Prediction device

Abstract

A prediction device (10) comprises a control unit (11) that, for a plurality of sites, uses a plurality of first learning groups, which are obtained by dividing a first learning data group indicating a plurality of indices related to topography and the presence / absence of damage due to landslide on the basis of the indices indicated by the first learning data group, to construct a prediction model for each of the first learning groups, said prediction model predicting the risk of a landslide, and, when a plurality of first indices related to topography are acquired for a first site for prediction, selects one first learning group as a first selected group on the basis of the first indices, and uses the prediction model of the first selected group to predict the risk of a landslide at the first site.

Description

Prediction device

[0001] This disclosure relates to a prediction device.

[0002] Various technologies are known for predicting the risk of landslides during heavy rainfall. For example, Non-Patent Document 1 discloses that on-site surveys are conducted when designating landslide hazard zones and special landslide hazard zones. For example, Non-Patent Document 2 discloses that warning information is issued based on soil moisture content and 60-minute cumulative rainfall predicted using a tank model. The parameters of this tank model are standardized throughout Japan, and topographic information is not used. For example, Non-Patent Document 3 discloses a proposed probability map that calculates the probability of landslides occurring throughout Japan in three stages based on three types of data: landslide hazard zones, deep-seated collapse frequency maps, and landslide topographic distribution maps. For example, Non-Patent Document 4 discloses a method for predicting landslides during heavy rainfall by predicting the number of damaged buildings and the rate of damage caused by typhoons, heavy rain, and earthquakes for each municipality and displaying it on a map. In Non-Patent Document 4, the pre-disaster prediction for the typhoon is up to 7 days in advance. For example, Non-Patent Document 5 discloses a landslide occurrence prediction support system using multi-parameter radar.

[0003] Ministry of Land, Infrastructure, Transport and Tourism, "Overview of the Landslide Disaster Prevention Act," [online], [searched March 3, 2025], Internet <URL: http: / / www.mlit.go.jp / river / sabo / sinpoupdf / gaiyou.pdf> Japan Meteorological Agency, "Landslide Disaster Warning Information / Landslide Kikikuru (Risk Distribution of Heavy Rain Warnings (Landslide Disasters))," [online], [searched March 3, 2025], Internet <URL: https: / / www.jma.go.jp / jma / kishou / know / bosai / doshakeikai.html> Matsuda Masayuki et al., "A Study on a Method for Estimating the Risk of Sediment-Related Disasters Nationwide Using Thematic Maps Related to Topography and Geology," National Institute for Land and Infrastructure Management Document, No. 1120, 2020, [online], [Retrieved March 3, 2025], Internet <URL: https: / / www.nilim.go.jp / lab / bcg / siryou / tnn / tnn1120.htm> "Real-time Damage Prediction Website / App cmap," [online], [Retrieved March 3, 2025], Internet <URL: https: / / aioinissaydowa.co.jp / corporate / service / cmap / > "Predicting the Risk of Sediment-Related Disasters and Disseminating Information with Map Images - Development of a 'Sediment-Related Disaster Occurrence Prediction Support System'," [online], [Retrieved March 3, 2025], Internet <URL: https: / / www.engineering-eye.com / interview / user / 14 / >

[0004] There is currently no method for predicting landslides occurring in mountainous areas that does not require special field surveys or measurements across a wide area such as the entire country of Japan, and that can be used even during peacetime.

[0005] In light of these points, the purpose of this disclosure is to enable efficient prediction of landslide risk during peacetime.

[0006] A prediction device according to one embodiment of the present disclosure includes a prediction model for each of several first learning groups that predicts landslide risk, using each of several first learning groups obtained by dividing a first learning data set, which includes several indicators related to topography and a first learning data set indicating whether or not there is damage from landslides, based on at least one indicator shown in the first learning data set, for each of several locations, and when several first indicators related to topography are obtained for a first location to be predicted, the device selects one of the several first learning groups as a first selection group based on one or more first indicators, and a control unit that predicts the landslide risk of the first location using the prediction model of the first selection group.

[0007] According to one embodiment of this disclosure, it becomes possible to efficiently predict the risk of landslides during normal times.

[0008] This is a block diagram showing the schematic configuration of the prediction device. This is a flowchart showing an example of the overall flow of the sediment disaster risk prediction process. This is a diagram for explaining the first target point, the sediment runoff route, and the second target point. This is a flowchart showing an example of the flow of the sediment collapse risk prediction process at the first target point. This is a flowchart showing an example of the sediment collapse risk prediction process at the first target point. This is a diagram showing an example of elevation information. This is a diagram for explaining example 1 of setting a predetermined area. This is a diagram for explaining example 2 of setting a predetermined area. This is a diagram for explaining example 3 of setting a predetermined area. This is a table showing an example of a first learning data group that is divided into multiple first learning groups. This is a diagram showing an example of an expanded prediction area. This is a table showing an example of first prediction data. This is a table showing an example of expanded prediction data. This is a table for explaining the method for selecting the first selection group. This is a table for explaining the method for selecting the first selection group. This is a diagram for explaining the method for selecting the first selection group. This is a diagram for explaining the method for selecting the first selection group. This is a diagram for explaining the method for selecting the first selection group. This is a diagram for explaining the method for selecting the first selection group. This is a flowchart showing an example of the sediment runoff route prediction process. This is a flowchart showing an example of the sediment arrival risk prediction process at the second target point. This flowchart shows an example of the process for predicting the risk of sediment impact at the second target location. This flowchart shows an example of the process for predicting sediment disaster risk.

[0009] The embodiments relating to this disclosure will be described below with reference to the figures.

[0010] In each figure, identical or corresponding parts are denoted by the same reference numerals. In the description of this embodiment, the description of identical or corresponding parts will be omitted or simplified as appropriate.

[0011] The outline of this embodiment will be described with reference to Figure 1. The prediction device 10 according to this embodiment is, for example, a general-purpose computer such as a PC or tablet, a server computer such as a cloud server, or a dedicated computer. "PC" is an abbreviation for Personal Computer.

[0012] The prediction device 10 constructs a prediction model (hereinafter referred to as the first prediction model) for each of the multiple locations (hereinafter referred to as multiple first locations) by dividing the first learning data set, which includes multiple indicators related to topography and the presence or absence of landslides, into multiple first learning groups based on at least one indicator shown in the first learning data set, and predicting the risk of landslides. When the prediction device 10 obtains multiple first indicators related to topography (hereinafter referred to as multiple first prediction indicators) for the first location to be predicted (hereinafter referred to as the first target location), it selects one of the multiple first learning groups as the first selection group based on one or more first prediction indicators, and predicts the risk of landslides at the first target location using the prediction model (hereinafter referred to as the first selection model) of the first selection group.

[0013] According to this embodiment, based on the topography of the first target site, a first prediction model that is best suited to the prediction of the first target site can be selected from among multiple first prediction models. Therefore, the risk of landslides in topography with various characteristics can be predicted accurately and flexibly. Consequently, efficient prediction of landslide risk during normal times becomes possible.

[0014] Referring to Figure 1, the configuration of the prediction device 10 according to this embodiment will be described. The prediction device 10 comprises a control unit 11, a storage unit 12, a communication unit 13, an input unit 14, and an output unit 15.

[0015] The control unit 11 includes at least one processor, at least one programmable circuit, at least one dedicated circuit, or any combination thereof. The processor is a general-purpose processor such as a CPU or GPU, or a dedicated processor specialized for a specific process. "CPU" is an abbreviation for Central Processing Unit. "GPU" is an abbreviation for Graphics Processing Unit. The programmable circuit is, for example, an FPGA. "FPGA" is an abbreviation for field-programmable gate array. The dedicated circuit is, for example, an ASIC. "ASIC" is an abbreviation for application specific integrated circuit. The control unit 11 controls each part of the prediction device 10 and executes processing related to the operation of the prediction device 10.

[0016] The memory unit 12 includes at least one semiconductor memory, at least one magnetic memory, at least one optical memory, or any combination thereof. The semiconductor memory is, for example, RAM, ROM, or flash memory. "RAM" is an abbreviation for Random Access Memory. "ROM" is an abbreviation for Read Only Memory. The RAM is, for example, SRAM or DRAM. "SRAM" is an abbreviation for Static Random Access Memory. "DRAM" is an abbreviation for Dynamic Random Access Memory. The ROM is, for example, EEPROM. "EEPROM" is an abbreviation for Electrically Erasable Programmable Read Only Memory. The flash memory is, for example, SSD. "SSD" is an abbreviation for Solid-State Drive. The magnetic memory is, for example, HDD. "HDD" is an abbreviation for Hard Disk Drive. The storage unit 12 functions, for example, as a main memory, auxiliary memory, or cache memory. The storage unit 12 stores information used for the operation of the prediction device 10 and information obtained through the operation of the prediction device 10.

[0017] The communication unit 13 includes at least one communication module. The communication module is, for example, a module compatible with a wired LAN communication standard such as Ethernet®, a wireless LAN communication standard such as IEEE 802.11, or a mobile communication standard such as LTE, 4G, or 5G. "LAN" is an abbreviation for local area network. "IEEE" is an abbreviation for Institute of Electrical and Electronics Engineers. "LTE" is an abbreviation for Long Term Evolution. "4G" is an abbreviation for 4th generation. "5G" is an abbreviation for 5th generation. The communication unit 13 receives information used in the operation of the prediction device 10 and transmits information obtained by the operation of the prediction device 10. The communication unit 13 enables the prediction device 10 to send and receive information with other devices via the network.

[0018] The network includes the Internet, at least one WAN, at least one MAN, or a combination thereof. "WAN" is an abbreviation for Wide Area Network. "MAN" is an abbreviation for Metropolitan Area Network. The network may also include at least one wireless network, at least one optical network, or a combination thereof. Wireless networks include, for example, ad hoc networks, cellular networks, wireless LANs, satellite communication networks, or terrestrial microwave networks.

[0019] The input unit 14 includes at least one input interface. The input interface may be, for example, a physical key, a capacitive key, a pointing device, a touchscreen integrated with a display, or a microphone. The input unit 14 accepts operations to input information used for the operation of the prediction device 10. Instead of being provided in the prediction device 10, the input unit 14 may be connected to the prediction device 10 as an external input device. Any connection method can be used, for example, USB, HDMI®, or Bluetooth®. "USB" is an abbreviation for Universal Serial Bus. "HDMI®" is an abbreviation for High-Definition Multimedia Interface.

[0020] The output unit 15 includes at least one output interface. The output interface is, for example, a display or a speaker. The display is, for example, an LCD or an organic EL display. "LCD" is an abbreviation for Liquid Crystal Display. "EL" is an abbreviation for Electro Luminescent. The output unit 15 outputs information obtained by the operation of the prediction device 10. Instead of being provided in the prediction device 10, the output unit 15 may be connected to the prediction device 10 as an external output device. Any connection method can be used, for example, USB, HDMI®, or Bluetooth®.

[0021] The functions of the prediction device 10 are realized by executing the program according to this embodiment on the processor acting as the control unit 11. In other words, the functions of the prediction device 10 are realized by software. The program causes the computer to perform the operations of the prediction device 10, thereby causing the computer to function as the prediction device 10. That is, the computer functions as the prediction device 10 by performing the operations of the prediction device 10 according to the program.

[0022] The program can be stored on a non-temporary computer-readable medium. Examples of non-temporary computer-readable mediums include flash memory, magnetic recording devices, optical discs, magneto-optical recording media, or ROM. The program can be distributed, for example, by selling, transferring, or lending portable media such as SD cards, DVDs, or CD-ROMs containing the program. "SD" is an abbreviation for Secure Digital. "DVD" is an abbreviation for Digital Versatile Disc. "CD-ROM" is an abbreviation for Compact Disc Read Only Memory. The program may also be distributed by storing it in server storage and transferring it from the server to other computers. The program may also be provided as a program product.

[0023] A computer, for example, stores a program stored on a portable medium or a program transferred from a server in its main memory. Then, the computer reads the program stored in the main memory with its processor and executes the processing according to the read program. The computer may also read the program directly from the portable medium and execute the processing according to the program. The computer may also execute the processing according to the received program sequentially each time a program is transferred to it from a server. Processing may also be performed by a so-called ASP type service, which does not transfer programs from the server to the computer, but realizes its function only through execution instructions and result acquisition. "ASP" is an abbreviation for Application Service Provider. A program includes information used for processing by an electronic computer that is equivalent to a program. For example, data that is not a direct instruction to the computer but has the nature of defining the computer's processing falls under "equivalent to a program".

[0024] Some or all of the functions of the prediction device 10 may be implemented by a programmable circuit or a dedicated circuit as the control unit 11. In other words, some or all of the functions of the prediction device 10 may be implemented by hardware.

[0025] Next, the operation of the prediction device 10 according to this embodiment will be described with reference to Figures 2 to 18.

[0026] [Prediction Process for Sediment Disaster Risk] Figure 2 is a flowchart showing an example of the overall flow of the sediment disaster risk prediction process. When the control unit 11 of the prediction device 10 receives an instruction from the user to execute the sediment disaster risk prediction process via the input unit 14, it starts the process of step S1.

[0027] In step S1, the control unit 11 predicts the risk of landslides at the first target location. The first target location is a location on the land to be predicted where the user wishes to predict the risk of landslides, as shown in Figure 3. The land to be predicted is, for example, a mountain slope. Although one first target location is shown in Figure 3, there may be multiple first target locations. Details of the process in step S1 will be described later with reference to Figures 4A and 4B.

[0028] In step S2, the control unit 11 predicts a sediment runoff route as shown in Figure 3 by extracting valley topography from the land to be predicted. When sediment runs off, the starting point of the sediment runoff route becomes the first target point, and the ending point of the sediment runoff route becomes the second target point, which will be described later. Details of the process in step S2 will be described later with reference to Figure 16.

[0029] In step S3, the control unit 11 predicts the risk of soil and sand reaching the second target location. The second target location is a location on the land to be predicted where the user wishes to predict the risk of soil and sand reaching, as shown in Figure 3. Although one second target location is shown in Figure 3, there may be multiple second target locations. Details of the process in step S3 will be described later with reference to Figures 17A and 17B.

[0030] In step S4, the control unit 11 predicts the risk of landslides in the target land based on the risk of landslides at the first target site, the route of soil runoff, and the risk of soil reaching the second target site. Details of the process in step S4 will be described later with reference to Figure 18.

[0031] [Prediction process for landslide risk at the first target site] Figures 4A and 4B are flowcharts illustrating an example of the flow of the landslide risk prediction process at the first target site. The flow shown in Figures 4A and 4B corresponds to the process of step S1 shown in Figure 2. The flow shown in Figures 4A and 4B includes the construction process of the first prediction model and the prediction process. The construction process of the first prediction model includes the processes of steps S11 to S15. The prediction process includes the processes of steps S21 to S25.

[0032] In step S11, the control unit 11 acquires landslide information indicating past landslides and elevation information indicating the elevation of the land. The control unit 11 may communicate with an external server device to receive landslide information and elevation information. The control unit 11 may accept user input of landslide information and elevation information via the input unit 14, or it may read landslide information and elevation information from the storage unit 12.

[0033] Landslide information consists of location information for multiple "first points" that are locations on land where landslides have occurred in the past, and information indicating whether or not damage has occurred at those multiple first points. Land where landslides have occurred in the past is, for example, a mountain slope. The multiple first points are, for example, points at predetermined intervals on land where landslides have occurred in the past. The land where landslides have occurred in the past may be multiple different plots of land. The land where landslides have occurred in the past may be the same plot of land where landslides occurred at different times. The presence or absence of damage at a first point specifically refers to whether or not a slope collapse occurred at that first point.

[0034] The elevation information is associated with position information of each point whose elevation is indicated by the elevation information. The position information is, for example, longitude and latitude. The elevation information may be information based on publicly available map information such as mesh information. As another example, the elevation information may be information on actually measured values obtained by surveying. In this case, by acquiring the position information of the surveyed location while surveying the elevation, the position information can be associated with the elevation. The elevation information may be information obtained by analyzing a captured image captured by a camera or a depth image acquired by a ToF (Time of Flight) camera. In this case, position information can be associated with elevation by analyzing the captured image or the depth image. The elevation information and the position information may be information measured by a satellite.

[0035] The elevation information may be information based on point cloud data obtained by laser measurement or the like. In this case, by adding position information to the point cloud data based on the reference position of the laser measuring instrument, position information can be associated with the elevation.

[0036] The size of each mesh indicated by the mesh information as the elevation information may be arbitrary. For example, mesh information as shown in FIG. 5 is acquired as the elevation information. In FIG. 5, the position information of the center of a rectangular mesh and the position information of each vertex of the rectangular mesh are each associated with each mesh. Specifically, in FIG. 5, the elevation of each mesh as the elevation information is elevation E ０ , elevation E １ , ..., elevation E ８ . These elevations E ０ , elevation E １ , ..., elevation E ８ are associated with position (X ０ , Y ０ ), position (X １ , Y １ ), ..., position (X ８ , Y ８ ), which are position information of the center of the mesh. Furthermore, these elevations E ０ , elevation E １ , ..., elevation E ８This includes the position information of each vertex of the mesh, which is the position (x ０１ y ０１ , x ０２ y ０２ , x ０３ y ０３ , x ０４ y ０４ ), ..., position (x ８１ y ８１ , x ８２ y ８２ , x ８３ y ８３ , x ８４ y ８４ ) is associated with it.

[0037] Landslide information may include information on the type of landslide, which will be the dependent variable described later. Examples of landslide types include debris flows, cliff collapses, and mudslides.

[0038] Landslide information may include rainfall amounts from past landslide events. The rainfall amounts from past landslide events may be arbitrary data, such as hourly rainfall or 24-hour rainfall.

[0039] Landslide information may include the name of the disaster in which the landslide occurred, or the name of the place where the disaster occurred. The place name may be a prefecture name or city name, etc.

[0040] In addition to elevation information, the control unit 11 may further acquire geological information that indicates the geology of the land and associates location information with it, and vegetation information that indicates the vegetation of the land and associates location information with it.

[0041] In step S12, the control unit 11 compares the location information of multiple first points indicated by the landslide information with the location information associated with the elevation information acquired in step S11, and acquires elevation information (hereinafter referred to as "first elevation information") for each of the multiple first points in a predetermined area including the first point. This predetermined area is the area including the first point and the area surrounding the first point. For example, the control unit 11 uses a GIS (Geographic Information System) to compare the location information of the first point with the location information associated with the elevation information and acquires the first elevation information. Here, the control unit 11 may arbitrarily set the predetermined area including the first point. Examples 1 to 3 of setting the predetermined area including the first point will be described below with reference to Figures 6, 7 and 8.

[0042] [Setting Example 1] As shown in Figure 6, the control unit 11 may set the mesh containing the first point and the meshes adjacent to the mesh containing the first point in a predetermined area indicated by a thick line. The meshes adjacent to the mesh containing the first point are those adjacent to the mesh containing the first point in the vertical, horizontal, and diagonal directions. In this case, the first elevation information includes the elevation associated with the mesh containing the first point and the meshes adjacent to the mesh containing the first point. For example, nine elevations related to nine meshes in the predetermined area indicated by a thick line in Figure 6 are acquired as the first elevation information.

[0043] [Setting Example 2] The control unit 11 may be set to a predetermined area, as shown in Figure 7, which is within a predetermined distance from the first point and is indicated by a thick line. The predetermined distance may be set according to the scale of the predicted landslide, etc. In this case, the first elevation information includes the elevation associated with each mesh that is included in whole or in part within the predetermined area.

[0044] [Setting Example 3] The control unit 11 may be set to a predetermined area indicated by a thick line, which is defined by a certain policy from the first point, as shown in Figure 8. The policy may be set according to the scale of the predicted landslide, etc. In this case, the first elevation information includes the elevation associated with each mesh that is included in whole or in part within the predetermined area.

[0045] If the elevation information is based on point cloud data, the control unit 11 may define a region including the first point and its surroundings by providing a buffer for each point within an arbitrary distance from the first point, and set it as a predetermined region. In this case, the control unit 11 acquires the elevation of each point included in the set predetermined region as first elevation information.

[0046] In step S13, the control unit 11 generates first learning data that indicates multiple indicators related to the terrain and whether or not there is damage from landslides, based on the first elevation information and the landslide information. The control unit 11 generates first learning data for each of the multiple first locations and generates a group of first learning data consisting of the multiple first learning data.

[0047] The first training data may include the elevation associated with a mesh containing a first point in a predetermined region, as the index "elevation". The first training data may include the maximum elevation E in the predetermined region. ｍａｘ and minimum elevation E ｍｉｎ The first training data may include at least one of the following as an indicator. The first training data may include at least one of the mean, median, and standard deviation of the elevation associated with each point of each mesh or point cloud data in a predetermined region as the indicator "elevation".

[0048] The first learning data may include at least one of the maximum and minimum elevation differences in a predetermined region as indicators. For example, the control unit 11 may use the maximum elevation E ｍａｘ From minimum elevation E ｍｉｎ By subtracting this, the maximum elevation difference in a given region is calculated. The first training data may include at least one of the maximum elevation difference and the minimum elevation difference as indicators.

[0049] The first training data is the maximum value a of the slope angle from the first point in a predetermined region. ｍａｘ , Minimum value of the slope angle a ｍｉｎ and the difference in inclination angle (a ｍａｘ -a ｍｉｎ ) may include at least one of the following as an indicator. For example, the control unit 11 may include the maximum elevation E ｍａｘ Based on the elevation associated with the mesh containing the first point, the maximum value of the slope angle aｍａｘ The control unit 11 calculates the minimum elevation E. ｍｉｎ Based on the elevation associated with the mesh containing the first point, the minimum value of the slope angle a ｍｉｎ The control unit 11 calculates the maximum value a of the inclination angle. ｍａｘ Minimum angle of inclination a ｍｉｎ By subtracting this, the difference in inclination angle (a ｍａｘ -a ｍｉｎ The first training data may include, as an index "slope angle", at least one of the mean, median, and standard deviation of the slope angle between the first point and the center of each mesh in a predetermined area other than the mesh containing the first point.

[0050] If the elevation information is based on point cloud data, the first training data may include at least one of the mean, median, and standard deviation of the elevation of each point indicated by the first elevation information as the index "elevation". The first training data may also include at least one of the mean, median, and standard deviation of the slope angle between the first point or the point closest to the first point and each point included in the predetermined area as the index "slope angle".

[0051] The first training data may include at least one of the following as indicators: distance from a valley to point 1, distance from a river to point 1, distance from a forest to point 1, distance from a sloping point to point 1, and distance from a sloping line to point 1. Based on the first elevation information, the control unit 11 may use GIS to detect valleys, rivers, forests, sloping points, and sloping lines. Topographic features are represented for valleys, rivers, forests, sloping points, and sloping lines. A sloping point is a point where the slope becomes steep. A sloping line is a line connecting multiple sloping points.

[0052] The first training data may include the distance from the point with the highest standard deviation of the plane curvature or cross-sectional curvature in a predetermined region to the first point as an index. The first training data may also include the maximum, minimum, mean, median, or standard deviation of the plane curvature as the index "plane curvature". The first training data may also include the maximum, minimum, mean, median, or standard deviation of the cross-sectional curvature as the index "cross-sectional curvature".

[0053] The first learning data may include slope classification of a predetermined area as an indicator. The control unit 11 calculates the planar curvature and cross-sectional curvature of the predetermined area based on the first elevation information, and may classify the predetermined area into one of a plurality of pre-set slope types based on the planar curvature and cross-sectional curvature. Slope types include, for example, concave valley slopes, concave ridge slopes, convex valley slopes, or convex ridge slopes.

[0054] The first training data may include water systems in a predetermined area as indicators. The control unit 11 may detect water systems using GIS based on the first elevation information.

[0055] The first learning data may include rainfall amounts in a predetermined region when landslides have occurred in the past, as an indicator. The control unit 11 may use the rainfall data included in the landslide information acquired in step S11 described above as the indicator.

[0056] The first learning data may include, as an indicator, the name of a disaster or the place name where a landslide occurred in a predetermined region in the past, which has been assigned to the disaster. The control unit 11 may use the name of the disaster or the place name included in the landslide information acquired in step S11 described above as the indicator.

[0057] The first learning data may include the geology and vegetation of a predetermined region as indicators. The control unit 11 may identify the geology and vegetation of a predetermined region based on the geological information and vegetation information acquired in step S11. The first learning data may include the most common geology or vegetation among multiple meshes included in the predetermined region as the indicator "geology" or "vegetation". The first learning data may also include the geology or vegetation at the location of the first point, identified based on the location information, geological information, and vegetation information of the first point, as the indicator "geology" or "vegetation".

[0058] In step S14, the control unit 11 divides the first learning data set generated in step S13 into a plurality of first learning groups based on at least one index. The control unit 11 stores the plurality of first learning groups in the storage unit 12. The control unit 11 may accept a user specification of the at least one index via the input unit 14 and divide the first learning data set into a plurality of first learning groups based on the user specification index. The control unit 11 may divide the first learning data set into a plurality of first learning groups based on one or more indexes common to the plurality of indexes of the second learning data set described below. One first learning data is associated with one first location. One first learning group contains at least one first learning data. Therefore, one first learning group is associated with one or more first locations from the plurality of first locations described above.

[0059] The control unit 11 divides the first learning data group based on the category or numerical value of the index. For example, the control unit 11 divides the first learning data group according to the numerical value of the index "altitude". For example, the control unit 11 divides the first learning data group into four first learning groups according to three thresholds from the first threshold to the third threshold. The threshold values ​​may be set in advance and stored in the memory unit 12. For example, if the first threshold is 200m, the second threshold is 400m, and the third threshold is 600m, the control unit 11 divides the first learning data where the index "altitude" is 0 or greater and less than the first threshold into the first first learning group, the first learning data where it is 1 or greater and less than the second threshold into the second first learning group, the first learning data where it is 2 or greater and less than the third threshold into the third first learning group, and the first learning data where it is 3 or greater and greater than the third threshold into the fourth first learning group.

[0060] As an example, the control unit 11 divides the first training data group according to the indicator "disaster name". For example, if the disaster name includes three items, such as "XX heavy rain", "YY heavy rain", and "ZZ heavy rain", the control unit 11 divides the first training data group into three first training groups according to the disaster name of each training data item.

[0061] As an example, the control unit 11 divides the first training data set according to the index "slope classification". For example, if the slope classification includes four types: "concave valley slope", "concave ridge slope", "convex valley slope", and "convex ridge slope", the control unit 11 divides the first training data set into four first training groups according to the slope classification of each training data.

[0062] As an example, the control unit 11 divides the first training data set according to the index "place name where the disaster occurred". For example, if the place name where the disaster occurred includes N prefecture names, the control unit 11 divides the first training data set into N first training groups according to the place name where the disaster occurred in each training data.

[0063] Not limited to the foregoing, the control unit 11 may divide the first learning data group according to the following indicators: "maximum elevation," "minimum elevation," "elevation difference," "maximum elevation difference," "minimum elevation difference," "angle of inclination," "angle of inclination difference," "maximum angle of inclination," "minimum angle of inclination," "distance from valley to first point," "distance from river to first point," "distance from forest to first point," "distance from breakpoint to first point," "distance from breakline to first point," "distance from point with high standard deviation of plane curvature to first point," "distance from point with high standard deviation of cross-sectional curvature to first point," "plane curvature," "cross-sectional curvature," "water system," "rainfall when landslides occurred in the past," "geology," or "vegetation."

[0064] Figure 9 shows the first set of training data, in which each training data is divided into multiple first training groups LG_1 to LG_4 according to the three threshold values ​​of the "altitude" index mentioned above. Referring to Figure 9, training data with an "altitude" index of 0m or more and less than 200m are divided into the first training group LG_1, training data with an "altitude" index of 200m or more and less than 400m are divided into the first training group LG_2, training data with an "altitude" index of 400m or more and less than 600m are divided into the first training group LG_3, and training data with an "altitude" index of 600m or more are divided into the first training group LG_4.

[0065] The control unit 11 may divide the first learning data set into multiple first learning groups based on all indicators. In this case, the control unit 11 divides the first learning data set separately based on multiple indicators. For example, when all indicators are "altitude" and "disaster name", the control unit 11 divides the first learning data set into four first learning groups LG_1 to LG_4 according to three threshold values ​​for the indicator "altitude". The control unit 11 may also separately divide the first learning data set into three first learning groups according to the indicator "disaster name". The control unit 11 may store each of the seven first learning groups in the storage unit 12.

[0066] In step S15, the control unit 11 constructs a first prediction model for each of the multiple first learning groups that were divided in step S14, in order to predict the risk of landslides. As a result, multiple first prediction models are constructed. Hereafter, these multiple first prediction models will be collectively referred to as the first prediction model group. Each constructed first prediction model outputs the risk of landslides when multiple indicators related to topography are input. Specifically, the risk of landslides is the probability that damage will occur due to a landslide. The control unit 11 stores the first prediction model group in the storage unit 12.

[0067] The control unit 11 constructs a first prediction model using machine learning or statistical methods, with each of the multiple indicators of each first learning group as explanatory variables and the presence or absence of landslide damage at the first location as the objective variable. The control unit 11 may also use the type of landslide disaster included in the landslide information obtained in step S11 as the objective variable. The control unit 11 may also use the three stages of debris flow, landslide, and mudslide as the objective variables. The machine learning used to construct the first prediction model is, for example, supervised machine learning such as gradient boosting decision trees, random forests, or neural networks.

[0068] The statistical methods used to construct the first prediction model include, for example, methods that use a model equation given by a function such as a linear or sigmoid curve. In this case, the first prediction model is given in the form of a model equation.

[0069] Referring to Figure 9, the control unit 11 constructs the first first prediction model using the first training data, which includes first training data numbers 1, 4, and 5 belonging to the first training group LG_1. The control unit 11 constructs the second first prediction model using the first training data, which includes first training data numbers 6 and 7 belonging to the first training group LG_2. The control unit 11 constructs the third first prediction model using the first training data, which includes first training data number 3 belonging to the first training group LG_3. The control unit 11 constructs the fourth first prediction model using the first training data, which includes first training data number 2 belonging to the first training group LG_4.

[0070] Even when the first training data set is divided based on multiple indicators, the control unit 11 constructs a first prediction model based on each first training group. For example, if the first training data set is divided into seven first training groups, the control unit 11 constructs seven first prediction models based on each first training group.

[0071] In step S21, the control unit 11 acquires location information for at least one first target point. The control unit 11 may accept user input of location information for the first target point via the input unit 14, or it may read the location information for the first target point from the storage unit 12. The control unit 11 may also receive the location information for the first target point from an external server device.

[0072] The first target point is, as described above with reference to Figure 3, a point on the land to be predicted where the user wishes to predict the risk of landslides. At least one first target point may be multiple first target points. For example, if the user wants to predict the risk of landslides at 10m intervals on the land to be predicted, location information for multiple first target points will be acquired. In this case, the location information for multiple first target points will be location information for each 10m interval.

[0073] The location of the first target point does not have to coincide with the location of the first point obtained in step S11, or it may coincide with it. Also, the land to be predicted may be different from or the same as land where landslides have occurred in the past.

[0074] In step S22, the control unit 11 compares the location information of the first target point with the location information associated with the elevation information and obtains elevation information (hereinafter referred to as the first predicted elevation information) indicating the elevation of a predetermined area including the first target point (hereinafter referred to as the first prediction area). For example, the control unit 11 obtains the first predicted elevation information by comparing the location information of the first target point with the location information associated with the elevation information using GIS in the same manner as the processing in step S12. Here, the control unit 11 may set the first prediction area and obtain the first predicted elevation information in the same manner as the processing in step S12. If the control unit 11 has obtained location information for each of the multiple first target points in step S21, it sets the first prediction area and obtains the first predicted elevation information for each first target point.

[0075] The first prediction area may be a wider area than the predetermined area set in step S12 described above. Specifically, the first prediction area may be an area consisting of multiple clusters of areas of the same size as the predetermined area set in step S12 described above (hereinafter referred to as the expanded prediction area). Any method may be used to set the expanded prediction area. For example, when the control unit 11 acquires the location information of the first target point in step S21, it first sets the predetermined area using one of the methods described in setting examples 1 to 3 above. Next, the control unit 11 may set the expanded prediction area by setting an area of ​​the same size as the predetermined area so as to surround the predetermined area. Figure 10 shows an example of an expanded prediction area. The expanded prediction area consists of a large mesh M1, which is an area consisting of multiple meshes including a mesh containing the first target point, and large meshes M2 to M9, which are areas surrounding the large mesh M1. Each of the large meshes M1 to M9 may have the same size as the predetermined area set in step S12 described above. In this case, the control unit 11 acquires first predicted elevation information for each of the large meshes M1 to M9 by the same process as in step S12 described above, and determines a representative value for the first predicted elevation information for each mesh. That is, the control unit 11 determines nine representative values. Each representative value is the median or average value of multiple elevations included in the first predicted elevation information. The control unit 11 acquires these nine representative values ​​as elevation information indicating the elevation of the expanded prediction area (hereinafter referred to as expanded predicted elevation information).

[0076] In step S23, the control unit 11 generates first prediction data indicating the topography of the first prediction area based on the first predicted elevation information. The first prediction data includes a plurality of first prediction indicators associated with the first target point. Each indicator of the first prediction indicator may correspond one-to-one with a plurality of indicators shown in the first training data set generated in step S13. The control unit 11 may derive a plurality of first prediction indicators associated with one first target point, similar to the processing in step S13. The first prediction indicator may specifically be "altitude," "maximum altitude," "minimum altitude," "altitude difference," "maximum altitude difference," "minimum altitude difference," "angle of inclination," "angle of inclination," "maximum angle of inclination," "minimum angle of inclination," "distance from valley to first target point," "distance from river to first target point," "distance from forest to first target point," "distance from breakpoint to first target point," "distance from breakline to first target point," "distance from point with high standard deviation of plane curvature to first target point," "distance from point with high standard deviation of cross-sectional curvature to first target point," "plane curvature," "cross-sectional curvature," "slope classification," "water system," "predicted rainfall," "disaster name," "place name where the disaster occurred," "geology," or "vegetation," etc. "Predicted rainfall" is the amount of rainfall predicted for the day or time on which the risk of landslides is predicted. The control unit 11 may accept user input of "predicted rainfall" via the input unit 14. The control unit 11 may also receive "predicted rainfall" from an external weather information server.

[0077] Figure 11 shows an example of data for the first prediction. In Figure 11, there are N first prediction indicators, with indicator 1 being "elevation," indicator 2 being "elevation difference," and indicator N being "geology."

[0078] In step S22, if an expanded prediction area is set as the first prediction area and expanded prediction elevation information is acquired, the control unit 11 may generate expanded prediction data showing the topography of the expanded prediction area based on the expanded prediction elevation information. The expanded prediction data, like the first prediction data, includes a plurality of first prediction indicators associated with the first target point. For example, based on the nine representative values ​​included in the expanded prediction elevation information related to the expanded prediction area in Figure 10, the control unit 11 may generate expanded prediction data by deriving a plurality of first prediction indicators for each of the nine large meshes M1 to large mesh M9 in the same manner as the processing in step S13.

[0079] Figure 12 shows an example of data for expanded prediction related to the expanded prediction area in Figure 10. In Figure 12, there are N first prediction indices, with indicator 1 being "elevation", indicator 2 being "elevation difference", and indicator N being "elevation difference". Data numbers 1 to 9 in each row of Figure 12 correspond to the large meshes M1 to M9 in Figure 10, respectively.

[0080] In step S24, the control unit 11 selects one of the multiple first learning groups as the first selection group based on one or more first prediction indicators of the first prediction data, and identifies the first selection model constructed in step S15 based on the first selection group from the first prediction model group. If the control unit 11 has acquired location information for each of the multiple first target locations in step S21, it identifies the first selection model for each first target location.

[0081] Any method may be used to select the first selection group. For example, the control unit 11 selects the first learning group in which the first prediction indicator of the first prediction data belongs to the same category as the first selection group. For example, if the indicator "Disaster Name" of the first prediction indicator is "XX Heavy Rain," the control unit 11 selects the first learning group in which the indicator "Disaster Name" is "XX Heavy Rain" as the first selection group. For example, if the indicator "Slope Classification" of the first prediction indicator is "Concave Valley Slope," the control unit 11 selects the first learning group in which the indicator "Slope Classification" is "Concave Valley Slope" as the first selection group.

[0082] For example, the control unit 11 selects the first learning group as the first selection group, which is the group whose first prediction index value matches the first prediction data. The first learning data set is divided into four first learning groups LG_1 to LG_4 shown in Figure 9, according to the three threshold values ​​of the index "altitude" mentioned above, and as shown in Figure 11, the first prediction index "altitude" of the first prediction data is assumed to be 150m. In this case, the control unit 11 selects the first learning group LG_1 as the first selection group. The control unit 11 identifies the first prediction model constructed based on the first learning group LG_1.

[0083] In step S14, if the first learning data set has been divided into first learning groups based on multiple indicators, the control unit 11 may accept user specification of a first prediction indicator to be used for selecting the first selection group and select the first selection group based on the first prediction indicator specified by the user. The control unit 11 may output the indicators used to divide the first learning data set in step S14 via the output unit 15 and output a screen prompting the user to select an indicator. For example, suppose that four first learning groups divided according to three threshold values ​​of the indicator "altitude" and three first learning groups divided according to the indicator "disaster name" are each stored in the storage unit 12, and the first prediction indicator specified by the user is "disaster name". Also, suppose that the first prediction indicator "altitude" of the first prediction data is 250m and the first prediction indicator "disaster name" is "XX heavy rain". In this case, the control unit 11 selects the first learning group related to "XX Heavy Rain" as the first selection group from among the three first learning groups divided according to the indicator "Disaster Name".

[0084] For example, the control unit 11 may select the first selection group based on the sum of the squares of the differences between the average value of the indicators of the first learning group and the average value of the first prediction indicators of the augmented prediction data. In this case, there is a one-to-one correspondence between the multiple indicators shown by the first learning data group and the multiple first prediction indicators. Specifically, there is a one-to-one correspondence between the multiple indicators of each first learning group and the multiple indicators of each data belonging to each first learning group and the multiple first prediction indicators.

[0085] The control unit 11 calculates the average value of each indicator based, for example, on the magnified prediction data in Figure 12. Figure 13 shows data to which the average value of each first prediction indicator has been assigned to the magnified prediction data in Figure 12. The control unit 11 divides the sum of the values ​​of "elevation" as indicator 1 for each of the nine data by 9, which is the number of meshes constituting the magnified prediction area, to obtain the average value P of "elevation". １ ＊ The control unit 11 calculates the average value P of the "tilt angle" as index 2. ２ ＊ , and the average value P of "altitude difference" as index N Ｎ ＊ The control unit 11 then derives the following. Next, the control unit 11 similarly calculates the average value for each indicator for each of the multiple first learning groups. Figure 14A shows data in which the average value of each indicator is assigned to multiple indicators of the first learning group LG_A, which consists of T learning data. T corresponds to the total number of first locations that correspond to the first learning group LG_A out of the multiple first locations. The control unit 11 divides the sum of the values ​​of "altitude" as indicator 1 for each of the T data by T to obtain the average value A of "altitude". １ ＊ Similarly, the control unit 11 calculates the average value A of the "tilt angle" as index 2. ２ ＊ , and the average value A of "altitude difference" as index N Ｎ ＊ The following is derived. Figure 14B shows data in which the average value of each indicator is assigned to multiple indicators of the first learning group LG_B, which consists of U learning data. U corresponds to the total number of first locations that correspond to the first learning group LG_B out of a plurality of first locations. The control unit 11 divides the sum of the "altitude" values ​​of indicator 1 for each of the U data by U to obtain the average value B of "altitude". １ ＊ The control unit 11 calculates the average value B of the "tilt angle" as index 2. ２ ＊ , and the average value B of "altitude difference" as index N Ｎ ＊is derived. The control unit 11 calculates an average value for each index in the same manner for all of the first learning groups divided and generated in step S14. In this way, for each of the plurality of first learning groups, the control unit 11 calculates the average value by dividing the total value of each of the plurality of indices indicated by the first learning data group by the total number of corresponding first points among the plurality of first points. For each index, the control unit 11 calculates a squared difference value obtained by squaring the difference between the average value of the corresponding first prediction index of the expansion prediction data and said average value. When there are N plurality of indices, the squared difference values are calculated up to the N-th index. The control unit 11 calculates the sum of the N calculated squared difference values. For example, the sum A of squared difference values with the first learning group LG_A ｒ is calculated by the following Equation 1, and the sum B of squared difference values with the second learning group LG_B ｒ is calculated by the following Equation 2.

[0086] [Formula 1] A ｒ = (P １ ＊ − A １ ＊ ) ２ + (P ２ ＊ − A ２ ＊ ) ２ + ... (P Ｎ ＊ − A Ｎ ＊ ) ２ (Equation 1)

[0087] [Formula 2] B ｒ = (P １ ＊ − B １ ＊ ) ２ + (P ２ ＊ − B ２ ＊ ) ２ + ... (P Ｎ ＊ − B Ｎ ＊ ) ２(Formula 2)

[0088] The control unit 11 selects the first learning group with the smallest sum of squared difference values ​​as the first selected group. For example, if the first learning groups generated by splitting in step S14 consist only of the first learning group LG_A and the first learning group LG_B, the sum B ｒ Total A ｒ When it is smaller, the control unit 11 selects the first learning group LG_B as the first selection group. In this way, the control unit 11 calculates the sum of the squared differences of the N first prediction indices A ｒ and B ｒ Based on this, the first selection group is selected. The control unit 11 identifies the first prediction model constructed based on the first learning group LG_B.

[0089] The control unit 11 may select the first selection group based on the sum of the squares of the difference between the average value of the indicators of the first learning group and the first prediction indicator of the first prediction data. For example, if the indicator 1 of the first prediction data is "altitude", １ , the "angle of inclination" as indicator 2 is S ２ , and the "altitude difference" as index N is S Ｎ Assumes that this is the case. The control unit 11 calculates a squared difference value for each indicator by squaring the difference between the first prediction data and the corresponding first prediction indicator. When there are N indicators, the squared difference value is calculated up to the Nth indicator. The control unit 11 calculates the sum of the calculated N squared difference values. Specifically, the control unit 11 performs the S １ S ２ From S Ｎ And, as mentioned above, the sum of the squared differences between the mean values ​​of each indicator of the first learning group LG_A and C. ｒ The sum of the squared differences between the mean values ​​of each indicator in the first learning group LG_B and the given values, D, is calculated using the following equation 3. ｒ This is calculated using the following formula 4.

[0090] [Math 3] C ｒ = (S １ ― A １ ＊ ) ２ + (S ２ ― A ２ ＊ ) ２ +...(S Ｎ ― A Ｎ ＊ ) ２ (Formula 3)

[0091] [Mathematics 4] D ｒ = (S １ ― B １ ＊ ) ２ + (S ２ ― B ２ ＊ ) ２ +...(S Ｎ ― B Ｎ ＊ ) ２ (Formula 4)

[0092] The control unit 11 selects the first learning group with the smallest sum of squared difference values ​​as the first selected group. For example, if the first learning groups generated by splitting in step S14 consist only of the first learning group LG_A and the first learning group LG_B, the sum D ｒ Total C ｒ When the value is smaller, the control unit 11 selects the first learning group LG_B as the first selection group. The control unit 11 identifies the first prediction model constructed based on the first learning group LG_B.

[0093] For example, the control unit 11 may select the first selection group based on a comparison between a regression line created based on two indicators of the first learning group (hereinafter referred to as the learning regression line) and a regression line created based on two first prediction indicators of the expansion prediction data (hereinafter referred to as the prediction regression line). The two indicators of the first learning group and the two first prediction indicators of the expansion prediction data are common. The control unit 11 may use one of the two common indicators as the horizontal axis and the other as the vertical axis to obtain the learning regression line and the prediction regression line, respectively, using the least squares method, and determine the degree of similarity between the learning regression line and the prediction regression line. Any method may be used to determine the degree of similarity. For example, the control unit 11 may determine that the degree of similarity is higher the smaller the difference in slope and intercept between the learning regression line and the prediction regression line. Figure 15A shows an example of a prediction regression line created based on expansion prediction data. Figure 15B shows an example of a learning regression line created based on the first learning group LG_A. Figure 15C shows an example of a learning regression line created based on the first learning group LG_B. The control unit 11 compares the regression lines in Figures 15A, 15B, and 15C and determines that the learning regression line in Figure 15B has a higher similarity to the prediction regression line in Figure 15A than the learning regression line in Figure 15C. In this case, the control unit 11 selects the first learning group LG_A as the first selection group. The control unit 11 identifies the first prediction model constructed based on the first learning group LG_A.

[0094] In step S25, the control unit 11 predicts the risk of landslide at the first target site using the first prediction model identified in step S24. The control unit 11 predicts the risk of landslide at the first target site by inputting the first prediction indicator of the first prediction data generated in step S23 into the first prediction model. If the control unit 11 generated expanded prediction data in step S23, it may also predict the risk of landslide at the first target site by inputting the average or median value of each indicator of the expanded prediction data into the first prediction model. If the first prediction indicator of the expanded prediction data includes indicators other than numerical values, such as "geology," the control unit 11 may input the most common indicator among the data constituting the expanded prediction data into the first prediction model. The control unit 11 may output the predicted risk of landslide at the first target site to the output unit 15 along with a map showing the location information of the first target site. If the control unit 11 has acquired location information for each of the multiple first target locations in step S21, it uses the first prediction model identified for each first target location to predict the risk of landslides at each first target location. After that, the prediction device 10 terminates the process of predicting the risk of landslides at the first target locations.

[0095] [Prediction process for sediment runoff routes] Figure 16 is a flowchart showing an example of the flow of the sediment runoff route prediction process. The flow shown in Figure 16 corresponds to the process in step S2 shown in Figure 2.

[0096] In step S31, the control unit 11 acquires location information for multiple candidate locations. Candidate locations are locations in the land to be predicted that are candidates for valleys. The control unit 11 may acquire location information for multiple candidate locations by receiving it from the user via the input unit 14.

[0097] In step S32, the control unit 11 compares the location information of the candidate point with the location information associated with the elevation information acquired in step S11, and acquires elevation information (hereinafter referred to as route elevation information) indicating the elevation of a predetermined area (hereinafter referred to as the route area) that includes the candidate point. For example, the control unit 11 may set the route area and acquire route elevation information in the same manner as the processing in step S12.

[0098] In step S33, the control unit 11 generates valley topography data that indicates the characteristics of valley topography in the route area based on the route elevation information. The control unit 11 may derive a plurality of indicators associated with a single candidate point in the same manner as the processing in step S13. These plurality of indicators are indicators that indicate the characteristics of valley topography. As an example, the valley topography data may include at least one of the slope angle, curvature, slope classification, and water system of the route area as indicators. The control unit 11 may calculate the slope angle and curvature of the route area based on the route elevation information. The control unit 11 may classify the route area into one of a plurality of pre-set slope types based on the route elevation information. The control unit 11 may identify the water system of the route area based on the route elevation information.

[0099] In step S34, the control unit 11 extracts valley topography from the land to be predicted based on the valley topography data generated in step S33, and determines the extracted valley topography as the sediment runoff route through which sediment will flow. As an example, the control unit 11 detects depressions in the topography based on at least one of the slope angle, curvature, slope classification, and water system of the topography data. The control unit 11 extracts valley topography by grouping the detected depressions into a single set. In extracting valley topography, the control unit 11 may group the depressions into a set of lines, into a set of meshes, or into a set of polygons. After that, the prediction device 10 finishes the sediment runoff route prediction process.

[0100] [Prediction process for the risk of sediment reaching the second target site] Figures 17A and 17B are flowcharts illustrating an example of the flow of the prediction process for the risk of sediment reaching the second target site. The flow shown in Figures 17A and 17B corresponds to the process of step S3 shown in Figure 2. The flow shown in Figures 17A and 17B includes the process of constructing the second prediction model and the prediction process. The process of constructing the second prediction model includes the processes of steps S41 to S45. The prediction process includes the processes of steps S51 to S55.

[0101] In step S41, the control unit 11 acquires sediment arrival information indicating past sediment arrivals. The control unit 11 may communicate with an external server device to receive sediment arrival information. The control unit 11 may accept user input of sediment arrival information via the input unit 14, or it may read sediment arrival information from the storage unit 12.

[0102] Sediment arrival information consists of location information for multiple second points, which are multiple locations on land where sediment arrival has occurred in the past, and information indicating whether or not damage has occurred at these multiple second points. The multiple second points are, for example, locations at predetermined intervals on land where sediment arrival has occurred in the past. The land where sediment arrival has occurred in the past may be the same land as the land where the aforementioned landslide occurred in the past, or it may be a different land. The land where sediment arrival has occurred in the past may be multiple different land areas. The land where sediment arrival has occurred in the past may be the same land where sediment arrival occurred at different dates and times. The presence or absence of damage at the second points specifically refers to whether or not sediment accumulation has occurred at those second points.

[0103] The sediment arrival information may include information on the type of sediment-related disaster that serves as the dependent variable. Examples of sediment-related disaster types include debris flows, landslides, and mudslides.

[0104] Information on the arrival of sediment may include rainfall amounts from past events when sediment arrivals occurred. The rainfall amounts from past events when sediment arrivals occurred may be any data, such as hourly rainfall or 24-hour rainfall.

[0105] Information on the arrival of sediment may include data on the name of the disaster at the time the sediment arrived, or the place name where the disaster occurred. The place name may be a prefecture name or city name, etc.

[0106] In step S42, the control unit 11 compares the location information of a plurality of second points indicated by the sediment arrival information with the location information associated with the elevation information acquired in step S11, and acquires elevation information (hereinafter referred to as second elevation information) for a predetermined area including the second point for each of the plurality of second points. The predetermined area is an area that includes the second point and the surrounding area of ​​the second point. For example, the control unit 11 uses GIS to compare the location information of the second point with the location information associated with the elevation information and acquires the second elevation information. Here, the control unit 11 may set a predetermined area including the second point in the same manner as the processing in step S12 described above and acquire the second elevation information.

[0107] In step S43, the control unit 11 generates second learning data based on the second elevation information and sediment arrival information, which includes multiple indicators related to the terrain and whether or not there is damage from sediment arrival. The control unit 11 generates second learning data for each of the multiple second locations, and generates a second learning data group consisting of multiple second learning data. The multiple indicators related to the terrain in the second learning data may include one or more indicators common to the multiple indicators in the first learning data. The control unit 11 may generate the second learning data group using the same method as in step S13 described above.

[0108] In step S44, the control unit 11 divides the second learning data group generated in step S43 into a plurality of second learning groups based on at least one index. The control unit 11 may divide the second learning data group into a plurality of second learning groups using the same method as in step S14 described above. The control unit 11 may also divide the second learning data group into a plurality of second learning groups based on one or more indexes common to a plurality of indexes in the first learning data group. The control unit 11 stores the plurality of second learning groups in the storage unit 12.

[0109] In step S45, the control unit 11 constructs a prediction model for each of the multiple second learning groups (hereinafter referred to as the second prediction model) that predicts the risk of landslide arrival, using each of the multiple second learning groups that were divided in step S44. As a result, multiple second prediction models are constructed. Hereafter, multiple second prediction models will be collectively referred to as the second prediction model group. Each constructed second prediction model outputs the risk of landslide arrival when multiple indicators related to topography are input. Specifically, the risk of landslide arrival is the probability that damage will occur due to landslide arrival. The control unit 11 may construct the second prediction models using the same method as the first prediction model in step S15 described above. The control unit 11 stores the second prediction model group in the storage unit 12.

[0110] In step S51, the control unit 11 acquires location information for at least one second target point. The control unit 11 may accept user input of location information for the second target point via the input unit 14, or it may read the location information for the second target point from the storage unit 12. The control unit 11 may also receive location information for the second target point from an external server device. The second target point is a point on the land to be predicted where the user wishes to predict the risk of landslide arrival. The at least one second target point may be multiple second target points. For example, if the user wants to predict the risk of landslide arrival at 10m intervals on the land to be predicted, location information for multiple second target points may be acquired. In this case, the location information for multiple second target points will be location information at 10m intervals.

[0111] The location of the second target point does not have to coincide with the location of the second point obtained in step S41, or it may coincide with it. Also, the land subject to prediction may be different from or the same as land where sediment intrusion has occurred in the past.

[0112] In step S52, the control unit 11 compares the location information of the second target point with the location information associated with the elevation information and obtains elevation information (hereinafter referred to as the second predicted elevation information) indicating the elevation of a predetermined area including the second target point (hereinafter referred to as the second prediction area). For example, the control unit 11 obtains the second predicted elevation information by comparing the location information of the second target point with the location information associated with the elevation information using GIS in the same manner as the processing in step S12. If the control unit 11 has obtained location information for each of the multiple second target points in step S51, it sets a second prediction area for each second target point and obtains the second predicted elevation information.

[0113] In step S53, the control unit 11 generates second prediction data showing the topography of the second prediction area based on the second prediction elevation information. The second prediction data includes a plurality of indicators (hereinafter referred to as second prediction indicators) associated with the second target point. Each indicator of the second prediction indicator may correspond one-to-one with a plurality of indicators shown in the second training data set generated in step S43. The control unit 11 may derive a plurality of second prediction indicators associated with a single second target point, similar to the processing in step S13. The second prediction indicator may specifically be "altitude," "maximum altitude," "minimum altitude," "altitude difference," "maximum altitude difference," "minimum altitude difference," "angle of inclination," "angle of inclination," "maximum angle of inclination," "minimum angle of inclination," "distance from valley to second target point," "distance from river to second target point," "distance from forest to second target point," "distance from breakpoint to second target point," "distance from breakline to second target point," "distance from point with high standard deviation of plane curvature to second target point," "distance from point with high standard deviation of cross-sectional curvature to second target point," "plane curvature," "cross-sectional curvature," "slope classification," "water system," "predicted rainfall," "disaster name," "place name where disaster occurred," "geology," or "vegetation," etc. "Predicted rainfall" is the amount of rainfall predicted for the day or time on which the risk of landslide arrival is predicted. The control unit 11 may accept user input of "predicted rainfall" via the input unit 14. The control unit 11 may also receive "predicted rainfall" from an external weather information server.

[0114] The control unit 11 may generate second expansion prediction data using the same method as described above for expansion prediction data.

[0115] In step S54, the control unit 11 selects one of the multiple second learning groups as the second selection group based on one or more second prediction indicators of the second prediction data, and identifies the model constructed in step S45 based on the second selection group (hereinafter referred to as the second selection model) from the group of second prediction models. Any method may be used to select the second selection group, similar to the first selection group described above. If the control unit 11 has acquired location information for each of the multiple second target points in step S51, it identifies the second selection model for each second target point.

[0116] The control unit 11 may select a second selection group based on the sum of the squares of the differences between the average value of the indicators of the second learning group and the average value of the second prediction indicators of the second expansion prediction data, using the same method as in step S24. The control unit 11 may also select a second selection group based on a comparison between a regression line created based on the two indicators of the second learning group and a regression line created based on the two second prediction indicators of the second expansion prediction data, using the same method as in step S24.

[0117] In step S55, the control unit 11 predicts the risk of landslides reaching the second target location using the second prediction model identified in step S54. The control unit 11 predicts the risk of landslides reaching the second target location by inputting the second prediction indicators of the second prediction data generated in step S53 into the second prediction model. If the control unit 11 generated second expanded prediction data in step S53, it may also predict the risk of landslides reaching the second target location by inputting the average or median value of each indicator of the second expanded prediction data into the second prediction model. The control unit 11 may output the predicted risk of landslides reaching the second target location to the output unit 15 along with a map showing the location information of the second target location. If the control unit 11 acquired location information for each of the multiple second target locations in step S51, it predicts the risk of landslides reaching each second target location using the second prediction model identified for each second target location.

[0118] [Prediction Process for Sediment-Related Disaster Risk] Figure 18 is a flowchart showing an example of the flow of the prediction process for sediment-related disaster risk. The flow shown in Figure 18 corresponds to the process in step S4 shown in Figure 2. The sediment-related disaster risk calculated by the following process is the risk calculated based on the risk of landslide and the risk of sediment reaching the area.

[0119] In step S61, the control unit 11 identifies the calculation unit for calculating the risk of sediment-related disasters. First, the control unit 11 detects a first target point connected to the upper part of the sediment runoff route determined in step S34, and a second target point connected to the lower part of the sediment runoff route. If the position of the upper part of the sediment runoff route and the position of the first target point coincide, the control unit 11 assumes that the first target point is connected to the upper part of the sediment runoff route. Even if the position of the upper part of the sediment runoff route and the position of the first target point do not coincide, the control unit 11 may consider that the upper part of the sediment runoff route and the first target point are connected if the first target point is located within a first distance from the upper part of the sediment runoff route. If the position of the lower part of the sediment runoff route and the position of the second target point coincide, the control unit 11 assumes that the second target point is connected to the lower part of the sediment runoff route. The control unit 11 may consider the lower part of the sediment outflow route and the second target point to be connected even if the lower part of the sediment outflow route and the second target point do not coincide, as long as the second target point is located within a second distance from the lower part of the sediment outflow route. The first distance and the second distance may be set in advance based on data from past sediment disasters and stored in the storage unit 12.

[0120] The control unit 11 identifies one of the four calculation units 1 to 4 described below as the unit for calculating the risk of sediment-related disasters.

[0121] Calculation unit 1 is a pattern of one first target point and one second target point connected by one sediment runoff route.

[0122] The calculation unit 2 is a pattern of one first target point connected by one sediment runoff route and F second target points (where F is an integer satisfying 2 ≤ F).

[0123] The calculation unit 3 is a pattern of G first target points (where G is an integer satisfying 2 ≤ G) connected by a single sediment runoff route, and one second target point.

[0124] The calculation unit 4 is a pattern of P first target points (where P is an integer satisfying 2 ≤ P) and Q second target points (where Q is an integer satisfying 2 ≤ Q) connected by a single sediment runoff route.

[0125] In step S62, the control unit 11 calculates the landslide risk based on the calculation unit identified in step S61.

[0126] [Calculation Unit 1] As described above, Calculation Unit 1 is a pattern of one first target point and one second target point connected by one sediment runoff route. Hereinafter, the risk of sediment collapse at the first target point is "Sediment Collapse Risk R １ It is stated that the risk of sediment reaching the second target site is "Sediment Reach Risk R ２ It is written as follows:

[0127] In calculation unit 1, the control unit 11 calculates the landslide risk R using equation (2). R = aR １ +bR ２ (2) In equation (2), the coefficient a is the risk of landslides R １ This is a weighting coefficient for R. The coefficient b is the risk of sediment intrusion. ２ These are weighting coefficients. Coefficients a and b may be set based on the type of landslide. However, coefficients a and b may also be set to 1 (a = b = 1).

[0128] As another example, the control unit 11 determines the risk of landslides R, and the risk of soil collapse R. １ and risk of sediment intrusion R ２ It may also be calculated as the maximum value among them (R = MAX(R １ , R ２ )). As another example, the control unit 11 determines the risk of sediment-related disasters R, and the risk of landslides R. １ and risk of sediment intrusion R ２ It may also be calculated as the average value (R = AVERAGE(R １ , R ２ )).

[0129] [Calculation Unit 2] As described above, Calculation Unit 2 is a pattern of one first target point and F second target points connected by one sediment runoff route. Below, the sediment reach risk R to the F second target points is calculated. ２ Of these, the risk R of soil and debris reaching the i-th (where i is an integer satisfying 1 ≤ i ≤ F) second target point. ２ "Risk of sediment arrival" ２ｉ It is written as follows:

[0130] In calculation unit 2, there is one first target location where the risk of landslide is predicted, while there are multiple second target locations where the risk of soil arrival is predicted. Therefore, the control unit 11 calculates the soil arrival risk R of the i-th second target location if a landslide occurs at one first target location. ３ｉ The sediment disaster risk R may be calculated as the sediment disaster risk R. In this case, the control unit 11 calculates the sediment disaster risk R as the sediment arrival risk R. ３１ ~R ３Ｆ This calculates the risk of soil landslides reaching each of the F second target locations when a landslide occurs at one first target location.

[0131] The control unit 11 calculates the risk R of soil reaching the i-th second target location if a landslide occurs at one first target location. ３ｉ This is calculated using equation (3). R ３ｉ = cR １ +dR ２ｉ (3) In equation (3), the coefficient c is the risk of landslides R １ This is a weighting coefficient for R. The coefficient d is the risk of sediment intrusion R. ２ｉ These are weighting coefficients. Coefficients c and d may be set based on the type of sediment-related disaster. However, coefficients c and d may also be set to 1 (c = d = 1).

[0132] As another example, the control unit 11 determines the risk of landslides R, and the risk of soil collapse R. １ and risk of sediment intrusion R ２１ ~R ２Ｆ It can also be calculated as the sum of (R = SUM(R １ , R ２１ ~R２Ｆ )). As another example, the control unit 11 determines the risk of sediment-related disasters R, and the risk of landslides R. １ And, the risk of sediment intrusion R ２１ ~R ２Ｆ It may also be calculated as the sum of the maximum value among them (R = R １ +MAX(R ２１ ~R ２Ｆ )). As another example, the control unit 11 determines the risk of sediment-related disasters R, and the risk of landslides R. １ And, the risk of sediment intrusion R ２１ ~R ２Ｆ It may also be calculated as the sum of the average values ​​(R = R １ +AVERAGE(R ２１ ~R ２Ｆ )).

[0133] [Calculation Unit 3] As described above, Calculation Unit 3 is a pattern of G first target points and one second target point connected by one sediment runoff route. Below, the sediment collapse risk R of the G first target points is calculated. １ Of these, the landslide risk R of the j-th (where j is an integer satisfying 1 ≤ j ≤ G) first target site １ "Landslide risk R １ｊ It is written as follows:

[0134] In calculation unit 3, there are multiple first target locations where the risk of landslides is predicted, while there is only one second target location where the risk of soil arrival is predicted. Therefore, the control unit 11 calculates the soil arrival risk R, which is the risk of soil reaching one second target location if a landslide occurs at the j-th first target location. ４ｊ The sediment disaster risk R may be calculated as the sediment disaster risk R. In this case, the control unit 11 calculates the sediment disaster risk R as the sediment arrival risk R. ４１ ~R ４Ｇ This calculates the risk of soil reaching one second target site if a landslide occurs at each of the G first target sites.

[0135] The control unit 11 determines the risk R of soil reaching one second target location if a landslide occurs at the j-th first target location. ４ｊ This is calculated using equation (4). R ４ｊ= eR １ｊ +fR ２ (4) In equation (4), the coefficient e is the risk of landslides R １ｊ This is a weighting coefficient for R. The coefficient f is the risk of sediment intrusion R. ２ These are weighting coefficients. Coefficients e and f may be set based on the type of landslide. However, coefficients e and f may also be set to 1 (e = f = 1).

[0136] As another example, the control unit 11 determines the risk of landslides R, and the risk of soil collapse R. １１ ~R １Ｇ and risk of sediment intrusion R ２ It can also be calculated as the sum of (R = SUM(R １１ ~R １Ｇ , R ２ )). As another example, the control unit 11 determines the risk of sediment-related disasters R, and the risk of landslides R. １１ ~R １Ｇ The maximum value among them, and the risk of sediment intrusion R ２ It can also be calculated as the sum of (R = MAX(R) １１ ~R １Ｇ ) + R ２ ). As another example, the control unit 11 determines the risk of sediment-related disasters R, and the risk of landslides R. １１ ~R １Ｇ The average value and the risk of sediment intrusion R ２ It may also be calculated as the sum of (R = AVERAGE(R) １１ ~R １Ｇ ) + R ２ ).

[0137] [Calculation Unit 4] As described above, Calculation Unit 4 is a pattern of P first target points and Q second target points connected by a single sediment runoff route. Below, the sediment collapse risk R of the P first target points is calculated. １ Of these, the landslide risk R of the first target site, which is the kth (where k is an integer satisfying 1 ≤ k ≤ P) １ "Landslide risk R １ｋ It is stated that the risk of soil and debris reaching Q second target locations is R. ２ Of these, the risk R of soil and debris reaching the m-th (where m is an integer satisfying 1 ≤ m ≤ Q) second target site. ２ "Risk of sediment arrival"２ｍ It is written as follows:

[0138] In calculation unit 4, there are multiple first target locations where the risk of landslides is predicted, and multiple second target locations where the risk of landslide arrival is predicted. Therefore, the control unit 11 calculates the landslide arrival risk R, which is the risk of landslides reaching the mth second target location if a landslide occurs at the kth first target location. ５ｋｍ The sediment disaster risk R may be calculated as the sediment disaster risk R. In this case, the control unit 11 calculates the sediment disaster risk R as the sediment arrival risk R. ５ｋｍ In other words, the risk of sediment intrusion R ５１１ ~R ５ＰＱ This calculates the risk of soil reaching the m-th target location if a landslide occurs at the k-th first target location.

[0139] The control unit 11 determines the risk R of soil reaching the m-th second target location if a landslide occurs at the k-th first target location. ５ｋｍ This is calculated using equation (5). R ５ｋｍ = gR １ｋ +hR ２ｍ (5) In equation (5), the coefficient g is the risk of landslides R １ｋ This is a weighting coefficient for R. The coefficient h is the risk of sediment intrusion R. ２ｍ These are weighting coefficients. Coefficients g and h may be set based on the type of sediment-related disaster. However, coefficients g and h may also be set to 1 (g = h = 1).

[0140] As another example, the control unit 11 determines the risk of landslides R, and the risk of soil collapse R. １１ ~R １Ｐ and risk of sediment intrusion R ２１ ~R ２Ｑ It can also be calculated as the sum of (R = SUM(R １１ ~R １Ｐ , R ２１ ~R ２Ｑ )). As another example, the control unit 11 determines the risk of sediment-related disasters R, and the risk of landslides R. １１ ~R １Ｐ The maximum value among them, and the risk of sediment intrusion R ２１ ~R ２ＱIt can also be calculated as the sum of the maximum value among them (R = MAX(R １１ ~R １Ｐ ) + MAX (R ２１ ~R ２Ｑ )). As another example, the control unit 11 determines the risk of sediment-related disasters R, and the risk of landslides R. １１ ~R １Ｐ The average value and the risk of sediment intrusion R ２１ ~R ２Ｑ It may also be calculated as the sum of (R = AVERAGE(R) １１ ~R １Ｐ )+AVERAGE(R ２１ ~R ２Ｑ )).

[0141] In step S62, the control unit 11 may display the calculated landslide risk on the display of the output unit 15.

[0142] In step S63, the control unit 11 determines the standardized landslide risk R １ and risk of sediment intrusion R ２ The safety level of the sediment runoff route extracted as a valley topography is calculated. The control unit 11 calculates the safety level of the sediment runoff route according to the calculation method corresponding to the calculation unit specified in step S61, as described below. The control unit 11 may display the calculated safety level of the sediment runoff route on the display of the output unit 15. After that, the operation of the prediction device 10 ends.

[0143] [Calculation Unit 1] The control unit 11 calculates the landslide risk R of the first target site. １ The risk of soil and debris reaching the second target site R ２ The values ​​are standardized so that each falls within the range of 0 to 1. The landslide risk R after standardization. １ "Landslide risk r １ It is stated as follows: Standardized soil and sediment arrival risk R ２ "Risk of sediment arrival r" ２ It is written as follows:

[0144] The control unit 11 calculates the safety level S of the soil runoff route using equation (6). S = (1 - r １ ) × (1 - r ２ ) (6)

[0145] [Calculation Unit 2] The control unit 11 calculates the landslide risk R of the first target site. １ And the risk R of soil and debris reaching each of the F second target locations. ２１ ~R ２Ｆ The values ​​are standardized so that each falls within the range of 0 to 1. The landslide risk R after standardization. １ As mentioned above, "landslide risk r １ It is stated that the normalized soil and sediment reach risk R of the i-th (where i is an integer satisfying 1 ≤ i ≤ F) second target site out of F second target sites. ２ｉ "Risk of sediment arrival r" ２ｉ It is written as follows:

[0146] The control unit 11 calculates the safety level S of the soil runoff route using equation (7). S = (1 - r １ ) × [1 - { (1 - r ２１ ) × (1 - r ２２ ) × ... × (1 - r ２Ｆ )}] (7)

[0147] [Calculation Unit 3] The control unit 11 calculates the landslide risk R of each of the G first target locations. １１ ~R １Ｇ The risk of soil and debris reaching the second target site R ２ The values ​​are normalized so that each falls within the range of 0 to 1. The normalized landslide risk R of the j-th (where j is an integer satisfying 1 ≤ j ≤ G) first target site out of G first target sites. １ｊ "Landslide risk r １ｊ It is stated as follows: Standardized soil and sediment arrival risk R ２ As mentioned above, "the risk of sediment intrusion r ２ It is written as follows:

[0148] The control unit 11 calculates the safety level S of the soil runoff route using equation (8). S = [1 - {(1 - r １１ ) × (1 - r １２ ) × ... × (1 - r １Ｇ )}] × (1-r ２ ) (8)

[0149] [Calculation Unit 4] The control unit 11 calculates the landslide risk R of each of the P first target locations. １１~R １Ｐ And the risk R of soil and debris reaching each of the Q second target locations. ２１ ~R ２Ｑ The values ​​are normalized so that each falls within the range of 0 to 1. The normalized landslide risk R of the k-th (where k is an integer satisfying 1 ≤ k ≤ P) first target site out of P first target sites. １ｋ "Landslide risk r １ｋ The normalized sediment intrusion risk R of the m-th (where m is an integer satisfying 1 ≤ m ≤ Q) second target site out of Q second target sites. ２ｍ "Risk of sediment arrival r" ２ｍ It is written as follows:

[0150] The control unit 11 calculates the safety level S of the soil runoff route using equation (9). S = [1 - {(1 - r １１ ) × (1 - r １２ ) × ... × (1 - r １Ｐ )}] ×[1-{(1-r ２１ ) × (1 - r ２２ ) × ... × (1 - r ２Ｑ )}] (9)

[0151] This disclosure is not limited to the embodiments described above. For example, two or more blocks described in the block diagram may be combined, or one block may be divided. Instead of executing two or more steps described in the flowchart in chronological order as described, they may be executed in parallel or in a different order, depending on the processing capacity of the device performing each step or as necessary. Other modifications are possible without departing from the spirit of this disclosure.

[0152] For example, in step S13 described above, first training data is generated based on first elevation information, and in step S23, first prediction data is generated based on first predicted elevation information. Also, in step S33 described above, valley topography data is generated based on route elevation information. Also, in step S43 described above, second training data is generated based on second elevation information, and in step S53, second prediction data is generated based on second predicted elevation information. However, the control unit 11 may generate each of the first training data, first prediction data, valley topography data, second training data, and second prediction data based on arbitrary information. For example, the control unit 11 may generate this data from publicly available data or image analysis, etc.

[0153] 10 Prediction device 11 Control unit 12 Storage unit 13 Communication unit 14 Input unit 15 Output unit

Claims

1. A prediction device comprising: a control unit that predicts the risk of landslides at a first location by constructing a prediction model for each of several first learning groups, each of which is divided into several first learning groups based on at least one indicator shown in the first learning data group, which includes several indicators related to topography and a first learning data group indicating whether or not there has been damage from landslides for each of several locations; and when several first indicators related to topography are obtained for a first location to be predicted, the control unit selects one of the several first learning groups as the first selected group based on one or more first indicators, and predicts the risk of landslides at the first location using the prediction model of the first selected group.

2. The control unit constructs a prediction model for each of several second learning groups that predict the risk of sediment intrusion, using each of several second learning groups obtained by dividing several indicators related to topography and a second learning data set indicating whether or not there is damage from sediment intrusion based on at least one indicator shown in the second learning data set, for each of several locations different from the aforementioned several locations; when several second indicators related to topography are obtained for a second location that is the target of prediction and is located in the same land as the first location, one of the several second learning groups is selected as the second selection group based on one or more second indicators; the prediction model for the second location is used to predict the risk of sediment intrusion; and the sediment disaster risk of the land is calculated based on the landslide risk of the first location and the sediment intrusion risk of the second location, according to claim 1.

3. The prediction device according to claim 2, wherein the plurality of indicators shown in the first learning data set and the plurality of indicators shown in the second learning data set include one or more common indicators, the plurality of first learning groups are groups obtained by dividing the first learning data set based on the one or more common indicators, and the plurality of second learning groups are groups obtained by dividing the second learning data set based on the one or more common indicators.

4. The prediction device according to any one of claims 1 to 3, wherein there is a one-to-one correspondence between the multiple indicators shown in the first learning data set and the multiple first indicators, and the control unit selects the first selection group based on the sum of the squared difference values ​​of the total of the multiple first indicators, when the average value obtained by dividing the sum of the multiple indicators shown in the first learning data set by the total number of corresponding locations among the multiple locations is defined as the squared difference value, and the corresponding first indicator.

Images