Forestry carbon sink dynamic metering method

Abstract

The invention discloses a forestry carbon sink dynamic metering method, and the method comprises the following steps: S100, obtaining a remote sensing image of a to-be-measured land; s200, processing the remote sensing image, and extracting feature data used for calculating carbon sink of the to-be-measured land; and S300, inputting the feature data into the calculation model, and calculating the carbon sink of the to-be-measured ground. According to the forestry carbon sink dynamic metering method, remote sensing data are obtained in a data processing mode, data in multiple aspects are extracted from remote sensing images, a calculation model is established, accurate monitoring of carbon sink is achieved, meanwhile, the defects that time and labor are consumed, and the range is small due to traditional carbon sink monitoring are greatly overcome, and the forestry carbon sink dynamic metering method is suitable for large-scale popularization and application. And the method is also suitable for monitoring forest carbon reserves and carbon sequestration on a large space scale.

Description

Technical Field

[0001] The invention belongs to the technical field of forestry data monitoring, and in particular relates to a dynamic measurement method for forestry carbon sinks. Background Art

[0002] Carbon sink refers to the offsetting of carbon dioxide emissions directly or indirectly generated by enterprises, groups or individuals within a certain period of time through afforestation, energy conservation and emission reduction. Traditional methods of monitoring carbon sinks mostly rely on field collection of plant data in the monitoring area. This method consumes a lot of manpower and material resources, causes great damage to forest vegetation, and is not suitable for monitoring forest carbon storage and carbon sinks on a large spatial scale. In addition, these methods may rely on theoretical formulas or previous experience in analysis, resulting in low accuracy. Another part uses machine learning methods to construct fitting models, but the fitting models constructed by these machine learning methods often have the problem of low model prediction accuracy. Summary of the invention

[0003] This application proposes a dynamic measurement method for forestry carbon sinks in response to the above technical problems. The specific technical solution is as follows:

[0004] The dynamic measurement method of forestry carbon sinks includes the following steps:

[0005] S100, obtaining a remote sensing image of the land to be measured;

[0006] S200, processing the remote sensing image to extract characteristic data for calculating the carbon sink of the measured area;

[0007] S300, inputting characteristic data into a calculation model to calculate the carbon sink of the land to be measured.

[0008] Preferably, the remote sensing images include satellite images, aerial images and radar images.

[0009] Preferably, the method for processing remote sensing images is:

[0010] Preprocess the remote sensing images by radiation correction, geometric correction, and noise removal, and put all data in the same coordinate system and resolution;

[0011] Extract features of remote sensing images and perform data normalization.

[0012] Preferably, the features extracted from the remote sensing images include vegetation index, leaf area index, biomass, crown height and diameter, soil organic carbon content, topographic factors and climate data.

[0013] Preferably, in step S300, a trained computing model is used to take multiple feature data as input to generate comprehensive features, and the computing model is used to train and verify the fused feature data with known carbon sink data, and carbon sink calculations are performed on the site to be tested based on the trained model.

[0014] Preferably, the computing model includes a linear regression model, a support vector machine model and a random forest model.

[0015] Preferably, between step S200 and step S300, a mapping relationship between features and models is also established, including the following steps:

[0016] Use correlation analysis, feature importance scoring, and recursive feature elimination to select the most appropriate features in the standardized data;

[0017] training a computational model using the selected features;

[0018] Evaluate the generalization ability and performance indicators of the model through cross-validation;

[0019] Use grid search or random search for hyperparameter tuning;

[0020] The best performing model and its corresponding features are selected.

[0021] The beneficial effects of the present invention are as follows: the dynamic measurement method of forestry carbon sinks of the present invention obtains remote sensing data by means of data processing, extracts data in multiple aspects from remote sensing images, establishes a calculation model, and thus realizes accurate monitoring of carbon sinks. At the same time, it greatly overcomes the shortcomings of traditional carbon sink monitoring such as time-consuming, labor-intensive, and small scope, and is also suitable for forest carbon storage and carbon sink monitoring on a large spatial scale. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Shown is a flow chart of the dynamic measurement method for forestry carbon sinks. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments.

[0024] In order to achieve the purpose of dynamic measurement of forestry carbon sinks, this application adopts the following steps:

[0025] S100, obtaining a remote sensing image of the land to be measured;

[0026] S200, processing the remote sensing image to extract characteristic data for calculating the carbon sink of the measured area;

[0027] S300, inputting characteristic data into a calculation model to calculate the carbon sink of the land to be measured.

[0028] Generally, in step S100, the remote sensing image needs to be preprocessed, including radiation correction, geometric correction, cropping and mosaicking. Radiation correction is used to eliminate the influence of the atmosphere and make the reflectance value of the image more accurate. Geometric correction is used to ensure that the image is consistent with the actual position on the ground and improve spatial accuracy. Cropping and mosaicking are used to crop the image according to the study area and stitch multiple images into a complete map.

[0029] Among them, the sources of remote sensing images include publicly available satellite images, aerial images and radar images. Multispectral or hyperspectral images provided by satellites such as Landsat and Sentinel-2 can provide wide coverage and periodic data updates. High-resolution images taken by drones or aircraft are suitable for detailed surveys in small areas, such as SAR (synthetic aperture radar) images, which can obtain data even when there is cloud cover.

[0030] In an implementation manner, the sources of remote sensing images include satellite images, aerial images and radar images, but different data sources have different data formats. Generally speaking, satellite images are usually stored in GeoTIFF, HDF or NetCDF format; aerial images may be JPEG2000, TIFF or other image formats; radar images may be in RAW, BIL (Band Sequential), BSQ (Band Sequential) or BIP (Band Sequential) format.

[0031] Therefore, in step S200, remote sensing data from different sources need to be unified into a standard format to facilitate subsequent processing and analysis.

[0032] First, the data needs to be preprocessed for radiation correction, geometric correction, and noise removal, and all data are in the same coordinate system and resolution.

[0033] Then, different features are extracted for different remote sensing data. In order to realize the calculation of carbon sinks in the measured area, the features to be extracted include vegetation index (NDVI, EVI), leaf area index (LAI), biomass (Biomass), crown height and diameter, soil organic carbon content, terrain factors (slope, aspect, altitude), and climate data (temperature, precipitation). Among them, vegetation index (NDVI, EVI), leaf area index (LAI), and soil organic carbon content are extracted from satellite images, biomass (Biomass), crown height and diameter are extracted from aerial images, and terrain factors (slope, aspect, altitude) and climate data (temperature, precipitation) are extracted from digital elevation model (DEM) and climate model data, respectively.

[0034] Next, data standardization is performed. Commonly used methods include: minimum-maximum standardization and Z-score standardization.

[0035] In step S300, a random forest model is used to take multiple feature data as input to generate comprehensive features, and the fused feature data and known carbon sink data are trained and verified using the random forest model. After that, carbon sink calculation is performed on the test site based on the trained model.

[0036] In the implementation of carbon sink calculation, there are many features that can be extracted from remote sensing images and many calculation models that can be used, but not all features are suitable for all models. Different models always have the most critical features that need to be used. How to establish a mapping relationship between features and models to ensure the selection of appropriate features, reduce the amount of calculation, and ensure the accuracy of the calculation results.

[0037] This application provides a method for evaluating the mapping relationship between features and models and optimizing the features adapted to the model, the steps are as follows:

[0038] Data preprocessing: standardize or normalize feature data;

[0039] Feature selection: Select the most appropriate features using methods such as correlation analysis, feature importance scoring, and recursive feature elimination (RFE);

[0040] Model training: train multiple models using the selected features and evaluate the performance of the models;

[0041] Performance evaluation: Evaluate the generalization ability and performance indicators of the model (such as accuracy, F1 score, AUC, etc.) through cross-validation;

[0042] Hyperparameter tuning: Use grid search or random search to tune hyperparameters to further optimize model performance;

[0043] Result selection: Select the best performing model and its corresponding features.

[0044] Specifically, data preprocessing includes standardization and normalization. Standardization standardizes all feature data so that the mean of each feature is 0 and the standard deviation is 1. Normalization normalizes all feature data to the interval [0,1].

[0045] Feature selection includes correlation analysis, feature importance scoring, and recursive feature elimination. Correlation analysis calculates the Pearson correlation coefficient between each feature and the target variable, and selects the top 10 features with the highest absolute value of the correlation coefficient. Feature importance scoring uses the random forest model training data to obtain the feature importance score of each feature. The top 10 features with the highest feature importance scores are selected. Recursive feature elimination (RFE) uses support vector machine (SVM) or other models for recursive feature elimination, and selects the 10 features that remain.

[0046] Select a model to train:

[0047] Random Forest: n_estimators = 100, max_depth = None, min_samples_split = 2, min_samples_leaf = 1, max_features = 'auto';

[0048] Logistic Regression: C = 1.0, penalty = 'l2', solver = 'liblinear';

[0049] Support vector machine (SVM): kernel = 'rbf', C = 1.0, gamma = 'scale'.

[0050] Five-fold cross validation was used to evaluate the generalization ability of the model; the evaluation indicators included accuracy, F1 score, and AUC (Area Under the ROC Curve).

[0051] Grid Search:

[0052] Grid search for random forest models with parameters ranging from:

[0053] n_estimators: [50, 100, 200]

[0054] max_depth: [None, 10, 20, 30]

[0055] min_samples_split: [2,5,10]

[0056] min_samples_leaf: [1,2,4]

[0057] max_features:['auto','sqrt']

[0058] Grid search for logistic regression models with parameters ranging from:

[0059] C: [0.1, 1.0, 10.0]

[0060] penalty:['l1','l2']

[0061] Grid search the support vector machine model with parameters ranging from:

[0062] C: [0.1, 1.0, 10.0]

[0063] gamma:['scale','auto'].

[0064] Evaluation results: Record the cross-validation scores of each model under different feature selection methods; select the model with the highest score and its corresponding feature selection method and features.

[0065] Hyperparameter selection: Record the cross-validation scores of each model under different hyperparameter settings; select the hyperparameter setting with the highest score.

[0066] Through the above steps, the mapping relationship between features and models can be systematically evaluated, and the most suitable combination of features and models can be selected to improve the accuracy and efficiency of carbon sink calculations.

[0067] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

[0068] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them.

Claims

1. A dynamic measurement method for forestry carbon sinks, characterized in that: The steps include: S100, obtaining a remote sensing image of the land to be measured; S200, processing the remote sensing image to extract characteristic data for calculating the carbon sink of the measured area; S300, inputting characteristic data into a calculation model to calculate the carbon sink of the land to be measured.

2. The forestry carbon sink dynamic measurement method according to claim 1 is characterized in that: The remote sensing images include satellite images, aerial images and radar images.

3. The method for dynamic measurement of forestry carbon sinks according to claim 1, characterized in that: The method for processing remote sensing images is: Preprocess the remote sensing images by radiation correction, geometric correction, and noise removal, and put all data in the same coordinate system and resolution; Extract features of remote sensing images and perform data normalization.

4. The method for dynamic measurement of forestry carbon sinks according to claim 3, characterized in that: The features extracted from remote sensing images include vegetation index, leaf area index, biomass, crown height and diameter, soil organic carbon content, topographic factors and climate data.

5. The method for dynamic measurement of forestry carbon sinks according to claim 1, characterized in that: In step S300, a trained computing model is used to take multiple feature data as input to generate comprehensive features, and the computing model is used to train and verify the fused feature data with known carbon sink data, and carbon sink calculations are performed on the site to be tested based on the trained model.

6. The method for dynamic measurement of forestry carbon sinks according to claim 1, characterized in that: The computing models include a linear regression model, a support vector machine model and a random forest model.

7. The method for dynamic measurement of forestry carbon sinks according to claim 1, characterized in that: The step between step S200 and step S300 also includes establishing a mapping relationship between features and models, including the following steps: Use correlation analysis, feature importance scoring, and recursive feature elimination to select the most appropriate features in the standardized data; training a computational model using the selected features; Evaluate the generalization ability and performance indicators of the model through cross-validation; Use grid search or random search for hyperparameter tuning; The best performing model and its corresponding features are selected.

Images