Arithmetic processing unit, rainfall estimation system, and rainfall estimation method

Abstract

To provide a rainfall estimation system capable of appropriately considering an influence on raindrops by the environment until the raindrops fall onto the earth's surface.SOLUTION: A data processing server 71 estimates a rainfall in an optional position on the basis of reception strength data obtained from a radar device 21 according to the intensity of reflection waves obtained by reflection of electric waves radiated from the radar device 21 by raindrops, ground rainfall data obtained from a ground rain gauge 31 for measuring a rainfall in a predetermined point on the earth's surface, and wind direction / wind velocity data obtained from a wind direction anemometer 41 for measuring an influence of the raindrop on the environment until the raindrop reaches the earth's surface.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to rainfall estimation using machine learning. [Background technology]

[0002] Weather radar systems for forecasting weather and radar rain gauge systems for measuring rainfall in preparation for disasters have been known for some time. Weather radar systems and radar rain gauge systems use radar, making it possible to grasp and estimate rainfall over a wide area. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 7-146375 [Patent Document 2] Japanese Patent Application Publication No. 9-138279 Summary of the Invention [Problem to be solved by the invention]

[0004] Weather radar estimates rainfall by detecting waves reflected by raindrops falling from rain clouds in the sky, which results in errors between the radar's estimates and the actual rainfall amount at the ground. Therefore, in References 1 and 2, rainfall is estimated by interpolating the observed values ​​obtained by ground-based rain gauges installed on the ground.

[0005] On the other hand, neither Cited Document 1 nor Cited Document 2 takes into consideration the influence of the environment on raindrops before they reach the ground. However, there are cases where an error occurs between the actual rainfall position and the predicted rainfall position based on radar observation due to the influence of wind, etc., and therefore, in order to accurately estimate the amount of rainfall, it is desirable to take into consideration the influence of the environment on raindrops before they reach the ground.

[0006] Therefore, an object of the present disclosure is to provide a rainfall estimation system that can appropriately consider the influence that raindrops receive from the environment before they fall to the ground. [Means for solving the problem]

[0007] In order to solve the above problem, the processing device of the present disclosure employs a method of estimating the amount of rainfall by taking into consideration the influence of the environment on raindrops until they fall to the ground.

[0008] Specifically, the processing device of the present disclosure includes: The amount of rainfall at an arbitrary position is estimated based on reception intensity data obtained from the radar device according to the intensity of the radio waves emitted from the radar device and reflected by raindrops, ground rainfall data obtained from a first measuring instrument that observes the amount of rainfall at a predetermined point on the ground surface, and environmental data obtained from a second measuring instrument that measures the influence of the environment on the raindrops before they reach the ground surface. It is characterized by:

[0009] The present disclosure also provides a radar device, The arithmetic processing device, A rainfall estimation system is provided.

[0010] The present disclosure also provides a method for detecting a reflected wave of a radio wave emitted from a radar device and reflected by raindrops, the reflected wave being received as reception intensity data corresponding to the intensity of the reflected wave; a step of estimating the amount of rainfall at an arbitrary position based on the reception intensity data, ground rainfall data obtained from a first measuring instrument that observes the amount of rainfall at a predetermined point on the ground surface, and environmental data obtained from a second measuring instrument that measures the influence of the environment on the raindrops until they reach the ground surface; A rainfall estimation method including:

[0011] Such a processing device, rainfall estimation system, and rainfall estimation method can accurately estimate rainfall based on the observation values ​​of the second measuring instrument, by appropriately taking into account the influence of the environment on raindrops before they fall to the ground.

[0012] The arithmetic processing device further includes a learning unit that learns a relationship between the reception intensity data, the ground rainfall data, and the environmental data, based on the reception intensity data, the ground rainfall data, and the environmental data, It is preferable that the amount of rainfall at an arbitrary position is estimated from the reception intensity data and the environmental data according to the result of the learning.

[0013] The rainfall amount estimation method includes a step of learning a relationship between the reception intensity data, the ground rainfall data, and the environmental data, based on the reception intensity data, the ground rainfall data, and the environmental data; It is preferable that the method further comprises a step of estimating the amount of rainfall at an arbitrary position from the reception intensity data and the environmental data according to the result of the learning.

[0014] With this configuration and method, it is possible to accurately estimate the amount of rainfall at any location by utilizing the results of learning that appropriately takes into account the influence of the environment on raindrops from the time they fall to the ground based on environmental data.

[0015] Preferably, the second measuring device measures a physical quantity related to wind until the raindrops reach the ground surface, and outputs the physical quantity related to wind as the environmental data.

[0016] With this configuration, the amount of rainfall can be estimated accurately by appropriately considering the influence of wind on raindrops before they fall to the ground.

[0017] Preferably, the arithmetic processing device estimates the amount of rainfall at any position based on the elevation angle of the antenna of the radar device in addition to the reception intensity data, the ground rainfall data, and the environmental data.

[0018] With this configuration, machine learning is performed based on the elevation angle of the antenna, i.e., the altitude of the rain clouds that produce the raindrops, allowing for accurate estimation of rainfall by appropriately taking into account the influence that the raindrops receive from the environment before they fall to the ground.

[0019] The arithmetic processing device further includes a learning unit that learns a relationship between the reception intensity data, the ground rainfall data, the environmental data, and the elevation angle and the rainfall amount based on the reception intensity data, the ground rainfall data, the environmental data, and the elevation angle, It is preferable that the amount of rainfall at an arbitrary position is estimated from the reception intensity data, the environmental data, and the elevation angle according to the result of the learning.

[0020] With this configuration, machine learning is performed based on the elevation angle of the antenna, i.e., the altitude of the rain clouds that produce the raindrops, allowing for accurate estimation of rainfall by appropriately taking into account the influence that the raindrops receive from the environment before they fall to the ground.

[0021] It is preferable that the system further includes an image analysis unit that outputs analysis data obtained by analyzing image data output from a camera that captures images during rainfall, learns the relationship between the analysis data and the amount of rainfall, and estimates the rainfall situation at any location.

[0022] With this configuration, it is possible to perform image analysis with high accuracy based on the amount of rainfall estimated by taking into consideration the influence of the environment on the raindrops until they fall to the ground.

[0023] The above disclosures can be combined as much as possible. [Effects of the Invention]

[0024] According to the present disclosure, it is possible to accurately estimate the amount of rainfall by appropriately considering the influence of the environment on raindrops until they fall to the ground. [Brief explanation of the drawings]

[0025] [Figure 1] 1 is a diagram illustrating a schematic configuration of a rainfall estimation system. [Figure 2] FIG. 1 is a diagram illustrating a problem in a related rainfall estimation system. [Figure 3] FIG. 1 is a diagram illustrating a problem in a related rainfall estimation system. [Figure 4] 10 is a flowchart illustrating machine learning and rainfall estimation by a data processing server. [Figure 5] FIG. 10 is a diagram illustrating machine learning based on the antenna elevation angle of a radar device. [Figure 6] 1 is a diagram illustrating a schematic configuration of a rainfall estimation system. [Figure 7] 10 is a flowchart illustrating machine learning and situation estimation by an image analysis unit. [Figure 8] FIG. 1 is a diagram illustrating an image of actual use of the rainfall estimation system. DETAILED DESCRIPTION OF THE INVENTION

[0026] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. Note that the present disclosure is not limited to the embodiments shown below. These implementation examples are merely illustrative, and the present disclosure can be implemented in various forms with various modifications and improvements based on the knowledge of those skilled in the art. Note that components with the same reference numerals in this specification and drawings indicate the same components.

[0027] (Embodiment 1) A rainfall estimation system 101 according to an embodiment of the present disclosure will be described with reference to Fig. 1 to Fig. 4. Fig. 1 is a diagram illustrating a schematic configuration of the rainfall estimation system 101. As shown in FIG. 1, the rainfall estimation system 101 mainly includes a radar device 21 that emits radio waves toward rain clouds in the sky, a control device 25 that controls the radar device 21, a data processing server 71, and a data collection server 61. The rainfall estimation system 101 detects reflected waves reflected by raindrops falling from rain clouds and estimates the amount of rainfall at any point on a two-dimensional map based on the signal strength of the reflected waves. Specifically, the rainfall estimation system 101 estimates the amount of rainfall at any position within a specified area according to the characteristics and layout of the radar device 21. The information provision system 201 may be, for example, the Automated Meteological Data System (AMeDAS) of the Japan Meteorological Agency. The data processing server 71 functions as a "processing device."

[0028] The rainfall estimation system 101 is linked to an information provision system 201 external to the rainfall estimation system 101. The information provision system 201 mainly comprises an external server 81, and a ground rain gauge 31 and an anemometer 41 connected to the external server 81 via a predetermined network. The ground rain gauge 31 functions as a "first measuring instrument." The anemometer 41 functions as a "second measuring instrument."

[0029] Specifically, the data processing server 71 estimates the amount of rainfall at any location based on reception intensity data obtained from the radar device 21 according to the intensity of the radio waves emitted from the radar device 21 and reflected by raindrops, ground rainfall data obtained from a ground rain gauge 31 that observes the amount of rainfall at a specified point on the ground surface, and wind direction and wind speed data obtained from an anemometer 41 that measures the influence that raindrops receive from the environment before they reach the ground surface.

[0030] A plurality of ground rain gauges 31 are installed at predetermined intervals on a two-dimensional map. Each time a predetermined amount of rainwater accumulates, the ground rain gauges 31 output a pulse signal containing the location information of each ground rain gauge to an external server 81. For ease of explanation, the following description will be given assuming that each ground rain gauge 31 is placed at a corresponding "predetermined position" on the two-dimensional map.

[0031] A plurality of anemometers 41 are installed at predetermined intervals on a two-dimensional map. The anemometers 41 are configured to be able to measure the wind direction and speed at any position on the two-dimensional map. Specifically, the anemometers 41 measure the distance the wind moves per unit time and the direction in which the wind moves. In other words, the anemometers 41 measure physical quantities related to the wind until raindrops reach the ground surface. For example, the anemometers 41 may be ultrasonic. The anemometers 41 output pulse signals indicating the wind direction and speed, including location information related to the point where the wind speed and wind direction were measured, to the external server 81.

[0032] Although the present embodiment employs an anemometer capable of measuring both "wind direction" and "wind speed," the scope of the present disclosure is not limited thereto. An anemometer that measures only wind speed may be combined with an anemometer that measures only wind direction. Furthermore, wind direction and speed constantly fluctuate. Therefore, the average values ​​of measurements taken over the three seconds immediately prior to any measurement by the anemometer 41 may be used as the "wind direction" and "wind speed." Furthermore, the average values ​​of measurements taken over the ten minutes immediately prior to any measurement by the anemometer 41 may be used as the "wind direction" and "wind speed." The anemometer 41 may also have an anemometer function that measures the flow rate per unit time in a unit volume. The scope of the present disclosure is not limited to the above, and the radar device 21 may also have a function to measure physical quantities related to wind. For example, the radar device 21 may measure physical quantities related to wind by observing the direction and speed of cloud movement, etc.

[0033] The external server 81 is configured to collect and store pulse signals from the ground rain gauge 31 as ground rainfall data. The external server 81 is also configured to collect and store pulse signals from the anemometer 41 as wind direction and wind speed data. The wind direction and wind speed data is an example of "environmental data."

[0034] The radar device 21 includes an antenna 22, a support 23, and a motor 24. The antenna 22 is a concave antenna with a parabolic reflective surface. The antenna 22 emits radio waves toward rain clouds in the sky and receives waves reflected by raindrops. The support 23 is installed on the ground surface and supports the antenna 22 so that it can rotate. The motor 24 is disposed within the support 23. The motor 24 is a stepping motor.

[0035] The control device 25 transmits the transmission wave as a pulse signal to the antenna 22. The antenna 22 radiates the pulse signal toward rain clouds in the sky. The control device 25 also detects and amplifies the reflected wave received by the antenna 22, and transmits it to the data processing server 71 as reception intensity data.

[0036] Furthermore, the control device 25 is configured to control the elevation angle of the antenna 22 according to the type and position of rain clouds, etc. Specifically, the control device 25 outputs a PWM (Pulse Width Modulation) signal to the motor 24. The motor 24 drives the antenna 22 according to the PWM signal output from the control device 25. The antenna 22 receives a driving force from the motor 24 and rotates around the support part 23.

[0037] The data collection server 61 collects data from the external server 81 of the information provision system 201 and provides the collected data to the data processing server 71 in response to a request from the data processing server 71. In this embodiment, the data collection server 61 collects ground rainfall data observed by each ground rain gauge 31 and wind direction and wind speed data observed by each anemometer 41 in real time as the ground rainfall data and wind direction and wind speed data are updated. The ground rainfall data includes location information of each ground rain gauge 31. The data collection server 61 sends the ground rainfall data and wind direction and wind speed data to the data processing server 71. Note that, although the data collection server 61 is provided for data collection in this embodiment, the data processing server 71 may be configured to collect data directly from the information provision system 201.

[0038] The data collection server 61 may also process the ground rainfall data and wind direction / speed data depending on the purpose of the data. For example, the data collection server 61 may generate data related to changes in ground rainfall. Similarly, the data collection server 61 may generate data related to changes in wind direction and speed.

[0039] The data processing server 71 includes a learning unit 71a and a calculation processing unit 71b. The learning unit 71a performs machine learning based on reception intensity data related to reflected waves from raindrops above a predetermined location on a two-dimensional map, ground rainfall data from a ground rain gauge 31 located at the predetermined location, and wind direction and wind speed data at the predetermined location measured by an anemometer 41. Specifically, the learning unit 71a receives the reception intensity data and wind direction and wind speed data as input signals, and ground rainfall data as a training signal, and learns the relationship between these. The calculation processing unit 71b estimates the rainfall amount at any location on the two-dimensional map from the reception intensity data and wind direction and wind speed data at the location based on the learning results of the learning unit 71a. As shown in FIG. 1A, the calculation processing unit 71b outputs the rainfall estimation results in correspondence with the two-dimensional map. The learning unit 71a may be a neural network configured using a general computer, a dedicated computer, or the like.

[0040] Specifically, the learning unit 71a learns the relationship between the reception intensity data and wind direction / speed data and the amount of rainfall based on the reception intensity data, ground rainfall data, and wind direction / speed data. Furthermore, the calculation processing unit 71b estimates the amount of rainfall at an arbitrary position from the reception intensity data and wind direction / speed data in accordance with the results of learning by the learning unit 71a.

[0041] Here, a problem with the existing rainfall estimation system 101A related to this embodiment will be described. Figures 2 and 3 are diagrams for explaining the problem with the existing rainfall estimation system 101A related to this embodiment. As shown in FIG. 2, in an existing rainfall estimation system 101A, a radar device 21A emits radio waves toward raindrops D falling from a raincloud C. The radar device 21A receives the reflected waves reflected by the raindrops. The rainfall estimation system 101A estimates the amount of rainfall based on the signal strength of the reflected waves. However, the existing rainfall estimation system 101A does not consider the influence of wind W (environmental influence) on the raindrops D before they fall to the ground. As a result, a discrepancy occurs between the predicted rainfall location B and the actual rainfall location A, making it impossible to accurately estimate the actual amount of rainfall at any point. This problem cannot be solved by machine learning using the reception intensity data of the reflected waves as an input signal and ground rainfall data from a ground rain gauge as a training signal, because the falling location of the raindrops D from the raincloud C changes due to the influence of wind W.

[0042] On the other hand, in order to minimize the influence of wind W, it is also possible to emit radio waves to raindrops D near the ground and process the radio waves reflected by the raindrops D as reception intensity data. However, as shown in FIG. 3, because the earth is round, the radar device 21A cannot observe areas near the ground surface at a distance farther than the line-of-sight distance E. Furthermore, even near the ground surface, there is a possibility that an error will occur between the predicted rainfall amount and the actual rainfall amount due to the influence of strong winds, etc.

[0043] In contrast, the rainfall estimation system 101 of this embodiment performs machine learning based on wind direction and speed data obtained by an anemometer 41 in addition to ground rainfall data obtained by a ground rain gauge 31, as shown in Fig. 1. This makes it possible to accurately estimate the amount of rainfall by taking into account the influence of wind on raindrops before they fall from rainclouds (influence from the environment).

[0044] Next, the processing by the data processing server 71 will be described with reference to FIG. Specifically, the rainfall estimation method of the present disclosure includes: A step of acquiring reflected waves, which are radio waves emitted from the radar device 21 and reflected by raindrops, as reception intensity data corresponding to the intensity of the reflected waves; a step of estimating the amount of rainfall at an arbitrary position based on the received intensity data, ground rainfall data obtained from a ground rain gauge 31 that observes the amount of rainfall at a predetermined point on the ground surface, and wind direction and wind speed data obtained from an anemometer 41 that measures the influence of the environment on raindrops before they reach the ground surface; Includes:

[0045] [Machine Learning] FIG. 4A is a flowchart illustrating machine learning by the learning unit 71a. First, the learning unit 71a acquires reception intensity data related to reflected waves from raindrops above a predetermined position on a two-dimensional map (step S1). The control device 25 controls the antenna 22 of the radar device 21 to acquire the reception intensity data. The antenna 22 emits a transmission wave toward raindrops falling from rainclouds above the predetermined position and receives the reflected wave reflected by the raindrops. The control device 25 detects and amplifies the reflected wave received by the antenna 22 and sends it to the learning unit 71a as reception intensity data. Next, the learning unit 71a transmits data related to the reception intensity data, including information on the time when the reception intensity data was acquired, to the data collection server 61, and requests ground rainfall data from a ground rain gauge 31 located at a predetermined position and wind direction and wind speed data from an anemometer 41 located at a predetermined position, from the data collection server 61.

[0046] In response to a request from the learning unit 71a, the data collection server 61 transmits to the learning unit 71a the ground rainfall data and wind direction / speed data at a predetermined position at the time when the learning unit 71a received the reception intensity data (step S2). The learning unit 71a uses the reception intensity data and wind direction / speed data as input signals and the ground rainfall data as a teacher signal to learn the relationship therebetween (step S3). In other words, the learning unit 71a learns the relationship between the reception intensity data and wind direction / speed data and the amount of rainfall.

[0047] At this time, the learning unit 71a may use the reception intensity data to calculate regional information such as latitude and longitude corresponding to the direction and distance from the radar device 21 to the rainclouds C and raindrops D, link the reception intensity data to this, and use this as an input signal. Also, the learning unit 71a may use the acquisition time of the reception intensity data to link the reception intensity data to regional information and time information, and use this as an input signal. Wind direction / speed data and ground rainfall data may also be linked to regional information and time information.

[0048] The learning unit 71a may acquire data on changes in ground rainfall and data on changes in wind direction and wind speed, and use these data as input signals to perform machine learning. This allows the amount of rainfall to be estimated while taking into account the amount of change in each measurement value and appropriately complementing it.

[0049] Thereafter, the learning unit 71a stores the learning results and sends them to the calculation processing unit 71b (step S4). The learning by the learning unit 71a is repeatedly performed for each predetermined position of each surface rain gauge 31, and the learning results are accumulated and stored. In this way, by accumulating learning results, it is expected that the weather conditions will be grasped more accurately. Furthermore, since the data collection server 61 provides data in real time so as to correspond to the reception intensity data, it is expected that the weather conditions will be grasped more accurately.

[0050] The machine learning performed by the learning unit 71a may be, for example, deep learning using a neural network. The learning unit 71a may also perform learning using a random forest that uses multiple decision trees.

[0051] [Rainfall estimation] FIG. 4B is a flowchart illustrating the rainfall amount estimation by the arithmetic processing unit 71b. First, the arithmetic processing unit 71b acquires reception intensity data at an arbitrary position on the two-dimensional map (step S5). The acquisition of the reception intensity data is performed by the control device 25 driving the antenna 22 so that the antenna 22 covers its scanning range. In other words, the antenna 22 thoroughly scans its scanning range. Next, the arithmetic processing unit 71b requests the data collection server 61 to provide wind direction and wind speed data at an arbitrary position corresponding to the acquired reception intensity data at the arbitrary position (step S6). In response to the request from the arithmetic processing unit 71b, the data collection server 61 sends the wind direction and wind speed data at the arbitrary position corresponding to the reception intensity data at the arbitrary position acquired by the arithmetic processing unit 71b to the arithmetic processing unit 71b.

[0052] The calculation processing unit 71b estimates the amount of rainfall at any location on the two-dimensional map from the reception intensity data and wind direction / speed data based on the learning results of the learning unit 71a (step S7). Specifically, the calculation processing unit 71b estimates the optimal amount of rainfall based on the correspondence between the reception intensity data, wind direction / speed data, and ground rainfall data learned during machine learning. The rainfall amount is estimated each time reception intensity data for any location is received, so that real-time rainfall conditions can be provided to the user. Furthermore, if the estimated rainfall amount at any location exceeds a predetermined threshold, an alert may be provided to the user, such as by flashing the corresponding location on the map. In this way, in this embodiment, the relationship between the reception intensity data, wind direction / speed data (input signal), and ground rainfall data (teacher signal) is learned, and then the rainfall amount can be estimated based on reception intensity data from the radar device 21, which detects reflected waves from raindrops in a planar manner, thereby enabling the amount of rainfall to be grasped in a planar manner in accordance with the two-dimensional map.

[0053] (Embodiment 2) A rainfall estimation system 102 according to an embodiment of the present disclosure will be described with reference to Fig. 5. Fig. 5 is a diagram illustrating machine learning based on the antenna elevation angle of the radar device 21. As described above, the amount of rainfall can be suitably estimated by taking into account the influence of the environment on the raindrops D before they fall to the ground. The longer the distance the raindrops D travel from the rain cloud to the ground, the greater the influence of the environment. The altitude of the rainclouds C that generate the raindrops D varies depending on the type of raincloud C. For example, in nimbostratus clouds, the bottom of the rainclouds C that generate the raindrops D is 3,000 to 4,000 meters above the ground, whereas in cumulonimbus clouds, the bottom of the rainclouds C that generate the raindrops D is approximately 2,000 meters above the ground. Therefore, in this embodiment, a method is adopted in which the altitude of the rainclouds C is taken into account during machine learning and rainfall amount estimation, thereby estimating the amount of rainfall more accurately. This will be explained below.

[0054] Specifically, the learning unit 71a in this embodiment learns the relationship between the reception intensity data, the ground rainfall data, the wind direction and wind speed data, and the elevation angle and the rainfall amount based on the reception intensity data, the ground rainfall data, the wind direction and wind speed data, and the elevation angle of the antenna 22 of the radar device 21, The amount of rainfall at any location is estimated from reception strength data, wind direction and speed data, and elevation angle.

[0055] As described above, the antenna 22 of the radar device 21 is configured to be rotatable relative to the support part 23. Here, since the rain cloud C is located in the sky, when transmitting (radiating) radio waves to the bottom of the rain cloud C and receiving reflected waves from raindrops, the elevation angle of the antenna 22 forms an angle θ with respect to the horizontal direction. The altitude of the rain cloud C changes depending on the situation, and the elevation angle θ of the antenna 22 changes in accordance with the change in the altitude of the rain cloud C. In other words, the altitude of the rain cloud C corresponds to the elevation angle θ of the antenna 22.

[0056] In this embodiment, the control device 25 sends the reception intensity data to the learning unit 71a and, at the same time, sends to the learning unit 71a data relating to the elevation angle θ at the time when the antenna 22 received the reflected wave. The learning unit 71a receives the reception intensity data, wind direction / speed data, and data relating to the elevation angle θ as input signals, and uses the ground rainfall data as a teacher signal to learn the relationship between these. In other words, the learning unit 71a learns the relationship between the reception intensity data, wind direction / speed data, and elevation angle θ and the amount of rainfall.

[0057] Based on the learning results of the learning unit 71a, the calculation processing unit 71b estimates the amount of rainfall at any position on the two-dimensional map from the reception intensity data, wind direction / speed data, and data related to the elevation angle θ. As described above, according to this embodiment, the amount of rainfall can be estimated more accurately by taking into account the elevation angle θ of the antenna 22 (i.e., the altitude to the rain cloud C).

[0058] (Embodiment 3) A rainfall estimation system 103 according to an embodiment of the present disclosure will be described with reference to Fig. 6 to Fig. 8. Fig. 6 is a diagram illustrating a schematic configuration of the rainfall estimation system 103. In this embodiment, the data processing server 72 includes a learning unit 72a similar to the learning unit 71a, a calculation processing unit 72b similar to the calculation processing unit 71b, and an image analysis unit 72c. The information provision system 203 includes an external server 81, a ground rain gauge 31, an anemometer 41, and a camera 51. The camera 51 is connected to the external server 81 via a predetermined communication network. A plurality of cameras 51 are installed at predetermined intervals on a two-dimensional map. Each camera 51 acquires an image and outputs the image as image data to the external server 81. The camera 51 acquires, for example, images of rivers and road conditions. The image data stored in the external server 81 is collected in real time by a data collection server 61. The image data includes location information of each camera 51. In the following description, each camera 51 is assumed to be located at a "predetermined position" on the two-dimensional map so as to correspond one-to-one to each ground rain gauge 31. However, each camera 51 may be located away from each ground rain gauge 31. Even in this case, the ground rainfall data includes the location information of the ground rain gauge 31, and the image data includes the location information of the camera 51, so machine learning can be performed with high accuracy by taking into account the distance between the ground rain gauge 31 and the camera 51.

[0059] Specifically, the image analysis unit 72c outputs analysis data obtained by analyzing image data output from the camera 51 that captures images during rainfall, learns the relationship between the analysis data and the amount of rainfall, and estimates the rainfall situation at any location.

[0060] The image analysis unit 72c receives the rainfall amount estimation result (hereinafter referred to as "rainfall amount data") from the calculation processing unit 72b. The image analysis unit 72c also requests the data collection server 61 for image data.

[0061] The image analysis unit 72c analyzes the image data. For example, as shown in FIG. 6(B), the image analysis unit 72c calculates the water level of a river from the image data. Here, the "water level" refers to the height from a reference position to the water surface. In the following description, the data obtained by analyzing the image data will be collectively referred to as "analysis data."

[0062] Next, the image analysis unit 72c uses the input signal as analysis data and the rainfall data as training signals to learn the relationship between them. Based on the learning results, the image analysis unit 72c estimates the water level and flooding situation at any position.

[0063] Next, the processing by the image analysis unit 72c will be described with reference to FIG. [Machine Learning] 7A is a flowchart illustrating machine learning by the image analysis unit 72c. First, the image analysis unit 72c receives rainfall data from the calculation processing unit 72b (step S8). The image analysis unit 72c extracts data from the rainfall data that is linked to information related to a predetermined position (step S9). Hereinafter, the "data from the rainfall data that is linked to information related to a predetermined position" will be referred to as "specific data."

[0064] Next, the image analysis unit 72c transmits data relating to the reception intensity data, including information on the time when the reception intensity data, which is the source of the specific data, was acquired, to the data collection server 61, and requests image data from the camera 51 located at a predetermined position from the data collection server 61. In response to the request from the image analysis unit 72c, the data collection server 61 transmits image data at the predetermined position at the time when the learning unit 72a received the reception intensity data to the image analysis unit 72c (step S10).

[0065] Next, the image analysis unit 72c analyzes the image data and generates analysis data indicating the water level and flooding status at a predetermined position (step S11). The image analysis unit 72c uses the specific data as an input signal and the analysis data as a teacher signal to learn the relationship between them (step S12). The image analysis unit 72c then stores the learning results (step S13). Learning by the image analysis unit 72c is repeatedly performed for each predetermined position of each camera 51, and the learning results are accumulated and stored. In this way, accumulating learning results can be expected to enable a more accurate understanding of the situation. Furthermore, since the data collection server 61 provides data in real time so as to correspond to the reception intensity data, a more accurate understanding of the situation can be expected.

[0066] [Situation Estimation] 7B is a flowchart illustrating the situation estimation by the image analysis unit 72c. First, the image analysis unit 72c acquires rainfall data from the calculation processing unit 72b (step S14). Based on the above learning results, the image analysis unit 72c estimates the water level, flooding status, etc. at an arbitrary position from the rainfall data (step S15). In this way, in this embodiment, the relationship between the rainfall data (specific data, input signal) and the analysis data (teacher signal) is learned, and then the water level, flooding status, etc. at an arbitrary position is estimated based on the rainfall data, thereby making it possible to grasp the disaster situation, etc. in a planar manner by corresponding it to a two-dimensional map.

[0067] Next, with reference to FIG. 8, an image of actual use of the rainfall estimation system 103 of this embodiment will be described. As shown in the figure, the information providing system 203 includes a river CCTV (Closed Circuit Television) or a simple river camera as an example of the camera 51. Also, an image collection and storage server owned by the Japan Meteorological Agency is included as an example of the external server 81. The actual information providing system 203 also includes an information providing server 82 for providing information to the public. The information providing server 82 acquires data from the image collection and storage server and makes it available to the public.

[0068] The image analysis unit 72c (data processing server 72) of the rainfall estimation system 102 acquires and analyzes image data from the data collection server 61. The image analysis unit 72c also learns the relationship between rainfall data and analysis data. For example, the image analysis unit 72c performs image analysis to determine the water level of a river, as shown by (1) in the figure. The analysis result of the river water level is displayed as a cross-sectional view, as shown by (1)' in the figure, so that the user can easily understand the water level. For example, the image analysis unit 72c also performs image analysis to determine the flood depth, as shown by (2) in the figure. Here, "flood depth" refers to the height from the ground to the water surface. The analysis result of the flood depth is displayed in different colors, as shown by (2)' in the figure, so that the user can easily understand the extent of the flooding. These analysis results are provided in conjunction with the information provision server 82 disclosing the information to the outside.

[0069] The device of the present invention can also be realized by a computer and a program, and the program can be recorded on a recording medium or provided via a network. The program of the present disclosure is a program for causing a computer to realize each function of the device according to the present disclosure, and a program for causing a computer to execute each procedure of the method executed by the device according to the present disclosure. [Industrial Applicability]

[0070] The processing in the rainfall estimation system of the present disclosure can be applied to a radar rain gauge system that uses a radar device. [Explanation of symbols]

[0071] 21, 21A: Radar equipment 22: Antenna 23: Support part 24: Motor 25: Control device 31: Ground rain gauge 41:Anemometer 51: Camera 61: Data collection server 71: Data processing server 71a: Learning Department 71b: Processing unit 71c: Image analysis department 81: External server 101, 101A, 102, 103: Rainfall estimation system 201, 203: Information provision system C: Rain clouds D: Raindrops W: Wind

Claims

1. The amount of rainfall at an arbitrary position is estimated based on reception intensity data obtained from the radar device according to the intensity of the radio waves emitted from the radar device and reflected by raindrops, ground rainfall data obtained from a first measuring instrument that observes the amount of rainfall at a predetermined point on the ground surface, and environmental data obtained from a second measuring instrument that measures the influence of the environment on the raindrops before they reach the ground surface. Processing unit.

2. a learning unit that learns a relationship between the reception intensity data and the environmental data and the amount of rainfall based on the reception intensity data, the ground rainfall data, and the environmental data; estimating the amount of rainfall at an arbitrary position from the reception strength data and the environmental data according to the learning result; The processor according to claim 1 .

3. The second measuring device measures a physical quantity related to wind until the raindrops reach the ground surface, and outputs the physical quantity related to wind as the environmental data. The processor according to claim 1 .

4. estimating the amount of rainfall at an arbitrary position based on the reception intensity data, the ground rainfall data, the environmental data, and also on the elevation angle of the antenna of the radar device; The processor according to claim 1 .

5. a learning unit that learns a relationship between the reception intensity data, the ground rainfall data, the environmental data, and the elevation angle and the rainfall amount based on the reception intensity data, the ground rainfall data, the environmental data, and the elevation angle, estimating the amount of rainfall at an arbitrary position from the reception strength data, the environmental data, and the elevation angle according to the learning result; The processor according to claim 4.

6. The image analysis unit further includes an image analysis unit that outputs analysis data obtained by analyzing image data output from a camera that acquires images, learns the relationship between the analysis data and the amount of rainfall, and estimates the rainfall situation at an arbitrary position. The processor according to claim 1 .

7. the radar device; and the processing device according to any one of claims 1 to 6. Rainfall estimation system.

8. acquiring reflected waves, which are radio waves emitted from a radar device and reflected by raindrops, as reception intensity data corresponding to the intensity of the reflected waves; a step of estimating the amount of rainfall at an arbitrary position based on the reception intensity data, ground rainfall data obtained from a first measuring instrument that observes the amount of rainfall at a predetermined point on the ground surface, and environmental data obtained from a second measuring instrument that measures the influence of the environment on the raindrops until they reach the ground surface; Rainfall estimation methods including.

9. learning a relationship between the reception intensity data, the ground rainfall data, and the environmental data based on the reception intensity data, the ground rainfall data, and the environmental data; and estimating the amount of rainfall at an arbitrary position from the reception strength data and the environmental data according to the result of the learning. The rainfall estimation method according to claim 8.

Images