Landslide susceptibility prediction method based on maximum slope ratio and gradient generalization

Abstract

This invention provides a landslide susceptibility prediction method based on maximum slope gradient generalization, comprising the following steps: S1. Randomly selecting an equal number of non-landslide samples around a landslide; S2. Randomly dividing the landslide and non-landslide samples in a ratio of 7:3, with 70% of the samples used as a training set and 30% as a test set; S3. Sampling the row samples in the training set with replacement, performing column sampling on the characteristic factors, and randomly selecting samples from the characteristic factors to obtain a sample set containing row and column samples; S4. Iterating through the column samples, selecting the most appropriate feature based on the Gini coefficient for splitting, and constructing a decision tree; S5. Repeating S3 and S4 to generate multiple decision trees to form a random forest; S6. Inputting the test set into the random forest for prediction. This invention effectively improves prediction accuracy.

Description

Technical Field

[0001] The invention relates to the technical field of landslide susceptibility prediction, in particular to a landslide susceptibility prediction method based on maximum slope rate and gradient generalization. Background Art

[0002] When evaluating landslide susceptibility, the choice of evaluation indicators significantly impacts the accuracy of model predictions. These indicators are often selected based on factors such as the terrain, lithology, geological structure, and meteorological conditions where landslides occur. Among the terrain evaluation indicators, slope gradient is the most important consideration in landslide susceptibility assessment, as the appropriate slope gradient range plays a crucial role in landslide development.

[0003] The current evaluation units for landslide susceptibility assessment mainly include grid units and slope units. Slope units are widely used in medium and large-scale evaluations because they can better reflect the characteristics of landslides as a slope hazard. In the evaluation of slope units, some non-single characteristic values ​​need to be generalized. Currently, the evaluation index selected considering the influence of slope gradient is mainly the average slope. Usually, the average slope within the slope unit is calculated as the slope generalization value and added to the slope unit. When selecting landslide samples, the terrain of the evaluation unit is the post-sliding terrain. This usually results in the slope at the rear edge of the landslide becoming steeper and the slope in the front accumulation area being smaller, which is significantly different from the terrain slope before the landslide sliding. It is inappropriate to use the average slope after sliding to characterize the terrain slope conditions before the slope sliding in the current landslide samples, as it cannot accurately represent the slope characteristics before the landslide sliding. Summary of the Invention

[0004] The present invention provides a landslide susceptibility prediction method based on maximum slope rate and gradient generalization to solve one or more of the above problems.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] The landslide susceptibility prediction method based on the maximum slope gradient generalization includes:

[0007] S1. Randomly select an equal number of non-landslide samples around the landslide;

[0008] S2. Randomly divide the landslide and non-landslide samples into a ratio of 7:3, with 70% as the training set and 30% as the test set;

[0009] S3. Perform row sampling with replacement on the row samples in the training set, perform column sampling on the characteristic factors, and randomly select samples from the characteristic factors to obtain a sample set containing row samples and column samples;

[0010] S4. Traverse the column samples, select the most appropriate features for splitting based on the Gini coefficient, and complete the construction of the decision tree;

[0011] S5. Repeat S3 and S4 to generate multiple decision trees to form a random forest;

[0012] S6. Input the test set into the random forest for prediction.

[0013] In this specification, the slope unit is selected as the evaluation unit, and the evaluation factors of the evaluation unit are obtained. The evaluation factors include average elevation, average slope aspect, stratum lithology, distance from the fault, average historical earthquake point density, distance from the water system and average annual rainfall; the average slope and maximum slope of the evaluation unit are combined with the evaluation factors to form a data set, and the landslide susceptibility evaluation model is trained and tested respectively.

[0014] In this specification, 133 landslides were selected, and an equal number of non-landslide samples were randomly selected around the landslides. The slopes where landslides occurred were marked as 1, and the slopes where landslides did not occur were marked as 0, resulting in 266 input samples.

[0015] In this specification, 186 row samples in the training set are sampled 186 times with replacement. Sampling with replacement will result in repeated samples in the collected row sample data, so that during training, the input samples of each decision tree are not all samples, making it relatively less likely to overfit. After the row sampling is completed, the characteristic factors are column sampled, and m samples are randomly selected from the 8 characteristic factors, m≤8, to form an input sample set containing 186 row samples and m column samples. The 8 characteristic factors are average slope, average elevation, average slope aspect, stratum lithology, distance from fault, average historical earthquake point density, distance from water system and average annual rainfall, or the 8 characteristic factors are maximum slope, average elevation, average slope aspect, stratum lithology, distance from fault, average historical earthquake point density, distance from water system and average annual rainfall.

[0016] In this specification, m feature vectors are traversed and the most appropriate feature is selected for splitting. The feature selection method is the Gini coefficient. Feature selection is performed based on the Gini coefficient. The value range of the Gini coefficient is 0 to 1, where 0 indicates the highest purity of the data set and 1 indicates the lowest purity of the data set. For each possible partition threshold, samples with a value of the feature in the data set less than or equal to the partition threshold are divided into one side l, and samples with a value greater than the partition threshold are divided into the other side r. The Gini coefficients are:

[0017] ;

[0018] ;

[0019] in, is the Gini coefficient corresponding to node l, is the Gini coefficient corresponding to node r, is the probability of the first type of data point in node l, is the probability of the second type of data point in node l, The probability of a data point of type 1 in node r, is the probability of the second type of data point in node r;

[0020] For each partition threshold, the Gini coefficients of the samples on both sides are calculated respectively, and the final partition threshold Gini coefficient is obtained by weighted average according to the number of samples. The partition threshold with the smallest Gini coefficient is selected as the optimal partition threshold for continuous features. Then, the Gini coefficients of m features are compared, and the feature vector with the smallest Gini coefficient is selected as the optimal split feature. Its optimal partition threshold is used as the partition point. The above node classification selection process is repeated until the minimum number of samples for internal node splitting is 2 or the depth of the tree reaches 10, and the splitting is stopped. The decision tree is constructed.

[0021] In this manual, the discontinuous factors (average slope, stratum lithology, distance from faults, and distance from water systems) are standardized to be in the range of [-1, 1].

[0022] In this manual, accuracy, recall, precision, F1 score, ROC curve and AUC value are used to evaluate model accuracy.

[0023] In this manual, two landslide susceptibility evaluation models were established based on the two experimental variables of maximum slope and average slope in the evaluation unit. After the landslide susceptibility evaluation model based on random forest was run normally, the susceptibility index p of each evaluation unit in the study area was calculated, that is, the landslide occurrence probability predicted by the random forest model. The landslide susceptibility was divided into very high susceptibility, very high susceptibility and very high susceptibility according to the evaluation unit susceptibility index p using ArcGIS 10.8 software. , Gao Yifa , Zhongyifa , low incidence , extremely low incidence , and generate landslide susceptibility evaluation maps of the study area under different slope gradient generalization methods.

[0024] In summary, the present invention has at least the following beneficial effects:

[0025] This paper uses a random forest model to assess the landslide susceptibility of a selected study area. Other parameters in the susceptibility assessment system are referenced to commonly used parameters from previous studies. The landslide sample slope generalization parameters are calculated using the two aforementioned methods, respectively, to evaluate the accuracy of different slope gradient estimation methods. The accuracy of the susceptibility assessment results is verified using accuracy, recall, precision, F1 score, and receiver operating characteristic (ROC) curves. This method effectively improves prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0027] Figure 1 Schematic diagram of the landslide susceptibility prediction method based on maximum slope rate and gradient generalization involved in the present invention.

[0028] Figure 2 Schematic diagram of the maximum slope calculation method involved in the present invention.

[0029] Figure 3 Schematic diagram of the decision tree model involved in the present invention.

[0030] Figure 4 The figure is a schematic diagram of the model construction process involved in the present invention.

[0031] Figure 5 Schematic diagram of the ROC curve of the susceptibility evaluation model involved in the present invention. DETAILED DESCRIPTION

[0032] Hereinafter, only certain exemplary embodiments are briefly described. As will be appreciated by those skilled in the art, the described embodiments may be modified in various ways without departing from the spirit or scope of the embodiments of the present invention. Therefore, the drawings and description are to be regarded as illustrative in nature and not restrictive.

[0033] The disclosure below provides many different embodiments or examples for implementing different structures of the embodiments of the present invention. In order to simplify the disclosure of the embodiments of the present invention, the components and configurations of specific examples are described below. Of course, these are merely examples and are not intended to limit the embodiments of the present invention. In addition, the embodiments of the present invention may repeat reference numerals and / or reference letters in different examples. Such repetition is for the purpose of simplicity and clarity and does not in itself indicate the relationship between the various embodiments and / or configurations discussed.

[0034] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0035] like Figure 1 As shown, this embodiment provides a landslide susceptibility prediction method based on maximum slope gradient generalization, including:

[0036] S1. Randomly select an equal number of non-landslide samples around the landslide;

[0037] S2. Randomly divide the landslide and non-landslide samples into a ratio of 7:3, with 70% as the training set and 30% as the test set;

[0038] S3. Perform row sampling with replacement on the row samples in the training set, perform column sampling on the characteristic factors, and randomly select samples from the characteristic factors to obtain a sample set containing row samples and column samples;

[0039] S4. Traverse the column samples, select the most appropriate features for splitting based on the Gini coefficient, and complete the construction of the decision tree;

[0040] S5. Repeat S3 and S4 to generate multiple decision trees to form a random forest;

[0041] S6. Input the test set into the random forest for prediction.

[0042] The present invention proposes a maximum slope index to characterize and generalize the slope index in landslide samples. The maximum slope index comprehensively considers the difference in terrain slope before and after the landslide sample slides. It is defined as the maximum slope value of the line connecting the landslide back wall (A) and the landslide front edge (B) in the sliding direction, and is expressed as Max(S), where A is the determined landslide back wall line and B is the determined landslide accumulation front line. Figure 2 As shown, the green line is the landslide back wall line A and the landslide front edge line B; the blue line is the slope line corresponding to the dividing point between the landslide back wall and the front edge; the red line is the maximum slope line, and Max(S) is the maximum slope gradient corresponding to the maximum slope line. Based on the ArcGIS 10.8 platform's point-along-line tool, fixed points A are generated at equal intervals on A and B. i、 B i (The default value is 10). The maximum slope value Max(S) is calculated by pairwise correspondence between the top and bottom. Compared with the traditional method, it can more accurately reflect the slope characteristics of the slope, which is expressed as:

[0043] ;

[0044] in: for and Point elevation difference; L for and Point distance difference.

[0045] In order to verify the effectiveness and advancement of this method, an experimental study was conducted in the Chaya area of ​​Qamdo, Tibet. The study area is located in the Chaya area of ​​Qamdo, Tibet Autonomous Region, which is a rectangular area from Qamdo Karo District to Chaya County, with an area of ​​4213.36 km 2The region is generally located on the eastern edge of the Qinghai-Tibet Plateau and the northern part of the Sanjiang area. It belongs to the tectonic-erosional and erosional alpine landforms. The topography and landforms vary greatly, the geological structure is complex, and faults, folds and other structures are distributed in northwest-trending strips. The main exposed strata are Quaternary alluvial-diluvial, Paleogene, Jurassic, Triassic, Permian, Devonian sandstone, mudstone, limestone, etc., and local igneous rocks are exposed. The region is generally high in the west and low in the east, with the lowest elevation of 3605m and the highest elevation of 5615m. The Lancang River runs through the western part of the study area from south to north, with gullies crisscrossing on both sides. The slope gradient is mostly between 20° and 40°, and locally greater than 40°.

[0046] The landslide susceptibility of the selected study area was assessed using a random forest model. Other parameters in the susceptibility assessment system were referenced to commonly used parameters from previous studies. The generalized slope parameters of the landslide samples were calculated using the two aforementioned methods to evaluate the accuracy of the different slope estimation methods. The accuracy of the susceptibility assessment results was verified using accuracy, recall, precision, F1 score, and receiver operating characteristic (ROC) curves.

[0047] Data source:

[0048] Landslide data were obtained from field surveys and remote sensing interpretation. A total of 133 landslide-prone slopes were developed in the study area. Fault and stratigraphic data were obtained from a 1:250,000 geological map; topographic data were based on a 1:50,000 topographic map; rainfall data were obtained from the Qinghai-Tibet Plateau National Data Center, with monthly average rainfall data from 1958 to 2022; and historical earthquake data were obtained from the National Earthquake Science Data Center.

[0049] Evaluation indicators selection:

[0050] The formation of landslides is mainly affected by factors such as topography, geological structure, and lithology. In addition to the two calculation methods for the slope factor mentioned above, in order to achieve a more accurate prediction model, this paper combines the conditions required for landslide occurrence and comprehensively evaluates the availability and computability of the factors to select the following evaluation indicators:

[0051] Factors related to topographic conditions

[0052] (1) Average elevation: refers to the average value of all elevation grids within the evaluation unit;

[0053] (2) Average slope aspect: refers to the average value of all slope aspect grids in the evaluation unit;

[0054] Geological condition-related factors

[0055] (1) Main lithologic properties: refers to the main lithologic combination in the study area. According to the characteristics of the lithologic properties in the study area, the main lithologic properties in the evaluation unit are extracted and divided into 15 categories, namely ①C1k.; ②D3z; ③E g;④J1w⑤J2d;⑥J3x;⑦P2j;⑧Pt 1-2 K.\Pt 1-2 J.\Pt3Y;⑨Qpal;⑩T 1-2 m; ⑪T3a; ⑫T3b; ⑬T3d; ⑭T3j; ⑮η\γ\δ\ψ l \;

[0056] (2) Distance from the fault: Divide the evaluation unit into intervals of 500 m, 1000 m, 2000 m, 4000 m, and >4000 m, and extract the main distance values ​​from the fault within the evaluation unit;

[0057] (3) Average historical earthquake point density: refers to the average value of the point density within the evaluation unit extracted using the point density tool in ArcGIS based on the national historical earthquake record data (2022 edition);

[0058] Hydrological condition-related factors

[0059] (1) Distance from water system: Divide the evaluation unit into intervals of 500 m, 1000 m, 1500 m, 2000 m, 2500 m, and >2500 m, and extract the main distance values ​​from the water system within the evaluation unit;

[0060] (2) Annual average rainfall: obtained by averaging the monthly average rainfall from 1958 to 2022. The calculation formula is as follows: Annual average rainfall m = total rainfall S / year n;

[0061] Random Forest Model

[0062] Random forest is an ensemble learning method consisting of multiple decision trees, each trained independently with a randomly selected subset of input features. By voting or averaging the results of multiple decision trees, random forests effectively avoid overfitting and exhibit good generalization and robustness.

[0063] First, a portion of samples is randomly selected from the training dataset to construct each decision tree. This process is called "bootstrap sampling" or "out-of-bag sampling," where the probability of each sample being selected is 1 / n, where n is the total number of samples. Each time a node in the decision tree is partitioned, a subset of features is randomly selected, and the best feature is chosen from them for partitioning. This process can reduce the correlation between features and enhance the stability and generalization ability of the model. Then, based on the above steps, multiple decision trees are constructed. Each decision tree is generated by continuously binary partitioning the features until the stopping condition is met. Typically, random forests use the CART algorithm to construct decision trees. Finally, for classification tasks, random forests use voting to predict the category of the target variable; for regression tasks, random forests use the average to predict the value of the target variable.

[0064] Model accuracy evaluation:

[0065] Commonly used methods for evaluating the accuracy of random forest models include accuracy, recall, F1 score, confusion matrix, ROC curve, AUC value, and cross-validation. This paper selects accuracy, recall, precision, and F1 score for model accuracy evaluation. The ROC curve is plotted to demonstrate the relationship between the true positive rate (TPR) and false positive rate (FPR) at different thresholds, and the area under the curve (AUC value) is calculated to measure the classification performance of the random forest model. AUC values ​​closer to 1 indicate better classifier performance; conversely, AUC values ​​closer to 0 indicate worse classifier performance. Theoretically, a perfect classifier has an AUC value of 1, while a random classifier has an AUC value of 0.5.

[0066] Model construction:

[0067] The evaluation unit used in this study was the slope unit. 133 landslides in the study area were selected as the research objects, and an equal number of non-landslide samples were randomly selected from the surrounding landslide areas. Slopes with landslides were marked as "1," and slopes without landslides were marked as "0." The average slope and maximum slope rate of the evaluation unit were combined with data from seven other evaluation factors to form a data set, which was then used to train and test the landslide susceptibility evaluation model.

[0068] like Figure 4 As shown, the model building process is as follows:

[0069] (1) Data processing

[0070] First, the discontinuous factors (average slope, stratum lithology, distance from faults, and distance from water systems) were standardized to be in the range of [-1, 1].

[0071] (2) Data division

[0072] The 266 input samples are randomly divided into a ratio of 7:3, of which 70% are the training set T (186 samples) and 30% are the test set C (80 samples).

[0073] (3) Random sampling

[0074] The model samples the rows in the training set T (i.e., 186 samples) 186 times with replacement. This sampling with replacement results in duplicate samples in the collected row data. This ensures that during training, the input samples for each decision tree are not all samples, making overfitting less likely. After row sampling, the model performs column sampling on the feature factors, randomly selecting m samples (m ≤ 8) from the eight feature factors mentioned above. This creates an input sample set consisting of 186 row samples and m column samples.

[0075] (4) Constructing a decision tree

[0076] Traverse the m feature vectors and select the most appropriate feature for splitting. The feature selection method is the Gini impurity. The model performs feature selection based on the Gini coefficient. The Gini coefficient is an indicator of the purity of a data set. It measures the probability that two samples randomly selected from the data set have inconsistent category labels. The value range of the Gini coefficient is 0 to 1, where 0 indicates the highest purity of the data set and 1 indicates the lowest purity of the data set. For example, m column samples are randomly selected in a certain sampling. Since non-continuous features are standardized to form continuous data, the model for continuous features will first sort all possible values. For each possible partition threshold, samples in the data set whose values ​​of the feature are less than or equal to the partition threshold are divided into one side l, and samples whose values ​​are greater than the partition threshold are divided into the other side r. Their Gini coefficients are:

[0077] ;

[0078] ;

[0079] in, is the Gini coefficient corresponding to node l, is the Gini coefficient corresponding to node r, is the probability of the first type of data point in node l, is the probability of the second type of data point in node l, The probability of a data point of type 1 in node r, is the probability of the second type of data point in node r; the evaluation unit where landslide has occurred is recorded as "1", and the evaluation unit where landslide has not occurred is recorded as "0".

[0080] For each partition threshold, the Gini coefficients of the samples on both sides are calculated separately, and the final partition threshold Gini coefficient is obtained by weighted average according to the number of samples. The partition threshold with the smallest Gini coefficient is selected as the optimal partition threshold for continuous features. Then the Gini coefficients of the m features are compared, and the feature vector with the smallest Gini coefficient is selected as the optimal split feature, and its optimal partition threshold is used as the partition point. Repeat the above node classification selection process until the minimum number of samples for internal node splitting is 2 or the depth of the tree reaches 10 to stop splitting. The decision tree is constructed, as shown in the figure below. Figure 3 As shown, blue represents classes with a label value of 1 (corresponding to landslides), and orange represents classes with a label of 0 (corresponding to non-landslides). Darker colors indicate higher classification purity. The closer the Gini coefficient is to 0, the higher the purity. The first line is the abbreviation for the evaluation factor, and the four parameters represent the Gini coefficient (gini), the number of samples (samples), the class count (value), and the final classification (class).

[0081] (5) Constructing a random forest

[0082] Repeat (3) and (4) 100 times to generate 100 decision trees. The 100 constructed decision trees are combined into a random forest. The mode of the prediction results of the 100 decision trees is used as the final output of the random forest prediction model.

[0083] (6) Prediction

[0084] The randomly divided test set C in (2) is brought into the random forest prediction model constructed in (5) to predict the results.

[0085] (7) Evaluate model performance

[0086] The accuracy of the model was evaluated using accuracy, recall, precision, F1 score, ROC curve and AUC value.

[0087] Evaluation results:

[0088] Two landslide susceptibility evaluation models were established based on the two experimental variables of maximum slope and average slope in the evaluation unit. After the landslide susceptibility evaluation model based on random forest was run normally, the susceptibility index p of each evaluation unit in the study area was calculated, that is, the landslide occurrence probability predicted by the random forest model. The landslide susceptibility was divided into extremely high susceptibility (low susceptibility) and low susceptibility (low susceptibility) according to the evaluation unit susceptibility index p using ArcGIS 10.8 software. )、High Yifa ( )、 Medium susceptibility )、 Low susceptibility )、 Very low susceptibility ) five categories.

[0089] Model accuracy evaluation:

[0090] In order to evaluate the modeling impact of two slope unit slope generalization methods on the random forest landslide susceptibility model, the accuracy, recall rate, precision, F1 score, ROC curve and AUC value are used to evaluate the model accuracy of the two models. The model accuracy evaluations of the two models are shown in Table 1.

[0091] Table 1 Model accuracy evaluation table

[0092]

[0093] The above table shows the classification evaluation indicators of the test set, and the classification effect of the random forest on the test data is measured by quantitative indicators. Among them, the accuracy rate is the proportion of correctly predicted samples in the total samples, and the larger the accuracy rate, the better. The recall rate is the proportion of the results that are actually positive samples and are predicted as positive samples, and the larger the recall rate, the better. The precision rate is the proportion of the results that are predicted as positive samples and are actually positive samples, and the larger the precision rate, the better. The F1 score is the harmonic mean of the precision rate and the recall rate.

[0094] In the model accuracy evaluation, the accuracy rate of the test set is usually used to evaluate the accuracy of the model. As can be seen from the above table, the accuracy rate, recall rate, precision rate and F1 score of the random forest landslide susceptibility evaluation model with the data processing method of the maximum slope rate (Max(S)) as the slope generalization are 91.25%, 91.54%, 91.22% and 91.22% respectively, which are 7.50%, 7.07%, 6.90% and 7.47% higher than the generalization processing method of the average slope, respectively, proving that the generalization processing method of the maximum slope rate can significantly improve the accuracy of the model.

[0095] The ROC curve is a curve commonly used in machine learning to evaluate the performance of classifiers. Since the ROC curve is obtained by gradually decreasing the landslide susceptibility threshold to obtain the coordinates corresponding to each point on the curve, the closer the curve is to the upper left corner, the more landslides are concentrated in a high-level and small-range interval, and the larger the area under the curve (AUC value), so the better the prediction effect. The ROC curves of the random forest models under the two slope generalization processing methods of the present invention are as Figure 5As shown in the figure, the AUC values ​​of the maximum slope generalization (Max(S)) and average slope (AS) models are 0.952 and 0.933, respectively. The AUC value of the maximum slope generalization model is significantly higher than that of the other two models, demonstrating that the slope data processing method of the maximum slope generalization method can improve the performance of the random forest landslide susceptibility prediction model.

[0096] The present invention proposes a slope generalization calculation method for landslide susceptibility assessment and compares this method with the commonly used methods. The results show that in model accuracy evaluation, the accuracy of the test set is usually used to evaluate the accuracy of the model. The random forest landslide susceptibility assessment model using the maximum slope gradient (Max(S)) as the data processing method for slope gradient generalization has an accuracy rate, recall rate, precision rate and F1 score of 91.25%, 91.54%, 91.22% and 91.22% respectively, which are 7.50%, 7.07%, 6.90% and 7.47% higher than the average slope generalization processing method, respectively. This proves that the maximum slope gradient generalization processing method can significantly improve the accuracy of the model.

[0097] The ROC curve results showed that the AUC values ​​of the maximum slope generalization model and the average slope generalization model were 0.952 and 0.933, respectively. The AUC value of the maximum slope generalization model was significantly higher than that of the other two models, demonstrating that the maximum slope generalization data processing method can improve the performance of the random forest landslide susceptibility prediction model.

[0098] The above embodiments are intended to illustrate the present invention, not to limit the present invention. Therefore, changes in illustrative values ​​or substitutions of equivalent components should still fall within the scope of the present invention.

[0099] From the above detailed description, it will be clear to those skilled in the art that the present invention can indeed achieve the aforementioned objectives and is in compliance with the provisions of the Patent Law.

[0100] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they become aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as covering the preferred embodiments and all changes and modifications that fall within the scope of the invention. The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

[0101] It should be noted that the above description of the relevant processes is for illustration and purpose only and does not limit the scope of application of this specification. For those skilled in the art, various modifications and changes can be made to the processes under the guidance of this specification. However, such modifications and changes are still within the scope of this specification.

[0102] The basic concepts have been described above. It will be apparent to those skilled in the art after reading this application that the above disclosures are merely illustrative and do not constitute limitations on this application. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and amendments to this application. Such modifications, improvements, and amendments are suggested in this application and remain within the spirit and scope of the exemplary embodiments of this application.

[0103] In addition, unless expressly stated in the claims, the order of the processing elements and sequences described in this application, the use of alphanumeric characters, or the use of other names are not intended to limit the order of the processes and methods of this application. Although the above disclosure discusses some embodiments of the invention that are currently considered useful through various examples, it should be understood that such details are only for illustrative purposes, and the attached claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that are consistent with the essence and scope of the embodiments of this application. For example, although the implementation of the various components described above can be embodied in a hardware device, it can also be implemented as a pure software solution, for example, installation on an existing server or mobile device.

[0104] Similarly, it should be noted that in order to simplify the presentation of this disclosure and thereby facilitate understanding of one or more of the invention's embodiments, the foregoing descriptions of the embodiments of this disclosure sometimes combine multiple features into a single embodiment, figure, or description thereof. However, this approach should not be interpreted as reflecting an intention that the claimed subject matter requires more features than expressly recited in each claim. Rather, the subject matter of the invention may possess fewer features than the single embodiment described above.

Claims

1. The landslide susceptibility prediction method based on the maximum slope gradient generalization is characterized by: include: S1. Randomly select an equal number of non-landslide samples around the landslide; S2. Randomly divide the landslide and non-landslide samples into a ratio of 7:3, with 70% as the training set and 30% as the test set; S3. Perform row sampling with replacement on the row samples in the training set and column sampling on the characteristic factors. Randomly select samples from the characteristic factors to obtain a sample set containing row samples and column samples. The characteristic factors include the maximum slope gradient, which is defined as the maximum slope value of the line connecting the landslide back wall line and the landslide accumulation front line in the sliding direction, represented by Max(S). Generate fixed points at equal intervals on A and B. and , calculate the maximum slope value Max(S) by pairwise correspondence between top and bottom, expressed as: ； Among them, A is the determined landslide back wall line, B is the determined landslide accumulation front line, for and Point elevation difference; for and Point distance difference; S4. Traverse the column samples, select the most appropriate features for splitting based on the Gini coefficient, and complete the construction of the decision tree; S5. Repeat S3 and S4 to generate multiple decision trees to form a random forest; S6. Input the test set into the random forest for prediction.

2. The landslide susceptibility prediction method based on maximum slope gradient generalization according to claim 1 is characterized in that: Slope units were selected as evaluation units, and characteristic factors of the evaluation units were obtained. The characteristic factors included average elevation, average slope aspect, stratum lithology, distance from faults, average historical earthquake point density, distance from water systems, and average annual rainfall. The average slope and maximum slope of the evaluation unit were combined with the characteristic factors to form data sets, and the landslide susceptibility evaluation model was trained and tested respectively.

3. The landslide susceptibility prediction method based on maximum slope gradient generalization according to claim 2 is characterized in that: 133 landslides were selected, and an equal number of non-landslide samples were randomly selected around the landslides. The slopes where landslides occurred were marked as 1, and the slopes where landslides did not occur were marked as 0, resulting in 266 input samples.

4. The landslide susceptibility prediction method based on maximum slope gradient generalization according to claim 3 is characterized in that: The 186 row samples in the training set were sampled 186 times with replacement. Sampling with replacement will result in duplicate samples in the collected row sample data. After the row sampling is completed, the characteristic factors are sampled in columns. M samples are randomly selected from the 8 characteristic factors, m≤8, to form an input sample set containing 186 row samples and m column samples. The 8 characteristic factors are average slope, average elevation, average slope aspect, stratum lithology, distance from faults, average historical earthquake point density, distance from water system and average annual rainfall, or the 8 characteristic factors are maximum slope, average elevation, average slope aspect, stratum lithology, distance from faults, average historical earthquake point density, distance from water system and average annual rainfall.

5. The landslide susceptibility prediction method based on maximum slope gradient generalization according to claim 4 is characterized in that: Traverse the m feature vectors and select the most appropriate feature for splitting. The feature selection method is the Gini coefficient. Feature selection is performed based on the Gini coefficient. The value range of the Gini coefficient is 0 to 1, where 0 indicates the highest purity of the data set and 1 indicates the lowest purity of the data set. For each possible partition threshold, samples with a feature value less than or equal to the partition threshold in the data set are divided into one side l, and samples with a value greater than the partition threshold are divided into the other side r. The Gini coefficients are: ; ； in, is the Gini coefficient corresponding to node l, is the Gini coefficient corresponding to node r, is the probability of the first type of data point in node l, is the probability of the second type of data point in node l, The probability of a data point of type 1 in node r, is the probability of the second type of data point in node r; For each partition threshold, the Gini coefficients of the samples on both sides are calculated respectively, and the final partition threshold Gini coefficient is obtained by weighted average according to the number of samples. The partition threshold with the smallest Gini coefficient is selected as the optimal partition threshold for continuous features. Then, the Gini coefficients of m features are compared, and the feature vector with the smallest Gini coefficient is selected as the optimal split feature. Its optimal partition threshold is used as the partition point. The above node classification selection process is repeated until the minimum number of samples for internal node splitting is 2 or the depth of the tree reaches 10, and the splitting is stopped. The decision tree is constructed.

6. The landslide susceptibility prediction method based on maximum slope gradient generalization according to claim 5 is characterized in that: The non-continuous factors are standardized so that they are in the range of [-1,1].

7. The landslide susceptibility prediction method based on maximum slope gradient generalization according to claim 6 is characterized in that: The accuracy of the model was evaluated using accuracy, recall, precision, F1 score, ROC curve and AUC value.

8. The landslide susceptibility prediction method based on maximum slope gradient generalization according to claim 7 is characterized in that: Two landslide susceptibility evaluation models were established based on the two experimental variables of maximum slope and average slope in the evaluation unit. After the landslide susceptibility evaluation model based on random forest was run normally, the susceptibility index p of each evaluation unit in the study area was calculated, that is, the landslide occurrence probability predicted by the random forest model. The landslide susceptibility was divided into extremely high susceptibility and extremely high susceptibility according to the evaluation unit susceptibility index p using ArcGIS10.8 software. , Gao Yifa , Zhongyifa , low incidence , extremely low incidence , and generate landslide susceptibility evaluation maps of the study area under different slope gradient generalization methods.

Images