trained machine learning models; S44: using the validation set to perform accuracy evaluation on the multiple trained machine learning models to obtain the optimal estimation model.

[0013] Furthermore, step S5 specifically includes: using the optimal estimation model and vegetation types and their areas, calculating the above-ground, surface and underground vegetation carbon storage of tree forests, bamboo forests, economic forests and shrub forests in the study area to obtain the total vegetation carbon storage in the study area.

[0014] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the above-mentioned vegetation carbon density and carbon storage estimation method based on machine learning are implemented.

[0015] The implementation of the vegetation carbon density and carbon storage estimation method and device based on machine learning provided by the present invention has the following beneficial effects: The present invention obtains land cover and vegetation type data of a study area; sets up standard sample plots in the study area and conducts surveys to obtain the carbon density and carbon storage data of the aboveground, surface and underground vegetation of the sample plots; preprocesses the land cover data of the study area and extracts information to obtain remote sensing ecological factors of the sample plots; constructs and optimizes a vegetation carbon density model based on the remote sensing ecological factors of the sample plots and the measured carbon density data of the sample plots to obtain an optimal estimation model; estimates the aboveground, surface and underground carbon density of various types of vegetation in the region based on the vegetation type and its area using the optimal carbon density estimation model to obtain the total carbon storage of the vegetation in the study area, thereby realizing rapid, dynamic and accurate estimation of the carbon density and carbon storage of multi-factor vegetation in forest, grassland, wetland and wasteland at a regional scale. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which: Figure 1 It is a flow chart of a vegetation carbon density and carbon storage estimation method based on machine learning provided by the present invention; Figure 2 It is a schematic diagram of sample plot setting provided by the present invention; Figure 3 It is a structural block diagram of the computer device provided by the present invention. DETAILED DESCRIPTION

[0017] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0018] Figure 1 A schematic diagram of a vegetation carbon density and carbon stock estimation method based on machine learning in this embodiment is shown. In this embodiment, the vegetation carbon density and carbon stock estimation method based on machine learning includes the following steps: S1: Obtain the land cover data of the study area, which includes the land type, vegetation type and area of ​​the study area; As an exemplary embodiment, the vegetation types in the study area include tree forest, bamboo forest, economic forest, shrub forest, grassland, wetland, and wasteland; In an exemplary embodiment, step S1 includes land cover data collection: downloading the latest Landsat8 remote sensing image data (spatial resolution 30 meters), land use data and digital elevation data of the study area from the geospatial data cloud, and dividing the land types of the study area into tree forest land, bamboo forest land, economic forest land, shrub forest land, grassland, wetland, wasteland, etc. according to the vegetation type, so as to form a vegetation type and area layer of the study area; S2: Set up standard plots according to the distribution of vegetation in the study area, conduct surveys on the standard plots, and obtain measured data on the carbon density and carbon storage of vegetation aboveground, on the surface and underground in the plots; As an exemplary embodiment, step S2 specifically includes: S21: setting a standard plot according to the distribution of vegetation in the study area, investigating the standard plot, taking above-ground, surface and underground plant samples, and obtaining the biomass of the standard plot according to the above-ground, surface and underground plant samples and the vegetation type and area; S22: taking the above-ground, surface and underground plant samples collected according to the survey of the standard plot back to the indoor constant temperature drying to determine the carbon content, and obtaining the carbon content coefficient of the above-ground, surface and underground plants in the standard plot; S23: according to the above-ground, surface and underground plant carbon content coefficient of the standard plot and the biomass of the standard plot, combined with the sampling area, obtaining the measured data of the carbon density and carbon storage of the above-ground, surface and underground vegetation in the plot; In an exemplary embodiment, standard plot data collection is performed in step S2, specifically including: Sample plot setting and investigation: Standard sample plots such as arbor forest, bamboo forest, economic forest, shrub forest, and herbaceous plantation were set up in the study area. The number of standard sample plots was determined according to the size of the study area. The number of sample plots was distributed as evenly as possible in the study area and covered various vegetation types such as forest, grassland, wetland, and wasteland. The aboveground, surface, and underground root biomass of the standard sample plots were investigated. The forest, grassland, and wetland survey sample plots were designed as an integrated composite sample plot, consisting of one area of ​​0.06-0.08hm 2 square plot (the size and shape of the fixed plots in each province of the Ninth National Forest Resources Inventory remain unchanged, and new plots may be added as needed), 1 100m 2 Large plot (10m×10m, when the shrub crown is small and evenly distributed, it can be reduced to 5m×5m), 3 1m 2 The composition of the yield test sample plot (such as Figure 2 As shown); among which, 0.06-0.08hm 2 The square plots were used to investigate forest vegetation such as arbor forests, bamboo forests and economic forests. 2 Large plots are used to survey shrubs, young trees under the forest, and tall shrubs. 1m 2 Small sample plots are used to survey herbaceous plants, small shrubs, litter and root systems; each tree in the arbor forest, bamboo forest and economic forest is measured, and data such as tree species name, breast diameter, tree height and number of plants are recorded; the shrub forest adopts the standard tree survey method, collects trunk, branch, leaf and root samples, and records shrub type, coverage, average height, number of plants and other data; the herb, litter, root system and small shrub are surveyed by the full harvest method, recording the herb name, coverage, average height and other data, and only roots with a diameter of 2mm or more are collected; Calculation of the biomass of trees and bamboo forests: calculated according to the allometric growth equation or the biomass expansion factor method; calculation of the biomass of economic forests, shrubs, herbs, surface vegetation (litter, dead standing trees, dead fallen trees) and roots: the above-ground part is harvested, and the root system is excavated to remove the soil. All plant samples in the sample plot are collected, and the fresh weight is recorded. The samples are brought back to the laboratory and baked at a constant temperature of 105℃ to a constant weight to obtain the dry weight of the biomass; Determination of carbon content: The average carbon content coefficient of trees and bamboo forests is determined based on empirical values, while the carbon content coefficient of shrubs, economic forests, herbs, litter and roots is determined in the laboratory through plant samples; Calculation of carbon density and carbon storage: Based on the sample area, vegetation biomass and carbon content coefficient, calculate the sum of the carbon density and carbon storage of the aboveground vegetation, surface litter and underground roots of the standard sample plot, so as to obtain the carbon density and carbon storage values ​​of the standard sample plot; As an exemplary embodiment, step S2 further includes: collecting data on carbon density and carbon storage of each standard sample plot, importing the data into ArcGIS, and establishing a Geodatabase of carbon density and carbon storage of the sample plot; S3: Preprocessing and information extraction of land cover data in the study area to obtain remote sensing ecological factors of the sample plots; As an exemplary embodiment, step S3 specifically includes: S31: preprocessing the land cover data of the study area to obtain processed land cover data of the study area; As an exemplary embodiment, step S31 includes: using ENVI software to perform radiation calibration, atmospheric correction, image mosaic, geometric correction, image clipping, image enhancement and supervised classification on the acquired Landsat8 remote sensing data and DEM terrain data of the study area, and performing supervised classification on the processed research data, specifically: S311: Perform radiometric calibration on the land cover data of the study area to obtain full-band and thermal infrared band radiometric correction data; specifically, in step S311, radiometric calibration is performed to eliminate and reduce the nonlinear distortion and image noise problems existing in the original remote sensing image. The Radiometric Calibration tool in the ENVI5.6 software is used to extract the imaging time and solar altitude angle in the remote sensing image, and the digital quantitative value (DN) of the image is converted into corresponding physical quantities such as radiometric brightness value, reflectivity and surface temperature, so as to complete the full-band and thermal infrared band radiometric correction of the remote sensing image; S312: Perform atmospheric correction on the land cover data of the study area to obtain full-band atmospheric correction data; specifically, in step S312, atmospheric correction is performed by using the FLAASH standard atmospheric correction model in the ENVI5.6 software, importing the radiometrically calibrated data, and completing the full-band atmospheric correction of the remote sensing image by setting a series of parameters such as the sensor type, the average ground elevation of the image area, the remote sensing image imaging time, and the average elevation of the study area; S313: Perform image mosaicking on the land cover data of the study area to obtain mosaicked image data; specifically, in step S313, image mosaicking is performed by using the Mosaicking / SeamlessMosaic tool in the ENVI5.6 software according to the image geographic coordinates; S314: geometrically correct the land cover data of the study area to obtain geometrically corrected image data; specifically, in step S314, geometric correction is performed to correct and eliminate the geometric distortion of the remote sensing image through a mathematical model. In this example, geometric correction is performed by using control points (or reference images); S315: perform image cropping on the land cover data of the study area to obtain cropped image data; specifically, in step S315, image cropping is performed, the shp file of the study area is imported into ENVI, and the Regions of Interest / Subset Data from ROIs tool of ENVI5.6 software is used to crop the remote sensing image data based on the vector mask data of the study area in the shp format; S316: image enhancement is performed on the land cover data of the study area to obtain enhanced image data; specifically, image enhancement is performed in step S316, and standard false color image synthesis is used to achieve the object enhancement effect and identify the target object; S317: supervised classification is performed on the land cover data of the study area to obtain classified image data; specifically, supervised classification is performed in step S317, ENVI5.6 software is used to select a certain number of sample ROIs, and different pixels are classified according to certain calculation rules; more than 100 samples are selected for each image, and accuracy verification is started after the classification results are obtained, until the classification is relatively accurate and meets the research needs; S318: Obtain the processed land cover data of the study area based on the full-band and thermal infrared band radiation correction data, full-band atmospheric correction data, mosaicked image data, geometrically corrected image data, cropped image data, enhanced image data, and classified image data; S32: extract information from the processed land cover data of the study area to obtain the remote sensing ecological factors of the sample plot; As an exemplary embodiment, step S32 specifically includes: S321: extracting band information from the processed land cover data of the study area to obtain band information of the study area; As an exemplary embodiment, the band information of the study area includes a coast band, a blue band, a green band, a red band, a near infrared wave, a shortwave infrared 1 and a shortwave infrared 2; S322: extracting vegetation index from the processed land cover data of the study area to obtain vegetation index of the study area; As an exemplary embodiment, step S322 specifically includes: extracting vegetation index from the processed land cover data of the study area to obtain vegetation index of the study area, such as formula: , , , , , , in, represents the normalized difference vegetation index; represents the ratio vegetation index; represents the difference vegetation index; represents the soil adjusted vegetation index; represents the enhanced vegetation index; Represents the normalized difference vegetation index of the green band; It is near infrared band; is the red band; is the soil adjustment coefficient; It is the blue band; It is the green band; S323: extracting texture features from the processed land cover data of the study area to obtain texture features of the study area; As an exemplary embodiment, step S323 specifically includes: using the gray level co-occurrence matrix method to extract texture features from the processed land cover data of the study area to obtain the texture features of the study area, such as the formula: , , , , , , , , in, represents variance; Indicates homogeneity; Indicates contrast; Indicates heterogeneity; represents entropy; represents the angular second moment; Indicates relevance: represents the mean; , is the gray level co-occurrence matrix Line Gray value of the column; The gray value is probability; is the number of rows and columns of the remote sensing image; and for The mean of and is the standard deviation; S4: Based on the remote sensing ecological factors of the sample plot and the measured data of vegetation carbon density of the sample plot, the vegetation carbon density model is constructed and optimized to obtain the optimal estimation model; As an exemplary embodiment, step S4 specifically includes: S41: Conduct correlation analysis on the remote sensing ecological factors of the sample plots and the measured data of vegetation carbon density of the sample plots to obtain remote sensing ecological factors with correlation and statistical significance; As an exemplary embodiment, step S41 specifically includes: correlation analysis, using SPSS software to perform correlation analysis on the carbon density of the tree forest, bamboo forest, economic forest, and shrub forest in the sample plot and the extracted remote sensing ecological factors, and selecting remote sensing ecological factors with correlation and statistical significance to participate in the construction of the carbon density accounting model; S42: Based on the remote sensing ecological factors with correlation and statistical significance, four machine learning models, including multivariate linear regression, random forest, support vector machine and extreme gradient boosting, were used to establish multiple machine learning models in combination with vegetation carbon density; As an exemplary embodiment, step S42 specifically includes: S421: Multiple linear regression model is a supervised learning algorithm, mainly used to solve regression problems. Its calculation formula can be expressed as:

[0019] In the formula, is the dependent variable, here vegetation carbon density; is the independent variable, here it is the remote sensing ecological factor; is the intercept; is the model regression coefficient, which indicates the influence of the independent variable on the dependent variable; S422: Random forest model is an ensemble learning method based on decision trees. It improves the accuracy and robustness of classification by constructing multiple decision trees and voting on their prediction results. It is mainly used for classification and regression problems and does not involve specific mathematical formulas. Its core steps are to construct multiple decision trees and then integrate the prediction results of the decision trees. A part of samples (vegetation carbon density) and features (remote sensing ecological factors) are randomly selected from the data set to obtain multiple sub-data sets and construct multiple decision trees. The prediction results of all decision trees are voted or the prediction results are averaged to obtain the final result.

[0020] In the formula, is the prediction result of random forest, is the total number of decision trees in the random forest, It is The prediction results of a decision tree; S423: Support vector machine model is a powerful supervised learning algorithm that maximizes the separation of two types of data by finding an optimal hyperplane, making classification more accurate. It is mainly used for classification problems and is suitable for classifiers trained with small sample models. The decision function is:

[0021] In the formula, is the input vector (sample feature vector); is the weight vector, is the bias top (scalar), a sample satisfies , it is judged as a positive sample, otherwise it is judged as a negative sample; S424: Extreme Gradient Boosting Model, a machine learning ensemble algorithm, is based on the gradient boosting algorithm. It gradually builds a series of weak learners and combines them into a strong learner to improve prediction performance. It is used to solve problems such as regression, classification, and ranking. It has the advantages of fast computing speed, good results, easy parameter adjustment, and massive data processing. Compared with other machine learning algorithms, it has stronger interpretability. The objective function of the model is:

[0022] In the formula, the first part on the right side of the equation is the loss function, and the second part is the regularization term. The loss function reveals the training error of the model (that is, the error between the predicted value and the measured value), and the regularization defines the complexity to avoid overfitting. It should be noted that S421-S424 are not executed in chronological order, and the above four machine learning models are built and run independently; S43: obtaining a training set and a validation set based on the measured data of vegetation carbon density in the sample plot, and using the training set to train multiple machine learning models to obtain multiple trained machine learning models; As an exemplary embodiment, step S43 specifically includes: randomly selecting 80% of the measured data in the study area as a training set and 20% as a validation set, training multiple machine learning models, and obtaining multiple trained machine learning models; S44: Use the validation set to evaluate the accuracy of multiple trained machine learning models and obtain the optimal estimation model; As an exemplary embodiment, step S44 specifically includes: comparing the predicted value calculated by the model with the actual measured value of the sample plot, using the determination coefficient (R 2 ) and root mean square error (RMSE) to evaluate the model accuracy and determine the optimal estimation model;

[0023]

[0024] in, It is the sum of the measured values ​​of aboveground, surface and underground carbon density (carbon storage); It is the sum of the model-predicted values ​​of aboveground, surface, and underground carbon density (carbon storage); The average value of the measured values ​​of aboveground, surface and underground carbon density (carbon storage); n is the number of samples. 2 The closer the value is to 1, the better the model fit is, and the smaller the RMSE value is, the higher the estimation accuracy is; S5: Based on the vegetation type and its area, the total carbon stock of vegetation in the study area was obtained using the optimal estimation model; As an exemplary embodiment, vegetation carbon density and carbon storage are estimated in step S5: using the optimal estimation model and vegetation type and area, the above-ground, surface and underground vegetation carbon storage of tree forests, bamboo forests, economic forests and shrub forests in the study area are calculated to obtain the total vegetation carbon storage in the study area.

[0025] In some embodiments, the above-mentioned vegetation carbon density and carbon storage estimation method based on machine learning can also be implemented in the following manner.

[0026] In this embodiment, the vegetation carbon density and carbon storage estimation method based on machine learning includes: Step 1. Obtain vegetation image data: Obtain the spatial distribution data of vegetation types at 30 meters in the study area from the geospatial data cloud website, classify the vegetation into tree forest, bamboo forest, shrub forest, economic forest, grassland, wetland, and wasteland, and establish vegetation type and area layers; Step 2. Sample plot layout and data collection: According to the size of the study area, the number of standard sample plots shall be determined and the sample plots shall be distributed as evenly as possible in the forest, grass and wetland vegetation coverage area of ​​the study area. The standard sample plots shall cover tree forest sample plots, bamboo forest sample plots, shrub forest sample plots, economic forest sample plots and herbaceous sample plots at the same time; the tree species, tree height, diameter at breast height, canopy density, number of bamboo trees and other indicators in the sample plots shall be investigated, and the biomass shall be converted into biomass according to the biomass expansion factor, and the biomass of shrub forests, economic forests, herbs and other vegetation shall be investigated; Calculation of tree and bamboo forest biomass: calculated according to the allometric growth equation or biomass expansion factor method; Calculation of biomass of shrubs, economic forests, herbs, ground vegetation (litter, dead standing trees, dead fallen trees) and roots: The aboveground parts were harvested, and the roots were excavated to remove mud. All plant samples in the sample plot were collected, and the fresh weight was recorded. The samples were brought back to the laboratory and baked at a constant temperature of 105℃ to a constant weight to obtain the dry weight of the biomass; Determination of carbon content: The average carbon content coefficient of trees and bamboo forests is determined based on empirical values, while the carbon content coefficient of shrubs, economic forests, herbs, litter and roots is determined in the laboratory through samples; Calculation of carbon density and carbon storage: Calculate the carbon density and carbon storage of the standard plot including aboveground, surface and underground vegetation based on the sample area, plant biomass and carbon content coefficient; Count the carbon density and carbon storage data of each plot, import them into ArcGIS, and establish a Geodatabase of carbon density and carbon storage of the plot; Step 3. Remote sensing image data processing: The Landsat remote sensing data and DEM terrain data of the study area are pre-processed by using ENVI software, such as geometric accuracy correction, radiation calibration, atmospheric correction, mosaicking, and cropping, and supervised classification is performed on the processed research data; Step 4. Construction and optimization of vegetation carbon density model: extract the remote sensing ecological factors of the sample plot, conduct correlation analysis, select the remote sensing ecological factors with correlation and statistical significance to participate in the construction of the carbon density inversion model, take the vegetation carbon density as the dependent variable and the remote sensing ecological factors of the sample plot as the independent variable, use four machine learning algorithms including multivariate linear regression, random forest, support vector machine and extreme gradient boosting to conduct vegetation carbon density traversal modeling, divide the data samples into training set and test set, input the model for training and testing, and determine the optimal estimation model based on the determination coefficient and root mean square error; Step 5. Estimation of vegetation carbon density and carbon storage: According to the vegetation carbon density model and vegetation type area, the total carbon storage of vegetation in the study area is calculated.

[0027] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the above-mentioned vegetation carbon density and carbon storage estimation method based on machine learning are implemented.

[0028] like Figure 3 As shown, the computer device 120 may include: at least one processor 121, such as a central processing unit (CPU), at least one communication interface 123, a memory 124, and at least one communication bus 122. Among them, the communication bus 122 is used to realize the connection and communication between these components. Among them, the communication interface 123 may include a display screen (Display), a keyboard (Keyboard), and the optional communication interface 123 may also include a standard wired interface and a wireless interface. The memory 124 may be a high-speed random access memory (Random Access Memory, RAM), or a non-volatile memory (non-volatile memory), such as at least one disk storage. The memory 124 may also be at least one storage device located away from the aforementioned processor 121. Among them, the memory 124 stores an application program, and the processor 121 calls the program code stored in the memory 124 to perform any of the above method steps. Among them, the communication bus 122 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The communication bus 122 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3Only one line is used to represent it, but it does not mean that there is only one bus or one type of bus. Among them, the memory 124 may include a volatile memory (volatile memory), such as a random-access memory (random-access memory, RAM); the memory may also include a non-volatile memory (non-volatile memory), such as a flash memory (flash memory), a hard disk drive (hard disk drive, HDD) or a solid-state drive (solid-state drive, SSD); the memory 124 may also include a combination of the above-mentioned types of memory. Among them, the processor 121 may be a central processing unit (central processing unit, CPU), a network processor (network processor, NP) or a combination of CPU and NP. Among them, the processor 121 may further include a hardware chip. The above-mentioned hardware chip may be an application-specific integrated circuit (application-specific integrated circuit, ASIC), a programmable logic device (programmable logic device, PLD) or a combination thereof. The above-mentioned PLD may be a complex programmable logic device (complex programmable logic device, CPLD), a field-programmable gate array (field-programmable gate array, FPGA), a generic array logic (generic array logic, GAL) or any combination thereof. Optionally, the memory 124 is also used to store program instructions. The processor 121 can call the program instructions to implement the vegetation carbon density and carbon storage estimation method based on machine learning as in this embodiment.

[0029] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A method for estimating vegetation carbon density and carbon storage based on machine learning, characterized in that: The following steps are involved: S1: Obtaining land cover data of the study area, wherein the land cover data of the study area includes land type, vegetation type and area of ​​the study area; S2: Set up standard plots according to the distribution of vegetation in the study area, conduct surveys on the standard plots, and obtain measured data on the carbon density and carbon storage of vegetation aboveground, on the surface and underground in the plots; S3: Preprocessing and information extraction of the land cover data of the study area to obtain remote sensing ecological factors of the sample plot; S4: constructing and optimizing a vegetation carbon density model based on the remote sensing ecological factors of the sample plot and the measured data of vegetation carbon density of the sample plot to obtain an optimal estimation model; S5: According to the vegetation type and its area, the total carbon storage of vegetation in the study area is obtained using the optimal estimation model.

2. The method for estimating vegetation carbon density and carbon storage based on machine learning according to claim 1, characterized in that: Step S2 specifically includes: S21: Setting a standard plot according to the distribution of vegetation in the study area, investigating the standard plot, taking above-ground, surface and underground plant samples, and obtaining the biomass of the standard plot according to the above-ground, surface and underground plant samples and the vegetation type and area; S22: The above-ground, surface and underground plant samples collected from the standard plot survey are brought back to the room for constant temperature drying to determine the carbon content, and the carbon content coefficients of the above-ground, surface and underground plants in the standard plot are obtained; S23: According to the carbon content coefficient of the aboveground, surface and underground plants in the standard sample plot and the biomass of the standard sample plot, combined with the sampling area, the measured data of the carbon density and carbon storage of the aboveground, surface and underground vegetation in the sample plot are obtained.

3. The vegetation carbon density and carbon storage estimation method based on machine learning according to claim 1 is characterized in that: Step S3 specifically includes: S31: preprocessing the land cover data of the study area to obtain processed land cover data of the study area; S32: extracting information from the processed land cover data of the study area to obtain remote sensing ecological factors of the sample plot.

4. The method for estimating vegetation carbon density and carbon storage based on machine learning according to claim 3 is characterized in that: Step S31 specifically includes: S311: performing radiation calibration on the land cover data of the study area to obtain full-band and thermal infrared band radiation correction data; S312: performing atmospheric correction on the land cover data of the study area to obtain full-band atmospheric correction data; S313: performing image mosaicking on the land cover data of the study area to obtain mosaicked image data; S314: geometrically correcting the land cover data of the study area to obtain geometrically corrected image data; S315: performing image cropping on the land cover data of the study area to obtain cropped image data; S316: performing image enhancement on the land cover data of the study area to obtain enhanced image data; S317: performing supervised classification on the land cover data of the study area to obtain classified image data; S318: Obtain processed land cover data of the study area based on the full-band and thermal infrared band radiation correction data, full-band atmospheric correction data, mosaicked image data, geometrically corrected image data, cropped image data, enhanced image data, and classified image data.

5. The method for estimating vegetation carbon density and carbon storage based on machine learning according to claim 3, characterized in that: Step S32 specifically includes: S321: extracting band information from the processed land cover data of the study area to obtain band information of the study area; S322: extracting vegetation index from the processed land cover data of the study area to obtain vegetation index of the study area; S323: Extracting texture features from the processed land cover data of the study area to obtain texture features of the study area.

6. The method for estimating vegetation carbon density and carbon storage based on machine learning according to claim 5, characterized in that: Step S322 specifically includes: extracting vegetation index from the processed land cover data of the study area to obtain vegetation index of the study area, such as formula: ， ， ， ， ， ， in, represents the normalized difference vegetation index; represents the ratio vegetation index; represents the difference vegetation index; represents the soil adjusted vegetation index; represents the enhanced vegetation index; Represents the normalized difference vegetation index of the green band; It is near infrared band; is the red band; is the soil adjustment coefficient; It is the blue band; It is the green band.

7. The method for estimating vegetation carbon density and carbon storage based on machine learning according to claim 5, characterized in that: Step S323 specifically includes: using the gray level co-occurrence matrix method to extract texture features from the processed land cover data of the study area to obtain texture features of the study area, such as formula: ， ， ， ， ， ， ， ， in, represents variance; Indicates homogeneity; Indicates contrast; Indicates heterogeneity; represents entropy; represents the angular second moment; Indicates relevance: represents the mean; , is the gray level co-occurrence matrix Line Gray value of the column; The gray value is probability; is the number of rows and columns of the remote sensing image; and for The mean of and is the standard deviation.

8. The method for estimating vegetation carbon density and carbon storage based on machine learning according to claim 1, characterized in that: Step S4 specifically includes: S41: performing correlation analysis on the remote sensing ecological factors of the sample plot and the measured data of vegetation carbon density of the sample plot to obtain remote sensing ecological factors with correlation and statistical significance; S42: Based on the remote sensing ecological factors with correlation and statistical significance, four machine learning models including multivariate linear regression, random forest, support vector machine and extreme gradient boosting were used in combination with vegetation carbon density to establish multiple machine learning models; S43: obtaining a training set and a validation set according to the measured data of vegetation carbon density of the sample plot, and using the training set to train the multiple machine learning models to obtain multiple trained machine learning models; S44: Use the validation set to evaluate the accuracy of the trained multiple machine learning models to obtain the optimal estimation model.

9. The method for estimating vegetation carbon density and carbon storage based on machine learning according to claim 1, characterized in that: Step S5 specifically includes: using the optimal estimation model and the vegetation types and their areas, calculating the above-ground, surface and underground vegetation carbon storage of tree forests, bamboo forests, economic forests and shrub forests in the study area to obtain the total vegetation carbon storage in the study area.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the vegetation carbon density and carbon stock estimation method based on machine learning are implemented as described in any one of claims 1-8.

Images