Debris flow susceptibility evaluation method based on progressive optimization framework

Abstract

The invention discloses a debris flow susceptibility evaluation method based on a progressive optimization framework. The debris flow susceptibility evaluation method comprises the steps of collecting and sorting basic data of a research area; selecting evaluation factors, and performing correlation analysis on the evaluation factors; dividing drainage basin units, constructing a frequency ratio model to carry out susceptibility evaluation, and selecting an optimal drainage basin unit; adopting semi-supervised learning to select negative samples, and constructing a machine learning data set; through comparison of different Boosting algorithms and hyper-parameter optimization algorithms, an evaluation model with the highest precision under a village and town scale is constructed through multiple evaluation indexes, and an SHAP technology is adopted, so that contribution of different evaluation factors to an evaluation model decision process is quantitatively analyzed. According to the method, accurate and explainable theoretical support can be provided for risk assessment of the geological disasters in the village and town areas through a progressive optimization framework method, and important practical reference can be provided for optimization of disaster management and prevention and control strategies, so that potential risks and economic losses are effectively reduced.

Description

Technical Field

[0001] The present invention relates to a debris flow susceptibility assessment method, which is a debris flow susceptibility assessment method based on a progressive optimization framework, is particularly suitable for township-scale geological disaster risk assessment and disaster prevention and control strategy optimization, and belongs to the technical field of geological disaster prediction and risk assessment. Background Art

[0002] Debris flow is a highly destructive geological disaster, usually triggered by extreme weather conditions such as heavy rain or snowmelt. It is sudden and highly harmful, seriously threatening the safety of life and property of mountain residents and the stability of the ecological environment. In recent years, with the intensification of climate change and the frequent human activities, the frequency and intensity of debris flows have shown an upward trend, especially in townships and mountainous areas, and the need for disaster risk assessment and management is becoming increasingly urgent. However, there are still many technical bottlenecks in the current assessment of debris flow susceptibility at the township scale.

[0003] In the existing technology, the methods for evaluating the susceptibility of debris flow are mostly based on the construction of statistical models based on factors such as topography, hydrogeology and engineering geology, such as FR, logistic regression and weighted regression. However, there are problems in the existing technology: first, many methods in the existing technology ignore the accurate selection of negative samples, resulting in the model's performance in areas where disasters do not occur being not stable enough, and the prediction results have large deviations; second, there is no clear comparative explanation for the selection of watershed units, and there is a lack of rigorous introduction.

[0004] In recent years, machine learning technology has been widely used in geological disaster risk assessment, especially the Boosting algorithm has attracted attention due to its powerful learning ability and excellent generalization performance. However, in the evaluation of debris flow susceptibility at the township scale, how to use optimization algorithms to further improve the model accuracy and generalization ability is still an urgent problem to be solved; at the same time, although complex machine learning models have high prediction accuracy, their black box characteristics make the decision-making process of the model difficult to explain, which limits its application in actual disaster management; in addition, the current debris flow susceptibility evaluation method lacks quantitative analysis of the contribution of different factors, and cannot provide a clear scientific basis for disaster prevention and control strategies.

[0005] Therefore, there is an urgent need for a technical framework that can integrate multi-source data, improve model transparency, and balance accuracy and interpretability to better serve the geological hazard risk assessment and management needs at the township level. Summary of the invention

[0006] The purpose of this invention is to solve the many technical bottlenecks in carrying out debris flow susceptibility assessment at the township scale, and proposes a debris flow susceptibility assessment method based on a progressive optimization framework. It aims to construct a high-precision and highly interpretable debris flow susceptibility assessment method through a progressive optimization framework combined with advanced machine learning technology and factor contribution analysis, so as to provide theoretical support and practical guidance for disaster prediction and management at the township scale.

[0007] The present invention achieves the above-mentioned purpose through the following technical scheme: a debris flow susceptibility evaluation method based on a progressive optimization framework, the debris flow susceptibility evaluation method comprising the following steps:

[0008] S100, collect and organize basic data of the study area;

[0009] S200, based on the collected and collated basic data, select evaluation factors and conduct correlation analysis on the evaluation factors;

[0010] S300, divide the watershed units based on ArcGIS, construct the frequency ratio (FR) model for susceptibility evaluation, and select the optimal watershed units;

[0011] S400, the optimal watershed unit selected based on the frequency ratio (FR) model, semi-supervised learning is used to select negative samples and construct a machine learning data set, where the training set: test set = 0.8:0.2;

[0012] S500, by comparing different Boosting algorithms and hyperparameter optimization algorithms, a variety of evaluation indicators are used to construct the most accurate evaluation model at the township scale.

[0013] S600 uses SHAP (SHapley Additive exPlanations) technology to quantitatively analyze the contribution of different evaluation factors to the decision-making process of the evaluation model, thereby enhancing the transparency and scientific nature of the evaluation model.

[0014] As a further technical solution of the present invention: in S100, the basic data includes regional geological data, field survey data, meteorological and hydrological data, remote sensing image data and related literature data.

[0015] As a further technical solution of the present invention: in S200, the evaluation factors include slope, aspect, curvature, elevation variation coefficient, terrain undulation, surface cutting depth, terrain undulation, stream intensity index (SPI), terrain wetness index (TWI), distance to road, distance to river, normalized difference vegetation index (NDVI), soil type, land use and lithology.

[0016] As a further technical solution of the present invention: In S200, the correlation between the evaluation factors is checked using the tolerance (TOL) and the variance inflation factor (VIF), and the calculation formulas are respectively:

[0017]

[0018] Where: TOL is the tolerance, VIF is the variance inflation factor, R J is the independent variable x i Multiple correlation coefficients of regression analysis for the remaining independent variables.

[0019] As a further technical solution of the present invention: In S300, dividing the watershed unit based on ArcGIS refers to dividing through basic DEM data, filling depressions with elevation data, calculating flow direction, cumulative watershed volume, setting thresholds, river links, generating watersheds and converting raster to surface.

[0020] As a further technical solution of the present invention: In S300, the susceptibility evaluation is performed based on the frequency ratio model, and the optimal watershed unit is selected, which includes:

[0021] Evaluate based on the area under the ROC curve (AUC);

[0022] The percentage of units spanned by disasters (PUC) is introduced, and the AUC-PUC scatter plot is constructed for comprehensive evaluation;

[0023] The area under the ROC curve is calculated by plotting the true positive rate against the false positive rate with sensitivity (true positive rate) as the X-axis and 1-specificity (false positive rate) as the Y-axis to show the performance of the frequency ratio model at different thresholds. The calculation formula is as follows:

[0024]

[0025] Where: TP represents true positive, TN represents true negative, FP represents false positive, and FN represents false negative;

[0026] PUC refers to the proportion of units spanned by the disaster, and its calculation formula is:

[0027]

[0028] Where: N is the number of units spanned by the disaster, and M is the total number of divisions in the study area.

[0029] As a further technical solution of the present invention: in S400, the negative samples are selected by sampling based on semi-supervised learning constructed based on a frequency ratio (FR) model, and sampling is performed in extremely low and low susceptibility intervals, so as to construct a high-quality negative sample set.

[0030] As a further technical solution of the present invention: in S500, constructing the evaluation model with the highest accuracy at the township scale includes the following steps:

[0031] S501. Based on the optimal watershed unit and high-quality negative sample set, four Boosting algorithms, XGBoost, LightGBM, CatBoost and NGBoost, are selected to build a high-precision evaluation model;

[0032] S502. Efficient tuning of hyperparameter optimization algorithms is achieved by introducing Bayesian optimization (BO), particle swarm optimization (PSO) and genetic algorithm (GA);

[0033] S503, using 10-fold cross-validation (CV) to ensure that the evaluation model obtains better average performance and generalization ability;

[0034] S504, conduct comprehensive evaluation and selection through ACC (accuracy), F1-score and AUC;

[0035] The calculation formula of ACC is:

[0036]

[0037] The calculation formula of F1-score is:

[0038]

[0039] S505. After comprehensive comparison, select the evaluation model with the highest accuracy at the township scale.

[0040] As a further technical solution of the present invention: in S501, XGBoost selects max_depth, learning_rate, n_estimators, Subsample, reg_lambda, gamma, min_child_weight and colsample_bytree;

[0041] LGBM selects max_depth, learning_rate, n_estimators, subsample, reg_lambda, min_child_samples, colsample_bytree and reg_alpha;

[0042] CatBoost selects depth, learning_rate, iterations, od_wait, l2_leaf_reg, bagging_temperature and colsample_bylevel;

[0043] NGBoost chooses n_estimators, learning_rate, minibatch_frac, and natural_gradient.

[0044] As a further technical solution of the present invention: in S600, when the SHAP technology is used for analysis, a scatter plot is selected to analyze the overall situation of each evaluation factor on the occurrence of debris flow through the SHAP value; a factor importance diagram is selected to rank the importance of each evaluation factor; a single factor dependency diagram is selected to further explore the specific impact mechanism of the influencing factors on the occurrence of debris flow through the SHAP value.

[0045] The beneficial effects of the present invention are:

[0046] 1) The method of the progressive optimization framework of the present invention can not only provide accurate and explainable theoretical support for the risk assessment of geological disasters in township areas, but also provide important practical references for the optimization of disaster management and prevention and control strategies, thereby effectively reducing potential risks and economic losses;

[0047] 2) The present invention can integrate multi-source data such as topography, hydrogeology and engineering geology, select the optimal watershed unit through the FR model, build semi-supervised learning for negative sample collection, and use advanced Boosting algorithms and BO, PSO and GA hyperparameter optimization techniques to construct a high-precision risk assessment model at the township scale, and use SHAP technology to quantify and analyze the contribution of evaluation factors, enhance the transparency and scientificity of the model, and thus provide a scientific basis and practical guidance for disaster risk assessment, management and prevention and control strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0049] Figure 2 This is a schematic diagram of the terrain elevation distribution of the study area in Example 1;

[0050] Figure 3 This is a schematic diagram of the process of dividing watershed units based on ArcGIS in Example 3;

[0051] Figure 4 It is the AUC-PUC scatter plot of the five threshold watershed units in Example 3;

[0052] Figure 5 It is a schematic diagram of debris flow susceptibility based on GA-CatBoost in Example 4;

[0053] Figure 6 It is a scatter point SHAP diagram of each evaluation factor in Example 5;

[0054] Figure 7 This is a schematic diagram of the importance of evaluation factors in Example 5;

[0055] Figure 8 This is a schematic diagram of the lithology, soil type and NDVI dependence in Example 5. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] Embodiment 1, as Figure 1 As shown, this embodiment provides a debris flow susceptibility evaluation method based on a progressive optimization framework, and the debris flow susceptibility evaluation method specifically includes the following steps:

[0058] S100. Collect and organize basic data of the study area.

[0059] The collection and organization of basic data includes regional geological data, field survey data, meteorological and hydrological data, remote sensing image data and relevant literature data collection; such as Figure 2 shown.

[0060] S200. Based on the collected and organized basic data, select evaluation factors from the aspects of topography, hydrogeology and engineering geology, and conduct correlation analysis on the evaluation factors.

[0061] S300, based on ArcGIS, divides the watershed units for the DEM (digital elevation model of the study area), constructs a frequency ratio (FR) model for susceptibility evaluation, and selects the optimal watershed unit for subsequent work.

[0062] S400. Based on the optimal watershed unit selected by the FR model, semi-supervised learning is used to select negative samples and construct a machine learning data set, in which the training set: test set = 0.8:0.2.

[0063] Among them, the selection of negative samples is based on sampling of semi-supervised learning constructed based on the frequency ratio model, and sampling is performed in the extremely low and low susceptibility intervals to construct a high-quality negative sample set.

[0064] S500, by comparing different Boosting algorithms and hyperparameter optimization algorithms, a variety of evaluation indicators are used to construct the most accurate evaluation model at the township scale.

[0065] S600 uses SHAP (SHapley Additive exPlanations) technology to quantitatively analyze the contribution of different evaluation factors to the decision-making process of the evaluation model, thereby enhancing the transparency and scientific nature of the evaluation model.

[0066] Embodiment 2: In S200, through the collection and organization of basic data, the selected evaluation factors include slope, aspect, curvature, elevation variation coefficient, terrain undulation, surface cutting depth, terrain roughness, stream intensity index (SPI), terrain wetness index (TWI), distance to road, distance to river, normalized difference vegetation index (NDVI), soil type, land use and lithology.

[0067] The correlation between the evaluation factors (influencing factors) was examined using tolerance (TOL) and variance inflation factor (VIF).

[0068] Table 1 lists the TOL and VIF values ​​of 15 influencing factors: the first statistical results show that the TOL of slope aspect, terrain roughness, surface cutting depth, terrain relief, SPI and TWI are less than 0.1 and the VIF is greater than 10, indicating that there is a certain degree of collinearity among these influencing factors, which may lead to overfitting problems; after excluding terrain roughness, surface cutting depth, terrain relief and TWI, the second statistical results show that the remaining 11 factors meet the requirements of TOL>0.1 and VIF<10 and can be used for subsequent research.

[0069] The calculation formulas are:

[0070]

[0071] Where: TOL is the tolerance, VIF is the variance inflation factor, R J is the independent variable x i Multiple correlation coefficients of regression analysis for the remaining independent variables.

[0072] Table 1 shows the multicollinearity analysis among the influencing factors.

[0073]

[0074] Embodiment 3: In S300, basin units are divided based on ArcGIS, mainly through basic DEM data, elevation data filling, flow direction calculation, cumulative watershed calculation, threshold setting, river linking, watershed generation and raster-to-surface division.

[0075] In this embodiment, a total of five watershed units with different thresholds are generated, and the thresholds are A1:500, A2:1000 (1k), A3:3000 (3k), A4:5000 (5k), and A5:10000 (10k). The specific extraction steps are as follows: Figure 3 shown.

[0076] Based on the FR model, the susceptibility of each threshold watershed unit is evaluated. In the embodiment, not only the area under the ROC curve (AUC) is used for evaluation, but also the proportion of units spanned by disasters (PUC) is introduced to construct an AUC-PUC scatter plot for comprehensive evaluation.

[0077] The area under the ROC curve (AUC) is a curve of true positive rate versus false positive rate with sensitivity (true positive rate) as the X-axis and 1-specificity (false positive rate) as the Y-axis to show the performance of the frequency ratio model at different thresholds. The area under the curve quantifies the discrimination ability of the model. The closer the AUC value is to 1, the better the performance of the frequency ratio model. The calculation formula is as follows:

[0078]

[0079] Where: TP represents true positive, TN represents true negative, FP represents false positive, and FN represents false negative.

[0080] PUC refers to the proportion of units spanned by disasters. Since debris flow surface files are used for susceptibility assessment, it is inevitable that disasters span multiple assessment units. Based on this, the study introduces PUC. The larger its proportion, the worse the division effect. Conversely, the smaller its proportion, the better the division effect. Its calculation formula is:

[0081]

[0082] Where: N is the number of units spanned by the disaster, and M is the total number of divisions in the study area.

[0083] like Figure 5 The figure shows the constructed AUC-PUC scatter plot. In order to quantitatively analyze the advantages and disadvantages of different threshold combinations, the AUC value and PUC value are mapped to a two-dimensional coordinate system, and the coordinate point (1,0) is set as the best combination point; then, the Euclidean distance from each combination point to (1,0) is calculated, and the combination with the shortest distance is selected as the optimal watershed unit. This method effectively converts the evaluation index into an intuitive spatial distance, which is convenient for comprehensive comparison and selection. Through calculation, it is found that the 1k threshold combination is closest to the point (1,0), so the 1k threshold is selected as the optimal watershed unit for subsequent evaluation.

[0084] In the fourth embodiment, in S500, constructing the evaluation model with the highest accuracy at the township scale mainly includes the following steps:

[0085] S501. Based on the optimal watershed unit and high-quality negative sample set, four advanced Boosting algorithms, XGBoost, LightGBM, CatBoost and NGBoost, are selected to build a high-precision evaluation model.

[0086] S502. Efficient tuning of the hyperparameter optimization algorithm is achieved by introducing Bayesian optimization (BO), particle swarm optimization (PSO) and genetic algorithm (GA).

[0087] For each hyperparameter optimization algorithm, the corresponding hyperparameters selected by Boosting are also different. For XGBoost, max_depth, learning_rate, n_estimators, Subsample, reg_lambda, gamma, min_child_weight and colsample_bytree are selected; for LGBM, max_depth, learning_rate, n_estimators, subsample, reg_lambda, min_child_samples, colsample_bytree and reg_alpha are selected; for CatBoost, depth, learning_rate, iterations, od_wait, l2_leaf_reg, bagging_temperature and colsample_bylevel are selected; for NGBoost, n_estimators, learning_rate, minibatch_frac and natural_gradient are selected. The search interval and the optimal hyperparameter values ​​are shown in Table 2.

[0088] Table 2 shows the search intervals and optimal hyperparameter values ​​of different Boosting algorithms

[0089]

[0090] S503, using 10-fold cross-validation (CV) to ensure that the evaluation model obtains better average performance and generalization ability;

[0091] S504, conduct comprehensive evaluation and selection through accuracy (ACC), F1-score and AUC;

[0092] The calculation formula of ACC is:

[0093]

[0094] The calculation formula of F1-score is:

[0095]

[0096] S505. After comprehensive comparison, select the evaluation model with the highest accuracy at the township scale.

[0097] Under the premise of ensuring the consistency of the search interval, for the same hyperparameter optimization algorithm, the optimal hyperparameter values ​​corresponding to different Boosting algorithms have great differences; in addition, for the same Boosting algorithm, the hyperparameter values ​​corresponding to different optimization algorithms also have certain differences. The evaluation indicators of each model are shown in Table 3.

[0098] Table 3 shows the evaluation indicators of different models

[0099]

[0100] The three hyperparameter optimization algorithms all have a certain degree of improvement on the Boosting algorithm. A comprehensive comparison shows that the GA hyperparameter optimization algorithm has the greatest improvement in the accuracy of the debris flow susceptibility mapping model, followed by PSO and BO. Among all the combined models, GA-CatBoost shows strong potential in debris flow susceptibility mapping, with the largest improvement in ACC, F1-score and AUC values, reaching 0.860, 0.880 and 0.910 respectively. GA-CatBoost was selected as the final debris flow susceptibility mapping model in the study area, and its susceptibility zoning is as follows: Figure 5 shown.

[0101] Embodiment 5: In S600, when the SHAP technology is used for analysis, a scatter plot is selected to analyze the overall situation of debris flow occurrence by each evaluation factor through the SHAP value, such as Figure 6 As shown in the figure, the importance of each evaluation factor is ranked by selecting the factor importance diagram, as shown in the figure. Figure 7 As shown in the figure, the single factor dependence diagram is used to further explore the specific impact mechanism of the most important influencing factors, namely lithology, soil type and NDVI on the occurrence of debris flow through the SHAP value, as shown in the figure. Figure 8 shown.

[0102] Among them, the size of the SHAP value has different meanings for the occurrence of disasters. The larger the SHAP value, the more favorable it is for disasters to occur; the smaller the SHAP value, the less favorable it is for disasters to occur, or it has a certain inhibitory effect.

[0103] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0104] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A debris flow susceptibility assessment method based on a progressive optimization framework, characterized in that: The debris flow susceptibility assessment method comprises the following steps: S100, collect and organize basic data of the study area; S200, based on the collected and sorted basic data, selecting evaluation factors, and performing correlation analysis on the evaluation factors; S300, divide the watershed units based on ArcGIS, build a frequency ratio model to evaluate the susceptibility, and select the optimal watershed unit; S400, the optimal watershed unit selected based on the frequency ratio model, semi-supervised learning is used to select negative samples, and a machine learning data set is constructed, where the training set: test set = 0.8:0.2; S500, by comparing different Boosting algorithms and hyperparameter optimization algorithms, a variety of evaluation indicators are used to construct the most accurate evaluation model at the township scale. S600 uses SHAP technology to quantitatively analyze the contribution of different evaluation factors to the decision-making process of the evaluation model, thereby enhancing the transparency and scientific nature of the evaluation model.

2. The debris flow susceptibility assessment method according to claim 1, characterized in that: In S100, the basic data include regional geological data, field survey data, meteorological and hydrological data, remote sensing image data and related literature data.

3. The debris flow susceptibility assessment method according to claim 1, characterized in that: In S200, the evaluation factors include slope, aspect, curvature, elevation variation coefficient, terrain undulation, surface cutting depth, terrain undulation, water flow intensity index, terrain moisture index, distance from road, distance from river, normalized difference vegetation index, soil type, land use and lithology.

4. The debris flow susceptibility assessment method according to claim 1, characterized in that: In S200, the correlation between the evaluation factors is checked using tolerance and variance inflation factor, and the calculation formulas are: Where: TOL is the tolerance, VIF is the variance inflation factor, R J is the independent variable x i Multiple correlation coefficients of regression analysis for the remaining independent variables.

5. The debris flow susceptibility assessment method according to claim 1, characterized in that: In the S300, dividing the watershed units based on ArcGIS refers to dividing through basic DEM data, filling depressions with elevation data, calculating flow direction, calculating cumulative watershed volume, setting thresholds, linking rivers, generating watersheds and converting raster to surface.

6. The debris flow susceptibility assessment method according to claim 1, characterized in that: In S300, the susceptibility evaluation is performed based on the frequency ratio model, and the optimal watershed unit is selected, which includes: Evaluate based on the area under the ROC curve; The proportion of units spanned by disasters is introduced, and the AUC-PUC scatter plot is constructed for comprehensive evaluation; The area under the ROC curve is based on the true positive rate as the X-axis and 1-false positive rate as the Y-axis. A curve of true positive rate versus false positive rate is plotted to show the performance of the frequency ratio model at different thresholds. The calculation formula is as follows: Where: TP represents true positive, TN represents true negative, FP represents false positive, and FN represents false negative; PUC refers to the proportion of units spanned by the disaster, and its calculation formula is: Where: N is the number of units spanned by the disaster, and M is the total number of divisions in the study area.

7. The debris flow susceptibility assessment method according to claim 1, characterized in that: In S400, the negative samples are selected by sampling based on the semi-supervised learning constructed based on the frequency ratio model, and the samples are taken in the extremely low and low susceptibility intervals, so as to construct a high-quality negative sample set.

8. The debris flow susceptibility assessment method according to claim 7, characterized in that: In S500, constructing the evaluation model with the highest accuracy at the township scale includes the following steps: S501. Based on the optimal watershed unit and high-quality negative sample set, four Boosting algorithms, XGBoost, LightGBM, CatBoost and NGBoost, are selected to build a high-precision evaluation model; S502. Efficient tuning of hyperparameter optimization algorithms is achieved by introducing Bayesian optimization, particle swarm optimization and genetic algorithms; S503, using 10-fold cross-validation to ensure that the evaluation model obtains better average performance and generalization ability; S504, conduct comprehensive evaluation and selection through ACC, F1-score and AUC; The calculation formula of ACC is: The calculation formula of F1-score is: S505. After comprehensive comparison, select the evaluation model with the highest accuracy at the township scale.

9. The debris flow susceptibility assessment method according to claim 8, characterized in that: In S501, the XGBoost selects max_depth, learning_rate, n_estimators, Subsample, reg_lambda, gamma, min_child_weight, and colsample_bytree; The LGBM selects max_depth, learning_rate, n_estimators, subsample, reg_lambda, min_child_samples, colsample_bytree, and reg_alpha; The CatBoost selects depth, learning_rate, iterations, od_wait, l2_leaf_reg, bagging_temperature, and colsample_bylevel; The NGBoost selects n_estimators, learning_rate, minibatch_frac, and natural_gradient.

10. The debris flow susceptibility assessment method according to claim 1, characterized in that: In S600, when the SHAP technology is used for analysis, a scatter plot is used to analyze the overall situation of each evaluation factor on the occurrence of debris flow through the SHAP value; a factor importance diagram is used to rank the importance of each evaluation factor; and a single factor dependency diagram is used to further explore the specific impact mechanism of the influencing factors on the occurrence of debris flow through the SHAP value.

Images