Soil organic carbon inversion method based on remote sensing data

Abstract

The invention provides a soil organic carbon inversion method based on remote sensing data. The method comprises the following steps: acquiring and preprocessing remote sensing data of different areas, wherein the remote sensing data comprises multispectral remote sensing data and digital elevation data; extracting a soil organic carbon remote sensing indicative index set from the remote sensing data; acquiring a spatial index weight by using the soil organic carbon remote sensing indicative index set; generating a soil organic carbon remote sensing inversion index based on the spatial index weight and the soil organic carbon remote sensing indicative index set; constructing an inversion model, and training the inversion model by using the soil organic carbon remote sensing inversion index and the corresponding soil organic carbon data to obtain a soil organic carbon inversion model; and inputting the soil organic carbon remote sensing inversion index of the target area into the soil organic carbon inversion model to obtain a soil organic carbon remote sensing inversion result of the target area.

Description

Technical Field

[0001] The present invention relates to the field of agricultural remote sensing technology, and in particular to a soil organic carbon inversion method based on remote sensing data. Background Art

[0002] Soil organic carbon (SOC) is an important indicator for assessing farmland fertility and soil quality. Its high content can not only promote crop growth and increase crop yields, but also improve soil structure and reduce soil erosion. With global climate change, the importance of environmental protection has become increasingly prominent. Accurately estimating soil organic carbon reserves and their dynamic changes are of great significance for formulating effective land management strategies and responding to climate change. For a long time, countries around the world have used different technical means to estimate or invert the content of SOC. However, due to the combined influence of natural soil-forming elements and human activities, the spatial distribution of farmland SOC often presents spatial non-stationary characteristics, which brings challenges to the precise prediction of SOC space.

[0003] As a means of quickly and widely obtaining information on the growth of surface vegetation, field temperature and humidity, remote sensing technology plays an important role in SOC monitoring, prediction and inversion. Commonly used statistical methods include ordinary linear regression, generalized linear prediction and generalized additive model. These methods are simple and easy to use, but they can only fit the linear relationship between environmental covariates and SOC. When the relationship between the two is nonlinear, this type of method is no longer applicable. In recent years, machine learning methods have been widely used to obtain and fit the relationship between environmental covariates and SOC through learning models such as artificial neural networks, random forests, and decision trees. The spatial distribution of SOC can be inverted based on the trained model. However, due to the differences in the spatial scale of remote sensing data, the complexity of texture feature information, the uncertainty of the prediction model and the influence of environmental factors, there is a certain uncertainty in the SOC remote sensing inversion results. Summary of the invention

[0004] In order to solve the inaccurate soil organic carbon inversion results caused by the difference in spatial scale of remote sensing data, the complexity of texture feature information, the uncertainty of prediction model and the influence of environmental factors in the prior art, the present invention proposes a soil organic carbon inversion method based on remote sensing data, including:

[0005] Step 1: Acquire remote sensing data of different regions and perform preprocessing, wherein the remote sensing data includes multispectral remote sensing data and digital elevation data;

[0006] Step 2: extracting a set of soil organic carbon remote sensing indicative indexes from the remote sensing data;

[0007] Step 3: Obtaining spatial index weights using the soil organic carbon remote sensing indicative index set;

[0008] Step 4: obtaining a soil organic carbon remote sensing inversion index based on the spatial index weight and the soil organic carbon remote sensing indicative index set;

[0009] Step 5: construct an inversion model, and use the soil organic carbon remote sensing inversion index and its corresponding soil organic carbon data to train the inversion model to obtain a soil organic carbon inversion model;

[0010] Step 6: Input the soil organic carbon remote sensing inversion index of the target area into the soil organic carbon inversion model to obtain the soil organic carbon remote sensing inversion result of the target area.

[0011] Further, in step 1, the preprocessing includes:

[0012] Multispectral remote sensing data are corrected by radiation and geometry.

[0013] Furthermore, in step 2, the soil organic carbon remote sensing indicative index set includes: digital elevation data, general normalized difference vegetation index, soil adjusted vegetation index, normalized difference moisture index and surface temperature index.

[0014] Furthermore, the land surface temperature index is calculated by Formula 4:

[0015]

[0016] Among them, TB represents the brightness temperature of the thermal infrared band, EM represents the emissivity, and λ represents the wavelength of the thermal infrared band.

[0017] EM=0.004×FV+0.86(5)

[0018]

[0019]

[0020] Among them, FV represents the surface vegetation coverage, NDVI represents the surface normalized vegetation index, and NDVI min Indicates the minimum value of NDVI, NDVI max Indicates the maximum value of NDVI, R nir , R r Represent the reflectivity of near-infrared light and red light bands respectively.

[0021] Furthermore, step 3 specifically includes:

[0022] Taking the pixel unit of multispectral remote sensing data as the benchmark, the multiples of the pixel size are used for movement, and the spatial indicator geographical weighted variance matrix S of the region is calculated using Formula 8:

[0023] S=D T M(x p ,y p )D(8)

[0024] Among them, D is the remote sensing indicative index set of soil organic carbon, M(x p ,y p ) is the distance weight diagonal matrix centered on pixel P;

[0025] The spatial indicator geographically weighted variance matrix S is decomposed by formula 10:

[0026] S=A(x p ,y p )B(x p ,y p )A(x p ,y p ) T (10)

[0027] Among them, A(x p ,y p ) represents the eigenvector matrix, B(x p ,y p ) represents the diagonal matrix of eigenvalues.

[0028] Calculate the eigenvalues ​​and corresponding eigenvectors of the decomposition S, and use the value of the first principal component as the spatial indicator weight.

[0029] Further, in step 4, the organic carbon remote sensing inversion index SOC-RIM is calculated by the following formula:

[0030]

[0031]

[0032] Among them, ξ represents the soil organic carbon remote sensing inversion index SOC-RIM, ξ0 represents the soil organic carbon remote sensing inversion index SOC-RIM before standardization, S_PCA1 represents the value of the first principal component, ξ 0_max and 0_min represent the maximum and minimum values ​​of ξ0 respectively, f(D',U',S',N',L') represents the standardized soil organic carbon remote sensing indicative index set, U A ,S A ,N A ,L AThey represent the eigenvector directions of UNDI, SAVI, NDMI and LST respectively.

[0033] Furthermore, in step 5, the soil organic carbon data is obtained by collecting soil in the field using a grid method and measuring it indoors using a carbon-nitrogen analyzer.

[0034] Furthermore, in step 5, the inversion model adopts a random forest model.

[0035] Beneficial effects of the present invention:

[0036] The invention provides a soil organic carbon inversion method based on remote sensing data, which comprehensively considers the key factors affecting soil organic carbon, such as topography, surface vegetation, soil properties, and soil temperature, and proposes a remote sensing inversion index SOC-RIM that can accurately invert the organic carbon content in the soil surface, providing strong technical support for improving the accuracy and reliability of soil organic carbon remote sensing inversion. At the same time, the method has the advantages of simple operation and strong practicality, and can be widely used in the field of soil organic carbon remote sensing inversion. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A schematic diagram of a flow chart of a soil organic carbon inversion method based on remote sensing data provided by an embodiment of the present invention;

[0038] Figure 2 A schematic diagram of the indicative weight results of soil organic carbon provided by an embodiment of the present invention;

[0039] Figure 3 A schematic diagram of soil organic carbon inversion results provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution in the embodiment of the present invention will be clearly described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] like Figure 1 As shown, an embodiment of the present invention provides a soil organic carbon inversion method based on remote sensing data, comprising:

[0042] Step 1: Obtain remote sensing data from different regions and preprocess them. The remote sensing data mainly include Landast-9 multispectral remote sensing image data and digital elevation (DEM) data from 2003 to 2023;

[0043] Specifically, Landsat-9 multispectral remote sensing data is downloaded free of charge from the dedicated portal website for sharing data of China Remote Sensing Satellite Ground Station. Multispectral remote sensing image preprocessing mainly undergoes operations such as radiation correction and geometric correction to eliminate errors caused by atmosphere and distortion. DEM data is extracted from the AOLS satellite synthetic aperture radar PALSAR data.

[0044] It can be understood that digital elevation data expresses the elevation information of the ground through the spatial distribution of a series of grid points.

[0045] Step 2: Extract a set of soil organic carbon remote sensing indices from remote sensing data; the set of soil organic carbon remote sensing indices includes digital elevation data (DEM), universal normalized vegetation index (UNDI), soil adjusted vegetation index (SAVI), normalized difference moisture index and land surface temperature index (NDMI).

[0046] Specifically, the preprocessed Landsat multispectral remote sensing data were imported into ENVI software or programmed to extract the universal normalized vegetation index (UNDI), soil adjusted vegetation index (SAVI) and normalized difference moisture index (NDMI) using the following calculation formula;

[0047]

[0048]

[0049]

[0050] Among them, R v , R s , R y , R w Respectively represent the sum of the reflectance of green vegetation, soil, yellow vegetation, and water, R nir , R r , R swir They represent the reflectance of near infrared light, red light, and short-wave infrared bands, respectively; a is the standard soil coefficient; and L is the soil adjustment factor, which is determined according to soil type and vegetation coverage.

[0051] The calculation method of land surface temperature (LST) is shown in formula (4):

[0052]

[0053] Among them, TB represents the brightness temperature of the thermal infrared band, EM represents the emissivity, and λ represents the wavelength of the thermal infrared band.

[0054] EM=0.004×FV+0.86(5)

[0055]

[0056]

[0057] Among them, FV represents the surface vegetation coverage, NDVI represents the surface normalized vegetation index, and NDVI min Represents the minimum surface normalized vegetation index, NDVI max Represents the maximum surface normalized difference vegetation index.

[0058] Step 3: Calculate the spatial index weight using the soil organic carbon remote sensing indicative index set;

[0059] Specifically, the spatial index weights of different regions are calculated as follows: Taking the pixel unit of Landsat-9 multispectral remote sensing data as the benchmark, the pixel size (15m) is used for movement. The coordinates of the central pixel P of a certain area are (x p ,y p ), its spatial indicator geographically weighted variance matrix S is:

[0060] S=D T M(x p ,y p )D(8)

[0061] Where D represents the soil organic carbon remote sensing indicative index set matrix, M(x p ,y p ) represents the distance weight diagonal matrix centered on pixel P;

[0062] The calculation formula for element distance weights of the distance weight diagonal matrix is ​​defined as:

[0063]

[0064] Among them, M pq is the distance weight of pixel Q, d pq is the distance between the Q pixel and the central pixel P in the window, θ is a fixed area width, and in the embodiment of the present invention, θ is set to 1 km.

[0065] The spatial indicator geographically weighted variance matrix S is decomposed, and the decomposition process is shown in formula (10):

[0066] S=A(x p ,y p )B(x p ,y p )A(x p ,y p ) T (10)

[0067] Among them, A(x p,y p ) represents the eigenvector matrix, which represents the main change trend of the data in different directions. p ,y p ) represents the diagonal matrix of eigenvalues, representing the importance or variance contribution of the corresponding eigenvector.

[0068] It can be understood that in the case of geographical weighting, the feature vector is related to the geographical spatial location (x p ,y p ) is related, reflecting the characteristic patterns of data in different geographical locations; the diagonal matrix form of eigenvalues ​​makes each eigenvalue correspond to a specific eigenvector, which facilitates the analysis of the importance of each eigenvector.

[0069] Calculate the eigenvalues ​​and corresponding eigenvectors of the decomposition S, and use the value S_PCA1 of the first principal component obtained by decomposing the spatial indicator geographically weighted variance matrix S as the spatial indicator weight of the central pixel P; the value of the first principal component contains the value of the first principal component of the decomposition of each indicator in the soil organic carbon remote sensing indicative index set.

[0070] like Figure 2 As shown, the soil organic carbon remote sensing indicative index weight results for 2008-2023 are obtained by the method provided in an embodiment of the present invention.

[0071] Step 4: Obtain soil organic carbon remote sensing inversion index based on spatial index weights and soil organic carbon remote sensing indicative index set;

[0072] Specifically, due to the non-uniqueness of the direction of the eigenvector in the above analysis process, when the eigenvectors of UNDI, SAVI, NDMI and LST obtained by decomposition are positive, S_PCA1 is directly used as the remote sensing inversion index of soil organic carbon; when the eigenvectors of UNDI, SAVI, NDMI and LST obtained by decomposition are negative, 1-S_PCA1 is used as the remote sensing inversion index of soil organic carbon.

[0073] Specifically, DEM, UNDI, SAVI, NDMI and LST are standardized to obtain D′, U′, S′, N′ and L′, and the soil organic carbon remote sensing inversion index SOC-RIM is established through formula (11) and formula (12):

[0074]

[0075]

[0076] Formula (12) can be transformed into:

[0077]

[0078] Among them, ξ represents the soil organic carbon remote sensing inversion index SOC-RIM, ξ0 represents the soil organic carbon remote sensing inversion index SOC-RIM before standardization, S_PCA1 represents the value of the first principal component, ξ 0_max and 0_min represent the maximum and minimum values ​​of ξ0 respectively, f(D',U',S',N',L') represents the standardized soil organic carbon remote sensing indicative index set, U A ,S A ,N A ,L A Represent the eigenvector directions of UNDI, SAVI, NDMI and LST, respectively, S_PCA1 D Indicates the spatial index weight of DEM, S_PCA1 U Represents the spatial indicator weight of UNDI, S_PCA1 S Represents the spatial index weight of SAVI, S_PCA1 N Represents the spatial indicator weight of NDMI, S_PCA1 L Represents the spatial indicator weight of LST.

[0079] Step 5: construct an inversion model, and use the soil organic carbon remote sensing inversion index and its corresponding measured soil organic carbon data to train the inversion model to obtain a soil organic carbon inversion model;

[0080] Specifically, an inversion model is constructed, and the soil organic carbon inversion index obtained in step 4 and the measured soil organic carbon data are used as training data. The random forest machine learning method is used to train the inversion model to obtain a soil organic carbon inversion model.

[0081] Furthermore, soil organic carbon data were obtained by collecting soil in the field using the grid method and measuring it indoors using a carbon-nitrogen analyzer.

[0082] Step 6: Input the soil organic carbon remote sensing inversion index of the target area into the soil organic carbon inversion model to obtain the soil organic carbon remote sensing inversion result of the target area.

[0083] Specifically, the soil organic carbon remote sensing inversion index of the target area to be detected is input into the soil organic carbon inversion model to obtain the soil organic carbon inversion result of the target area. Figure 3 As shown, the soil organic carbon inversion result of the target area obtained by the embodiment of the present invention.

[0084] The method provided in the embodiment of the present invention comprehensively considers the key factors affecting soil organic carbon such as long-term surface vegetation coverage, topography, soil properties, soil temperature, etc., and proposes a remote sensing inversion index SOC-RIM that can accurately invert the organic carbon content in the soil surface, providing strong technical support for improving the accuracy and reliability of remote sensing inversion of soil organic carbon.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A soil organic carbon inversion method based on remote sensing data, characterized in that: include: Step 1: Acquire remote sensing data of different regions and perform preprocessing, wherein the remote sensing data includes multispectral remote sensing data and digital elevation data; Step 2: extracting a set of soil organic carbon remote sensing indicative indexes from the remote sensing data; Step 3: Obtaining spatial index weights using the soil organic carbon remote sensing indicative index set; Step 4: obtaining a soil organic carbon remote sensing inversion index based on the spatial index weight and the soil organic carbon remote sensing indicative index set; Step 5: construct an inversion model, and use the soil organic carbon remote sensing inversion index and its corresponding soil organic carbon data to train the inversion model to obtain a soil organic carbon inversion model; Step 6: Input the soil organic carbon remote sensing inversion index of the target area into the soil organic carbon inversion model to obtain the soil organic carbon remote sensing inversion result of the target area.

2. The soil organic carbon inversion method based on remote sensing data according to claim 1, characterized in that: In step 1, the pretreatment includes: Multispectral remote sensing data are corrected by radiation and geometry.

3. The soil organic carbon inversion method based on remote sensing data according to claim 1, characterized in that: In step 2, the soil organic carbon remote sensing indicative index set includes: digital elevation data, general normalized difference vegetation index, soil adjusted vegetation index, normalized difference moisture index and surface temperature index.

4. The soil organic carbon inversion method based on remote sensing data according to claim 3 is characterized in that: The land surface temperature index is calculated by Formula 4: Among them, TB represents the brightness temperature of the thermal infrared band, EM represents the emissivity, and λ represents the wavelength of the thermal infrared band. EM=0.004×FV+0.86(5) Among them, FV represents the surface vegetation coverage, NDVI represents the surface normalized vegetation index, and NDVI min Indicates the minimum value of NDVI, NDVI max Indicates the maximum value of NDVI, R nir , R r Represent the reflectivity of near-infrared light and red light bands respectively.

5. The soil organic carbon inversion method based on remote sensing data according to claim 1, characterized in that: Step 3 specifically includes: Taking the pixel unit of multispectral remote sensing data as the benchmark, the multiples of the pixel size are used for movement, and the spatial indicator geographical weighted variance matrix S of the region is calculated using Formula 8: S=D T M(x p ,y p )D(8) Among them, D is the remote sensing indicative index set of soil organic carbon, M(x p ,y p ) is the distance weight diagonal matrix centered on pixel P; The spatial indicator geographically weighted variance matrix S is decomposed by formula 10: S=A(x p ,and p )B(x p ,and p )A(x p ,and p ) T (10) Among them, A(x p ,y p ) represents the eigenvector matrix, B(x p ,y p ) represents the diagonal matrix of eigenvalues. Calculate the eigenvalues ​​and corresponding eigenvectors of the decomposition S, and use the value of the first principal component as the spatial indicator weight.

6. The soil organic carbon inversion method based on remote sensing data according to claim 1, characterized in that: In step 4, the organic carbon remote sensing inversion index SOC-RIM is calculated by the following formula: Among them, ξ represents the soil organic carbon remote sensing inversion index SOC-RIM, ξ0 represents the soil organic carbon remote sensing inversion index SOC-RIM before standardization, S_PCA1 represents the value of the first principal component, ξ 0_max and 0_min represent the maximum and minimum values ​​of ξ0 respectively, f(D',U',S',N',L') represents the standardized soil organic carbon remote sensing indicative index set, U A ,S A ,N A ,L A They represent the eigenvector directions of UNDI, SAVI, NDMI and LST respectively.

7. The soil organic carbon inversion method based on remote sensing data according to claim 1, characterized in that: In step 5, the soil organic carbon data is obtained by collecting soil in the field using a grid method and measuring it indoors using a carbon-nitrogen analyzer.

8. The soil organic carbon inversion method based on remote sensing data according to claim 1, characterized in that: In step 5, the inversion model adopts a random forest model.

Images