Rock-soil physical parameter calibration method under rainfall condition and rainfall landslide disaster risk prediction method

Abstract

The invention discloses a rock-soil physical parameter calibration method and a rainfall landslide disaster risk prediction method under the rainfall condition, and relates to the technical field of geological disaster risk assessment and early warning. The rock-soil physical parameter calibration method under the rainfall condition depends on a TRIGRS model to take rock-soil body cohesion and an internal friction angle as parameters to be calibrated; aiming at the problems of uncertainty and spatial heterogeneity of rock-soil body parameters, the rock-soil body parameters are divided into different engineering geological rock group areas according to rock-soil body characteristics in a research area, and underground water level dynamic response is used as rainfall process constraint; according to the method, the cohesion and internal friction angle parameters of different engineering geological rock composition subareas in the research area are subjected to inversion calibration by combining historical rainfall landslide distribution in the research area, so that the key physical and mechanical parameters of the rock-soil body can be determined more efficiently and accurately in a more time-saving and labor-saving manner; based on the TRIGRS model after parameter calibration, the spatial distribution characteristics of rainfall landslide disasters in a research area can be predicted more truly and accurately.

Description

Methods for calibrating geotechnical physical parameters under rainfall conditions and methods for predicting the risk of rainfall-induced landslide disasters. Technical Field

[0001] This invention relates to the field of geological disaster risk assessment and early warning technology, specifically to a method for calibrating geophysical parameters under rainfall conditions and a method for predicting the risk of rainfall-induced landslide disasters. Background Technology

[0002] Global warming is an undeniable fact. In its 2013 assessment report, *Climate Change 2013*, the IPCC (Intergovernmental Panel on Climate Change) clearly stated that many changes observed in the climate system since 1950 are unprecedented in recent decades or even the last century, with almost all regions of the globe experiencing warming. The frequency of extreme weather events has shown a significant upward trend, especially the frequent occurrence of extreme heavy rainfall, which greatly increases the risk of geological disasters such as landslides. Statistics show that 70% of landslides are triggered by rainfall, and as many as 95% of landslides occur during the rainy season. Rainfall-induced landslides are often characterized by clustering and regional occurrences; a single heavy rainfall event can trigger hundreds or even thousands of landslides. Therefore, strengthening risk assessment of regional rainfall-induced landslides is of great significance.

[0003] Currently, research methods for regional rainfall-induced landslide stability and risk assessment mainly fall into two categories: methods based on mathematical statistics and methods based on physical mechanics. The latter, typically based on limit equilibrium theory, combines geomechanical models with hydrological models. By analyzing the impact of rainfall infiltration on soil matrix suction, stress state, and deformation, it studies the spatiotemporal variations of slope stability under rainfall conditions. Because it considers the mechanical mechanisms of landslide occurrence and characterizes the intrinsic laws of rainfall-induced landslides, it represents the main direction of current regional rainfall-induced landslide research both domestically and internationally. Currently, the physical and mechanical models for regional rainfall-induced landslides proposed by scholars at home and abroad mainly include the SINMAP model, SHALTAB model, dSLAM model, Yeh-Lee model, SLIP model, Iverson model, Li Ning model, Xie-Esaki-Cai model, Rosso-Rulli-Vannucchi model, and TRIGRS model. Among them, the TRIGRS model can simulate the transient process of rainfall infiltration that varies with time, assess the impact of continuous rainfall on slope stability, and handle the spatial variability of topography, soil type, and rainfall intensity, thereby improving the accuracy of local landslide prediction. It has become an important model for current regional rainfall-induced landslide risk assessment and early warning.

[0004] For any of the aforementioned physical models, the selection of physical and mechanical parameters of the soil and rock mass has a significant impact on the final evaluation results. At the regional scale, soil and rock mass parameters exhibit uncertainty and spatial heterogeneity, making their selection a challenge faced by every physical model application. Currently, the determination of soil and rock mass parameters, both domestically and internationally, generally involves literature review, indoor and outdoor experiments, engineering geological analogies, and reference to global soil databases. However, these methods are generally highly subjective, time-consuming, and labor-intensive, and fail to yield truly objective soil and rock mass parameters. In practical applications of physical models, finding suitable parameters often requires extensive and repeated trial calculations to obtain a satisfactory parameter combination. Furthermore, most physical and mechanical evaluation models do not adequately consider the spatial heterogeneity of soil and rock mass parameters. Therefore, there is an urgent need to propose practical, time-efficient, and effective methods for determining key soil and rock mass parameters for landslide physical and mechanical models to further improve the evaluation accuracy and applicability of such models. Summary of the Invention

[0005] In view of the above-mentioned shortcomings in the prior art, the method for calibrating geotechnical physical parameters under rainfall conditions and the method for predicting the risk of landslide disasters caused by rainfall provided by the present invention solves the problem that the prior art is unable to efficiently obtain the real and objective key physical and mechanical parameters of regional geotechnical bodies.

[0006] To achieve the aforementioned objectives, the present invention provides a method for calibrating geotechnical physical parameters under rainfall conditions, comprising the following steps: using cohesion and internal friction angle among the geotechnical physical parameters as parameters to be calibrated, determining the initial range and interval of cohesion and internal friction angle; calculating the stability coefficients of each engineering geological rock group in the study area under different groundwater depth conditions and all parameter combinations using the TRIGRS model, and creating ROC curves for each parameter group under each condition using different stability coefficient thresholds based on historical landslide data; eliminating unreasonable parameter combinations by setting initial area ratio and true positive rate indicators under rainfall-triggered landslide conditions; evaluating the performance of the TRIGRS model under the remaining parameter combinations to obtain the optimal geotechnical physical parameters combination for the entire study area and different engineering geological rock group zones, thus obtaining the geotechnical physical parameters calibration results under rainfall conditions; wherein the true positive rate is the ordinate value of the point in the ROC curve where the stability coefficient threshold is equal to 1, representing the ratio of accurately predicted landslides to actual landslides.

[0007] Furthermore, the specific method for determining the initial value range and value interval of cohesion and internal friction angle is as follows: based on the geotechnical test and engineering geological analogy method of the study area, determine the initial value range and value interval of the calibration parameters cohesion and internal friction angle of the study area.

[0008] Furthermore, the engineering geological rock group is divided into loose rock group, weak rock group, soft-hard composite rock group and hard rock group.

[0009] Furthermore, the different groundwater level depth conditions include groundwater level depths of Z, 0.9Z, 0.7Z, 0.4Z, 0.1Z and 0, respectively; where Z represents the soil layer thickness.

[0010] Furthermore, the specific method for creating ROC curves for each set of parameters under each condition based on historical landslide data using different stability coefficient thresholds includes: based on historical landslide data of each engineering geological rock group, the TRIGRS model is used to calculate the stability coefficient corresponding to each set of parameters under each condition; the stability coefficient values ​​of all effective pixels are extracted from the stability coefficient raster and sorted and deduplicated to obtain a monotonically increasing sequence; each stability coefficient value in the sequence is used as a threshold to calculate the true positive rate and false positive rate in sequence, thereby obtaining a series of coordinate points in the ROC map under each condition; the coordinate points are connected to form a line to create the ROC curve for each set of parameters under each condition.

[0011] Furthermore, the formula for calculating the initial area ratio index is as follows:

[0012] in This represents the initial area ratio; This represents the area of ​​the unstable region predicted by the TRIGRS model for each parameter combination under conditions of no rainfall and a groundwater level depth equal to the soil layer thickness. This represents the total area of ​​the entire study region.

[0013] Furthermore, specific methods for eliminating unreasonable parameter combinations by setting initial area ratio and true positive rate indicators include: parameter combinations with an initial area ratio greater than or equal to 0.03 or a true positive rate greater than or equal to 0.45 when the groundwater level depth reaches 0.1Z are recorded as unreasonable parameter combinations and eliminated.

[0014] Furthermore, under rainfall-triggered landslide conditions, the specific methods for evaluating the performance of the TRIGRS model under the remaining parameter combinations and obtaining the optimal combination of physical and mechanical parameters of soil and rock masses for the entire study area and different engineering geological rock component zones include: under rainfall-triggered landslide conditions, inputting the remaining parameter combinations into the TRIGRS model to obtain the stability coefficients output by the TRIGRS model; where grids with stability coefficients greater than 1 are non-landslide areas, and grids with stability coefficients less than or equal to 1 are landslide areas; comparing the grid results output by the TRIGRS model with historical landslide layers to obtain the prediction result categories, including true positives, false positives, true negatives, and false negatives; and creating a confusion matrix M based on the prediction result categories in each engineering geological rock component zone of the study area.

[0015]

[0016]

[0017]

[0018]

[0019] TP represents a true positive, which is the number of rasters that are predicted to be unstable and are actually unstable; FP represents a false positive, which is the number of rasters that are predicted to be unstable but are actually stable; FN represents a false negative, which is the number of rasters that are predicted to be stable but are actually unstable; and TN represents a true negative, which is the number of rasters that are predicted to be stable and are actually stable. Grids identified by the TRIGRS model with a stability coefficient less than or equal to 1; Grids with a stability coefficient greater than 1 identified by the TRIGRS model; The grid occupied by historical landslides; The grid outside the historical landslide area; Refers to the grid position; the overall accuracy is obtained based on the confusion matrix M. Recall and area ratio :

[0020]

[0021] Among the parameter combinations with a recall rate greater than 0.5 and an overall accuracy greater than 0.5, the cohesion and internal friction angle corresponding to the smallest area ratio will be used as the calibration results of the physical and mechanical parameters of the soil and rock mass in the entire study area and different engineering geological rock component zones under rainfall conditions.

[0022] A method for predicting landslide disaster risk based on calibration of geophysical parameters under rainfall conditions is provided. The method includes: using DEM data of the study area, soil layer thickness, engineering geological rock group type, rainfall scenario, weight density of soil and rock mass, slope, weight density of water body, and calibrated cohesion and internal friction angle as inputs to the TRIGRS model to predict the risk of landslide disaster under different rainfall scenarios in the future.

[0023] Furthermore, the slope was calculated from the DEM data using the ArcGIS platform; the relationship between soil layer thickness and slope is: Z = -0.04822θ + 3.5, where Z is the soil layer thickness and θ is the slope.

[0024] The beneficial effects of this invention are as follows: The method for calibrating the physical and mechanical parameters of soil and rock under rainfall conditions relies on the TRIGRS model to use the cohesion and internal friction angle of soil and rock as parameters to be calibrated. In view of the uncertainty and spatial heterogeneity of soil and rock parameters, the method divides the soil and rock in the study area into different engineering geological rock groups according to the characteristics of the soil and rock. The dynamic response of groundwater level is used as the constraint of rainfall process. Combined with the historical landslide distribution in the study area, the cohesion and internal friction angle parameters of the entire study area and different engineering geological rock groups are inverted and calibrated. This method can provide a new approach for the automated determination of parameters when using physical and mechanical models to conduct regional rainfall-induced landslide risk assessment in areas with scarce data.

[0025] This method for predicting landslide disaster risks caused by rainfall can determine key physical and mechanical parameters of soil and rock more efficiently and accurately, and can more realistically and accurately predict the spatial distribution characteristics of landslide disasters caused by rainfall in the study area. Attached Figure Description

[0026] Figure 1 is a flowchart of this method; Figure 2 shows the sensitivity analysis results of the TRIGRS model parameters; Figure 3 shows the variation of the true positive rate for all parameter combinations; where (a) represents the entire region, (b) represents the loose rock group, (c) represents the weak rock group, (d) represents the soft-hard rock combination group, and (e) represents the hard rock group; Figure 4 shows the variation of the true positive rate for parameter combinations after screening by the initial area ratio index and the true positive rate index; where (a) represents the entire region, (b) represents the loose rock group, (c) represents the weak rock group, (d) represents the soft-hard rock combination group, and (e) represents the hard rock group; Figure 5 shows the parameter combination results; where (a) represents the entire region, (b) represents the loose rock group, (c) represents the weak rock group, (d) represents the soft-hard rock combination group, and (e) represents the hard rock group. Figure 6 shows the calibration simulation results of potential landslides across the entire study area under a 50-year return period rainfall condition; Figure 7 shows the zonal calibration simulation results of potential landslides under a 50-year return period rainfall condition in the study area. Detailed Implementation

[0027] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0028] As shown in Figure 1, the calibration method for geotechnical physical parameters under rainfall conditions includes the following steps: S1, using cohesion and internal friction angle as parameters to be calibrated, determining the initial range and interval of cohesion and internal friction angle; S2, calculating the stability coefficients of each engineering geological rock group in the study area under different groundwater depth conditions and all parameter combinations using the TRIGRS model, and creating ROC curves for each parameter group under each condition using different stability coefficient thresholds based on historical landslide data; S3, in the absence of rainfall, eliminating unreasonable parameter combinations by setting initial area ratio and true positive rate indicators; S4, under rainfall-triggered landslide conditions, evaluating the performance of the TRIGRS model under the remaining parameter combinations, obtaining the optimal geotechnical physical parameter combinations for the entire study area and different engineering geological rock group zones, thus obtaining the calibration results of geotechnical physical parameters under rainfall conditions; where the true positive rate is the ordinate value of the point in the ROC curve where the stability coefficient threshold is equal to 1, representing the ratio of accurately predicted landslides to actual landslides.

[0029] In this embodiment, when collecting historical data, high-precision DEM data, geological maps, rainfall, and other natural geological environment background data of the study area can be collected. Simultaneously, the spatial distribution of historical rainfall-induced landslides in the study area can be determined based on county-level geological hazard distribution data and detailed geological hazard investigation reports, or the spatial distribution of rainfall-induced landslides in the study area can be determined based on field surveys and remote sensing interpretation after an extreme rainfall event. This data forms a historical rainfall-induced landslide surface distribution map (including landslide point latitude and longitude, landslide area, etc.), which serves as the inversion layer for key physical and mechanical parameters of the TRIGRS model.

[0030] It is also possible to calculate the maximum daily rainfall under a certain rainfall scenario for each meteorological station by using the Pearson Type III frequency curve based on the historical rainfall data of the study area and its adjacent meteorological stations over many years, and obtain the spatial heterogeneity distribution map of rainfall intensity in the study area based on the IDW inverse distance weighted spatial interpolation method, or to form a spatial distribution map of rainfall intensity under a certain rainfall scenario in the study area based on a single extreme rainfall data, and rasterize it into an ASC file based on the ArcGIS platform.

[0031] In the specific implementation process, the initial range and interval of cohesion and internal friction angle are determined as follows: based on the geotechnical test and engineering geological analogy methods in the study area, the initial range of the calibration parameters cohesion and internal friction angle in the study area is initially determined, and the interval can be taken as 1.

[0032] In this embodiment, based on the differences in engineering geological characteristics of the soil and rock masses in the study area (such as strength, hardness, weathering resistance, lithology, etc.), the lithology of the study area can be divided into four rock groups: loose rock group, weak rock group, soft-hard composite rock group, and hard rock group. Different groundwater level depths include Z, 0.9Z, 0.7Z, 0.4Z, 0.1Z, and 0, where Z represents the soil layer thickness. The groundwater level depth is the distance from the groundwater level to the surface.

[0033] In the specific implementation process, the TRIGRS model is used as an existing model, and this embodiment uses its key control parameter cohesion. and internal friction angle As the primary calibration parameter, other parameters (such as severe gamma) s saturated permeability coefficient K s The hydraulic diffusion coefficient (D0, etc.) are non-calibration parameters. The values ​​of non-calibration parameters are mainly determined through indoor and outdoor geotechnical tests, engineering geological analogy methods, or empirical formulas. This application will not elaborate further on these parameters, except for cohesion. and internal friction angle The process of obtaining or selecting external parameters.

[0034] In this embodiment, the initial ranges of cohesion and internal friction angle can be determined through geotechnical tests and engineering geological analogy methods, such as cohesion. The internal friction angle is [1, 10]. If the range is [10, 30], and the interval between each value is set to 1, then there are a total of 210 values. , Parameter combinations.

[0035] In the specific implementation process, the method for creating ROC curves for each set of parameters under each condition based on historical landslide data using different stability coefficient thresholds includes: based on historical landslide data of each engineering geological rock group, the TRIGRS model is used to calculate the stability coefficient corresponding to each set of parameters under each condition; the stability coefficient values ​​of all effective pixels are extracted from the stability coefficient raster and sorted and deduplicated to obtain a monotonically increasing sequence; each stability coefficient value in the sequence is used as a threshold to calculate the true positive rate and false positive rate in sequence, thereby obtaining a series of coordinate points in the ROC map under each condition; finally, the coordinate points are connected into a line to create the ROC curve for each set of parameters under each condition.

[0036] Since no landslides or only a few landslides occur in the initial state without rainfall, and regional landslides are only triggered after the groundwater level reaches a certain level, two indicators, the initial area ratio and the true positive rate, are set to eliminate unreasonable parameter combinations. Simultaneously, under the condition of rainfall-triggered landslides, i.e., when the groundwater level reaches the surface (depth = 0, where depth is the groundwater level depth), the confusion matrix index is used to evaluate the performance of the TRIGRS model under the remaining parameter combinations, thereby selecting the optimal combination of physical and mechanical parameters for the soil and rock masses in the entire study area and different engineering geological rock groups.

[0037] Specifically, the formula for calculating the initial area ratio index is as follows:

[0038] in This represents the initial area ratio; This represents the area of ​​the unstable region predicted by the TRIGRS model for each parameter combination under conditions of no rainfall and a groundwater level depth equal to the soil layer thickness. This represents the total area of ​​the entire study region.

[0039] The specific method for eliminating unreasonable parameter combinations by setting initial area ratio and true positive rate indicators includes: parameter combinations with an initial area ratio greater than or equal to 0.03 or a true positive rate greater than or equal to 0.45 when the groundwater level depth reaches 0.1Z are marked as unreasonable parameter combinations and eliminated. That is, the remaining parameters must meet the following conditions:

[0040] in This is a combination of residual cohesion and internal friction angle parameters.

[0041] Under rainfall-triggered landslide conditions, the specific method for evaluating the performance of the TRIGRS model under residual parameter combinations and obtaining the optimal combination of physical and mechanical parameters of soil and rock masses for the entire study area and different engineering geological rock component zones includes: inputting the residual parameter combinations into the TRIGRS model under rainfall-triggered landslide conditions to obtain the stability coefficients output by the TRIGRS model; grids with stability coefficients greater than 1 are non-landslide areas, and grids with stability coefficients less than or equal to 1 are landslide areas; comparing the grid results output by the TRIGRS model with historical landslide layers to obtain the prediction result categories, including true positives, false positives, true negatives, and false negatives; and creating a confusion matrix M based on the prediction result categories in each engineering geological rock component zone of the study area.

[0042]

[0043]

[0044]

[0045]

[0046] TP represents a true positive, which is the number of raster points that are predicted to be unstable and are actually unstable; FP represents a false positive, which is the number of raster points that are predicted to be unstable but are actually stable; FN represents a false negative, which is the number of raster points that are predicted to be stable but are actually unstable; and TN represents a true negative, which is the number of raster points that are predicted to be stable and are actually stable. Grids identified by the TRIGRS model with a stability coefficient less than or equal to 1; Grids with a stability coefficient greater than 1 identified by the TRIGRS model; The grid is occupied by historical landslides (the actual unstable slope grid). The grid outside the historical landslide area (actual slope stability grid). Refers to the grid position; the overall accuracy is obtained based on the confusion matrix M. Recall and area ratio :

[0047]

[0048] Among the parameter combinations with a recall rate greater than 0.5 and an overall accuracy greater than 0.5, the cohesion and internal friction angle corresponding to the smallest area ratio are taken as the calibration results of the physical and mechanical parameters of the soil and rock mass in the entire study area and different engineering geological rock component zones under rainfall conditions. The corresponding expressions are:

[0049] .

[0050] The rainfall-based method for predicting landslide disaster risk involves using DEM data of the study area, soil layer thickness, engineering geological rock group type, rainfall scenario, weight density of soil and rock mass, slope, weight density of water body, and calibrated cohesion and internal friction angle as inputs to the TRIGRS model to predict the risk of landslide disaster under different rainfall scenarios in the future.

[0051] The slope was calculated from the DEM data using the ArcGIS platform; the relationship between soil thickness and slope is: Z = -0.04822θ + 3.5, where Z is the soil thickness and θ is the slope.

[0052] In one embodiment of the present invention, to evaluate the responsiveness of the TRIGRS model to changes in input parameters, a local sensitivity analysis was performed on the TRIGRS model. This involved perturbing the parameter to be evaluated near a specific point while keeping other input parameters constant, and observing the impact of the changed parameter on the model output. Based on the soil characteristics of the study area, the variable values ​​and default values ​​of each parameter during the sensitivity analysis were determined (see Table 1). Simultaneously, to measure the stability coefficient... The relative change in the stability coefficient is defined by the sensitivity to each parameter. and relative changes in parameter values Two indicators. Among them... This represents the stability coefficient value for each simulation. The stability coefficient value is calculated using the default parameter values. These are the parameter values ​​that change for each simulation. These are the default values ​​for the corresponding changing parameters. These represent the maximum and minimum values ​​of the corresponding changing parameters, respectively.

[0053] Table 1: Parameter Range and Default Values

[0054] The sensitivity analysis results of the TRIGRS model parameters are shown in Figure 2. According to the results, the model is significantly more sensitive to the physical and mechanical parameters of the soil and rock mass than to hydrological parameters. The stability of the slope is mainly controlled by the physical and mechanical parameters of the soil and rock mass. Among the physical and mechanical parameters, the sensitivity from largest to smallest is soil thickness, cohesion, internal friction angle, and unit weight. The sensitivity curve for soil thickness is non-linear, increasing with decreasing soil thickness, while the sensitivity curves for cohesion and internal friction angle are linear. Among the hydrological parameters, the sensitivity from largest to smallest is groundwater level depth, saturated permeability coefficient, SWRC index, and water storage capacity. The sensitivity curve for groundwater level depth is linear, decreasing by 30% when the water level reaches the surface. The sensitivity curves for saturated permeability coefficient and SWRC index are non-linear, and their sensitivity no longer changes after reaching a certain value. The stability coefficient is not sensitive to water storage capacity.

[0055] In this embodiment, the cohesion and internal friction angle of a certain area are calibrated. The true positive rate of each parameter combination in the whole area and each sub-area changes with the groundwater depth as shown in Figure 3. After the unreasonable parameter combinations are eliminated by the initial area ratio index and the true positive rate index, the true positive rate of the remaining selectable parameter combinations changes with the groundwater depth as shown in Figure 4.

[0056] For the remaining parameter combinations after screening, under the condition of rainfall-triggered landslides, the three evaluation statistics based on the confusion matrix are calculated for each parameter combination, as shown in Figure 5. Further screening is conducted to select parameter groups with recall and precision greater than 0.5 in the whole area and in each sub-area. Finally, considering the above three indicators, the parameter combination that best matches the actual geological conditions of the study area is selected as the calibration result (diamond in Figure 5) while keeping the area ratio as small as possible.

[0057] The calibration parameters obtained based on the above calibration procedure are shown in Table 2. According to the potential landslide classification criteria, the simulation results are divided into extremely unstable regions (…). Unstable regions ), stable region ( The determined parameters and raster data were input into the TRIGRS model to obtain the dynamic changes of potential landslide areas within 24 hours under a 50-year return period rainfall scenario. Figures 6 and 7 show these results.

[0058] Table 2: Results of physical and mechanical parameters of soil and rock mass calibrated based on the TRIGRS model

[0059] Simulation results show that, after calibration, both single-parameter and zonal-parameter simulations predict potential landslide areas with good dynamic change effects, which are relatively consistent with the actual situation. The study area showed few unstable areas in the initial stage of rainfall, with large-scale instability only appearing after 12 hours of continuous rainfall. To compare the full-area and zonal calibration simulation results, Table 3 shows the area changes of each level of region under different rainfall durations. Under a 50-year return period extreme heavy rainfall scenario, no extremely unstable areas appeared in the study area during the initial stage of rainfall (0-1 hour), and the proportion of unstable areas was less than 1%. As rainfall continued, the unstable areas in the study area gradually expanded, with larger areas of instability appearing after 8 hours of rainfall. It was observed that the area of ​​extremely unstable areas in the study area remained unchanged after 12 hours of rainfall, and the area of ​​unstable areas remained constant after 16 hours of rainfall. The total unstable area tended to stabilize after 16 hours of rainfall, indicating that the soil and rock masses in the area reached saturation after 16 hours of rainfall. Furthermore, as shown in Table 3, there are differences between the full-area calibration simulation results and the zonal calibration simulation results. During the same rainfall duration, the area of ​​the highly unstable region predicted by the zonal simulation is consistently lower than that predicted by the full-area simulation. The unstable region area begins to slightly exceed that of the full-area simulation 12 hours after rainfall, but the total unstable region area predicted by the zonal simulation remains lower than that predicted by the full-area simulation. The final zonal prediction of the unstable region area is 2235.9 km². 2This accounts for 26.97% of the total study area, which is lower than the predicted unstable area of ​​2381.7 km² for the entire region. 2 Its total area accounts for 28.73%.

[0060] Table 3: Area changes at different rainfall durations under 50-year return period rainfall conditions

[0061] In this embodiment, without considering parameter calibration, the cohesion of the soil and rock in the study area was determined to be 6 kPa and the internal friction angle to be 29° by consulting the data of the study area and relevant literature. The DEM data, soil layer thickness, rainfall data, soil and rock mass density, slope, water mass density, and the uncalibrated cohesion and internal friction angle of the study area were used as inputs to the TRIGRS model to obtain the simulation results of a 16-hour 50-year return period rainfall under uncalibrated conditions. Based on the simulation results of the 16-hour 50-year return period rainfall under each condition, the model performance was comprehensively evaluated using the confusion matrix, receiver operating characteristic curve, and area under the curve (AUC). The comparison of the model performance evaluation indicators for each condition is shown in Table 4.

[0062] The results show that parameter calibration significantly improves the model's landslide identification capability. Compared with the uncalibrated simulation, the calibrated model achieves a significant improvement in both recall and precision, indicating that parameter calibration can effectively enhance the model's accuracy in identifying landslide-prone areas. Among different calibration strategies, the zonal calibration simulation performs best, with recall, precision, and AUC values ​​reaching 55%, 73%, and 0.71, respectively, all significantly higher than the corresponding indicators of the uncalibrated simulation and the whole-area calibration simulation. Compared with the other two scenarios, zonal calibration not only has a lower false alarm rate but also correctly captures more actual landslide events with fewer predicted instability areas. In terms of parameter calibration strategies, the zonal calibration method, which fully considers the spatial heterogeneity of parameters, is significantly better than the whole-area calibration method based on a single parameter. By dividing the study area into zones and calibrating parameters separately, the model can more accurately characterize the differences in physical and mechanical properties of different geological rock groups, thereby significantly improving the efficiency and accuracy of regional rainfall-induced landslide risk assessment.

[0063] Table 4: Comparison of model evaluation indicators after 16 hours of rainfall under different conditions

[0064] Related studies have shown that the resolution of the Digital Elevation Model (DEM) used in the TRIGRS model calculations affects the simulation predictions. High-resolution raster calculations yield more accurate results, with 10m and 15m raster sizes considered optimal. However, due to the large study area provided in this embodiment, and considering computational efficiency, a DEM resolution of 90m was used. Furthermore, the historical landslide data does not represent landslides induced by a single rainfall event, all of which affect the calibration accuracy of the model parameters. Therefore, accurate rainfall-induced landslide sample data, accurate estimation of soil thickness spatial distribution, and high-resolution DEM data will all improve the calibration accuracy of the model parameters to varying degrees, further enhancing the evaluation and predictive performance of the calibrated model.

Claims

1. A method for calibrating geophysical parameters under rainfall conditions, characterized in that, Includes the following steps: Using cohesion and internal friction angle from the physical and mechanical parameters of soil and rock as parameters to be calibrated, the initial range and interval of cohesion and internal friction angle are determined. The TRIGRS model was used to calculate the stability coefficients of each engineering geological rock component zone in the study area under different groundwater levels and depths, and under all parameter combinations. Based on historical landslide data, ROC curves were created for each parameter group under each condition using different stability coefficient thresholds. In the absence of rainfall, unreasonable parameter combinations were eliminated by setting initial area ratio and true positive rate indices. Under rainfall-triggered landslide conditions, the performance of the TRIGRS model under the remaining parameter combinations was evaluated to obtain the optimal combination of physical and mechanical parameters of soil and rock in different engineering geological rock component zones in the study area, i.e., the calibration results of physical and mechanical parameters of soil and rock under rainfall conditions were obtained. The true positive rate is the ordinate value of the point in the ROC curve where the stability coefficient threshold is equal to 1, representing the ratio of accurately predicted landslides to actual landslides.

2. The method for calibrating geophysical parameters under rainfall conditions according to claim 1, characterized in that, The specific method for determining the initial range and interval of cohesion and internal friction angle is as follows: based on the geotechnical test and engineering geological analogy method of the study area, determine the initial range and interval of the calibration parameters cohesion and internal friction angle of the study area.

3. The method for calibrating geophysical parameters under rainfall conditions according to claim 1, characterized in that, The engineering geological rock group is divided into loose rock group, weak rock group, soft-hard composite rock group and hard rock group.

4. The method for calibrating geotechnical physical parameters under rainfall conditions according to claim 1, characterized in that, Different groundwater level depth conditions include groundwater level depths of Z, 0.9Z, 0.7Z, 0.4Z, 0.1Z and 0, respectively; where Z represents the soil layer thickness.

5. The method for calibrating geophysical parameters under rainfall conditions according to claim 1, characterized in that, The specific method for creating ROC curves for each set of parameters under each condition based on historical landslide data using different stability coefficient thresholds includes: based on historical landslide data of each engineering geological rock group, the TRIGRS model is used to calculate the stability coefficient corresponding to each set of parameters under each condition; the stability coefficient values ​​of all effective pixels are extracted from the stability coefficient raster and sorted and deduplicated to obtain a monotonically increasing sequence; each stability coefficient value in the sequence is used as a threshold to calculate the true positive rate and false positive rate in sequence, thereby obtaining a series of coordinate points in the ROC map under each condition; the coordinate points are connected to form a line to create the ROC curve for each set of parameters under each condition.

6. The method for calibrating geotechnical physical parameters under rainfall conditions according to claim 1, characterized in that, The formula for calculating the initial area ratio index is: in This represents the initial area ratio; This represents the area of ​​the unstable region predicted by the TRIGRS model for each parameter combination under conditions of no rainfall and a groundwater level depth equal to the soil layer thickness. This represents the total area of ​​the entire study region.

7. The method for calibrating geotechnical physical parameters under rainfall conditions according to claim 6, characterized in that, The specific methods for eliminating unreasonable parameter combinations by setting initial area ratio and true positive rate indicators include: parameter combinations with an initial area ratio greater than or equal to 0.03 or a true positive rate greater than or equal to 0.45 when the groundwater level depth reaches 0.1Z are recorded as unreasonable parameter combinations and eliminated.

8. The method for calibrating geotechnical physical parameters under rainfall conditions according to claim 1, characterized in that, Under rainfall-triggered landslide conditions, the specific method for evaluating the performance of the TRIGRS model under residual parameter combinations and obtaining the optimal combination of physical and mechanical parameters of soil and rock masses in different engineering geological rock component zones of the study area includes: inputting the residual parameter combinations into the TRIGRS model under rainfall-triggered landslide conditions and obtaining the stability coefficients output by the TRIGRS model; where grids with stability coefficients greater than 1 are non-landslide areas, and grids with stability coefficients less than or equal to 1 are landslide areas; comparing the grid results output by the TRIGRS model with historical landslide layers to obtain the prediction result categories, including true positives, false positives, true negatives, and false negatives; and creating a confusion matrix M based on the prediction result categories in each engineering geological rock component zone of the study area. TP represents a true positive, which is the number of rasters that are predicted to be unstable and are actually unstable; FP represents a false positive, which is the number of rasters that are predicted to be unstable but are actually stable; FN represents a false negative, which is the number of rasters that are predicted to be stable but are actually unstable; and TN represents a true negative, which is the number of rasters that are predicted to be stable and are actually stable. Grids identified by the TRIGRS model with a stability coefficient less than or equal to 1; Grids with a stability coefficient greater than 1 identified by the TRIGRS model; The grid occupied by historical landslides; The grid outside the historical landslide area; Refers to the grid position; the overall accuracy is obtained based on the confusion matrix M. Recall and area ratio ： Among the parameter combinations with a recall rate greater than 0.5 and an overall accuracy greater than 0.5, the cohesion and internal friction angle corresponding to the smallest area ratio will be used as the calibration results of the physical and mechanical parameters of the soil and rock mass in different engineering geological rock component zones under rainfall conditions.

9. A method for predicting the risk of landslide disasters caused by rainfall, based on the calibration method of geotechnical physical parameters under rainfall conditions as described in any one of claims 1 to 8, characterized in that, include: The DEM data, soil thickness, engineering geological rock group type, rainfall scenario, weight density of soil and rock mass, slope, weight density of water body, and calibrated cohesion and internal friction angle of the study area are used as inputs to the TRIGRS model to predict the risk of landslide disaster under different rainfall scenarios in the future.

10. The method for predicting landslide disaster risk based on rainfall according to claim 9, characterized in that, The slope was calculated from the DEM data using the ArcGIS platform; the relationship between soil thickness and slope is: Z = -0.04822θ + 3.5, where Z is the soil thickness and θ is the slope.

Images