Precipitable water amount estimation system and method, learning system and method, storage medium

Abstract

Provided is a precipitable water amount estimation system and method, a learning system and method, and a storage medium, which can observe water vapor in a local range without using liquid nitrogen for calibration. A learning system (4) of a precipitable water amount estimation model according to the present application includes: a radio wave intensity acquisition unit (40) that acquires radio wave intensities of a plurality of frequencies in radio waves received by a microwave radiometer (3); a precipitable water amount acquisition unit (41) that acquires a precipitable water amount calculated based on an atmospheric delay of a GNSS signal received by a GNSS receiver (2); and a learning unit (43) that causes a learning model (43a) to perform machine learning in a manner in which input data based on the radio wave intensities of the plurality of frequencies is input and the precipitable water amount is output, based on the radio wave intensities of the plurality of frequencies and the precipitable water amount at a plurality of time points in a prescribed period.

Description

Technical Field

[0001] This disclosure relates to a system and method for estimating precipitable water, a learning system and method, and a storage medium. Background Technology

[0002] It is known that the observation of precipitable water, i.e. water vapor observation, uses Global Navigation Satellite System (GNSS) receivers, microwave radiometers, etc.

[0003] Water vapor observation using a GNSS receiver utilizes multi-frequency radio waves radiated from satellites. If radio waves of two or more different frequencies radiated from four or more satellites can be received, the delay in these radio waves can be captured. This delay corresponds to the amount of water vapor, thus allowing for the observation of water vapor levels. Water vapor observation using GNSS can be performed stably and without calibration. However, because GNSS uses a variety of satellites deployed throughout the day, while an average value of water vapor over a wide area can be obtained, it is impossible to observe water vapor in localized areas. Furthermore, Patent Document 1 describes water vapor observation using GNSS.

[0004] Water vapor observation using a microwave radiometer involves measuring the radio waves radiated from water vapor or clouds in the atmosphere. Due to the directionality of the receiver's antenna or horn, it is possible to measure water vapor over a localized area compared to GNSS-based observations. However, to prevent machine drift and to accurately measure brightness and temperature, periodic calibration with liquid nitrogen is required. Liquid nitrogen is difficult to transport or handle. Furthermore, Patent Document 2 specifically describes a microwave radiometer.

[0005] [Existing technical documents]

[0006] [Patent Literature]

[0007] Patent Document 1: Japanese Patent Application Publication No. 2010-60444

[0008] Patent Document 2: U.S. Patent Application Publication No. 2014 / 0035779 Summary of the Invention

[0009] [The problem the invention aims to solve]

[0010] This disclosure provides a technique for observing water vapor in a localized area without the need for calibration using liquid nitrogen.

[0011] [Technical means to solve the problem]

[0012] The learning system of the precipitable water estimation model disclosed herein includes: a radio wave intensity acquisition unit for acquiring the radio wave intensity of multiple frequencies in the radio waves received by a microwave radiometer; a precipitable water acquisition unit for acquiring the precipitable water calculated based on the atmospheric delay of the GNSS signal received by a GNSS receiver; and a learning unit for performing machine learning on the estimation model by calculating the precipitable water using input data based on the radio wave intensity of the multiple frequencies at multiple time points within a specified period and the precipitable water. Attached Figure Description

[0013] Figure 1 This is a block diagram illustrating the learning system and structure of the precipitation estimation system for the implementation method of the precipitation estimation model.

[0014] Figure 2 It is a flowchart representing the processes performed by the learning system.

[0015] Figure 3 This is a flowchart illustrating the processes performed by the precipitation estimation system.

[0016] Figure 4 It is a graph representing the frequency spectrum of the radio wave intensity received by the microwave radiometer.

[0017] Figure 5 It is a graph that compares the precipitable amount estimated by the precipitable amount estimation system for a certain period with the precipitable amount obtained based on the detection data for the same period.

[0018] [Explanation of Symbols]

[0019] 4: Learning System

[0020] 40: Radio Wave Intensity Acquisition Unit

[0021] 41: Precipitation Acquisition Department

[0022] 43: Study Department

[0023] 44: Dimensional Reduction Department

[0024] 45: Standardization Processing Department

[0025] 5: Precipitation estimation system

[0026] 50: Presumption Department

[0027] 51: Standardization Processing Department

[0028] 52: Dimensional Reduction Department Detailed Implementation

[0029] Hereinafter, an embodiment of the present disclosure will be described with reference to the drawings.

[0030] Figure 1 This is a diagram showing the structure of the learning system 4 and the precipitation estimation system 5 (also called estimation system 5) of the precipitation estimation model in this embodiment.

[0031] like Figure 1 As shown, in this embodiment, the learning system 4 and the precipitation estimation system 5 of the precipitation estimation model are built on the same computer system, but they can be used independently. That is, only the learning system 4 or only the precipitation estimation system 5 can be installed.

[0032] <Learning System 4>

[0033] Figure 1 The learning system 4 shown includes: a radio wave intensity acquisition unit 40, a precipitation data acquisition unit 41, and a learning unit 43.

[0034] Figure 1 The radio wave intensity acquisition unit 40 shown acquires the radio wave intensity of multiple frequencies among the radio waves received by the microwave radiometer 3. In this embodiment, it acquires the radio wave intensity of N (N=30) different frequencies between 18 GHz and 26.5 GHz. The radio wave intensity acquisition unit 40 acquires the radio wave intensity of 30 different frequencies (f1, f2, ..., f29, f30) [p(f1), p(f2), ..., p(f29), p(f30)]. Here, the radio wave intensity is represented as p(f), where f represents the frequency. The radio wave intensity of multiple frequencies acquired by the radio wave intensity acquisition unit 40 is stored in the storage unit 42 in the form of radio wave intensity time sequence data D2.

[0035] like Figure 4 As shown, the peak intensity of the electromagnetic waves radiated from water vapor and cloud water in the upper atmosphere is 22 GHz. Figure 4 In this embodiment, the received intensity of the microwave radiometer 3 is denoted as p(f), where f represents the frequency. For example, a 22 GHz radio wave contains precipitable water, i.e., water vapor and cloud water components. To remove the cloud water component contained in the 22 GHz radio wave, the cloud water component is calculated based on the radio wave intensity at frequencies other than 22 GHz. Therefore, multiple different radio wave intensities are required. Furthermore, while 22 GHz is shown as an example, water vapor and cloud water components are also included at frequencies other than 22 GHz, so the combination of frequencies is not limited to a combination of 22 GHz and frequencies other than 22 GHz. In this embodiment, N is set to 30, but the number of N can be appropriately changed. Moreover, the frequency range preferably includes 22 GHz or ±1 GHz before and after 22 GHz. In this embodiment, N = 30, but it is not limited to this. To improve the specific accuracy of water vapor and cloud water components, N is preferably a natural number of 3 or more.

[0036] In this embodiment, an actuator is used to periodically pass the blackbody through the receiving range of the antenna of the microwave radiometer 3, receiving radio waves from the blackbody of known intensity and from the sky. The received intensity p(f) of the microwave radiometer 3 is the difference between the intensity of the radio waves from the sky (ps(f)) and the intensity of the radio waves from the blackbody (pb(f)). Of course, the microwave radiometer 3 is not limited to this; it can also periodically manipulate a mirror to receive radio waves from the blackbody.

[0037] Figure 1 The precipitable water volume acquisition unit 41 shown acquires the precipitable water volume calculated based on the atmospheric delay (strictly speaking, tropospheric delay) of the GNSS signal received by the GNSS receiver 2. It is known that precipitable water volume (PWV) obtained using GNSS can be calculated based on the GNSS signal, altitude and other coordinate values, temperature, and air pressure. The precipitable water volume acquisition unit 41 acquires the GNSS precipitable water volume using the GNSS signal and altitude information obtained from the GNSS receiver 2, and the temperature and air pressure obtained from the weather sensor 1. The GNSS precipitable water volume acquired by the precipitable water volume acquisition unit 41 is stored in the storage unit 42 in the form of a time series data D1 of GNSS precipitable water volume.

[0038] Figure 1 The learning unit 43, as shown, enables the estimation model 43a to perform machine learning based on the time series data D1 of precipitable water and the time series data D2 of radio wave intensity. Specifically, the learning unit 43 enables the estimation model 43a to perform machine learning based on the radio wave intensity and precipitable water at multiple frequencies at multiple time points within a specified period, by taking the input data based on the radio wave intensity at multiple frequencies as input and outputting the precipitable water. The teacher dataset used by the learning unit 43 is data that establishes a correlation between the precipitable water at a certain time t and the input data based on the radio wave intensity at multiple frequencies [p(f1), p(f2), ..., p(f29), p(f30)] at the same time t. The input data, as long as it is data based on the radio wave intensity at multiple frequencies, can be the radio wave intensity at multiple frequencies itself, or it can be data obtained by reducing the dimensionality of the radio wave intensity at multiple frequencies. If model 43a is presumed to be a teacher-assisted machine learning model, then various models such as linear regression, regression tree, random forest, support vector machine, neural network, and ensemble can be used. In this embodiment, which will be explained in detail below, multinomial regression with terms of quadratic or higher is used, employing complex regression with multiple variables, but it is not limited to this.

[0039] like Figure 1As shown, the learning system 4 preferably includes a dimensionality reduction unit 44, which performs dimensionality reduction processing on the radio wave intensities of multiple frequencies to calculate the dimensionality-reduced input data representing the radio wave intensities of multiple frequencies. By performing dimensionality reduction, the original features represented by the radio wave intensities of multiple frequencies can be reproduced and the number of dimensions can be reduced, thereby reducing computational costs and avoiding the curse of dimensionality (overlearning). The dimensionality reduction method in this embodiment is Principal Component Analysis (PCA), but it is not limited to this. For example, other algorithms such as factor analysis, multifactor analysis, autoencoder, independent component analysis, and nonnegative matrix factorization can be used.

[0040] In this embodiment, the dimensionality reduction unit 44 uses principal component analysis, selecting the first principal component, the second principal component, and the third principal component as input data. Of course, this is not a limitation, and various modifications can be made. For example, the input data can be set to only the first principal component of the principal component analysis, or it can be set to both the first and second principal components. That is, a predetermined number (any natural number greater than 1) of principal components after the first priority is selected as input data. The predetermined number can be appropriately set according to the required accuracy. The reason why the first principal component must be included is that the original features of the first principal component have the highest reproducibility.

[0041] Figure 1 Before performing dimensionality reduction using principal component analysis, the standardization processing unit 45 performs standardization processing on the radio wave intensities [p(f1), p(f2), ..., p(f29), p(f30)] at multiple times and frequencies at multiple time points. The standardization processing unit 45 performs standardization processing on the time series data D2 of the radio wave intensities stored in the storage unit 42, and stores the standardized time series data D3 of the radio wave intensities in the storage unit 42. Standardization processing is used for centering (setting the mean to 0) and scaling (setting the standard deviation to 1). The standardization processing calculates the mean and standard deviation for each radio wave intensity at multiple time points, subtracts the mean from the original data, and divides the result by the standard deviation, thereby converting each original radio wave intensity into a standardized radio wave intensity. The calculated mean and standard deviation are stored in the storage unit 42 as standardization parameters used in the standardization processing of the precipitable water estimation system 5 described below (see reference). Figure 1 ).

[0042] In addition, the learning system 4 of this embodiment has a dimensionality reduction unit 44 and a standardization processing unit 45, but these can be omitted.

[0043] <Specific examples of Study Section 43 and Presumed Model 43a>

[0044] Figure 1 The learning unit 43 shown uses the first principal component PC1, the second principal component PC2, and the third principal component PC3 as input data to construct a presumption model 43a for calculating precipitable water volume (PWV). The presumption model 43a is a transformation formula using complex regression, represented by equation (1) shown below. By using the least squares method for fitting, the unknown coefficients S1 to S10 are calculated, and the presumption model 43a is constructed.

[0045] [Formula 1]

[0046]

[0047] <Precipitation estimation system 5>

[0048] Figure 1 The illustrated precipitable water estimation system 5 includes a radio wave intensity acquisition unit 40 and an estimation unit 50. The estimation unit 50 uses the estimation model 43a constructed by the learning unit 43, takes input data based on the radio wave intensity of multiple frequencies acquired by the radio wave intensity acquisition unit 40, and outputs the corresponding precipitable water. For the estimation unit 50, the radio wave intensity of multiple frequencies at the estimation point [p(f1), p(f2), ..., p(f29), p(f30)] can also be input, but to improve accuracy, a standardization processing unit 51 and a dimension reduction unit 52 are preferably provided.

[0049] Figure 1 The standardization processing unit 51 shown performs standardization processing on the intensity of radio waves at multiple frequencies using preset parameters before the dimensionality reduction processing is performed by the dimensionality reduction unit 52. The standardization parameters are the parameters (mean value, standard deviation) calculated by the standardization processing unit 45 of the learning system 4. The standardization processing unit 51 does not calculate the parameters (mean value, standard deviation), but the other processing is the same as that of the standardization processing unit 45 of the learning system 4.

[0050] Figure 1 The dimension reduction unit 52 shown performs dimension reduction processing on the radio wave intensity of multiple frequencies, and calculates the dimension-reduced input data representing the radio wave intensity of multiple frequencies. The dimension reduction unit 52 uses the same parameters calculated by the dimension reduction unit 44 of the learning system 4.

[0051] <Learning Methods for Precipitation Estimation Models>

[0052] use Figure 2 The learning method for the precipitable water estimation model is explained. For example... Figure 2As shown, in step ST100, the radio wave intensity acquisition unit 40 acquires the radio wave intensity of multiple frequencies in the radio waves received by the microwave radiometer. In step ST101, the precipitable water acquisition unit 41 acquires the precipitable water calculated based on the atmospheric delay of the GNSS signal received by the GNSS receiver. The order of steps ST100 and ST101 is different.

[0053] In the next step ST102, the standardization processing unit 45 performs standardization processing on the radio wave intensity at multiple frequencies at multiple time points.

[0054] In the next step ST103, the dimensionality reduction unit 44 performs dimensionality reduction processing on the radio wave intensity of multiple frequencies through principal component analysis, and calculates the dimensionality-reduced input data representing the radio wave intensity of multiple frequencies.

[0055] In the next step ST104, the learning unit 43 performs machine learning on the estimation model by taking the input data based on the radio wave intensity and precipitable water at multiple times during the specified period and outputting the precipitable water as input.

[0056] <Methods for estimating precipitation>

[0057] use Figure 3 The method for estimating precipitable water is explained. For example... Figure 3 As shown, in step ST201, the radio wave intensity acquisition unit 40 acquires the radio wave intensity of multiple frequencies in the radio waves received by the microwave radiometer.

[0058] In the next step ST202, the standardization processing unit 51 performs standardization processing on the radio wave intensity of multiple frequencies.

[0059] In the next step ST203, the dimensionality reduction unit 52 performs dimensionality reduction processing on the radio wave intensity of multiple frequencies through principal component analysis, and calculates the dimensionality-reduced input data representing the radio wave intensity of multiple frequencies.

[0060] In the next step ST204, the estimation unit 50 uses the estimation model 43a to output the precipitable water amount. The estimation model 43a is machine learning in a manner that outputs the precipitable water amount by taking input data based on the intensity of radio waves at multiple frequencies as input. The precipitable water amount corresponds to the input data based on the acquired intensity of radio waves at multiple frequencies.

[0061] Figure 5This is a graph comparing the precipitable rainfall for a certain period, estimated by the prediction model constructed by learning system 4 and the precipitable rainfall estimation system 5, with the precipitable rainfall obtained based on Sonde data for the same period. The Sonde data is data released by the Japan Meteorological Agency and consists of actual meteorological observations taken by sending a sensor-equipped balloon into the air. Figure 5 As shown, the root mean square error (RMSE) is 1.8 mm, which achieves a certain level of accuracy.

[0062] Furthermore, this method acquires radio wave intensities at multiple frequencies. Therefore, even if some frequencies contain noise due to the use of a common amplifier with high noise temperatures, the use of multiple frequencies helps suppress the noise's impact. Thus, it is considered more noise-resistant than, for example, using two specific frequencies to estimate precipitation using a predetermined formula. In other words, even if some noise is present, it can be covered by multiple frequencies, thus reducing the need for high-performance equipment and enabling cost reduction.

[0063] As described above, the learning system 4 of the precipitable water estimation model in this embodiment includes: a radio wave intensity acquisition unit 40, which acquires the radio wave intensity of multiple frequencies in the radio waves received by the microwave radiometer 3; a precipitable water acquisition unit 41, which acquires the precipitable water calculated based on the atmospheric delay of the GNSS signal received by the GNSS receiver 2; and a learning unit 43, which performs machine learning on the estimation model 43a by taking the input data based on the radio wave intensity of multiple frequencies at multiple time points in a specified period and the precipitable water as input and outputting the precipitable water.

[0064] The learning method of the precipitable water estimation model in this embodiment includes: acquiring the radio wave intensity of multiple frequencies in the radio waves received by the microwave radiometer 3; acquiring the precipitable water based on the atmospheric delay of the GNSS signal received by the GNSS receiver 2; and performing machine learning on the estimation model 43a by taking the input data based on the radio wave intensity of multiple frequencies and the precipitable water at multiple time points in a specified period as input and outputting the precipitable water.

[0065] The precipitation estimation system of this embodiment includes: a radio wave intensity acquisition unit 40, which acquires the radio wave intensity of multiple frequencies in the radio waves received by the microwave radiometer 3; and an estimation unit 50, which uses an estimation model 43a to output precipitation amount, wherein the estimation model 43a performs machine learning in a manner that outputs precipitation amount by taking input data based on the radio wave intensity of multiple frequencies as input, and the precipitation amount corresponds to the input data based on the acquired radio wave intensity of multiple frequencies.

[0066] The method for estimating precipitable water volume in this embodiment includes: acquiring the radio wave intensity of multiple frequencies in the radio waves received by the microwave radiometer 3; and using an estimation model 43a to output precipitable water volume, wherein the estimation model 43a is machine learning-based in a manner that takes input data based on the radio wave intensity of multiple frequencies as input and outputs precipitable water volume, wherein the precipitable water volume corresponds to the input data based on the acquired radio wave intensity of multiple frequencies.

[0067] According to the aforementioned learning method, estimation method, and system, because machine learning is performed using input data based on radio wave intensity at multiple frequencies, the correlation between radio wave intensity and precipitable water content, which cannot be elucidated using a single frequency due to the inclusion of both water vapor and cloud water in the radio wave intensity, can be clarified through machine learning, thereby enabling the estimation of water vapor content (precipitable water content). Furthermore, because precipitable water content is obtained using radio wave intensity and GNSS data at multiple time points within a specified period, local water vapor data based on microwave radiometers, which lacks absolute values, can be converted into reliable local water vapor data with consistent absolute values. Even without liquid-based calibration of the microwave radiometer, highly reliable data can be obtained.

[0068] As shown in this embodiment, it preferably includes a dimension reduction unit 44 and a dimension reduction unit 52, which perform dimension reduction processing on the radio wave intensity of multiple frequencies and calculate the dimension-reduced input data representing the radio wave intensity of multiple frequencies.

[0069] Thus, by reducing the dimensionality, frequencies with good sensitivity are selected from multiple frequencies after processing by the high-performance portion of the receiver. Therefore, estimation can be performed even using a general-purpose, inexpensive amplifier. In contrast, without dimensionality reduction, estimation directly uses the low-sensitivity frequency bands processed by the low-performance portion of the receiver, which negatively impacts estimation accuracy. Dimensionality reduction saves the effort of manually removing low-sensitivity frequencies from multiple frequencies; however, it also avoids deterioration in estimation accuracy.

[0070] As shown in this embodiment, the dimensionality reduction units 44 and 52 preferably perform dimensionality reduction through principal component analysis and select a predetermined number of principal components after the first order as input data.

[0071] Therefore, principal component analysis is suitable for dimensionality reduction.

[0072] As shown in the learning system 4 of this embodiment, it preferably includes a standardization processing unit 45, which performs standardization processing on the radio wave intensity of multiple frequencies at multiple time points before performing dimensionality reduction processing using the dimensionality reduction unit 44.

[0073] As shown in the precipitation estimation system 5 of this embodiment, it preferably includes a standardization processing unit 51. Before performing dimensionality reduction processing using the dimensionality reduction unit 52, the standardization processing unit 51 performs standardization processing on the intensity of radio waves at multiple frequencies using pre-set standardization parameters.

[0074] This allows for appropriate dimensionality reduction, thereby improving estimation accuracy.

[0075] As shown in this embodiment, the preferred method is to acquire N radio wave intensities at different frequencies, where N is a natural number greater than or equal to 3, and the dimension reduction units 44 and 52 reduce the N radio wave intensities at different frequencies to input data that are smaller than N numbers.

[0076] Thus, by reducing dimensionality, it is possible to reproduce the original features represented by the intensity of radio waves at N frequencies and reduce the number of dimensions, thereby avoiding the reduction of computational costs and the curse of dimensionality (overlearning).

[0077] The program in this embodiment is a program that causes a computer (one or more processors) to execute the method. Furthermore, the program is stored on a computer-readable non-temporary storage medium in this embodiment.

[0078] The embodiments of this disclosure have been described above based on the drawings; however, it should be understood that the specific structure is not limited to these embodiments. The scope of this disclosure is defined by the claims and not merely by the description of the embodiments, and therefore includes all modifications within the meaning and scope equivalent to the claims.

[0079] The structures used in the above embodiments can be applied to other arbitrary embodiments.

[0080] The specific structure of each part is not limited to the described implementation method, and various modifications can be made without departing from the spirit of this disclosure.

Claims

1. A learning system for a precipitation estimation model, comprising: The radio wave intensity acquisition unit acquires the radio wave intensity of multiple frequencies in the radio waves received by the microwave radiometer at multiple time points during a specified period. The precipitable water acquisition unit acquires precipitable water calculated based on the atmospheric delay of the Global Navigation Satellite System (GNSS) signal received by the GNSS receiver at each of the plurality of time points during the specified period. as well as The learning department, based on the radio wave intensity of the multiple frequencies corresponding to the multiple time points within the specified period and the precipitable water amount, uses input data based on the radio wave intensity of the multiple frequencies at each of the multiple time points as input and outputs the corresponding precipitable water amount to enable the estimation model to perform machine learning. The estimation model, after undergoing machine learning, is used to obtain the radio wave intensity of multiple frequencies in the radio waves received by the microwave radiometer at a specific time point as input data and output the corresponding estimated precipitation.

2. The learning system for the precipitable water estimation model according to claim 1, comprising: The dimension reduction unit performs dimension reduction processing on the radio wave intensity of the plurality of frequencies at each of the plurality of time points, and calculates the dimension-reduced input data representing the radio wave intensity of the plurality of frequencies at each of the plurality of time points.

3. The learning system for the precipitable water estimation model according to claim 2, wherein... The dimensionality reduction unit performs principal component analysis on the radio wave intensity of the multiple frequencies at each of the multiple time points, and selects a predetermined number of principal components after the first order from the multiple principal components corresponding to the radio wave intensity of the multiple frequencies at each of the multiple time points as the input data.

4. The learning system for the precipitable water estimation model according to claim 2, comprising: The standardization processing unit performs standardization processing on the radio wave intensity of the plurality of frequencies at each of the plurality of time points before performing the dimensionality reduction processing using the dimensionality reduction unit.

5. The learning system for the precipitation estimation model according to claim 2, wherein... The radio wave intensity acquisition unit acquires the radio wave intensity of N different frequencies at each of the plurality of time points, where N is a natural number greater than or equal to 3. The dimensionality reduction unit reduces the radio wave intensity of the N frequencies at each of the plurality of time points to input data that is smaller than N.

6. A system for estimating precipitable water, comprising: The radio wave intensity acquisition unit acquires the radio wave intensity of multiple frequencies in the radio waves received by the microwave radiometer at a specific time point; as well as The estimation unit uses an estimation model to receive the radio wave intensity of the plurality of frequencies acquired at the specific time point as input data and outputs an estimated precipitable amount of water. The estimation model is based on the radio wave intensity of the plurality of frequencies corresponding to the plurality of time points in a specified period and the precipitable amount of water, and is machine learning performed in a manner that takes the input data of the radio wave intensity of the plurality of frequencies at each of the plurality of time points as input and outputs the corresponding precipitable amount of water.

7. The precipitation estimation system according to claim 6, comprising: The dimension reduction unit performs dimension reduction processing on the radio wave intensity of the plurality of frequencies and calculates the dimension-reduced input data representing the radio wave intensity of the plurality of frequencies at the specific time point.

8. The precipitation estimation system according to claim 7, wherein... The dimensionality reduction unit performs principal component analysis on the radio wave intensities of the multiple frequencies, and selects a predetermined number of principal components from the multiple principal components corresponding to the radio wave intensities of the multiple frequencies as the input data.

9. The precipitation estimation system according to claim 7, comprising: The standardization processing unit performs standardization processing on the radio wave intensity of the multiple frequencies at the specific time point using pre-set standardization parameters before performing the dimensionality reduction processing using the dimensionality reduction unit.

10. The precipitation estimation system according to claim 7, wherein... The radio wave intensity acquisition unit acquires the radio wave intensity of N different frequencies at the specific time point, where N is a natural number greater than or equal to 3. The dimensionality reduction unit reduces the radio wave intensity of the N frequencies at the specific time point to input data that is smaller than N.

11. A learning method for a predictable precipitation model, comprising: At multiple points in time during a specified period, the intensity of radio waves at multiple frequencies received by the microwave radiometer is acquired. At each of the plurality of time points during the specified period, the precipitable water amount is obtained based on the atmospheric delay of the Global Navigation Satellite System (GNSS) signal received by the GNSS receiver. as well as Based on the radio wave intensity of the multiple frequencies corresponding to the multiple time points within the specified period and the precipitable water amount, the estimation model performs machine learning by taking the input data of the radio wave intensity of the multiple frequencies at each of the multiple time points as input and outputting the corresponding precipitable water amount respectively. The estimation model, after undergoing machine learning, is used to obtain the radio wave intensity of multiple frequencies in the radio waves received by the microwave radiometer at a specific time point as input data and output the corresponding estimated precipitation.

12. A method for estimating precipitable water, comprising: To obtain the intensity of multiple frequencies of radio waves received by a microwave radiometer at a specific time point; as well as An estimation model is used to receive the radio wave intensity of multiple frequencies acquired at a specific time point as input data and output an estimated precipitation amount. The estimation model is based on the radio wave intensity and precipitation amount corresponding to multiple times within a specified period, and is performed by machine learning in a manner that takes the input data based on the radio wave intensity of multiple frequencies at each of the multiple times as input and outputs the corresponding precipitation amount. The estimation model, after undergoing machine learning, is used to obtain the radio wave intensity of multiple frequencies in the radio waves received by the microwave radiometer at a specific time point as input data and output the corresponding estimated precipitation.

13. A computer-readable storage medium storing a program that causes one or more processors to execute the learning method of the precipitation estimation model as claimed in claim 11 or the precipitation estimation method as claimed in claim 12.

Images