A decoupling analysis method for soil erosion based on PLS-SEM model

Abstract

The present invention discloses a soil erosion decoupling analysis method based on a PLS-SEM model, the method comprising the steps of obtaining data for preprocessing, data preprocessing, Group Lasso variable feature screening, and establishing a PLS-SEM model for decoupling. The present invention proposes a method for decoupling the soil erosion mechanism, analyzing the correlation between soil erosion and various driving factors such as precipitation, soil type, and runoff depth, revealing the mutual relationship and potential influencing mechanism between them, and making up for the current domestic insufficient attribution analysis of soil erosion influencing factors; constructing a PLS-SEM model by extracting effective influencing factors, which can handle non-normal data, has strong interpretability, a wide range of applications, and is easy to use and implement; revealing the coordinated regulation of various factors in the soil and water loss mechanism, can provide important support for the rational use of water and soil resources and the alleviation of land degradation; the method has the universality and applicability to be promoted to regional and large-scale ranges.

Description

Technical Field

[0001] The invention relates to qualitative and quantitative analysis of regional soil erosion mechanism, and in particular to a soil erosion decoupling analysis method based on a PLS-SEM model. Background Art

[0002] Soil erosion is a global ecological and environmental problem, mainly caused by unreasonable human activities. Soil erosion is affected by many factors, including climate, topography, soil, vegetation, land use, etc. At present, most studies usually only focus on the direct impact of certain factors on soil erosion, while ignoring the impact of other factors on these factors in various ways, or the complex interactions between them. This research method has led to an incomplete understanding of the mechanism and evolution of soil erosion, and has also limited the proposal of effective soil protection and governance strategies. Therefore, a method that can analyze the interrelationships and indirect effects between factors affecting soil erosion is needed to better understand and predict the process of soil erosion.

[0003] Current research on soil erosion decoupling analysis faces a variety of methodological choices, including physical models, statistical analysis, geographic information systems and remote sensing technology, mathematical models, and machine learning methods; however, each of these methods has some shortcomings: physical models and complex statistical analysis methods have high computational costs; mathematical models have strict data requirements, and different methods are limited by model assumptions; at the same time, machine learning methods have poor interpretability; in addition, data acquisition and processing are also common difficulties.

[0004] In this case, structural equation modeling can make up for these shortcomings. Structural equation modeling (SEM) is a method for dealing with multivariate relationships in complex systems, which can clearly identify direct and indirect effects and conduct a comprehensive analysis of complex models. As a special form of SEM, partial least squares-structural equation modeling (PLS-SEM) has the advantages of wide applicability, suitability for small samples, ability to handle non-normal data, strong interpretability, and ease of use and implementation. Therefore, considering the advantages and disadvantages of various methods and combining the specific research objects and purposes, it is an effective method to select the PLS-SEM model for soil erosion decoupling analysis. The PLS-SEM model can not only effectively solve the complex relationship problems in soil erosion decoupling analysis, but also provide more flexible and effective solutions in data acquisition and processing, providing a deeper and more comprehensive understanding for the research. Summary of the invention

[0005] The technical problem to be solved by the present invention is: to propose a soil erosion decoupling analysis method based on the PLS-SEM model, and to establish a decoupling system for exploring the mechanism of soil erosion with regional universality and applicability. The method is proposed to solve the problems existing in the prior art, such as: different research methods have their own limitations when dealing with soil erosion problems, such as physical models have high requirements for terrain and soil data, statistical methods may not be able to capture complex spatial relationships, and the black box nature of machine learning methods makes the model difficult to interpret. These limitations limit the applicability of the method in different regions.

[0006] The present invention solves the technical problem by adopting the following technical solutions:

[0007] The soil erosion decoupling analysis method based on the PLS-SEM model provided by the present invention comprises the following steps:

[0008] Step S1, obtaining data for preprocessing:

[0009] The digital data include elevation model data, topographic map data, land use data, soil data, rainfall data, soil and water conservation measures data and other required data;

[0010] Step S2, data preprocessing:

[0011] First, multiple categorical variables including "rainfall, slope, land use type, soil and water conservation measures, soil erosion modulus, runoff depth, and soil type" were independently coded so that they could be used to screen the effective influencing factors below;

[0012] Step S3, Group Lasso variable feature screening:

[0013] The feature set with natural grouping structure includes independent variables except continuous variables, as well as multiple categorical variables such as "soil and water conservation measures type" and "land use type". The transformed categorical variables need to be considered as a group for screening, and Group-Lasso selects variables at the group level. Therefore, this method is used to extract effective factors, and its expression is:

[0014]

[0015] Where y is the observed response variable vector; X is the design matrix, containing all the features; β is the model parameter vector; G is the number of groups; P g is the number of features in the gth group; β g is the subset of parameters associated with the gth group; λ is the regularization parameter used to control sparsity;

[0016] Step S4, establish a PLS-SEM model for decoupling:

[0017] According to the final effective influencing factors, combined with the research purpose and theoretical basis, the research framework is determined and the latent variable model is constructed; the partial least squares PLS-SEM method is used to estimate the model parameters, including path coefficients and loading coefficients, and corresponding interaction terms are added as needed to form a basic model.

[0018] Next, the basic model is evaluated, including model fit, significance test of path coefficients and reliability test of latent variables; based on the evaluation, the model is modified and optimized, including deleting irrelevant paths, adding new paths or modifying the measurement model, so as to maximize the goodness of fit under the condition of conforming to the soil erosion mechanism, thereby generating a soil erosion decoupling model with the highest prediction accuracy.

[0019] The present invention may adopt the following method to independently encode the categorical variables in step S2, and adopt corresponding independent encoding methods according to different data types collected.

[0020] The present invention can independently encode categorical variables based on the SPSS method.

[0021] The present invention may include the following steps:

[0022] (1) Open the data file:

[0023] Open the data file containing categorical variables and ensure that the data file contains the categorical variables that need to be independently coded;

[0024] (2) Convert categorical variables: In SPSS, go to the "Convert" menu and select "Recode variables";

[0025] (3) Select categorical variables: In the pop-up dialog box, add the categorical variables that need to be independently coded to the "Variable" list on the right;

[0026] (4) Set the encoding rules: In the "Variable" list on the right, select the categorical variable to be encoded, and then click the "Transform" button; in the dialog box that pops up, select "Customize required values" and then click "Continue";

[0027] (5) Set the code value: In the "Customize Required Value" dialog box, set the code value for each level;

[0028] (6) Apply coding rules: After completing the coding value setting, click the "Add" button to apply the coding rules to the selected categorical variable, and then click "Continue";

[0029] (7) Save the converted data: After completing the coding of all categorical variables, click the "OK" button. SPSS will apply the set coding rules to generate the converted data file and save the converted data file for future use.

[0030] In step (5), when soil types are used, each level thereof is encoded as a binary value, with one level being set to 1 and the other levels being set to 0; the remaining multi-class categorical variables are encoded by selecting custom encoding rules to adapt to specific needs.

[0031] The present invention can convert the data file in step (7) into virtual variables for use in the modeling process.

[0032] In step S3 of the present invention, when λ is larger, the sparsity of the model is stronger, and more feature coefficients will be compressed to zero, thereby achieving the effect of feature selection.

[0033] In step S3, the Akaike information criterion can be used as a screening criterion to measure the quality and complexity of the model fitting data, and the best lambda value is selected by cross-validation AIC to determine the effective influencing variables to be retained.

[0034] The present invention can estimate the soil erosion decoupling model in step S4, analyze the relationship between potential variables, and the relationship between observed variables and potential variables, and write a research report or paper to provide theoretical and technical support for soil erosion and soil and water conservation planning in the sample area.

[0035] Any of the soil erosion decoupling analysis methods described above provided by the present invention is used for qualitative and quantitative analysis of soil erosion mechanisms in regional and large-scale ranges.

[0036] The present invention has the following main advantages:

[0037] 1. A method for decoupling soil erosion mechanisms is proposed to analyze the correlation between soil erosion and driving factors such as precipitation, soil type, and runoff depth, revealing their interrelationships and potential influencing mechanisms, making up for the current insufficient attribution analysis of soil erosion influencing factors in China.

[0038] 2. In view of the previous analytical methods, the PLS-SEM model was constructed by extracting effective influencing factors to quantify the contribution of natural factors (slope, precipitation, soil type) and human factors (water conservation measures, land use type) in driving soil erosion, and comprehensively evaluate their role in soil erosion. Therefore, the PLS-SEM model method can handle non-normal data, has strong interpretability, a wide range of applicability, and is easy to use and implement.

[0039] 3. Study the interaction between soil erosion and its potential driving factors, focus on the quantitative assessment of the joint impact of multiple factors, and reveal the coordinated regulation of various factors in the mechanism of soil and water loss. Studying these goals is expected to provide important support for a deeper understanding of the soil erosion mechanism, the rational use of water and soil resources, and the mitigation of land degradation.

[0040] 4. The proposed decoupling method has universality and applicability to regional and large-scale areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is the PLS-SEM basic model diagram from 2008 to 2013.

[0042] Figure 2 This is the PLS-SEM basic model diagram from 2014 to 2022.

[0043] Figure 3 This is the PLS-SEM decoupling model diagram from 2008 to 2013.

[0044] Figure 4 This is the PLS-SEM decoupling model diagram from 2014 to 2022.

[0045] Figure 5 It is a training flow chart of the PLS-SEM decoupling model of the present invention.

[0046] Figure 6 This is the PLS-SEM result chart from 2008 to 2013.

[0047] Figure 7 This is the PLS-SEM result chart from 2014 to 2022. DETAILED DESCRIPTION

[0048] The present invention is further described below in conjunction with embodiments and drawings, but is not limited to the following contents.

[0049] The present invention provides a soil erosion decoupling analysis method based on a PLS-SEM model, the method comprising the following steps:

[0050] Step S1, obtaining digital elevation model (DEM) data, topographic map data, land use data, soil data, rainfall data, soil and water conservation measures data and other required data.

[0051] The digital elevation model data is: digital elevation model (DEM, Digital Elevation Model) data processed based on ASTGTM data with a resolution of 30m.

[0052] For example, this implementation method selected Chongqing as the study area, and its data were all from the Chongqing Soil and Water Conservation Bulletin, including rainfall, slope, land use type, soil and water conservation measures, soil erosion modulus, runoff depth, soil type, etc.

[0053] Step S2: data preprocessing.

[0054] First, multiple categorical variables such as "rainfall, slope, land use type, soil and water conservation measures, soil erosion modulus, runoff depth, soil type" were independently coded so that they could participate in the screening of effective influencing factors below.

[0055] The following steps were used to independently encode the above categorical variables:

[0056] Different independent coding methods can be adopted according to the different types of data collected. Taking the data collected in this case as an example, the categorical variables are independently coded based on SPSS.

[0057] 1. Open the data file: First, open the data file containing the categorical variables and make sure that the data file contains the categorical variables that need to be independently coded. (1) In the 2008-2013 study, the categorical variables included soil type (purple soil and yellow loam), soil and water conservation measures (no water conservation measures, biological measures, farming measures, and engineering measures), and topography (straight gentle slope, shallow hills, hills, low mountains, and hillsides); (2) In the 2014-2022 study, in addition to soil type and soil and water conservation measures, the categorical variables also included land use types (grassland, cultivated land, forest land, garden land, and other land).

[0058] 2. Convert categorical variables: In SPSS, go to the "Convert" menu and select "Recode variables".

[0059] 3. Select categorical variables: In the pop-up dialog box, select the categorical variables that need to be independently coded and add them to the "Variable" list on the right. Select them by double-clicking the categorical variable or dragging it to the list on the right.

[0060] 4. Set the encoding rules: In the "Variable" list on the right, select the categorical variable to be encoded, and then click the "Transform" button. In the dialog box that pops up, select "Customize required values" and then click "Continue".

[0061] 5. Set the code value: In the "Customize Required Values" dialog box, set the code value for each level. Generally, for example, soil type can be coded as a binary value for each level, with one level set to 1 and the other levels set to 0. For other multi-class categorical variables, you can choose to customize the coding rules to suit specific needs.

[0062] 6. Apply the coding rules: After completing the coding value setting, click the "Add" button to apply the coding rules to the selected categorical variable. Then click "Continue".

[0063] 7. Save the converted data: After completing the coding of all categorical variables, click the "OK" button, SPSS will apply the set coding rules and generate the converted data file. Save the converted data file for subsequent modeling.

[0064] Through the above steps, categorical variables can be independently encoded in SPSS and converted into dummy variables for use in the modeling process.

[0065] Step S3: Group Lasso variable feature screening.

[0066] The Group Lasso (Group Lasso regularization) is a statistical and machine learning technique for feature selection and sparse modeling, which is designed to process feature sets with natural grouping structures. In addition to continuous variables, the independent variables of the present invention also include multiple categorical variables such as "soil and water conservation measures type" and "land use type". The transformed categorical variables need to be considered as a group for screening, and Group-Lasso selects variables at the group level. Therefore, this method is used to extract effective factors. Its expression is as follows:

[0067]

[0068] Where y is the observed response variable vector; X is the design matrix, containing all the features; β is the model parameter vector; G is the number of groups; P g is the number of features in the gth group; β g is the parameter subset associated with the gth group; λ is the regularization parameter used to control sparsity. When λ is larger, the sparsity of the model is stronger, and more feature coefficients will be compressed to zero, thus achieving the effect of feature selection.

[0069] Akaike information criterion (AIC) is used as the screening criterion. AIC can measure the quality and complexity of the model fitting data. The best lambda value is selected by cross-validation AIC to determine the effective influencing variables to be retained.

[0070] Step S4, establishing a PLS-SEM model for decoupling.

[0071] Based on the final effective influencing factors, combined with the research purpose and theoretical basis, the research framework is determined and the latent variable model is constructed. The partial least squares (PLS-SEM) method is used to estimate the model parameters, including the path coefficient and the loading coefficient, and the corresponding interaction terms are added as needed to form a Figure 1 and Figure 2 The base model shown.

[0072] Next, the basic model is evaluated, including the model fit, significance test of path coefficients and reliability test of latent variables; based on the evaluation, the model is modified and optimized, including deleting irrelevant paths, adding new paths or modifying the measurement model, so as to maximize the goodness of fit under the condition of conforming to the soil erosion mechanism, thereby generating the most accurate prediction model. Figure 3 and Figure 4 The decoupled soil erosion model is shown.

[0073] Finally, interpret the model estimation results, analyze the relationship between latent variables and the relationship between observed variables and latent variables, and write a research report or paper, and provide theoretical and technical support for soil erosion and soil and water conservation planning in Chongqing.

[0074] Embodiment 1:

[0075] The present invention decouples soil erosion mechanism based on PLS-SEM model method, see Figure 5 . First, clarify the research purpose and theoretical basis, determine the research framework and build a latent variable model; second, collect relevant data and perform preprocessing, including missing value processing, outlier detection and data standardization; then, use the partial least squares (PLS-SEM) method to estimate model parameters, including path coefficients and loading coefficients; then, evaluate the model, including model fit, significance test of path coefficients and reliability test of latent variables; based on the evaluation, modify and optimize the model, including deleting irrelevant paths, adding new paths or modifying the measurement model; finally, interpret the model estimation results, analyze the relationship between latent variables and the relationship between observed variables and latent variables, and write a research report or paper to summarize the research conclusions and implications (refer to the above content for the specific process and formulas).

[0076] Embodiment 2:

[0077] The present invention selects Chongqing, China as the study area, collects the monitoring data in the "Soil and Water Conservation Bulletin" issued by the Chongqing Municipal Government, takes the soil erosion modulus as the explained variable, and the slope, precipitation, soil type, soil and water conservation measures, etc. as the explanatory variables (see Table 1 and Table 2). Due to the differences in the types of influencing factors of soil erosion in the original data of different years, in order to ensure the consistency of subsequent data processing and comparison, we studied each period separately. First, the non-numerical data are One-Hot encoded, and MATLAB is used to perform Group-Lasso feature selection, combined with Akaike informationcriterion (AIC) as the screening criterion. AIC can measure the excellence and complexity of the model fitting data, and the best lambda value is selected by cross-validation AIC to determine the effective influencing factors. According to the final effective influencing factors, combined with the research purpose and theoretical basis, the research framework is determined and the potential variable model is constructed. The partial least squares (PLS-SEM) method was used to estimate the model parameters, including the path coefficient and the loading coefficient. The corresponding interaction terms were added as needed, and the model was adjusted according to the goodness of fit. The goodness of fit was maximized under the condition that it met the soil erosion mechanism, thus generating a soil erosion decoupling model with the highest prediction accuracy. The research results are shown in Figure 6 , Figure 7 The PLS-SEM model can be used to intuitively understand the soil erosion mechanism in Chongqing and provide theoretical and technical support for soil erosion and soil and water conservation planning in Chongqing.

[0078] Table 1 Influence of PLS-SEM model in 2008-2013

[0079]

[0080] Note: NA indicates no direct or indirect effect; single asterisk (*), p<0.10; double asterisk (**), p<0.05; triple asterisk (***), p<0.01; NS indicates non-significant effect.

[0081] Table 2 Influence of PLS-SEM model in 2008-2013

[0082]

[0083] Note: NA indicates no direct or indirect effect; single asterisk (*), p<0.10; double asterisk (**), p<0.05; triple asterisk (***), p<0.01; NS indicates non-significant effect.

[0084] The present invention is not limited to the embodiments described above. The above description of the specific embodiments is intended to describe and illustrate the technical solution of the present invention. The above specific embodiments are merely illustrative and not restrictive. Without departing from the scope of the present invention and the scope of protection of the claims, a person of ordinary skill in the art can also make many forms of specific changes under the guidance of the present invention, which all fall within the scope of protection of the present invention.

[0085] In addition, the differences in effective influencing factors screened due to different data sources in the PLS-SEM model method should fall within the scope of the present invention.

Claims

1. A soil erosion decoupling analysis method based on PLS-SEM model, characterized in that: The following steps are involved: Step S1, obtaining data for preprocessing: Including elevation model data, topographic map data, land use data, soil data, rainfall data, soil and water conservation measures data and other required data; Step S2, data preprocessing: First, multiple categorical variables including "rainfall, slope, land use type, soil and water conservation measures, soil erosion modulus, runoff depth, and soil type" were independently coded so that they could be used to screen effective influencing factors later; Step S3, Group Lasso variable feature screening: The feature set with natural grouping structure includes independent variables except continuous variables, as well as multiple categorical variables such as "soil and water conservation measures type" and "land use type". The transformed categorical variables need to be considered as a group for screening, and Group-Lasso selects variables at the group level; therefore, this method is used to extract effective factors, and its expression is: Where y is the observed response variable vector; X is the design matrix, containing all the features; β is the model parameter vector; G is the number of groups; P g is the number of features in the gth group; β g is the subset of parameters associated with the gth group; λ is the regularization parameter used to control sparsity; Step S4, establish a PLS-SEM model for decoupling: Based on the final effective influencing factors, combined with the research purpose and theoretical basis, the research framework is determined and the latent variable model is constructed; the partial least squares PLS-SEM method is used to estimate the model parameters, including the path coefficient and the loading coefficient, and the corresponding interaction terms are added as needed to form a basic model; Next, the basic model is evaluated, including model fit, significance test of path coefficients and reliability test of latent variables; based on the evaluation, the model is modified and optimized, including deleting irrelevant paths, adding new paths or modifying the measurement model, so as to maximize the goodness of fit under the condition of conforming to the soil erosion mechanism, thereby generating a soil erosion decoupling model with the highest prediction accuracy.

2. The soil erosion decoupling analysis method according to claim 1, characterized in that: The following method is used to independently encode the categorical variables in step S2, and the corresponding independent coding method is adopted according to the different data types collected.

3. The soil erosion decoupling analysis method according to claim 2, characterized in that: Categorical variables were independently coded using the SPSS method.

4. The soil erosion decoupling analysis method according to claim 3, characterized in that: The following steps are involved: (1) Open the data file: Open the data file containing categorical variables and ensure that the data file contains categorical variables that need to be independently coded; (2) Convert categorical variables: In SPSS, go to the "Convert" menu and select "Recode variables"; (3) Select categorical variables: In the pop-up dialog box, add the categorical variables that need to be independently coded to the "Variable" list on the right; (4) Set the encoding rules: In the "Variable" list on the right, select the categorical variable to be encoded, and then click the "Transform" button; in the dialog box that pops up, select "Customize required values" and then click "Continue"; (5) Set the code value: In the "Customize Required Value" dialog box, set the code value for each level; (6) Apply coding rules: After completing the coding value setting, click the "Add" button to apply the coding rules to the selected categorical variable, and then click "Continue"; (7) Save the converted data: After completing the coding of all categorical variables, click the "OK" button. SPSS will apply the set coding rules to generate the converted data file and save the converted data file for future use.

5. The soil erosion decoupling analysis method according to claim 4, characterized in that: In step (5), when soil type is used, each level is encoded as a binary value, with one level set to 1 and the other levels set to 0; the remaining multi-class categorical variables are customized by selecting custom encoding rules to adapt to specific needs.

6. The soil erosion decoupling analysis method according to claim 4, characterized in that: Convert the data file in step (7) into dummy variables for use in the modeling process.

7. The soil erosion decoupling analysis method according to claim 1, characterized in that: In step S3, when λ is larger, the sparsity of the model is stronger, and more feature coefficients will be compressed to zero, achieving the effect of feature selection.

8. The soil erosion decoupling analysis method according to claim 7, characterized in that: The Akaikeinformation criterion is used as the screening criterion to measure the quality and complexity of the model fitting data. The optimal lambda value is selected through cross-validation AIC to determine the effective influencing variables to be retained.

9. The soil erosion decoupling analysis method according to claim 1, characterized in that: Estimate the soil erosion decoupling model in step S4, analyze the relationship between latent variables, as well as the relationship between observed variables and latent variables, and write a research report or paper to provide theoretical and technical support for soil erosion and soil and water conservation planning in the sample area.

10. The soil erosion decoupling analysis method according to any one of claims 1 to 9, characterized in that: Used for qualitative and quantitative analysis of soil erosion mechanisms at regional and large scales.

Images