Early hail cloud recognition method based on polarization radar

Abstract

The invention discloses a hail cloud early recognition method based on a polarization radar. The method comprises the following steps: establishing a sample library; extracting a plurality of decision factors of hail cloud monomer radar data 20 minutes before hail occurs in the sample library, establishing a logistic regression model for the plurality of decision factors in the sample library, selecting an optimal weight algorithm by using a cross validation mechanism, calculating weights of the plurality of decision factors, and determining a hail early warning index threshold; the early warning index is compared with the hail early warning index threshold value, if the early warning index is larger than or equal to the hail early warning index threshold value, it is judged that the hail cloud is an early-stage hail cloud monomer, and if not, it is judged that the hail cloud is a general thunderstorm cloud. Based on polarization radar data and three key judgment factors, the early recognition accuracy of hail clouds is effectively improved, the hail clouds and general thunderstorm clouds can be accurately distinguished, and weather forecast and manual hail suppression operation are assisted.

Description

TECHNICAL FIELD

[0001] The present application relates to the meteorological early identification technical field, and particularly relates to a hail cloud early identification method based on a polarization radar. BACKGROUND

[0002] Hail cloud and other thunderstorm clouds (thunderstorm clouds or weak convective clouds) all belong to strong convective weather. Compared with strong convective clouds, weak convective clouds are generally short in life cycle, small in cloud range, weak in radar echo intensity, and rarely form lightning and precipitation, and thus are easier to identify. However, hail clouds and thunderstorm clouds have many similarities in the development process, and it is difficult to distinguish them in the early stage. In the past, due to the limitation of detection means, it is difficult to understand the microphysical change process in the cloud in real time, and thus the problem of early identification of hail clouds has not been well solved. Nowadays, with the development and use of new detection instruments such as dual-polarization weather radar, it is possible to solve the problem of early identification of hail clouds.

[0003] If the hail cloud is prevented in the early stage of development, that is, when large hail has not yet been formed in the cloud, the environmental conditions for generating large hail are destroyed, and a hail prevention effect can be achieved with half the effort. Hail cloud will be accompanied by strong lightning and gale during the development process, and will form hail, local rainstorm and lightning and other disastrous weather. In addition, early identification of hail cloud can also gain more time for hail prevention and disaster reduction preparation. Therefore, a hail cloud early identification method based on a polarization radar is needed. SUMMARY

[0004] The purpose of the present application is to provide a hail cloud early identification method based on a polarization radar to solve the problems existing in the prior art.

[0005] To achieve the above purpose, the present application is implemented according to the following technical scheme:

[0006] On the one hand, the present application comprises the following steps:

[0007] Step 1: screening hail events verified by ground observation based on historical polarization radar data to establish a sample library;

[0008] Step 2: extracting a plurality of determination factors of hail cloud monomer radar data 20 minutes before hail occurrence in the sample library, and performing standardization processing on the determination factors; the determination factors include a vertical integral liquid water VIL factor FVIL, a differential reflectivity column height factor F ZDRc , and a 45dBZ reflectivity echo top height and 0℃ layer height difference factor F ΔH45 .

[0009] Step three, a logistic regression model is established for a plurality of the decision factors in the sample library, the best weight algorithm is selected by using a cross-validation mechanism, the weight of a plurality of the decision factors is calculated, and a hail warning index threshold is determined;

[0010] Step four, the decision factors are extracted from the strong convective cloud detection data and are standardized, the processed decision factors are brought into the logistic regression model to calculate a warning index, the warning index is compared with the hail warning index threshold, if the warning index is greater than or equal to the hail warning index threshold, it is determined that it is an early hail cloud monomer, otherwise it is determined that it is a general thunderstorm cloud;

[0011] The logistic regression model calculates a linear weighted sum F of each decision factor, and then brings F into a sigmoid function to obtain a warning index P 冰雹 ; the calculation formula of the linear weighted sum F 冰雹 is as follows:

[0012] F 冰雹 = W VIL × F VIL + W ZDRc × F ZDRc + W ΔH45 × F ΔH45

[0013] In the formula, W VIL is the weight of F VIL , W ZDRc is the weight of F ZDRc , and W ΔH45 is the weight of F ΔH4 , and the minimum value of F is 0, and the maximum value is dynamically changed by data input; the calculation formula of the sigmoid function is as follows:

[0014]

[0015] In the formula, F 冰雹 is a linear weighted sum of decision factors, and e is a natural constant.

[0016] Further, 30 dBZ is used as a recognition threshold to recognize a strong return monomer, monomer information is confirmed, the maximum value of vertical integral liquid water VIL in the monomer is calculated by using the monomer information and VIL grid data, and a VIL factor FVIL is finally determined, and the calculation formula of the vertical integral liquid water VIL factor FVIL is as follows:

[0017]

[0018] In the formula, VIL max is the maximum value of VIL in a hail cloud monomer, VIL max L is the minimum VIL maximum value of a hail case sample library, and VILmax H is the maximum VIL of the hail case sample library.

[0019] Further, the calculation formula of the differential reflectivity column height factor FZDRc is:

[0020]

[0021] H ZDRc is the maximum height of the differential reflectivity column within the hail cloud monomer that vertically extends above the 0℃ layer and has ZDR≥1dB, H ZDRc L is the minimum differential reflectivity column height of the hail cases in the sample library, H ZDRc H is the maximum differential reflectivity column height of the hail cases in the sample library.

[0022] Further, the calculation formula of the 45dBZ reflectivity echo top height and 0℃ layer height difference factor FΔH45 is:

[0023]

[0024] H ΔH45 is the difference value of the 45dBZ reflectivity echo top height and the 0℃ layer height within the hail cloud monomer, H ΔH45 L is the minimum 45dBZ reflectivity height difference of the hail cases in the sample library, H ΔH45 H is the maximum 45dBZ reflectivity height difference of the hail cases in the sample library.

[0025] Further, the standardization processing is min-max standardization; and before the standardization processing, the step of removing extreme abnormal values in the determination factor by using 3σ method is further included.

[0026] Further, the polarimetric radar data needs to be gridded; the gridding processing includes: (1) according to the longitude, latitude, height and radar position of the grid point in the Cartesian coordinate system, combined with the elevation, azimuth and slant range of the radar echo data in the polar coordinate system, the adjacent point assignment method is used to assign values to the grid points, and the clutter is filtered in the assignment process, when the number of valid data in 8 directions around a reflectivity data is ≤3, it is determined as clutter and removed;

[0027] (2) using the eight-point Barnes interpolation algorithm to process the data of the adjacent elevation layer in the polar coordinate system.

[0028] Further, in step three, the best weight algorithm uses one of L1 regularization logistic regression algorithm, genetic algorithm and grid algorithm, and the algorithm with the highest AUC score is selected as the best weight algorithm through the AUC score of the cross-validation mechanism.

[0029] Further, in step three, the determination method of the hail early warning index threshold G hail includes: extracting the early warning index calculated by the 20-minute determination factor before the hail occurs in the sample library, and extracting the early warning index increment calculated by the 30-minute and 20-minute determination factors before the hail occurs, and combining the early warning index and the early warning index increment to divide the threshold.

[0030] Further, the early warning level division method corresponding to the hail early warning index threshold G hail includes: when the early warning index increment is >0.06 or the early warning index is >0.6, it is a red early warning, indicating that hail may occur in the next 20 minutes; when the early warning index increment is between 0.05-0.06 or the early warning index is between 0.5-0.6, it is an orange early warning; and the rest is a low risk level.

[0031] The beneficial effects of the present application are:

[0032] The present application is a hail cloud early identification method based on polarization radar, which has the following technical effects compared with the prior art:

[0033] The present application is based on polarization radar data and three key determination factors, which effectively improves the early identification accuracy of hail cloud, can accurately distinguish hail cloud from general thunderstorm cloud, and helps weather forecasting and artificial hail prevention operation. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 The figure is a schematic diagram of the adjustment system framework of the early identification method of the hail cloud based on the polarization radar of the present application;

[0035] Figure 2 The figure is a VIL time sequence change broken line of the early identification method of the hail cloud based on the polarization radar of the present application in the A area hail process on July 24, 2022;

[0036] Figure 3 The figure is a VIL time sequence change broken line of the early identification method of the hail cloud based on the polarization radar of the present application in the B area hail process on June 5, 2024;

[0037] Figure 4 The figure is a differential reflectivity radar echo profile of the early identification method of the hail cloud based on the polarization radar of the present application in the hail process of a certain place on August 9, 2022; DETAILED DESCRIPTION

[0038] The present application will be further described below through specific embodiments, and the illustrative embodiments of the present application and the description are used to explain the present application, but not as a limitation of the present application.

[0039] As Figure 1As shown, a hail cloud early identification method based on polarization radar includes the following steps:

[0040] The development of hail cloud needs sufficient water vapor supply. In order to use the three-dimensional space grid water vapor content information to represent two-dimensional data, the more vigorous the convection develops, the larger the VIL value is, and the large VIL value has a strong correlation with the ground hail. For example, Figure 2 As shown, the VIL time sequence change broken line of the A area hail process on July 24, 2022;

[0041] The VIL maximum value in the cell is calculated by using the cell information and the VIL grid data, and the VIL factor is finally determined. In this paper, the VIL factor is defined as F VIL , which is represented by using W VIL to determine the influence weight of the VIL factor in the hail cloud early identification system, and the maximum VIL value VIL max in the cell area is used as the influence parameter of the factor.

[0042]

[0043] VIL max L is the minimum VIL maximum value of the hail case sample library, VIL max H is the maximum VIL maximum value of the hail case sample library.

[0044] The difference reflectivity column height H ZDRc factor

[0045] The dual-polarization radar transmits horizontal and vertical polarized electromagnetic waves, and in addition to obtaining the reflectivity factor (ZH), it can also obtain the differential reflectivity factor (ZDR), the differential propagation phase shift rate (KDP) and the correlation coefficient (CC) and other dual-polarization parameters. These parameters are closely related to the phase state, shape, spatial orientation and distribution of precipitation particles, and can be used to identify specific microphysical and dynamic processes. Among them, ZDR column as one of the most significant dual-polarization radar features in convective storms, its development and evolution has a close relationship with the dynamic and microphysical structure inside the convective storm, and has important indicative significance for the development and evolution of convective storm.

[0046] The core logic of ZDR column height as the basis for determining hail cloud is that its height directly reflects the strength of the strong updraft in the cloud, the vertical transport capacity of large water droplets, and whether it crosses the 0℃ layer into the freezing zone. Higher ZDR column means that the cloud has sufficient liquid water supply, strong updraft support and freezing conditions required for hail formation, so it becomes one of the important indicators for radar monitoring hail cloud. In actual business, meteorological personnel can quickly judge whether the cloud cluster has hail potential by analyzing the relative position of ZDR column height and 0℃ layer, which provides a key basis for hail warning, such as Figure 3The difference reflectivity height-time series of a hail process on June 5, 2024 is shown as a broken line, Figure 4 The radar echo profile of a hail process on August 9, 2022 is shown.

[0047] Using the cell information and the difference reflectivity grid data, the columnar difference reflectivity factor enhancement zone vertically extending above the 0°C layer within the cell, the maximum height of ZDR≥1dB, is calculated to determine the final ZDRc factor F ZDRc :

[0048]

[0049] H ZDRc L is the minimum difference reflectivity column height of the hail case sample library, H ZDRc H is the maximum difference reflectivity column height of the hail case sample library.

[0050] 45dBZ reflectivity height difference factor

[0051] The formation of hail needs to meet two core conditions: sufficient liquid water is transported to the freezing zone and strong updraft supports particle growth. Height difference directly quantifies these two conditions by combining radar observations with atmospheric thermodynamic parameters: 45dBZ is a key threshold of radar reflectivity factor, which usually corresponds to strong precipitation particles (such as large water droplets, hail embryos or small hail), and its existence indicates that there are enough large particles in the cloud, which is the "material basis" for hail formation. The 0°C layer is the boundary between liquid water (warm area) and freezing zone (cold area) in the atmosphere, and particles can only freeze and grow into hail above the 0°C layer.

[0052] When the 45dBZ echo top height exceeds the 0°C layer height (i.e. the height difference is positive), it means that the large particles (liquid water droplets or embryos) corresponding to the strong reflectivity are transported to the freezing zone by the updraft, which has the condition of freezing growth; if the height difference is negative (the 45dBZ echo top height is lower than the 0°C layer), the particles cannot enter the freezing zone and can only form rain or small sleet, which is difficult to develop into hail. Through a large amount of observation data statistics, meteorologists can determine the critical height difference threshold in different regions and different seasons, and when the actual height difference exceeds the threshold, the probability of hail occurrence significantly increases, providing a clear quantitative index for prediction.

[0053] Using the cell information and the reflectivity grid data, the 45dBZ reflectivity height and the 0°C layer height difference are calculated to finally determine the 45dBZ reflectivity height difference factor F ΔH45 .

[0054]

[0055] H ΔH45L is the minimum 45dBZ reflectivity height difference of hail case sample library, H ΔH45 H is the maximum 45dBZ reflectivity height difference of hail case sample library.

[0056] After the function value of each sensitive factor is calculated, the linear weighted sum F of each decision factor is calculated 冰雹 , the early warning index P is calculated by bringing the sigmoid function 冰雹 , P 冰雹 is compared with the hail cloud early warning index threshold G 冰雹 , the cell greater than the threshold is determined as early hail cloud, otherwise as general thunderstorm cloud;

[0057] F 冰雹 = W VIL × F VIL + W ZDRc × F ZDRc + W ΔH45 × F ΔH45

[0058] F 冰雹 The maximum value of F is dynamically changed by data input, and the minimum value is 0

[0059]

[0060] If P 冰雹 ≥ G 冰雹 , it is determined as hail cloud, otherwise as other thunderstorm cloud.

[0061] Radar data gridding processing

[0062] The gridding processing of radar data is as follows: firstly, the detection value of the grid point is obtained by assigning the adjacent point to the grid point according to the longitude, latitude and height of the grid point in the Cartesian coordinate system, the elevation, azimuth and slant range of the radar echo data in the polar coordinate system, and the adjacent point is assigned to the grid point in the gridding processing process. In the assignment process, the de-noising operation is performed on the base data. That is, when there are 3 or less data around the 8 azimuths of a reflectivity data, it is considered to be single noise and is filtered out, so as to ensure the continuity of the data image. In order to ensure the continuity and integrity of the data, the eight-point Barnes interpolation algorithm is used by using the adjacent elevation layer data in the polar coordinate system.

[0063] Extraction of hail cloud early warning index threshold

[0064] Based on historical polarization radar data, hail events verified by ground observations are screened to establish a sample library; the determination factors of hail cloud monomer radar data 20 minutes before hail occurrence are extracted and min-max standardized; a logistic regression model is established for each determination factor in the sample library, the best weight algorithm is selected by cross-validation mechanism, the weight of each determination factor is calculated, and the threshold of hail warning index is determined;

[0065] Early hail cloud recognition effect verification and analysis

[0066] In this embodiment, the hail data sets of dual-polarization radar from 2022 to 2024 in a certain city and the hail data sets of dual-polarization radar from 2018 to 2023 in other regions are used to select appropriate hail cases, and the key time points of each case are labeled, including the first hail time, 30 minutes before hail, and 20 minutes before hail. The hail cloud radar detection data at the key time points are extracted to determine the factors, including the maximum value of vertical integrated liquid water VIL, the column height of differential reflectivity, and the 45dBZ threshold echo top height. The 0℃ height is determined by using sounding data, and finally the difference between the 45dBZ top height and the 0℃ height is determined.

[0067] The maximum value of vertical integrated liquid water VIL, the column height of differential reflectivity, and the difference in 45dBZ echo top height 30 minutes and 20 minutes before hail are subjected to min-max standardization, and then the extreme abnormal values are removed by 3σ method to make a training sample library for hail cloud determination logistic regression model.

[0068] The standardization formula in this example is as follows (based on the last 20 minutes of data):

[0069] F VIL =(VIL max –6.48) / (45.42-6.48)

[0070] F ZDRc =(H ZDRc -0.00) / (11.68-0.00)

[0071] F ΔH45 =(H ΔH45 --1.47) / (6.82--1.47)

[0072] The logistic regression coefficient method is used to calculate the weight of the determination factors extracted from the sample library 20 minutes before hail, the best weight optimization algorithm is selected according to the cross-validation mechanism AUC score, and L1 regularization logistic regression algorithm, genetic algorithm, and grid algorithm are used in this example. The determination factor weights calculated by the three algorithms and their AUC scores are as follows:

[0073] L1 regularization logistic regression algorithm (AUC: 0.7882)

[0074] WVIL : 0.1153; W ZDRC : 0.2383; W ΔH45 : 0.6464

[0075] Genetic algorithm (AUC: 0.8118)

[0076] W VIL : 0.1947; W ZDRC : 0.1888; W ΔH45 : 0.6164

[0077] Grid algorithm (AUC: 0.8750)

[0078] W VIL : 0.3401; W ZDRC : 0.0993; W ΔH45 : 0.5607

[0079] Therefore, the grid algorithm with the highest AUC score is used to determine the weights of the decision factors, and the following logistic regression model is obtained:

[0080] F 冰雹 = 0.3401 x F VIL + 0.0993 x F ZDRc + 0.5607 x F ΔH45

[0081] The weighted sum of the decision factors obtained by the logistic regression model is brought into the sigmoid function to convert it into a hail warning index:

[0082]

[0083] The G 冰雹 threshold value is extracted from the hail data set sample instance. In this example, the hail warning index threshold value is extracted from the decision factors 20 minutes before the hail falls, and the hail warning increment index threshold value is extracted from the difference between the decision factors 30 minutes and 20 minutes before the hail falls: when the index increment is > 0.06 or the index is > 0.6, it is a red warning, and hail may occur in the next 20 minutes; the index increment is between 0.05 and 0.06 or the index is between 0.5 and 0.6, it is an orange warning.

[0084] As Figures 2-4 shown, 16 hail clouds occurring in different regions are selected in addition to the training samples, and the evaluation factors 30 minutes and 20 minutes before the hail falls or the maximum echo intensity of the thunderstorm cloud are extracted, the warning index is calculated, and the following table is obtained. 15 out of 16 hail instances are correctly identified as red warnings, i.e. hail will occur in the next 20 minutes, and 1 hail instance is falsely reported as an orange warning, with a hail recognition accuracy of 93.75%.

[0085] Table 1 Example Model Test Results

[0086]

[0087]

[0088] The above description is only the preferred embodiment of the application, and is not used to limit the application, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application should be included in the protection scope of the application.

Claims

1. A method for early identification of hail clouds based on polarization radar, characterized in that: The following steps are involved: Step 1: Screen hail events verified by ground observations based on historical polarimetric radar data and establish a sample library; Step 2: extract multiple determination factors of hail cloud monomer radar data 20 minutes before the hail occurs in the sample library, and standardize the determination factors; the determination factors include vertically integrated liquid water VIL factor FVIL, differential reflectivity column height factor F ZDRc , 45dBZ reflectivity echo top height and 0℃ layer height difference factor F ΔH45 ; Step 3: Establish a logistic regression model for the multiple determination factors in the sample library, select the best weight algorithm using a cross-validation mechanism, calculate the weights of the multiple determination factors, and determine the hail warning index threshold; Step 4: extracting the determination factors from the severe convective cloud detection data and performing normalization processing, bringing the processed determination factors into the logistic regression model to calculate the warning index, and comparing the warning index with the hail warning index threshold. If the warning index is greater than or equal to the hail warning index threshold, it is determined to be an early hail cloud cell; otherwise, it is determined to be a general thunderstorm cloud; The logistic regression model calculates the linear weighted sum F of each decision factor, and then brings F into the sigmoid function to obtain the warning index P 冰雹 ; The linear weighted sum F 冰雹 The calculation formula is: F 冰雹 =W VIL ×F VIL +W ZDRc ×F ZDRc +W ΔH45 ×F ΔH45 Where W VIL F VIL The weight, W ZDRc F ZDRc The weight, W ΔH45 F ΔH4 The weight of Fhail is 0, and the maximum value changes dynamically according to the amount of data input. The calculation formula of the sigmoid function is: Among them F 冰雹 is the linear weighted sum of the determination factors, and e is a natural constant.

2. The method for early identification of hail clouds based on polarization radar according to claim 1, characterized in that: Use 30dBZ as the identification threshold to identify strong echo monomers, confirm monomer information, use monomer information and VIL grid data to calculate the maximum vertical integrated liquid water VIL value in the monomer, and finally determine the VIL factor FVIL. The calculation formula of the vertical integrated liquid water VIL factor FVIL is: Where VIL max is the maximum value of VIL in a hail cloud cell, VIL max L is the minimum VIL value of the hail case sample library, VIL max H is the maximum VIL value of the hail case sample library.

3. The method for early identification of hail clouds based on polarization radar according to claim 1, characterized in that: The calculation formula of the differential reflectivity column height factor FZDRc is: Where H ZDRc H is the maximum height of the differential reflectivity column that extends vertically above the 0°C layer in the hail cloud cell and has a ZDR ≥ 1dB. ZDRc L is the minimum differential reflectivity column height of the hailstone case in the sample library, H ZDRc H is the maximum differential reflectivity column height of the hailstone case in the sample library.

4. The method for early identification of hail clouds based on polarization radar according to claim 1, characterized in that: The calculation formula of the height difference factor FΔH45 between the 45dBZ reflectivity echo top height and the 0°C layer height is: Where H ΔH45 H is the difference between the top height of the 45dBZ reflectivity echo and the 0℃ layer height in the hail cloud cell. ΔH45 L is the minimum 45dBZ reflectivity height difference of hailstones in the sample library, G ΔH45 H is the maximum 45dBZ reflectivity height difference of the hailstone cases in the sample library.

5. The method for early identification of hail clouds based on polarization radar according to claim 1, characterized in that: The standardization process is min-max standardization; and before the standardization process, it also includes a step of using the 3σ method to eliminate extreme outliers in the determination factor.

6. The method for early identification of hail clouds based on polarization radar according to claim 1, characterized in that: The polarization radar data needs to be gridded; the gridding process includes: (1) Based on the longitude, latitude, altitude and radar position of the grid point in the Cartesian coordinate system, combined with the elevation angle, azimuth and slant range of the radar echo data in the polar coordinate system, the grid point is assigned a value using the neighboring point assignment method. During the assignment process, clutter is filtered out. When the number of valid data in the eight azimuths around a reflectivity data is ≤3, it is determined to be clutter and removed; (2) Using the adjacent elevation layer data in the polar coordinate system, the eight-point Barnes interpolation algorithm is used for processing.

7. The method for early identification of hail clouds based on polarization radar according to claim 1, characterized in that: In step 3, the optimal weight algorithm adopts one of the L1 regularized logistic regression algorithm, the genetic algorithm, and the grid algorithm, and the algorithm with the highest AUC score is selected as the optimal weight algorithm through the AUC score of the cross-validation mechanism.

8. The method for early identification of hail clouds based on polarization radar according to claim 1, characterized in that: In step three, the hail warning index threshold G hail is determined by extracting the warning index calculated by the determination factor 20 minutes before the hail occurs in the sample library, and at the same time extracting the warning index increment calculated by the determination factor 30 minutes and 20 minutes before the hail occurs, and dividing the threshold by combining the warning index and the warning index increment.

9. The method for early identification of hail clouds based on polarization radar according to claim 8, characterized in that: The hail warning index threshold G hail warning level classification method includes: when the warning index increment > 0.06 or the warning index > 0.6, it is a red warning, indicating that hail may occur in the next 20 minutes; when the warning index increment is between 0.05-0.06 or the warning index is between 0.5-0.6, it is an orange warning; the rest of the cases are low risk levels.

Images