A method for predicting regional landslide susceptibility based on frequency ratio-random forest model

Abstract

The present invention relates to a method for predicting the susceptibility of regional landslides based on an improved frequency ratio-random forest model and a geographic detector, comprising the following steps: S1: spatially analyzing the inventory of landslides and related influencing factors in the study area; S2: calculating the contribution of influencing factors to the spatial distribution of landslides based on the geographic detector, and selecting influencing factors with significant contribution to the spatial distribution of landslides as subsequent evaluation indicators. S3: establishing an initial landslide susceptibility level based on the improved frequency ratio model, and further determining non-landslide grid unit samples; S4: establishing an improved frequency ratio-random forest model based on the improved frequency ratios of each evaluation indicator, landslide grid units, and randomly selected non-landslide grid units. The present invention makes full use of continuous data and retains more information on evaluation indicators, thereby greatly improving the susceptibility prediction performance of the machine learning model.

Description

Technical Field

[0001] The present invention relates to the field of natural disaster risk management, and in particular to a method for predicting regional landslide susceptibility based on a frequency ratio-random forest model and a geographic detector. Background Art

[0002] The collapse of ridges seriously deteriorates the local ecological environment and also contributes to geological disasters such as landslides, which seriously threaten the safety of human life and property. The study of the susceptibility of collapse of ridges can accurately predict the spatial probability of collapse of ridges in a specific area. Therefore, it is of great significance to carry out the study of spatial prediction of collapse of ridges susceptibility.

[0003] Since there is no strict standard for the selection of evaluation indicators, and the selection of evaluation indicators is affected by the professional knowledge of the evaluator, the evaluation results of the susceptibility of landslides are somewhat subjective, which limits the reliability of the evaluation results to a certain extent. The geographic detector is a new statistical method to reveal the influencing factors behind a certain phenomenon. Since it does not require any assumptions about the data, this method is widely used in remote sensing and earth science. In addition, since the number of landslide grid cells in different intervals of the evaluation indicator attribute values ​​needs to be counted, traditional frequency ratio, information volume, entropy index, evidence weight and other models need to fuzzify the evaluation indicators of continuous attribute values ​​into discrete data through reclassification, thereby reducing the accuracy of the data. The frequency ratio solves the above problems well by introducing the idea of ​​neighborhood. The original attribute value of the evaluation indicator is generally not directly related to whether the landslide occurs, so it is necessary to use models such as frequency ratio to reassign the original attributes for input into machine learning models such as random forests.

[0004] The random forest model (RF) is a classification method, which is composed of multiple weak decision tree models. The random forest model is an easy-to-use and relatively excellent machine learning algorithm with strong prediction accuracy and stability. Therefore, the present invention provides a regional landslide susceptibility prediction method based on frequency ratio-random forest model and geographic detector. Summary of the invention

[0005] In order to address these shortcomings of the existing technology, this paper provides a regional landslide susceptibility prediction method based on frequency ratio-random forest model and geographic detector.

[0006] To achieve the above-mentioned object of the invention, the technical solution of the present invention is as follows:

[0007] A method for predicting regional landslide susceptibility based on a frequency ratio-random forest model and a geographic detector comprises the following steps:

[0008] S1: Taking the grid unit as the basic evaluation unit, the ENVI and ArcGIS platforms are used to manage and analyze the collapsing gully catalog and related influencing factors in the study area. The influencing factors are four categories: topography, basic geology, hydrological environment, and surface cover;

[0009] S2: Calculate the contribution of the influencing factors in the basic evaluation unit to the spatial distribution of collapsing gullies based on the geographic detector, and select the influencing factors with significant contribution to the spatial distribution of collapsing gullies as the evaluation indicators for the subsequent evaluation of the susceptibility of collapsing gullies;

[0010] S3: Based on the frequency ratio model, establish the initial susceptibility level of landslides and further determine the non-landslide grid unit samples;

[0011] S4: Based on the frequency ratio of each evaluation index, the collapsing ridge grid unit, and the randomly selected non-collapsed ridge grid unit, a frequency ratio-random forest model is established.

[0012] The step S1 is to obtain the four major influencing factors of the study area, including topography, hydrological environment, stratigraphic lithology, and surface cover, based on the basic geological data of the study area, ArcGIS, ENVI, and Ovi map platforms and visual interpretation of remote sensing images.

[0013] In the step S2, the attribute values ​​of the evaluation indicators of the basic evaluation unit are partitioned into numerical ranges according to the numerical values, and the q values ​​in different partitions are calculated; wherein the calculation formula of the q value is:

[0014]

[0015] Among them, i is the number of partitions of the evaluation index, N is the number of basic evaluation units in the area to be evaluated, is the variance of the landslide rate R in the ith sub-area, σ 2 is the variance of the gully collapse rate R in the entire area. The q value is used to characterize the contribution of the evaluation index to the spatial distribution of gully collapse. The larger the q value, the greater the contribution of the evaluation index to the spatial distribution of gully collapse.

[0016] The step S3 adopts a frequency ratio model to obtain the frequency ratio of each indicator; the grid calculator is used in ArcGIS software to superimpose these frequency ratios to obtain an initial landslide susceptibility value, and the area with a larger initial landslide susceptibility value indicates a greater possibility of landslide; the natural break point method is used in ArcGIS software to divide the initial landslide susceptibility value into five types of landslide susceptibility levels: 1-extremely low susceptibility area, 2-low susceptibility area, 3-medium susceptibility area, 4-high susceptibility area, and 5-extremely high susceptibility area; non-landslide grid cells that are equal to the landslide grid cells are randomly selected in the 1-extremely low susceptibility area and the 2-low susceptibility area.

[0017] The specific method of the frequency ratio method in step S3 is as follows:

[0018] (1) Normalization: Normalize the evaluation index of the continuity attribute value. The normalization formula is as follows:

[0019] value i =value i / (value max -value min ) (2)

[0020] (2) Set the number of decimal places to be retained for the normalized evaluation index attribute value. For example, if the precision is set to 3, the normalized value i Accurate to 3 decimal places;

[0021] (3) Set the neighborhood width within 0-1, count the number of collapsing ridge grid cells that fall within the neighborhood interval of the normalized evaluation index attribute value and the number of grid cells distributed in the neighborhood, and divide the "ratio of the two" by the "ratio of the total number of collapsing ridge grid cells in the study area to the total number of grid cells in the study area" to obtain the frequency ratio of the attribute value neighborhood. The frequency ratio calculation formula is:

[0022]

[0023] Among them, FR i represents the frequency ratio of the neighborhood of the i-th attribute value in the normalized evaluation index, S Di represents the number of collapsing grid cells in the neighborhood of the i-th attribute value, S i The number of grid cells representing the neighborhood of the i-th attribute value, S D represents the total number of grid cells in the study area, and S represents the total number of grid cells in the study area.

[0024] The natural break point method used in step S3 is not an equidistant classification, but is based on the natural classification inherent in the data, and then the classification interval is identified.

[0025] The frequency ratio of the evaluation index in step S4 is used as the input variable X of the frequency ratio-random forest model. At the same time, the collapsing ridge grid unit and the randomly selected non-collapsed ridge grid unit form a training and testing data set for the input y of the frequency ratio-random forest model, and further randomly divide it into two parts: 70% of the data set is used for training, and the remaining 30% is used for testing; in the training and testing process of the frequency ratio-random forest model, the known collapsing ridge grid unit of the positive sample is represented by 1, and the randomly selected non-collapsed ridge grid unit of the negative sample is represented by 0; the frequency The output variable of the frequency ratio-random forest model is the probability value of each grid cell between 0 and 1, and the distribution of these 0-1 probability values ​​reflects the distribution law of regional landslide susceptibility. The trained and tested frequency ratio-random forest model is used to predict the grid cells of the entire study area to obtain the initial landslide susceptibility value of the entire study area, and then the natural break point classification method in ArcGIS software is used to divide it into 5 levels of landslide susceptibility: 1-extremely low susceptibility area, 2-low susceptibility area, 3-medium susceptibility area, 4-high susceptibility area and 5-extremely high susceptibility area.

[0026] In step S4, the frequency ratio of each evaluation index is used as the input of the random forest model to evaluate the susceptibility of landslide, the ROC curve and the AUC value are calculated, and the AUC value is used to characterize the prediction accuracy of the random forest model.

[0027] The beneficial effects of the present invention are:

[0028] The present invention is based on the method for evaluating the susceptibility of landslides and ridges, which is based on the frequency ratio-random forest and geographic detector. The contribution of the influencing factors in the grid unit to the spatial distribution is calculated by the geographic detector, and the influencing factors with significant contribution to the spatial distribution of landslides and ridges are selected as the evaluation indicators of the frequency ratio-random forest model for evaluating the susceptibility of landslides and ridges. Since the influencing factors with significant contribution to landslides and ridges are considered, the modeling accuracy is improved in the subsequent modeling process; on the other hand, non-landslides and ridges grid data with higher credibility are selected in the extremely low and low-prone areas in the initial landslides and ridges susceptibility zoning results of the frequency ratio, which reduces the errors of the training and test data sets and improves the model accuracy; at the same time, the frequency ratio of the evaluation indicator is used as the input of the random forest model, and continuous data is used, which further improves the reliability of the landslides and ridges susceptibility prediction model. From the above analysis, it can be seen that the geographic detector can well select the influencing factors with significant contribution to the distribution of landslides and ridges, and the frequency ratio-random forest model can well use the continuous data and select non-landslides and ridges grid units with higher credibility, which can significantly improve the prediction accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a flow chart of the method of the present invention;

[0030] Figure 2is a spatial location map of the study area in the example of the present invention;

[0031] Figure 3 It is a graph showing the contribution of 19 influencing factors to the spatial distribution of gully collapse in the example of the present invention;

[0032] Figure 4 The eight influencing factors with the highest contribution to the spatial distribution of gully collapse selected in the example of the present invention are used as evaluation indicators for evaluating the susceptibility of gully collapse, wherein a, b, c, d, e, f, g, and h respectively represent the spatial distribution of the values ​​of each evaluation indicator;

[0033] Figure 5 : The ROC curves and AUC values ​​of the traditional frequency ratio-random forest model (a) and the frequency ratio-random forest model (b) in the examples of the present invention;

[0034] Figure 6 The results of the landslide susceptibility zoning in Yongfeng Township using the traditional frequency ratio-random forest model (a) and the frequency ratio-random forest model (b) in the examples of the present invention. DETAILED DESCRIPTION

[0035] The present invention discloses a method for predicting regional landslide susceptibility based on a frequency ratio-random forest model and a geographic detector, comprising the following steps:

[0036] The object of the present invention is to achieve the goal through a method for predicting regional landslide susceptibility based on a frequency ratio-random forest model and a geographic detector, comprising the following steps:

[0037] S1: Taking the grid unit as the basic evaluation unit, the ENVI and ArcGIS platforms are used to manage and analyze the collapsing gully catalog and related influencing factors in the study area. The influencing factors are four categories: topography, basic geology, hydrological environment, and surface cover;

[0038] The data quality of gully collapse cataloging and the resolution quality of influencing factors have a significant impact on the prediction performance of gully collapse susceptibility in a study area. The gully collapse cataloging is conducive to understanding the location and size of gully collapse and analyzing the main controlling factors.

[0039] In the process of predicting the susceptibility of landslides, representative evaluation indicators such as topography, basic geology, hydrological environment and surface cover should be selected according to the factors that cause landslide disasters in the study area to predict the susceptibility of landslides.

[0040] S2: Calculate the contribution of the influencing factors in the basic evaluation unit to the spatial distribution of landslides based on the geographic detector, and select the influencing factors with significant contribution to the spatial distribution of landslides as evaluation indicators for subsequent landslide susceptibility evaluation.

[0041] The attribute values ​​of the evaluation indicators of the basic evaluation unit are divided into numerical ranges according to the numerical values, and the q values ​​in different partitions are calculated; wherein the calculation formula of the q value is:

[0042]

[0043] Among them, i is the number of partitions of the evaluation index, N is the number of basic evaluation units in the area to be evaluated, is the variance of the landslide rate R in the ith sub-area, σ 2 is the variance of the gully collapse rate R in the entire area. The q value is used to characterize the contribution of the evaluation index to the spatial distribution of gully collapse. The larger the q value, the greater the contribution of the evaluation index to the spatial distribution of gully collapse.

[0044] S3: Based on the frequency ratio model, establish the initial susceptibility level of landslides and further determine the non-landslide grid cells:

[0045] The frequency ratio model was used to obtain the frequency ratio of each indicator. The raster calculator in ArcGIS software was used to superimpose these frequency ratios to obtain the initial susceptibility value of landslides. The larger the initial susceptibility value of landslides, the greater the possibility of landslides. The natural break point method was used in ArcGIS software to divide the initial landslide susceptibility values ​​into five levels of landslide susceptibility: 1-extremely low susceptibility area, 2-low susceptibility area, 3-medium susceptibility area, 4-high susceptibility area, and 5-extremely high susceptibility area. Non-collapse grid cells that were equal to the landslide grid cells were randomly selected in the 1-extremely low susceptibility area and the 2-low susceptibility area.

[0046] The specific method of its frequency ratio is as follows:

[0047] (1) Normalization: normalize the evaluation index of the continuous attribute value (such as elevation). The normalization formula is as follows:

[0048] value i =value i / (value max -value min ) (2)

[0049] (2) Set the number of decimal places to be retained for the normalized evaluation index attribute value. For example, if the precision is set to 3, the normalized value i Accurate to 3 decimal places.

[0050] (3) Set the neighborhood width within 0-1, count the number of collapsing ridge grid cells that fall within the neighborhood interval of the normalized evaluation index attribute value and the number of grid cells distributed in the neighborhood, and divide the "ratio of the two" by the "ratio of the total number of collapsing ridge grid cells in the study area to the total number of grid cells in the study area" to obtain the frequency ratio of the attribute value neighborhood. The frequency ratio calculation formula is:

[0051]

[0052] Among them, FR i represents the frequency ratio of the neighborhood of the i-th attribute value in the normalized evaluation index, S Di represents the number of collapsing grid cells in the neighborhood of the i-th attribute value, S i The number of grid cells representing the neighborhood of the i-th attribute value, S D represents the total number of grid cells in the study area, and S represents the total number of grid cells in the study area.

[0053] S4: Based on the frequency ratio of each evaluation index, the collapsing grid unit, and the randomly selected non-collapsing grid unit, a frequency ratio-random forest model is established:

[0054] The frequency ratio of the evaluation index is used as the input variable X of the frequency ratio-random forest model. At the same time, the collapsing ridge grid unit and the randomly selected non-collapsed ridge grid unit constitute the training and testing data set for the input y of the frequency ratio-random forest model, and further randomly divided into two parts: 70% of the data set is used for training, and the remaining 30% is used for testing; in the training and testing process of the frequency ratio-random forest model, the known collapsing ridge grid unit of the positive sample is represented by 1, and the non-collapsed ridge grid unit randomly selected by the negative sample is represented by 0; the output variable of the frequency ratio-random forest model is the probability value of each grid unit between 0 and 1, and the distribution of these 0-1 probability values ​​reflects the distribution law of regional collapsing ridge susceptibility; the trained and tested frequency ratio-random forest model is used to predict the grid units of the entire study area to obtain the initial collapsing ridge susceptibility value of the entire study area, and then the natural break point classification method in ArcGIS software is used to divide it into 5 categories of collapsing ridge susceptibility levels: 1-extremely low susceptibility area, 2-low susceptibility area, 3-medium susceptibility area, 4-high susceptibility area and 5-extremely high susceptibility area;

[0055] The frequency ratio of each evaluation index is used as the input of the random forest model to evaluate the susceptibility of landslide, and the ROC curve and AUC value are calculated, and the AUC value is used to characterize the prediction accuracy of the random forest model.

[0056] The present invention mainly utilizes the GD function package in R language to run the geographic detector to select the evaluation index; uses ArcGIS and GIS plug-in ALSA to perform frequency ratio calculation and initial landslide susceptibility zoning while selecting non-landslide grid units with higher credibility; and uses the scikit-learn model library in Python to construct a random forest model.

[0057] The susceptibility of gully collapse refers to the spatial probability of gully collapse. The spatial location and distribution probability of gully collapse events in the future can be predicted by the gully collapse events and their environmental conditions that have already occurred. Therefore, the selection of evaluation indicators is very important for a reliable evaluation of gully collapse susceptibility.

[0058] The data quality of gully collapse cataloging and the resolution quality of influencing factors have a significant impact on the prediction performance of gully collapse susceptibility in a study area. The gully collapse cataloging is conducive to understanding the location and size of gully collapse and analyzing the main controlling factors.

[0059] In the process of predicting the susceptibility of landslides, representative evaluation indicators such as topography, basic geology, hydrological environment and surface cover should be selected according to the factors that cause landslide disasters in the study area to predict the susceptibility of landslides.

[0060] In order to make those skilled in the art better understand the scheme of the present invention, the following Figure 1 The flowchart and specific implementation methods shown in the figure further describe the present invention in detail. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The embodiments based on the present invention all belong to the protection scope of the present invention.

[0061] (1) Step 1: Determine the grid unit with a precision of 10 m*10 m as the basic evaluation unit, and select the influencing factors of the landslide susceptibility evaluation for subsequent screening of the influencing factors that contribute more to the spatial distribution of landslides as the evaluation indicators of the landslide susceptibility evaluation model.

[0062] In the specific implementation, take Yongfeng Township, Xingguo County, Jiangxi Province as an example ( Figure 2 ), according to the development characteristics and influencing factors of gullies in the area and the natural geographical characteristics of the study area, a total of 19 influencing factors were selected as the input of the geographic detector, including topography (LS-factor is slope length factor, Slope is slope, Aspect is slope aspect, ProfC is profile curvature, PlanC is plane curvature), basic geology (Clay is clay, Sand is sand, GZK is erodibility), hydrological environment (Prec is the average annual rainfall, GZR is rainfall erosivity, SPI is hydrodynamic index, TWI is terrain wetness index) and surface cover (VH is standardized VH channel backscatter coefficient, FVCx is vegetation coverage, Lai is leaf area index, VV is standardized VV channel backscatter coefficient, CI is coloring index, BI is red edge index, MSAVI is modified soil adjusted vegetation index).

[0063] (2) Step 2: Based on the geographic detector, the contribution q value of 19 influencing factors to the spatial distribution of landslides was calculated, and the first 8 influencing factors (annual average rainfall (Prec), rainfall erosivity (GZR), erodibility (GZK), slope length factor (LS-factor), clay (Clay), slope (Slope), sand (Sand), and standardized VH channel backscatter coefficient (VH)) were selected as evaluation indicators for the subsequent landslide susceptibility evaluation ( Figure 4 ), the q value of each influencing factor is as follows Figure 3 shown.

[0064] In the specific implementation, ArcGIS software was used to reclassify the influencing factors of all 10m precision grid cells into 8 categories according to the attribute values, and the natural break point method was used for classification; then the values ​​were assigned from 1 to 8 from low to high to represent different categories; then a fishing net was created, and the reassigned grid cell influencing factors were converted into point data and spatially connected to the fishing net; then the collapsing ridge grid cells were spatially connected to the fishing net, and the collapsing ridge grid cells were set to 1; finally, the attribute table of the fishing net file was exported to the GD function of the R language for calculation.

[0065] (3) Step 3: Before conducting landslide susceptibility prediction modeling, it is necessary to identify reliable non-landslide grid cells.

[0066] In the specific implementation, the frequency ratio model was used to obtain the frequency ratios of the eight evaluation indicators; the raster calculator in ArcGIS software was used to superimpose these frequency ratios to obtain the initial susceptibility value of landslides. The larger the initial susceptibility value of landslides, the greater the possibility of landslides. The natural break point method was used in ArcGIS software to divide the initial landslide susceptibility values ​​into five levels of landslide susceptibility: 1-extremely low susceptibility area, 2-low susceptibility area, 3-medium susceptibility area, 4-high susceptibility area, and 5-extremely high susceptibility area; non-collapse grid cells that were equal to the landslide grid cells were randomly selected in the 1-extremely low susceptibility area and the 2-low susceptibility area.

[0067] In the specific implementation, the first eight influencing factors with the largest q values ​​(annual average rainfall (Prec), rainfall erosivity (GZR), erodibility (GZK), slope length factor (LS-factor), clay (Clay), slope (Slope), sand (Sand), and standardized VH channel backscattering coefficient (VH)) were selected as the evaluation indicators of the landslide susceptibility evaluation model. i =value i / (value max -value min ) Normalize the original grid cell attribute values ​​of the evaluation index, and set the normalized precision to 3, that is, retain three significant digits; then set the neighborhood width to 0.1, according to the formula The frequency ratios of the neighborhoods of different attribute values ​​of the evaluation index are calculated to obtain the frequency ratios of each index. A frequency ratio greater than 1 indicates that it is conducive to the occurrence of gully collapse. The larger the frequency ratio, the greater the effect on the development of gully collapse.

[0068] (4) Step 4: Based on the frequency ratios of each evaluation index, collapsing ridge grid cells, and randomly selected non-collapsed ridge grid cells, a frequency ratio-random forest model was established.

[0069] In the specific implementation, the frequency ratio of the selected 8 evaluation indicators is first used as the input variable X of the frequency ratio-random forest model, and the 745 collapsed ridge grid units that have occurred in Yongfeng Township, Xingguo County and the 745 randomly selected non-collapsed ridge samples are used as the input variable y of the frequency ratio-random forest model; and then further randomly divided into two parts: 70% of the data set is used for training, and the remaining 30% is used for testing; in the training and testing process of the frequency ratio-random forest model, the collapsed ridge grid unit with known positive samples is represented by 1, and the non-collapsed ridge grid unit randomly selected by the negative sample is represented by 1. The output variable of the frequency ratio-random forest model is the probability value of each grid cell between 0 and 1, and the distribution of these 0-1 probability values ​​reflects the distribution law of regional landslide susceptibility. The trained and tested frequency ratio-random forest model is used to predict the landslide susceptibility value of the entire study area using 1153221 grid cells, and then the natural break point classification method in ArcGIS software is used to divide it into 5 levels of landslide susceptibility: 1-extremely low susceptibility area, 2-low susceptibility area, 3-medium susceptibility area, 4-high susceptibility area and 5-extremely high susceptibility area ( Figure 6 b);

[0070] (5) Step 5: In order to compare the advantages of the frequency ratio-random forest model, this example also uses the traditional frequency ratio-random forest model to carry out the landslide susceptibility assessment in Yongfeng Township.

[0071] In the specific implementation, the 8 evaluation indicators are first reclassified into 5 categories using the natural break point method, and then the formula is used The traditional frequency ratio of them was counted and used as the input variable X of the traditional frequency ratio-random forest model. Meanwhile, the 745 collapsed ridge grid units that had occurred in Yongfeng Township, Xingguo County and the 745 randomly selected non-collapsed ridge samples were used as the input variable y of the traditional frequency ratio-random forest model. The grid units were further randomly divided into two parts: 70% of the data set was used for training and the remaining 30% for testing. In the training and testing process of the frequency ratio-random forest model, the collapsed ridge grid units with known positive samples were represented by 1, and the non-collapsed ridge grid units randomly selected from the negative samples were represented by 0. The output variable of the frequency ratio-random forest model is the probability value of each grid cell between 0 and 1, and the distribution of these 0-1 probability values ​​reflects the distribution law of regional landslide susceptibility. The trained and tested frequency ratio-random forest model is used to predict the landslide susceptibility value of the entire study area using 1153221 grid cells, and then the natural break point classification method in ArcGIS software is used to divide it into 5 levels of landslide susceptibility: 1-extremely low susceptibility area, 2-low susceptibility area, 3-medium susceptibility area, 4-high susceptibility area and 5-extremely high susceptibility area ( Figure 6 a);

[0072] Finally, the receiver operating characteristic curve (ROC) and area under the curve (AUC) were used to evaluate the accuracy of the two models. The AUC values ​​of the frequency ratio-random forest model and the traditional frequency ratio-random forest model were 0.8930 ( Figure 5 b) and 0.8483( Figure 5 a). It shows that the frequency ratio model significantly improves the accuracy of landslide susceptibility assessment of the traditional frequency ratio-random forest model. It further shows that the use of continuous data through frequency ratio can retain more information of evaluation indicators, thereby significantly improving the susceptibility prediction performance of machine learning models.

Claims

1. A method for predicting regional landslide susceptibility based on frequency ratio-random forest model, characterized in that: The following steps are involved: S1: Taking the grid unit as the basic evaluation unit, the ENVI and ArcGIS platforms are used to manage and analyze the collapsing gully catalog and related influencing factors in the study area. The influencing factors are four categories: topography, basic geology, hydrological environment, and surface cover; S2: Calculate the contribution of the influencing factors in the basic evaluation unit to the spatial distribution of collapsing gullies based on the geographic detector, and select the influencing factors with significant contribution to the spatial distribution of collapsing gullies as the evaluation indicators for the subsequent evaluation of the susceptibility of collapsing gullies; S3: Based on the frequency ratio model, establish the initial susceptibility level of landslides and further determine the non-landslide grid unit samples; S4: Based on the frequency ratio of each evaluation index, the collapsing ridge grid unit, and the randomly selected non-collapsing ridge grid unit, a frequency ratio-random forest model is established; In step S2, the attribute values ​​of the evaluation indicators of the basic evaluation unit are divided into numerical ranges according to the numerical values, and the q value; wherein, the q The value is calculated as: （1）； In the formula, i is the number of partitions for the evaluation index, N is the number of basic evaluation units in the area to be evaluated, For the i The landslide rate in each zone R The variance of The landslide rate in the whole area R The variance of q The value is used to characterize the contribution of the evaluation index to the spatial distribution of collapse gullies, and q The larger the value, the greater the contribution of the evaluation index to the spatial distribution of collapsing gullies; The specific method of the frequency ratio method in step S3 is as follows: (1) Normalization: Normalize the evaluation index of the continuity attribute value. The normalization formula is as follows: （2）； (2) Set the number of decimal places to be retained for the normalized evaluation index attribute value. If the precision is set to 3, the normalized value i Accurate to 3 decimal places; (3) Set the neighborhood width within 0-1, count the number of collapsing ridge grid cells that fall within the neighborhood interval of the normalized evaluation index attribute value and the number of grid cells distributed in the neighborhood, and divide the "ratio of the two" by the "ratio of the total number of collapsing ridge grid cells in the study area to the total number of grid cells in the study area" to obtain the frequency ratio of the attribute value neighborhood. The frequency ratio calculation formula is: （3）； in, Represents the normalized evaluation index i The frequency ratio of the attribute value neighborhood, S Di Representative i The number of collapsing grid cells in the neighborhood of the attribute value, S i Representative i The number of grid cells in the neighborhood of the attribute value, S D Represents the total number of Benggang grid cells in the study area, S Represents the total number of grid cells in the study area.

2. The method for predicting regional landslide susceptibility based on frequency ratio-random forest model according to claim 1 is characterized in that: The step S1 is to obtain the four major influencing factors of the study area, including topography, hydrological environment, stratigraphic lithology, and surface cover, based on the basic geological data of the study area, ArcGIS, ENVI, and Ovi map platforms and visual interpretation of remote sensing images.

3. The method for predicting regional landslide susceptibility based on frequency ratio-random forest model according to claim 1 is characterized in that: The step S3 adopts a frequency ratio model to obtain the frequency ratio of each indicator; the grid calculator is used in ArcGIS software to superimpose these frequency ratios to obtain an initial landslide susceptibility value, and the area with a larger initial landslide susceptibility value indicates a greater possibility of landslide; the natural break point method is used in ArcGIS software to divide the initial landslide susceptibility value into five types of landslide susceptibility levels: 1-extremely low susceptibility area, 2-low susceptibility area, 3-medium susceptibility area, 4-high susceptibility area, and 5-extremely high susceptibility area; non-landslide grid cells that are equal to the landslide grid cells are randomly selected in the 1-extremely low susceptibility area and the 2-low susceptibility area.

4. The method for predicting regional landslide susceptibility based on frequency ratio-random forest model according to claim 1 is characterized in that: The natural break point method used in step S3 is not an equidistant classification, but is based on the natural classification inherent in the data, and then the classification interval is identified.

5. The method for predicting regional landslide susceptibility based on frequency ratio-random forest model according to claim 1 is characterized in that: The frequency ratio of the evaluation index in step S4 is used as the input variable X of the frequency ratio-random forest model. At the same time, the collapsing ridge grid unit and the randomly selected non-collapsed ridge grid unit constitute a training and testing data set for the input y of the frequency ratio-random forest model, and further randomly divided into two parts: 70% of the data set is used for training, and the remaining 30% is used for testing; in the training and testing process of the frequency ratio-random forest model, the known collapsing ridge grid unit of the positive sample is represented by 1, and the non-collapsed ridge grid unit randomly selected by the negative sample is represented by 0; the output variable of the frequency ratio-random forest model is the probability value of each grid unit between 0 and 1, and the distribution of these 0-1 probability values ​​reflects the distribution law of regional collapsing ridge susceptibility; the trained and tested frequency ratio-random forest model is used to predict the grid units of the entire study area to obtain the initial collapsing ridge susceptibility value of the entire study area, and then the natural break point classification method in ArcGIS software is used to divide it into 5 types of collapsing ridge susceptibility levels: 1-extremely low susceptibility area, 2-low susceptibility area, 3-medium susceptibility area, 4-high susceptibility area and 5-extremely high susceptibility area.

6. The method for predicting regional landslide susceptibility based on frequency ratio-random forest model according to claim 1 is characterized in that: In step S4, the frequency ratio of each evaluation index is used as the input of the random forest model to evaluate the susceptibility of landslide, the ROC curve and the AUC value are calculated, and the AUC value is used to characterize the prediction accuracy of the random forest model.

Images