Soil organic matter content prediction method based on mixed soil environment
By using post-trained extended arc length prediction model and machine learning technology in the soil organic matter prediction method, the problem of low prediction accuracy under mixed soil species is solved, and higher prediction accuracy and model adaptability are achieved.
Patent Information
- Application Number
- CN202510121748.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-26
- Publication Date
- 2025-06-06
AI Technical Summary
The existing soil organic matter prediction methods have low prediction accuracy under mixed soil species conditions, making it difficult to stably monitor soil organic matter content.
The soil organic matter content prediction method based on the mixed soil environment is adopted. By obtaining soil reflectivity spectral data, normalized projection and arc length extension are carried out, and the extended arc length prediction model after training is constructed. Combined with machine learning technology, the end element sample selection is optimized to improve prediction accuracy.
The accuracy and reliability of soil organic matter prediction are improved, especially under the conditions of soil species mixing, and the adaptability and stability of the model are enhanced.
Smart Images
Figure CN120108536A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of soil organic matter content estimation, and in particular to a soil organic matter content prediction method based on a mixed soil environment. Background Art
[0002] In recent years, soil degradation has become increasingly serious, especially in black soil areas. The loss of soil organic matter has led to problems such as decreased soil fertility and structural degradation. Soil organic matter content is an important indicator for measuring soil quality and directly determines the level of soil productivity. In order to improve agricultural production efficiency and protect soil resources, soil organic matter monitoring based on remote sensing technology has gradually become a research hotspot. Hyperspectral remote sensing technology, with its advantages of high efficiency and wide coverage, plays an important role in large-scale and long-term monitoring.
[0003] However, the existing soil organic matter prediction methods mainly focus on the prediction accuracy of a single type of soil, lacking in-depth research on the mixed conditions of multiple types of soil, resulting in unsatisfactory application results. Existing related schemes mainly use optical remote sensing technology combined with machine learning, semi-empirical models and spectral index methods to monitor soil organic matter content. These methods use spectral features or band combinations in remote sensing data to predict soil composition, but in field applications, prediction accuracy is still challenging due to factors such as measurement conditions and soil diversity. Although optical remote sensing technology has made significant progress in soil organic matter estimation, the mixing of soil types, especially in large-scale SOM mapping, the mixing of different soil types will significantly affect the accuracy and reliability of predictions, which is still a key issue that needs to be solved urgently.
[0004] Therefore, developing a new inversion method that can stably monitor soil organic matter, especially under mixed soil type conditions, has important practical significance and application value. Summary of the invention
[0005] In order to solve the above problems, the present invention provides a method for predicting soil organic matter content based on a mixed soil environment.
[0006] The present invention aims to provide a method for predicting soil organic matter content based on a mixed soil environment, which specifically comprises the following steps: S1. Obtain measured soil reflectance spectral data and construct a data set; S2. Perform normalized projection on soil reflectance spectral data and construct traditional normalized arc length; S3. Extend the end member arc line of the traditional normalized arc length established in step S2; after the extension, reset the end member samples on the arc line; S4. Divide the data set into training set and test set in proportion, determine the appropriate end member samples through the determination coefficient R², and construct a soil organic matter prediction model; judge the accuracy of the soil organic matter prediction model, and obtain the prediction results of soil organic matter content through accuracy adjustment and optimization.
[0007] Preferably, step S2 includes the following sub-steps: S201. Consider the soil reflectance spectrum data as a point in the high-dimensional Hibbert space and project it onto the high-dimensional unit sphere through unitization; S202. Consider the soil sample as a mixture of two end member samples A and E, and calculate the shortest arc length of the sphere between the soil prediction sample B and the two end member samples; S203. Normalized spectral reflectance of known end member sample A, soil prediction sample B and end member sample E R A , R B and R E , find the lengths of arcs AB, BE, and AE by the cosine theorem of space vectors; S204. Using the spherical cosine theorem, arcs AB' and B'E are obtained through trigonometric transformation; S205. The organic matter content of sample A and sample E is known A SOM and E SOM Get the organic matter content of soil prediction sample B B SOM .
[0008] Preferably, the end member sample A in step S202 is a sample with a soil organic matter content of 0, and the end member sample E is a sample with a saturated soil organic matter content.
[0009] Preferably, in step S205, the organic matter content of soil sample B is predicted B SOM The calculation formula is as follows: ; In the formula, A SOM and E SOM represents the organic matter content of end member sample A and end member sample E, respectively, b 1 Represents the shortest arc AB' from the end member sample A to the projection point B' on the arc AE, b 2 Represents the shortest arc B'E from the end member sample E to the projection point B' on the arc AE.
[0010] Preferably, step S4 includes the following sub-steps: S401. Divide the data set into a training set and a test set in proportion, and select end-member samples in the training set that can accurately predict the soil organic matter content of other samples by calculating the determination coefficient; S402. Establishing a soil organic matter prediction model based on the end member samples selected in step S401 and the reflectance spectrum data of the soil samples; S403. Fit the selected end member sample data to obtain a linear regression equation; use the linear regression equation to predict the soil organic matter content of the test set data to obtain a fitted prediction value; S404. Calculate the coefficient of determination, interquartile range and root mean square error between the fitted prediction value and the actual measurement value, perform optimization screening, and obtain the extended arc length prediction model to predict the soil organic matter content of the sample to be tested.
[0011] Preferably, the fitting in step S403 adopts partial least squares method.
[0012] Preferably, the ratio of the training set to the test set is 1:1.
[0013] Preferably, the soil reflectance spectral data in step S1 is obtained from a soil spectral dataset jointly created by the Consultative Group on International Agricultural Research and the International Centre for Soil Research and Information.
[0014] Compared with the prior art, the present invention can achieve the following beneficial effects: (1) Advantages of geometric modeling: Compared with existing technologies such as machine learning, semi-empirical models, and spectral index methods, the post-training extended arc length method effectively reduces the fluctuations under different soil types and measurement conditions through mathematical normalization, making the model more stable and reliable. The normalized arc length method avoids the endmember selection problem in traditional methods and improves the accuracy of soil organic matter prediction.
[0015] (2) Combining the equal soil organic matter circle with machine learning: The post-training extended arc length method introduced the idea of the equal soil organic matter circle and combined it with machine learning technology to successfully solve the interference of the mixture of two soil types on the prediction results. This method optimizes the prediction effect when mixing different soil types by adjusting the end member selection direction and enhances the adaptability of the model. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 4 is a schematic diagram of the principle of a traditional normalized arc length method provided according to an embodiment of the present invention.
[0017] Figure 2 3 is a schematic diagram of the principle of extending the end member arc provided according to an embodiment of the present invention.
[0018] Figure 3It is a schematic diagram of the influence of different end member selections on the projection point deviation provided in an embodiment of the present invention.
[0019] Figure 4 It is a schematic diagram of the principle of equal soil organic matter rings provided according to an embodiment of the present invention. DETAILED DESCRIPTION
[0020] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the following description, the same modules are represented by the same reference numerals. In the case of the same reference numerals, their names and functions are also the same. Therefore, the detailed description thereof will not be repeated.
[0021] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not constitute a limitation of the present invention.
[0022] The present invention provides a method for predicting soil organic matter content based on a mixed soil environment, which specifically comprises the following steps: S1. Obtain measured soil reflectance spectral data and construct a data set; A mixed dataset of different types of soils was selected from the soil spectral dataset jointly created by the International Consultative Group on Agricultural Research (ICRAF) and the International Soil Research and Information Centre (ISRIC) to obtain soil reflectance spectral data.
[0023] Specifically, soil spectral datasets from six countries were selected, namely Australia, France, Jamaica, Japan, Nigeria and Romania. The basis for selecting these datasets is that the method can theoretically effectively reduce the impact of soil type mixing on prediction accuracy, and by observing the spectral curves, national datasets with obvious soil type mixing were selected for verification.
[0024] S2. Perform normalized projection on soil reflectance spectral data to construct traditional normalized arc length; specifically, the following sub-steps are included: S201. Consider the soil reflectance spectrum data as a point in the high-dimensional Hibbert space and project it onto the high-dimensional unit sphere through unitization; S202. The soil sample is regarded as a mixture of two end-member samples, where the end-member samples include sample A with a soil organic matter content of 0 and sample E with a saturated (i.e., sufficiently large) soil organic matter content; the shortest arc length of the sphere between the soil prediction sample B and the two end-member samples is calculated; S203. Normalized spectral reflectance of known samples A, B and E R A , R Band R E , the lengths c, c' and b of arcs AB, BE and AE are obtained by the cosine theorem of space vectors 1 +b 2 : (1) (2) (3) Among them, b 1 Represents the shortest arc AB' from sample A to the projection point B' on arc AE, b 2 Represents the shortest arc B'E from sample E to the projection point B' on arc AE; S204. By using the spherical cosine theorem, arcs AB' and B'E are obtained through trigonometric transformation; the spherical cosine theorem is as follows: (4) (5) Where, d is the distance from sample B to the projection point B' on arc AE; The expressions for arc AB' and arc B'E are as follows: (6) (7) Among them, b 1 represents the length of arc AB', b 2 represents the length of arc B'E; S205. The organic matter content of sample A and sample E is known A SOM and E SOM Get the organic matter content of soil prediction sample B B SOM , the calculation formula is as follows: (8) Since the normalized spectral reflectance is positive and all arcs are located on the unit sphere, the lengths of all arcs used in this method are less than π / 2.
[0025] Brief description of the principle: Due to the non-collinear changes in reflectance in different bands caused by changes in soil organic matter, the projection point moves on the sphere; using the inverse idea of spectral unmixing, the soil sample is regarded as a mixture of soil organic matter content of 0 and content of sufficiently large; by calculating the shortest arc length of the sphere between the predicted sample and the two end member samples, its soil organic matter can be predicted. However, the sample projection point is not always located on the shortest arc connecting the two end members, so the normalized arc length method uses the cosine theorem for correction. By estimating the actual arc length, the normalized arc length is obtained to more accurately reflect the organic matter of the soil. Figure 1 The principle of estimating the organic matter content of sample B by using sample A with zero organic matter content and soil sample E with sufficiently large content combined with the normalized arc length method is demonstrated.
[0026] The normalized projection of soil hyperspectral data proposed in step S2 of the present invention is carried out to construct a normalized arc length method, the soil spectral data is regarded as a point in the high-dimensional Hibbert space, and it is projected onto the unit sphere by unitization. The non-collinear change of reflectance caused by the change of soil organic matter is used to regard the sample as a mixture with an organic matter content of 0 and a sufficiently large amount. The soil organic matter content is predicted by calculating the shortest arc length of the sphere between the sample point and the end member sample (organic matter content is 0 and a sufficiently large amount). If the sample projection point is not on the end member arc line, a more accurate normalized arc length is calculated by correction through the cosine theorem to reflect the content of soil organic matter.
[0027] S3. Extending the end member arc of the traditional normalized arc length established in step S2; after the extension, resetting the sample B' with a soil organic matter content of 0 and the sample D' with a saturated (ie, sufficiently large) soil organic matter content on the arc as the end members; Figure 2 This extended process is shown, illustrated using samples with different soil organic matter contents (B, D, C). Figure 2 All solid lines in can be obtained by using the law of cosines combined with the normalized spectra of the samples.
[0028] According to the soil organic matter content of sample B and sample D, as well as sample B' with a soil organic matter content of 0 and sample D' with a saturated (i.e., large enough) soil organic matter content, the following expression can be obtained: (9) (10) In the above formula, B'D' , B'B , BD Both represent arcs, B SOM , B’ SOM , D SOM ,D’ SOM Represent the soil organic matter contents of sample B, sample B', sample D, and sample D', respectively; C 1 Point is sample C on the end member arc B'D' The projection point on the arc BC 1 and DC 1 The length of can be solved by formula (6) and formula (7), if BC 1 + DC 1 > BD ,and BC < DC Projection point C 1 Located on the dotted line B'B superior: (11) In other cases: (12) At this time, the predicted soil organic matter content of sample C is: (13) The soil organic matter content can be obtained by formula (13): C SOM and D’ SOM There is no conclusion about the relationship, so D’ SOM You can choose any one.
[0029] Principle: When the normalized arc length method is applied to soil organic matter prediction, it is difficult to select end members; because all arcs in the derivation of the normalized arc length method have no direction restrictions, when any two samples with medium content are used as end members, the prediction point may fall on the extension line of the arc. To solve this problem, this step theoretically extends the end member arc, constructs samples with soil organic matter content of 0 and large enough as end members, so as to ensure the accuracy of the prediction.
[0030] S4. Divide the data set into training set and test set in proportion, determine the appropriate end member samples through the determination coefficient R², and build a soil organic matter prediction model; determine the accuracy of the soil organic matter prediction model, and obtain the prediction results of soil organic matter content through accuracy adjustment and optimization; specifically include the following sub-steps: S401. Divide the data set into a training set and a test set in a 1:1 ratio, and select the end-member samples in the training set that can accurately predict the soil organic matter (SOM) content of other samples by calculating the coefficient of determination (R²); S402. Establishing a soil organic matter prediction model based on the end member samples selected in step S401 and the reflectance spectrum data of the soil samples; S403. Use partial least squares (PLS) method to fit the selected end member sample data to obtain a linear regression equation; use the linear regression equation to predict the soil organic matter content of the test set data to obtain a fitted prediction value; S404. Calculate the coefficient of determination (R²), interquartile range and root mean square error (RPIQ P ), optimize and screen, and obtain the extended arc length prediction model to predict the soil organic matter content of the tested samples; in this step, the R² and RPIQ of the soil organic matter prediction model P The higher the value, the higher the prediction accuracy of soil organic matter content.
[0031] Principle: After the end member arc is extended, two end members can be selected from known samples to predict the soil organic matter content of unknown samples. However, this does not guarantee that the selected samples are the most suitable end members in the data set; this is because the projection points of all samples are not exactly located on the same shortest arc on the unit sphere. After principal component analysis and dimensionality reduction of normalized spectral data of different types of soil, it can be found that differences in soil organic matter content and soil types will cause projection points to shift in different directions. Figure 3 The figure shows the effect of selecting different samples as end members using 8 samples (A, B, B', C, C', D, D', E).
[0032] exist Figure 3 In (a), when sample A and sample E are selected as end members, different types of soil samples cannot be projected to the same position on the end member arc even if they have the same soil organic matter content. This is because the selection of end members directly affects the direction of the end member arc, causing the projection position to deviate from expectations and making it difficult to accurately reflect the characteristic differences between soil types. Figure 3 In (b), by selecting sample A and sample D' as end members, the prediction effect is significantly optimized, so that different types of soil samples can be more accurately projected onto the end member arc at the same organic matter content. Based on this theory, this study proposed the idea of equal soil organic matter rings. The principle of this idea is as follows Figure 4 shown.
[0033] Due to the difference in soil types, the sample points will shift somewhat overall, but these sample points are still located on the unit sphere. There are many types of soil, but if the shape of the spectral curve of the soil is roughly the same, the position of the projection point on the unit sphere will not deviate much. On the contrary, if the shape of the curve changes significantly, it will cause a more serious shift on the unit sphere; however, when there are no more than two rough shapes of the soil spectral curve in the data set, the sample projection points with the same soil organic matter content will always find a ring and gather near this ring, that is, Figure 4 Since the soil organic matter content is the ratio of organic matter to the total mass of soil, these circles are parallel to each other. By selecting different end member combinations to adjust the direction of the end member arcs, we can eventually find an end member arc perpendicular to all parallel circles ( Figure 4 The dotted line in the meridian direction in the figure makes the deviation of the projection of samples with the same soil organic matter content on the end member arc line minimal.
[0034] Since it is theoretically difficult to select the best end-member sample, this study chose to use a statistical method. However, random division of the training set and the test set will lead to large variations in the results. To ensure the uniqueness of the results, this study chose to divide the data set into a training set (50%) to determine a pair of end-member samples that can accurately predict the soil organic matter of other samples in the training set based on the coefficient of determination (R²). The remaining 50% of the data was used as a test set to evaluate the suitability of the end-member samples selected from the training set. The partial least squares method (PLS) was then used to further fit the results with the predicted soil organic matter as a feature and the actual soil organic matter of the sample as the output, and finally the training extended arc length prediction model was obtained.
[0035] Table 1 Prediction accuracy results
[0036] Table 1 shows the prediction accuracy results of the extended arc length prediction model after training. It can be seen from the table that the ERALAT method performs stably and excellently on different data sets, with an RPIQ range of 1.56 to 9.43, showing significant differences in prediction accuracy under different soil types, among which the JP data set has the highest RPIQ value (9.43) and the best performance; the R² values are all maintained at a high level of 62.32% to 88.41%, especially reaching 88.41% on the JP data set, showing excellent prediction accuracy and generalization ability, proving its adaptability and robustness under complex soil conditions. .
[0037] The present invention proposes the idea of equal soil organic matter circles and combines machine learning to predict soil organic matter content; through the above steps, a post-training extended arc length method based on end member arc extension and statistical method optimization is finally obtained.
[0038] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the disclosure of the present invention can be performed in parallel, sequentially or in different orders, as long as the desired results of the technical solution disclosed in the present invention can be achieved, and this document does not limit this.
[0039] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for predicting soil organic matter content based on a mixed soil environment, characterized in that: The specific steps include: S1. Obtain measured soil reflectance spectral data and construct a data set; S2. Perform normalized projection on soil reflectance spectral data and construct traditional normalized arc length; S3. Extend the end member arc line of the traditional normalized arc length established in step S2; after the extension, reset the end member samples on the arc line; S4. Divide the data set into training set and test set in proportion, determine the appropriate end member samples through the determination coefficient R², and construct a soil organic matter prediction model; judge the accuracy of the soil organic matter prediction model, and obtain the prediction results of soil organic matter content through accuracy adjustment and optimization.
2. The method for predicting soil organic matter content based on a mixed soil environment according to claim 1, characterized in that: The step S2 includes the following sub-steps: S201. Consider the soil reflectance spectrum data as a point in the high-dimensional Hibbert space and project it onto the high-dimensional unit sphere through unitization; S202. Consider the soil sample as a mixture of two end member samples A and E, and calculate the shortest arc length of the sphere between the soil prediction sample B and the two end member samples; S203. Normalized spectral reflectance of known end member sample A, soil prediction sample B and end member sample E R A , R B and R E , find the lengths of arcs AB, BE, and AE by the cosine theorem of space vectors; S204. Using the spherical cosine theorem, arcs AB' and B'E are obtained through trigonometric transformation; S205. The organic matter content of sample A and sample E is known A SOM and E SOM Get the organic matter content of soil prediction sample B B SOM .
3. The method for predicting soil organic matter content based on a mixed soil environment according to claim 2, characterized in that: The end member sample A in step S202 is a sample with a soil organic matter content of 0, and the end member sample E is a sample with a saturated soil organic matter content.
4. The method for predicting soil organic matter content based on a mixed soil environment according to claim 3, characterized in that: The organic matter content of the soil sample B predicted in step S205 B SOM The calculation formula is as follows: ； In the formula, A SOM and E SOM They represent the organic matter contents of end member sample A and end member sample E respectively, b1 represents the shortest arc AB' from end member sample A to the projection point B' on arc AE, and b2 represents the shortest arc B'E from end member sample E to the projection point B' on arc AE.
5. The method for predicting soil organic matter content based on a mixed soil environment according to claim 1, characterized in that: The step S4 includes the following sub-steps: S401. Divide the data set into a training set and a test set in proportion, and select end-member samples in the training set that can accurately predict the soil organic matter content of other samples by calculating the determination coefficient; S402. Establishing a soil organic matter prediction model based on the end member samples selected in step S401 and the reflectance spectrum data of the soil samples; S403. Fit the selected end member sample data to obtain a linear regression equation; use the linear regression equation to predict the soil organic matter content of the test set data to obtain a fitted prediction value; S404. Calculate the coefficient of determination, interquartile range and root mean square error between the fitted prediction value and the actual measurement value, perform optimization screening, and obtain the extended arc length prediction model to predict the soil organic matter content of the sample to be tested.
6. The method for predicting soil organic matter content based on a mixed soil environment according to claim 5, characterized in that: The fitting in step S403 adopts the partial least squares method.
7. The method for predicting soil organic matter content based on a mixed soil environment according to claim 6, characterized in that: The ratio of the training set to the test set is 1:
1.
8. The method for predicting soil organic matter content based on a mixed soil environment according to claim 1, characterized in that: The soil reflectance spectral data in step S1 is obtained from a soil spectral dataset jointly created by the Consultative Group on International Agricultural Research and the International Centre for Soil Research and Information.