Landslide susceptibility evaluation method

Abstract

The invention discloses a landslide susceptibility evaluation method. The method comprises the following steps: exporting original values of all evaluation factors; training the random forest model; the trained random forest model is used to replace a basic learning device of the Adaboost model, and an optimized Adaboost model is obtained; the optimized Adaboost model is trained, and a trained landslide susceptibility evaluation model is obtained; and performing grid division on the landslide disaster research area, inputting all evaluation factor original values corresponding to grid points into the trained landslide susceptibility evaluation model, and obtaining the landslide occurrence probability at the corresponding grid points through the trained landslide susceptibility evaluation model so as to complete landslide susceptibility evaluation of the landslide disaster research area. According to the method, the landslide susceptibility of a landslide disaster research area can be accurately and rapidly evaluated, a specific evaluation result is given, and the landslide disaster can be conveniently and rapidly found and coped with.

Description

Technical Field

[0001] The invention relates to the field of landslide prediction, and in particular to a landslide susceptibility evaluation method. Background Art

[0002] In geomorphology, landslide usually refers to the sliding of rocks and soil down a slope. Landslides are dangerous natural hazards that can cause serious damage to natural ecosystems and human production facilities, often resulting in severe casualties and economic losses worldwide. Therefore, reducing landslide losses is an important research issue. Modeling landslide susceptibility is a useful tool to obtain spatially detailed and attribute-accurate landslide susceptibility maps, which can reduce the losses caused by landslides. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a landslide susceptibility evaluation method which can perform accurate and efficient landslide susceptibility evaluation.

[0004] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0005] A landslide susceptibility assessment method is provided, which comprises the following steps:

[0006] S1. Select sample points in the landslide disaster research area and derive the original values ​​of all the evaluation factors corresponding to them; the sample points include sample points that have experienced landslide and sample points that have not experienced landslide;

[0007] S2. Based on the original values ​​of all evaluation factors corresponding to the derived sample points, the random forest model is trained by a random search algorithm with AUC as the training indicator to obtain a trained random forest model;

[0008] S3, using the Adaboost model as an external model, replacing the basic learner of the Adaboost model with the trained random forest model to obtain an optimized Adaboost model;

[0009] S4, based on the original values ​​of all evaluation factors corresponding to the derived sample points, the optimized Adaboost model is trained to obtain a trained landslide susceptibility evaluation model;

[0010] S5. Divide the landslide hazard study area into grids, and input the original values ​​of all evaluation factors corresponding to the grid points into the trained landslide susceptibility evaluation model. The trained landslide susceptibility evaluation model is used to obtain the landslide occurrence probability at the corresponding grid points, thereby completing the landslide susceptibility evaluation of the landslide hazard study area.

[0011] Further, in step S1, the evaluation factors include elevation, slope, aspect, surface cut coefficient, profile curvature, elevation variation coefficient, ground roughness, curvature, terrain moisture index, lithology, distance from faults, distance from roads, distance from rivers, rainfall, normalized difference vegetation index and land type.

[0012] Furthermore, the specific method of selecting sample points in the landslide hazard research area and deriving the original values ​​of all evaluation factors corresponding to the sample points includes the following sub-steps:

[0013] S1-1, set the ratio of sample points with landslide to sample points without landslide to 1:1, and randomly select sample points without landslide;

[0014] S1-2, obtain the original values ​​of all evaluation factors corresponding to each sample point;

[0015] S1-3, unify the spatial resolution and coordinate system of the evaluation factors, and convert them into layers of the same grid size through resampling, that is, obtain data with a unified coordinate system;

[0016] S1-4. Use the multi-value extraction to point tool in ArcGIS 10.5 software to extract the original data table of all evaluation factors according to the landslide point data and complete the export of the original numerical values ​​of the evaluation factors.

[0017] Furthermore, the specific method of unifying the evaluation factor coordinate system is:

[0018] The original values ​​of the evaluation factors were imported into ArcGIS software, and based on the data management tools, projection and transformation tools, and definition projection tools of ArcGIS software, the coordinate system of the evaluation factors was changed to WGS_1984_UTM_Zone_46N.

[0019] Furthermore, the specific method of training the random forest model by using the random search algorithm with AUC as the training indicator is:

[0020] Use RandomizedSearchCV in the scikit-learn library to execute the random search algorithm, take AUC as the training indicator, and perform random search on the four parameters of the random forest model: max_depth, min_samples_splitn_estimators, and min_samples_leaf to complete the training of the random forest model; the ranges of the four parameters max_depth, min_samples_splitn_estimators, and min_samples_leaf are (2, 30), (2, 20), (25, 500), and (1, 20), respectively.

[0021] Furthermore, the max_depth, min_samples_splitn_estimators, and min_samples_leaf parameters of the trained random forest model are 18, 2, 460, and 1, respectively.

[0022] Furthermore, in step S5, the landslide occurrence probabilities at the grid points are sorted in ascending order, and the sorted landslide occurrence probabilities are divided into five levels, namely, relatively low, low, medium, high and extremely high, by a natural breakpoint method.

[0023] An electronic device is provided, comprising:

[0024] Processor; and

[0025] A memory for storing executable instructions of the processor;

[0026] The processor is configured to implement the landslide susceptibility assessment method by executing executable instructions.

[0027] The beneficial effects of the present invention are as follows: the present invention obtains the original data of the landslide disaster research area, uses the multi-value extraction to point tool of ArcGIS 10.5 software, extracts the original data table of all evaluation factors of the sample set according to the landslide point data (positive and negative samples 1:1), and packages and constructs the experimental data set. The data set first uses a random search algorithm to optimize the parameters of the random forest model, and then the optimized random forest model is adaptively enhanced and integrated, and then the sample set is substituted into the optimized Adaboost model (Adaboost-RF) for training, and then the values ​​of all grid points are imported into the trained Adaboost-RF model for landslide susceptibility analysis, and the landslide susceptibility prediction map is output in five levels in the research area, which can intuitively evaluate the landslide susceptibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a schematic diagram of the process of this method;

[0029] Figure 2 It is a landslide disaster susceptibility prediction map of the present invention;

[0030] Figure 3 Schematic diagram of ROC curve. DETAILED DESCRIPTION

[0031] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0032] like Figure 1 As shown, the landslide susceptibility assessment method includes the following steps:

[0033] S1. Select sample points in the landslide disaster research area and derive the original values ​​of all the evaluation factors corresponding to them; the sample points include sample points that have experienced landslide and sample points that have not experienced landslide;

[0034] S2. Based on the original values ​​of all evaluation factors corresponding to the derived sample points, the random forest model is trained by a random search algorithm with AUC as the training indicator to obtain a trained random forest model;

[0035] S3, using the Adaboost model as an external model, replacing the basic learner of the Adaboost model with the trained random forest model to obtain an optimized Adaboost model;

[0036] S4, based on the original values ​​of all evaluation factors corresponding to the derived sample points, the optimized Adaboost model is trained to obtain a trained landslide susceptibility evaluation model;

[0037] S5. Divide the landslide hazard study area into grids, and input the original values ​​of all evaluation factors corresponding to the grid points into the trained landslide susceptibility evaluation model. The trained landslide susceptibility evaluation model is used to obtain the landslide occurrence probability at the corresponding grid points, thereby completing the landslide susceptibility evaluation of the landslide hazard study area.

[0038] In step S1, the evaluation factors include elevation, slope, aspect, surface cut coefficient, profile curvature, elevation variation coefficient, ground roughness, curvature, terrain wetness index, lithology, distance from fault, distance from road, distance from river, rainfall, normalized difference vegetation index and land type. The digital elevation model (DEM) data comes from the geospatial data cloud, and the environmental factors that can be directly obtained from the DEM are elevation, slope, aspect, surface cut coefficient, profile curvature, elevation variation coefficient, ground roughness, curvature and terrain wetness index (TWI). Lithology and distance from fault come from the geological cloud of the China Geological Survey. Road and hydrological data come from the geospatial data cloud. Precipitation and normalized difference vegetation index (NDVI) come from the Qinghai-Tibet Plateau Data Center. Land use type data comes from the publication of Professor Huang Xin of Wuhan University.

[0039] The specific method of selecting sample points in the landslide hazard research area and deriving the original values ​​of all evaluation factors corresponding to them includes the following sub-steps:

[0040] S1-1, set the ratio of sample points with landslide to sample points without landslide to 1:1, and randomly select sample points without landslide;

[0041] S1-2, obtain the original values ​​of all evaluation factors corresponding to each sample point;

[0042] S1-3, unify the spatial resolution and coordinate system of the evaluation factors, and convert them into layers of the same grid size through resampling, that is, obtain data with a unified coordinate system;

[0043] S1-4. Use the multi-value extraction to point tool in ArcGIS 10.5 software to extract the original data table of all evaluation factors according to the landslide point data and complete the export of the original numerical values ​​of the evaluation factors.

[0044] The specific method to unify the coordinate system of the evaluation factors is as follows: import the original values ​​of the evaluation factors into ArcGIS software, and based on the data management tool, projection and transformation tool and define projection tool (Data Management Tool—Projections and Transformations—Define Projection) of ArcGIS software, unify the coordinate system of the evaluation factors to WGS_1984_UTM_Zone_46N.

[0045] In this embodiment, resampling can be performed based on Data Management Tools—Raster—RasterProcessing—Resample. Select the elevation raster as the reference image, where Input Raster is set to the raster image to be resampled, Output Raster Dataset is set to "newraster1", that is, the raster image after resampling, Output Cell Size is set to "Same as layer elevation raster", and Resample is set to "NEAREST"; click OK to obtain the converted raster data of all evaluation factors, that is, to obtain layers of the same raster size.

[0046] The specific method of training the random forest model through the random search algorithm with AUC as the training indicator is as follows: use the RandomizedSearchCV in the scikit-learn library to execute the random search algorithm, use AUC as the training indicator, and perform random search on the four parameters of the random forest model, max_depth, min_samples_splitn_estimators, and min_samples_leaf, to complete the training of the random forest model; the ranges of the four parameters max_depth, min_samples_splitn_estimators, and min_samples_leaf are (2, 30), (2, 20), (25, 500), and (1, 20), respectively.

[0047] The max_depth, min_samples_splitn_estimators, and min_samples_leaf parameters of the trained random forest model are 18, 2, 460, and 1 respectively.

[0048] In the specific training process, you need to import the libraries required for RandomForestClassifier, AdaBoostClassifier, GridSearchCV, roc_auc_score and dataset loading. Then load the landslide susceptibility assessment dataset. Divide the dataset into training set and test set. Define the grid of random forest parameters that you want to randomly search: param_dist = {'n_estimators':np.arange(25,500),'max_depth':np.arange(2,30)'min_samples_split':np.arange(2,20),'min_samples_leaf':np.arange(1,20)}.

[0049] The relevant parameters of the random search algorithm when executing the random search algorithm through RandomizedSearchCV are as follows: random_search = RandomizedSearchCV(estimator = RandomForestClassifier(), param_distributions = param_dist, n_iter = 100, number of random searches cv = 5, number of cross-validation folds scoring = 'roc_auc', using AUC as the scoring standard random_state = 42, n_jobs = -1.

[0050] When training the optimized Adaboost model, the data set can be processed based on Python software, and the evaluation factor values ​​​​except ID and attribute label Y are set as training samples, and the attribute data Y is the training label Y (positive samples are 1, and negative samples are 0).

[0051] In step S5, the landslide occurrence probabilities at the grid points are sorted in ascending order, and the sorted landslide occurrence probabilities are divided into five levels, namely, relatively low, low, medium, high and extremely high, by a natural breakpoint method.

[0052] The electronic device includes:

[0053] Processor; and

[0054] A memory for storing executable instructions of the processor;

[0055] The processor is configured to implement the landslide susceptibility assessment method by executing executable instructions.

[0056] In one embodiment of the present invention, Luolong County of the Tibet Autonomous Region is used as the landslide disaster research area, and the local geological bureau data, the Resource and Environmental Science and Data Center of the Chinese Academy of Sciences, and Landslides_SL are integrated to obtain a total of 109 landslides, which are randomly divided into two groups of training data and verification data using code, with a ratio of 7:3. The training data is used as the input data of this method for corresponding training. After the training is completed, the original values ​​of all evaluation factors corresponding to all grid points are used as the input of this method for verification. The evaluation results obtained are as follows Figure 2 As shown in Figure 1, the extremely low, low, medium, high and extremely high landslide hazard study areas account for 37.68%, 27.73%, 15.75%, 12.64% and 6.2% of the total landslide hazard study area, respectively. Among the landslide geological disasters that have occurred, 92.83% of them occurred in the high-risk areas; on the other hand, Figure 3 As shown, the AUC value obtained by this method (AdaBoost (RandomForest)) is 0.970, indicating that the trained landslide susceptibility assessment model has an excellent prediction effect and is better than directly using the trained random forest model (RandomForest).

[0057] In conclusion, the present invention can accurately and quickly evaluate the landslide susceptibility of a landslide disaster research area and provide specific evaluation results, thereby facilitating the rapid discovery and response to landslide disasters.

Claims

1. A landslide susceptibility assessment method, characterized in that: The following steps are involved: S1. Select sample points in the landslide disaster research area and derive the original values ​​of all the evaluation factors corresponding to them; the sample points include sample points that have experienced landslide and sample points that have not experienced landslide; S2. Based on the original values ​​of all evaluation factors corresponding to the derived sample points, the random forest model is trained by a random search algorithm with AUC as the training indicator to obtain a trained random forest model; S3, using the Adaboost model as an external model, replacing the basic learner of the Adaboost model with the trained random forest model to obtain an optimized Adaboost model; S4, based on the original values ​​of all evaluation factors corresponding to the derived sample points, the optimized Adaboost model is trained to obtain a trained landslide susceptibility evaluation model; S5. Divide the landslide hazard study area into grids, and input the original values ​​of all evaluation factors corresponding to the grid points into the trained landslide susceptibility evaluation model. The trained landslide susceptibility evaluation model is used to obtain the landslide occurrence probability at the corresponding grid points, thereby completing the landslide susceptibility evaluation of the landslide hazard study area.

2. The landslide susceptibility assessment method according to claim 1, characterized in that: In step S1, the evaluation factors include elevation, slope, aspect, surface cut coefficient, profile curvature, elevation variation coefficient, ground roughness, curvature, terrain moisture index, lithology, distance from faults, distance from roads, distance from rivers, rainfall, normalized difference vegetation index and land type.

3. The landslide susceptibility assessment method according to claim 2, characterized in that: The specific method of selecting sample points in the landslide hazard research area and deriving the original values ​​of all the evaluation factors corresponding to them includes the following sub-steps: S1-1, set the ratio of sample points with landslide to sample points without landslide to 1:1, and randomly select sample points without landslide; S1-2, obtain the original values ​​of all evaluation factors corresponding to each sample point; S1-3, unify the spatial resolution and coordinate system of the evaluation factors, and convert them into layers of the same grid size through resampling, that is, obtain data with a unified coordinate system; S1-4. Use the multi-value extraction to point tool in ArcGIS 10.5 software to extract the original data table of all evaluation factors according to the landslide point data and complete the export of the original numerical values ​​of the evaluation factors.

4. The landslide susceptibility assessment method according to claim 3, characterized in that: The specific method of unifying the evaluation factor coordinate system is: The original values ​​of the evaluation factors were imported into ArcGIS software, and based on the data management tools, projection and transformation tools, and definition projection tools of ArcGIS software, the coordinate system of the evaluation factors was changed to WGS_1984_UTM_Zone_46N.

5. The landslide susceptibility assessment method according to claim 1, characterized in that: The specific method of training the random forest model using the random search algorithm with AUC as the training indicator is: Use RandomizedSearchCV in the scikit-learn library to execute the random search algorithm, take AUC as the training indicator, and perform random search on the four parameters of the random forest model: max_depth, min_samples_splitn_estimators, and min_samples_leaf to complete the training of the random forest model; the ranges of the four parameters max_depth, min_samples_splitn_estimators, and min_samples_leaf are (2, 30), (2, 20), (25, 500), and (1, 20), respectively.

6. The landslide susceptibility assessment method according to claim 5, characterized in that: The max_depth, min_samples_splitn_estimators, and min_samples_leaf parameters of the trained random forest model are 18, 2, 460, and 1 respectively.

7. The landslide susceptibility assessment method according to claim 1, characterized in that: In step S5, the landslide occurrence probabilities at the grid points are sorted in ascending order, and the sorted landslide occurrence probabilities are divided into five levels, namely, relatively low, low, medium, high and extremely high, by a natural breakpoint method.

8. An electronic device, characterized in that: include: processor; as well as A memory for storing executable instructions of the processor; The processor is configured to implement the landslide susceptibility assessment method as claimed in any one of claims 1 to 7 by executing executable instructions.

Images