A method for evaluating the carbon sink potential after ecological restoration of quarry benches

Abstract

The present invention provides a method for evaluating the carbon sink potential after ecological restoration of a quarry pit. First, ecological parameters of the quarry pit and the surrounding mountains are collected by using an unmanned aerial vehicle, and the measured carbon density values of the quarry pit and the surrounding mountains are obtained by using publicly released carbon density data. Secondly, the area in the surrounding mountains of the quarry pit that has not been disturbed by humans is used as a reference ecosystem, and the ecological parameters of the reference ecosystem are used as variable factors, and the measured carbon density value of the reference ecosystem is used as the model supervision value. The variable factors and the model supervision value form a data set for training a machine learning model. Finally, by using the trained qualified machine learning model, with the ecological parameters of the quarry pit as variable factors, the predicted carbon density value after ecological restoration of the quarry pit is output, and the carbon sink potential of the quarry pit is obtained by subtracting the measured carbon density value from the predicted carbon density value after ecological restoration. This method can meet the accuracy requirements for rapid estimation of the carbon sink potential after quarry pit restoration.

Description

Technical Field

[0001] The present invention relates to the technical field of land reclamation and ecological restoration, and particularly relates to a method for evaluating the carbon sink potential after ecological restoration of a quarry. Background Art

[0002] Carbon sink refers to a CO2 pool formed when the carbon intake of an ecosystem exceeds its release capacity, that is, the net ecosystem productivity NEP (Net Ecosystem Productivity). Its calculation formula is NEP = GPP - ER, where GPP refers to the gross primary productivity, that is, the total carbon fixed by plant photosynthesis, and ER refers to the ecosystem respiration. The calculation formula of NEP shows that the carbon sink value depends on the combined action of the plant carbon absorption process and the respiration process. The carbon sink potential refers to the amount of CO2 that an ecosystem can absorb and store after ecological restoration.

[0003] During the rapid development of our country, a large number of mineral resources have been exploited and utilized, which has led to a series of ecological and environmental problems. Therefore, after the exploitation of mineral resources, the environmental improvement and restoration work of abandoned mines has received extensive attention. As a quarry formed by open-pit mining, a quarry is a common type of abandoned mine landform. Through afforestation, grassland restoration, and biodegradation, etc., its ecological restoration can increase the carbon sequestration capacity of vegetation and soil, enabling the restored quarry to play a direct carbon sequestration role, absorbing a large amount of carbon dioxide in the atmosphere to increase the carbon sink, and having great potential for exploration. Therefore, it is of great significance to evaluate the carbon sink capacity of a quarry and give the planning and design of ecological restoration.

[0004] Currently, there are various technologies for evaluating the CO2 flux, carbon storage, or carbon density of an ecosystem. For example, an inverse model constructed using atmospheric CO2 observations and atmospheric transport data is used to infer the average spatial distribution of CO2 flux; the resource inventory method is used to calculate the current carbon sink of forests based on fossil fuel emissions, land use change sources, and data on ocean and atmospheric carbon sinks; a multiple stepwise regression model and an ordinary Kriging residual correction model are constructed by combining plot survey data and Landsat OLI remote sensing images to evaluate the carbon storage of urban forests; an improved BIOMEBGC model is used to estimate the average aboveground carbon density of bamboo forests by combining plot surveys and remote sensing data.

[0005] However, the above-mentioned existing technologies only evaluate the CO2 flux, carbon storage, or carbon density at the current time point, do not directly evaluate the future carbon sink potential of the ecosystem, and cannot efficiently and highly accurately evaluate the carbon sink potential of a quarry after ecological restoration. Therefore, in order to objectively reflect the benefits generated by the ecological restoration of a quarry and provide a reference for the ecological restoration planning and design of a quarry, this application proposes a method for evaluating the carbon sink potential of a quarry after ecological restoration. Summary of the Invention

[0006] To solve the problems existing in the above-mentioned prior art, the present invention provides a method for evaluating the carbon sink potential after ecological restoration of quarry rock faces.

[0007] The technical solution of the present invention is as follows:

[0008] A method for evaluating the carbon sink potential after ecological restoration of quarry rock faces, comprising the following steps:

[0009] S1, obtaining the ecological parameters and the measured carbon density values of the quarry rock face and the reference ecosystem;

[0010] S2, based on the machine learning method and the ecological parameters and the measured carbon density values of the reference ecosystem, establishing a random forest regression model for predicting the carbon density of the quarry rock face, and training the random forest regression model;

[0011] S3, using the ecological parameters of the quarry rock face as variable factors, and obtaining the predicted carbon density value of the quarry rock face through the trained random forest regression model;

[0012] S4, calculating the difference between the predicted carbon density value and the measured carbon density value of the quarry rock face, and the difference is the carbon sink potential of the quarry rock face.

[0013] Further, the ecological parameters include terrain parameters and soil type parameters.

[0014] Further, the method for obtaining the ecological parameters is: obtaining the remote sensing images of the quarry rock face and the reference ecosystem, and extracting the terrain parameters and soil type parameters from the remote sensing images.

[0015] Further, the remote sensing image is a high-resolution multi-spectral remote sensing image; the terrain parameters are obtained by processing with PhotoScan or ArcGIS software, and the processing path is: data import → initialization processing → point cloud encryption → generation of digital surface model and orthophoto → generation of high-precision digital orthophoto (DOM), digital surface model (DSM) and digital elevation model (DEM); the soil type parameters are obtained by visual interpretation based on regional conditions.

[0016] Further, the random forest regression model is constructed and trained using a data set, the data set includes model supervision values and variable factors, the model supervision values are the measured carbon density values of the reference ecosystem, and the variable factors are the ecological parameters of the reference ecosystem.

[0017] Further, the training steps of the random forest regression model are:

[0018] a. In the original dataset, use the bootstrapping method to randomly draw K new sample sets with replacement, and use the new samples to construct K classification and regression trees;

[0019] b. Assume there are n features. At each node of each classification and regression tree, select Mtry features, where Mtry features < n. Calculate the squared error of each Mtry feature, and select the Mtry feature with the strongest regression ability for node splitting based on the squared error value;

[0020] c. Directly generate a random forest composed of multiple classification and regression trees to perform regression on new data, and the value obtained by weighted averaging the regression results is the final value for output; the weights for weighted averaging are determined according to the out-of-bag data error of each classification and regression tree.

[0021] Furthermore, the regression decision method for the new data is as follows:

[0022]

[0023] In the formula, H(X) is the identification combined regression algorithm; h i is the single decision tree regression algorithm; W i is the weight of the i-th tree; / is the discriminant function. If h i (X) is equal to the target variable Y, the value of I is 1, otherwise it is 0; Y is the target variable; K is the number of trees in the random forest.

[0024] Furthermore, in the random forest regression model, set the Ntree value and the Mtry value. The Ntree value is set to 500, and the Mtry value is set to 2.

[0025] Furthermore, it is characterized in that the expression of the random forest regression model is:

[0026] CD = F R (AS, SL, SO)

[0027] In the formula, CD is the predicted value of the quarry carbon density estimated based on the random forest regression model; F R is the random forest machine learning algorithm; AS is the terrain aspect; SL is the terrain slope; SO is the soil type.

[0028] Furthermore, the dataset is also used to evaluate the prediction accuracy of the random forest regression model.

[0029] Furthermore, the prediction accuracy evaluation method of the random forest regression model is the coefficient of determination R 2 and the root mean square error RMSE, where,

[0030]

[0031]

[0032] Where R 2 is the linear correlation value; RMSE is the deviation value; is the predicted value; is the sample mean; y i is the supervision value; n is the number of samples.

[0033] Furthermore, the evaluation method is:

[0034] When the linear correlation degree value reaches or exceeds the preset linear value, and the deviation degree value does not exceed the preset deviation value, the random forest regression model is determined to be qualified; otherwise, if the linear correlation degree value is lower than the preset linear value, or the deviation degree value is higher than the preset deviation value, the random forest regression model is determined to be unqualified;

[0035] If the random forest regression model fails to meet the requirements, the random forest regression model is reconstructed based on the ecological parameters and carbon density measured values of the reference ecosystem based on a machine learning method until the accuracy of the random forest regression model is judged to be qualified. If the model is qualified, a sensitivity test is performed on the random forest regression model to obtain a test result, which is qualified or unqualified.

[0036] Beneficial effects of the present invention:

[0037] (1) The acquisition of data sources is low-cost and easy to operate through the combination of UAV remote sensing image detection and open-source remote sensing data. (2) The carbon sequestration potential assessment model trained by the reference ecosystem around the quarry pit has high accuracy and can ensure that the residual error is within 10%, which can meet the accuracy requirements for rapid estimation of carbon sequestration potential after quarry pit restoration. (3) By predicting the carbon sequestration potential of the quarry pit after restoration, it can provide data support for the ecological restoration benefit assessment and planning design of the quarry pit. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 :Schematic diagram of the changing trend of carbon sinks in the ecosystem succession hypothesis;

[0039] Figure 2 :Schematic diagram of the steps of carbon sink estimation method;

[0040] Figure 3 : Schematic diagram of carbon sink estimation output;

[0041] Figure 4 : Schematic diagram of residuals of carbon sequestration potential assessment model;

[0042] Figure 5: Schematic diagram comparing the predicted values and measured values of the carbon sequestration potential assessment model. DETAILED DESCRIPTION

[0043] The present invention will be described in detail below by way of embodiments with reference to the accompanying drawings.

[0044] As attached Figure 1 To the attached Figure 5 As shown, the present application provides a method for evaluating the carbon sequestration potential of a quarry pit after ecological restoration, comprising the following steps:

[0045] S11, obtaining remote sensing images of the quarry pit and the reference ecosystem through drone aerial survey, and obtaining ecological parameters of the quarry pit and the reference ecosystem from the remote sensing images, the ecological parameters including but not limited to terrain parameters and soil type parameters.

[0046] S12, obtain the measured values of carbon density of the quarry pit and the reference ecosystem.

[0047] In step S11 and step S12,

[0048] a. The reference ecosystem is preferably the mountain surrounding the quarry pit, and more preferably the area of the surrounding mountain that has not been disturbed by human beings.

[0049] The mountains surrounding the quarry pit refer to the mountains within a range of 1000m or less radiating outward from the quarry pit.

[0050] Selecting the mountains around the quarry pit as the reference ecosystem can make the reference ecosystem and the quarry pit consistent in ecological parameters.

[0051] b. The remote sensing images obtained by UAV aerial survey are preferably high-resolution multispectral remote sensing images.

[0052] c. Use PhotoScan or ArcGIS software to obtain the spectrum, texture, and / or terrain characteristics of high-resolution multispectral remote sensing images and extract terrain parameters and soil type parameters.

[0053] in,

[0054] The terrain parameters are obtained by processing with PhotoScan or ArcGIS software. The processing path is: data import → initialization processing → point cloud encryption → generation of digital surface model and orthophoto map → generation of high-precision digital orthophoto map (DOM), digital surface model (DSM) and digital elevation model (DEM);

[0055] Soil type parameters are based on regional conditions and are obtained through visual interpretation.

[0056] d. The drone model used is DJI-P4 Multispectral.

[0057] e. The measured carbon density value is obtained from the publicly released carbon density data, and data queries can be specifically conducted in the "Carbon Density Dataset of Chinese Terrestrial Ecosystems in the 2010s".

[0058] S2. Based on the machine learning method, establish a random forest regression model for the predicted carbon density value of the quarry pit, and train the random forest regression model. The random forest regression model is used to obtain the predicted carbon density value after the ecological restoration of the quarry pit.

[0059] S21. Use the ecological parameters of the reference ecosystem as variable factors, and use the measured carbon density value of the reference ecosystem as the supervised value of the model. The variable factors and the model supervised value constitute a dataset. The dataset is used to construct a random forest regression model for the predicted carbon density value of the quarry pit.

[0060] As a preferred method, a part of the data in the dataset can also be composed into a training set, and another part of the data can be composed into a test set. Among them, the training set is used to construct a random forest regression model for the predicted carbon density value of the quarry pit.

[0061] The expression of the random forest regression model is:

[0062] CD = F R (AS, SL, SO)

[0063] In the formula,

[0064] CD is the predicted carbon density value of the quarry pit estimated based on the random forest regression model;

[0065] F R is the random forest machine learning algorithm;

[0066] AS is the terrain aspect;

[0067] SL is the terrain slope;

[0068] SO is the soil type.

[0069] S22. Use the bootstrap sampling method to train the random forest regression model, which specifically includes the following steps:

[0070] a. In the original dataset, use the bootstrapping method to randomly draw K new sample sets with replacement, and use the new samples to construct K classification regression trees;

[0071] b. Assume n features, select Mtry features at each node of each classification regression tree, where Mtry features < n, calculate the squared error of each Mtry feature, and select the Mtry feature with the strongest regression ability for node splitting through the magnitude of the squared error value;

[0072] c. Directly generate multiple classification and regression trees to form a random forest for regression on new data, and perform arithmetic averaging on the regression results. The obtained arithmetic mean value is the final value for output. The weights for weighted averaging can be determined based on the out-of-bag (OOB) errors of each classification and regression tree.

[0073] The final regression decision method is as follows:

[0074]

[0075] In the formula,

[0076] H(X) represents the combined regression algorithm;

[0077] h i represents the regression algorithm of a single decision tree;

[0078] W i is the weight of the i-th tree;

[0079] I is the discriminant function. If h i (X) is equal to the target variable Y, the value of I is 1, otherwise it is 0;

[0080] Y is the target variable;

[0081] K is the number of trees in the random forest.

[0082] In step S22,

[0083] To improve the prediction accuracy of the random forest regression model, the values of the model parameters Ntree and Mtry are set to 500 and 2 respectively. Here, Ntree represents the number of decision trees included in the random forest, and Mtry represents the number of nodes included in each decision tree.

[0084] The present invention selects such parameter settings for the following reasons:

[0085] Generally, the default value of Ntree is 500, and the default value of Mtry is logN. When Ntree is set to 500, the OOB error tends to be stable, so the present invention adopts the value of Ntree as 500; while as the value of Mtry increases, the OOB error also increases. Therefore, the present invention adopts the value of Mtry as 2.

[0086] S23. To evaluate the prediction accuracy of the random forest regression model, the test set in step S21 is used to evaluate the prediction accuracy of the random forest regression model.

[0087] The evaluation is carried out through the coefficient of determination R 2 and the root mean square error RMSE. Among them,

[0088]

[0089]

[0090] Wherein,

[0091] R 2 is the linear correlation degree value;

[0092] RMSE is the deviation degree value;

[0093] is the predicted value;

[0094] is the sample mean;

[0095] y i is the supervised value;

[0096] n is the number of samples.

[0097] The evaluation method is as follows:

[0098] If the linear correlation degree value reaches or exceeds the preset linear value and the deviation degree value does not exceed the preset deviation value, the model is determined to be qualified; otherwise, the model is determined to be unqualified.

[0099] If the model is unqualified, based on the machine learning method, on the basis of the ecological parameters of the reference ecosystem and the measured carbon density values, a random forest regression model is reconstructed until the accuracy judgment result of the model shows that the model is qualified.

[0100] If the model is qualified, further sensitivity testing is performed on the random forest regression model and the test results are obtained. The test results are qualified or unqualified.

[0101] S3. Taking the ecological parameters of the quarry as variable factors, using the trained random forest regression model in step S2, output the predicted carbon density value CD after ecological restoration of the quarry.

[0102] S4. Subtract the measured carbon density value CD o of the quarry from the predicted carbon density value CD of the quarry to obtain the carbon sink potential CP of the quarry, that is, C P = CD - CD o .

[0103] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed.

Claims

1. A method for evaluating the carbon sink potential after ecological restoration of a quarry pit, characterized in that, It includes the following steps: S1. Obtain the ecological parameters and the measured carbon density values of the quarry pit and the reference ecosystem; S2. Based on the machine learning method, the ecological parameters of the reference ecosystem, and the measured carbon density values, establish a random forest regression model for predicting the carbon density of the quarry pit, and train the random forest regression model; The expression of the random forest regression model is: CD = F R (AS, SL, SO), where CD is the predicted value of the carbon density of the quarry opening estimated based on the random forest regression model, and F R is the random forest machine learning algorithm, AS is the terrain aspect, SL is the terrain slope, and SO is the soil type; The random forest regression model is constructed and trained using a data set. The data set includes model supervision values and variable factors. The model supervision values are the measured carbon density values of the reference ecosystem, and the variable factors are the ecological parameters of the reference ecosystem. The training steps of the random forest regression model are as follows: a. In the original data set, randomly extract K new sample sets with replacement using the bootstrapping method, and construct K classification and regression trees with the new samples; b. Assume n features. At each node of each classification and regression tree, select Mtry features, where Mtry features < n. Calculate the squared error of each Mtry feature, and select the Mtry feature with the strongest regression ability for node splitting based on the squared error value; c. Directly generate multiple classification and regression trees to form a random forest for regression on new data, and the value obtained by weighted averaging the regression results is the final value for output; The weights for weighted averaging are determined according to the out-of-bag data error of each classification and regression tree; S3. Using the ecological parameters of the quarry pit as variable factors, obtain the predicted carbon density value of the quarry pit through the trained random forest regression model; S4. Calculate the difference between the predicted carbon density value and the measured carbon density value of the quarry pit, and this difference is the carbon sink potential of the quarry pit.

2. The method for evaluating the carbon sink potential after ecological restoration of a quarry bench according to claim 1, wherein The ecological parameters include terrain parameters and soil type parameters.

3. The method for evaluating the carbon sink potential after ecological restoration of a quarry pit according to claim 2, characterized in that, The method for obtaining the ecological parameters is: Obtain the remote sensing images of the quarry pit and the reference ecosystem, and extract the terrain parameters and soil type parameters from the remote sensing images.

4. The method for evaluating the carbon sink potential after ecological restoration of a quarry pit according to claim 3, wherein the remote sensing image is a high-resolution multispectral remote sensing image; the terrain parameters are obtained by processing with PhotoScan or ArcGIS software. The processing path is: data import → initialization processing → point cloud encryption → generation of digital surface model and orthophoto → generation of high-precision digital orthophoto, digital surface model, and digital elevation model; the soil type parameters are obtained by visual interpretation based on regional conditions.

5. The method for evaluating the carbon sink potential after ecological restoration of a quarry opening according to claim 1, characterized in that, The regression decision method for regression of the new data is as follows: Where, H(X) is the identification combined regression algorithm; h i is the single decision tree regression algorithm; W i is the weight of the i-th tree; I is the discriminant function. If h i (X) is equal to the target variable Y, the value of I is 1, otherwise it is 0; Y is the target variable; K is the number of trees in the random forest.

6. The method for evaluating the carbon sink potential after ecological restoration of a quarry opening according to claim 5, characterized in that, In the random forest regression model, set the Ntree value and the Mtry value. The Ntree value is set to 500, and the Mtry value is set to 2.

7. A method for evaluating the carbon sink potential after ecological restoration of a quarry pit according to claim 1, characterized in that, The data set is also used to evaluate the prediction accuracy of the random forest regression model.

8. The method for evaluating the carbon sink potential after ecological restoration of a quarry pit according to claim 7, wherein The prediction accuracy evaluation method of the random forest regression model is the coefficient of determination R 2 and the root mean square error RMSE, where Wherein, R 2 is the linear correlation degree value; RMSE is the deviation degree value; is the predicted value; is the sample mean; y i is the supervised value; n is the number of samples.

9. The method for evaluating the carbon sink potential after ecological restoration of a quarry as claimed in claim 8, wherein The evaluation method is: When the linear correlation degree value reaches or exceeds the preset linear value and the deviation degree value does not exceed the preset deviation value, it is determined that the random forest regression model is qualified; otherwise, if the linear correlation degree value is lower than the preset linear value or the deviation degree value is higher than the preset deviation value, it is determined that the random forest regression model is unqualified; If the random forest regression model is unqualified, based on the machine learning method, the random forest regression model is reconstructed on the basis of the ecological parameters of the reference ecosystem and the measured carbon density values until the judgment result of the accuracy of the random forest regression model is that the model is qualified; if the model is qualified, a sensitivity test is performed on the random forest regression model and a test result is obtained, and the test result is qualified or unqualified.

Images