Landslide susceptibility mapping method, equipment and medium

Abstract

The present invention relates to the field of digital mapping and discloses a landslide susceptibility mapping method, device and medium. The method comprises the following steps: acquiring multi-source remote sensing data and historical landslide data to produce a hazard factor dataset and a landslide sample dataset; inputting all landslide sample datasets into an AGNES clustering algorithm and a VIIQ feature optimization strategy module to obtain an optimal feature combination; dividing the optimized data features into a training set and a test set, inputting the data features into a machine learning and deep learning classifier to perform model accuracy evaluation to obtain a preliminary landslide susceptibility map; inputting the landslide susceptibility map into a blending heterogeneous integrated model to obtain a final landslide susceptibility mapping result; and combining a SHAP algorithm to deeply interpret the integrated model and optimize the landslide susceptibility result map. The present invention can more effectively predict landslide susceptibility and provide a scientific basis for risk assessment, early warning and prevention of geological disasters.

Description

Technical Field

[0001] The present invention relates to the field of digital mapping, and in particular to a landslide susceptibility mapping method, equipment and medium. Background Art

[0002] Among natural disasters, landslides occur far more frequently than other types of geological hazards and are extremely destructive. Earthquake-induced landslides, in particular, are often extremely sudden and present a significant threat to human survival and development. Earthquake-induced landslides not only cause direct damage but can also trigger secondary disasters such as dam breaches and floods, further exacerbating the impact.

[0003] To assist relevant departments in carrying out effective and rational disaster prevention and mitigation, landslide susceptibility mapping has become a key research method. By studying environmental factors related to landslides, the probability of recurrence in similar environments is analyzed, and a landslide susceptibility map is created. Based on the resulting landslide susceptibility zoning map, the risk of a regional geological hazard and the potential impact area can be determined.

[0004] Data-driven models are based on objective, complex geographic data (such as elevation, geology, and precipitation) and comprehensive, detailed historical landslide data, reducing the influence of subjective factors and uncertainty. Currently, two main types of models are used: traditional statistical methods and machine learning algorithms. Traditional statistical models focus on linear relationships between landslide characteristic factors, whereas the relationship between landslide events and these factors is primarily nonlinear. Machine learning and deep learning algorithms can better construct nonlinear relationships between landslide characteristic factors, effectively capturing and simulating the complex mechanisms of landslide occurrence, and significantly improving their prediction accuracy. Given the complexity of landslide susceptibility mapping, a single machine learning algorithm cannot fully and accurately reflect the inherent connections between landslide influencing factors and actual catastrophic events in all scenarios.

[0005] Furthermore, landslides have complex mechanisms of formation, and the regional characteristics of various hazard factors significantly influence their distribution. Currently, research on landslide susceptibility primarily focuses on model selection and optimization, while in-depth study of the model's prediction principles and algorithmic details is relatively limited. Previous approaches have emphasized the correlation between landslide characteristic factors and landslide events, but have insufficiently considered the impact of the spatiotemporal characteristics and spatial heterogeneity of the geographic environment on landslide events.

[0006] Therefore, in landslide susceptibility mapping technology, how to effectively utilize spatiotemporal characteristics and consider spatial heterogeneity is a technical problem that needs to be solved urgently. Summary of the Invention

[0007] The purpose of the present invention is to propose a landslide susceptibility mapping method to solve the technical problems of how to effectively utilize spatiotemporal characteristics and how to consider spatial heterogeneity.

[0008] Specifically, the present invention provides a method for mapping landslide susceptibility, comprising the following steps:

[0009] S1: Acquire multi-source remote sensing data and historical landslide data to create hazard factor datasets and landslide sample datasets;

[0010] S2: All landslide sample data sets are input into the AGNES clustering algorithm and VIIQ feature optimization strategy module to obtain the optimal feature combination;

[0011] S3: Divide the optimized data features into training and test sets, input them into machine learning and deep learning classifiers, evaluate the model accuracy, and obtain a preliminary landslide susceptibility map;

[0012] S4: Input the preliminary landslide susceptibility map of S3 into the blending heterogeneous integration model to obtain the final landslide susceptibility mapping result;

[0013] S5: The blending heterogeneous integration model of S4 is combined with the SHAP algorithm to provide an in-depth interpretation of the integration model and optimize the landslide susceptibility result map.

[0014] A storage medium stores instructions and data for implementing a landslide susceptibility mapping method.

[0015] A landslide susceptibility mapping device comprises: a processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement a landslide susceptibility mapping method.

[0016] The beneficial effects provided by the present invention are:

[0017] (1) The AGNES clustering algorithm and VIIQ (VIF-IGR-Importance-Qvalue) feature optimization strategy module provided by the present invention are helpful to consider and deal with the spatial heterogeneity problem in landslide susceptibility evaluation. The AGNES clustering method can capture the characteristics of the local area by dividing the global grid unit of the study area into several local sub-areas, thereby laying the foundation for selecting the optimal combination of landslide condition factors for each sub-area. The VIIG feature combination method is used to screen the optimal spatial features of each sub-area, reduce feature redundancy and improve the efficiency of landslide susceptibility evaluation. Therefore, through the AGNES clustering algorithm and the VIIQ (VIF-IGR-Importance-Qvalue) feature optimization strategy module, the spatial characteristics of different disaster-causing factors can be better understood in a comprehensive manner to better analyze the landslide susceptibility in the region.

[0018] (2) The present invention predicts landslide events by constructing machine learning and deep learning models, covering a variety of advanced classifiers to improve the accuracy and generalization ability of landslide susceptibility assessment. Specifically, the machine learning models include decision trees, random forests (RF), support vector machines (SVM), gradient boosted regression trees (GBRT), and K nearest neighbors (KNN), while the deep learning models include artificial neural networks (ANN), convolutional neural networks (CNN), and recurrent neural networks (RNN). These models perform well in processing the optimal feature combinations obtained by the AGNES clustering algorithm and the VIIQ (VIF-IGR-Importance-Qvalue) feature optimization strategy module, and can further perform classification tasks. In addition, the present invention combines random forests (RF), gradient boosted trees (GBRT), and K nearest neighbors (KNN) through the blending algorithm to construct a heterogeneous integrated model based on the VIIQ feature combination. This integration method not only improves the accuracy of the model, but also enhances the generalization ability of the model, enabling it to better adapt to different landslide susceptibility assessment scenarios. In this way, the present invention can more effectively predict the susceptibility of landslides, provide a scientific basis for risk assessment, early warning and prevention of geological disasters, thereby protecting people's lives and property and reducing losses caused by disasters.

[0019] (3) The SHAP algorithm provided by the present invention makes the model's decision-making process more transparent and improves the model's interpretability by quantifying the contribution of each feature to the model's prediction. SHAP values ​​can reveal which features have the greatest impact on the model's predictions, providing a quantitative method for identifying key factors in landslide susceptibility assessment. The SHAP algorithm helps analyze whether model predictions are fair and unbiased, which is crucial for high-risk decision-making and ensures the reliability and impartiality of model predictions.

[0020] (4) The technology provided by the present invention for landslide susceptibility mapping that combines the spatiotemporal characteristics and spatial heterogeneity of the geographical environment can take into account the particularity and diversity of landslide events in a specific area, supplement the shortcomings of existing machine learning and deep learning technologies applied to landslide susceptibility mapping, and more comprehensively discover areas with landslide tendencies, providing assistance for landslide geological disaster prevention and control work in the region. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic flow chart of the method of the present invention;

[0022] Figure 2 This is an AGNES algorithm structure diagram of a landslide susceptibility mapping method that integrates spatiotemporal characteristics and spatial heterogeneity according to the present invention;

[0023] Figure 3This is a schematic diagram and flow chart of an optimal combination method of VIIQ features that integrates spatiotemporal features and spatial heterogeneity in the present invention;

[0024] Figure 4 This is a structural diagram of a landslide susceptibility mapping method using a blending heterogeneous integrated model combining machine learning and deep learning;

[0025] Figure 5 This is a landslide susceptibility result diagram of a blending heterogeneous integrated model of the present invention;

[0026] Figure 6 This is a post-global interpretation result diagram of landslide susceptibility results based on the blending heterogeneous integrated model of the present invention;

[0027] Figure 7 It is a schematic diagram of the working of the hardware device of an embodiment of the present invention. DETAILED DESCRIPTION

[0028] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0029] Before formally explaining the present invention, the scheme of the present invention is first generally explained for easy understanding.

[0030] Please refer to Figure 1 The present invention provides a method for mapping landslide susceptibility, comprising the following steps:

[0031] S1: Acquire multi-source remote sensing data and historical landslide data to create hazard factor datasets and landslide sample datasets;

[0032] As an example, multi-source remote sensing data includes but is not limited to remote sensing image data. Hazard factor data includes, but is not limited to, depending on the selected study area: lithology, elevation, curvature, plan curvature, profile curvature, slope, aspect, surface roughness, terrain wetness index, normalized difference vegetation index, precipitation, distance from faults, distance from rivers, distance from roads, and surface cover data. The landslide sample dataset is derived from accumulated historical landslide data within the study area, as well as remote sensing data combined with landslide data and non-landslide data annotated by Google Earth, ensuring a balanced ratio of landslide samples to non-landslide samples of approximately 1:1.

[0033] S2: All landslide sample data sets are input into the AGNES clustering algorithm and VIIQ feature optimization strategy module to obtain the optimal feature combination;

[0034] As an example, all landslide samples are input into the AGNES clustering algorithm to obtain different partitions. Figure 2 As shown in Figure 1, it aggregates data points into clusters by stepwise merging. This method starts with each data point as a separate cluster and then iteratively merges the most similar cluster pairs until all data points are aggregated into a large cluster or the preset number of clusters is reached. The application of the AGNES clustering algorithm can effectively identify landslide areas with similar characteristics, which may be affected by similar geological conditions and environmental factors. The advantages and disadvantages of different clustering schemes are evaluated by calculating the CH index, which is an indicator of clustering effectiveness based on intra-cluster similarity and inter-cluster dissimilarity.

[0035] In each cluster, landslide samples were classified based on the VIIQ (VIF-IGR-Importance-Qvalue) method. Figure 3 As shown, the most characteristic combination is performed.

[0036] During feature selection, the variance inflation factor (VIF) was used to test for multicollinearity among the characteristic factors to ensure model stability. The information gain ratio (IGR) was then calculated to help determine which factors contributed most to the model's predictions. Subsequently, the Importance algorithm was introduced to calculate the importance of landslide features and further identify features that contribute to the landslide susceptibility assessment model. Finally, the geographic detector Qvalue was used for further optimization. By spatially partitioning the study area and fitting a local linear model, the most discriminative and explanatory features were selected.

[0037] S3: Divide the optimized data features into training and test sets, input them into machine learning and deep learning classifiers, evaluate the model accuracy, and obtain a preliminary landslide susceptibility map;

[0038] As an example, the model process of machine learning and deep learning classifiers combines the AGNES clustering algorithm and the VIIQ feature optimization strategy module to obtain the spatial characteristics of landslides. Decision trees, random forests (RF), support vector machines (SVM), gradient boosted regression trees (GBRT), K-nearest neighbors (KNN), artificial neural networks (ANN), convolutional neural networks (CNN), and recurrent neural networks (RNN) are then used to classify all data into landslide and non-landslide categories. Finally, the probability of a landslide occurring is output, resulting in a single model landslide susceptibility map.

[0039] S4: Input the preliminary landslide susceptibility map of S3 into the blending heterogeneous integration model to obtain the final landslide susceptibility mapping result;

[0040] It should be noted that in order to construct the optimal landslide susceptibility model, the invention adopts a method that integrates machine learning and deep learning models. First, the machine learning model and the deep learning model are compared and analyzed to determine the performance of each. Figure 4 As shown in Figure 1, the optimal three learning models were selected for heterogeneous ensemble. This process requires a deep understanding of the different models to ensure that the integrated model can maximize the accuracy of landslide susceptibility prediction. Subsequently, this paper compared the ensemble model with a single learning model to verify the effectiveness of the blending ensemble strategy for coupling multiple machine learning or deep learning algorithms. By combining the prediction results of different models, the blending ensemble strategy aims to improve the model's generalization ability and prediction accuracy. This ensemble approach can fully leverage the strengths of each model and compensate for the shortcomings of a single model, thereby obtaining more reliable results in landslide susceptibility assessment.

[0041] In this example, to test the performance of the proposed landslide susceptibility mapping model, the region of Nepal affected by the Gorkha earthquake was selected as the research object to illustrate the model's performance and mapping results. To comprehensively assess the reliability of each method in landslide susceptibility mapping and select the optimal method for accurate mapping, this paper uses five evaluation methods to assess each model: Accuracy, Precision, Recall, F1 score, and AUC value.

[0042] First, the positive sample P (Positive) is recorded as 1, and the negative sample (Negative) is recorded as 0. TP (True Positive) means that the predicted value and the actual value are both 1, that is, the classification is correct; FP (False Positive) means that the predicted value is 1 and the actual value is 0, that is, the prediction is wrong; FN (False Negative) means that the predicted value is 0 and the actual value is 1, that is, the prediction is wrong; TN (True Negative) means that the predicted value and the actual value are both 0, that is, the prediction is correct.

[0043] Table 1 shows a comparison of model performance using the proposed method. The blending ensemble model based on feature optimization achieved a 3.65% improvement in AUC, a 3.71% improvement in accuracy, a 0.72% improvement in precision, a 1.98% improvement in recall, and a 0.41% improvement in F1-Score. This demonstrates that the addition of VIIQ for feature optimization further improves model performance.

[0044] Table 1 Comparison of model performance of the method of the present invention

[0045]

[0046]

[0047] The preliminary mapping results of the present invention using the VIIQ and Blending heterogeneous integrated model are as follows Figure 5 As shown, all landslide susceptibility values ​​are classified as very low (0 to 0.2), low (0.2 to 0.4), medium (0.4 to 0.6), high (0.6 to 0.8) and very high (0.8 to 1).

[0048] S5: The blending heterogeneous integration model of S4 is combined with the SHAP algorithm to provide an in-depth interpretation of the integration model and optimize the landslide susceptibility result map.

[0049] The present invention takes the area of ​​Nepal that was hit by the Gorkha earthquake as the research object, uses the SHAP algorithm, selects the landslide condition factors based on this paper, and combines the blending integrated model to study the factor contribution degree, the local interpretation and the global interpretation of the model. The results of the SHAP method in the present invention are as follows Figure 6 shown.

[0050] See Figure 7 , Figure 7 4 is a schematic diagram of the working of the hardware device of an embodiment of the present invention, wherein the hardware device specifically comprises: a landslide susceptibility mapping device 401, a processor 402 and a storage medium 403.

[0051] A landslide susceptibility mapping device 401: the landslide susceptibility mapping device 401 implements the landslide susceptibility mapping method.

[0052] Processor 402: The processor 402 loads and executes instructions and data in the storage medium 403 to implement the landslide susceptibility mapping method.

[0053] Storage medium 403: The storage medium 403 stores instructions and data; the storage medium 403 is used to implement the landslide susceptibility mapping method.

[0054] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for mapping landslide susceptibility, characterized by: The method comprises the following steps: S1: Acquire multi-source remote sensing data and historical landslide data to create hazard factor datasets and landslide sample datasets; S2: All landslide sample data sets are input into the AGNES clustering algorithm and VIIQ feature optimization strategy module to obtain the optimal feature combination; S3: Divide the optimized data features into training and test sets, input them into machine learning and deep learning classifiers, evaluate the model accuracy, and obtain a preliminary landslide susceptibility map; S4: Input the preliminary landslide susceptibility map of S3 into the blending heterogeneous integration model to obtain the final landslide susceptibility mapping result; S5: The blending heterogeneous integration model of S4 is combined with the SHAP algorithm to further interpret the integration model and optimize the landslide susceptibility result map; The working process of the VIIQ feature optimization strategy module in step S2 is as follows: The variance inflation factor (VIF) was used to test the multicollinearity of the characteristic factors. The information gain ratio (IGR) is used to assist in determining the contribution to the model prediction effect. The information gain ratio (IGR) is used to assist in determining the feature factor group that contributes the best to the model prediction effect. Calculate the importance of landslide features and optimize feature combinations based on importance; The optimized feature combination is further screened out using the geographic detector Qvalue to obtain the final feature combination.

2. A landslide susceptibility mapping method according to claim 1, characterized in that: Step S1 is specifically as follows: the data set in step S2 is divided into discrete data and continuous data, and input into the AGNES clustering algorithm and VIIQ feature optimization strategy module.

3. A landslide susceptibility mapping method according to claim 1, characterized in that: The AGNES clustering algorithm determines the inter-cluster distance measurement criterion and the optimal number of clusters based on the CH index. The CH index calculation formula is: in, is the sum of the distances between classes, i.e., the inter-cluster dispersion; is the sum of the intra-cluster distances, i.e., the intra-cluster dispersion; is the total number of samples, is the number of clusters, It is The number of samples in a cluster; It is The center point of each cluster; is the sample point.

4. A landslide susceptibility mapping method according to claim 1, characterized in that: The blending heterogeneous ensemble model in step S4 consists of a base learner and a meta-learner.

5. A landslide susceptibility mapping method according to claim 4, characterized in that: The base learners include random forest, GBRT and KNN; the meta learner adopts logistic regression.

6. A landslide susceptibility mapping method according to claim 1, characterized in that: In step S5, the SHAP method explains the model's prediction results by decomposing the prediction value of the evaluation model into the sum of the contribution values ​​of each input feature. The calculation formula is: in, It is a feature of value; is the set of all features; Does not contain features The feature subset of In the feature subset The predicted value of the model below; In the feature subset Add features The predicted value of the posterior model; is a feature subset The number of features in is the total number of features.

7. A storage medium, characterized in that: The storage medium stores instructions and data for implementing a landslide susceptibility mapping method according to any one of claims 1 to 6.

8. A landslide susceptibility mapping device, characterized by: include: A processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement a landslide susceptibility mapping method according to any one of claims 1 to 6.

Images