Information processing device, weather radar system, composite processing station, method and program

Abstract

The objective is to provide an information processing device, method, and program capable of detecting in advance the occurrence of characteristic precipitation areas. [Solution] The information processing device according to the embodiment includes a processing unit that detects a first precipitation area based on a first hourly cumulative value of rainfall at each of the multiple locations, which is based on rainfall data indicating hourly rainfall at the multiple locations, and detects the occurrence of a characteristic precipitation area by inputting first and second feature quantities calculated with respect to the detected first precipitation area into a first machine learning model.

Description

Technical Field

[0001] Embodiments of the present invention relate to an information processing apparatus, a weather radar system, a synthesis processing station, a method, and a program.

Background Art

[0002] In recent years, for example, in a characteristic precipitation area such as a linear precipitation band, it is known that a great deal of damage occurs.

[0003] In order to reduce such damage, it is useful to detect in advance the occurrence of the above-described characteristic precipitation area.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] Therefore, the problem to be solved by the present invention is to provide an information processing apparatus, a weather radar system, a synthesis processing station, a method, and a program capable of detecting in advance the occurrence of a characteristic precipitation area.

Means for Solving the Problems

[0006] The information processing apparatus according to the embodiment includes a processing unit that detects a first precipitation area based on a first time integrated value of rainfall amounts at each of the plurality of points based on rainfall data indicating rainfall amounts at each point at each time, and inputs first and second feature amounts calculated for the detected first precipitation area into a first machine learning model to detect the occurrence of a characteristic precipitation area.

Brief Description of the Drawings

[0007] [Figure 1]A block diagram showing an example of the functional configuration of the information processing device according to the embodiment. [Figure 2] A diagram showing an example of the system configuration of an information processing device. [Figure 3] A diagram illustrating an example of rainfall data. [Figure 4] A flowchart illustrating an example of the processing procedure for an information processing device. [Figure 5] A diagram illustrating precipitation areas. [Figure 6] A diagram to explain rainfall statistics. [Figure 7] A diagram showing an example of how detection results are displayed. [Figure 8] A block diagram showing another example of the functional configuration of an information processing device. [Figure 9] A diagram illustrating an example of an upwind region. [Figure 10] A block diagram showing yet another example of the functional configuration of an information processing device. [Figure 11] A diagram illustrating the features derived from prediction data. [Figure 12] A diagram showing an example of features input to the first machine learning model. [Figure 13] A diagram showing an example of features input to the second machine learning model. [Figure 14] A diagram showing an example of the configuration of a weather radar system. [Figure 15] A diagram showing an example of the configuration of a synthesis processing unit. [Modes for carrying out the invention]

[0008] The embodiments will be described below with reference to the drawings. Figure 1 is a block diagram showing an example of the functional configuration of the information processing device according to this embodiment. As shown in Figure 1, the information processing device 10 includes a rainfall data storage unit 11 and a processing unit 12.

[0009] The rainfall data storage unit 11 stores rainfall data indicating the rainfall for each time at a plurality of locations. The rainfall data stored in the rainfall data storage unit 11 is data indicating the rainfall calculated based on data observed at each time (hereinafter referred to as observation data) in a weather radar system not shown, for example, and is acquired from the weather radar system. In the weather radar system, as the weather radar for acquiring observation data, for example, a parabolic weather radar, a multi-parameter phased array weather radar (MP-PAWR), a phased array weather radar (PAWR), or the like may be adopted, or other types of weather radars may be adopted.

[0010] The processing unit 12 includes a precipitation area detection unit 121, a feature amount calculation unit 122, and a detection unit 123, and executes processing for detecting in advance the occurrence of a characteristic precipitation area using the rainfall data stored in the rainfall data storage unit 11. In other words, it can be said that the information processing apparatus 10 according to the present embodiment has a function as a weather prediction apparatus for predicting the occurrence of a characteristic precipitation area.

[0011] The characteristic precipitation area in the present embodiment corresponds to a rain area (dangerous precipitation area) accompanied by strong precipitation where damage occurs, and is, for example, an area (region) affected by a linear precipitation band, local heavy rain, or concentrated heavy rain.

[0012] In the following, it will be described as detecting the occurrence of a linear precipitation band. The linear precipitation band can be defined as a rain area with strong precipitation that extends linearly, about 50 to 300 km in length and about 20 to 50 km in width, created by an organized cumulonimbus cloud group in which developed rain clouds (cumulonimbus clouds) occur one after another in a row and pass through or stagnate at substantially the same location over several hours.

[0013] The precipitation area detection unit 121 acquires, for each of a plurality of points, a value obtained by time-integrating the rainfall amount at each point (hereinafter referred to as the time-integrated rainfall value) based on the rainfall data stored in the rainfall data storage unit 11. Based on the acquired time-integrated rainfall value, the precipitation area detection unit 121 detects, for example, an area including a plurality of points with a large time-integrated rainfall value as a precipitation area.

[0014] The feature quantity calculation unit 122 calculates feature quantities related to the precipitation area detected by the precipitation area detection unit 121. The feature quantities calculated by the feature quantity calculation unit 122 in this way correspond to an index (determination index) for determining whether a linear precipitation band occurs.

[0015] Note that there are still many unclear parts regarding the occurrence of a linear precipitation band, and it is difficult to detect the occurrence of the linear precipitation band with high accuracy using only a single feature quantity (determination index). Therefore, in this embodiment, at least two or more feature quantities are calculated.

[0016] The detection unit 123 holds a machine learning model prepared in advance, and detects the future occurrence of a linear precipitation band (hereinafter simply referred to as the occurrence of a linear precipitation band) by inputting the feature quantities calculated by the feature quantity calculation unit 122 into the machine learning model. Note that detecting the occurrence of a linear precipitation band in this embodiment corresponds to determining (discriminating) whether the precipitation area detected by the precipitation area detection unit 121 described above is a precipitation area leading to a linear precipitation band.

[0017] FIG. 2 shows an example of the system configuration of the information processing apparatus 10 shown in FIG. 1. The information processing apparatus 10 includes a CPU 10a, a nonvolatile memory 10b, a RAM 10c, a communication device 10d, and the like.

[0018] The CPU 10a is a processor for controlling the operation of various components within the information processing device 10. The CPU 10a may be a single processor or may consist of multiple processors. The CPU 10a executes various programs loaded from the non-volatile memory 10b into the RAM 10c. The programs executed by the CPU 10a include weather forecasting programs for detecting the occurrence of linear precipitation bands (i.e., predicting the weather).

[0019] The non-volatile memory 10b is a storage medium used as an auxiliary storage device. The RAM 10c is a storage medium used as a main storage device. Although only the non-volatile memory 10b and RAM 10c are shown in Figure 2, the information processing device 10 may also include other storage devices such as an HDD (Hard Disk Drive) and an SSD (Solid State Drive).

[0020] The communication device 10d is a device configured to perform wired or wireless communication.

[0021] Although not shown in Figure 2, the information processing device 10 may include, for example, input devices such as a mouse and keyboard, and display devices such as a display.

[0022] In this embodiment, the rainfall data storage unit 11 shown in Figure 1 is implemented by, for example, a non-volatile memory 10b or another storage device.

[0023] Furthermore, in this embodiment, the processing unit 12 shown in Figure 1 is implemented by at least one processor. The processor includes, for example, a control unit and an arithmetic unit, and is implemented by analog or digital circuits. The processor may be the CPU 10a described above, or it may be a general-purpose processor, a microprocessor, a digital signal processor (DSP), an ASIC, an FPGA, or a combination thereof.

[0024] Furthermore, part or all of the processing unit 12 may be implemented by having the CPU 10a (i.e., the computer of the information processing device 10) execute the weather forecast program described above, that is, by software. This weather forecast program may be stored and distributed on a computer-readable storage medium, or it may be downloaded to the information processing device 10 via a network. Part or all of the processing unit 12 may also be implemented by dedicated hardware, etc.

[0025] Here, with reference to Figure 3, an example of rainfall data stored in the rainfall data storage unit 11 will be explained.

[0026] As shown in Figure 3, the rainfall data in this embodiment is data in which the amount of rainfall at each of the multiple locations placed on the map is assigned to that location. In such rainfall data, each of the multiple locations can be represented by a color corresponding to the amount of rainfall. Each of the multiple locations placed on the map in the rainfall data corresponds to a grid point represented by the latitude and longitude defined on the map.

[0027] Furthermore, Figure 3 shows, for example, rainfall data indicating the amount of rainfall over a predetermined period of time, and the rainfall data storage unit 11 stores rainfall data like that shown in Figure 3 for each period of time. Specifically, the rainfall data storage unit 11 stores rainfall data from the past to the present, for example, at 5-minute intervals. In other words, in this embodiment, for example, rainfall data acquired from the weather radar system is sequentially stored in the rainfall data storage unit 11 as time progresses.

[0028] Note that the rainfall data shown in Figure 3 is just one example; the rainfall data can be in any format as long as it shows the rainfall at each of multiple locations.

[0029] Furthermore, the rainfall data shown in this embodiment may be, for example, ground rainfall (mm / h) corresponding to the amount of rain that falls on the ground (surface), or it may be vertically integrated liquid water content (VIL) including rainfall in the upper atmosphere.

[0030] Furthermore, although this description assumes that rainfall data is obtained from a weather radar system, the rainfall data in this embodiment may be, for example, rainfall data for each of multiple locations corresponding to grid points represented by latitude and longitude, generated based on data observed by multiple weather radars. Such rainfall data may also be obtained from an external server device or the like. Additionally, the rainfall data may be data measured (observed) using instruments other than weather radar, or processed data obtained by processing (correcting) such data.

[0031] Below, an example of the processing procedure of the information processing device 10 according to this embodiment will be described with reference to the flowchart in Figure 4.

[0032] First, the precipitation area detection unit 121 included in the processing unit 12 acquires rainfall data stored in the rainfall data storage unit 11 (step S1). As described above, the rainfall data storage unit 11 stores hourly rainfall data, and in step S1, rainfall data for each hour corresponding to a predetermined time period (for example, the time period from the current time to 60 minutes prior) is acquired.

[0033] Here, the rainfall data in this embodiment is data indicating the amount of rainfall at each of a plurality of points corresponding to grid points represented by latitude and longitude. The precipitation area detection unit 121 calculates the time-integrated rainfall value at one of the plurality of points (hereinafter referred to as the target point) based on the rainfall data acquired in step S1 (step S2). The time-integrated rainfall value calculated in step S2 is a value obtained by sequentially adding up the rainfall at the target point extracted from the rainfall data for each time acquired in step S1 (i.e., integrating over time). Specifically, the time-integrated rainfall value (mm) is calculated by the following formula (1).

number

[0034] In equation (1) above, T is the cumulative time, and R is the total time. t Δt represents the instantaneous value of rainfall (rainfall intensity) at time t, and Δt represents the observation frequency of the instantaneous value of rainfall to be accumulated over time. In this case, the accumulation time T is assumed to be, for example, 60 minutes or 180 minutes, and the observation frequency Δt is assumed to be, for example, 1 minute.

[0035] When the process in step S2 is executed, it is determined whether the process in step S2 has been executed for all of the above-mentioned locations (i.e., whether the cumulative rainfall value over time has been calculated for all locations) (step S3).

[0036] If it is determined that time-cumulative rainfall values ​​have not been calculated for any location (NO in step S3), the process returns to step S2 and is repeated. In this case, the process in step S2 is executed with the locations for which time-cumulative rainfall values ​​have not been calculated as target locations.

[0037] On the other hand, if it is determined that time-cumulative rainfall values ​​have been calculated at all locations (YES in step S3), the precipitation area detection unit 121 detects a precipitation area that includes at least some of the locations based on the time-cumulative rainfall values ​​at each of the locations calculated by repeatedly executing the process in step S2 (step S4). In step S4, multiple precipitation areas may be detected.

[0038] Here, with reference to Figure 5, the precipitation area detected in step S4 will be described. Figure 5 shows an example of hourly cumulative rainfall values ​​at each of several locations (i.e., grid points represented by latitude and longitude).

[0039] In this embodiment, for example, a region (a group of adjacent grid points) where points with a cumulative time rainfall value equal to or greater than a threshold value is detected as a precipitation area. The threshold value for cumulative time rainfall is, for example, 30 mm. According to this, the region 121a shown in Figure 5, where points with a cumulative time rainfall value of 30 mm or more are consecutive, can be detected as a precipitation area.

[0040] Here, we have explained that a region (a cluster of grid points) where the cumulative rainfall value over time exceeds a threshold is detected as a precipitation area. However, it is preferable to also set a threshold for the area of ​​this region. This area threshold could be, for example, 50 km². 2 According to this, for example, region 121b shown in Figure 5 (that is, area of ​​50 km²) 2 Areas smaller than a certain value are not detected as precipitation areas, and very small areas can be excluded from the precipitation area.

[0041] For example, the Japan Meteorological Agency defines a 3-hour cumulative rainfall value as 100 mm (33 mm / h) or more, and an area of ​​500 km². 2 While the above-mentioned region is detected as a linear precipitation band, in this embodiment, in order to detect the occurrence of the linear precipitation band at an early stage, a configuration is adopted that detects the precipitation area with a shorter accumulation time and targeting a smaller area.

[0042] Furthermore, although this explanation describes detecting a precipitation area as a continuous region (a cluster of grid points) where the cumulative hourly rainfall value is 30 mm or more, for example, points surrounding the detected precipitation area may also be included in the precipitation area. Specifically, as shown in Figure 5, a region 121c including points where the cumulative hourly rainfall value is 30 mm or more and points adjacent to those points where the cumulative hourly rainfall value is 20 mm or more may be detected as a precipitation area. In this case, a region that does not include any points where the cumulative hourly rainfall value is 30 mm or more, and only includes points where the cumulative hourly rainfall value is 20 mm or more, will not be detected as a precipitation area.

[0043] Furthermore, in the example shown in Figure 5, regions 121d and 121e will be detected as precipitation areas, similar to region 121c described above. However, if multiple regions are located close together (the distance between multiple regions is below a threshold), such as regions 121d and 121e, region 121f, which includes regions 121d and 121e, may be detected as a single precipitation area. In this case, regions 121d, 121e, and 121f may each be detected as different precipitation areas.

[0044] Furthermore, as described above, the detected precipitation area may be corrected to approximate a predetermined shape. Note that the region 121f shown above illustrates an example where the region including regions 121d and 121e is corrected to an ellipse shape. The corrected shape of the precipitation area may be any shape other than an ellipse.

[0045] Returning to Figure 4, the feature calculation unit 122 calculates features related to the precipitation area detected in step S4 (hereinafter referred to as the target precipitation area) (step S5). In step S5, for example, the rainfall statistics at each of the multiple locations included in the target precipitation area (hereinafter referred to as rainfall statistics) are calculated as features related to the target precipitation area.

[0046] The following explains the rainfall statistics calculated as characteristic features for the target precipitation area, referring to Figure 6.

[0047] First, the feature calculation unit 122 obtains rainfall data for each time period from the rainfall data storage unit 11 and calculates the time-cumulative rainfall value at each of the multiple locations by accumulating the rainfall (ground rainfall or VIL) at each of those locations.

[0048] Next, the feature calculation unit 122 extracts (acquires) the time-cumulative rainfall values ​​at each point included in the target precipitation area from among the multiple points where time-cumulative rainfall values ​​have been calculated as described above.

[0049] Furthermore, the feature calculation unit 122 calculates rainfall statistics based on the extracted time-cumulative rainfall values. In this case, the feature calculation unit 122 calculates the maximum value, average value, or sum of time-cumulative rainfall values ​​in the target precipitation area (a two-dimensional space corresponding to an area defined by latitude and longitude).

[0050] Generally, the average value of the time-cumulative rainfall in a target precipitation area is calculated, for example, by dividing the sum of the time-cumulative rainfall values ​​at each point included in the target precipitation area by the number of such points. However, the number of points from which this sum is divided may be the number of points among the multiple points included in the target precipitation area where the time-cumulative rainfall value is not zero (i.e., grid points where there are rainfall amounts greater than zero).

[0051] In this embodiment, the rainfall statistics of the target precipitation area calculated in this manner (i.e., the maximum value, average value, or sum of the time-integrated rainfall values ​​in the target precipitation area) are used as features related to the target precipitation area. However, the features related to the target precipitation area used in this embodiment may be other than rainfall statistics. Furthermore, in this embodiment, the above-mentioned rainfall statistics and at least two features selected from other features are used.

[0052] Returning to Figure 4, the detection unit 123 performs a process to detect the occurrence of a linear rainband based on the feature quantities calculated in step S5. In this case, the detection unit 123 uses a machine learning model to determine whether the target precipitation area is a precipitation area connected to a linear rainband (step S6).

[0053] The machine learning model used in this embodiment will be described below. For example, the occurrence of linear rainbands is announced by the Japan Meteorological Agency, but in this embodiment, the precipitation area that leads to a linear rainband is defined as the precipitation area that exists near the location where the linear rainband is to occur before the occurrence of the linear rainband (before the announcement by the Japan Meteorological Agency).

[0054] In this case, the machine learning model is assumed to be pre-generated (prepared) by using past rainfall data, for example, by tracing back through precipitation areas near a linear rainband in 5-minute increments from the time a linear rainband occurred in the past, and learning these precipitation areas as precipitation areas connected to the linear rainband.

[0055] Specifically, for example, if the rainfall data at the time the Japan Meteorological Agency announces the occurrence of a linear rainband is taken as the rainfall data at time t=0, then a precipitation area that overlaps with the linear rainband, at least partially, is detected based on the rainfall data at time t=-5 (i.e., 5 minutes before the occurrence of the linear rainband). The process for detecting this precipitation area is the same as the process in step S4 described above, and the detected precipitation area includes multiple locations where the time-cumulative rainfall value is equal to or greater than a threshold. Furthermore, a precipitation area that overlaps with the linear rainband, at least partially, may be, for example, a precipitation area that overlaps with the linear rainband by a predetermined percentage or more, or a precipitation area whose distance from the linear rainband (or its center or end) is less than or equal to a predetermined value.

[0056] As described above, if the detected precipitation area is designated as a training precipitation area, training data for the training precipitation area is created, which is a combination of features related to the training precipitation area (for example, rainfall statistics for the training precipitation area) and a label (for example, label "1") indicating that the training precipitation area is a precipitation area connected to a linear precipitation band (i.e., a linear precipitation band occurred). The rainfall statistics for the training precipitation area are the maximum, average, or total value of the time-cumulative rainfall in the training precipitation area, and correspond to the rainfall statistics calculated as features related to the target precipitation area as described above.

[0057] Here, we have described the case where training data for a training precipitation area is created using rainfall data at time t=-5. However, the creation of this training data for a training precipitation area is further carried out using rainfall data from earlier times than time t=-5 (for example, rainfall data at time t=-10, rainfall data at time t=-15, etc.). When using rainfall data at time t=-10, for example, a precipitation area that at least partially overlaps with the precipitation area detected based on the rainfall data at time t=-5 (i.e., the previous precipitation area) is detected, and training data for a training precipitation area is created using this detected precipitation area as the training precipitation area.

[0058] Furthermore, in this embodiment, the rainfall data (training data) used to train the machine learning model is, for example, rainfall data from the time a linear rainband occurred up to a maximum of 3 hours prior. Also, if no rainfall area is detected that overlaps with at least a portion of the previous rainfall area based on the above-mentioned rainfall data, training data for the training rainfall area using that rainfall data and rainfall data from earlier times does not need to be created.

[0059] Furthermore, while this section described the creation of training data for precipitation areas used by machine learning models to learn about precipitation areas connected to linear precipitation bands, training data for precipitation areas not connected to linear precipitation bands will also be created. Precipitation areas not connected to linear precipitation bands are defined as precipitation areas where linear precipitation bands did not form nearby even after a period of time had passed, or precipitation areas that are clearly located or time-dependent from linear precipitation bands. In this case, training data for precipitation areas not connected to linear precipitation bands will be created, consisting of a combination of features related to the precipitation area not connected to a linear precipitation band and a label indicating that the precipitation area is not connected to a linear precipitation band (for example, label "0").

[0060] By training a machine learning model based on the training data of the precipitation areas created above, and the training data of precipitation areas that do not connect to linear precipitation bands, it is possible to generate a machine learning model that, for example, outputs a label indicating whether or not a linear precipitation band will occur as a result of the development of a given precipitation area, given that a feature related to that precipitation area is input.

[0061] In this embodiment, as described above, the system is configured to detect (analytically extract) the precipitation area of ​​interest and learn its features (e.g., rainfall statistics). Therefore, compared to learning directly from rainfall data as shown in Figure 3, for example, it is possible to generate a highly accurate machine learning model even when the amount of training data is small.

[0062] Furthermore, in this embodiment, it is sufficient for the detection unit 123 to hold the machine learning model that has been trained, and the training of the machine learning model may be performed, for example, inside the information processing device 10 or outside the information processing device 10.

[0063] Furthermore, the training of the machine learning model in this embodiment is performed based on machine learning algorithms such as logistic regression, support vector machines, random forests, gradient boosting decision trees, or neural networks.

[0064] According to such a machine learning model, the detection unit 123 can determine whether the target precipitation area is a precipitation area connected to a linear precipitation band based on the output from the machine learning model (e.g., a label) when the features calculated in step S5 are input to the machine learning model.

[0065] Specifically, assuming the learning described above is performed, if the machine learning model outputs the label "1", it is determined that the target precipitation area is a precipitation area that will lead to a linear precipitation band (YES in step S6). The fact that the target precipitation area is a precipitation area that will lead to a linear precipitation band means that a linear precipitation band will occur in the future, and the detection unit 123 can detect the occurrence of the linear precipitation band.

[0066] If the occurrence of a linear rainband is detected in this manner, the detection result is output (step S7). In step S7, for example, the detection result may be output to the communication device 10d in order to transmit the detection result to an external device, or the detection result may be output to the display device in order to display the detection result on the display device. The detection result output in step S7 only needs to be information indicating that a linear rainband will occur in the future, but for example, as shown in Figure 7, by outputting information that represents the target precipitation area 123a, which is a precipitation area connected to a linear rainband, on a map as the detection result, users who use this information can intuitively grasp the area where the occurrence of a linear rainband is predicted.

[0067] On the other hand, if the machine learning model outputs the label "0", it is determined that the target precipitation area is not a precipitation area that leads to a linear precipitation band (NO in step S6). In this case, the process in step S7 is omitted, and the process shown in Figure 4 is terminated. Here, the explanation assumes that the process in step S7 is omitted, but if it is determined that the target precipitation area is not a precipitation area that leads to a linear precipitation band, it is acceptable for information indicating that a linear precipitation band will not occur (or is unlikely to occur) to be output.

[0068] Although not shown in Figure 4, if multiple precipitation areas are detected in step S4 as described above, the processes in steps S5 to S7 will be repeatedly executed with each of the multiple precipitation areas as the target precipitation area.

[0069] As described above, the information processing device 10 (processing unit 12) according to this embodiment detects a precipitation area (first precipitation area) based on the hourly cumulative rainfall value (first hourly cumulative value of rainfall) at each of the multiple locations, which is based on rainfall data indicating the hourly rainfall at the multiple locations, and detects the occurrence of a linear precipitation band (characteristic precipitation area) by inputting a plurality of feature quantities (first and second feature quantities) calculated with respect to the detected precipitation area into a machine learning model.

[0070] Generally, the Japan Meteorological Agency (JMA) announces the occurrence of a linear rainband, for example, as part of its information on such bands. However, such announcements by the JMA are based on, for example, three-hour cumulative rainfall data and are equivalent to real-time information. In other words, when the JMA announces the occurrence of a linear rainband, heavy rainfall has often continued for a certain period of time, and damage (disaster) may have already occurred.

[0071] In contrast, in this embodiment, with the configuration described above, if rainfall data for about 60 minutes is available, for example, the occurrence of a linear rainband can be detected in advance (for example, several hours before the Japan Meteorological Agency makes an announcement).

[0072] Furthermore, the Japan Meteorological Agency issues warnings to areas where linear rainbands are likely to form about half a day in advance. However, the occurrence of linear rainbands in these cases is predicted based on numerical weather prediction models used in weather forecasts, for example. As a result, the accuracy of these predictions is low, and there are many cases where linear rainbands are missed or the prediction is wrong. Missing a linear rainband means that a linear rainband occurred, but its occurrence could not be predicted. A wrong prediction means that a linear rainband was predicted, but it did not occur.

[0073] In contrast, in this embodiment, it is possible to detect the occurrence of linear rainbands with high accuracy using a machine learning model that utilizes multiple features, thereby reducing the chances of missing or inaccurately predicting such linear rainbands.

[0074] In other words, in this embodiment, it becomes possible to detect the occurrence of a linear rainband with high accuracy, for example, several hours before the Japan Meteorological Agency makes an announcement regarding the occurrence of such a linear rainband.

[0075] By the way, in this embodiment, the occurrence of a linear rainband has been described as being detected using at least two feature quantities, including rainfall statistics (for example, the maximum value, average value, or sum of the time-cumulative rainfall values). However, other feature quantities besides rainfall statistics may be used.

[0076] Below, we will describe examples of features other than rainfall statistics that can be used in this embodiment.

[0077] First, Figure 8 is a block diagram showing another example of the functional configuration of the information processing device 10 according to this embodiment. In Figure 8, the same reference numerals are used for parts that are the same as those in Figure 1, and their detailed explanation is omitted. Only parts that differ from Figure 1 will be described.

[0078] As shown in Figure 8, the information processing device 10 includes a wind condition data storage unit 13. The wind condition data storage unit 13 stores wind condition data, for example, showing the wind speed and wind direction for each hour at multiple locations. The wind condition data stored in the wind condition data storage unit 13 may be acquired from a system other than the weather radar system (for example, a weather observation system).

[0079] In this embodiment, feature quantities calculated from the wind condition data stored in the wind condition data storage unit 13 may be used. In this case, the feature quantities include, for example, statistical values ​​of wind speed and wind direction at each of the multiple locations included in the target precipitation area described above (hereinafter referred to as wind condition statistics).

[0080] The wind condition statistics are assumed to be, for example, the average wind speed or average wind direction in the target precipitation area, but they may also be the angle (angle difference) between the average wind direction and the direction of travel (movement) of the target precipitation area.

[0081] Furthermore, if the wind condition data is, for example, data indicating wind speed and wind direction at each grid point defined in three-dimensional space (i.e., three-dimensional grid data of latitude, longitude, and altitude), then the wind condition statistics may be calculated for each of the multiple altitude planes, or they may be the difference or dot product of the average wind direction between the multiple altitude planes in the target precipitation area.

[0082] The wind condition statistics may be one of the values ​​described here, or a combination of two or more values.

[0083] In this embodiment, the detection accuracy can be improved by utilizing the wind condition statistics (wind speed and wind direction information) described above to detect the occurrence of linear rainbands.

[0084] Furthermore, it is generally known that rain clouds (cumulonimbus clouds) move as they are carried by the wind (that is, they move from upwind to downwind), and in order to detect the occurrence of linear rainbands, for example (that is, to determine whether or not the target precipitation area is a precipitation area connected to a linear rainband), it is considered useful to consider information about the area located upwind of the target precipitation area (hereinafter referred to as the upwind area).

[0085] In this case, in addition to the features related to the target precipitation area described above (rainfall statistics or wind condition statistics for the target precipitation area), the accuracy of detecting the occurrence of linear precipitation bands can be improved by further utilizing features related to the upwind region identified based on wind condition data, for example (rainfall statistics or wind condition statistics for the said upwind region).

[0086] An example of the upwind region will be explained with reference to Figure 9. In this embodiment, the upwind region 202 of the target precipitation area 201 is defined as, for example, a region within a distance of 150 km from the center of the target precipitation area 201, and within an angle of 30° around the upwind vector.

[0087] Here, we have described a case where the upwind region is statically determined according to the location of the target precipitation area. However, the upwind region may also be dynamically determined based on wind condition data, for example. In this case, the upwind region may be adaptively determined by, for example, determining the area of ​​rain clouds that are expected to reach the location of the target precipitation area within a specified time, based on the average wind speed in the upwind region of the target precipitation area. Specifically, for example, if the average wind speed is 20 m / s and the specified time is 1 hour, the upwind region can be determined to be the area within 72 km upwind from the center of the target precipitation area.

[0088] In this embodiment, as described above, it is believed that the lead time can be extended by utilizing features related to the upwind region (focusing on information upwind of the rain area). The lead time is an evaluation value defined by "the time when the linear rainband occurred (for example, the time when the occurrence of the linear rainband was announced by the Japan Meteorological Agency) - the time when the occurrence of the linear rainband was detected," and the longer the lead time, the more it can be evaluated that the occurrence of the linear rainband was detected (predicted) well in advance of the actual time when it occurred.

[0089] In this explanation, wind condition data is used to calculate features related to the target precipitation area. However, this wind condition data (wind condition statistics) may also be used to detect the precipitation area described above. In this case, for example, an area where wind speed or wind direction satisfies certain conditions may be detected as a precipitation area.

[0090] Furthermore, in this embodiment, when rainfall statistics are used as a characteristic quantity for the target precipitation area, it has been explained that these rainfall statistics are calculated based on the time-cumulative rainfall value in the target precipitation area. However, a configuration in which multiple cumulative time periods are prepared for calculating the time-cumulative rainfall value is also possible.

[0091] Specifically, assuming that a first and second cumulative time are provided as multiple cumulative time options as described above, the first rainfall statistic calculated from the hourly cumulative rainfall value in the target precipitation area based on the first cumulative time (e.g., 60 minutes) and the second rainfall statistic calculated from the hourly cumulative rainfall value in the target precipitation area based on the second cumulative time (e.g., 180 minutes) can be used as feature quantities related to the target precipitation area.

[0092] The reason for using rainfall statistics based on multiple cumulative time periods is that it can be difficult to distinguish between precipitation areas connected to linear rainbands and those that are not based on rainfall statistics based on a single cumulative time period alone. Specifically, it is difficult to distinguish between rainfall areas caused by individual cumulonimbus clouds and rainfall areas caused by groups of precipitation areas connected to linear rainbands using only rainfall statistics based on long cumulative time periods. On the other hand, if the rainfall statistics are large due to previous rainfall even if the recent rainfall has been small, it is possible to mistakenly identify low-risk precipitation areas as precipitation areas connected to linear rainbands.

[0093] With this configuration, recent rainfall conditions can be considered using rainfall statistics with short cumulative time periods, while the duration of that rainfall can also be considered using rainfall statistics with longer cumulative time periods. By combining rainfall statistics based on multiple cumulative time periods, the accuracy of detecting the occurrence of linear rainbands can be improved.

[0094] Next, Figure 10 is a block diagram showing yet another example of the functional configuration of the information processing device 10 according to this embodiment. In Figure 10, the same reference numerals are used for parts that are the same as those in Figure 1, and their detailed explanation is omitted. Parts that differ from Figure 1 will be explained.

[0095] As shown in Figure 10, the information processing device 10 includes a prediction data storage unit 14. The prediction data storage unit 14 stores prediction data that shows, for example, the predicted hourly rainfall at multiple locations (hereinafter referred to as predicted rainfall). The prediction data stored in the prediction data storage unit 14 is assumed to be data generated by an external system, for example.

[0096] In this embodiment, rainfall statistics calculated based on rainfall data from the past (for example, 60 minutes ago) to the present were described as being used as features related to the target precipitation area. However, according to the information processing device 10 shown in Figure 10 above, rainfall statistics calculated from prediction data (predicted distribution of rainfall) can be used as features related to the target precipitation area.

[0097] As shown in Figure 11, the feature quantities (rainfall statistics) in this case may be calculated from, for example, prediction data showing the predicted rainfall at each of the multiple locations included in the target precipitation area from the present (time of detection processing) to 60 minutes later, or from rainfall data showing the rainfall (actually observed rainfall) at each of the multiple locations included in the target precipitation area from 30 minutes ago to the present, and prediction data showing the predicted rainfall at each of the multiple locations included in the target precipitation area from the present to 30 minutes later.

[0098] In other words, in the information processing device 10 shown in Figure 10, rainfall statistics (statistics for the future only) may be calculated as features related to the target precipitation area from only the prediction data stored in the prediction data storage unit 14, or rainfall statistics (statistics for the past and future) may be calculated from the prediction data and the rainfall data stored in the rainfall data storage unit 11. Furthermore, both the rainfall statistics calculated from the prediction data and the rainfall statistics calculated from the rainfall data may be used as different features related to the target precipitation area.

[0099] With this configuration, by considering the predicted rainfall shown by the forecast data, it is possible to improve the accuracy of detecting the occurrence of linear rainbands and to extend the lead time mentioned above.

[0100] In this embodiment, the features related to the target precipitation area have been described as rainfall statistics for the target precipitation area, wind condition statistics for the target precipitation area, statistics for the windward region of the target precipitation area (rainfall statistics or wind condition statistics), rainfall statistics based on multiple cumulative time periods, and rainfall statistics calculated from forecast data. However, it is sufficient to use at least two of these features related to the target precipitation area.

[0101] Furthermore, in order to improve the accuracy of detecting linear rainbands, it is necessary to select a combination of features from among the various features mentioned above that is useful for improving detection accuracy. However, it is difficult for users to perform manual analysis to select such a combination of features from among various features.

[0102] In contrast, this embodiment uses a machine learning model. For example, by generating a machine learning model that has learned an arbitrary combination of features from among various features, and then evaluating the accuracy of the machine learning model, it becomes possible to easily select (determine) a useful combination of features compared to the manual analysis described above.

[0103] Furthermore, although this embodiment describes the detection of linear rainbands using a machine learning model, there are multiple factors that contribute to the occurrence of such rainbands, and it is difficult to detect the occurrence of all linear rainbands with only a single model. In addition, it is conceivable that there may be variability in the discrimination tendencies of different machine learning models. For example, one machine learning model may have high accuracy in discriminating rapidly developing linear rainbands, while another machine learning model may have high accuracy in discriminating linear rainbands that develop gradually over a longer period of time. For this reason, it is also possible to have a configuration in which multiple machine learning models (hereinafter referred to as the first and second machine learning models) that have learned different combinations of features are prepared. With this configuration, for example, the final detection result (i.e., whether or not a linear rainband will occur) can be determined based on the determination result of whether or not the target precipitation area is a precipitation area that leads to a linear rainband, obtained by inputting features (first and second features) into the first machine learning model (hereinafter referred to as the determination result by the first machine learning model), and the determination result of whether or not the target precipitation area is a precipitation area that leads to a linear rainband, obtained by inputting different features (third and fourth features) into the second machine learning model (hereinafter referred to as the determination result by the second machine learning model).

[0104] In this case, the final detection result can be determined, for example, by a logical AND or OR operation on the detection results from the first and second machine learning models. Specifically, in a configuration where the final detection result is determined by a logical AND operation on the judgment results from the first and second machine learning models, the occurrence of a linear rainband will be detected if both the first and second machine learning models determine that the target precipitation area is a precipitation area connected to a linear rainband. Also, in a configuration where the final detection result is determined by a logical OR operation on the judgment results from the first and second machine learning models, the occurrence of a linear rainband will be detected if at least one of the first and second machine learning models determines that the target precipitation area is a precipitation area connected to a linear rainband. The final detection result may also be determined by considering the weighting of each of the judgment results from the first and second machine learning models.

[0105] Figure 12 shows an example of a combination of features input to the first machine learning model, and Figure 13 shows an example of a combination of features input to the second machine learning model. However, since the first and second machine learning models use different features, the results of their judgments are likely to show different trends. Therefore, by comprehensively considering the judgment results of the first and second machine learning models to detect the occurrence of linear precipitation bands, it may be possible to improve the detection accuracy.

[0106] Furthermore, generally speaking, there is a trade-off between the probability of a prediction being incorrect (the rate of incorrect predictions) and the lead time. Specifically, determining the final detection result by performing a logical AND operation on the judgment results of the first and second machine learning models described above shortens the lead time but reduces the rate of incorrect predictions. On the other hand, determining the final detection result by performing a logical OR operation on the judgment results of the first and second machine learning models increases the rate of incorrect predictions but lengthens the lead time.

[0107] Therefore, in the configuration using the first and second machine learning models as described above, the method for determining the final detection result may be changed depending on whether the false alarm rate or lead time is prioritized. Furthermore, the configuration may be such that the machine learning model (or combination of machine learning models) used to detect the occurrence of linear rainbands is selected depending on whether the false alarm rate or lead time is prioritized. Note that the method for determining the final detection result and the selection of the machine learning model used to detect the occurrence of linear rainbands may be done, for example, depending on the type of linear rainband or the user's needs.

[0108] Furthermore, while this explanation primarily focuses on the use of two machine learning models (the first and second machine learning models), configurations using three or more machine learning models are also acceptable. In cases where three or more machine learning models are used, the final detection result may be determined by a majority vote of the judgment results from those models.

[0109] Furthermore, while we have described multiple machine learning models with different input features, these multiple machine learning models may differ not in their input features, but in the machine learning algorithms used to generate them.

[0110] In this embodiment, we have described several feature quantities that can be used to detect the occurrence of linear rainbands. However, depending on the combination of feature quantities used, the configuration of the information processing device 10 described in this embodiment may be combined. Specifically, when rainfall data, wind condition data, and forecast data are used to calculate the feature quantities, the configurations described in Figures 8 and 10 can be combined to form an information processing device 10 that includes a rainfall data storage unit 11, a wind condition data storage unit 13, and a forecast data storage unit 14.

[0111] Furthermore, although this embodiment describes detecting the occurrence of a linear rainband using a machine learning model that has learned the precipitation areas connected to the linear rainband, if the machine learning model learns precipitation areas connected to other characteristic precipitation areas (such as dangerous precipitation areas that do not fall under the category of linear rainbands but are accompanied by heavy rainfall that can cause damage), it becomes possible to detect the occurrence of such dangerous precipitation areas using this machine learning model. In other words, in this embodiment, it is also possible to use the machine learning model to detect, for example, a uniquely defined characteristic precipitation area other than a linear rainband.

[0112] Furthermore, in this embodiment, the system may be configured to detect the occurrence of linear rainbands without using a machine learning model. Specifically, for example, the occurrence of linear rainbands may be detected based on whether or not the rainfall statistics of the target precipitation area are above a threshold.

[0113] In this embodiment, a precipitation area is detected based on the time-cumulative rainfall values ​​at each of several locations, and rainfall statistics are calculated based on the time-cumulative rainfall values ​​at each of the several locations included in the precipitation area. However, the cumulative time used to calculate the rainfall statistics may differ from the cumulative time used to detect the precipitation area. With such a configuration, for example, a relatively long cumulative time can be used to detect an area where rainfall has occurred for a long period of time as a precipitation area, and a relatively short cumulative time can be used to detect the occurrence of a linear rainband when it is determined that there has been heavy rainfall in a short period of time. This makes it possible to detect the occurrence of a linear rainband early with high accuracy.

[0114] Furthermore, in configurations that do not use machine learning models, the occurrence of linear rainbands may be detected based on rainfall statistics derived from each of the multiple cumulative time periods prepared in advance, as described above.

[0115] Incidentally, in this embodiment, for example, the information processing device 10 was described as being located outside the weather radar system and acquiring rainfall data from the weather radar system, but the information processing device 10 may also be mounted on the weather radar system.

[0116] Figure 14 shows an example of the configuration of a weather radar system equipped with an information processing device 10. As shown in Figure 14, the weather radar system 300 comprises a transceiver 301, a signal processing device 302, and an information processing device 10.

[0117] The transceiver 301 is configured to transmit a radar signal (radar wave) and receive a reflected wave signal based on the reflected wave of the radar signal. The transceiver 301 outputs the received reflected wave signal to the signal processing device 302.

[0118] The signal processing device 302 processes the reflected wave signal output from the transceiver 301 to acquire observation data, such as the received power (echo intensity) of the reflected wave signal, and generates rainfall data based on the acquired observation data. The signal processing device 302 outputs the generated rainfall data to the information processing device 10.

[0119] The transceiver 301 and signal processing device 302 described above constitute a weather radar, such as a phased array weather radar (MP-PAWR or PAWR).

[0120] The rainfall data output from the signal processing device 302 is stored in the rainfall data storage unit 11 included in the information processing device 10 and used to detect the occurrence of linear rainbands as described above. Furthermore, at least some of the transmitting / receiving device 301, signal processing device 302, and information processing device 10 may be implemented together.

[0121] This description has focused on the case where the information processing device 10 is installed in a weather radar system. However, in environments where multiple weather radar systems are deployed, for example, the observation data acquired by each of the weather radar systems (signal processing devices) may be combined and processed at a synthesis processing station. The information processing device 10 according to this embodiment may be installed (deployed) in such a synthesis processing station.

[0122] Figure 15 shows an example of the configuration of a composite processing station equipped with an information processing device 10. As shown in Figure 15, the composite processing station 400 is communicatively connected to multiple weather radar systems 500 and includes a data receiving device 401, a radar combining device 402, and an information processing device 10.

[0123] Each of the multiple weather radar systems 500 is equipped with a weather radar consisting of the above-described transmitting and receiving device and signal processing device, and is configured to transmit observation data acquired by the operation of the weather radar.

[0124] The data receiver 401 is configured to receive observational data transmitted from each of the multiple weather radar systems 500. The data receiver 401 outputs the received observational data to the radar combining unit 402.

[0125] The radar combining device 402 generates rainfall data based on observation data output from the data receiving device 401. The rainfall data generated by the radar combining device 402 is a wide-ranging and highly reliable rainfall data obtained, for example, by combining observation data from multiple weather radar systems 500. The radar combining device 402 outputs the generated rainfall data to the information processing device 10.

[0126] The rainfall data output from the radar combining device 402 is stored in the rainfall data storage unit 11 included in the information processing device 10 and used to detect the occurrence of linear rainbands as described above. Furthermore, at least some of the data receiving device 401, radar combining device 402, and information processing device 10 may be implemented together.

[0127] While several embodiments of the present invention have been described, these embodiments are presented as examples only and are not intended to limit the scope of the invention. These embodiments can be carried out in a variety of other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims and their equivalents.

[0128] With regard to the embodiments described above, the following additional information is disclosed. [1] Based on rainfall data showing hourly rainfall at multiple locations, the first precipitation area is detected based on the first hourly cumulative value of rainfall at each of those multiple locations. The system includes a processing unit that detects the occurrence of a characteristic precipitation area by inputting the first and second feature quantities calculated for the detected first precipitation area into a first machine learning model. Information processing device. [2] The rainfall data is an information processing device as described in [1], which shows ground rainfall or vertically integrated rainfall calculated based on observation data observed by a weather radar system. [3] The first machine learning model is constructed to output whether or not the characteristic precipitation area occurs when the first and second features calculated for the first precipitation area are input, by learning the first and second features calculated for the second precipitation area that was detected before the occurrence of a characteristic precipitation area that occurred in the past, and a label indicating that the characteristic precipitation area has occurred. [1] or [2] [4] An information processing device according to any one of the following [1] to [3], wherein at least a portion of the first and second features includes a first statistical value calculated from the rainfall at each of the multiple locations included in the first precipitation area indicated by the rainfall data. [5] The first statistical value includes the maximum, average, or total value of the second-hour cumulative rainfall at each of the multiple locations included in the first precipitation area. The time used to calculate the second cumulative value is different from the time used to calculate the first cumulative value. [4] The information processing device described above. [6] The first statistical value includes the maximum value, average value, or total value of the second-hour cumulative rainfall at each of the multiple locations included in the first precipitation area, and the maximum value, average value, or total value of the third-hour cumulative rainfall at each of the multiple locations included in the first precipitation area. The time used to calculate the second time cumulative value is different from the time used to calculate the third time cumulative value. The information processing device described in [4] or [5]. [7] The first statistical value is calculated from forecast data showing the predicted rainfall at each of the multiple locations included in the first precipitation area. [4] to [6] Information processing device according to any one of the above. [8] An information processing device according to any one of the following [1] to [7], wherein at least a portion of the first and second features includes a second statistical value calculated from wind condition data indicating wind speed and wind direction at each of a plurality of locations included in the first precipitation area. [9] The processing unit detects an area located upwind of the first precipitation area based on wind condition data indicating hourly wind speed and wind direction at the plurality of locations, and further inputs a third feature calculated with respect to the detected area into the first machine learning model to detect the occurrence of the characteristic precipitation area, as described in any one of the [1] to [8].

[10] The processing unit detects the occurrence of the characteristic precipitation area by inputting the first and second features calculated with respect to the detected first precipitation area into the first machine learning model, and inputting the third and fourth features calculated with respect to the detected first precipitation area into a second machine learning model different from the first machine learning model [1] to [9].

[11] The first and second machine learning models are information processing devices as described in

[10] , wherein the machine learning algorithms for generating the first and second machine learning models are different.

[12] The processing unit detects a first region and a second region different from the first region based on the first time cumulative value, and the distance between the first region and the second region is less than a predetermined value, and detects the first precipitation region including the first region and the second region, according to any one of the information processing devices in any one of the [1] to

[11] paragraphs.

[13] The processing unit corrects the first precipitation area to a predetermined shape, as described in any one of the items [1] to

[12] .

[14] A transmitting and receiving device that transmits a radar signal and receives a reflected wave signal based on the reflected wave of the radar signal, A signal processing device that processes the received reflected wave signal to acquire observation data and generates rainfall data based on the acquired observation data, The information processing device described in any one of the items [1] to

[13] and A weather radar system equipped with the following features.

[15] A synthesis device that generates the rainfall data by combining observation data transmitted from each of multiple weather radar systems, The information processing device described in any one of the items [1] to

[13] and A synthesis processing station equipped with the following.

[16] Based on rainfall data showing hourly rainfall at multiple locations, the first precipitation area is detected based on the first hourly cumulative value of rainfall at each of those multiple locations. The occurrence of characteristic precipitation areas is detected by inputting the first and second features calculated for the detected first precipitation area into the first machine learning model. method.

[17] On the computer, The first precipitation area is detected based on the first hourly cumulative value of rainfall at each of the multiple locations, which is based on rainfall data showing hourly rainfall at multiple locations. The occurrence of characteristic precipitation areas is detected by inputting the first and second features calculated for the detected first precipitation area into the first machine learning model. A program to execute. [Explanation of Symbols]

[0129] 10...Information processing device, 10a...CPU, 10b...Non-volatile memory, 10c...RAM, 10d...Communication device, 11...Rainfall data storage unit, 12...Processing unit, 13...Wind condition data storage unit, 14...Prediction data storage unit, 121...Precipitation area detection unit, 122...Feature calculation unit, 123...Detection unit, 300, 500...Weather radar system, 301...Transmitting and receiving device, 302...Signal processing device, 400...Synthesis processing station, 401...Data receiving device, 402...Radar synthesis device.

Claims

1. Based on rainfall data showing hourly rainfall at multiple locations, the first precipitation area is detected based on the first hourly cumulative value of rainfall at each of those multiple locations. The system includes a processing unit that detects the occurrence of a characteristic precipitation area by inputting the first and second feature quantities calculated with respect to the detected first precipitation area into a first machine learning model. Information processing device.

2. The information processing device according to claim 1, wherein the rainfall data indicates ground rainfall or vertically integrated rainfall calculated based on observation data observed by a weather radar system.

3. The information processing device according to claim 1, wherein the first machine learning model is constructed to output whether or not the characteristic precipitation area occurs when the first and second features calculated for the first precipitation area are input, by learning the first and second features calculated for the second precipitation area that was detected before the occurrence of a characteristic precipitation area that occurred in the past, and a label indicating that the characteristic precipitation area has occurred.

4. The information processing apparatus according to claim 1, wherein at least a portion of the first and second feature quantities includes a first statistical value calculated from the rainfall at each of the plurality of points included in the first precipitation area indicated by the rainfall data.

5. The first statistical value includes the maximum, average, or total value of the second-hour cumulative rainfall at each of the multiple locations included in the first precipitation area. The time used to calculate the second cumulative value is different from the time used to calculate the first cumulative value. The information processing apparatus according to claim 4.

6. The first statistical value includes the maximum value, average value, or total value of the second hourly cumulative rainfall at each of the multiple locations included in the first precipitation area, and the maximum value, average value, or total value of the third hourly cumulative rainfall at each of the multiple locations included in the first precipitation area. The time used to calculate the second time cumulative value is different from the time used to calculate the third time cumulative value. The information processing apparatus according to claim 4.

7. The information processing device according to claim 4, wherein the first statistical value is calculated from prediction data showing the predicted rainfall at each of the multiple locations included in the first precipitation area.

8. The information processing apparatus according to claim 1, wherein at least a portion of the first and second feature quantities includes a second statistical value calculated from wind condition data indicating wind speed and wind direction at each of a plurality of locations included in the first precipitation area.

9. The information processing apparatus according to claim 1, wherein the processing unit detects an area located upwind of the first precipitation area based on wind condition data indicating hourly wind speed and wind direction at the plurality of locations, and further inputs a third feature calculated with respect to the detected area into the first machine learning model to detect the occurrence of the characteristic precipitation area.

10. The information processing apparatus according to claim 1, wherein the processing unit inputs the first and second feature quantities calculated with respect to the detected first precipitation area into the first machine learning model, and inputs the third and fourth feature quantities calculated with respect to the detected first precipitation area into a second machine learning model different from the first machine learning model, thereby detecting the occurrence of the characteristic precipitation area.

11. The information processing apparatus according to claim 10, wherein the first and second machine learning models are different in that the machine learning algorithms for generating the first and second machine learning models are different.

12. The information processing apparatus according to claim 1, wherein the processing unit detects a first region and a second region different from the first region based on the first time cumulative value, and the distance between the first region and the second region is less than a predetermined value, and detects the first precipitation region including the first region and the second region.

13. The processing unit corrects the first precipitation area to a predetermined shape, as described in claim 1.

14. A transmitting and receiving device that transmits a radar signal and receives a reflected wave signal based on the reflected wave of the radar signal, A signal processing device that processes the received reflected wave signal to acquire observation data and generates rainfall data based on the acquired observation data, The information processing apparatus according to any one of claims 1 to 13 and A weather radar system equipped with the following features.

15. A synthesis device that generates the rainfall data by combining observation data transmitted from each of multiple weather radar systems, The information processing apparatus according to any one of claims 1 to 13 and A synthesis processing station equipped with the following.

16. Based on rainfall data showing hourly rainfall at multiple locations, the first precipitation area is detected based on the first hourly cumulative value of rainfall at each of those multiple locations. The occurrence of characteristic precipitation areas is detected by inputting the first and second feature quantities calculated for the detected first precipitation area into the first machine learning model. method.

17. On the computer, The first precipitation area is detected based on the first hourly cumulative value of rainfall at each of the multiple locations, which is based on rainfall data showing hourly rainfall at multiple locations. The occurrence of a characteristic precipitation area is detected by inputting the first and second feature quantities calculated for the detected first precipitation area into the first machine learning model. A program to execute.

Images