KR20230068481A - System and method for diagonising the possibility of heavy rainfall and computer readable program for the same - Google Patents

Abstract

Disclosed are a device for diagnosing a possibility of heavy rainfall, a method thereof, and a computer-readable program for the same. The device for diagnosing a possibility of heavy rainfall diagnoses a possibility of heavy rainfall by using a rainfall material from one or more automatic ground observation devices and a geostationary satellite. The device for diagnosing a possibility of heavy rainfall includes: a rainfall material collecting unit collecting the rainfall material from the automatic ground observation device and the geostationary satellite; a preprocessing unit preprocessing the collected rainfall material; a rainfall diagnosis model generating unit selecting at least one or more preprocessed rainfall materials as training data and generating a rainfall diagnosis model diagnosing the possibility of heavy rainfall by each type of a preset cloud system based on the training data; and a diagnosis unit diagnosing the possibility of heavy rainfall based on a rainfall area and the amount of rainfall which are output by inputting the preprocessed rainfall material in the generated rainfall diagnosis model. Therefore, the device for diagnosing a possibility of heavy rainfall can accurately diagnose the possibility of rainfall by each type of cloud system generated on the Korean Peninsula by finding a pattern from the rainfall material collected by the automatic ground observation device and the geostationary satellite.

Description

Apparatus for diagnosing the possibility of heavy rain, method and computer readable program therefor

The present invention relates to an apparatus, method, and computer readable program for diagnosing the possibility of heavy rain, and more particularly, to diagnose the possibility of heavy rain through at least one automatic ground observation data and precipitation data from a geostationary satellite. , It relates to a device and method for diagnosing the possibility of heavy rain, and a computer readable program therefor.

In the Korean Peninsula, precipitation is formed by various cloud types, and strong precipitation occurs frequently in a short period of time in some cloud systems. Cloud systems that develop rapidly and cause strong precipitation are very difficult to predict, and heavy rains caused by these cloud systems cause enormous property and human damage.

Therefore, a study was conducted on the cloud system that causes heavy rainfall in the Korean Peninsula. Representatively, in the study of "Heavy precipitation systems over the Korean Penisula and their Classification. J. Korean Meteor, Soc., 43(4), 367-396, Lee and Kim (2007)", 7 years from 2000 to 2006 Radar and satellite images, weather charts, rainfall data, and thermodynamic chart data from 86 cases of heavy rain in South Korea were analyzed to classify the HPS throughout the Korean Peninsula and identify the characteristics of the precipitation system.

As a result, 4 major types of cloud systems were identified through the phenomenological analysis of HPS in 86 heavy rain cases. These four cloud systems include isolated thunderstorm (IS), convective band (CB), squall line (SL), and cloud cluster (CC). First, an isolated thunderstorm (IS) is the smallest cloud system among the four cloud systems, and its width generally ranges from several kilometers to several tens of kilometers. Also, the convection band (CB) and the squall line (SL) are similar to each other in that the convection system develops into a line and its width is much smaller than its length. A cloud community (CC) is an elliptical region with low cloud top temperature in satellite IR images, which may consist of several HPS or meso-β regions included in a continuous region of stratified precipitation. The scales of these cloud clusters (CCs) generally range from large meso-β to small meso-α scales, and are the most frequent cloud systems, accounting for about 47% of HPS from 2000 to 2006. Meanwhile, isolated thunderstorms (IS) and squall lines (SL) accounted for 12% and 7%, respectively.

As described above, morphological studies of cloud systems that cause heavy rain in the Korean Peninsula have been conducted, but there is a need for a method for diagnosing the possibility of heavy rain occurrence for various precipitation pictures occurring in the Korean Peninsula.

Domestic Patent No. 10-1986025

One aspect of the present invention is a Convolution Neural Network (CNN) that collects precipitation data from ground automatic observation equipment and geostationary orbit satellites, pre-processes them, and selects the pre-processed precipitation data as training data to diagnose the possibility of heavy rain by type of cloud system. )-based heavy rain diagnosis model, and based on this, it is an object of the present invention to provide a device, method, and computer readable program for diagnosing the possibility of heavy rain occurrence by discriminating precipitation amount and heavy rain area.

An apparatus for diagnosing the possibility of occurrence of heavy rain according to an embodiment of the present invention is an apparatus for diagnosing the possibility of occurrence of heavy rain through at least one automatic ground observation equipment and precipitation data from a geostationary orbit satellite, wherein the automatic ground observation equipment and the geostationary orbit A precipitation data collection unit that collects precipitation data from satellites, a preprocessor that preprocesses the collected precipitation data, selects at least one or more of the preprocessed precipitation data as training data, and selects at least one of the preprocessed precipitation data as training data. A heavy rain diagnosis model generating unit for generating a heavy rain diagnosis model for diagnosing the possibility of heavy rain occurrence by type and inputting the preprocessed precipitation data into the generated heavy rain diagnosis model to diagnose the possibility of heavy rain occurrence based on the output precipitation and heavy rain area includes diagnostics.

Meanwhile, the precipitation data collection unit includes multi-channel data including visible ray, near infrared ray, infrared ray, and water vapor channels from the geostationary satellite, cloud top height data, cloud top temperature data, and solar zenith angle. Geostationary orbit precipitation data including data may be collected, and AWS grid data interpolated by radial basis function interpolation as AWS (Automatic Weather Station) observation data from the ground automatic observation equipment may be collected.

In addition, the preprocessor may match and align the space-time of the geostationary-orbit precipitation data and the AWS grid data, and normalize the geostationary-orbit precipitation data based on the aligned geostationary-orbit precipitation data and the AWS grid data.

In addition, the heavy rain diagnosis model generation unit may classify the AWS grid data into at least one layer based on the amount of precipitation, and select the classified AWS grid data as training data by sampling differently for each layer.

In addition, the heavy rain diagnosis model may be a Convolution Neural Network (CNN) architecture composed of three 1D convolution layers.

In addition, the heavy rain diagnosis model generation unit may generate a heavy rain diagnosis model for predicting precipitation and classifying heavy rain areas for each type of mesoscale convective system (MCS) cloud system.

In addition, the diagnosis unit may diagnose the possibility of occurrence of heavy rain by assigning weights optimized for each type of the MCS cloud system to the output amount of precipitation.

A method for diagnosing the possibility of heavy rain occurrence according to another embodiment of the present invention is a method for diagnosing the possibility of heavy rain occurrence through at least one ground automatic observation equipment and precipitation data from a geostationary orbit satellite. As a method of diagnosing the possibility of occurrence, collecting precipitation data from the ground automatic observation equipment and geostationary satellites, pre-processing the collected precipitation data, selecting at least one or more of the pre-processed precipitation data as training data, , Generating a heavy rain diagnosis model for diagnosing the possibility of heavy rain occurrence for each type of preset cloud system based on the training data, and inputting the preprocessed precipitation data to the generated heavy rain diagnosis model to determine the output precipitation and heavy rain area and diagnosing the possibility of heavy rain on the basis of

On the other hand, in the step of collecting precipitation data from ground automatic observation equipment and geostationary satellites,

Geostationary orbit precipitation including multi-channel data including visible ray, near infrared ray, infrared ray, and water vapor channels, cloud top height data, cloud top temperature data, and solar zenith angle data from the geostationary satellite Data may be collected, and AWS grid data interpolated by radial basis function interpolation as AWS (Automatic Weather Station) observation data from the ground automatic observation equipment may be collected.

In addition, the generating of the heavy rain diagnosis model may include classifying the AWS grid data into at least one layer based on the amount of precipitation, and selecting the classified AWS grid data as training data by sampling differently for each layer.

In addition, the heavy rain diagnosis model is a CNN (Convolutinal Neural Network) architecture composed of three 1D convolutional layers, and the step of generating the heavy rain diagnosis model is to predict precipitation for each type of Mesoscale Convective System (MCS) cloud system. It may be to create a heavy rain diagnosis model that classifies the heavy rain area.

According to another embodiment of the present invention, it may be a computer readable program stored in a computer readable recording medium configured to execute a method for diagnosing a possibility of heavy rain.

According to one aspect of the present invention described above, it has the advantage of accurately diagnosing the possibility of heavy rain occurrence for each type of various cloud systems occurring on the Korean Peninsula by finding a pattern from precipitation data collected from ground automatic observation equipment and geostationary satellites. .

1 is a diagram illustrating a system for diagnosing a possibility of heavy rain occurrence according to an embodiment of the present invention.
FIG. 2 is a diagram showing an observation analysis field area of a geostationary satellite and ground automatic observation equipment shown in FIG. 1;
FIG. 3 is a diagram showing an example of precipitation data collected by the precipitation data collection unit shown in FIG. 1 .
FIG. 4 is a diagram showing an example of precipitation data collected by the precipitation data collection unit shown in FIG. 1 .
FIG. 5 is a diagram illustrating the architecture of a heavy rain diagnosis model generated in the heavy rain diagnosis model generator shown in FIG. 1 .
6 is a flowchart illustrating a method for diagnosing a possibility of heavy rain occurrence according to another embodiment of the present invention.
7 is a diagram showing the result of diagnosing the possibility of occurrence of heavy rain according to the present invention in an SLP precipitation case.
8 is a diagram showing the result of diagnosing the possibility of occurrence of heavy rain according to the present invention in an SLD precipitation case.
9 is a diagram showing the result of diagnosing the possibility of occurrence of heavy rain according to the present invention in an MCC precipitation case.
10 is a diagram showing the result of diagnosing the possibility of occurrence of heavy rain according to the present invention in a CC precipitation case.
11 is a diagram showing the result of diagnosing the possibility of occurrence of heavy rain according to the present invention in a CC precipitation case.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The detailed description of the present invention which follows refers to the accompanying drawings which illustrate, by way of illustration, specific embodiments in which the present invention may be practiced. These embodiments are described in sufficient detail to enable one skilled in the art to practice the present invention. It should be understood that the various embodiments of the present invention are different from each other but are not necessarily mutually exclusive. For example, specific shapes, structures, and characteristics described herein may be implemented in another embodiment without departing from the spirit and scope of the invention in connection with one embodiment. Additionally, it should be understood that the location or arrangement of individual components within each disclosed embodiment may be changed without departing from the spirit and scope of the invention. Accordingly, the detailed description set forth below is not to be taken in a limiting sense, and the scope of the present invention, if properly described, is limited only by the appended claims, along with all equivalents as claimed by those claims. Like reference numbers in the drawings indicate the same or similar function throughout the various aspects.

Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the drawings.

1 is a diagram illustrating a system for diagnosing a possibility of heavy rain occurrence according to an embodiment of the present invention.

Referring to FIG. 1 , the system for diagnosing the possibility of occurrence of heavy rain according to the present embodiment includes a geostationary satellite system 10, an automatic ground observation equipment system 20, and a device 100 for diagnosing the possibility of occurrence of heavy rain.

The geostationary satellite system 10 is a system capable of obtaining precipitation data in a wide range such as a full disk area or East Asia. Meanwhile, the geostationary satellite system 10 according to this embodiment may be a GK2A (Geostationary Korea multi purpose satellite 2A, Geo-Kompsat 2A) satellite. The region of the observation and analysis field of the geostationary satellite system 10 may be a KO domain and may be a region of Northeast Asia including the Korean Peninsula. More specifically, as shown in FIG. 2, the geostationary satellite system 10 according to the present embodiment has an observation analysis field region with a black border, preferably, the latitude range is 29 to 46 °N, The hardness range can be from 114 to 138 °E.

The automatic ground observation equipment system 20 can observe meteorological data as Automatic Weather Stations (AWS). There are about 713 such ground automatic observation equipment on the ground as detailed observation data for each point, and they observe precipitation, wind, temperature, relative humidity, atmospheric pressure, and presence or absence of precipitation. Precipitation-related data include 15-minute precipitation accumulation, 1-hour precipitation accumulation, 12-hour precipitation accumulation, and 24-hour precipitation accumulation data, which are calculated in 1-minute cycles. In order to compare the observed precipitation values by these points with radar or satellite data, interpolation can be used to grid them. In the ground automatic observation system 20, 10-minute precipitation intensity is averaged by 10 1-minute difference values of 24-hour accumulated precipitation data, and these values can be produced by interpolating to a 1km spatial resolution grid.

On the other hand, as shown in FIG. 2, the automatic ground observation equipment system 20 according to this embodiment may have a relatively smaller analysis field area than a geostationary orbit satellite system as an analysis field area corresponding to a small box area. there is.

The apparatus 100 for diagnosing the possibility of occurrence of heavy rain according to the present embodiment is a device for diagnosing the possibility of occurrence of heavy rain through at least one automatic ground observation equipment and precipitation data from a geostationary orbit satellite. To this end, a precipitation data collection unit ( 110), a pre-processing unit 120, a heavy rain diagnosis model generating unit 130, and a diagnosis unit 140.

The precipitation data collection unit 110 collects precipitation data from ground automatic observation equipment and geostationary satellites.

The precipitation data collection unit 110 includes multi-channel data including visible ray, near infrared ray, infrared ray, and water vapor channels from geostationary satellites, cloud top height data, cloud top temperature data, and sun Geostationary orbital precipitation data including zenith angle data can be collected. The precipitation data collection unit 110 according to the present embodiment includes multi-channel data including Level 1b visible ray, near infrared ray, infrared ray, and water vapor channels provided by the GK2A satellite, and Level 2 cloud top height and cloud top temperature. (Cloud top temperature) data and geostationary precipitation data including solar zenith angle data to reflect the temporal change effect of satellite data can be collected. Such geostationary precipitation data may preferably have a spatial resolution of 1 km, a temporal resolution of 10 minutes, and an analysis area of 1800×1800 ㎢.

In addition, the precipitation data collection unit 110 collects AWS grid data grid-interpolated by radial basis function interpolation as AWS (Automatic Weather Station) observation data from ground automatic observation equipment. These AWS grid data may preferably have a spatial resolution of 13 km and a temporal resolution of 1 minute, and the average rainfall intensity calculated based on the observed data for 10 minutes is interpolated into a grid of 480 × 600 ㎢ observation area It can be an AWS grid data.

3 is a diagram showing an example of Level 1B (July 31, 2019 09:00 KST) precipitation data collected by the precipitation data collection unit. Referring to FIG. 3, in the precipitation data of Level 1B (July 31, 2019 09:00 KST), VIS (0.6 μm ) data as shown in (a) of FIG. 3 and SW (3.8 μm) as shown in (b) of FIG. ㎛ ) data, as shown in (c) of Figure 3, WV (6.9 ㎛ ) data, as shown in Figure 3 (d) IR (10.5 ㎛ ) data, as shown in (e) of Figure 3 IR (11.2 ㎛ ) data and IR (13.3 ㎛ ) data as shown in (f) of FIG.

4 is a diagram showing Level 2 precipitation data, solar zenith angle, and AWS rainfall intensity data among precipitation data collected by the precipitation data collection unit. Referring to FIG. 4, precipitation data include cloud top height data as shown in FIG. 4 (a), cloud top temperature data as shown in FIG. 4 (b), solar zenith angle data and solar zenith angle data as shown in FIG. , AWS rainfall intensity data may be further included as shown in (d) of FIG.

The pre-processing unit 120 pre-processes the precipitation data collected by the precipitation data collection unit 110 .

More specifically, the pre-processing unit 120 matches and aligns the space-time of the geostationary-orbit precipitation data and the AWS grid data, and obtains the geostationary-orbit precipitation data based on the aligned geostationary-orbit precipitation data and the AWS grid data. normalize At this time, the pre-processing unit 120 may set geostationary precipitation data as an independent variable and arrange the AWS grid data acquired at the same point as a dependent variable. Here, the geostationary precipitation data, which are independent variables, are 19 GK2A variables, and may include 16 bands including visible, infrared, and shortwave infrared wavelengths, cloud top altitude, cloud top temperature, and solar zenith angle. These 19 GK2A Variables can be normalized.

The heavy rain diagnosis model generation unit 130 selects at least one or more of the precipitation data preprocessed by the preprocessing unit 120 as training data, and diagnoses the possibility of occurrence of heavy rain for each type of preset cloud system based on the training data. Generate a diagnostic model.

Meanwhile, the heavy rain phenomenon for modeling in this embodiment generally occurs in a limited area, so most areas of the domain have a low AWS value. For example, most of the AWS grid data has an AWS value of less than 10 mmh ^-1 , and in some cases, 80% of all pixels were concentrated in a section with less than 5 mmh ^-1 . Therefore, if a section of low precipitation is randomly sampled (sampled) for AWS grid data having a skewed distribution, oversampling (oversampling) occurs only in the low precipitation range and undersampling (undersampling) occurs in the high precipitation range. ) has a problem. Therefore, the heavy rain diagnosis model generation unit 130 according to the present embodiment selects training data through stratified sampling in order to solve the problem of being biased toward low values of training data. More specifically, the heavy rain diagnosis model generation unit 130 classifies the AWS grid data into at least one layer based on the amount of precipitation, and selects the classified AWS grid data as training data by sampling differently for each layer.

For example, as shown in Table 1 below, the heavy rain diagnosis model generation unit 130 stratifies and classifies the precipitation of the AWS grid data into four ranges, samples the AWS grid data belonging to these ranges at different rates for each layer, and uses the data as training data. can be selected

Value Range [mm/h] Sampling Percentage < 10 20% 10 - 20 40% 20 - 30 60% > 30 80%

As described above, when selecting training data by applying stratified sampling, it is possible to limit the amount of training data having a low value of precipitation, whereas the training data applied to the training process for generating a heavy rain diagnosis model is It becomes possible to apply data with values of medium and high precipitation.

The heavy rain diagnosis model generator 130 generates a heavy rain diagnosis model by applying the selected training data to a machine learning algorithm. The heavy rain diagnosis model trained by applying these training data can predict precipitation and classify heavy rain areas for each type of mesoscale convective system (MCS) cloud system. For example, the heavy rain diagnosis model generation unit 130 may learn precipitation cases for each of the four types (MCS types) of cloud systems in which strong precipitation occurred in the Korean Peninsula during 2018 and 2019, and the learned precipitation cases are It may be Squall Line Diagonal (SLD), Squall Line Parallel (SLP), Mesoscale Convective Complex (MCC), or Cloud Cluster (CC).

Meanwhile, the heavy rain diagnosis model generated by the heavy rain diagnosis model generation unit 130 may be a Convolutinal Neural Network (CNN) architecture composed of three 1D convolutional layers, as shown in FIG. 5 . More specifically, the heavy rain diagnosis model according to the present embodiment is composed of three convolutional layers to which different numbers of filters of 64, 64, and 32 are applied, respectively, and the kernel size may be set to two. The features extracted through these convolution layers are converted into 512 nodes through flattening, and can be fully connected to an output layer composed of one node through a dense layer composed of 32 nodes. In addition, a Rectified Linear Unit (ReLU) activation function may be applied to each convolution layer and a dense layer, and a linear activation function may be applied to an output node.

The diagnosis unit 140 inputs the preprocessed precipitation data to the heavy rain diagnosis model generated by the heavy rain diagnosis model generation unit 130, outputs the amount of precipitation and the heavy rain area, and calculates the possibility of occurrence of heavy rain based on the output amount of precipitation and heavy rain area. Diagnose.

The diagnosis unit 140 diagnoses the possibility of heavy rain occurrence by assigning weights optimized for each type of the MCS cloud system to the output amount of precipitation.

6 is a flowchart illustrating a method for diagnosing a possibility of heavy rain occurrence according to another embodiment of the present invention.

Referring to FIG. 6 , the method for diagnosing the possibility of heavy rain occurrence according to the present embodiment is performed in an apparatus for diagnosing the possibility of heavy rain occurrence through precipitation data from at least one or more ground automatic observation equipment and a geostationary orbit satellite. As a method for diagnosing the possibility of heavy rain, collecting precipitation data from the ground automatic observation equipment and geostationary satellite (S10), pre-processing the collected precipitation data (S20), and at least one of the pre-processed precipitation data Selecting one or more as training data and generating a heavy rain diagnosis model for diagnosing the possibility of heavy rain occurrence for each type of preset cloud system based on the training data (S30) and the preprocessed precipitation in the created heavy rain diagnosis model. and diagnosing the possibility of heavy rain based on the amount of rainfall and the heavy rain area output by inputting the data (S40).

The operation of the method for diagnosing the possibility of heavy rain according to the embodiments described above may be at least partially implemented as a computer program and recorded on a computer-readable recording medium. A program for implementing an operation by a deep learning-based super-resolution image processing method according to embodiments is recorded and a computer-readable recording medium includes all types of recording devices in which data readable by a computer is stored. include Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage devices. In addition, computer-readable recording media may be distributed in computer systems connected through a network, and computer-readable codes may be stored and executed in a distributed manner. In addition, functional programs, codes, and code segments for implementing this embodiment can be easily understood by those skilled in the art to which this embodiment belongs.

Hereinafter, a performance evaluation method for a result of diagnosing the possibility of heavy rain occurrence according to the present invention will be described. First, performance evaluation was performed by applying actual precipitation cases in 2020 as verification data. On the other hand, heavy rain advisories and heavy rain special warnings were classified in the form of heavy rain indices given specific numbers, and performance evaluation was performed by comparing them with the classification results of the heavy rain critical index (CI) used by the National Meteorological Satellite Center. The reference value of the heavy rain critical index used by the National Meteorological Satellite Center is as follows.

CI < 30)(1) Special notice not expressed = 0 (0

CI < 30)

CI < 60)(2) Heavy rain advisory = 1 (30

CI < 60)

60)(3) Heavy rain alert = 2 (CI

60)

On the other hand, the precipitation intensity threshold values used for heavy rain advisories and heavy rain warnings are as follows.

(1) Special alert not expressed = 0 (0~10 mmh ^-1 precipitation section)

(2) Monitoring = 1 (10~20 mmh ^-1 precipitation section)

(3) Heavy rain warning = 2 (20~30 mmh ^-1 precipitation section)

(4) Heavy rain warning = 3 (30 mmh ^-1 or more precipitation section)

In addition, AWS grid data was used as the reference data used for performance evaluation, and the CI classification result provided by the National Meteorological Satellite Center using the actual heavy rain area classified in the AWS grid data and the verification data for the heavy rain diagnosis model according to the present invention The performance evaluation was performed on the ability to diagnose the possibility of heavy rain by MCS type by comparing the characteristics of the spatial distribution of the heavy rain area, which is the result of applying the output.

Hereinafter, performance evaluation results performed as described above for each precipitation case will be described with reference to FIGS. 7 to 11 .

7 is a diagram showing the result of diagnosing the possibility of occurrence of heavy rain according to the present invention in an SLP precipitation case. 7 shows the AWS heavy rain critical index (FIG. 7 (a)), the GK2A heavy rain critical index (FIG. 7 (b)), and heavy rain diagnosis for the SLP precipitation case (July 30, 2020 06:10 KST) The heavy rain critical index in the model (Fig. 7(c)), the rainfall intensity of AWS, which is the actual precipitation (Fig. 7(d)), and the distribution of rainfall intensity simulated in the heavy rain diagnosis model with a weight of 2 (Fig. 7 (e)) of is shown. According to FIG. 7, it can be confirmed that although the heavy rain area diagnosed by the heavy rain diagnosis model according to the present invention is over-simulated in some Chungcheong-do areas, it appears similar to the pattern of the heavy rain area of AWS, which is relatively actual precipitation compared to the GK2A heavy rain critical index. can

8 is a diagram showing the result of diagnosing the possibility of occurrence of heavy rain according to the present invention in an SLD precipitation case.

8 shows the AWS heavy rain critical index (Fig. 8 (a)), the GK2A heavy rain critical index (Fig. 8 (b)) and the heavy rain for the SLD precipitation case (August 2, 2020 15:30 KST) The heavy rain critical index in the diagnostic model (Fig. 8(c)), the rainfall intensity of AWS, which is the actual precipitation (Fig. 8(d)), and the distribution of rainfall intensity simulated in the heavy rain diagnostic model with a weight of 2 (Fig. 8(e)) is shown. According to FIG. 8, similar to FIG. 7, the heavy rain critical index of GK2A was extensively overestimated over a wide area from the inland to the west coast of the Korean Peninsula compared to the AWS pattern, which is the actual precipitation, but diagnosed in the heavy rain diagnosis model according to the present invention It can be seen that the heavy rain area almost coincides with the pattern of the heavy rain area of AWS, which is the actual precipitation.

9 is a diagram showing the result of diagnosing the possibility of heavy rain occurrence according to the present invention in an MCC precipitation case. 9 shows the AWS heavy rain critical index (FIG. 9 (a)), the GK2A heavy rain critical index (FIG. 9 (b)), and heavy rain for the MCC precipitation case (August 3, 2020, 18:00 KST) The heavy rain critical index in the diagnostic model (Fig. 9(c)), the rainfall intensity of AWS, which is the actual precipitation (Fig. 9(d)), and the distribution of rainfall intensity simulated in the heavy rain diagnostic model with a weight of 2 (Fig. 9 (e)) is shown. According to FIG. 9, it can be seen that the heavy rain critical index of GK2A was simulated more widely than the area where MCC, the actual amount of precipitation, occurred, and did not coincide with the area of more than the actual heavy rain warning. On the other hand, in the diagnosis result of the heavy rain diagnosis model according to the present invention, it can be confirmed that the location and pattern of the heavy rain alarm abnormality almost coincide with the heavy rain alarm area in AWS, which is the actual precipitation.

10 is a diagram showing the result of diagnosing the possibility of occurrence of heavy rain according to the present invention in a CC precipitation case. 10 shows the AWS heavy rain critical index (FIG. 10 (a)), the GK2A heavy rain critical index (FIG. 10 (b)), and heavy rain for CC precipitation cases (August 21, 2020 18:00 KST) The heavy rain critical index in the diagnostic model (Fig. 10(c)), the rainfall intensity of AWS, which is the actual precipitation (Fig. 10(d)), and the distribution of rainfall intensity simulated in the heavy rain diagnostic model with a weight of 5 (Fig. 10(e)) is shown. According to FIG. 10, the heavy rain critical index of GK2A does not match the area where the actual CC occurred, whereas the diagnosis result of the heavy rain diagnosis model according to the present invention shows that the heavy rain alarm abnormality is found in the localization pattern in the actual AWS heavy rain alarm You can see the similarities with the area.

11 is a diagram showing the result of diagnosing the possibility of occurrence of heavy rain according to the present invention in a CC precipitation case. 11 shows the AWS heavy rain critical index (FIG. 11 (a)), the GK2A heavy rain critical index (FIG. 11 (b)), and heavy rain for CC precipitation cases (August 3, 2019, 19:00 KST) The heavy rain critical index in the diagnostic model (Fig. 11(c)), the rainfall intensity of AWS, which is the actual precipitation (Fig. 11(d)), and the distribution of rainfall intensity simulated in the heavy rain diagnostic model with a weight of 2 (Fig. 11 (e)) is shown. According to FIG. 11 , it can be confirmed that the heavy rain area diagnosed by the heavy rain diagnosis model according to the present invention almost coincides with the pattern of the heavy rain area of AWS, which is the actual precipitation.

According to one aspect of the present invention described above, it has the advantage of accurately diagnosing the possibility of heavy rain occurrence for each type of various cloud systems occurring on the Korean Peninsula by finding a pattern from precipitation data collected from ground automatic observation equipment and geostationary satellites. .

Although the above has been described with reference to embodiments, it will be understood that those skilled in the art can variously modify and change the present invention without departing from the spirit and scope of the present invention described in the claims below. You will be able to.

100: device for diagnosing the possibility of heavy rain
110: precipitation data collection unit
120: pre-processing unit
130: heavy rain diagnosis model generation unit
140: diagnosis unit

Claims (12)

A device for diagnosing the possibility of heavy rain through precipitation data from at least one ground automatic observation equipment and geostationary satellite,
a precipitation data collection unit that collects precipitation data from the ground automatic observation equipment and the geostationary satellite;
a pre-processing unit pre-processing the collected precipitation data;
a heavy rain diagnosis model generating unit selecting at least one of the preprocessed precipitation data as training data and generating a heavy rain diagnosis model for diagnosing a possibility of occurrence of heavy rain for each type of preset cloud system based on the training data; and
and a diagnosis unit configured to input the preprocessed precipitation data into the generated heavy rain diagnosis model and diagnose the possibility of heavy rain occurrence based on the output amount and heavy rain area. According to claim 1,
The precipitation data collection unit,
Geostationary orbit precipitation including multi-channel data including visible ray, near infrared ray, infrared ray, and water vapor channels, cloud top height data, cloud top temperature data, and solar zenith angle data from the geostationary satellite collect data,
A device for diagnosing the possibility of heavy rain, which collects grid interpolated AWS grid data by radial basis function interpolation as AWS (Automatic Weather Station) observation data from the ground automatic observation equipment. According to claim 1,
The pre-processing unit,
Align the space-time of the geostationary precipitation data and the AWS grid data to match each other,
An apparatus for diagnosing the possibility of occurrence of heavy rain, which normalizes the geostationary-orbit precipitation data based on the aligned geostationary-orbit precipitation data and the AWS grid data. According to claim 1,
The heavy rain diagnosis model generating unit,
Classifying the AWS grid data into at least one layer based on precipitation,
An apparatus for diagnosing the possibility of heavy rain occurrence by sampling the classified AWS grid data differently for each layer and selecting them as training data. According to claim 4,
The heavy rain diagnosis model,
A device for diagnosing the possibility of heavy rain, which is a CNN (Convolution Neural Network) architecture consisting of three 1D convolutional layers. According to claim 1,
The heavy rain diagnosis model generating unit
Mesoscale Convective System (MCS) A device for diagnosing the possibility of heavy rain, which creates a heavy rain diagnosis model that predicts precipitation and classifies heavy rain areas by type of cloud system. According to claim 6,
The diagnosis unit,
An apparatus for diagnosing the possibility of heavy rain occurrence by assigning weights optimized for each type of the MCS cloud system to the output precipitation amount to diagnose the possibility of heavy rain occurrence. A method for diagnosing the possibility of heavy rain occurrence in an apparatus for diagnosing the possibility of heavy rain occurrence through at least one ground automatic observation equipment and precipitation data from geostationary satellites,
collecting precipitation data from the ground automatic observation equipment and the geostationary satellite;
pre-processing the collected precipitation data;
selecting at least one of the preprocessed precipitation data as training data, and generating a heavy rain diagnosis model for diagnosing a possibility of occurrence of heavy rain for each type of preset cloud system based on the training data; and
and inputting the preprocessed precipitation data into the generated heavy rain diagnosis model and diagnosing a possibility of heavy rain occurrence based on the output rainfall amount and heavy rain area. According to claim 8,
The step of collecting precipitation data from the ground automatic observation equipment and the geostationary orbit satellite,
Geostationary orbit precipitation including multi-channel data including visible ray, near infrared ray, infrared ray, and water vapor channels, cloud top height data, cloud top temperature data, and solar zenith angle data from the geostationary satellite collect data,
A method for diagnosing the possibility of heavy rain, which collects grid interpolated AWS grid data by radial basis function interpolation as AWS (Automatic Weather Station) observation data from the ground automatic observation equipment. According to claim 9,
The step of generating the heavy rain diagnosis model,
Classifying the AWS grid data into at least one layer based on precipitation,
A method for diagnosing the possibility of heavy rain, wherein the classified AWS grid data is sampled differently for each layer and selected as training data. According to claim 10,
The heavy rain diagnosis model is a CNN (Convolutinal Neural Network) architecture composed of three 1D convolutional layers,
The step of generating the heavy rain diagnosis model is to generate a heavy rain diagnosis model for predicting precipitation and classifying heavy rain areas for each type of mesoscale convective system (MCS) cloud system. How to diagnose the possibility of heavy rain. A computer readable program stored on a computer readable recording medium, configured to execute the method for diagnosing a possibility of occurrence of heavy rain according to claims 8 to 11.

Images