Landslide susceptibility evaluation method integrating slope unit heterogeneity and negative sample optimization

Abstract

The invention provides a landslide susceptibility evaluation method integrating slope unit heterogeneity and negative sample optimization. The method comprises the following steps: acquiring an environment factor based on multi-source data; dividing slope units based on terrain and slope structure data; constructing an environmental factor evaluation index based on slope unit heterogeneity; a negative sample selection strategy is optimized in combination with stability constraint and spatial distance constraint conditions; constructing, training and predicting a landslide susceptibility evaluation model; and carrying out susceptibility analysis by utilizing the constructed model. According to the method, five statistical variables of the internal environment factors of the slope unit are extracted, heterogeneity characteristics of the internal environment factors of the slope unit are effectively represented, the slope stability value of the research area is calculated in combination with the TRIGRS model, negative sample selection is optimized in combination with stability value constraint and landslide point spatial distance constraint, and the stability of the slope is improved. And the distinction degree of positive and negative samples and the reliability of model training are improved, so that the precision of landslide susceptibility prediction is improved.

Description

Technical Field

[0001] The present invention relates to the field of geological disaster assessment, and specifically to a landslide susceptibility assessment method that comprehensively considers the heterogeneity of slope units and the optimization of negative samples. Background Technique

[0002] Landslide susceptibility assessment is a core content in geological disaster research, which refers to predicting the likelihood level of landslide disasters in a certain area in the future by comprehensively analyzing factors such as geology, topography, hydrology, and human activities. The core lies in constructing a quantitative model with multi-factor coupling to provide a scientific basis for disaster warning and national land spatial planning.

[0003] The existing landslide susceptibility assessment methods mainly include methods based on statistical models (such as logistic regression, frequency ratio method) and methods based on machine learning models (such as random forest, support vector machine, etc.). These methods generally select environmental factors (such as slope, aspect, lithology, etc.) closely related to landslides based on evaluation units (such as slope units) for modeling. However, the existing technologies have the following main defects:

[0004] (1) Ignoring the heterogeneity of internal factors in the evaluation unit: Traditional methods usually represent environmental factors by the average value or single statistical value of slope units, ignoring the variation characteristics of internal factors in the unit, resulting in the model being unable to comprehensively reflect the distribution law of factors.

[0005] (2) The randomness of negative sample selection is relatively large: Currently, the selection of negative samples mostly uses random methods, lacking scientific constraints on regional stability and the distribution of disaster points, which affects the training effect and prediction accuracy of the model. Summary of the Invention

[0006] Aiming at the deficiencies of the existing technologies, the purpose of the present invention is to provide a landslide susceptibility assessment method that comprehensively considers the heterogeneity of slope units and the optimization of negative samples to solve the problems raised in the above background technique. The evaluation factors of the present invention are more comprehensive and accurate, which is conducive to more scientifically reflecting the environmental background characteristics of landslide occurrence and effectively improving the scientificity and prediction accuracy of the model.

[0007] To achieve the above purpose, the present invention is realized through the following technical solutions: A landslide susceptibility assessment method that comprehensively considers the heterogeneity of slope units and the optimization of negative samples, including the main body of the assessment method, which comprises the following steps:

[0008] S1: Obtain environmental factors based on multi-source data, and obtain multi-source data of landslide influencing factors and historical landslide catalog data in the study area;

[0009] [[ID=!31]]S2: Divide slope units based on terrain and slope structure data. Based on the high-resolution terrain data of the study area and combined with the slope structure characteristics, divide the study area into multiple slope units;

[0010] S3: Construct an environmental factor evaluation index based on the heterogeneity of slope units, select environmental factors closely related to landslides, and extract five statistical variables, namely the mean value, standard deviation, change value, central point value, and mode value, for the heterogeneity characteristics of environmental factors within each slope unit. Conduct correlation analysis and multicollinearity analysis to screen effective factor groups and optimize the factor index;

[0011] S4: Optimize the negative sample selection strategy by combining stability constraints and spatial distance constraints. Use the TRIGRS model to calculate the slope stability value, and based on the calculation results, optimize the negative sample selection by combining stability constraints and spatial distance constraints;

[0012] S5: Construct, train, and predict a landslide susceptibility evaluation model. Select the classic machine learning model of random forest (RF) to construct the landslide susceptibility model, and draw the landslide susceptibility distribution map after training and prediction;

[0013] S6: Use the constructed model to conduct susceptibility analysis and analyze the landslide susceptibility results.

[0014] Furthermore, in step S1, the spatial resolution, projection coordinate system, and geographic coordinate system of the multi-source data of landslide influencing factors are unified.

[0015] Furthermore, the multiple slope units in step S2 are used as the evaluation basic units.

[0016] Furthermore, the specific process of step S3 is as follows:

[0017] S3.1: Select environmental factors closely related to landslides;

[0018] S3.2: Extract statistical variables for the heterogeneity characteristics of environmental factors within each slope unit;

[0019] S3.3: Conduct correlation analysis and multicollinearity analysis on the extracted statistical variables to screen effective factor groups and optimize the factor index.

[0020] Furthermore, the environmental factors include slope, aspect, lithology, soil type, vegetation coverage, etc.; the statistical variables include mean value, standard deviation, change value, central point value, and mode value.

[0021] Furthermore, the specific process of step S4 is as follows:

[0022] S4.1: Use the TRIGRS model to calculate the slope stability value, and based on the calculation results, optimize the negative sample selection by combining stability constraints and spatial distance constraints;

[0023] S4.2: Stability constraint means removing areas with low stability to ensure that negative samples do not contain potential landslide points;

[0024] S4.3: Spatial distance constraint means randomly selecting negative samples in the area outside the 200m buffer zone of the disaster point.

[0025] Furthermore, the range of the low-stability area to be removed is Fs < 2.25; Step S4.3 is used to avoid the positions of positive and negative samples being too close, and improve the rationality of the sample distribution.

[0026] Furthermore, the specific process of Step S5 is as follows:

[0027] S5.1: Select the classic machine learning model of random forest (RF) to construct the landslide susceptibility model;

[0028] S5.2: Input the optimized positive and negative sample data into the model for training. The model generates a landslide susceptibility prediction model by learning the relationship between environmental factors and the occurrence of landslides in the sample data;

[0029] S5.3: Combine the prediction results, divide the study area according to the magnitude of the landslide susceptibility probability value, and draw the landslide susceptibility distribution map.

[0030] Furthermore, the study area in Step S5.3 is divided into extremely high susceptibility area, high susceptibility area, medium susceptibility area, low susceptibility area and extremely low susceptibility area, and the division method adopts the natural break method.

[0031] Advantages of the present invention:

[0032] 1. Based on the method for constructing environmental factor evaluation indexes with slope unit heterogeneity, by extracting five statistical variables (mean value, standard deviation, change value, center point value and mode value) of the internal environmental factors of slope units, the heterogeneity characteristics of the internal environmental factors of slope units are effectively characterized, making the evaluation factors more comprehensive and accurate, and being conducive to more scientifically reflecting the environmental background characteristics of landslide occurrence.

[0033] 2. Based on the TRIGRS model to calculate the slope stability value of the study area, and optimize the selection of negative samples through stability value constraint and landslide point spatial distance constraint, overcome the problem of low quality of negative samples in the traditional random selection method of negative samples, improve the discrimination between positive and negative samples and the reliability of model training, and thus improve the accuracy of landslide susceptibility prediction. Description of the drawings

[0034] Figure 1 It is the specific flow chart of a landslide susceptibility evaluation method integrating slope unit heterogeneity and negative sample optimization of the present invention;

[0035] Figure 2This is the flow chart for optimizing the negative sample selection strategy of the present invention. Detailed implementation manners

[0036] To make the technical means, creative features, achieved purposes and effects of the present invention easy to understand, the present invention will be further described below in conjunction with the detailed implementation manners.

[0037] Please refer to Figures 1 to 2 , the present invention provides the following technical solutions: A landslide susceptibility evaluation method that comprehensively considers the heterogeneity of slope units and optimizes negative samples. This method effectively characterizes the heterogeneity characteristics of the internal environmental factors of slope units by extracting five statistical variables (mean value, standard deviation, change value, center point value, and mode value) of the internal environmental factors of slope units, making the evaluation factors more comprehensive and accurate, and facilitating a more scientific reflection of the environmental background characteristics of landslide occurrence. In this embodiment, the specific steps of this method are as follows:

[0038] Step 1: Obtain environmental factors based on multi-source data.

[0039] Obtain multi-source data of landslide influencing factors and historical landslide catalog data in the study area, and unify the spatial resolution, projection coordinate system, and geographic coordinate system of the multi-source data of landslide influencing factors.

[0040] Step 2: Divide slope units based on terrain and slope structure data.

[0041] Based on the high-resolution terrain data in the study area and combined with the slope structure characteristics, divide the study area into multiple slope units as the basic evaluation units.

[0042] Step 3: Construct environmental factor evaluation indicators based on the heterogeneity of slope units.

[0043] Select environmental factors closely related to landslides, such as slope, aspect, lithology, soil type, vegetation coverage, etc. For the heterogeneity characteristics of the internal environmental factors of each slope unit, extract five statistical variables: mean value, standard deviation, change value, center point value, and mode value;

[0044] Table 1 Statistical variable table

[0045]

[0046] Conduct correlation analysis and multicollinearity analysis on the extracted statistical variables to screen effective factor groups and optimize the factor indicators. Use the Pearson correlation coefficient method to analyze the correlation between factors. The calculation formula of the Pearson coefficient P is as follows:

[0047]

[0048] Where n is the number of samples; Xi and Yi are the observed values of variables X and Y corresponding to point i; are the average values of variables X and Y.

[0049] The VIF (Variance Inflation Factor) test method is selected for multicollinearity analysis. The VIF calculation formula is as follows:

[0050]

[0051] Where Ri is the correlation coefficient, which is obtained by linear regression with the i-th variable selected as the dependent variable and all other variables as independent variables. TOL is the tolerance of the sample, which is the reciprocal of VIF.

[0052] Step 4: Optimize the negative sample selection strategy by combining stability constraints and spatial distance constraints.

[0053] Use the TRIGRS model to calculate the slope stability value. Based on the calculation results, optimize the negative sample selection by combining stability constraints and spatial distance constraints. The TRIGRS model is a raster-based rainfall-induced slope stability calculation model developed by the United States Geological Survey, which consists of a slope stability calculation module, an infiltration module, and a hydrological module. It is assumed that the deep layer is the weathered bedrock layer with a very small permeability coefficient. Therefore, the calculation of the pressure head uses the lower boundary finite depth condition, and the expression is as follows:

[0054]

[0055] Where: ψ is the groundwater hydraulic head; t is the total time for calculating ψ; Z is the soil layer thickness in the vertical direction, and z is the soil layer thickness perpendicular to the slope surface; d is the buried depth of the vertical groundwater level measured under steady state; β is a specific parameter, and the calculation formula is β = cos2α - (IZLT / Ks), where Ks is the saturated vertical permeability coefficient; Inz is the surface infiltration amount corresponding to the rainfall intensity in the n-th time period, and IZLT is the stable (initial) surface infiltration amount, which can generally be obtained according to the average rainfall in the recent few weeks or months; D1 = D0 / cos2α, where D0 is the saturated hydraulic diffusivity (D0 = Ks / Ss, and Ss is the specific storage coefficient); N is the total number of rainfall duration intervals; H(t - tn) is the Heaviside step function, and tn is the rainfall duration in the n-th stage during the rainfall period. ierfc(n) is the first integral value of the Gaussian complementary error function for variable n.

[0056] The stability calculation model introduces an infinite slope model and combines the water pressure head to calculate the safety factor Fs of the slope at different time periods:

[0057]

[0058] Where: Fs is the safety factor; φ is the internal friction angle; α is the slope; c is the cohesion; ψ is the water pressure head; γw is the unit weight of groundwater; γs is the unit weight of soil; Z is the vertical thickness of the soil layer.

[0059] Fs = 2.25 is generally considered as the dividing line for judging whether a landslide is unstable. When Fs > 2.25, the slope is considered to be in a stable state, and when Fs < 2.25, it is in an unstable state.

[0060] Input the soil parameters and rainfall data, simulate and calculate the stability values of each slope unit, input the rainfall data, calculate the slope stability values of the obtained area, and extract the stable state slope areas where Fs > 2.25. The spatial distance constraint means randomly selecting negative samples in the area outside the 200m buffer zone of the disaster point to avoid the positions of positive and negative samples being too close, improving the rationality of the sample distribution, and selecting the positive and negative sample ratio as 1:1.

[0061] Step 5: Construct, train, and predict the landslide susceptibility evaluation model.

[0062] Select the classic machine learning model of random forest (RF) to construct the landslide susceptibility model. Based on slope units and considering the heterogeneity of environmental factors within slope units, combined with stability constraints and spatial distance constraints, optimize the selection of negative samples to construct the model. Input the optimized positive and negative sample data into the model for training. The model learns the relationship between environmental factors and the occurrence of landslides in the sample data to generate a landslide susceptibility prediction model; combined with the prediction results, use the natural breakpoint method to divide the study area into extremely high susceptibility areas, high susceptibility areas, medium susceptibility areas, low susceptibility areas, and extremely low susceptibility areas according to the size of the landslide susceptibility probability value, and draw the landslide susceptibility distribution map.

[0063] Step 6: Use the constructed model for susceptibility analysis and analyze the landslide susceptibility results.

[0064] The above shows and describes the basic principles, main features, and advantages of the present invention. For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic features of the present invention, the present invention can be implemented in other specific forms.

[0065] In addition, it should be understood that although this specification is described according to embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A landslide susceptibility evaluation method that comprehensively considers the heterogeneity of slope units and negative sample optimization, including the ontology of the evaluation method, characterized in that, It includes the following steps: S1: Obtain environmental factors based on multi-source data, and acquire multi-source data of landslide influencing factors and historical landslide catalog data in the study area; S2: Divide slope units based on terrain and slope structure data. Based on the high-resolution terrain data of the study area and combined with the slope structure characteristics, divide the study area into multiple slope units; S3: Construct an environmental factor evaluation index based on the heterogeneity of slope units. Select environmental factors closely related to landslides. For the heterogeneity characteristics of environmental factors within each slope unit, extract five statistical variables: mean value, standard deviation, change value, central point value, and mode value, and conduct correlation analysis and multicollinearity analysis to screen effective factor groups and optimize factor indicators; S4: Optimize the negative sample selection strategy by combining stability constraints and spatial distance constraints. Use the TRIGRS model to calculate the slope stability value, and based on the calculation results, optimize the negative sample selection by combining stability constraints and spatial distance constraints; S5: Construct, train, and predict the landslide susceptibility evaluation model. Select the classic machine learning model of random forest (RF) to construct the landslide susceptibility model, and draw the landslide susceptibility distribution map after training and prediction; S6: Use the constructed model to conduct susceptibility analysis and analyze the landslide susceptibility results.

2. A landslide susceptibility assessment method that integrates slope unit heterogeneity and negative sample optimization according to claim 1, characterized in that: In step S1, the spatial resolution, projection coordinate system, and geographic coordinate system of the multi-source data of landslide influencing factors are unified.

3. A landslide susceptibility evaluation method that comprehensively optimizes the heterogeneity of slope units and negative samples according to claim 1, characterized in that: The multiple slope units in step S2 are used as the evaluation basic units.

4. A landslide susceptibility evaluation method that comprehensively considers the heterogeneity of slope units and negative sample optimization, as described in claim 1, wherein The specific process of step S3 is as follows: S3.1: Select environmental factors closely related to landslides; S3.2: Extract statistical variables for the heterogeneity characteristics of environmental factors within each slope unit; S3.3: Conduct correlation analysis and multicollinearity analysis on the extracted statistical variables to screen effective factor groups and optimize factor indicators.

5. A landslide susceptibility evaluation method that synthesizes slope unit heterogeneity and negative sample optimization according to claim 4, characterized in that: The environmental factors include slope, aspect, lithology, soil type, vegetation coverage, etc.; the statistical variables include mean value, standard deviation, change value, central point value, and mode value.

6. A landslide susceptibility evaluation method that synthesizes slope unit heterogeneity and negative sample optimization according to claim 1, characterized in that The specific process of step S4 is as follows: S4.1: Use the TRIGRS model to calculate the slope stability value, and based on the calculation results, optimize the negative sample selection by combining stability constraints and spatial distance constraints; S4.2: The stability constraint is to eliminate areas with low stability to ensure that negative samples do not contain potential landslide points; S4.3: The spatial distance constraint is to randomly select negative samples in the area outside the 200m buffer zone of the disaster point.

7. A landslide susceptibility assessment method that synthesizes slope unit heterogeneity and negative sample optimization according to claim 6, characterized in that: The range of the eliminated low-stability area is Fs < 2.25; step S4.3 is used to avoid the positions of positive and negative samples being too close and improve the rationality of sample distribution.

8. A landslide susceptibility evaluation method that integrates slope unit heterogeneity and negative sample optimization according to claim 1, characterized in that, The specific process of step S5 is as follows: S5.1: Select the classic machine learning model of random forest (RF) to construct the landslide susceptibility model; S5.2: Input the optimized positive and negative sample data into the model for training. The model generates a landslide susceptibility prediction model by learning the relationship between environmental factors and the occurrence of landslides in the sample data; S5.3: Combine the prediction results, divide the study area according to the magnitude of the landslide susceptibility probability value, and draw the landslide susceptibility distribution map.

9. A landslide susceptibility evaluation method that synthesizes slope unit heterogeneity and negative sample optimization according to claim 8, characterized in that: The research area in step S5.3 is divided into extremely high susceptibility areas, high susceptibility areas, medium susceptibility areas, low susceptibility areas, and extremely low susceptibility areas, and the division method uses the natural break point method.

Images