A landslide susceptibility comprehensive evaluation method, system, device, medium and product
By introducing SBAS-InSAR deformation rate and static environmental factors, combined with recursive feature elimination and XGBoost model, dynamic and static data fusion for landslide susceptibility assessment was achieved, improving the timeliness and accuracy of the assessment, and providing a high-precision landslide susceptibility classification map and interpretable analysis.
Patent Information
- Application Number
- CN202610475237.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-10
- Publication Date
- 2026-07-07
AI Technical Summary
Traditional landslide susceptibility assessment methods cannot effectively integrate dynamic deformation information with static environmental factors, resulting in limited timeliness and accuracy of assessment results and difficulty in capturing the early signs of dynamic landslide evolution.
Using SBAS-InSAR deformation rate as the dynamic deformation factor and combining it with static environmental factors, the optimal feature subset is selected through recursive feature elimination algorithm, and the RFE-RF-XGBoost cascade ensemble model is trained to achieve feature-level fusion and collaborative learning of dynamic and static data, generating a spatial classification map of landslide susceptibility.
It significantly improves the timeliness and accuracy of landslide susceptibility assessment, effectively captures the precursors of landslide dynamic evolution, generates high-precision landslide susceptibility classification maps, and provides reliable decision-making basis through model interpretive analysis.
Smart Images
Figure CN122346642A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of geological disaster prediction technology, and in particular to a comprehensive evaluation method, system, equipment, medium and product for landslide susceptibility. Background Technology
[0002] Landslide susceptibility assessment is the cornerstone of geological hazard risk management, and its core lies in quantitatively evaluating the spatial probability of landslides occurring under specific geological environmental conditions. Traditional methods mainly rely on static environmental factors such as elevation, slope, lithology, and distance from water systems, and employ statistical models or shallow machine learning models such as logistic regression, information content method, and coefficient of determination. However, these traditional methods have two inherent drawbacks: First, at the model level, traditional statistical models struggle to characterize the complex nonlinear interactions between factors, while single machine learning models (such as support vector machines and decision trees) are prone to overfitting or insufficient generalization ability when dealing with high-dimensional, imbalanced sample data, resulting in limited partitioning accuracy and often exhibiting "salt-and-pepper noise" or blurred boundaries. Second, at the data level, relying entirely on static environmental factors fundamentally ignores the nature of landslides as a dynamic geological process. Landslides are usually preceded by continuous surface deformation, and this deformation information is the most direct and valuable dynamic precursor revealing the slope instability stage. Traditional static evaluation systems cannot integrate this dynamic time-series information, resulting in poor timeliness of evaluation results. They cannot reflect the current activity status and evolution trend of landslide bodies, making it difficult to provide timely warnings for landslides that are deforming or reactivating.
[0003] In recent years, temporal synthetic aperture radar interferometry (InSAR), particularly small baseline set interferometry, has been attempted to be introduced into landslide research due to its ability to monitor millimeter-level surface deformation with high precision and over a wide area. However, existing studies often use InSAR deformation data in the post-processing stage, such as updating landslide inventories or simply overlaying empirical matrices with static susceptibility maps. These methods fail to deeply embed deformation rate as a key input feature into the training process of machine learning models, thus hindering feature-level fusion and collaborative learning between dynamic and static data. This deficiency prevents the models from effectively capturing the precursors of dynamic landslide evolution, limiting the timeliness and accuracy of landslide susceptibility assessment results. Therefore, constructing an intelligent assessment model that can deeply integrate dynamic deformation information and static environmental factors, and possess strong nonlinear fitting capabilities and generalization performance, has become a key scientific problem and technological bottleneck for improving the accuracy of landslide susceptibility assessment. Summary of the Invention
[0004] The purpose of this application is to provide a comprehensive evaluation method, system, equipment, medium, and product for landslide susceptibility, in order to solve the problem of limited timeliness and accuracy of landslide susceptibility evaluation results.
[0005] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a comprehensive evaluation method for landslide susceptibility, including: The SBAS-InSAR deformation rate was used as a dynamic deformation factor, and combined with static environmental factors, the original feature dataset was constructed. A recursive feature elimination algorithm is used to rank and iteratively eliminate evaluation factors in the original feature dataset to determine the optimal feature subset; the evaluation factors include dynamic deformation factors and static environmental factors. Train the RFE-RF-XGBoost cascaded ensemble model based on the optimal feature subset; The optimal feature subset corresponding to all grid cells in the entire study area is input into the trained RFE-RF-XGBoost cascaded ensemble model to determine the spatial classification map of landslide susceptibility. The SBAS-InSAR deformation rate is used to verify and locally correct the different risk level areas divided in the landslide susceptibility spatial classification map, and the landslide susceptibility assessment results are determined.
[0006] Secondly, this application provides a comprehensive landslide susceptibility assessment system, including: The original feature dataset construction module is used to construct the original feature dataset by taking the SBAS-InSAR deformation rate as a dynamic deformation factor and combining it with static environmental factors. The optimal feature subset determination module is used to employ a recursive feature elimination algorithm to rank and iteratively eliminate evaluation factors in the original feature dataset to determine the optimal feature subset; the evaluation factors include dynamic deformation factors and static environmental factors; The model training module is used to train the RFE-RF-XGBoost cascaded ensemble model based on the optimal feature subset. The landslide susceptibility spatial classification map determination module is used to input the optimal feature subset corresponding to all grid cells in the entire study area into the trained RFE-RF-XGBoost cascade ensemble model to determine the landslide susceptibility spatial classification map. The landslide susceptibility assessment result determination module is used to verify and locally correct the different risk level areas divided in the landslide susceptibility spatial classification map using the SBAS-InSAR deformation rate, and determine the landslide susceptibility assessment result.
[0007] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described comprehensive evaluation method for landslide susceptibility.
[0008] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned comprehensive evaluation method for landslide susceptibility.
[0009] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned comprehensive evaluation method for landslide susceptibility.
[0010] According to the specific embodiments provided in this application, this application has the following technical effects: This application uses SBAS-InSAR deformation rate as a dynamic deformation factor and combines it with static environmental factors to construct an original feature dataset. A recursive feature elimination algorithm is employed to rank and iteratively eliminate evaluation factors in the original feature dataset based on their importance, determining the optimal feature subset. An RFE-RF-XGBoost cascade ensemble model is then trained. This application achieves feature-level fusion and collaborative learning of dynamic and static data by directly inputting SBAS-InSAR deformation rate as the core dynamic deformation factor into the RFE-RF-XGBoost cascade ensemble model for training. Subsequently, the optimal feature subset corresponding to all grid cells in the entire study area is input into the trained RFE-RF-XGBoost cascade ensemble model to determine a spatial classification map of landslide susceptibility. The SBAS-InSAR deformation rate is used to verify and locally correct the different risk level areas divided in the spatial classification map of landslide susceptibility, determining the landslide susceptibility assessment results. This effectively captures the dynamic evolution precursors of landslides, improving the timeliness and accuracy of the landslide susceptibility assessment results. Attached Figure Description
[0011] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0012] Figure 1 A schematic diagram of a comprehensive evaluation method for landslide susceptibility provided in an embodiment of this application; Figure 2 A flowchart illustrating another comprehensive landslide susceptibility assessment method provided in an embodiment of this application; Figure 3A detailed flowchart illustrating the construction and training of the RFE-RF-XGBoost model according to an embodiment of this application; Figure 4 The annual average deformation rate map of Yongshan County obtained by SBAS-InSAR technology processing is provided in an embodiment of this application. Figure 5 An RFE-screened ranking chart of evaluation factors (based on SHAP values) provided for an embodiment of this application. Figure 6 A comparison chart of ROC curves for different machine learning models provided in an embodiment of this application. Detailed Implementation
[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0014] To make the objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0015] Before describing this embodiment, the following key terms will be explained: Small Baseline Subset Interferometric Synthetic Aperture Radar (SBAS-InSAR) interferometric technology is used to acquire high-precision, large-scale temporal deformation information of the Earth's surface.
[0016] Recursive Feature Elimination (RFE) is a method for selecting the optimal feature subset by recursively reducing the number of features.
[0017] Random Forest (RF): An ensemble learning algorithm based on decision trees.
[0018] Extreme Gradient Boosting (XGBoost): An efficient gradient boosting decision tree algorithm.
[0019] Shapley Additive Explanations (SHAP): A method for interpreting the output of machine learning models based on game theory.
[0020] Area Under the Curve (AUC): Used to comprehensively evaluate the performance of classification models; the closer the value is to 1, the better the performance.
[0021] like Figure 1 As shown in the embodiment of this application, a comprehensive evaluation method for landslide susceptibility is provided, including: S1: Using the SBAS-InSAR deformation rate as a dynamic deformation factor and combining it with static environmental factors, the original feature dataset is constructed.
[0022] S2: A recursive feature elimination algorithm is used to sort and iteratively eliminate the evaluation factors in the original feature dataset to determine the optimal feature subset; the evaluation factors include dynamic deformation factors and static environmental factors.
[0023] S3: Train the RFE-RF-XGBoost cascaded ensemble model based on the optimal feature subset.
[0024] S4: Input the optimal feature subset corresponding to all grid cells in the entire study area into the trained RFE-RF-XGBoost cascaded ensemble model to determine the spatial classification map of landslide susceptibility.
[0025] S5: The SBAS-InSAR deformation rate is used to verify and locally correct the different risk level areas divided in the landslide susceptibility spatial classification map to determine the landslide susceptibility assessment results.
[0026] This application introduces SBAS-InSAR deformation rate as a dynamic deformation factor and constructs an innovative RFE-RF-XGBoost cascaded ensemble machine learning model to achieve full-process optimization from data fusion and feature optimization to model prediction, thereby significantly improving the accuracy, reliability and timeliness of landslide susceptibility assessment.
[0027] In an exemplary embodiment, S1 specifically includes: S11: Based on SBAS-InSAR technology, a time baseline threshold and a spatial vertical baseline threshold are set, phase unwrapping is performed using the minimum cost flow method, and the terrain phase is removed using an external digital elevation model to generate the annual average surface deformation rate of the study area along the radar line of sight; the annual average surface deformation rate is the SBAS-InSAR deformation rate.
[0028] In practical applications, the acquisition and preprocessing of static environmental factors specifically includes: acquiring data on various static environmental factors in the study area, such as topography, geological conditions, hydrology and meteorology, human activities, and vegetation ecology; performing unified processing on spatial resolution, coordinate system, and raster size to form a standardized static factor layer set. Therefore, static environmental factors include at least: elevation, slope, aspect, stratigraphic lithology, distance from faults, annual average rainfall, normalized difference vegetation index (NDVI), distance from roads, distance from rivers, topographic humidity index, and land use type.
[0029] The process of constructing the original feature dataset includes: determining the location of positive samples based on the historical landslide inventory and landslide points fused with optical remote sensing images and InSAR deformation information interpretation; using the spatial constraint buffering method, generating non-landslide points as negative samples in areas far from known landslides; and deriving the dynamic deformation factor values and static environmental factor values corresponding to all sample points to construct the original feature dataset.
[0030] In an exemplary embodiment, S2 specifically includes: S21: Using a random forest model as the basic evaluator, calculate the importance scores of the evaluation factors in the original feature dataset; S22: A recursive feature elimination algorithm is used to iteratively remove features corresponding to importance scores below a set score threshold. After each removal, the random forest model is retrained, and the performance metrics of the random forest model on the validation set are recorded. The features are evaluation factors. The performance metrics include accuracy, precision, recall, and F1 score. S23: Compare the performance indicators under different numbers of features, select the feature combination that achieves the optimal performance indicator, and take the feature combination as the optimal feature subset.
[0031] In practical applications, this application adopts the following core metrics: Area under the receiver operating characteristic curve (AUC), used to comprehensively evaluate the classification performance of the model; Accuracy, reflecting the overall prediction accuracy of the model; Precision, measuring the proportion of samples predicted as positive that are actually positive; Recall, measuring the proportion of samples that are actually positive that are correctly predicted; and F1 score, the harmonic mean of precision and recall, used to comprehensively evaluate the model's performance on imbalanced data.
[0032] Replace all features in the original feature dataset with the feature data corresponding to the optimal feature subset to form a training dataset with lower dimensionality and less redundancy.
[0033] In an exemplary embodiment, the RFE-RF-XGBoost model construction and training process is as follows: Figure 2As shown, S3 specifically includes: S31: Based on the optimal feature subset, perform model fitting using the random forest algorithm to generate a random forest model.
[0034] S32: Determine the feature input and initial learning weights of the XGBoost model based on the training results of the random forest model; the training results include feature importance and data distribution learning.
[0035] S33: Based on the feature input and the initial learning weights, predict the XGBoost model as the core, and optimize the hyperparameters using a grid search method to determine the trained RFE-RF-XGBoost cascade ensemble model; the hyperparameters include learning rate, maximum depth, subsampling rate and number of weak learners.
[0036] In practical applications, the optimal feature subset is used to fit the model using the random forest algorithm, generating a model for evaluating feature stability and providing a preliminary classification benchmark.
[0037] The training results of the random forest model (feature importance, data distribution learning) provide higher quality feature inputs and initial learning weights for the subsequent XGBoost model. The XGBoost model performs more refined nonlinear fitting and prediction optimization based on the initial learning of the random forest.
[0038] Using the same training set, with the XGBoost model as the core predictor, a grid search method was employed to optimize key hyperparameters such as learning rate, maximum depth, and subsampling rate. An early stopping strategy was also applied to prevent overfitting, thus completing model training. Specifically, the optimization process involved iterating through a pre-defined range of hyperparameter values, evaluating the model performance for each set of parameters through cross-validation, and selecting the optimal parameter combination.
[0039] The hyperparameter optimization range for the XGBoost model includes: learning rate in the range of [0.01, 0.2], maximum depth in the range of [3, 9], subsampling rate in the range of [0.8, 1.0], and the number of weak learners in the range of [100, 300] for grid search; the early stopping strategy is set to automatically terminate training when the validation set loss function does not decrease for 10 consecutive iterations.
[0040] In an exemplary embodiment, S4 specifically includes: S41: Input the optimal feature subset corresponding to all grid cells in the entire study area into the trained RFE-RF-XGBoost cascade ensemble model to determine the landslide probability value of each grid cell; S42: The landslide occurrence probability value is divided into multiple susceptibility levels using the natural breakpoint method to determine the spatial classification map of landslide susceptibility.
[0041] In an exemplary embodiment, S5 specifically involves result verification and dynamic correction: the generated landslide susceptibility classification map is spatially overlaid and compared with the deformation rate obtained by SBAS-InSAR technology, and the susceptibility zoning results are verified and locally corrected using the deformation rate; and high-susceptibility areas are confirmed by combining evidence from field investigations; wherein, the susceptibility zoning results are the different risk level areas divided in the landslide susceptibility spatial classification map.
[0042] In one exemplary embodiment, S5, further comprising: S6: The SHAP model is used to interpret the trained RFE-RF-XGBoost cascade ensemble model, calculate the contribution of each evaluation factor to the spatial classification map of landslide susceptibility, and generate a factor contribution summary map and a dependency map.
[0043] In practical applications, this application also includes: model interpretability analysis and factor contribution visualization. Specifically, the SHAP model is used to interpret the trained RFE-RF-XGBoost model, the contribution of each evaluation factor to the final prediction result is calculated, and a factor contribution summary plot and a dependency plot are generated. The final prediction result includes the landslide occurrence probability values for each grid cell output by the model and the resulting spatial classification map of landslide susceptibility.
[0044] In practical applications, this application also includes: dynamic model updates and maintenance. Specifically, new landslide occurrence records, optical remote sensing images, and time-series InSAR data are acquired periodically, and S1 to S2 are repeated to supplement and update the sample dataset. The deployed RFE-RF-XGBoost model is then incrementally trained or periodically retrained using the new dataset to maintain the model's adaptability to environmental changes and landslide dynamics.
[0045] like Figure 3 As shown in the figure, this embodiment provides another comprehensive evaluation method for landslide susceptibility. This method integrates dynamic deformation factors and static environmental factors, and its implementation steps are as follows: Step 101: Obtain the dynamic deformation factor.
[0046] Sentinel-1A ascending-orbit SAR imagery of the study area (County A) from May 5, 2022 to March 26, 2025, totaling 147 scenes, was acquired. Gamma remote sensing processing software was used for processing with SBAS-InSAR technology. Temporal baseline thresholds were set to 12-96 days, and spatial vertical baseline thresholds to 0.4-309.2 meters. Phase unwrapping was performed using the minimum cost flow method, and topographic phase was removed using an external 30-meter resolution Digital Elevation Model (DEM). Finally, the annual average surface deformation rate field along the radar line of sight for the study area was obtained, as shown in the figure. Figure 4 As shown. The deformation rate ranges from -118.6 mm / year (far from the satellite) to 84.8 mm / year (near the satellite), and this deformation rate raster layer is used as the core dynamic deformation factor.
[0047] Step 102: Acquisition and preprocessing of static environmental factors.
[0048] The system collected data on 12 static environmental factors in County A, including: elevation, slope, aspect, stratigraphic lithology, distance from faults, annual average rainfall, normalized difference vegetation index (NDVI), distance from roads, distance from rivers, topographic humidity index, land use type, and peak ground acceleration (PGA). All data were standardized to the WGS_1984_UTM Zone_48N coordinate system and resampled to a 30m×30m raster size in ArcGIS Pro software, forming a spatially registered static factor layer set. ArcGIS Pro is a next-generation desktop Geographic Information System (GIS) software.
[0049] Step 103: Sample dataset construction.
[0050] By integrating 455 historical landslide logging points in County A and combining the InSAR deformation rate map obtained in step S1 with Gaofen-2 optical remote sensing imagery, 73 new landslide or potential landslide areas were identified. These 528 points were used as positive samples. A 1-kilometer radius buffer was established centered on each positive sample, and 528 non-landslide points were randomly generated outside the buffer as negative samples. Using ArcGIS's "Multivalue Extraction to Point" tool, the 13 factor values (12 static + 1 dynamic) corresponding to all 1056 sample points were exported to construct the original feature dataset.
[0051] Step 104: Feature selection and dataset optimization.
[0052] The Recursive Feature Elimination (RFE) algorithm, using a Random Forest (RF) model as the evaluator, is employed to rank and iteratively filter the evaluation factors in the original dataset based on their importance. This process can be formally represented as: in: : Represents the original dataset, which is a data matrix consisting of all sample points (including positive and negative samples) and their corresponding 13 factor values (12 static and 1 dynamic). : This represents the optimal feature subset obtained by the recursive feature elimination algorithm, that is, the combination of features that is most important for landslide susceptibility prediction after optimization and retention; : This indicates the recursive feature elimination process using a random forest as the evaluator.
[0053] By removing low-importance features through multiple rounds of iteration, the optimal subset is finally determined based on model performance, and an optimized training dataset is constructed.
[0054] The model's performance was evaluated using 5-fold cross-validation with varying numbers of features (AUC as the metric). The optimal feature subset was determined by retaining 12 factors (excluding "topographic relief"). SHAP analysis was then used to analyze this subset, and the factor importance ranking is as follows: Figure 5 As shown, elevation, stratigraphy, and NDVI rank in the top three.
[0055] Step 105: Ensemble Model Training. First, train a baseline random forest model based on the optimal feature subset. Then, construct an XGBoost model with the objective function: , in: The XGBoost model in the first The overall objective function value in each iteration is used to train the model by optimizing this value; This represents the set of all parameters to be optimized in the current iteration; : indicates that for all training samples Summation; Loss function: used to measure the model's predicted values. Compared with the true label of the sample The difference between them is that in the binary classification task of landslide susceptibility, log loss or cross-entropy loss is usually used; : indicates that for all weak learners (i.e., decision trees) Summation; : Regularization term, used to control the first... The complexity of the tree structure helps prevent overfitting of the model. (The specific expression for the regularization term); : Represents a decision tree (weak learner); : Complexity control parameter (gamma), used to penalize the number of leaf nodes in the tree; The total number of leaf nodes in the decision tree; L2 regularization weights (lambda) are used to penalize the sum of squares of the weights of leaf nodes to prevent the weights from becoming too large. : Represents the weight vector of the leaf node The square of the L2 norm (i.e., the sum of the squares of all weights). It is an L2 regularization term; : A vector consisting of the weights (or output values) of all leaf nodes in the decision tree.
[0056] Other symbol explanations: : No. The true label for each sample (1 for landslides, 0 for non-landslides).
[0057] Model for the first The predicted probability value of each sample.
[0058] : No. A decision tree (weak learner).
[0059] The set of parameters to be optimized during model training includes the tree structure (split features, thresholds) and leaf node weights.
[0060] During training, grid search was used to optimize hyperparameters, and an early stopping strategy was employed to prevent overfitting. The hyperparameter tuning ranges are shown in Table 1.
[0061] Table 1. XGBoost Hyperparameter Tuning Range
[0062] The final optimal parameters were determined to be: learning_rate=0.1, max_depth=7, subsample=0.9, colsample_bytree=0.9, n_estimators=200. The final XGBoost model was trained using these parameters.
[0063] Step 106: Landslide susceptibility assessment and mapping.
[0064] The optimal feature subset data of 12 features from a total of 3,154,444 grid cells in County A are input into a pre-trained RFE-RF-XGBoost model to predict the probability of landslide occurrence for each grid cell. The probability value is calculated using the following formula: in, For logical functions, The total number of trees, This is the optimal feature subset of data; Indicates that given a feature vector Under these conditions, the grid cell experienced a landslide. The probability of ). It is a logical function (Sigmoid function), and its expression is: , is used to map the weighted sum of the model output to the [0,1] interval, representing the probability; This represents the total number of weak learners (decision trees) in the XGBoost model. For the first The prediction output function of a decision tree; Represents the optimal feature subset of the input model. The corresponding feature data vector; Indicates all The outputs of each tree are summed to obtain the model's final prediction score for that sample.
[0065] Subsequently, the probability values were divided into 5 levels using the natural breakpoint method: extremely low, low, medium, high, and extremely high, generating a spatial classification map of landslide susceptibility.
[0066] Step 107: Result verification, analysis and dynamic correction.
[0067] Multi-dimensional validation and closed-loop analysis were performed on the model results: Model performance validation: Evaluated on 30% of the independent test set, the RFE-RF-XGBoost model of this application achieved an AUC of 0.9495, with accuracy, precision, recall, and F1 score of 0.8096, 0.8014, 0.8204, and 0.8049, respectively. Figure 6 The comparison of Receiver Operating Characteristic (ROC) curves for different machine learning models (including RF, XGBoost, CatBoost, LightGBM, and RFE-RF-XGBoost) shows that its performance is significantly better than that of single Random Forest, XGBoost, Categorical Boosting (CatBoost), and Light Gradient Boosting Machine (LightGBM) models.
[0068] Spatial consistency verification: Comparing the susceptibility grading map with the SBAS-InSAR deformation rate map (i.e. Figure 4Spatial overlay analysis was performed. The results showed that approximately 85% of the "high" and "extremely high" susceptibility zones and the significant deformation zones with an absolute deformation rate greater than 40 mm / year were spatially highly correlated, demonstrating the effectiveness of dynamic factor fusion.
[0069] Field verification: Four high-risk locations predicted by the model, namely Yinggepingzi and Luoqiu Village, were selected for field verification. The field investigation revealed varying degrees of instability signs, such as landslide backwalls, cracks, or building deformation, confirming the reliability of the model predictions.
[0070] Dynamic update mechanism: To maintain the model's timeliness, an update process is established. Every six months or one year, new Sentinel-1 SAR images are collected, steps S1-S3 are repeated, newly added landslide points are interpreted, and the sample dataset is updated. The deployed RFE-RF-XGBoost model is incrementally trained or periodically retrained using the new dataset, enabling the evaluation system to adapt to environmental changes and the dynamic evolution of landslides.
[0071] Step 108: Model interpretability analysis and factor contribution visualization.
[0072] After completing model training and validation, interpretability analysis was implemented to enhance the transparency and decision credibility of the model. The SHAP model was used to interpret the trained RFE-RF-XGBoost model. The specific process is as follows: The SHAP value of all samples in the test set was calculated using the shap library of Python. The SHAP value quantitatively characterizes the positive or negative contribution of each evaluation factor (such as elevation, lithology, deformation rate, etc.) to a single prediction result (probability of landslide occurrence). Based on this, two key visualizations were generated: (1) Factor contribution summary diagram: such as Figure 5 As shown, it displays the importance ranking of all factors based on the global average absolute SHAP value, intuitively revealing that elevation, stratigraphic lithology, and normalized difference vegetation index (NDVI) are the core driving factors; (2) Factor dependency diagram: further analyze the nonlinear relationship between key factors (such as deformation rate) and prediction probability, and reveal its influence mechanism. This step transforms the "black box" model decision-making process into an understandable factor contribution map, providing geological disaster experts with reliable decision-making basis and mechanism explanation.
[0073] Step 109: Dynamic updating and maintenance of the model.
[0074] To maintain the timeliness of the evaluation system in this application and to address environmental changes and new landslide events, a dynamic updating and maintenance mechanism for the model is established. The process is as follows: Data Update: Collect new cycles of Sentinel-1 SAR imagery, optical remote sensing imagery, and the latest landslide logging data.
[0075] Sample set update: Repeat steps 101 to 103, interpret the new landslide points using the new InSAR deformation data and remote sensing images, and generate new non-landslide points accordingly to supplement and update the original sample dataset.
[0076] Model iterative training: Using the updated sample dataset, the deployed RFE-RF-XGBoost model is incrementally trained (Fine-tuning) or periodically retrained. Incremental training can fine-tune the original model parameters, which is more efficient; full retraining starts from scratch and is better able to adapt to significant changes in the data.
[0077] Update Validation and Deployment: The updated model undergoes performance validation (as in step S7). After confirming that its accuracy has not decreased or has even improved, it is deployed to the application system, replacing the old model. This closed-loop update mechanism ensures that the landslide susceptibility assessment results continuously reflect the latest geological dynamics and risk status of the region.
[0078] In summary, this embodiment successfully integrates SBAS-InSAR dynamic deformation data with static environmental factors. Through the innovative RFE-RF-XGBoost cascaded ensemble model, it achieves a high-precision and dynamic assessment of landslide susceptibility in County A, forming a complete technical closed loop from data preparation, model training, prediction mapping, multi-source verification, mechanism interpretation to dynamic updates, significantly improving the accuracy and practicality of geological hazard risk assessment.
[0079] This application can achieve the following effects: 1. Achieved deep integration of "dynamic and static": For the first time, the temporal surface deformation rate obtained by SBAS-InSAR was used as the core dynamic deformation factor and fused with multi-source static environmental factors at the feature level. This enabled the model to not only learn the static environmental background of landslide occurrence, but also to capture its dynamic evolution precursors, promoting a substantial shift in the evaluation paradigm from "static probability" to "dynamic risk".
[0080] 2. An innovative cascaded ensemble learning framework is proposed: the constructed RFE-RF-XGBoost model combines the advantages of feature selection and high-precision prediction. The front-end RFE-RF module effectively eliminates redundant noise factors through data-driven recursive screening, improving the model's interpretability and generalization ability; the back-end XGBoost module achieves high-precision fitting of complex nonlinear relationships through gradient boosting and regularization techniques. This architecture overcomes the limitations of single models, achieving an AUC value of over 0.95 in the publicly available case (County A, Yunnan Province), significantly outperforming the comparison models.
[0081] 3. A complete closed-loop process of "evaluation-verification-update" was established: not only were susceptibility grading maps generated, but emphasis was also placed on using InSAR deformation itself for spatial consistency verification and result correction, combined with field surveys for confirmation. Simultaneously, a periodic model update mechanism was designed, iteratively optimizing the model by supplementing it with new samples and data, ensuring the long-term effectiveness and adaptability of the evaluation system.
[0082] 4. Enhanced engineering practicality and interpretability: By analyzing the model's decisions through interpretable artificial intelligence technologies such as SHAP, the contribution of each factor is quantified, making the "black box" model transparent and providing geological disaster prevention and control experts with intuitive and reliable decision-making basis, which facilitates the formulation of targeted prevention and control measures.
[0083] This application provides a comprehensive landslide susceptibility assessment system, including: The original feature dataset construction module is used to construct the original feature dataset by taking the SBAS-InSAR deformation rate as a dynamic deformation factor and combining it with static environmental factors.
[0084] The optimal feature subset determination module is used to use a recursive feature elimination algorithm to sort and iteratively eliminate evaluation factors in the original feature dataset to determine the optimal feature subset; the evaluation factors include dynamic deformation factors and static environmental factors.
[0085] The model training module is used to train the RFE-RF-XGBoost cascaded ensemble model based on the optimal feature subset.
[0086] The landslide susceptibility spatial classification map determination module is used to input the optimal feature subset corresponding to all grid cells in the entire study area into the trained RFE-RF-XGBoost cascaded ensemble model to determine the landslide susceptibility spatial classification map.
[0087] The landslide susceptibility assessment result determination module is used to verify and locally correct the different risk level areas divided in the landslide susceptibility spatial classification map using the SBAS-InSAR deformation rate, and determine the landslide susceptibility assessment result.
[0088] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments. The computer device can be a server or a terminal. The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, memory, and I / O interface are connected via a system bus, and the communication interface is connected to the system bus via the I / O interface. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device stores data to be processed. The I / O interface of the computer device is used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communicating with an external terminal via a network connection. When the computer program is executed by the processor, it implements the above-described methods.
[0089] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0090] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0091] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.
[0092] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by hardware related to computer program instructions. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).
[0093] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0094] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0095] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A comprehensive evaluation method for landslide susceptibility, characterized in that, include: The SBAS-InSAR deformation rate was used as a dynamic deformation factor, and combined with static environmental factors, the original feature dataset was constructed. A recursive feature elimination algorithm is used to rank and iteratively eliminate evaluation factors in the original feature dataset to determine the optimal feature subset; the evaluation factors include dynamic deformation factors and static environmental factors. Train the RFE-RF-XGBoost cascaded ensemble model based on the optimal feature subset; The optimal feature subset corresponding to all grid cells in the entire study area is input into the trained RFE-RF-XGBoost cascaded ensemble model to determine the spatial classification map of landslide susceptibility. The SBAS-InSAR deformation rate is used to verify and locally correct the different risk level areas divided in the landslide susceptibility spatial classification map, and the landslide susceptibility assessment results are determined.
2. The comprehensive evaluation method for landslide susceptibility according to claim 1, characterized in that, Using SBAS-InSAR deformation rate as a dynamic deformation factor specifically includes: Based on SBAS-InSAR technology, a temporal baseline threshold and a spatial vertical baseline threshold are set, phase unwrapping is performed using the minimum cost flow method, and the terrain phase is removed using an external digital elevation model to generate the annual average surface deformation rate of the study area along the radar line of sight; the annual average surface deformation rate is the SBAS-InSAR deformation rate.
3. The comprehensive evaluation method for landslide susceptibility according to claim 1, characterized in that, A recursive feature elimination algorithm is used to rank and iteratively eliminate evaluation factors in the original feature dataset to determine the optimal feature subset, specifically including: Using a random forest model as the basic evaluator, the importance scores of the evaluation factors in the original feature dataset are calculated. A recursive feature elimination algorithm is used to iteratively remove features corresponding to importance scores below a set score threshold. After each removal, the random forest model is retrained, and the performance metrics of the random forest model on the validation set are recorded. The features are evaluation factors. The performance metrics include accuracy, precision, recall, and F1 score. By comparing the performance metrics under different numbers of features, the feature combination that achieves the optimal performance metrics is selected, and this feature combination is taken as the optimal feature subset.
4. The comprehensive evaluation method for landslide susceptibility according to claim 1, characterized in that, Training the RFE-RF-XGBoost cascaded ensemble model based on the optimal feature subset specifically includes: Based on the optimal feature subset, a random forest algorithm is used to fit the model and generate a random forest model; The feature inputs and initial learning weights of the XGBoost model are determined based on the training results of the random forest model; the training results include feature importance and data distribution learning. Based on the feature input and the initial learning weights, the XGBoost model is used as the core for prediction. The hyperparameters are optimized using a grid search method to determine the trained RFE-RF-XGBoost cascade ensemble model. The hyperparameters include the learning rate, maximum depth, subsampling rate, and number of weak learners.
5. The comprehensive evaluation method for landslide susceptibility according to claim 1, characterized in that, The optimal feature subset corresponding to all grid cells in the entire study area is input into the trained RFE-RF-XGBoost cascaded ensemble model to determine the spatial classification map of landslide susceptibility, specifically including: The optimal feature subset corresponding to all grid cells in the entire study area is input into the trained RFE-RF-XGBoost cascade ensemble model to determine the landslide occurrence probability value of each grid cell; The landslide occurrence probability value was divided into multiple susceptibility levels using the natural discontinuity method, and a spatial classification map of landslide susceptibility was determined.
6. The comprehensive evaluation method for landslide susceptibility according to claim 1, characterized in that, The SBAS-InSAR deformation rate is used to verify and locally correct the different risk level areas divided in the landslide susceptibility spatial classification map to determine the landslide susceptibility assessment results. This process then includes: The SHAP model is used to interpret the trained RFE-RF-XGBoost cascade ensemble model, calculate the contribution of each evaluation factor to the spatial classification map of landslide susceptibility, and generate a factor contribution summary map and a dependency map.
7. A comprehensive evaluation system for landslide susceptibility, characterized in that, The landslide susceptibility comprehensive evaluation system executes the landslide susceptibility comprehensive evaluation method according to any one of claims 1-6, and the landslide susceptibility comprehensive evaluation system includes: The original feature dataset construction module is used to construct the original feature dataset by taking the SBAS-InSAR deformation rate as a dynamic deformation factor and combining it with static environmental factors. The optimal feature subset determination module is used to employ a recursive feature elimination algorithm to rank and iteratively eliminate evaluation factors in the original feature dataset to determine the optimal feature subset; the evaluation factors include dynamic deformation factors and static environmental factors; The model training module is used to train the RFE-RF-XGBoost cascaded ensemble model based on the optimal feature subset. The landslide susceptibility spatial classification map determination module is used to input the optimal feature subset corresponding to all grid cells in the entire study area into the trained RFE-RF-XGBoost cascade ensemble model to determine the landslide susceptibility spatial classification map. The landslide susceptibility assessment result determination module is used to verify and locally correct the different risk level areas divided in the landslide susceptibility spatial classification map using the SBAS-InSAR deformation rate, and determine the landslide susceptibility assessment result.
8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the landslide susceptibility comprehensive evaluation method according to any one of claims 1-6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the comprehensive evaluation method for landslide susceptibility as described in any one of claims 1-6.
10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the comprehensive evaluation method for landslide susceptibility as described in any one of claims 1-6.