A slope hazard assessment method based on a Bayesian hierarchical spatiotemporal model

Abstract

This invention discloses a slope hazard risk assessment method based on a Bayesian hierarchical spatiotemporal model. The method utilizes a Bayesian hierarchical spatiotemporal model, based on remote sensing data, to establish a regional geological FEM model in both temporal and spatial dimensions. It then analyzes the spatiotemporal distribution and dynamic changes of the model in real time, providing a comprehensive and thorough analysis of regional environmental changes and assessing the region's geological hazard risk. This method primarily targets the temporal and spatial trends of geological hazards such as landslides. It adds a multi-level Bayesian spatiotemporal model component to the traditional evaluation model, constructing a spatiotemporal algorithm model composed of factors highly correlated with geological hazards. This model is embedded into the multi-level Bayesian hierarchical spatiotemporal model and integrated with existing geological hazard systems to obtain a multi-level, high-precision hazard risk assessment system with spatiotemporal effects.

Description

Technical Field

[0001] This invention belongs to the field of geoscience remote sensing technology and relates to a slope hazard risk assessment method, specifically a slope hazard risk assessment method based on a Bayesian hierarchical spatiotemporal model. Background Technology

[0002] Landslides, as one of the most frequent and destructive natural disasters globally, pose a significant threat to people's lives and property and hinder local economic development. Landslide displacement prediction models include not only complex physical models but also mathematical models based on monitoring datasets. Compared to physical models, mathematical models are simpler and more accurate to establish. However, landslide deformation evolution is a nonlinear dynamic process, influenced by factors such as topography, soil and rock structure, hydrogeology, climate, and human activities, exhibiting spatiotemporal correlations. However, these prediction models only consider the temporal correlation of displacement monitoring data, enabling displacement prediction at a single typical monitoring point, but neglecting the spatial correlation between monitoring points. This limits the improvement of prediction accuracy to some extent and fails to accurately determine the overall deformation trend of the landslide, thus leading to the overlooking of potential threats. Summary of the Invention

[0003] To address the lack of spatiotemporal effects in existing disaster assessment systems, this invention provides a slope hazard assessment method based on a Bayesian hierarchical spatiotemporal model. This method primarily targets the temporal and spatial variation trends of geological hazards such as landslides. It adds a multi-level Bayesian spatiotemporal model component to the traditional evaluation model, constructing a spatiotemporal algorithm model composed of factors highly correlated with geological hazards. This model is then embedded into the multi-level Bayesian hierarchical spatiotemporal model, integrating existing geological hazard systems to obtain a multi-level, high-precision disaster risk assessment system with spatiotemporal effects.

[0004] The objective of this invention is achieved through the following technical solution:

[0005] A slope hazard risk assessment method based on a Bayesian hierarchical spatiotemporal model includes the following steps:

[0006] Step 1: Study Area Division

[0007] The area to be monitored is divided into k non-overlapping 1km×1km grid regions. Satellite imagery, topography, and meteorological data of the monitored areas are collected to obtain uniform soil and climate condition indicators, surface temperature, vegetation index, etc. Displacement data are recorded continuously over time t to obtain a K×N dataset. Then:

[0008] Y(k,n) = {y(k,n)};

[0009] Step 2: Integrate sequence data and establish a regional geological information FEM model.

[0010] An FEM model is established based on satellite imagery data, topography, and meteorological data. Data preprocessing is performed on the collected time series data, including outlier removal and missing value filling.

[0011] Step 3: Construct a multi-level Bayesian spatiotemporal model

[0012] Nov is an unknown surface deformation parameter representing its potential spatiotemporal process. A Bayesian spatial model for geological hazard risk assessment is established using the natural logarithm of Nov.

[0013] log(nov) = a + A + B + δ k,n +ε

[0014] log(nov) = a + u k +v k +g n +t n +δ k,n +ε

[0015] In the formula, a is the intercept, representing the overall level; A represents the spatial effect, A = u k +v k u k Representing structural spatial effects, v k B represents the unstructured spatial effect; B represents the time effect, B = g n +t n g n Prior information representing time, t n Represents a time-dependent random effect; δ k,n This represents a spatiotemporal interaction effect, where ε is random noise.

[0016] Step 4: Establish disaster assessment levels using a multi-level Bayesian spatiotemporal model.

[0017] Step 41: Use the Monte Carlo method to sample the FEM model of the slope in the area to be tested and obtain the probability density function of the instability failure of the FEM model;

[0018] Step 42: Select WinBUGS software for data analysis, use Gibbs sampling and Metropolis algorithms to generate Markov chains; add regular constraints to core key indicators to prevent overfitting of the multi-level Bayesian spatiotemporal model.

[0019] Step 43: Use a multi-level Bayesian spatiotemporal model to predict the time series of disaster occurrences, and classify the disaster risk level of the region based on spatial hotspot areas;

[0020] Step 5: Conduct an overall assessment and early warning of the geological disaster risk level in the area to be tested.

[0021] Step 51: Design an early warning threshold based on the landslide disaster development mechanism. When the predicted surface deformation of the landslide body is greater than the threshold, a landslide disaster early warning will be issued.

[0022] Step 52: Calculate the relative risk of disasters in each region using a multi-level Bayesian spatiotemporal model, identify high-risk areas, and then display them hierarchically on the regional map. Based on the calculation results, integrate long-term regional data to achieve autonomous processing and decision-making, and issue early warnings for areas experiencing sudden or unconventional changes.

[0023] Compared with the prior art, the present invention has the following advantages:

[0024] 1. This invention combines the advantages of spatiotemporal Bayesian models in analyzing spatiotemporal data to analyze the distribution pattern and trend of disaster occurrence risk in the study area in time, space, and spatiotemporal interaction; combined with the MCMC algorithm, it integrates information from multiple sources and the uncertainty of parameters into a single model, making full use of prior information and continuously updating it with information from sampling studies; it fully considers uncertainty and overcomes the shortcomings caused by selecting small samples.

[0025] 2. This invention can meet the requirements of simulating the real damage process of geological disasters, accurately expressing the random parameters of dynamic processes, and evaluating the spatiotemporal nature of disaster levels, thereby obtaining multi-dimensional, high-precision geological disaster assessment data with rich spatial information and spatiotemporal effects. Attached Figure Description

[0026] Figure 1 This is a flowchart of the slope hazard risk assessment method based on the Bayesian hierarchical spatiotemporal model of the present invention;

[0027] Figure 2 To illustrate the relationship between the slope hazard risk assessment method based on the Bayesian hierarchical spatiotemporal model and the various modules of the autonomous decision-making system;

[0028] Figure 3 This refers to the spatiotemporal changes and relationships of data information. Detailed Implementation

[0029] The technical solution of the present invention will be further described below with reference to the accompanying drawings, but it is not limited thereto. Any modifications or equivalent substitutions to the technical solution of the present invention that do not depart from the spirit and scope of the technical solution of the present invention should be covered within the protection scope of the present invention.

[0030] This invention provides a slope hazard risk assessment method based on a Bayesian hierarchical spatiotemporal model. Utilizing this model and remote sensing data, a regional geological FEM model is established in both temporal and spatial dimensions. The model is then analyzed in real-time to assess its spatiotemporal distribution and dynamic changes, providing a comprehensive and thorough analysis of regional environmental changes and evaluating the region's geological hazard risk. Figure 1-3 As shown, the method includes the following steps:

[0031] Step 1: Study Area Division

[0032] The area to be measured was divided into multiple 1km × 1km grid regions. Environmental conditions in this area were monitored using remote sensing technology. Monitoring indicators included topography, soil conditions, meteorological data, surface temperature, and vegetation index. The slope to be measured and its surrounding area were divided into k non-overlapping spatial units, and data were recorded over a continuous time period t, resulting in a K × N dataset with K rows (spatial units) and N columns (time units).

[0033] Y(k,n)={y(k,n)}

[0034] In this step, the monitoring data (topography, soil conditions, meteorological data, surface temperature, vegetation index) comes from one or more of the following devices: remote sensing equipment, rain gauges, surface displacement monitoring stations, crack gauges, inclinometers, etc.

[0035] Step 2: Sequence data integration and FEM model establishment

[0036] An FEM model was established based on satellite imagery data, topographic data, and meteorological data. Data preprocessing was performed on the collected time-series data, including outlier removal and missing value imputation. The specific steps are as follows:

[0037] Step 2: 1. Establish an FEM model based on satellite imagery data, geological data, and meteorological data;

[0038] Step 22: Reconstruct time series data that are affected by random factors such as sensor performance and data transmission errors using the mean iterative filtering method;

[0039] Steps 2 and 3: Interpolate meteorological indicators using the spatiotemporal kriging method;

[0040] Step 24: Assuming the spatial process follows an autoregressive dynamic change, considering the correlation and variability of variables, construct a theoretical spatiotemporal variability function for fitting:

[0041]

[0042] Where y(d,t) is the theoretical spatiotemporal semivariogram with spatial distance d and temporal distance t;S (d) is the pure-space semivariogram; y T (t) is the pure time semivariogram; y ST α is the spatiotemporal semivariogram; α is the anisotropy ratio, through which time distance units are converted into spatial distance units.

[0043] Step 3: Construct a multi-level Bayesian spatiotemporal model

[0044] A multi-level Bayesian model based on Markov chain Monte Carlo simulation is analyzed by combining prior information, data information, and model information. Taking into account both temporal and spatial effects, the multi-level Bayesian spatiotemporal model is constructed as follows:

[0045] log(nov) = a + A + B + δ k,n +ε

[0046] log(nov) = a + u k +v k +g n +t n +δ k,n +ε

[0047] In the formula, a is the intercept, representing the overall level; A represents the spatial effect, A = u k +v k u k Representing structural spatial effects, v k B represents the unstructured spatial effect; B represents the time effect, B = g n +t n g n Prior information representing time, t n Represents a time-dependent random effect; δ k,n This represents a spatiotemporal interaction effect, where ε is random noise.

[0048] Spatial effects consider the impact of regional deformation on surrounding areas; prior information is expressed through conditional autoregression. ω ij =ω ji Let be the spatial adjacency coefficient between region j and region i. This can be understood as the j-th region and the i-th region being "neighbors" in a Markov random field. Let u be the set of integers before and after i. i u represents the spatial effect of the i-th region. j φ(u) represents the spatial effect of the j-th region. i -u j This is a formula to measure the spatial effect relationship between region j and region i. The spatial effect is specifically modeled as follows:

[0049]

[0050] The magnitude of the mutual influence between two spatial effects is defined as:

[0051]

[0052] Where k is an unknown constant.

[0053] Considering the potential autocorrelation of time series data, the prior distribution of time random effects parameters adopts a conditional autoregressive prior form, and the time effects are modeled as follows:

[0054] g t =γ0+γ1v t-1 +γ2v t-2 +…+γ l v t-l +u t

[0055]

[0056] Where γ0, γ1, γ2, ..., γ l u is the regression coefficient in the time-effect model. t The error can be roughly attributed to factors in the model that are independent of time, and follows a normal distribution with a mean of 0.

[0057] random effects of spatiotemporal interaction δ k,n The noise follows a Gaussian distribution.

[0058] The prior distributions of the parameters in the time-effect model, spatial-effect model, and spatiotemporal-effect model are based on uninformative priors, as follows:

[0059] The uninformative priors for the time-effects model are as follows:

[0060] π(γ i )∝1,i=0,1,…l

[0061]

[0062] Spatial effect model:

[0063] π(u i )∝1,i=0,1,2

[0064]

[0065] Spatiotemporal effect model:

[0066] π(γ i )∝1,i=0,1,…l

[0067]

[0068] π(u i )∝1,i=0,1,2

[0069]

[0070] Step 4: Establish disaster assessment levels using a multi-level Bayesian spatiotemporal model.

[0071] Step 41: Use the Monte Carlo method to sample the FEM model of the slope in the area to be tested and obtain the probability density function of the instability failure of the FEM model.

[0072] Step 42: Select WinBUGS software for data analysis, use Gibbs sampling and Metropolis algorithms to generate Markov chains; add regular constraints to core key indicators to prevent the multi-level Bayesian spatiotemporal model from overfitting.

[0073] Step 43: Use a multi-level Bayesian spatiotemporal model to predict the time series of disasters and classify the disaster risk level of the region based on spatial hotspot areas.

[0074] In this step, the method for time series prediction of disaster occurrence using a multi-level Bayesian spatiotemporal model is as follows: Based on the established multi-level Bayesian spatiotemporal model, the existing data is used as training data to train the multi-level Bayesian spatiotemporal model, and then the probability of disaster occurrence (system instability and damage) in the whole region is predicted by the multi-level Bayesian spatiotemporal model.

[0075] In this step, a multi-level Bayesian spatiotemporal model is used to estimate the posterior parameters of different factors, evaluate the contribution of different factors to instability and failure, and conduct a comprehensive effect assessment. The relative contribution rate (RC) is used to explain the actual effect of each variable on instability and failure; when RC > 1, it indicates that the variable has a promoting effect on instability and failure.

[0076] In this step, the DIC is calculated to measure the model fit. The DIC value represents the best balance between model fit and complexity; the smaller the value, the better the model fit.

[0077] In this step, the probability of disasters occurring (system instability and damage) in the entire region is regressed and predicted using a multi-level Bayesian spatiotemporal model.

[0078] Step 5: Conduct an overall assessment and early warning of the geological disaster risk level in the area to be tested.

[0079] Step 51: Design an early warning threshold based on the landslide disaster development mechanism. When the predicted surface deformation of the landslide body is greater than the threshold, a landslide disaster early warning will be issued.

[0080] Step 52: Calculate the relative risk of disasters in each region using the model, identify high-risk areas, and then display them in a hierarchical manner on the regional map. Based on the calculation results and comprehensive long-term regional data, autonomous processing and decision-making are achieved, and early warnings are issued for areas experiencing sudden or unconventional changes.

[0081] Step 53: Through a visual platform combined with real-time updated data, the early warning information is delivered to the system users in an intuitive, vivid, and graphic form using platform-based graphics and tables.

[0082] In this step, the system itself is upgraded, integrated, and optimized by combining the results of the on-site investigation and utilizing a large amount of real-time comparative data.

Claims

1. A slope hazard risk assessment method based on a Bayesian hierarchical spatiotemporal model, characterized in that... The method includes the following steps: Step 1: Study Area Division The area to be monitored is divided into k non-overlapping 1km×1km grid regions. Satellite imagery, topographic data, and meteorological data of the monitored areas are collected to obtain uniform soil and climate condition indicators, surface temperature, and vegetation index. Displacement data are recorded continuously over time t to obtain a K×N dataset. Then: Y(k,n) = {y(k,n)}; Step 2: Integrate sequence data and establish a regional geological information FEM model. An FEM model is established based on satellite imagery data, topography, and meteorological data, and the collected time-series data is preprocessed. Step 3: Construct a multi-level Bayesian spatiotemporal model Nov is an unknown surface deformation parameter representing its potential spatiotemporal process. A Bayesian spatial model for geological hazard risk assessment is established using the natural logarithm of Nov. log(nov)=a+A+B+δ k,n +e log(nov)=a+u k +v k +g n +t n +d k,n +e In the formula, a is the intercept, representing the overall level; A represents the spatial effect, A = u k +v k u k Representing structural spatial effects, v k B represents the unstructured spatial effect; B represents the time effect, B = g n +t n g n Prior information representing time, t n Represents a time-dependent random effect; δ k,n This represents a spatiotemporal interaction effect, where ε is random noise. Step 4: Establish disaster assessment levels using a multi-level Bayesian spatiotemporal model. Step 41: Use the Monte Carlo method to sample the FEM model of the slope in the area to be tested and obtain the probability density function of the instability failure of the FEM model; Step 42: Select WinBUGS software for data analysis, use Gibbs sampling and Metropolis algorithms to generate Markov chains; add regular constraints to core key indicators to prevent overfitting of the multi-level Bayesian spatiotemporal model. Step 43: Use a multi-level Bayesian spatiotemporal model to predict the time series of disaster occurrences, and classify the disaster risk level of the region based on spatial hotspot areas; Step 5: Conduct an overall assessment and early warning of the geological disaster risk level in the area to be tested. Step 51: Design an early warning threshold based on the landslide disaster development mechanism. When the predicted surface deformation of the landslide body is greater than the threshold, a landslide disaster early warning will be issued. Step 52: Calculate the relative risk of disasters in each region using a multi-level Bayesian spatiotemporal model, identify high-risk areas, and then display them hierarchically on the regional map. Based on the calculation results, integrate long-term regional data to achieve autonomous processing and decision-making, and issue early warnings for areas experiencing sudden or unconventional changes.

2. The slope hazard risk assessment method based on a Bayesian hierarchical spatiotemporal model according to claim 1, characterized in that... The specific steps of step two are as follows: Step 2:

1. Establish an FEM model based on satellite imagery data, geological data, and meteorological data; Step 22: Reconstruct time series data that are affected by sensor performance issues and random factors such as data transmission errors using the mean iterative filtering method. Steps 2 and 3: Interpolate meteorological indicators using the spatiotemporal kriging method; Step 24: Assuming the spatial process follows an autoregressive dynamic change, considering the correlation and variability of variables, construct a theoretical spatiotemporal variability function for fitting: Where y(d,t) is the theoretical spatiotemporal semivariogram with spatial distance d and temporal distance t; S (d) is the pure-space semivariogram; y T (t) is the pure time semivariogram; y ST α is the spatiotemporal semivariogram; α is the anisotropy ratio, through which time distance units are converted into spatial distance units.

3. The slope hazard risk assessment method based on a Bayesian hierarchical spatiotemporal model according to claim 1, characterized in that... In step three, the spatial effect specifically adopts the following spatial effect model: Where, ω ij =ω ji Let be the spatial adjacency coefficient between region j and region i. Let region j and region i be "neighbors" in a Markov random field. Let i be the set of integers before and after i, and u be the set of integers before and after i. i u represents the spatial effect of the i-th region. j φ(u) represents the spatial effect of the j-th region. i -u j ) is a formula for measuring the spatial effect relationship between region j and region i.

4. The slope hazard risk assessment method based on a Bayesian hierarchical spatiotemporal model according to claim 1, characterized in that... In step three, the time effect adopts the following time effect model: g t =γ0+γ1v t-1 +γ2v t-2 +…+c l v t-l +u t Where γ0, γ1, γ2, ..., γ l u is the regression coefficient in the time-effect model. t The error caused by factors unrelated to time in the time-effect model follows a normal distribution with a mean of 0.

5. The slope hazard risk assessment method based on a Bayesian hierarchical spatiotemporal model according to claim 4, characterized in that... The uninformative prior of the time-effect model is as follows: p(g i )∝1,i=0,1,…l; 6. The slope hazard risk assessment method based on a Bayesian hierarchical spatiotemporal model according to claim 3, characterized in that... The uninformed priors for the spatial effects model are as follows: π(in i )∝1,i=0,1,2; 7. The slope hazard risk assessment method based on a Bayesian hierarchical spatiotemporal model according to claim 1, characterized in that... The information-free prior of the multi-level Bayesian spatiotemporal model is as follows: p(g i )∝1,i=0,1,…l; π(in i )∝1,i=0,1,2;

Images