A landslide meteorological risk early warning method and system combining statistics and mechanism models

Abstract

This invention provides a landslide meteorological risk early warning method and system that integrates statistical and mechanistic models, belonging to the field of geological disaster early warning technology. The method involves calculating and correcting historical rainfall-type landslide samples and key factor data to obtain a corrected base probability; using high-resolution soil moisture data and a coupled mechanistic model, a landslide safety factor is obtained; based on the landslide safety factor, rainfall infiltration lag effect, and the cumulative rate of surface deformation, a dynamic weighting coefficient is calculated using a nonlinear lag suppression function and a deformation acceleration amplification function; this dynamic weighting coefficient is then used to perform Bayesian correction on the corrected base probability to obtain the final landslide probability; based on the final landslide probability, risk warning levels are classified to obtain the landslide meteorological risk early warning result. This invention solves the problem in existing landslide meteorological risk early warning systems where purely statistical models are insufficient to reflect the physical mechanisms of landslide occurrence, leading to false alarms under special conditions.

Description

Technical Field

[0001] This invention relates to the field of geological disaster early warning technology, and in particular to a landslide meteorological risk early warning method and system that integrates statistical and mechanistic models. Background Technology

[0002] Landslides are one of the common geological disasters in mountainous areas, especially in areas with complex geological conditions and abundant rainfall. Landslides seriously threaten the safety of people's lives and property as well as the operational safety of major projects. Therefore, conducting research on landslide meteorological risk early warning is of great practical significance. Currently, landslide meteorological risk early warning methods are mainly divided into two categories: the first category is the data-driven statistical model method, which constructs an early warning model by analyzing the statistical relationship between historical landslide samples and factors such as rainfall and topography; the second category is the physical mechanism model method, which simulates the impact of rainfall infiltration on slope stability by coupling hydrological models and slope stability models. However, the existing technologies still have the following shortcomings: (1) Pure statistical models are highly dependent on the quality and quantity of historical samples, making it difficult to truly reflect the physical mechanism of landslides. They are prone to false alarms under special conditions such as steep terrain but dry soil; (2) Pure mechanism models have high requirements for the accuracy of soil and rock parameters, and the calculation process is complex. They face efficiency bottlenecks in large-scale regional applications; (3) Existing methods have not yet been able to effectively integrate the rapid prediction capability of statistical models with the physical constraint advantages of mechanism models, making it difficult to achieve a good balance between early warning accuracy and reliability. Summary of the Invention

[0003] To address the aforementioned shortcomings in existing technologies, this invention provides a landslide meteorological risk early warning method and system that integrates statistical and mechanistic models. This solves the problems in existing landslide meteorological risk early warning systems, where pure statistical models are difficult to reflect the physical mechanisms of landslide occurrence, leading to false alarms under special conditions, while pure mechanistic models have low computational efficiency and are difficult to apply on a large scale. Furthermore, existing methods of integrating the two models are difficult to balance the accuracy and reliability of early warning, do not address the mismatch between spatiotemporal scales, and lack dynamic physical mechanism correction.

[0004] To achieve the aforementioned objectives, the technical solution adopted by this invention is: a landslide meteorological risk early warning method integrating statistical and mechanistic models, comprising: S1: Based on historical rainfall-type landslide samples and key factor data, a statistical model was constructed using logistic regression to perform calculations, and a Gaussian membership function was used to perform slope correction. The mean of the Gaussian membership function is related to the average internal friction angle of the regional soil and rock mass, and the standard deviation is related to the statistical dispersion of the regional landslide points in the slope dimension, thus obtaining the corrected basic probability. S2: Based on high-resolution soil moisture data and meteorological data, a mechanism model including hydrological model and landslide dynamic model is used for coupled calculation. In the coupled calculation, bilinear spatiotemporal interpolation with topographic moisture index correction factor and boundary flux smoothing algorithm are used to handle spatiotemporal scale differences and obtain landslide safety factor. S3: Based on the landslide safety factor, rainfall infiltration lag effect, and the cumulative rate of surface deformation in the early stage, the dynamic weighting coefficient is calculated by nonlinear lag suppression function and deformation acceleration amplification function. The dynamic weighting coefficient is then used to perform Bayesian correction on the corrected base probability to obtain the final landslide probability. S4: Based on the final landslide probability, classify the risk warning level, obtain the landslide meteorological risk warning result, and complete the landslide meteorological risk warning.

[0005] Further, S1 includes: Based on historical rainfall-type landslide samples, effective rainfall and slope were selected as key factor data. A statistical model was established using logistic regression to calculate the initial probability of landslides. The mean of the Gaussian membership function is calibrated based on the peak slope distribution of historical landslide points in the statistical region, and the upper bound constraint of the mean is set based on the average internal friction angle of the soil and rock mass in the region. The standard deviation is determined based on the statistical dispersion of the landslide sample points in the slope dimension, and the Gaussian membership function is constructed. The initial probability of the landslide is corrected by applying a Gaussian membership function to obtain the corrected base probability.

[0006] Furthermore, the expression for the corrected base probability is: ; ; in, This represents the corrected base probability. This represents the initial probability of a landslide calculated based on historical rainfall-type landslide samples. This represents the slope correction coefficient for Gaussian membership functions. Represents slope data. This represents the mean of the Gaussian membership function, calibrated based on the peak slope distribution of historical landslide points in the statistical region. It represents the standard deviation determined based on the statistical dispersion of regional landslide sample points in the slope dimension.

[0007] Further, S2 includes: S21: Based on the high-resolution soil moisture data output by the hydrological model driven by meteorological data, in the coupled calculation using a mechanism model that includes the hydrological model and the landslide dynamic model, the bilinear spatiotemporal interpolation algorithm with the introduction of the topographic moisture index correction factor and the boundary flux smoothing algorithm are used to process the spatiotemporal scale differences between the hydrological model and the landslide dynamic model, so as to obtain continuous soil moisture field data. S22: Based on continuous soil moisture field data, the landslide safety factor is calculated using the landslide dynamic model in the mechanism model.

[0008] Further, S21 includes: Based on high-resolution soil moisture data driven by meteorological data-driven hydrological model output, in the coupled calculation using a mechanism model that includes hydrological model and landslide dynamic model, a bilinear interpolation method with topographic moisture index correction factor is used for spatial downscaling to obtain downscaled soil moisture data. Based on the downscaled soil moisture data, time interpolation was performed using a first-order linearized approximation equation to obtain the time-interpolated soil moisture data. Based on the time-interpolated soil moisture data, the boundary flux smoothing process is performed using the anisotropic diffusion filter operator to handle the spatiotemporal scale differences between the hydrological model and the landslide dynamic model, thereby obtaining continuous soil moisture field data.

[0009] Further, S22 includes: Hydrostatic pressure data are obtained based on continuous soil moisture field data. Based on hydrostatic pressure data, combined with landslide body weight data, slope data, soil internal friction angle data, soil cohesion data, and slope length data, the landslide safety factor is calculated using the landslide dynamic mode in the mechanism model.

[0010] Furthermore, the expression for the landslide safety factor is: ; in, Indicates the landslide safety factor. This represents soil cohesion data. This represents the slope length data. This represents the self-weight data of the landslide body. Represents slope data. This represents hydrostatic pressure data. This represents the friction angle data within the soil.

[0011] Further, S3 includes: The basic weights are determined based on the threshold range in which the landslide safety factor falls; The rainfall infiltration retardation effect is calculated by using the ratio of soil volumetric water content to soil saturated hydraulic conductivity. The cumulative deformation rate of the land surface in the early stage is calculated by using the ratio of the cumulative deformation in the early stage to the critical deformation threshold before the occurrence of historical landslide events. The hysteresis suppression function is used to process the hysteresis effect of rainfall infiltration, and the deformation acceleration amplification function is used to process the cumulative rate of surface deformation in the early stage. The processing results are combined with the basic weight to calculate the dynamic weight coefficient. The corrected base probability is corrected using dynamic weighting coefficients to obtain the final landslide probability.

[0012] Furthermore, the expression for the final landslide probability is: ; ; ; ; ; ; in, Indicates the final probability of a landslide. This represents the corrected base probability. Indicates dynamic weighting coefficients. Indicates the basic weight. This represents the hysteresis suppression function, which represents the nonlinearity. This represents the deformation acceleration amplification function. This indicates the amount of rainfall infiltration lag effect. This represents the soil volumetric water content data at the current moment, output by the hydrological model simulation. This represents the saturated hydraulic conductivity data of soil obtained from tests on the physical properties of soil and rock. This represents the first regional empirical parameters obtained based on the inversion of the relationship between historical rainfall and landslide response in the region. This represents the threshold for the delayed response time of rainfall infiltration, set based on regional experience. It represents the cumulative rate of early surface deformation. This represents the cumulative deformation variables obtained from previous periods based on time series analysis. This represents the critical deformation threshold obtained based on statistics of historical landslide events. This represents the second regional empirical parameter obtained by inversion based on the historical landslide deformation evolution patterns in the region. This represents the critical threshold for deformation acceleration, set based on regional experience.

[0013] This invention provides a landslide meteorological risk early warning system that integrates statistical and mechanistic models, comprising: The data acquisition module is used to acquire historical rainfall-type landslide samples, meteorological data, key factor data, and high-resolution soil moisture data; The statistical model calculation module is used to construct a statistical model based on historical rainfall-type landslide samples and key factor data, and to perform slope correction using a Gaussian membership function. The mean of the Gaussian membership function is related to the average internal friction angle of the regional soil and rock mass, and the standard deviation is related to the statistical dispersion of the regional landslide points in the slope dimension, so as to obtain the corrected basic probability. The mechanism model calculation module is used to perform coupled calculations based on high-resolution soil moisture data and meteorological data, using a mechanism model that includes hydrological and landslide dynamic models. In the coupled calculation, bilinear spatiotemporal interpolation with topographic moisture index correction factor and boundary flux smoothing algorithm are used to handle spatiotemporal scale differences and obtain the landslide safety factor. The Bayesian correction module is used to calculate dynamic weighting coefficients based on the landslide safety factor, rainfall infiltration hysteresis effect, and the cumulative rate of surface deformation in the early stage. It uses a nonlinear hysteresis suppression function and a deformation acceleration amplification function to calculate the dynamic weighting coefficients and then uses the dynamic weighting coefficients to perform Bayesian correction on the corrected base probability to obtain the final landslide probability. The early warning output module is used to classify risk warning levels based on the final landslide probability, obtain landslide meteorological risk warning results, and complete the landslide meteorological risk warning.

[0014] The beneficial effects of this invention are as follows: This invention provides a landslide meteorological risk early warning method that integrates statistical and mechanistic models. In the stage of constructing the statistical model, a Gaussian membership function is used to perform slope correction, and the mean is correlated with the average internal friction angle of the regional soil and rock mass, while the standard deviation is correlated with the statistical dispersion of landslide points in the slope dimension. This setting avoids the arbitrariness of experience in the selection of mathematical functions and parameter setting, establishes an intrinsic physical connection between the slope correction mechanism and the regional geological environment characteristics, and improves the transferability and physical interpretability of the model across different regions.

[0015] In the coupled calculation stage of the mechanism model, a bilinear spatiotemporal interpolation algorithm with a topographic moisture index correction factor and a boundary flux smoothing algorithm are used to handle the spatiotemporal scale differences. This setting effectively solves the problem of spatial resolution and time step mismatch between the hydrological model and the landslide dynamic model, ensuring that the soil moisture distribution and local water catchment characteristics remain physically consistent after downscaling, and providing high-precision physical-hydrological boundary conditions for subsequent landslide safety factor calculation.

[0016] In the Bayesian correction stage, dynamic weighting coefficients are calculated based on the rainfall infiltration hysteresis effect and the cumulative rate of anterior surface deformation using a nonlinear hysteresis suppression function and a deformation acceleration amplification function. This setup incorporates the infiltration hysteresis process and the progressive damage characteristics of deformation as physical state variables into the probabilistic update framework, achieving mechanism-driven dynamic probabilistic correction. While maintaining a high detection rate, it effectively suppresses high-risk false alarms in special scenarios such as steep terrain but dry soil, significantly reducing the false alarm rate. Attached Figure Description

[0017] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein: Figure 1 This is a schematic diagram illustrating an application scenario of a landslide meteorological risk early warning system that integrates statistical and mechanistic models, as shown in some embodiments of this specification. Figure 2 This is an exemplary flowchart of a landslide meteorological risk early warning method that integrates statistical and mechanistic models, as shown in some embodiments of this specification. Figure 3 This is an exemplary schematic diagram of the hysteresis suppression function curve shown in some embodiments of this specification; Figure 4 This is an exemplary schematic diagram of the deformation acceleration amplification function curve shown in some embodiments of this specification; Figure 5 This is an exemplary schematic diagram of the relationship curve between basic weight and safety factor shown in some embodiments of this specification. Detailed Implementation

[0018] 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.

[0019] Example 1 Figure 1 This is a schematic diagram of a landslide meteorological risk early warning system that integrates statistical and mechanistic models, as shown in some embodiments of this specification.

[0020] In some embodiments, the landslide meteorological risk early warning system integrating statistical and mechanistic models may include a data acquisition module for acquiring historical rainfall-type landslide samples, meteorological data, key factor data, and high-resolution soil moisture data; a statistical model calculation module for constructing a statistical model based on historical rainfall-type landslide samples and key factor data using logistic regression, and performing slope correction using a Gaussian membership function. The mean of the Gaussian membership function is correlated with the average internal friction angle of the regional soil and rock mass, and the standard deviation is correlated with the statistical dispersion of the regional landslide points in the slope dimension, to obtain the corrected basic probability; and a mechanistic model calculation module for calculating the underlying probability based on high-resolution soil moisture data and meteorological data. According to the method, a coupled calculation is performed using a mechanistic model that includes a hydrological model and a landslide dynamic model. In the coupled calculation, bilinear spatiotemporal interpolation with a topographic humidity index correction factor and a boundary flux smoothing algorithm are used to handle spatiotemporal scale differences to obtain the landslide safety factor. The Bayesian correction module is used to calculate the dynamic weight coefficient based on the landslide safety factor, rainfall infiltration hysteresis effect, and the cumulative rate of surface deformation in the early stage. The dynamic weight coefficient is then used to perform Bayesian correction on the corrected base probability to obtain the final landslide probability. The early warning output module is used to classify the risk warning level based on the final landslide probability to obtain the landslide meteorological risk warning result and complete the landslide meteorological risk warning.

[0021] In some embodiments, a landslide meteorological risk early warning system integrating statistical and mechanistic models can be used to implement a landslide meteorological risk early warning method integrating statistical and mechanistic models, including: S1: Based on historical rainfall-type landslide samples and key factor data, a statistical model is constructed using logistic regression for calculation, and a Gaussian membership function is used to perform slope correction. The mean of the Gaussian membership function is correlated with the average internal friction angle of the regional soil and rock mass, and the standard deviation is correlated with the statistical dispersion of the regional landslide points in the slope dimension, to obtain the corrected basic probability; S2: Based on high-resolution soil moisture data, a method integrating hydrological models and landslide dynamics is used to perform a landslide meteorological risk early warning method integrating statistical and mechanistic models. The mechanism model of the force model is coupled for calculation. In the coupled calculation, bilinear spatiotemporal interpolation with topographic humidity index correction factor and boundary flux smoothing algorithm are used to handle spatiotemporal scale differences to obtain the landslide safety factor; S3: Based on the landslide safety factor, rainfall infiltration lag effect and surface deformation accumulation rate, dynamic weight coefficients are calculated through nonlinear lag suppression function and deformation acceleration amplification function. The corrected base probability is then Bayesian corrected using the dynamic weight coefficients to obtain the final landslide probability; S4: Based on the final landslide probability, risk warning levels are divided to obtain landslide meteorological risk warning results, thus completing the landslide meteorological risk warning.

[0022] In some embodiments of this specification, the processor utilizes a landslide meteorological risk early warning system that integrates statistical and mechanistic models to execute a landslide meteorological risk early warning method that integrates statistical and mechanistic models. Through the coordinated operation of the data acquisition module, statistical model calculation module, mechanistic model calculation module, Bayesian correction module, and early warning output module, the organic integration and calculation of historical rainfall pattern mining and real-time physical hydrological mechanisms are achieved. This system architecture clearly defines the data flow and business logic of each processing stage, improving the automation level of the landslide meteorological risk early warning process and the reliability of its business applications.

[0023] Example 2 Figure 2 This is an exemplary flowchart illustrating a landslide meteorological risk early warning method that integrates statistical and mechanistic models, according to some embodiments of this specification. Figure 2 As shown, the process includes the following steps. In some embodiments, the process may be executed by a processor.

[0024] S1: Based on historical rainfall-type landslide samples and key factor data, a statistical model was constructed using logistic regression for calculation, and a Gaussian membership function was used to perform slope correction. The mean of the Gaussian membership function is related to the average internal friction angle of the regional soil and rock mass, and the standard deviation is related to the statistical dispersion of the regional landslide points in the slope dimension, thus obtaining the corrected basic probability.

[0025] Historical rainfall-induced landslide samples refer to a collection of records of past landslide events within a target area that were clearly induced by precipitation. For example, a historical rainfall-induced landslide sample may include specific records of each landslide event in the target area within the past ten years, such as the exact time of occurrence, precise geographical coordinates, size of the landslide, cumulative rainfall prior to the landslide, and the amount of short-duration heavy rainfall on the day of the landslide.

[0026] In some embodiments, the processor can obtain historical rainfall-induced landslide sample data by calling the geological disaster monitoring database. Specifically, it can retrieve the historical disaster ledger of the geological department, compare it with the precipitation observation records of meteorological stations during the same period, extract detailed records of slope instability events caused by rainfall, and store them in a structured manner to obtain historical rainfall-induced landslide samples.

[0027] Key factor data refers to the set of core natural environmental variables that influence the probability of rainfall-induced landslides. For example, key factor data may include hourly rainfall data from meteorological stations, surface slope values ​​extracted based on digital elevation models, and deformation characteristic data such as the annual average surface deformation rate obtained from radar satellite image analysis.

[0028] In some embodiments, the processor can obtain key factor data through a multi-source data fusion module. Specifically, it extracts real-time and forecast precipitation sequences from a meteorological forecasting platform, reads the slope attributes of the target grid from a geographic information system, processes satellite image sequences using synthetic aperture radar interferometry time series analysis technology to extract the previous cumulative deformation values ​​of the land surface, and finally aligns the above data into the same spatial grid to obtain key factor data.

[0029] A statistical model refers to a data-driven logistic regression framework used to uncover the statistical associations between historical landslide events and environmental impact factors. For example, a statistical model may include a logistic regression network with effective rainfall parameters as independent variables and landslide occurrence probability as the dependent variable, but it does not include the specific physical mechanisms of landslide occurrence and the geotechnical calculation process.

[0030] In some embodiments, the processor can obtain a statistical model through the maximum likelihood estimation method. Specifically, it extracts effective rainfall parameters from historical rainfall-induced landslide samples as input features, uses whether a landslide has occurred as a label, and determines the regression coefficients of each influencing factor by iteratively fitting the model parameters, thereby establishing a statistical mapping relationship of the initial probability of rainfall-induced landslides and obtaining a statistical model.

[0031] A Gaussian membership function is a bell curve correction mechanism used to characterize the unimodal distribution pattern of landslide frequency, which initially increases and then decreases with increasing surface slope. For example, a Gaussian membership function may include a mean parameter to control the center position of the curve and a standard deviation parameter to control the width of the curve distribution. This function avoids artificially set piecewise linear rules or step abrupt changes.

[0032] In some embodiments, the processor can obtain a Gaussian membership function by combining regional geological statistical characteristics. Specifically, it can use the slope distribution peaks at the locations of historical landslide points in the region to calibrate the mean of the function, and use the average internal friction angle of the regional soil and rock mass to verify the physical upper limit constraint of the mean. It can also calculate the statistical dispersion of the regional landslide sample points in the slope dimension to determine the standard deviation of the function, thus obtaining the Gaussian membership function.

[0033] The regional average internal friction angle of soil and rock mass refers to a macroscopic mechanical statistical index of the ability of soil or rock materials constituting a slope within a target area to resist shear failure. For example, the regional average internal friction angle of soil and rock mass can include the average shear strength parameters of typical slippery strata such as silty clay and gravelly soil within a specific area. In a physical sense, this index corresponds to the critical slope value for maintaining stability of a homogeneous infinite slope.

[0034] In some embodiments, the processor can obtain the average internal friction angle data of regional soil and rock masses through field exploration and laboratory tests. Specifically, undisturbed soil and rock samples are drilled from typical strata in the target area, and shear tests are conducted on the soil samples in a laboratory environment using a direct shear apparatus or a triaxial compression apparatus with different normal stresses. The shear stress at the time of soil failure is recorded, thereby determining and calculating the average internal friction angle of the soil and rock masses in the area.

[0035] The statistical dispersion of landslide points in the slope dimension refers to the degree of divergence of the surface slope values ​​of all historical landslide locations within a target area from their average slope value. For example, the statistical dispersion of landslide points in the slope dimension can include the slope variance or standard deviation of all recorded points in a historical landslide sample set, used to quantify the concentration or dispersion of the slope range of landslide occurrences.

[0036] In some embodiments, the processor can obtain statistical dispersion data of landslide points in the slope dimension through spatial statistical analysis. Specifically, it can extract the specific coordinates of all historical rainfall-type landslide samples, read the corresponding slope values ​​on the digital elevation model to form a slope set, calculate the arithmetic mean of the set, calculate the sum of squares of the difference between the slope of each sample and the mean, and then obtain the standard deviation as a statistical dispersion index.

[0037] The corrected baseline probability refers to the landslide risk benchmark value obtained after calibration with geological environmental constraints, based on the initial landslide probability output by the statistical model. For example, the corrected baseline probability may include a comprehensive probability value that integrates rainfall driving factors and slope topographic constraints, and its magnitude reflects the historical statistical risk level without considering real-time physical hydrological infiltration processes.

[0038] In some embodiments, the processor can obtain the corrected base probability data through matrix multiplication. Specifically, it receives the initial landslide probability calculated by the statistical model based on rainfall data, extracts the slope data of the target grid and inputs it into the constructed Gaussian membership function to calculate the slope correction coefficient, multiplies the initial landslide probability with the slope correction coefficient, and outputs the final corrected base probability value.

[0039] In some embodiments, the processor can select effective rainfall and slope as key factor data based on historical rainfall-type landslide samples, and use logistic regression to establish a statistical model for calculation to obtain the initial probability of landslide; calibrate the mean of the Gaussian membership function based on the peak slope distribution of historical landslide points in the statistical area, set an upper bound constraint on the mean based on the average internal friction angle of the regional soil and rock mass, and determine the standard deviation based on the statistical dispersion of the regional landslide sample points in the slope dimension to construct the Gaussian membership function; and use the Gaussian membership function to perform slope correction on the initial probability of landslide to obtain the corrected base probability.

[0040] The initial probability of a landslide refers to a preliminary prediction of slope instability based solely on historical rainfall patterns and current meteorological conditions, without calibration for other factors such as topography. For example, the initial probability of a landslide can include a continuous floating-point number between zero and one, output by a logistic regression network, reflecting the statistical probability that precipitation alone could trigger a disaster.

[0041] In some embodiments, the processor can obtain initial landslide probability data through logistic regression model calculations. Specifically, the forecast rainfall on the warning day and the actual cumulative rainfall of the previous few days are converted into effective rainfall parameters, input into a logistic regression equation with pre-fitted regression coefficients, and the output of the exponential function is used as the initial landslide probability. The expression for the initial landslide probability is: ; in, This represents the initial probability of a landslide. This represents the effective rainfall parameters extracted and converted from historical rainfall-type landslide samples. Represents the first regression coefficient. The second regression coefficient is represented by the first and second regression coefficients, which are determined by fitting the maximum likelihood estimate of historical rainfall-type landslide samples.

[0042] In some embodiments, the expression for the corrected base probability is: ; ; in, This represents the corrected base probability. This represents the initial probability of a landslide calculated based on historical rainfall-type landslide samples. This represents the slope correction coefficient for Gaussian membership functions. Represents slope data. This represents the mean of the Gaussian membership function, calibrated based on the peak slope distribution of historical landslide points in the statistical region. It represents the standard deviation determined based on the statistical dispersion of regional landslide sample points in the slope dimension.

[0043] S2: Based on high-resolution soil moisture data and meteorological data, a mechanistic model that includes hydrological and landslide dynamic models is used for coupled calculation. In the coupled calculation, bilinear spatiotemporal interpolation with topographic moisture index correction factor and boundary flux smoothing algorithm are used to handle spatiotemporal scale differences and obtain the landslide safety factor.

[0044] Meteorological data refers to a collection of data that reflects the state and variation patterns of atmospheric precipitation in a target area. For example, meteorological data can include real-time meteorological monitoring data collected by automatic weather stations, as well as numerical weather forecast data issued by meteorological departments.

[0045] In some embodiments, the processor can acquire meteorological data through a multi-channel data interface, specifically by receiving real-time rain gauge feedback monitoring data through a meteorological dedicated line interface, and simultaneously calling numerical weather forecast grid data from the National Meteorological Center, and obtaining meteorological data after comprehensive processing.

[0046] Soil moisture data refers to a set of physical quantities that characterize the moisture content in the surface and deep soil layers of a slope. For example, soil moisture data can include volumetric water content measured in real time by soil moisture sensors buried at meteorological observation stations, as well as large-scale soil moisture raster products generated by meteorological satellite remote sensing.

[0047] In some embodiments, the processor can acquire soil moisture data through multi-channel fusion. Specifically, it can receive the timed measured values ​​transmitted back by soil moisture sensors from various automatic weather stations in the region via a communication interface, download soil moisture inversion image files released by the remote sensing satellite center, and parse both into a standard data matrix for subsequent assimilation and calculation.

[0048] A mechanistic model is a numerical computation framework built upon the physical laws of fluid mechanics and geotechnical mechanics to simulate the rainfall infiltration process and its dynamic impact on slope stability. For example, a mechanistic model may include a cascaded combination of a hydrological model component and a landslide dynamic model component. The former is used to calculate the confluence of water at the surface and the vertical infiltration distribution underground, while the latter is used to calculate the slope's safety factor based on the infiltration results.

[0049] In some embodiments, the processor can obtain the mechanism model through unidirectional coupling of modes. Specifically, it deploys a hydrological model module and a landslide dynamics model module, uses the high-resolution soil moisture field data output by the hydrological model as the hydrological boundary condition, and inputs it into the landslide dynamics model to update the pore water pressure distribution of the soil, thereby completing the coupling of physical mechanisms and obtaining the mechanism model.

[0050] The topographic moisture index correction factor is a spatial adjustment parameter used to adjust the weighting coefficients of bilinear interpolation, ensuring that the spatial distribution of soil moisture after downscaling maintains physical consistency with local runoff and water runoff characteristics. For example, the topographic moisture index correction factor may include a multiplicative adjustment coefficient that assigns higher weights to valley areas and lower weights to ridge areas.

[0051] In some embodiments, the processor can obtain the terrain humidity index correction factor through spatial terrain analysis. Specifically, it reads high-resolution digital elevation model data, calculates the catchment area and local slope of each grid cell, calculates the terrain humidity index based on the logarithm of the ratio of the catchment area to the slope tangent, and then normalizes it into an adjustment weight to obtain the terrain humidity index correction factor.

[0052] The landslide safety factor is a quantitative mechanical index that measures the relative magnitude of the slope's resistance to sliding and the sliding force. For example, the landslide safety factor can include dimensionless values ​​where a value greater than 1 indicates physical stability, a value equal to 1 indicates a critical state, and a value less than 1 indicates instability and failure.

[0053] In some embodiments, the processor can calculate the landslide safety factor through a dynamic model. Specifically, it substitutes the soil cohesion, internal friction angle, slope, landslide body weight, and hydrostatic pressure updated by hydrological boundary conditions into the slope stability limit equilibrium equation for solution to obtain the landslide safety factor.

[0054] In some embodiments, the processor may implement S2 based on the following steps.

[0055] S21: Based on high-resolution soil moisture data driven by meteorological data and output from hydrological models, in the coupled calculation using a mechanism model that includes hydrological and landslide dynamic models, bilinear spatiotemporal interpolation with a topographic moisture index correction factor and boundary flux smoothing algorithm are used to process the spatiotemporal scale differences between the hydrological and landslide dynamic models, thus obtaining continuous soil moisture field data.

[0056] Soil moisture field data refers to a soil moisture state matrix continuously distributed on a three-dimensional spatial grid of the target area. For example, soil moisture field data can include a high-resolution spatiotemporal moisture tensor that is perfectly matched to the computational grid of the landslide dynamic model after downscaling, temporal interpolation, and spatial smoothing.

[0057] In some embodiments, the processor can obtain soil moisture field data through data interpolation and filtering. Specifically, it receives discrete moisture data after time interpolation, uses a diffusion filter operator to smooth the water flux at the grid boundary to eliminate abrupt changes, and obtains soil moisture field data.

[0058] In some embodiments, the processor can drive high-resolution soil moisture data output from a hydrological model based on meteorological data. In the coupled calculation using a mechanistic model that includes both the hydrological model and the landslide dynamics model, a bilinear interpolation method with a topographic moisture index correction factor is used for spatial downscaling to obtain downscaled soil moisture data. Based on the downscaled soil moisture data, a first-order linearized approximation equation is used for temporal interpolation to obtain temporally interpolated soil moisture data. Based on the temporally interpolated soil moisture data, an anisotropic diffusion filter operator is used for boundary flux smoothing to handle the spatiotemporal scale differences between the hydrological model and the landslide dynamics model, resulting in continuous soil moisture field data.

[0059] Downscaled soil moisture data refers to moisture values ​​obtained by converting coarse spatial resolution hydrological model outputs onto a fine spatial resolution grid and adding topographic physical constraints. For example, downscaled soil moisture data may include soil moisture content values ​​converted from one-kilometer resolution to thirty-meter resolution, reflecting the actual effects of micro-topographic undulations.

[0060] In some embodiments, the processor can obtain downscaled soil moisture data through bilinear interpolation. Specifically, it reads the coarse grid moisture data output by the hydrological model, introduces a topographic moisture index correction factor to correct the distance weight of the traditional bilinear interpolation, and redistributes the moisture values ​​of each grid to obtain downscaled soil moisture data.

[0061] Temporally interpolated soil moisture data refers to a sequence of soil moisture states with higher temporal resolution, simulated and calculated between two discrete output time steps of a hydrological model. For example, temporally interpolated soil moisture data can include vertical moisture distribution profiles that encrypt hourly output moisture data to the minute level.

[0062] In some embodiments, the processor can obtain time-interpolated soil moisture data through numerical simulation. Specifically, it uses the coarse-resolution moisture of adjacent time steps as initial and boundary conditions, and employs the first-order linearized approximation of the Richards equation to perform differential solution on the vertical redistribution process of soil moisture to obtain time-interpolated soil moisture data.

[0063] An anisotropic diffusion filter operator is a spatial smoothing matrix used to suppress abrupt changes in numerical values ​​perpendicular to the slope while keeping the water gradient constant along the slope direction. For example, an anisotropic diffusion filter operator can include a 3x3 convolution kernel with the largest weight at the center, a larger weight along the contour line direction, and a smaller weight perpendicular to the contour line direction.

[0064] In some embodiments, the processor can obtain an anisotropic diffusion filter operator through matrix construction. Specifically, it reads the surface slope aspect attribute of the target grid, sets diffusion coefficients in different directions according to the slope aspect, assembles a smooth weight matrix with direction selectivity, and obtains the anisotropic diffusion filter operator.

[0065] S22: Based on continuous soil moisture field data, the landslide safety factor is calculated using the landslide dynamic model in the mechanism model.

[0066] In some embodiments, the processor can acquire hydrostatic pressure data based on continuous soil moisture field data; based on the hydrostatic pressure data, combined with landslide body weight data, slope data, soil internal friction angle data, soil cohesion data, and slope length data, the processor can calculate the landslide safety factor using the landslide dynamic mode in the mechanism model.

[0067] Hydrostatic pressure data refers to the hydrodynamic load generated by pore water within soil, which attempts to reduce the effective stress of the soil and promote sliding. For example, hydrostatic pressure data can include excess pore water pressure values ​​formed at potential sliding surfaces after rainfall infiltration causes a rise in the groundwater level.

[0068] In some embodiments, the processor can obtain hydrostatic pressure data by solving hydrophysical equations. Specifically, it reads continuous soil moisture field data, calculates the vertical infiltration depth of pore water and the thickness of the saturated zone, and multiplies it by the unit weight parameter of water to obtain hydrostatic pressure data.

[0069] Landslide self-weight data refers to the downward and outward gravitational loads exerted on a potential sliding soil mass by Earth's gravity. For example, landslide self-weight data can include the product of the landslide soil volume and the soil's natural unit weight.

[0070] In some embodiments, the processor can obtain landslide body self-weight data by calculating spatial geometry and physical parameters. Specifically, it reads the burial depth data and slope length data of the potential sliding surface, calculates the soil volume, and multiplies it by the soil unit weight determined by geological survey to obtain the landslide body self-weight data.

[0071] Slope data refers to the angle between a slope surface and the horizontal plane. For example, slope data can include the slope inclination angle expressed in degrees.

[0072] In some embodiments, the processor can obtain slope data through terrain elevation difference, specifically by reading the elevation values ​​of the target grid and its neighboring grids in the digital elevation model, calculating the maximum elevation change rate and its inverse trigonometric function, and obtaining the slope data.

[0073] Internal friction angle data refers to the strength index of soil particles against shear deformation due to friction and dilatation. For example, internal friction angle data can include angle values ​​reflecting the surface roughness and arrangement of soil particles, determined by direct shear tests.

[0074] In some embodiments, the processor can obtain soil internal friction angle data by reading a preset geological parameter library. Specifically, it can obtain soil internal friction angle data by matching and extracting the corresponding standard internal friction angle parameters from the geological attribute database according to the geological lithology classification of the target grid.

[0075] Soil cohesion data refers to the attractive forces between soil particles due to physicochemical interactions. For example, soil cohesion data can include the shear strength intercept values ​​in fine-grained soils caused by water film bonding and electrical charge.

[0076] In some embodiments, the processor can obtain soil cohesion data by reading a preset geological parameter library. Specifically, based on the soil type of the target grid, the processor queries and extracts the cohesion measurement values ​​archived in the regional geological survey report to obtain soil cohesion data.

[0077] Slope length data refers to the physical distance a slope extends along its direction of inclination. For example, slope length data can include the straight-line distance from the top of the slope to the bottom.

[0078] In some embodiments, the processor can obtain slope length data through geometric projection calculation. Specifically, it reads the horizontal width data and slope data of the target grid, performs projection transformation calculation using the cosine theorem in trigonometric functions, and obtains the slope length data.

[0079] In some embodiments, the expression for the landslide safety factor is: ; in, Indicates the landslide safety factor. This represents soil cohesion data. This represents the slope length data. This represents the self-weight data of the landslide body. Represents slope data. This represents hydrostatic pressure data. This represents the friction angle data within the soil.

[0080] S3: Based on the landslide safety factor, the rainfall infiltration lag effect, and the cumulative rate of surface deformation in the early stage, the dynamic weighting coefficient is calculated by the nonlinear lag suppression function and the deformation acceleration amplification function. The corrected basic probability is then corrected using the dynamic weighting coefficient to obtain the final landslide probability.

[0081] The rainfall infiltration lag effect refers to the time lag required for the infiltration front generated by rainfall to reach the potential sliding surface from the land surface. For example, the rainfall infiltration lag effect can include a time quantity in minutes, represented by the ratio of the current soil volumetric water content to the saturated hydraulic conductivity.

[0082] In some embodiments, the processor can obtain the rainfall infiltration lag effect by division operation. Specifically, it extracts the soil volumetric water content data at the current moment from the hydrological model simulation output, extracts the soil saturated hydraulic conductivity data, divides the two to calculate the lag time index, and obtains the rainfall infiltration lag effect.

[0083] The cumulative rate of surface deformation in the early stages refers to the relative approximation between the actual deformation of a slope within a certain time window and the critical deformation of historical landslide failure. For example, the cumulative rate of surface deformation in the early stages can include the dimensionless ratio of the cumulative deformation over the past thirty days to the critical deformation threshold.

[0084] In some embodiments, the processor can obtain the cumulative rate of early surface deformation by division operation. Specifically, it reads the early cumulative deformation data obtained by satellite remote sensing inversion, extracts the critical deformation threshold obtained by historical statistics, calculates the ratio of the two, and obtains the cumulative rate of early surface deformation.

[0085] Hysteresis suppression functions are mathematical models used to characterize the nonlinear reduction of landslide risk weights when rainfall infiltration is insufficient. For example, ... Figure 3 As shown, the hysteresis suppression function can include a Sigmoid-type smooth curve centered on the hysteresis response time threshold.

[0086] In some embodiments, the processor can obtain the hysteresis suppression function through function mapping. Specifically, it inputs the rainfall infiltration hysteresis effect, reads the empirical parameters of the first region obtained by inversion and the set response time threshold, substitutes them into the preset natural base exponential function equation for solution, and obtains the value of the hysteresis suppression function.

[0087] The deformation acceleration amplification function is a mathematical model used to linearly amplify the risk weights when the slope deformation exceeds the critical state and enters the accelerated creep period. For example, ... Figure 4 As shown, the deformation acceleration amplification function can include a piecewise linear increasing curve with a set critical starting point.

[0088] In some embodiments, the processor can obtain the deformation acceleration amplification function through piecewise linear calculation. Specifically, it inputs the early-stage cumulative deformation rate of the land surface, determines whether it is greater than the set deformation acceleration critical threshold, and if it is greater, calculates the difference amplification factor in combination with the empirical parameters of the second region; otherwise, it keeps it as one and obtains the value of the deformation acceleration amplification function.

[0089] Dynamic weighting coefficients are scaling factors used to adjust basic probabilities, taking into account the basic physical state, hydrological infiltration process, and deformation evolution stage. For example, dynamic weighting coefficients can include the floating-point product of basic weights after hysteresis suppression and deformation amplification adjustments.

[0090] In some embodiments, the processor can obtain dynamic weight coefficients through a multiplication operation. Specifically, it reads the set base weights, reads the calculated hysteresis suppression function value and deformation acceleration amplification function value, and multiplies the three to obtain the dynamic weight coefficients.

[0091] The final landslide probability refers to the probability of a landslide occurring after correction for physical hydrological mechanisms and real-time deformation state constraints. For example, the final landslide probability can include a posterior probability between zero and one, obtained by updating the statistical baseline probability using Bayes' theorem.

[0092] In some embodiments, the processor can obtain the final landslide probability through Bayesian update operation, specifically by inputting the corrected base probability and dynamic weight coefficients, updating the weight coefficients as a likelihood function according to the Bayesian correction formula, and obtaining the final landslide probability.

[0093] In some embodiments, the processor can determine the basic weights based on the threshold range of the landslide safety factor; calculate the rainfall infiltration lag effect using the ratio of soil volumetric water content to soil saturated hydraulic conductivity; calculate the surface deformation accumulation rate using the ratio of the accumulated deformation in the previous period to the critical deformation threshold before the occurrence of historical landslide events; process the rainfall infiltration lag effect using a nonlinear lag suppression function and process the surface deformation accumulation rate using a deformation acceleration amplification function, and combine the processing results with the basic weights to calculate the dynamic weight coefficients; and use the dynamic weight coefficients to perform Bayesian correction on the corrected basic probability to obtain the final landslide probability.

[0094] The basic weight refers to the initial correction coefficients allocated solely based on the physically stable state range in which the landslide safety factor lies. For example, the basic weights may include an amplification factor set for the instability range and a reduction factor set for the stable range.

[0095] In some embodiments, the processor can obtain the basic weight through interval threshold matching, specifically by reading the landslide safety factor, determining the numerical interval in which it falls, and matching it with a preset mapping rule table to obtain the basic weight.

[0096] Soil volumetric water content refers to the proportion of liquid water in a unit volume of soil. For example, soil volumetric water content can include the numerical state that 30% of the pores in the topsoil are occupied by water.

[0097] In some embodiments, the processor can obtain soil volumetric water content data through hydrological simulation. Specifically, it reads forced meteorological precipitation data, drives the hydrological model to calculate the water generation, runoff, and infiltration process on the land surface, outputs the volumetric water content status of each grid node, and obtains soil volumetric water content data.

[0098] Saturated hydraulic conductivity of soil refers to the rate at which water seeps vertically downwards under the influence of gravity when the soil pores are completely filled with water. For example, saturated hydraulic conductivity of soil can include the permeability coefficient of a specific silty clay at 0.04 cm / min.

[0099] In some embodiments, the processor can obtain soil saturated hydraulic conductivity data by reading a preset geological parameter library. Specifically, based on the soil type map of the target area, the processor retrieves and extracts the physical parameters recorded in the laboratory double-ring permeability test to obtain soil saturated hydraulic conductivity data.

[0100] Previous cumulative deformation refers to the total displacement of the slope surface within a specified observation time window. For example, previous cumulative deformation can include millimeter-level subsidence or slip distance of the target grid along the line of sight over the past thirty days.

[0101] In some embodiments, the processor can obtain the previous cumulative deformation data through radar interferometry analysis. Specifically, it collects multi-period synthetic aperture radar images of the target area, performs phase unwrapping and atmospheric delay correction, accumulates the deformation phase within a specified time period and converts it into spatial displacement to obtain the previous cumulative deformation data.

[0102] The critical deformation threshold refers to the displacement limit value corresponding to the qualitative change of a slope from the constant creep stage to the accelerated failure stage. For example, the critical deformation threshold may include the 5.5 mm cumulative deformation standard that a certain type of landslide generally reaches before historical instability.

[0103] In some embodiments, the processor can obtain critical deformation threshold data through historical inversion statistics. Specifically, it can extract the deformation time series curve of the area before the landslide event, identify the time point when the curvature of the curve changes abruptly, extract the corresponding cumulative displacement, and perform average calculation to obtain the critical deformation threshold.

[0104] In some embodiments, the final landslide probability is expressed as: ; ; ; ; ; ; in, Indicates the final probability of a landslide. This represents the corrected base probability. Indicates dynamic weighting coefficients. Indicates the basic weight. This represents the hysteresis suppression function, which represents the nonlinearity. This represents the deformation acceleration amplification function. This indicates the amount of rainfall infiltration lag effect. This represents the soil volumetric water content data at the current moment, output by the hydrological model simulation. This represents the saturated hydraulic conductivity data of soil obtained from tests on the physical properties of soil and rock. This represents the first regional empirical parameters obtained based on the inversion of the relationship between historical rainfall and landslide response in the region. This represents the threshold for the delayed response time of rainfall infiltration, set based on regional experience. It represents the cumulative rate of early surface deformation. This represents the cumulative deformation variables obtained from previous periods based on time series analysis. This represents the critical deformation threshold obtained based on statistics of historical landslide events. This represents the second regional empirical parameter obtained by inversion based on the historical landslide deformation evolution patterns in the region. This represents the critical threshold for deformation acceleration, set based on regional experience.

[0105] S4: Based on the final landslide probability, classify the risk warning level, obtain the landslide meteorological risk warning result, and complete the landslide meteorological risk warning.

[0106] Risk warning levels refer to standardized warning levels formed by classifying and categorizing the probability of eventual landslide occurrence. For example, risk warning levels may include four danger levels: blue, yellow, orange, and red, corresponding to different probability ranges.

[0107] In some embodiments, the processor can obtain risk warning levels by thresholding. Specifically, it reads the final landslide probability value and compares it with the landslide disaster probability grading standard range set by national or local regulations to obtain the risk warning level. For example, the warning level can be divided according to the final probability Pf: 0.3 ≤ Pf < 0.38: Some risk (blue warning); 0.38 ≤ Pf < 0.5: High risk (yellow warning); 0.5 ≤ Pf < 0.7: High risk (orange warning); Pf ≥ 0.7: Very high risk (red warning).

[0108] Landslide meteorological risk warning results refer to comprehensive output information that combines specific spatial location, forecast time, and corresponding risk warning level. For example, landslide meteorological risk warning results may include data packets containing latitude and longitude coordinates, warning trigger time, and a red warning label.

[0109] In some embodiments, the processor can obtain landslide meteorological risk warning results through data encapsulation. Specifically, it can extract the geographic coordinates of the target grid, merge the current system timestamp with the risk warning level generated, and package the data according to the standard communication protocol format to obtain the landslide meteorological risk warning results.

[0110] In some embodiments, the range or constraints for the constant parameters and thresholds involved in the above formulas are as follows: the mean of the Gaussian membership function. The average internal friction angle of the regional soil and rock mass As an upper bound constraint, the preferred option is to satisfy... .like Figure 5 As shown, the preferred threshold for determining and assigning basic weights is the lower limit threshold of the safety factor. Values Upper limit threshold Values Correspondingly, the first basic weight Values The third basic weight Values In a specific application scenario, set , , , The first region empirical parameter in the hysteresis suppression function The range of values ​​is Rainfall infiltration hysteresis response time threshold The range of values ​​is Minutes. Empirical parameters of the second region in the deformation acceleration amplification function. The range of values ​​is Deformation acceleration critical threshold The range of values ​​is .

[0111] It should be noted that different embodiments may produce different beneficial effects. In different embodiments, the beneficial effects may be any one or a combination of the above, or any other possible beneficial effects.

Claims

1. A landslide meteorological risk early warning method combining statistics and mechanism model, characterized in that, include: S1: Based on historical rainfall-type landslide samples and key factor data, a statistical model was constructed using logistic regression to perform calculations, and a Gaussian membership function was used to perform slope correction. The mean of the Gaussian membership function is related to the average internal friction angle of the regional soil and rock mass, and the standard deviation is related to the statistical dispersion of the regional landslide points in the slope dimension, thus obtaining the corrected basic probability. S2: Based on high-resolution soil moisture data and meteorological data, a mechanism model including hydrological model and landslide dynamic model is used for coupled calculation. In the coupled calculation, bilinear spatiotemporal interpolation with topographic moisture index correction factor and boundary flux smoothing algorithm are used to handle spatiotemporal scale differences and obtain landslide safety factor. S3: Based on the landslide safety factor, rainfall infiltration lag effect, and the cumulative rate of surface deformation in the early stage, the dynamic weighting coefficient is calculated by nonlinear lag suppression function and deformation acceleration amplification function. The dynamic weighting coefficient is then used to perform Bayesian correction on the corrected base probability to obtain the final landslide probability. S4: Based on the final landslide probability, classify the risk warning level, obtain the landslide meteorological risk warning result, and complete the landslide meteorological risk warning.

2. The landslide meteorological risk early warning method based on the fusion of statistical and mechanistic models according to claim 1, characterized in that, S1 includes: Based on historical rainfall-type landslide samples, effective rainfall and slope were selected as key factor data. A statistical model was established using logistic regression to calculate the initial probability of landslides. The mean of the Gaussian membership function is calibrated based on the peak slope distribution of historical landslide points in the statistical region, and the upper bound constraint of the mean is set based on the average internal friction angle of the soil and rock mass in the region. The standard deviation is determined based on the statistical dispersion of the landslide sample points in the slope dimension, and the Gaussian membership function is constructed. The initial probability of the landslide is corrected by applying a Gaussian membership function to obtain the corrected base probability.

3. The method according to claim 1, characterized in that, The expression for the corrected base probability is: ； ； in, This represents the corrected base probability. This represents the initial probability of a landslide calculated based on historical rainfall-type landslide samples. This represents the slope correction coefficient for Gaussian membership functions. Represents slope data. This represents the mean of the Gaussian membership function, calibrated based on the peak slope distribution of historical landslide points in the statistical region. It represents the standard deviation determined based on the statistical dispersion of regional landslide sample points in the slope dimension.

4. The landslide meteorological risk early warning method based on the fusion of statistical and mechanistic models according to claim 1, characterized in that, S2 includes: S21: Based on the high-resolution soil moisture data output by the hydrological model driven by meteorological data, in the coupled calculation using a mechanism model that includes the hydrological model and the landslide dynamic model, the bilinear spatiotemporal interpolation algorithm with the introduction of the topographic moisture index correction factor and the boundary flux smoothing algorithm are used to process the spatiotemporal scale differences between the hydrological model and the landslide dynamic model, so as to obtain continuous soil moisture field data. S22: Based on continuous soil moisture field data, the landslide safety factor is calculated using the landslide dynamic model in the mechanism model.

5. The landslide meteorological risk early warning method based on the fusion of statistical and mechanistic models according to claim 4, characterized in that, S21 includes: Based on high-resolution soil moisture data driven by meteorological data-driven hydrological model output, in the coupled calculation using a mechanism model that includes hydrological model and landslide dynamic model, a bilinear interpolation method with topographic moisture index correction factor is used for spatial downscaling to obtain downscaled soil moisture data. Based on the downscaled soil moisture data, time interpolation was performed using a first-order linearized approximation equation to obtain the time-interpolated soil moisture data. Based on the time-interpolated soil moisture data, the boundary flux smoothing process is performed using the anisotropic diffusion filter operator to handle the spatiotemporal scale differences between the hydrological model and the landslide dynamic model, thereby obtaining continuous soil moisture field data.

6. The landslide meteorological risk early warning method based on the fusion of statistical and mechanistic models according to claim 4, characterized in that, S22 includes: Hydrostatic pressure data are obtained based on continuous soil moisture field data. Based on hydrostatic pressure data, combined with landslide body weight data, slope data, soil internal friction angle data, soil cohesion data, and slope length data, the landslide safety factor is calculated using the landslide dynamic mode in the mechanism model.

7. The method according to claim 6, characterized in that, The expression for the landslide safety factor is: ； in, Indicates the landslide safety factor. This represents soil cohesion data. This represents the slope length data. This represents the self-weight data of the landslide body. Represents slope data. This represents hydrostatic pressure data. This represents the friction angle data within the soil.

8. The landslide meteorological risk early warning method based on the fusion of statistical and mechanistic models according to claim 1, characterized in that, S3 includes: The basic weights are determined based on the threshold range in which the landslide safety factor falls; The rainfall infiltration retardation effect is calculated by using the ratio of soil volumetric water content to soil saturated hydraulic conductivity. The cumulative deformation rate of the land surface in the early stage is calculated by using the ratio of the cumulative deformation in the early stage to the critical deformation threshold before the occurrence of historical landslide events. The hysteresis suppression function is used to process the hysteresis effect of rainfall infiltration, and the deformation acceleration amplification function is used to process the cumulative rate of surface deformation in the early stage. The processing results are combined with the basic weight to calculate the dynamic weight coefficient. The corrected base probability is corrected using dynamic weighting coefficients to obtain the final landslide probability.

9. The method according to claim 1, characterized in that, The expression for the final landslide probability is: ； ； ； ； ； ； in, Indicates the final probability of a landslide. This represents the corrected base probability. Indicates the dynamic weighting coefficient. Indicates the basic weight. This represents the hysteresis suppression function, which represents the nonlinearity. This represents the deformation acceleration amplification function. This indicates the amount of rainfall infiltration lag effect. This represents the soil volumetric water content data at the current moment, output by the hydrological model simulation. This represents the saturated hydraulic conductivity data of soil obtained from tests on the physical properties of soil and rock. This represents the first regional empirical parameters obtained based on the inversion of the relationship between historical rainfall and landslide response in the region. This represents the threshold for the delayed response time of rainfall infiltration, set based on regional experience. It represents the cumulative rate of early surface deformation. This represents the cumulative deformation variables obtained from previous periods based on time series analysis. This represents the critical deformation threshold obtained based on statistics of historical landslide events. This represents the second regional empirical parameter obtained by inversion based on the historical landslide deformation evolution patterns in the region. This represents the critical threshold for deformation acceleration, set based on regional experience.

10. A landslide meteorological risk early warning system integrating statistical and mechanistic models, used to execute the landslide meteorological risk early warning method integrating statistical and mechanistic models as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to acquire historical rainfall-type landslide samples, meteorological data, key factor data, and high-resolution soil moisture data; The statistical model calculation module is used to construct a statistical model based on historical rainfall-type landslide samples and key factor data, and to perform slope correction using a Gaussian membership function. The mean of the Gaussian membership function is related to the average internal friction angle of the regional soil and rock mass, and the standard deviation is related to the statistical dispersion of the regional landslide points in the slope dimension, so as to obtain the corrected basic probability. The mechanism model calculation module is used to perform coupled calculations based on high-resolution soil moisture data and meteorological data, using a mechanism model that includes hydrological and landslide dynamic models. In the coupled calculation, bilinear spatiotemporal interpolation with topographic moisture index correction factor and boundary flux smoothing algorithm are used to handle spatiotemporal scale differences and obtain the landslide safety factor. The Bayesian correction module is used to calculate dynamic weighting coefficients based on the landslide safety factor, rainfall infiltration hysteresis effect, and the cumulative rate of surface deformation in the early stage. It uses a nonlinear hysteresis suppression function and a deformation acceleration amplification function to calculate the dynamic weighting coefficients and then uses the dynamic weighting coefficients to perform Bayesian correction on the corrected base probability to obtain the final landslide probability. The early warning output module is used to classify risk warning levels based on the final landslide probability, obtain landslide meteorological risk warning results, and complete the landslide meteorological risk warning.

Images