A regional rainfall landslide monitoring and early warning method based on multi-source data fusion

Abstract

The application provides a regional rainfall landslide monitoring and early warning method based on multi-source data fusion, comprising: constructing a landslide space-time database; acquiring multi-source data of the landslide, and adding the multi-source data into the landslide space-time database; performing correlation analysis on the landslide space-time database added with the multi-source data, and acquiring all high-correlation factors of landslide disasters; constructing a regional rainfall landslide space-time prediction model, inputting the high-correlation factors into the regional rainfall landslide space-time prediction model, and acquiring the occurrence time and spatial region of a landslide event. The application can contribute to the sustainable development of the economy in the region, especially in the effective management of geological disaster risks.

Description

Technical Field

[0001] This invention belongs to the field of geological disaster monitoring and early warning technology, and in particular relates to a regional rainfall-induced landslide monitoring and early warning method based on multi-source data fusion. Background Technology

[0002] Against the backdrop of global climate change and economic integration, rainfall-induced landslides exacerbate population and economic losses. The direct economic losses caused by landslides are showing an increasing trend year by year, with hundreds of millions of yuan spent annually on slope stabilization and post-disaster reconstruction. Landslide hazards vary across different regions, making the monitoring, prediction, and prevention of regional rainfall-induced landslides an urgent need.

[0003] Regarding the research on the mechanism of rainfall-induced landslides, Qiao Jianping et al. analyzed the spatiotemporal probability and warning probability of landslides based on the correlation between landslides and rainfall, and the initiation value of rainfall-induced landslides. They calculated the frequency of landslides that have occurred in the danger zone and the frequency of rainfall-induced landslides. Chi Yongxiang et al. analyzed the relationship between various characteristic elements of typhoon rainstorms and landslides. The results showed that when the rainfall on the day of typhoon landfall reaches the extreme value of 100 mm or 60 mm in 3 hours, it is easy to trigger a cluster of small shallow soil landslides that occur immediately after rain. Luo Yu et al. selected uniform, increasing, decreasing, and peak rainfall types as objects. Based on the rainfall intensity and groundwater relationship model proposed by Rosso, they studied the influence of different rainfall types on the groundwater level changes of shallow landslides. Then, combined with the infinite slope theory, they established a stability calculation model for shallow landslides and studied the influence of different rainfall types on the stability of shallow landslides. The results show that rainfall type has a significant impact on the stability of shallow landslides. The safety factor is lowest under increasing rainfall, followed by uniform rainfall, then peak rainfall, and highest under decreasing rainfall. These findings indicate that rainfall-induced landslides exhibit regional differences, complex causes, and complex mechanisms.

[0004] Regarding regional landslide disaster prediction models, Liu Xiangyu et al. established a GIS-based rainfall infiltration-stability evaluation model to quantitatively evaluate the stability of slopes; Bhardwaj et al. studied the role of extreme rainfall and physical factors in landslides in the Indian Himalayas, indicating that the pre-rainfall six days before extreme rainfall events had considerable depth, which may be the main cause of landslides in the region; Yin Kunlong et al. [7-8] summarized landslide disaster early warning and prediction models in recent years, believing that combining landslide disaster early warning and forecasting with risk management is a necessity for disaster reduction and prevention, and also a future trend in this field. This study also established a WebGIS-based landslide disaster information management system and a real-time early warning release system.

[0005] The above research progress, both domestically and internationally, indicates significant advancements in the understanding and prediction of regional rainfall-induced landslides. However, spatiotemporal prediction studies of geological hazards specific to regional rainfall-induced landslides still have shortcomings. These include insufficient integration of historical and real-time monitoring data, inadequate analysis of influencing factors, limited equipment indicators for slope monitoring, and a lack of spatiotemporal prediction models. Consequently, the root causes of slope instability remain unclear, the evolutionary process leading to landslides is ambiguous, and effective prevention of slope hazards is not feasible. Summary of the Invention

[0006] To address the aforementioned technical issues, this invention proposes a regional rainfall-based landslide monitoring and early warning method based on multi-source data fusion, which can contribute to the sustainable economic development of the study area, especially in the effective management of geological disaster risks.

[0007] To achieve the above objectives, this invention provides a regional rainfall-induced landslide monitoring and early warning method based on multi-source data fusion, comprising:

[0008] Construct a spatiotemporal database of landslides;

[0009] Acquire multi-source data on landslides and add the multi-source data to the landslide spatiotemporal database;

[0010] Correlation analysis was performed on the landslide spatiotemporal database containing multi-source data to obtain all highly correlated factors of landslide disasters;

[0011] A spatiotemporal prediction model for regional rainfall-induced landslides is constructed. The highly correlated factors are input into the spatiotemporal prediction model to obtain the occurrence time and spatial region of landslide events.

[0012] Optionally, constructing the landslide spatiotemporal database includes:

[0013] Obtain local landslide-related data; the landslide-related data includes: geographic information, geological information, hydrological environment, historical landslide data, and landslide hazard point data;

[0014] Construct a geospatial database;

[0015] The landslide-related data is imported into the geospatial database to obtain the landslide spatiotemporal database; the landslide spatiotemporal database supports functions based on rainfall events, spatiotemporal clustering, and attribute classification.

[0016] Optionally, acquiring the multi-source data on the landslide includes:

[0017] Construct an integrated monitoring system based on BeiDou satellite computing;

[0018] Based on the fusion monitoring system, multi-source data at key landslide locations in the study area are collected; wherein, the multi-source data includes: slope deformation rate prediction data, physical field data, and groundwater data.

[0019] Optionally, constructing the integrated monitoring system includes:

[0020] The first step is to construct a fusion algorithm. The Kalman filter algorithm is used, with the IMU model as the system state and differential positioning data based on GPS and BeiDou systems as the observations for filtering and correction. Environmental information is introduced into the fusion algorithm. By defining the sensor confidence domain, the fusion weight coefficients are assigned based on the confidence level determined by the membership function.

[0021] The second step involves incorporating multiple sensors. Based on the improved coupling framework, a GNSS / IMU fusion and high-precision positioning model is obtained. Pore water pressure, seepage pressure, geothermal, and rain gauge sensors are incorporated. This sensor set is deployed at multiple points in a self-organizing and multi-hop manner to form a unified system that integrates displacement monitoring, pore water pressure monitoring, and monitoring of other indicators calculated by BeiDou satellites, thus constructing the fusion monitoring system.

[0022] Optionally, obtaining all the highly correlated factors for landslide disasters includes:

[0023] The data factors in the landslide spatiotemporal database containing multi-source data are fused pairwise. Pearson correlation analysis is performed on the fused data factors to obtain the correlation coefficients between the data factors. The data factors with correlation coefficients higher than a preset threshold are selected as the highly correlated factors.

[0024] Optionally, the highly correlated factors include: a highly correlated factor of landslide spatial susceptibility and a highly correlated factor of landslide temporal susceptibility;

[0025] The highly relevant factors for landslide spatial susceptibility include: elevation, water content, rainfall, slope toe, slope, aspect, curvature, landform type, soil and rock type, distance from fault, vegetation, distance from river, soil erosion intensity, population density, distance from road, soil use type, and landslide occurrence events.

[0026] The highly relevant factors for landslide temporality include: daily rainfall, water content, pore water pressure, and groundwater level.

[0027] Optionally, the spatiotemporal prediction model for regional rainfall-induced landslides includes: a spatial susceptibility prediction model for landslides and a temporal prediction model for regional rainfall-induced landslides;

[0028] The landslide spatial susceptibility prediction model is used to determine the spatial region of landslide events;

[0029] The regional rainfall-induced landslide timing prediction model is used to determine the timing of landslide events.

[0030] Optionally, constructing the spatiotemporal prediction model for the regional rainfall-induced landslide includes:

[0031] A landslide spatial susceptibility prediction model is established based on support vector machines;

[0032] Based on the highly correlated factors of landslide spatial susceptibility, the landslide spatial susceptibility prediction model is trained.

[0033] A regional rainfall-induced landslide timing prediction model was constructed based on long short-term memory networks.

[0034] Based on the highly correlated factors of landslide time susceptibility, the prediction model for the time of rainfall-induced landslides in the region was trained.

[0035] The trained regional rainfall-induced landslide temporal prediction model and the trained landslide spatial susceptibility prediction model are coupled to obtain the regional rainfall-induced landslide spatiotemporal prediction model.

[0036] Compared with the prior art, the present invention has the following advantages and technical effects:

[0037] This invention integrates regional historical data and real-time monitoring data to establish a spatiotemporal database of regional rainfall-induced landslide disasters. It addresses the challenge that susceptibility models cannot predict the stability of rainfall-induced landslide time series, improving the accuracy and reliability of spatiotemporal analysis of rainfall-induced landslides and enabling accurate and effective prediction of when and where landslides will occur. This can contribute to the sustainable economic development of the study area, particularly in the effective management of geological hazard risks. Attached Figure Description

[0038] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0039] Figure 1 This is a schematic diagram of the regional rainfall-induced landslide monitoring and early warning method according to an embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram illustrating the working principle of the spatiotemporal prediction model for regional rainfall-induced landslides in an embodiment of the present invention. Detailed Implementation

[0041] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0042] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0043] This invention proposes a regional rainfall-induced landslide monitoring and early warning method based on multi-source data fusion, comprising:

[0044] Construct a spatiotemporal database of landslides;

[0045] Acquire multi-source data on landslides and add the multi-source data to the landslide spatiotemporal database;

[0046] Correlation analysis was performed on the landslide spatiotemporal database containing multi-source data to obtain all highly correlated factors of landslide disasters;

[0047] A spatiotemporal prediction model for regional rainfall-induced landslides is constructed. The highly correlated factors are input into the spatiotemporal prediction model to obtain the occurrence time and spatial region of landslide events.

[0048] Furthermore, constructing the landslide spatiotemporal database includes:

[0049] Obtain local landslide-related data; the landslide-related data includes: geographic information, geological information, hydrological environment, historical landslide data, and landslide hazard point data;

[0050] Construct a geospatial database;

[0051] The landslide-related data is imported into the geospatial database to obtain the landslide spatiotemporal database; the landslide spatiotemporal database supports functions based on rainfall events, spatiotemporal clustering, and attribute classification.

[0052] Furthermore, obtaining the multi-source data on the landslide includes:

[0053] Construct an integrated monitoring system based on BeiDou satellite computing;

[0054] Based on the fusion monitoring system, multi-source data at key landslide locations in the study area are collected; wherein, the multi-source data includes: slope deformation rate prediction data, physical field data, and groundwater data.

[0055] Furthermore, constructing the integrated monitoring system includes:

[0056] The first step is to construct a fusion algorithm. The Kalman filter algorithm is used, with the IMU model as the system state and differential positioning data based on GPS and BeiDou systems as the observations for filtering and correction. Environmental information is introduced into the fusion algorithm. By defining the sensor confidence domain, the fusion weight coefficients are assigned based on the confidence level determined by the membership function.

[0057] The second step involves incorporating multiple sensors. Based on the improved coupling framework, a GNSS / IMU fusion and high-precision positioning model is obtained. Pore water pressure, seepage pressure, geothermal, and rain gauge sensors are incorporated. This sensor set is deployed at multiple points in a self-organizing and multi-hop manner to form a unified system that integrates displacement monitoring, pore water pressure monitoring, and monitoring of other indicators calculated by BeiDou satellites, thus constructing the fusion monitoring system.

[0058] Furthermore, all the highly correlated factors for landslide disasters are obtained, including:

[0059] The data factors (elevation, rainfall, water content, groundwater level, pore water pressure, slope, aspect, curvature, landform type, soil and rock type, distance from fault, vegetation, distance from river, soil erosion intensity, population density, distance from road, and soil use type) in the landslide spatiotemporal database containing multi-source data are fused pairwise. Pearson correlation analysis is then performed on the fused data factors to obtain the correlation coefficients between each pair of data factors. Data factors with correlation coefficients higher than a preset threshold are selected as the highly correlated factors.

[0060] Furthermore, the highly correlated factors include: highly correlated factors of landslide spatial susceptibility and highly correlated factors of landslide temporal susceptibility;

[0061] The highly relevant factors for landslide spatial susceptibility include: elevation, water content, rainfall, slope toe, slope, aspect, curvature, landform type, soil and rock type, distance from fault, vegetation, distance from river, soil erosion intensity, population density, distance from road, soil use type, and landslide occurrence events.

[0062] The factors highly correlated with landslide temporality include: daily rainfall, water content, pore water pressure, and groundwater level.

[0063] Furthermore, the spatiotemporal prediction model for regional rainfall-induced landslides includes: a spatial susceptibility prediction model for landslides and a temporal prediction model for regional rainfall-induced landslides;

[0064] The landslide spatial susceptibility prediction model is used to determine the spatial region of landslide events;

[0065] The regional rainfall-induced landslide timing prediction model is used to determine the timing of landslide events.

[0066] Furthermore, constructing the spatiotemporal prediction model for rainfall-induced landslides in the region includes:

[0067] A landslide spatial susceptibility prediction model is established based on support vector machines;

[0068] Based on highly correlated factors of landslide spatial susceptibility, such as elevation, moisture content, rainfall, slope toe, slope, aspect, curvature, landform type, soil and rock type, distance from fault, vegetation, distance from river, soil erosion intensity, population density, distance from road, soil use type, and landslide occurrence events, the landslide spatial susceptibility prediction model was trained using 70% of historical data as the training set, and the model was validated using 30% of historical data.

[0069] A regional rainfall-induced landslide timing prediction model was constructed based on long short-term memory networks.

[0070] Based on factors highly correlated with landslide timing, such as daily rainfall, moisture content, pore water pressure, and groundwater level, 70% of historical data was used to train the rainfall-induced landslide timing prediction model for the region, and 30% of historical data was used for model validation.

[0071] The trained regional rainfall-induced landslide temporal prediction model and the trained landslide spatial susceptibility prediction model are coupled to obtain the regional rainfall-induced landslide spatiotemporal prediction model.

[0072] Example

[0073] like Figure 1 As shown in the figure, this embodiment provides a regional rainfall-induced landslide monitoring and early warning method based on multi-source data fusion, which specifically includes the following steps:

[0074] 1. Geological Exploration, Survey, and Literature Collection: In collaboration with the regional meteorological bureau and natural resources bureau, obtain local geographic information, geological information, hydrological environment, historical landslides, and landslide hazard points data. Organize the data, create data tables in a MySQL or PostgreSQL database, and then import the data of specific storage types into the data tables. In addition, create a geospatial database in ArcGIS software, and then import the data from the data tables into the geospatial database. Combining these two databases forms a landslide spatiotemporal database that supports functions such as rainfall event-based analysis, spatiotemporal clustering, and attribute classification.

[0075] 2. Develop a multi-source data fusion monitoring device based on BeiDou satellite computing. This primarily involves the layout design of sensors and the application of BeiDou satellite differential computing algorithms to achieve high-precision landslide parameter fusion. This fusion monitoring system can simultaneously monitor soil pore water pressure, infiltration pressure, moisture content, rainfall, ground temperature, displacement, and other indicators. Specifically, a Kalman filter algorithm is used, with an IMU model as the system state and differential positioning data based on GPS (Global Positioning System) and BeiDou systems as observations for filtering and correction. Environmental information is incorporated into the fusion algorithm, and fusion weight coefficients are assigned based on the confidence level determined by the membership function by defining the sensor confidence domain. Based on this improved coupling framework, a GNSS / IMU (Global Navigation Satellite System / Inertial Unit) fusion and high-precision positioning model is obtained. Other sensors, such as pore water pressure, infiltration pressure, ground temperature gauges, and rain gauges, are incorporated. This sensor set is deployed at multiple points in a self-organizing and multi-hop manner to integrate BeiDou satellite-calculated displacement monitoring, pore water pressure monitoring, and other indicator monitoring, thus constructing a fusion device. Key parameters of the equipment include: waterproof rating of IP2 or higher; effective data acquisition time of less than 20 seconds; measurement accuracy (under normal conditions) of ±1 cm horizontally and ±2 cm vertically; slope deformation rate prediction function with an accuracy of over 80%; volumetric moisture content testing range of 0-100% with a response time as fast as 1 second and an accuracy of ±3%; pore water pressure testing range of 0-100 kPa with a testing time as fast as 1 second and an accuracy of ±0.1% FS. This monitoring equipment will be deployed at key landslide locations in the study area for data acquisition and monitoring, and the data will be transmitted to a spatiotemporal database.

[0076] 3. Data Mining and Correlation Analysis: The data factors influencing slope landslides were analyzed. First, Pearson correlation analysis was performed pairwise on each factor using SPSSAU software. The correlation module was opened, and the correlation coefficients between all influencing factors were calculated. A positive correlation coefficient (0, 1) indicates a linear positive correlation between the two factors; a correlation coefficient of 0 indicates no linear correlation; and a negative correlation coefficient (-1, 0) indicates a linear negative correlation. The closer the absolute value of the correlation coefficient is to 1, the stronger the correlation. Factors with a correlation coefficient above 0.7 were selected as highly correlated factors. All highly correlated factors of landslide disasters were obtained, and these highly correlated influencing factors served as the basis for data analysis and prediction.

[0077] 4. A landslide spatial susceptibility prediction model is established based on the Support Vector Machine (SVM) machine learning algorithm: Highly correlated influencing factors are selected as input data for the SVM, followed by multiple iterative optimization operations to obtain the mapping relationship between the correlated factors and slope stability, thus establishing a spatial prediction model for slope stability. The constructed spatial prediction model for slope stability is used for prediction, and the prediction results are compared with the actual results. If the error is within the allowable range for engineering, it meets the engineering requirements, and the prediction model is reliable.

[0078] 5. Establishing a Spatiotemporal Prediction Model for Regional Rainfall-Induced Landslides Based on Deep Learning: First, establish a model showing the relationship between highly correlated influencing factors and time. Then, use a Long Short-Term Memory (LSTM) network to predict the time series of these influencing factors. Continuously learn and train on multi-source monitoring data to obtain a series of predictable values ​​for time parameters. If the error between the predicted values ​​and the actual results is within the acceptable range for engineering purposes, the time prediction model is reliable. In establishing the deep learning LSTM network, it is necessary to first determine the model's input and the number of neurons, then establish the network structure, which generally includes hidden layers, fully connected layers, dropout layers, etc., and finally determine the gradient function. The model output is then obtained, and the accuracy of various errors is calculated by comparing the model output with the actual values ​​to evaluate the model's performance. The obtained optimal time prediction model is coupled into a spatial susceptibility prediction model to establish a temporal-spatial correlation, forming the final spatiotemporal prediction model for regional rainfall-induced landslides. The working principle of the spatiotemporal prediction model for regional rainfall-induced landslides is as follows: Figure 2 As shown.

[0079] This embodiment integrates regional historical data and real-time monitoring data to establish a spatiotemporal database of rainfall-induced landslide disasters. This addresses the challenge that susceptibility models cannot predict the stability of rainfall-induced landslide time series, improving the accuracy and reliability of spatiotemporal analysis of rainfall-induced landslides and enabling accurate and effective prediction of when and where landslides will occur. This can contribute to the sustainable economic development of the study area, particularly in the effective management of geological hazard risks.

[0080] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for monitoring and early warning of regional rainfall-induced landslides based on multi-source data fusion, characterized in that, include: Construct a spatiotemporal database of landslides; Acquire multi-source data on landslides and add the multi-source data to the landslide spatiotemporal database; Correlation analysis was performed on the landslide spatiotemporal database containing multi-source data to obtain all highly correlated factors of landslide disasters; A spatiotemporal prediction model for regional rainfall-induced landslides is constructed. The highly correlated factors are input into the spatiotemporal prediction model to obtain the occurrence time and spatial region of landslide events. The multi-source data for acquiring landslide data includes: Construct an integrated monitoring system based on BeiDou satellite computing; Based on the fusion monitoring system, multi-source data at key landslide locations in the study area are collected; wherein, the multi-source data includes: slope deformation rate prediction data, physical field data, and groundwater data; The construction of the integrated monitoring system includes: The first step is to construct a fusion algorithm. The Kalman filter algorithm is used, with the IMU model as the system state and differential positioning data based on GPS and BeiDou systems as the observations for filtering and correction. Environmental information is introduced into the fusion algorithm. By defining the sensor confidence domain, the fusion weight coefficients are assigned based on the confidence level determined by the membership function. The second step is to incorporate multiple sensors. Based on the improved coupling framework, the GNSS / IMU fusion and high-precision positioning model is obtained. Pore water pressure, seepage pressure, geothermal and rain gauge sensors are incorporated. This sensor set is deployed at multiple points in a self-organizing and multi-hop manner to form a fusion monitoring system that integrates displacement monitoring, pore water pressure monitoring and other indicators calculated by Beidou satellites. All of the highly relevant factors for landslide disasters include: The data factors in the landslide spatiotemporal database containing multi-source data are fused pairwise. Pearson correlation analysis is performed on the fused data factors to obtain the correlation coefficients between the data factors. The data factors with correlation coefficients higher than a preset threshold are selected as the highly correlated factors. The highly correlated factors include: highly correlated factors of landslide spatial susceptibility and highly correlated factors of landslide temporal susceptibility; The highly relevant factors for landslide spatial susceptibility include: elevation, water content, rainfall, slope toe, slope, aspect, curvature, landform type, soil and rock type, distance from fault, vegetation, distance from river, soil erosion intensity, population density, distance from road, soil use type, and landslide occurrence events. The factors highly correlated with landslide temporality include: daily rainfall, water content, pore water pressure, and groundwater level; The spatiotemporal prediction model for regional rainfall-induced landslides includes: a spatial susceptibility prediction model for landslides and a temporal prediction model for regional rainfall-induced landslides. The landslide spatial susceptibility prediction model is used to determine the spatial region of landslide events; The regional rainfall-induced landslide timing prediction model is used to determine the timing of landslide events. The construction of the spatiotemporal prediction model for rainfall-induced landslides in the region includes: A landslide spatial susceptibility prediction model is established based on support vector machines; Based on the highly correlated factors of landslide spatial susceptibility, the landslide spatial susceptibility prediction model is trained. A regional rainfall-induced landslide timing prediction model was constructed based on long short-term memory networks. Based on the highly correlated factors of landslide time susceptibility, the prediction model for the time of rainfall-induced landslides in the region was trained. The trained regional rainfall-induced landslide temporal prediction model and the trained landslide spatial susceptibility prediction model are coupled to obtain the regional rainfall-induced landslide spatiotemporal prediction model.

2. The regional rainfall-induced landslide monitoring and early warning method based on multi-source data fusion according to claim 1, characterized in that, Constructing the landslide spatiotemporal database includes: Obtain local landslide-related data; the landslide-related data includes: geographic information, geological information, hydrological environment, historical landslide data, and landslide hazard point data; Construct a geospatial database; The landslide-related data is imported into the geospatial database to obtain the landslide spatiotemporal database; the landslide spatiotemporal database supports functions based on rainfall events, spatiotemporal clustering, and attribute classification.

Images