Selection device, selection method, and selection program

Abstract

A selection device 1 comprises: a generation part 101 that generates an initial value candidate for prediction simulation at each position in an observation area by using an observation value of weather estimated by a weather estimation model; and a selection part 102 that selects initial value candidates up to the number of thresholds with high observation accuracy as initial values on the basis of statistical information related to observation accuracy at each position of the weather estimation model.

Description

Selection device, selection method, and selection program

[0001] The present disclosure relates to a selection device, a selection method, and a selection program.

[0002] In the field of weather and oceanographic prediction, data assimilation, which is a process of generating initial values for prediction simulation of observed values using observation data, is known as a technology that complementarily connects observation and prediction simulation by a numerical prediction model (hereinafter referred to as prediction simulation) (see Non-Patent Documents 1 and 2).

[0003] The initial value is generated by combining, when the time of initial value generation is T, the observed value at time T obtained from observation data, and the analysis value which is the output value of prediction simulation executed from before time T and output at time T.

[0004] Japan Meteorological Agency, "Chapter 3 Data Assimilation", [online], [retrieved February 28, 2025], <URL: https: / / www.jma.go.jp / jma / kishou / books / nwptext / 45 / 1_chapter3.pdf> Miyoshi, et al., "Data Assimilation in Meteorology", [online], [retrieved February 28, 2025], <URL: https: / / www.metsoc.jp / tenki / pdf / 2007 / 2007_04_0015.pdf>

[0005] To improve the prediction accuracy of prediction simulation, it is desirable to sufficiently prepare observation data with high accuracy for generating initial values. However, the amount of observation data in sea areas and the like is insufficient, making it difficult to improve prediction accuracy.

[0006] The present disclosure has been made in view of the above circumstances, and an object of the present disclosure is to provide a technology capable of selecting initial values effective for prediction.

[0007] A selection device according to one aspect of the present disclosure includes: a generation unit that generates initial value candidates for prediction simulation at each position in an observation area using weather observation values estimated by a weather estimation model; and a selection unit that selects, as initial values, initial value candidates up to the number of thresholds with high observation accuracy based on statistical information related to observation accuracy at each position of the weather estimation model.

[0008] A selection method in one aspect of this disclosure involves a selection method performed by a selection device, in which initial value candidates for prediction simulations at each location in the observation area are generated using weather observation values ​​estimated by a weather estimation model, and initial value candidates up to a number of thresholds with high observation accuracy are selected as initial values ​​based on statistical information related to the observation accuracy at each location of the weather estimation model.

[0009] A selection program in one aspect of this disclosure causes a computer to perform the following processes: generate candidate initial values ​​for prediction simulations at each location in an observation area using weather observation values ​​estimated by a weather estimation model; and select candidate initial values ​​up to a number of threshold values ​​with high observation accuracy as initial values, based on statistical information related to the observation accuracy at each location in the weather estimation model.

[0010] According to this disclosure, initial values ​​that are effective for prediction can be selected.

[0011] Figure 1 shows an example of the overall system configuration. Figure 2 is a flowchart showing an example of the overall system operation. Figure 3 is a reference diagram used when explaining the overall system operation. Figure 4 shows an example of the configuration of the selected device. Figure 5 is a sequence diagram showing an example of the overall system operation. Figure 6 is a reference diagram used when explaining a specific example. Figure 7 is a reference diagram used when explaining a specific example. Figure 8 is a reference diagram used when explaining a specific example. Figure 9 shows an example of the hardware configuration of the selected device.

[0012] Embodiments of this disclosure will be described below with reference to the drawings. In the drawings, the same parts are denoted by the same reference numerals and their descriptions are omitted.

[0013] This disclosure aims to select initial values ​​that are effective for prediction.

[0014] To achieve this objective, this disclosure generates initial value candidates using weather observations estimated by a weather estimation model, and selects initial values ​​from among the candidates based on statistical information related to the observation accuracy of the weather estimation model. This makes it possible to select initial values ​​that are more effective for prediction than when using observation data alone.

[0015] Figure 1 shows an example of the overall configuration of the system according to this embodiment.

[0016] The system comprises a selection device 1 that selects initial values ​​for a prediction simulation using a weather estimation model, a prediction device 2 that executes a prediction simulation using the initial values ​​selected by the selection device 1, a client terminal 3 that instructs the start or end of observation, and an observation device 4 that performs observation.

[0017] Figure 2 is a flowchart showing an example of the overall operation of the system. Figure 3 is a reference diagram used to explain the overall operation of the system.

[0018] At time T when generating initial values, the selection device 1 obtains the observed values ​​at each observation location estimated by the weather estimation model, and statistical information related to the observation accuracy of the weather estimation model at each observation location (for example, the frequency distribution of observation locations, and the error between the observation data from the observation instrument 4 and the observed values ​​estimated by the weather estimation model) (step S1).

[0019] Next, the selection device 1 uses the observed values ​​at each observation location to generate candidate initial values ​​for prediction simulation at each observation location (step S2).

[0020] Finally, based on the above statistical information, the selection device 1 selects the top n initial value candidates with high observation accuracy as initial values ​​(step S3). Alternatively, the selection device 1 evaluates the difference between the selected initial values ​​and the analyzed values ​​analyzed by the prediction sim, and if the difference is above a certain level, adopts those initial values ​​as initial values ​​(step S3).

[0021] Subsequently, the prediction device 2 executes a prediction simulation using the initial values ​​selected by the selection device 1. The prediction device 2 may use the selected initial values ​​as they are, or it may use a value obtained by combining the selected initial values ​​with the analysis values ​​which are the output values ​​of the prediction simulation as the initial values ​​for the prediction simulation, or it may use a value obtained by combining the selected initial values ​​with both the said analysis values ​​and the observation data as the initial values ​​for the prediction simulation.

[0022] From this point onward, the selection device 1 executes steps S1 to S3 each time the time for generating the initial value arrives.

[0023] The prediction simulation and initial value generation performed by prediction device 2 are based on existing technologies. The prediction simulation is a process that uses a numerical weather prediction model to simulate the prediction result at a predicted time (T+n) from initial values ​​at a predetermined time T. Initial value generation is the process called "data assimilation" mentioned at the beginning. In other words, it is a process that generates initial values ​​for prediction at time T by combining the observed data at time T with the analyzed values ​​at time T output from the prediction simulation.

[0024] Conventionally, initial values ​​were generated using observation data from observation instrument 4, as shown by the dashed arrow in Figure 3. In contrast, this embodiment selects initial values ​​with high observation accuracy using a weather estimation model. This makes it possible to select initial values ​​that are effective for prediction.

[0025] Figure 4 shows an example of the configuration of the selection device 1.

[0026] The selection device 1 is connected to the prediction device 2, the client terminal 3, and the observation equipment 4 via a wireless / wired network N.

[0027] The selection device 1 comprises a processing unit 10 and a storage unit 20. The processing unit 10 and the storage unit 20 are mutually communicative. The processing unit 10 comprises a generation unit 101 and a selection unit 102. The storage unit 20 comprises a first storage unit 201, a second storage unit 202, and a third storage unit 203.

[0028] The generation unit 101 has the function of executing steps S1 and S2 in Figure 2.

[0029] For example, the generation unit 101 has the function of reading observation data from the observation instrument 4 from the first storage unit 201 and constructing a weather estimation model for estimating desired weather observation values ​​using the read observation data.

[0030] For example, the generation unit 101 has a function to store in the second storage unit 202 the observed weather values ​​estimated by the constructed weather estimation model and statistical information related to the observation accuracy at each observation location of the weather estimation model (for example, the frequency distribution of observation locations, and the error between the observation data from the observation instrument 4 and the observed values ​​estimated by the weather estimation model).

[0031] For example, the generation unit 101 reads weather observation values ​​estimated by the weather estimation model and statistical information related to the observation accuracy at each observation location of the weather estimation model from the second storage unit 202, generates initial value candidates for prediction simulation at each observation location in the observation area using the weather observation values ​​estimated by the weather estimation model, and has a function to set a priority for each observation location based on the statistical information related to the observation accuracy. The generation unit 101 also has a function to store the initial value candidates for each observation location and the priority for each observation location in the third storage unit 203.

[0032] The selection unit 102 has the function of executing step S3 in Figure 2.

[0033] For example, the selection unit 102 reads initial value candidates for each observation position and the priority of each observation position from the third storage unit 203, and has the function of selecting initial value candidates up to a number of thresholds with high observation accuracy as initial values ​​based on the priority of each observation position.

[0034] For example, the selection unit 102 has a function to adopt the selected initial value as the initial value if the difference between the selected initial value and the analyzed value analyzed by the prediction sim is greater than or equal to a threshold. In other words, the selection unit 102 calculates the amount of correction (the difference between the initial value and the analyzed value) when changing the analyzed value to the initial value, and determines whether or not to adopt the initial value based on this amount of correction.

[0035] The first memory unit 201 has, for example, a function to store observation data from the observation instrument 4.

[0036] The second memory unit 202 has a function for storing, for example, weather observation values ​​estimated by a weather estimation model, and statistical information related to the observation accuracy of the weather estimation model.

[0037] The third memory unit 203 has a function to store, for example, initial value candidates for each observation position and the priority of each observation position.

[0038] Figure 5 is a sequence diagram showing an example of the overall system operation.

[0039] The client terminal 3 transmits a start signal to the observation instrument 4 to begin accumulating observation data (step S101).

[0040] The observation device 4 determines whether it is necessary to continue observation (step S102).

[0041] When it is determined that observation needs to be continued, the observation device 4 performs observation at the current observation position (step S103), and transmits the observed observation data to the selection device 1 (step S104). Thereafter, the first storage unit 201 stores the observation data of the observation device 4 (step S105).

[0042] When it is determined that observation does not need to be continued, the observation device 4 notifies the client terminal 3 that no observation will be performed (step S106). Thereafter, the process is terminated.

[0043] After step S105, the generation unit 101 determines whether it is necessary to generate an initial value (step S107).

[0044] When it is determined that an initial value needs to be generated, the generation unit 101 reads observation data from the first storage unit 201 (step S108), and reads various parameters necessary for constructing a weather estimation model from the second storage unit 202 (step S109). The generation unit 101 constructs a weather estimation model for estimating desired weather observation values using the read observation data and the various parameters (step S110).

[0045] The generation unit 101 generates initial value candidates for prediction simulation at each observation position in the observation area using the weather observation values estimated by the weather estimation model, and sets a priority for each observation position based on statistical information related to observation accuracy at each observation position of the weather estimation model (step S111). The generation unit 101 transmits the initial value candidates for each observation position and the priority of each observation position to the third storage unit 203 (step S112). Thereafter, the third storage unit 203 stores the initial value candidates for each observation position and the priority of each observation position (step S113).

[0046] The selecting unit 102 reads the initial value candidates of each observation position and the priority of each observation position from the third storage unit 203 (step S114), and based on the priority of each observation position, selects the initial value candidates up to the number of thresholds with high observation accuracy as initial values (step S115). Alternatively, the selecting unit 102 evaluates the difference between the selected initial value and the analysis value analyzed by the prediction sim, and when the difference is equal to or greater than a certain value, adopts the initial value as the initial value (step S115).

[0047] After step S115, the process returns to step S107 (step S116). The process also returns to step S107 if it is determined in step S107 that initial value generation is not required (step S117).

[0048] A modified example of steps S110 to S111 will be described.

[0049] In the operation example of FIG. 5, the case where the selection apparatus 1 constructs a weather estimation model has been described, but the selection apparatus 1 may utilize an existing weather estimation model constructed by another apparatus. In that case, the selection apparatus 1 stores, in the first storage unit 201, the observed weather values estimated by the existing weather estimation model and statistical information related to the observation accuracy of the existing weather estimation model. Step S110 is not performed. In step S111, the selection apparatus 1 generates initial value candidates for each observation position using the observation values and observation accuracy read from the first storage unit 201, and sets a priority for each observation position.

[0050] Next, a specific example will be described.

[0051] The generating unit 101 generates initial value candidates for each observation position using observation values estimated by a weather estimation model. For example, the generating unit 101 uses, as initial value candidates for each observation position, observation values of each observation position estimated by the weather estimation model at the time of construction. When the weather estimation model is updated with newly observed observation data, observation values of each observation position estimated by the updated weather estimation model are used as initial value candidates for each observation position. The generating unit 101 may use, as an initial value candidate, a value obtained by performing arbitrary arithmetic processing on an observation value estimated by the weather estimation model.

[0052] Furthermore, the generation unit 101 sets a priority for each observation location based on statistical information related to the observation accuracy of each observation location in the weather estimation model. In this case, the generation unit 101 sets a higher priority for observation locations with higher observation quality. For example, if the statistical information is the frequency distribution of observation locations, a higher priority is set for observation areas with a high frequency of observation locations (areas with a large number of data points). If the statistical information is the error between the observation data from the observation instrument 4 and the observation values ​​estimated by the weather estimation model, a higher priority is set for observation locations with a small error.

[0053] Figure 6 shows an example of setting priorities on a grid of sea surface data. Figure 7 shows an example of setting priorities when using a weather estimation model to estimate typhoon intensity. "Number of data points" corresponds to the frequency distribution of observation locations, and "1 - error rate" corresponds to the error. Priorities may also be set using both "Number of data points" and "1 - error rate". For example, priority can be determined based on the score S calculated using equation (1).

[0054]

[0055] M is the "number of data points". E is "1 - error rate".

[0056] Priority can be determined by ranking based on observation quality, or by the score value of S itself. Priority can be set arbitrarily.

[0057] Next, the selection unit 102 selects initial value candidates up to a number of thresholds with high observation accuracy based on the priority of each observation position, and adopts initial values ​​for which the difference between the selected initial values ​​and the analyzed values ​​analyzed by the prediction sim (the amount of correction when changing the analyzed values ​​to initial values; see Figure 8) is greater than or equal to the threshold. If the amount of correction is small, the impact on the prediction sim is small, so initial values ​​with a small amount of correction (difference from the analyzed value) are excluded.

[0058] In this case, in the typhoon coordinate system shown in Figure 7(a), an upper limit on the number of elements to be adopted may be set for each region based on the positional relationship (quadrant Q and distance r* from the center O), and initial values ​​up to the upper limit may be selected from those with higher priority.

[0059] Alternatively, typhoon forecast information published by government agencies such as the Japan Meteorological Agency can be obtained, and observation data for observation locations included in the typhoon forecast information can be used as initial values ​​as is. For observation locations other than those mentioned, initial values ​​may be selected based on observation values ​​estimated by a weather estimation model.

[0060] As described above, according to this embodiment, (1) initial value candidates for prediction simulation at each location in the observation area are generated using weather observation values ​​estimated by the weather estimation model, and (2) initial value candidates up to a number of high observation accuracy thresholds are selected as initial values ​​based on statistical information related to the observation accuracy of the weather estimation model at each observation location, so that initial values ​​that are effective for prediction can be selected.

[0061] This disclosure is not limited to the embodiments described above. Numerous modifications are possible within the scope of the gist of this disclosure. For example, it can be applied to observing atmospheric temperature and humidity using a drone.

[0062] The selection device 1 of this embodiment described above can be realized using a general-purpose computer system, for example, as shown in Figure 9, which includes a CPU 901, a memory 902, a storage 903, a communication device 904, an input device 905, and an output device 906. The memory 902 and the storage 903 are storage devices. In this computer system, each function of the selection device 1 is realized when the CPU 901 executes a predetermined program loaded onto the memory 902.

[0063] The selection device 1 may be implemented on a single computer. The selection device 1 may be implemented on multiple computers. The selection device 1 may also be a virtual machine implemented on a computer. The program for the selection device 1 can be stored on a computer-readable recording medium such as an HDD, SSD, USB memory, CD, or DVD. A computer-readable recording medium is, for example, a non-transitory recording medium. The program for the selection device 1 can also be distributed via a communication network.

[0064] 1 Selection device 2 Prediction device 3 Client terminal 4 Observation equipment 10 Processing unit 20 Storage unit 101 Generation unit 102 Selection unit 201 First storage unit 202 Second storage unit 203 Third storage unit 901 CPU 902 Memory 903 Storage 904 Communication device 905 Input device 906 Output device

Claims

1. A selection device comprising: a generation unit that generates initial value candidates for prediction simulations at each location in an observation area using weather observation values ​​estimated by a weather estimation model; and a selection unit that selects initial value candidates up to a number of high observation accuracy thresholds as initial values ​​based on statistical information related to the observation accuracy at each location of the weather estimation model.

2. The selection device according to claim 1, wherein the selection unit selects the initial value when the difference between the initial value and the analyzed value analyzed in the prediction simulation is greater than or equal to a threshold.

3. A selection method performed by a selection device, wherein the selection method generates initial value candidates for prediction simulations at each location in the observation area using weather observation values ​​estimated by a weather estimation model, and selects initial value candidates up to a number of high observation accuracy thresholds as initial values ​​based on statistical information related to the observation accuracy at each location of the weather estimation model.

4. A selection program that causes a computer to perform the following: generate candidate initial values ​​for prediction simulations at each location in the observation area using weather observation values ​​estimated by a weather estimation model; and select candidate initial values ​​up to a number of high-precision thresholds based on statistical information related to the observation accuracy at each location of the weather estimation model.

Images