Landslide distribution probability prediction method and system based on remote sensing data

Abstract

The invention belongs to the technical field of landslide identification and prediction, and particularly relates to a landslide distribution probability prediction method and system based on remote sensing data, and the prediction method comprises the following steps: step 1 to step 6, selecting a research area, capturing remote sensing data, describing a curved surface form of the research area, and constructing a digital elevation model; after screening, introducing elevation data into MATLAB software, and drawing a three-dimensional topographic map in a three-dimensional coordinate system; constructing a landslide probability model, simulating variables in the landslide, deducing seismic data, and deducing slope data; constructing a WGEN model, and deducing a precipitation density function of the research area based on GAMMA distribution to realize precipitation simulation; the prediction system comprises a web crawler module, a local database, a processor and a touch screen. According to the method, the elevation and rainfall of the plateau area are associated, multiple models are constructed, the landslide distribution probability accuracy is high, and risk early warning can be carried out within 3 hours after an earthquake.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of landslide identification and prediction, and particularly relates to a landslide distribution probability prediction method and system based on remote sensing data. BACKGROUND

[0002] Geological disasters are one of the natural disaster types that exist globally, among which secondary landslide disasters are of great concern due to their strong suddenness and destructiveness. Landslides in plateau regions are particularly destructive due to large drop-offs and strong kinetic energy, which can cause strong structural damage to buildings and roads. If the landslide area can be identified and predicted through technical means, and appropriate measures are taken in advance, the property damage and casualties can be greatly reduced.

[0003] In the early stage of landslide prediction, traditional geological surveys are the main method, for example, Liu Chuanzheng's team first systematically demonstrated the feasibility of remote sensing technology in landslide survey in China, and completed the first 1:50000 landslide distribution map in the Three Gorges Reservoir area, establishing the "morphology-structure" dual identification criteria (Reference: Liu Chuanzheng, Li Tiefeng, Wen Mingsheng, et al. Spatial evaluation and early warning of geological disasters in the Three Gorges Reservoir area [J]. Hydrogeology Engineering Geology, 2004, (04): 9-19). With the opening of domestic satellite data, Liao Mingsheng's team from Wuhan University was the first to realize the engineering application of InSAR technology in landslide prediction, which improved the deformation monitoring accuracy from centimeter level to millimeter level (Reference: Liao Mingsheng, Dong Jie, Li Menghua, et al. Radar remote sensing landslide hazard identification and deformation monitoring [J]. Remote Sensing, 2021, 25 (01): 332-341). In the intelligent fusion stage, for example, the TransGEO model developed by Zhang Jianping's team from Tsinghua University can improve the timeliness of landslide prediction in plateau regions to 72 hours (Reference: Deng Li, Yuan Hongyong, Zhang Mingzhi, et al. Research progress of landslide deformation monitoring and early warning technology [J]. Journal of Tsinghua University (Science and Technology Edition), 2023, 63 (06): 849-864).

[0004] The current landslide prediction technology not only establishes a big data model, but also considers environmental factors, for example, a landslide prediction method with the announcement number CN108052761B collects precipitation (R), terrain (S), soil type (T), sand content percentage (sand), silt content percentage (silt), clay content percentage (clay), vegetation coverage (N), soil moisture content (P), and geographic location (W) as control factors, combines landslide curve with Euclidean distance, obtains landslide probability-distance relationship curve, and further obtains landslide probability curve.

[0005] In summary, although the big data model incorporates precipitation to analyze the influence of precipitation on landslide phenomena, the precipitation data is relatively rough and cannot be combined with the topography of high-altitude areas, and further analysis of the correlation is required, and there is a certain data loss, therefore, the present application proposes a landslide distribution probability prediction method and system based on remote sensing data, which is related to the elevation and precipitation of plateau areas, and can perform risk warning within 3 hours after the earthquake. SUMMARY

[0006] The purpose of the present application is to provide a landslide distribution probability prediction method and system based on remote sensing data, which is related to the elevation and precipitation of plateau areas, and can perform risk warning within 3 hours after the earthquake.

[0007] The technical scheme adopted by the present application is as follows: A landslide distribution probability prediction method based on remote sensing data, comprising the following steps: Step 1: Select the study area, grab the remote sensing data, describe the curved surface form of the study area, and construct a digital elevation model; Step 2: Screen the high elevation data, import it into MATLAB software, and draw a three-dimensional topographic map in a three-dimensional coordinate system; Step 3: Construct a landslide probability model, simulate the variables in the landslide, and deduce the slope data and earthquake data; Step 4: Construct a WGEN model, based on the GAMMA distribution, deduce the precipitation density function of the study area, and realize the simulation of precipitation; Step 5: Preprocess the precipitation data and store it in the local database for backup, and the preprocessed parameters are used to deduce the precipitation model and calculate the probability of heavy rain, moderate rain and light rain; Step 6: In the MATLAB software, analyze the DEM data of the study area, predict the landslide probability, and finally compare the landslide simulation results with the actual situation of the study area and visualize the display.

[0008] As an optional solution, the study area is provided with unknown coefficients (i,j=0,1,2,...,m), and the three-dimensional coordinates of reference points in the study area are measured: , , .

[0009] As an optional solution, the remote sensing data includes elevation data, slope data, slope direction data, earthquake fissure and deformation data.

[0010] As an optional scheme, in the landslide probability model construction, the earthquake time domain model is constructed based on Poisson theorem, the landslide distribution prediction after the earthquake is described in the MATLAB software, after the elevation data is imported, the earthquake data is deduced to quantify the relationship between the magnitude and the earthquake frequency, and a short-term earthquake warning mechanism is constructed.

[0011] As an optional scheme, in the slope data deduction, based on the DEM, the surface fluctuation state in space is described by using a binary function, the slope is defined as , and is converted into an angle system, and is divided into stable slope, transition slope, high-risk slope and extremely dangerous slope according to the angle range.

[0012] As an optional scheme, in the density function, the shape parameter is defined , the maximum likelihood estimation method is used for parameter estimation of , After the parameters are determined, the simulation of the precipitation is realized.

[0013] As an optional scheme, in the WGEN model, the 25% quantile, the 50% quantile and the 75% quantile are calculated, and less than 25% quantile is set as light rain, 25% to 75% quantile is set as moderate rain, and greater than 75% quantile is set as heavy rain.

[0014] As an optional scheme, in the establishment of the precipitation infiltration model of landslide in high-altitude area, the preprocessing includes the following steps: Data screening: the time range of remote sensing data is screened, the precipitation data and the temperature data in the key period are extracted, and the interference of irrelevant period data is excluded; Data cleaning: the collected precipitation data and temperature data are detected by setting a reasonable threshold, and the abnormal values are excluded; Data conversion: the daily cumulative precipitation data of the original precipitation data is converted into monthly precipitation data; the daily temperature data of the original temperature data is converted into monthly maximum temperature data, monthly minimum temperature data and monthly average temperature data; Data storage: the converted monthly precipitation data, monthly maximum temperature data, monthly minimum temperature data and monthly average temperature data are stored in a local database.

[0015] As an optional scheme, it further includes establishing a precipitation infiltration model of landslide in high-altitude area, deducing a slope stability quantification formula, considering the freezing and thawing effect of infinite slope safety factor FS, establishing a critical precipitation threshold model, establishing a comprehensive risk assessment index RI, performing multi-factor weighted quantization, and setting a risk level.

[0016] A landslide distribution probability prediction system based on remote sensing data, comprising.

[0017] The technical effects obtained by the present application are: The application is based on remote sensing data, associates the elevation and precipitation in plateau areas, constructs multiple models, fuses to predict, compares with actual data, and has high landslide distribution probability accuracy, can carry out risk early warning within 3h after the earthquake, so as to increase the valuable golden rescue time.

[0018] The application also establishes a precipitation infiltration model of landslide in high altitude area, deduces a quantitative formula of slope stability, considers the safety factor FS of infinite slope of freeze-thaw effect, establishes a critical precipitation threshold model, establishes a comprehensive risk assessment index RI, carries out multi-factor weighted quantization, sets the risk level, so as to clearly understand the emergency degree of risk early warning, so as to correspond to the rescue force. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a flow chart of a landslide distribution probability prediction method based on remote sensing data in embodiment one of the application; Figure 2 is a statistical diagram of the number of earthquakes of 3.0 level and above in embodiment one of the application; Figure 3 is a statistical diagram of the number of earthquakes below 3.0 level in embodiment one of the application; Figure 4 is a diagram of earthquake cracks and deformation data of the first satellite image in embodiment one of the application; Figure 5 is a diagram of earthquake cracks and deformation data of the second satellite image in embodiment one of the application; Figure 6 is a 3601th column elevation profile in embodiment one of the application; Figure 7 is a 3552th row elevation profile in embodiment one of the application; Figure 8 is a contour line filled topographic map of Dingri County in embodiment one of the application; Figure 9 is a false color topographic map of Dingri County in embodiment one of the application; Figure 10 is a three-dimensional topographic map of Dingri County in embodiment one of the application; Figure 11 is a slope classification map of Dingri County in embodiment one of the application; Figure 12 is a slope distribution histogram of Dingri County in embodiment one of the application; Figure 13 is a statistical diagram of monthly precipitation of Dingri County from 2023 to 2025 in embodiment one of the application; Figure 14 is a statistical diagram of monthly precipitation of Dingri County from 2023 to 2025 in embodiment one of the application; Figure 15 is a monthly maximum temperature data graph of Dingri County in 2023-2025 in Embodiment I of the present application; Figure 16 is a monthly minimum temperature data graph of Dingri County in 2023-2025 in Embodiment I of the present application; Figure 17 is a monthly average temperature statistical graph of Dingri County in 2023-2025 in Embodiment I of the present application; Figure 18 is a landslide distribution probability prediction graph after an earthquake in Dingri County in Embodiment I of the present application; Figure 19 is a part of landslide area statistical annotation graph selected after actual data is summarized in Embodiment I of the present application; Figure 20 is a prediction and actual comparison graph in Embodiment I of the present application; Figure 21 is a system flowchart of a landslide distribution probability prediction system based on remote sensing data in Embodiment III of the present application. DETAILED DESCRIPTION

[0020] In order to make the objects and advantages of the present application clearer and more apparent, the present application will be specifically described below in combination with embodiments. It should be understood that the following text is only used to describe one or several specific embodiments of the present application, and does not strictly limit the specific protection scope of the present application.

[0021] In recent years, the frequency of earthquake and landslide disasters in the Qinghai-Tibet Plateau region has gradually increased, with wide distribution, fast occurrence and great harm. The 72-hour golden rescue is extremely important, and the number of landslides triggered by earthquakes in the Qinghai-Tibet Plateau region is extremely large, with high altitude, sparse population and high investigation difficulty. Therefore, predicting the landslide distribution probability after an earthquake is very helpful for rescue operations.

[0022] According to the National Natural Disaster Assessment Report released by the Office of the National Disaster Prevention and Mitigation Committee and the Emergency Management Department in January 2025, the earthquake in Dingri County, Tibet, caused direct economic losses of 89.45 billion yuan, and the disaster chain characteristics were as follows: the strike of the earthquake caused significant damage to regional buildings, with about 2.69 thousand masonry structure houses collapsed that did not meet the seismic fortification requirements; at the same time, the earthquake triggered secondary landslides that destroyed 8 sections (cumulative damage of 3.2 km) of G219 National Highway, causing the interruption of the disaster relief channel for 76 hours. Therefore, how to predict the landslide distribution after an earthquake in a timely manner, shorten the rescue time, and even make disaster relief work deployment in advance, has become the main problem in dealing with secondary landslide disasters.

[0023] The application is based on a deep learning algorithm, extracts key elevation data mainly in Tibet Dingri County through remote sensing data in the national public geographic spatial data cloud, adds real earthquake data and simulates a fixed precipitation model, constructs a landslide distribution probability prediction model after an earthquake in Dingri County, pre-processes the elevation data, earthquake data and precipitation data, analyzes the topographic data characteristics of Dingri County, predicts the landslide distribution by using a convolutional neural network, and compares the prediction result with the actual landslide data of the national Qinghai-Tibet Plateau scientific data center, so that risk warning can be carried out within 3 hours after an earthquake through landslide distribution probability prediction.

[0024] Embodiment one: As Figures 1-20 shown, a landslide distribution probability prediction method based on remote sensing data comprises the following steps: Step one: constructing a digital elevation model; The overall interpolation method is mainly used to reflect the macro trend of terrain undulation, and the overall interpolation method is also called global function interpolation. The principle is to use a polynomial function to perform global interpolation operation by comprehensively sampling and observing the sample points in the studied area, so as to establish a complete interpolation fitting model. The curved surface form describing the research area is expressed as the following binary polynomial: Formula (1) In the formula, there are undetermined coefficients (i,j=0,1,2,...,m), and the three-dimensional coordinates of reference points in the research area are measured: , , , On this basis, they are substituted into formula (1) to obtain a unique m-order linear equation group, and the coordinates of the interpolation points are converted into the height values of the measured points by formula (1); the interpolation fitting model can effectively integrate the information of the discrete sample points in the research area, and then present the overall situation of the terrain undulation of the entire research area. Elevation is one of the key factors affecting the occurrence of landslides, and it does not work independently, but through the mutual coupling with many other factors, it jointly affects the occurrence of landslides. For example, the spatial differences of geology and weathering, the vertical differentiation of hydrological conditions, the vertical differentiation of human activities, etc. will affect the occurrence of landslides to different degrees; and the elevation data is the original data of DEM (digital elevation model), and DEM can comprehensively and accurately describe the terrain undulation of the ground through the integration of a large number of discrete elevation points. Based on this elevation data, a number of parameters important to landslide research can be further derived, among which slope is one of the important factors affecting slope stability. From a mechanical perspective, the stability of a slope is essentially the result of the interaction of slope angle and friction angle, permeability, and cohesive force, etc. In this study, slope data is derived from DEM with a resolution of 30m; Grabbing elevation data released by the National Earthquake Science Data Center, such as GCEMV3 data, which is an important part of satellite remote sensing data, has a resolution of 30m, and GCEMV3 data covers most of the area of Dingri County, stored in TIF format, and the specific form is as follows: Elevation data: records the elevation value of each pixel point on the ground, unit: meters (m); Slope data: records the slope value of each pixel point on the ground, unit: degrees (°); Aspect data: records the aspect value of each pixel point on the ground, unit: degrees (°), where 0° represents the north direction, and the angle increases clockwise; Also includes earthquake cracks and deformation data, mainly from the earthquake monitoring network and satellite remote sensing technology of the center, the data is stored in the form of pictures, including satellite images, aerial photos and ground photos, such as detailed information of M6.8 main shock and subsequent aftershocks in Dingri County, Tibet, on January 7, 2025, including time, latitude, longitude, depth, magnitude and location, etc.; As of 16:00 on February 10, 2025, a total of 9451 aftershocks were recorded, as shown in Figure 2 The number of earthquakes with a magnitude of 3.0 or above (a total of 70 times), and Figure 3 The number of earthquakes with a magnitude below 3.0 (a total of 9381 times) is shown in; Earthquake crack and deformation data are presented in the form of pictures, see Figure 4 and Figure 5 , mainly including the following contents: Earthquake crack pictures: record the location, shape, size, etc. of surface cracks after an earthquake, including high-resolution satellite images, aerial photos and ground photos; Before and after the earthquake comparison pictures: record the vertical and horizontal displacement of the ground before and after the earthquake, including InSAR (Interferometric Synthetic Aperture Radar) deformation map, GPS (Global Positioning System) observation point map, etc.; Slope aspect, also a very important topographic factor, has a significant impact on the infiltration and runoff of precipitation, and the absorption of solar radiation is also deeply affected by the slope aspect, which will indirectly affect the occurrence of landslides. In general, without considering other complex geological environmental factors, due to the fact that the windward slope is more likely to collect precipitation, and the infiltration of rainwater may cause the rock-soil body to be saturated, the shear strength is reduced; the sunny slope is due to the strong solar radiation, the weathering of the rock-soil body is relatively intense, and the structure may be more loose, so the windward slope and the sunny slope are more suitable for the development of landslides compared to other slope aspects to a certain extent; In this study, the slope aspect of the terrain surface is divided into 9 categories according to the direction: flat, N, EN, E, ES, S, WS, W and WN; Elevation, slope and aspect are the core parameter set of terrain analysis. By combining slope and aspect with DEM, the spatial distribution of landslide risk can be accurately quantified from multiple dimensions. Step 2: Select the elevation data after filtering; Total data amount: In this embodiment, the elevation data of the Rikaze City Dingri County in Tibet Autonomous Region can be sourced from the geographic spatial data cloud, and the two main categories of data with the central latitude of 28°50'N and the central longitude of 86°50'E and 87°50'E are mainly in the range of 86°20'E~87°70'E, 27°80'N~29°10'N, covering most of the Dingri County area; Remote sensing data conversion: Using the data import function of MATLAB software, the elevation data in the remote sensing data is converted into digital form corresponding to the elevation of each latitude and longitude; Distribution of data after filtering: First, analyze the terrain of Dingri County in MATLAB software in the longitudinal and transverse directions, draw the row and column profile graphs based on latitude and longitude, and take the 3601th column elevation profile graph as shown in Figure 6 and the 3552nd row elevation profile graph as shown in Figure 7 as examples; Secondly, the elevation data is summarized in MATLAB software to form the contour filled terrain map as shown in Figure 8 and the pseudo-color terrain map as shown in Figure 9 Then, in order to facilitate our observation of the terrain changes in Dingri County, a three-dimensional terrain map is drawn in MATLAB software based on the three-dimensional coordinate system; Step 3: Construction of landslide probability model, simulation of landslide variables; Derivation of earthquake data: ​The Poisson theorem is used to construct a time-domain model of earthquakes to reveal the potential laws of the Dingri County earthquake sequence in the time dimension. The distribution of landslides after earthquakes is described using MATLAB software to provide a reliable theoretical basis for predicting the distribution range, size, and occurrence probability of landslides after earthquakes, thereby helping relevant departments to develop more efficient and accurate disaster prevention and mitigation strategies to minimize the losses caused by secondary disasters after earthquakes; The data used here were sampled from January 7, 2025 to February 9, 2025, a total of 34 days in this earthquake interval, with a total of 9451 earthquakes of all sizes; The probability formula for h earthquakes occurring in t days is: Formula (2) Where the average occurrence rate times / day, is a constant; Further analysis of the correlation patterns of earthquake magnitudes in Dingri area in multiple key dimensions; on the one hand, we hope to quantify the relationship between magnitude and earthquake frequency to provide a reliable basis for predicting the distribution probability of landslides caused by future earthquake activity trends; on the other hand, we reveal the energy release rules hidden behind the magnitude, accurately assess the potential disaster energy scale, and analyze the dynamic evolution trajectory of the magnitude over time to assist in building a short-term earthquake warning mechanism; From Gutenberg-Richter G-R's magnitude relationship, the following formula is derived: Formula (3) Where represents the small interval centered on the magnitude ; ; is the seismicity parameter of the area; is the proportional relationship between the number of large and small earthquakes, if the number of large earthquakes is larger, the value of is smaller; represents the magnitude; The number of earthquakes occurring in a certain period is: Formula (4) Where , ; The probability density function of the magnitude can be expressed as: Formula (5) Where is the minimum magnitude during the earthquake, represents the differential in calculus, emphasizing that the continuous probability needs to be calculated by integration (i.e., the cumulative of ); The cumulative distribution function can be obtained: Formula (6) in, For any earthquake magnitude; By combining data such as earthquake magnitude and time, we can derive the probability density function, cumulative function, and transcendental function for Tingri County: Using the above formula, the probability of different magnitude areas within the earthquake interval of magnitude 3 or above in Dingri County can be calculated: The equation is consistent with the statistical probability of the data we obtained, therefore it has practical utility and conforms to the law of earthquake magnitude variation; Slope data derivation: In topography and geological hazard assessment, slope is an important indicator reflecting topographic features. Its accurate acquisition and reasonable classification are of great significance. DEM, as a commonly used data form for recording surface elevation information, provides a basic data source for slope calculation. Q1. Derivation of slope from DEM; The elevation distribution of the Earth's surface can be considered as relative to plane coordinates. The relevant bivariate functions are: Formula (7) in, and These are the x-coordinate and y-coordinate on the plane, respectively. The elevation of the corresponding point, formula (7) describes the undulation of the earth's surface in space, and provides a theoretical basis for subsequent derivation; Taking the total differential of formula (7), we get: Formula (8) Equation (8) reflects the displacement in a small plane. Below, elevation Change This reflects the rate of change of elevation with respect to plane coordinates; On curved surfaces For any point on the plane, its tangent plane can be represented by two linearly independent direction vectors, as follows: along Direction, Elevation Changes can be represented by vectors To describe, assuming in There is a tiny increment in direction. ,and Increment in direction The elevation change at this time is as follows: Formula (9) make Then it can be simplified to: Similarly, along the y-direction: Let the normal vector of the tangent plane at this point be... Since the normal vector is perpendicular to any vector on the tangent plane, it satisfies the following equation: Formula (10) Solving for: slope Defined as normal vector and vertical direction The cosine of the included angle, according to the dot product formula: Formula (11) in, for and The included angle; Therefore, we can conclude that: And slope satisfy: so: Formula (12) Due to the slope calculated above It is measured in radians, but in practical applications it needs to be converted to degrees. The conversion formula is: Formula (13) Q2. Slope distribution probability statistics; Considering the unique topography of the Qinghai-Tibet Plateau, it is divided into four categories based on the slope, and the proportion of each category in the entire study area is calculated using formula (14). ,as follows: Formula (14) in, The total number of pixels in the study area is categorized as follows: Stable slope : Such slope area is relatively gentle, and the possibility of landslide is small; the number of pixels in this slope range is , and the proportion is Transition slope : The slope is in the middle transition state, and has a certain landslide potential risk, and the number of pixels is , and the proportion is High-risk slope : The landslide risk of the slope range is high, and the number of pixels is , and the proportion is Extremely dangerous slope : The slope is steep, and the possibility of landslide is great, and the number of pixels is , and the proportion is In summary, the slope classification of Dingri County is shown in Figure 11 , and the slope distribution of Dingri County is shown in Figure 12 , so that users can intuitively see the main range of different types of slopes, anchor the rescue destination, and play a role in early warning and concentrated rescue forces. Step four: build a WGEN model (Weather Generator); The rock-soil structure is destroyed after the earthquake, and its stability is greatly reduced. In such a fragile state, once encountering a heavy rainfall process, the rainwater quickly infiltrates, not only increases the weight of the rock-soil body, but also increases the pore water pressure, which weakens the shear strength of the rock-soil body, and further greatly increases the probability of landslide. Therefore, in order to study the landslide disaster phenomenon occurring within 72 hours after the earthquake, it is necessary to consider factors such as precipitation, temperature and other factors, so the WGEN model needs to be introduced in this embodiment. The WGEN model, also known as the weather data simulation model, its core function is to study the general characteristics of the climate of a certain area, and through the in-depth mining and analysis of statistical characteristics, it can simulate the daily weather data of the area within a certain time. It can simulate precipitation, temperature, solar radiation and other climatic factors. It can generate daily weather scenarios in accordance with the statistical rules presented by historical climate data, which provides important data support and analysis tools for our research on the influence of meteorological factors on landslides after the earthquake. The simulation of precipitation can use the GAMMA distribution, and the corresponding density function is: Equation (15) In the formula: wherein, is a shape parameter, the size of which is used to determine the basic shape of the distribution density curve; is a scale parameter, the size of which is used to determine the scale of the GAMMA distribution; the parameter estimation of , is carried out by using the maximum likelihood estimation method, and after the parameters are determined, the simulation of the precipitation can be realized; Step five, precipitation data derivation; L1, precipitation data preprocessing: The monthly precipitation data and temperature data of Dingyi County from January 2023 to February 2025 are collected (from the National Meteorological Administration), as shown in Table 1. Many months have a precipitation of 0 mm. Since the gamma distribution is suitable for continuous positive data, we need to preprocess the precipitation data. First, we need to distinguish between rainy and non-rainy months. Second, to better fit regional research, we will use the quantile division method based on historical data to sort the non-zero precipitation months according to the precipitation from small to large, and calculate the 25% quantile, 50% quantile and 75% quantile. We set less than 25% quantile as light rain, 25% to 75% quantile as moderate rain, and greater than 75% quantile as heavy rain. Table 1: Precipitation summary Date Monthly Precipitation (mm) Is there precipitation this month? Precipitation level January 2023 0 No - February 2023 0 No - March 2023 0 No - April 2023 0 No - May 2023 11 Yes Moderate rain June 2023 0.4 Yes Light rain July 2023 86.8 Yes Heavy rain August 2023 69.2 Yes Heavy rain September 2023 5.2 Yes Moderate rain October 2023 0.2 Yes Light rain November 2023 0 No - December 2023 3.7 Yes Moderate rain January 2024 0 No - February 2024 0.2 Yes Light rain March 2024 2.1 Yes Moderate rain April 2024 0 No - May 2024 2.5 Yes Moderate rain June 2024 10.7 Yes Moderate rain July 2024 61.2 Yes Heavy rain August 2024 57 Yes Heavy rain September 2024 22 Yes Moderate rain October 2024 0 No - November 2024 1 Yes Moderate rain December 2024 0 No - January 2025 0 No - February 2025 0 No - The preprocessing approach is as follows: L101, data screening: Since the research focuses on the correlation between meteorological factors and landslides within 72 hours after the earthquake in Dingyi County, we first need to filter the data by time range. We accurately extracted the precipitation data and temperature data in the key period from one week before the earthquake to 72 hours after the earthquake, ensuring that the analyzed data closely meet the research needs and excluding irrelevant period data to improve the research relevance. L102, data cleaning: There may be outliers and missing values in the collected data. For outliers, we detect them by setting a reasonable threshold. L103, data conversion: The original precipitation data is mostly presented in daily cumulative precipitation data, which is converted to monthly precipitation data. The original temperature data is mostly presented in daily temperature data, which is converted to monthly maximum temperature data, monthly minimum temperature data, and monthly average temperature data. For example, the monthly maximum temperature data of Dingyi County from 2023 to 2025 converted in this embodiment is shown in Table 2, and the monthly minimum temperature data of Dingyi County from 2023 to 2025 is shown in Table 3. Figure 15 Table 2: Monthly maximum temperature data of Dingyi County from 2023 to 2025 Figure 16As shown, the monthly average temperature data of Dingjia County from 2023 to 2025 is as follows Figure 17 As shown; L104, data storage: the monthly precipitation data, monthly maximum temperature data, monthly minimum temperature data and monthly average temperature data after screening, cleaning and conversion are stored in a specially constructed local database; Combined with Table 1, this embodiment summarizes the monthly precipitation statistics of Dingjia County from 2023 to 2025 as shown Figure 13 As shown, the monthly precipitation statistics of Dingjia County from 2023 to 2025 is as follows Figure 14 As shown; L2, precipitation model parameter derivation: It is mentioned in the selection of WGEN model that the shape parameter and scale parameter in gamma function need to be estimated and solved by maximum likelihood method, so the sample mean and sample geometric mean need to be calculated first, as follows: Sample mean: Among them, represents the month from 2023, represents the monthly precipitation of the month; Geometric mean: geometric mean is the nth root of the product of each data, and as the data set increases, the product is obviously very large, so log transformation is needed, that is: Equation (16) Solve equation (16) to get: Equation (17) Solve equation (17) by Newton iteration method 0.45, ; To calculate the probability of heavy rain, moderate rain and light rain under the condition of rain, that is, to calculate the interval probability, the cumulative distribution function is also needed, and the cumulative distribution function (CDF) of gamma distribution is: Equation (18) Among them, is the lower incomplete gamma function, is the gamma function; Since less than 25% quantile is light rain, 25% to 75% quantile is moderate rain, and more than 75% quantile is heavy rain, the grade division standard is calculated as follows: 1mm is light rain, 57mm 1mm is moderate rain, 57mm is heavy rain, the specific calculation is as follows: Light rain probability: Bring in parameters , , get Probability of moderate rain: Probability of heavy rain: Step six, comparison and analysis of results; Landslide distribution probability simulation: Using MATLAB software, analyze the DEM data of Dingri County, assisted by 6.5 magnitude and 0mm precipitation and predict landslide probability, and finally visualize the results as shown in Figure 18 , Figure 18 The position of the red dot marked is the direction where the landslide is predicted to occur; Comparison of landslide simulation and actual situation: According to the satellite image comparison chart before and after the earthquake from the National Qinghai-Tibet Plateau Scientific Data Center, as well as the landslide point density distribution chart obtained from other literature, some landslide distribution charts are selected to make the actual data summary selected part of the landslide area statistics annotation chart as shown in Figure 19 , Then compare the actual data statistics chart with the prediction chart, and the comparison result is shown in Figure 20 , Figure 20 The left is the prediction chart, Figure 20 The right is the actual data statistics chart, according to Figure 20 , and the simulation distribution Figure 18 , it can be seen that the accuracy is about 80%, the accuracy is relatively high, and it can be optimized and actually used.

[0025] In summary, based on remote sensing data, the elevation and precipitation of the plateau region are associated, multiple models are constructed, and the prediction is made after fusion. After comparing with the actual data, the landslide distribution probability accuracy is high, and the risk warning can be made within 3h after the earthquake, thereby increasing the valuable golden rescue time.

[0026] Example two: In high-altitude areas, landslides induced by precipitation are affected by multiple factors such as low-temperature freezing and thawing, seasonal snow melting, and rock fracture development, and need to be further quantitatively analyzed based on example one, the difference lies in: Establishment of precipitation infiltration model for landslides in high-altitude areas: The Green-Ampt infiltration formula is corrected, and the surface soil in high-altitude areas often contains freeze-thaw cracks, which need to be corrected for infiltration capacity, as follows: Formula (19) in, for Infiltration rate at any given time (mm / h). The value is the saturated permeability coefficient (increased by 20% to 50% after freeze-thaw disturbance). The wet frontal suction force is taken as 0.1 kPa to 0.3 kPa for weathered rocks and soils at high altitudes. Due to poor pore water saturation, To accumulate infiltration, It is the fracture water conductivity (positively correlated with fracture density). This is the fracture closure attenuation coefficient (usually taken as 0.05). ~0.2 ); Applicable to calculating the dynamic infiltration process of precipitation in freeze-thaw fractured soil; The formula for quantifying slope stability is derived, and the safety factor FS for an infinite slope considering freeze-thaw effects is as follows: Formula (20) in, Effective cohesion (reduced by 30% to 70% after freeze-thaw cycles). The dry density of the soil and rock For the density of water in rock and soil, This refers to the potential thickness of the landslide mass (typically 1m to 5m for shallow landslides at high altitudes). This refers to the groundwater level. Additional shear force generated by the ice lens ( ); A critical precipitation threshold model is established, such as a two-parameter threshold based on cumulative precipitation and intensity. The expression for the critical condition in high-altitude areas is as follows: Formula (21) in, The critical energy index for landslide triggering. Rainfall intensity (mm / h) The duration of precipitation is in mm. This represents the equivalent of accumulated snow water in the previous period (mm). For the thickness of the active layer, and These are all regional parameters (for example, obtained through historical landslide inversion fitting, such as those from the eastern part of the Qinghai-Tibet Plateau). ≈12.3, ); Derivation of the dynamic response formula for pore water pressure; high-altitude rock mass with well-developed joints; pore water pressure satisfy: Equation (22) wherein, is the consolidation coefficient (for jointed rock mass ), is the joint permeability coefficient (proportional to the cube of the opening degree), is the joint water pressure (can be calculated by the cubic law ), is the vertical depth, and the numerical solution can be obtained by finite difference method or COMSOL and other physical field software; An integrated risk assessment index RI is established, and multi-factor weighted quantification is performed, as follows: Equation (23) wherein, is the 24h cumulative precipitation, is the initial safety factor, is the real-time safety factor, is the air temperature (unit ℃, reflecting the freeze-thaw state), and the weight =0.5, =0.3, =0.2; Based on the above, the risk level is set as follows: extremely high risk, >1.5; high risk, 1.5> >1.0; low risk, 1.0> .

[0027] Example Three: As Figure 21 shown, a landslide distribution probability prediction system based on remote sensing data includes: A network crawler module can be a pre-written crawler program that automatically crawls data from the cloud within a specified time period; A local database is used to store data, including elevation data, slope data, slope direction data, seismic crack and deformation data, and hardware configuration: Intel Core i5 processor, 4T normal hard disk memory, 512G-SSD (solid state disk), 16G flash memory, independent 500W power supply and matching liquid cooling system, such as MySQL database; A processor is separately configured with Intel Core i9 or AMD Ryzen 7, which can be high-performance loaded with MATLAB software, digital elevation model, landslide probability model, WGEN model, and precipitation model; Touch screen, fast communication through RS485, communication and interaction with processor and local database, used for visual display of MATLAB software, digital elevation model, landslide probability model, WGEN model, and precipitation model, and after predicting landslide probability, comparison of landslide simulation results with actual situation of the study area for visual display; wherein, The main code for landslide probability distribution prediction written by MATLAB software is as follows: Suppose the column order of sample feature data is: magnitude, slope, elevation, aspect, and precipitation magnitude = 6.5 * ones(num_samples, 1); Magnitude slope = sample_features(:, 2); Slope elevation = sample_features(:, 3); Elevation aspect = sample_features(:, 4); Aspect precipitation = sample_features(:, 5); Precipitation Construct feature matrix and label real_features= [magnitude, slope, elevation, aspect, precipitation]; Adjust the label construction formula real_labels = 0.5 * magnitude + 0.3 * slope + 0.1 * elevation + 0.05* aspect + 0.05 * randn(num_samples, 1); real_labels = max(min(real_labels, 1), 0); % Limit the probability to the range [0, 1] Divide the training set and test set train_ratio = 0.8; train_indices = 1:round(train_ratio * num_samples); test_indices = (round(train_ratio * num_samples) + 1):num_samples; train_features = real_features(train_indices, :); train_labels = real_labels(train_indices, :); test_features = real_features(test_indices, :); test_labels = real_labels(test_indices, :); Ensure data type is double train_features = double(train_features); train_labels = double(train_labels); test_features = double(test_features); test_labels = double(test_labels); Create a more complex convolutional neural network layers = [ featureInputLayer(num_features) fullyConnectedLayer(32) reluLayer() fullyConnectedLayer(16) reluLayer() fullyConnectedLayer(8) reluLayer() fullyConnectedLayer(1) regressionLayer()]; Set training options options = trainingOptions('adam',... 'InitialLearnRate', 0.001,... 'MaxEpochs', 10,... 'MiniBatchSize', 16,... 'Verbose', false,... 'Plots', 'training-progress'); Training the neural network net = trainNetwork(train_features, train_labels, layers, options); Making predictions on the test set predictions = predict(net, test_features); Assuming each sample corresponds to a location in the DEM matrix (need to match according to actual situation) sample_indices = randi([1, rows * cols], num_samples, 1); [sample_rows, sample_cols] = ind2sub([rows, cols], sample_indices); Finding samples with high landslide probability and their corresponding terrain locations Adjusting the probability threshold high_prob_indices = find(predictions > 0.3); high_prob_rows = double(sample_rows(test_indices(high_prob_indices))); high_prob_cols=double(sample_cols(test_indices(high_prob_indices)))。

[0028] The above is only an optional embodiment of the present application. It should be noted that for those skilled in the art, without departing from the principles of the present application, several improvements and refinements can be made, which should also be considered within the scope of protection of the present application. The structures, devices and operation methods not specifically described and explained in the present application, if not specially described and limited, are implemented according to the conventional means in the art.

Claims

1. A landslide distribution probability prediction method based on remote sensing data, characterized in that, It comprises the following steps: Step 1: select the study area, grab remote sensing data, describe the curved surface form of the study area, and build a digital elevation model; Step 2: filter the high elevation data, import it into MATLAB software, and draw a 3D topographic map in a 3D coordinate system; Step 3: landslide probability model construction, variable simulation in landslide, earthquake data derivation, and slope data derivation; Step 4: build WGEN model, based on GAMMA distribution, derive rainfall density function of the study area, and realize rainfall simulation; Step 5: pre-process the rainfall data and store it in the local database for future use. The pre-processed parameters are used to derive the rainfall model and calculate the probability of heavy rain, moderate rain, and light rain; Step 6: analyze the DEM data of the study area in MATLAB software, predict the landslide probability, and finally compare the landslide simulation results with the actual situation of the study area and visualize the results. 2.The landslide probability prediction method based on remote sensing data according to claim 1, characterized in that: The research region is set with a to-be-determined coefficient (i,j=0,1,2,...,m), in the research region, the three-dimensional coordinates of a reference point are measured: , , . 3.The landslide distribution probability prediction method based on remote sensing data according to claim 1, characterized in that: The remote sensing data includes elevation data, slope data, slope direction data, earthquake fissure and deformation data.

4. The landslide distribution probability prediction method based on remote sensing data according to claim 1, characterized in that: In the landslide probability model construction, the earthquake time domain model is built based on Poisson's theorem, the landslide distribution prediction after earthquake is described in MATLAB software, the earthquake data is derived after importing the elevation data, and the relationship between magnitude and earthquake frequency is quantified to assist in building a short-term earthquake warning mechanism.

5. The landslide probability prediction method based on remote sensing data according to claim 1, characterized in that: In the slope data derivation, based on DEM, the spatial fluctuation of the ground surface is described by a binary function, and the slope is defined And then, it is converted into angle system, and is divided into stable slope, transition slope, high-risk slope and extreme-risk slope according to the angle range.

6. The landslide probability prediction method based on remote sensing data according to claim 1, characterized in that: In the density function, the shape parameter is defined The maximum likelihood estimation method is used to estimate the parameters , After the parameters are determined, the precipitation is simulated.

7. The landslide probability prediction method based on remote sensing data according to claim 1, characterized in that: In the WGEN model, the 25th percentile, 50th percentile and 75th percentile are calculated, and less than 25% is set as light rain, 25% to 75% is set as moderate rain, and more than 75% is set as heavy rain. 8.The landslide probability prediction method based on remote sensing data according to claim 1, wherein, The preprocessing of the rainfall infiltration model for landslides in high-altitude areas comprises the following steps: Data screening: time range screening of remote sensing data, extraction of precipitation data and air temperature data in the key period, and exclusion of irrelevant period data interference; Data cleaning: detecting abnormal values by setting reasonable thresholds for collected precipitation data and air temperature data; Data conversion: converting daily cumulative precipitation data of original precipitation data into monthly precipitation data; converting daily air temperature data of original air temperature data into monthly maximum temperature data, monthly minimum temperature data, and monthly average temperature data; Data storage: storing the converted monthly precipitation data, monthly maximum temperature data, monthly minimum temperature data, and monthly average temperature data in the local database. 9.The landslide probability prediction method based on remote sensing data according to claim 1, characterized in that: It also includes building a rainfall infiltration model for landslides in high-altitude areas, deriving a slope stability quantification formula, considering the freezing and thawing effect of infinite slope safety factor FS, building a critical rainfall threshold model, building a comprehensive risk assessment index RI, and performing multi-factor weighted quantification to set risk levels.

10. A landslide distribution probability prediction system based on remote sensing data, applied to the landslide distribution probability prediction method based on remote sensing data according to any one of claims 1-9, characterized in that, It comprises: A web crawler module that automatically crawls data from the cloud within a specified time period; A local database for storing data; A processor equipped with MATLAB software, a digital elevation model, a landslide probability model, a WGEN model, and a rainfall model; A touch screen for visual display.

Images