A Hail Warning Method and System Based on Lightning Jump Increase and Support Vector Machine

Abstract

The present invention discloses a hail warning method and system based on lightning jump and support vector machine. The method obtains the combined reflectivity factor, vertical integrated liquid water content and echo top height data through radar network data, identifies the strong convective echo area and calculates the lightning time variability and its standard deviation. The method combines the lightning jump condition to screen potential hail events and further extracts the maximum value of vertical integrated liquid water content (VIL). max , jump increment GVIL and density VILD as feature parameters, and input them into the support vector machine classification model trained based on historical samples for secondary judgment, which effectively reduces the false alarm rate of the traditional 2σ algorithm and improves the accuracy of hail warning.

Description

Technical Field

[0001] The present invention relates to the technical field of severe convective weather forecast and warning services, and in particular to a hail warning method and system based on lightning surge and support vector machine. Background Art

[0002] Hail is a common meteorological natural disaster, often accompanied by lightning activity. With the improvement of lightning detection capabilities, many scholars at home and abroad have confirmed that the frequency of lightning also increases suddenly before and after hailstorms.

[0003] Currently, the main methods used to calculate lightning jumps are the Gatlin algorithm and the σ algorithm. Gatlin proposed an algorithm based on lightning jumps to identify and warn severe storms, namely the Gatlin algorithm.

[0004] This algorithm, as a primary hail warning method based on lightning data, has achieved good results, but it still suffers from a high false alarm rate. However, dual-polarization radars have yet to achieve full operational coverage, and areas without dual-polarization radar data cannot utilize this method to improve hail warning effectiveness. Furthermore, dual-polarization radar hydrometeor phase identification results rely heavily on the quality of polarization parameters. During actual radar operation, polarization data is susceptible to various factors and can result in significant errors, requiring careful and comprehensive quality control.

[0005] In recent years, with the gradual improvement of support vector machines (SVMs), they have become a new computer learning method for handling highly nonlinear classification and regression problems. A novel small-sample learning method with a solid theoretical foundation, SVMs enable efficient "transductive inference" from training samples to prediction samples, greatly simplifying common classification and regression problems. They excel when handling high-dimensional data and small sample sizes.

[0006] In summary, based on lightning and radar data, in-depth research on hail warning methods that integrate lightning jump and support vector machines can improve the forecast and warning level of severe convective weather such as hail, which is of great significance for effectively enhancing disaster prevention and mitigation capabilities and reducing people's lives and property losses. Summary of the Invention

[0007] In order to effectively make up for the shortcomings of hail weather warning methods in existing weather forecast services and solve the problem of high false alarm rate of hail warnings in existing technologies, a hail warning method based on lightning surge and support vector machine is provided, which includes the following steps:

[0008] Based on the radar network mosaic data of the warning area, obtain the two-dimensional grid data of the radar combination reflectivity factor CR, vertical integrated liquid water content VIL and echo top height TOP;

[0009] Based on the two-dimensional grid point data of the radar combined reflectivity factor CR, grid point position information of the connected area of the strong convective echo is obtained;

[0010] Based on the lightning data in the warning area and the grid location information of the strong convective echo connection area, the lightning time variability D and the standard deviation of the lightning time variability D are calculated. ;

[0011] Based on the two-dimensional grid data of the vertically integrated liquid water content VIL and the echo top height TOP and the grid position information of the strong convection echo connected area, the maximum value VIL of the vertically integrated liquid water content VIL is obtained. max And the VIL max The jump amount GVIL and density VILD;

[0012] Based on the lightning time variation rate D and the standard deviation of the lightning time variation rate D Make lightning jump condition judgment;

[0013] For the times that meet the lightning surge conditions, the maximum value VIL of the vertically integrated liquid water content VIL is max And the VIL max The jump increment GVIL and density VILD are used to judge the hail warning conditions using the support vector machine to achieve hail warning.

[0014] Preferably, the grid point location information of the connected area of the strong convection echo is obtained by using a neighborhood method, and the steps include:

[0015] performing binarization image processing on the two-dimensional grid point data of the radar combined reflectivity factor CR;

[0016] The neighborhood method is used to identify the strong convective echo connected area of the data after the binary image processing, and the grid point position information of the strong convective echo connected area is obtained.

[0017] Preferably, the step of identifying the connected area of strong convective echoes using the neighborhood method includes:

[0018] With the current strong convective echo grid point as the center, the eight adjacent grid points (upper, lower, left, right, upper left, upper right, lower left, and lower right) are used as the neighborhood of the current grid point. Determine whether any grid point in the neighborhood is a strong convective echo grid point. The judgment conditions include:

[0019] If the grid point is a strong convective echo grid point, it belongs to the same connected area as the central grid point;

[0020] If the grid point is not a strong convective echo grid point, it is a non-connected grid point;

[0021] In this way, domain grid point judgment is performed on all grid points in the same connected area until all domain grid points in the connected area are non-connected grid points, and the judgment and identification of a connected area is completed.

[0022] Preferably, the lightning time variation rate D and the standard deviation of the lightning time variation rate D are calculated The steps include:

[0023] Based on the lightning data in the warning area and the grid point location information of the strong convective echo connected area, the lightning frequency in the strong convective echo connected area is counted minute by minute;

[0024] Based on the lightning frequency in the strong convective echo connected area, the lightning time variation rate D and the standard deviation of the lightning time variation rate D are calculated. .

[0025] Preferably, the step of calculating the lightning frequency in the connected area of severe convective echoes includes:

[0026] Based on the lightning data of the warning area and the grid point location information of the strong convective echo connected area, it is determined whether the lightning occurrence location is within the strong convective echo area, and if so, the lightning frequency count is increased by 1; wherein the lightning includes: cloud-to-ground lightning;

[0027] The lightning frequency is calculated at minute intervals.

[0028] Preferably, the lightning time variation rate D and the standard deviation of the lightning time variation rate D are calculated The steps include:

[0029] Step S61: Calculate the average minute lightning frequency within the last two time units based on the lightning frequency in the connected area of the severe convective echo;

[0030] Step S62: Repeat step S61 to calculate the average lightning frequency of multiple minutes within a certain time unit period from the current time, where each average value corresponds to a specific time interval;

[0031] Step S63: Based on the multiple minute lightning frequency averages obtained in steps S61 and S62, the differences between two adjacent minute lightning frequency averages are calculated to obtain multiple lightning time variation rates D;

[0032] Step S64: Based on the set of lightning time variation rates D obtained in step S63, calculate the standard deviation that represents the degree of dispersion of these time variation rates. .

[0033] Preferably, the method for judging hail warning conditions using a support vector machine includes: inputting radar characteristic parameters, if the output is 1, a hail warning is issued, and if the output is 0, no hail warning is issued; wherein, the support vector machine classification model adopts a radial basis kernel function and is trained based on lightning jump data samples.

[0034] The present invention also provides a hail warning system based on lightning surge and support vector machine, the system is used to implement the above method, including: an acquisition module, a position acquisition module, a first calculation module, a second calculation module, a judgment module and an early warning module;

[0035] The acquisition module is used to obtain two-dimensional grid data of radar combined reflectivity factor CR, vertical integrated liquid water content VIL and echo top height TOP based on the radar network mosaic data of the warning area;

[0036] The position acquisition module is used to obtain grid point position information of the strong convective echo connected area based on the two-dimensional grid point data of the radar combined reflectivity factor CR;

[0037] The first calculation module is used to calculate the lightning time variability D and the standard deviation of the lightning time variability D based on the lightning data in the warning area and the grid position information of the strong convection echo connected area. ;

[0038] The second calculation module is used to obtain the maximum value VIL of the vertically integrated liquid water content VIL based on the two-dimensional grid data of the vertically integrated liquid water content VIL and the echo top height TOP and the grid position information of the strong convection echo connected area. max And the VIL max The jump amount GVIL and density VILD;

[0039] The judgment module is used to determine the lightning time variation rate D and the standard deviation of the lightning time variation rate D based on the lightning time variation rate D. Make lightning jump condition judgment;

[0040] The warning module is used to detect the times when the lightning surge condition is met based on the maximum value VIL of the vertical integrated liquid water content VIL. max And the VIL max The jump increment GVIL and density VILD are used to judge the hail warning conditions using the support vector machine to achieve hail warning.

[0041] Preferably, the position acquisition module obtains the grid point position information of the strong convection echo connected area using a neighborhood method, and the process includes:

[0042] performing binarization image processing on the two-dimensional grid point data of the radar combined reflectivity factor CR;

[0043] The neighborhood method is used to identify the strong convective echo connected area of the data after the binary image processing, and the grid point position information of the strong convective echo connected area is obtained.

[0044] Preferably, the process of identifying the connected area of strong convective echoes using the neighborhood method includes:

[0045] With the current strong convective echo grid point as the center, the eight adjacent grid points (upper, lower, left, right, upper left, upper right, lower left, and lower right) are used as the neighborhood of the current grid point. Determine whether any grid point in the neighborhood is a strong convective echo grid point. The judgment conditions include:

[0046] If the grid point is a strong convective echo grid point, it belongs to the same connected area as the central grid point;

[0047] If the grid point is not a strong convective echo grid point, it is a non-connected grid point;

[0048] In this way, domain grid point judgment is performed on all grid points in the same connected area until all domain grid points in the connected area are non-connected grid points, and the judgment and identification of a connected area is completed.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] This invention is based on historical hail weather process and through statistical analysis The lightning jump method uses the difference in the distribution of characteristic parameters of the convective system radar in the case of effective and invalid hail warnings to build a classification model based on support vector machines, which eliminates invalid hail warnings and effectively makes up for the shortcomings of hail warning methods in existing weather forecast services, and improves the application level of lightning data in severe convective weather monitoring and warning. The technical solution provided by the present invention improves The hail warning effect of the lightning jump method effectively reduces the false alarm rate of hail warnings and can be used in existing hail warning business services. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0052] Figure 1 This is a flow chart of a hail warning method based on lightning surge and support vector machine, taking Province A as an example in an embodiment of the present invention;

[0053] Figure 2This is a time distribution diagram of the minute lightning frequency of the convective system in County B, Province A on May 4, 2020, according to an embodiment of the present invention;

[0054] Figure 3 This is a time distribution diagram of the strongest echo of the convective system in County B, Province A on May 4, 2020, according to an embodiment of the present invention;

[0055] Figure 4 This is a time distribution diagram of hail warning condition judgment by the convective system support vector machine in County B, Province A on May 4, 2020, according to an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0057] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0058] Example 1

[0059] This embodiment provides a hail early warning method based on lightning surge and support vector machine, the steps including:

[0060] S1. Based on the radar network mosaic data of the warning area, obtain the two-dimensional grid data of the radar combination reflectivity factor CR, the vertical integrated liquid water content VIL and the echo top height TOP.

[0061] S2. Based on the two-dimensional grid data of the radar combined reflectivity factor CR, obtain the grid position information of the connected area of the strong convective echo.

[0062] S201. Performing binarization image processing on the two-dimensional grid data of the radar combined reflectivity factor CR;

[0063] S202. Using a neighborhood method to identify the strong convective echo connected area in the binarized data, and obtaining the grid point location information of the strong convective echo connected area.

[0064] With the current strong convective echo grid point as the center, the eight adjacent grid points (upper, lower, left, right, upper left, upper right, lower left, and lower right) are used as the neighborhood of the current grid point. Determine whether any grid point in the neighborhood is a strong convective echo grid point. The judgment conditions include:

[0065] If the grid point is a strong convective echo grid point, it belongs to the same connected area as the central grid point;

[0066] If the grid point is not a strong convective echo grid point, it is a non-connected grid point;

[0067] In this way, domain grid point judgment is performed on all grid points in the same connected area until all domain grid points in the connected area are non-connected grid points, and the judgment and identification of a connected area is completed.

[0068] S3. Calculate the lightning time variability D and the standard deviation of the lightning time variability D based on the lightning data in the warning area and the grid location information of the strong convective echo connection area. .

[0069] S301. Based on the lightning data in the warning area and the grid location information of the strong convective echo connected area, the lightning frequency LF of the strong convective echo connected area is counted minute by minute.

[0070] Based on the lightning data in the warning area and the grid location information of the strong convective echo connection area, it is determined whether the lightning occurrence location is within the strong convective echo area. If so, the lightning frequency count is increased by 1. Among them, lightning includes: cloud-to-ground flash;

[0071] The minute-by-minute lightning frequency LF is calculated at intervals of one minute.

[0072] S302. Based on the lightning frequency LF in the connected area of the severe convective echo, calculate the lightning time variation rate D and its standard deviation σ.

[0073] S3021. Calculate the mean minute lightning frequency in the last two time units based on the lightning frequency LF in the connected area of the severe convective echo. , that is, take the average of the minute lightning frequency of the current time unit and the previous time unit;

[0074] Step S3022: Repeat step S3021 to calculate the average lightning frequency of multiple minutes within a certain time unit period from the current time, where each average value corresponds to a specific time interval;

[0075] Step S3023: Based on the multiple minute lightning frequency averages obtained in steps S3021 and S3022, the differences between two adjacent minute lightning frequency averages are calculated to obtain multiple lightning time variation rates D;

[0076] Step S3024: Based on the set of lightning time variability D obtained in step S3023, calculate the standard deviation that characterizes the dispersion of these time variability .

[0077] S4. Based on the two-dimensional grid data of vertically integrated liquid water content VIL and echo top height TOP and the grid position information of the strong convection echo connection area, obtain the maximum value VIL of vertically integrated liquid water content VIL max and VIL max The jump amount GVIL and density VILD.

[0078] Based on the two-dimensional grid data of vertically integrated liquid water content VIL and the grid position information of the strong convection echo connected area, the VIL value in the strong convection echo area is determined grid by grid, and finally the maximum value of VIL VIL is obtained. max ;

[0079] Based on the two-dimensional grid data of echo top height TOP and the grid position information of the strong convection echo connected area, the size of the TOP value in the strong convection echo area is judged grid by grid, and finally the maximum TOP value TOP is obtained. max ;

[0080] Based on the maximum VIL max , calculate the jump amount GVIL of VIL, the formula is as follows:

[0081] GVIL=VIL max (i)-VIL max (i-1),

[0082] Among them, VIL max (i) VIL max (i-1) represents the VIL of two adjacent radar times max .

[0083] Based on the maximum VIL max and TOP max , calculate the density VILD of VIL, the formula is as follows:

[0084] VILD=VIL max / TOP max .

[0085] S5. Based on the lightning time variation rate D and the standard deviation of the lightning time variation rate D Determine the lightning surge conditions.

[0086] Based on the average frequency of lightning strikes per minute , lightning time variation rate D and standard deviation of lightning time variation rate , determine whether the lightning jump condition is met, that is: and , where LF thres is the lightning frequency activation threshold; if the current minute lightning frequency mean is greater than the activation threshold, and the lightning frequency time variation rate D is greater than , it indicates that a lightning jump occurs, that is, the lightning jump condition is met.

[0087] S6. For the times that meet the lightning surge conditions, the maximum value of the vertically integrated liquid water content VIL is VIL max and VIL max The jump increment GVIL and density VILD are used to judge the hail warning conditions using the support vector machine to achieve hail warning.

[0088] The maximum value of the vertically integrated liquid water content VIL is VIL max The support vector machine classification model is used to determine whether the hail warning conditions are met, namely: input radar characteristic parameters (VIL max , GVIL and VILD), if the output is 1, a hail warning is issued, and if the output is 0, no hail warning is issued; the support vector machine classification model adopts the radial basis kernel function and is trained based on the lightning jump data samples; the lightning jump data samples are obtained by the radar characteristic parameters (VIL) near the time of the lightning jump in the historical hail weather process. max , GVIL, and VILD), which include both valid warning samples of hail occurring within 1 hour after a lightning surge, and invalid warning samples of no hail occurring within 1 hour after a lightning surge.

[0089] Example 2

[0090] The following will describe in detail the technical problems solved by the present invention in practical work in conjunction with the present embodiment.

[0091] Take Province A as an example. Figure 1 As shown, the specific steps of implementing a hail warning method based on lightning surge and support vector machine include:

[0092] Step S1: Obtain radar network mosaic data generated by the Severe Weather Nowcasting System (SWAN) operated by the meteorological station in Province A, and obtain the two-dimensional grid data of the radar composite reflectivity factor CR, the vertically integrated liquid water content VIL, and the echo top height TOP;

[0093] Step S2: Obtain lightning data detected by the three-dimensional lightning location network operated by the lightning protection center of Province A and perform quality control;

[0094] Step S3: performing binary image processing on the CR two-dimensional grid point data obtained in step S1, and using the "8" neighborhood method to identify the connected area of strong convective echoes, and obtain the grid point position information of the connected area;

[0095] Step S4: Based on the lightning data obtained in step S2 and the grid point positions of the strong convective echo area obtained in step S3, the lightning frequency LF in the strong convective echo area is counted minute by minute;

[0096] Step S5: Based on the VIL and TOP two-dimensional grid data obtained in step S1 and the grid position of the strong convection echo area obtained in step S3, calculate the maximum VIL value VIL of the strong convection echo area max 、VIL max Jump increments GVIL and VIL max densityVILD;

[0097] Step S6: Based on the minute-by-minute lightning frequency LF obtained in step S4, calculate the lightning time variation rate D and its standard deviation ;

[0098] Step S7: Based on the D and , to judge the lightning jump condition;

[0099] Step S8: For the times that meet the lightning surge condition in step S7, based on the VIL obtained in step S5 max , GVIL and VILD, hail warning conditions are judged through support vector machine classification model to obtain hail warning results.

[0100] Furthermore, in step S3, the specific steps of identifying the strong convective echo connected area are:

[0101] Step S31: performing binary image processing on the CR two-dimensional grid point data obtained in step S1, that is, setting the grid points with CR ≥ 40 dBZ to 1; and setting the grid points with CR < 40 dBZ to 0;

[0102] Step S32: The "8" neighborhood method is used to identify the strong convective echo connected area for the CR two-dimensional grid point data obtained after the binarization processing in step S31, that is, taking the current strong convective echo grid point as the center, the eight grid points of upper, lower, left, right, upper left, upper right, lower left, and lower right are the "8" neighborhood of the current grid point, and it is judged whether any grid point in the "8" neighborhood is a strong convective echo grid point (the value is 1). If so, this grid point and the central grid point belong to the same connected area, and if not, it is a non-connected grid point. The domain grid point judgment is performed on all grid points in the same connected area in this cycle until all the domain grid points in the connected area are all non-connected grid points, and the judgment and identification of a strong convective echo connected area is completed;

[0103] Furthermore, in step S4, the method for counting the lightning frequency LF in the severe convective echo area minute by minute is:

[0104] Step S41: Based on the quality-controlled three-dimensional lightning location data obtained in step S2 and the grid point locations of the strong convective echo area obtained in step S3, determine whether the location (latitude and longitude) of the lightning (cloud-to-ground lightning) is within the strong convective echo area. If so, increment the lightning frequency count by 1.

[0105] Step S42: Execute step S41 at minute intervals to obtain the minute-by-minute lightning frequency LF.

[0106] Furthermore, in step S5, the VIL of the strong convective echo area is calculated. max The specific steps of GVIL and VILD are:

[0107] Step S51: Based on the VIL two-dimensional grid data obtained in step S1 and the grid position of the strong convection echo area obtained in step S3, the VIL value in the strong convection echo area is determined grid by grid point, and finally the maximum value of VIL VIL is obtained. max , that is: VIL max =max(VIL); where max is the maximum value;

[0108] Step S52: Based on the TOP two-dimensional grid point data obtained in step S1 and the grid point position of the strong convection echo area obtained in step S3, the TOP value in the strong convection echo area is determined grid by grid point, and finally the maximum value of TOP is obtained. max , that is: TOP max =max(TOP); where max is the maximum value;

[0109] Step S53: Based on the VIL obtained in step S51 max , calculate VIL max The jump increment GVIL, that is: GVIL=VIL max (i)-VIL max (i-1), where VIL max (i) VIL max (i-1) represents the VIL of two adjacent radar times max ;

[0110] Step S54: Based on the VIL obtained in step S51 max and TOP obtained in step S52 max , calculate VIL max The density VILD, that is: VILD=VIL max / TOP max ;

[0111] Furthermore, in step S6, D and The specific steps for the value are:

[0112] Step S61: Based on the minute-by-minute lightning frequency LF obtained in step S4, calculate the average minute lightning frequency of the last 2 minutes (the previous 0–2 minutes) (Unit: min -1 ),Right now: , where LF(0) and LF(-1) represent the minute lightning frequencies in the previous 0–1 min and the previous 1–2 min, respectively;

[0113] Step S62: Repeat step S61 to calculate the average lightning frequency for 6 minutes in the first 2-14 minutes period, namely: 、 、 、 、 、 ;

[0114] Step S63: Based on the 7-minute lightning frequency averages obtained in steps S61 and S62, the difference between the two adjacent minute lightning frequency averages is calculated to obtain the lightning frequency time variation rate D (unit: min) which reflects the temporal trend of lightning activity. -2 ),Right now: ; Thus, 6 D values can be obtained, namely: D(0), D(-2), D(-4), D(-6), D(-8), D(-10);

[0115] Step S64: Based on the set of lightning frequency time variation rates D obtained in step S63, calculate the standard deviation representing the dispersion of lightning frequency time variation rates (Unit: min -2 ),Right now: ; Among them, std is the standard deviation;

[0116] Furthermore, in step S7, the specific steps of determining the lightning surge condition are as follows:

[0117] Step S71: Based on the result obtained in step S6 , D(0) and , determine whether the lightning jump condition is met, that is: and , where LF thres is the lightning frequency activation threshold (value is 2 min -1 ); If the current minute lightning frequency mean is greater than the activation threshold, and the lightning frequency time variation rate D is greater than , it indicates that a lightning jump occurs, that is, the lightning jump condition is met;

[0118] Furthermore, support vector machines (SVMs) are a machine learning algorithm widely used in classification and regression problems, performing exceptionally well with high-dimensional data and small sample sizes. The core idea of SVMs is to find a hyperplane that perfectly separates data of different categories while ensuring that the closest data points are as far away from the hyperplane as possible. For linearly inseparable data, support vector machines utilize kernel techniques, mapping the data into a high-dimensional space using a kernel function, thereby transforming the originally linearly inseparable problem into a linearly separable problem in that high-dimensional space. Whether a lightning surge can provide a hail warning is also a binary classification problem: effective or ineffective, and hail samples are relatively rare compared to other classification problems. Therefore, based on radar feature parameters corresponding to effective and ineffective lightning surges in historical hail cases, a support vector machine classification model is constructed to assist the lightning surge algorithm in providing hail warnings, further improving the high false alarm rate of the lightning surge algorithm.

[0119] Furthermore, in step S8, the specific steps of determining hail warning conditions are as follows:

[0120] Step S81: Based on the historical hail weather process, the radar characteristic parameters (VIL max , GVIL and VILD) to form a lightning jump sample data set;

[0121] Step S82: Based on the lightning surge sample data set obtained in step S81, each sample is divided into a valid warning sample and an invalid warning sample according to whether hail occurs within 1 hour after the lightning surge occurs. That is, a valid warning sample is one in which hail occurs within 1 hour after the lightning surge occurs, and conversely, an invalid warning sample is one in which no hail occurs within 1 hour after the lightning surge occurs.

[0122] Step S83: Distribute the lightning surge sample data set obtained in step S81 in a ratio of 8:2 to form a training sample data set and a validation sample data set, and ensure that the ratio of valid warning samples to invalid warning samples in the two sample sets is consistent with that of the entire lightning surge sample data set;

[0123] Step S84: The support vector machine uses a nonlinear function to map the nonlinear separable problem from the original feature space to a higher-dimensional Hilbert space, thereby converting it into a linear separable problem. At this time, the hyperplane serving as the decision boundary is expressed as follows:

[0124] ,

[0125] In the formula is the mapping function; represents the normal vector of the hyperplane; b represents the intercept of the hyperplane; T represents the transpose operation; X represents the input sample data.

[0126] Since the mapping function has a complex form and it is difficult to calculate its inner product, the kernel method can be used, that is, the inner product of the mapping function is defined as a kernel function to avoid the explicit calculation of the inner product:

[0127] ,

[0128] Where, represents the kernel function; X1 and X2 both represent input sample data;

[0129] Based on the training sample data set and the validation sample data set obtained in step S83, a support vector machine with a radial basis kernel function (RBF kernel or Gaussian kernel) is trained to obtain a support vector machine classification model. The radial basis kernel function is:

[0130] ,

[0131] Where, A parameter representing the radial basis kernel function, which controls the speed at which the function decays.

[0132] Step S85: Based on the support vector machine classification model obtained in step S84, the hail warning condition is judged by the support vector machine (SVM) for the times that meet the lightning surge condition in step S7. The judgment function is as follows:

[0133] ,

[0134] That is: if the lightning jump condition is met ( ), then the radar characteristic parameter (VIL max , GVIL, and VILD) are input into the support vector machine classification model; if the output is 1, the hail warning condition is met; if the output is 0, the hail warning condition is not met.

[0135] In this way, the method of the present invention can effectively reduce the traditional There are many false alarms when using the lightning jump hail warning method.

[0136] The radar network mosaic data selected in this example comes from the Severe Weather Nowcasting System (SWAN) operated by the Meteorological Observatory of Province A. The data includes the combined reflectivity factor (MCR), vertically integrated liquid water content (MVIL), and echo top height (MTOP). Its grid resolution is 1 km, and its temporal resolution is 6 minutes. Radar network mosaic data is the most frequently used and important data type in severe convective weather forecasting and warning services, and has become a crucial basis for improving disaster weather preparedness and accurately issuing warnings.

[0137] The lightning data selected in this example comes from the Province A Three-Dimensional Lightning Location Network (HBLLS), operated by the Province A Lightning Protection Center. Its overall detection efficiency exceeds 95%, with a horizontal positioning error of less than 200 meters and a vertical positioning error of less than 500 meters. Currently, this lightning location network consists of 19 VLF / LF (Very Low Frequency / Low Frequency) lightning locators developed by the Institute of Electrical Engineering of the Chinese Academy of Sciences. This network enables real-time, high-precision networked observation of total lightning flashes (cloud-to-ground lightning flashes) across Province A.

[0138] The hail weather data selected in this example comes from direct hail disaster reports verified by county, city, and district-level meteorological departments and reported through the China Meteorological Administration's meteorological disaster management system, as well as eyewitness reports from official media outlets and new media platforms such as Weibo. To ensure the authenticity of the hailstorm process and the usability of the early warning method evaluation, only hailstorm records with accurate photographs and real-time location information were selected. Reports with ambiguous hail size, unclear time and location information, and incomplete lightning data were excluded.

[0139] The following analyzes the application effect of a hail warning method based on lightning surge and support vector machine involved in this application, combining a hail process in County B, Province A on May 4, 2020 and the overall evaluation results of 116 hail weather processes in Province A from 2015 to 2023.

[0140] On the afternoon of May 4th, severe convective activity occurred in County B, Province A, leading to a sudden hailstorm in Village D, Township C, County B. At 4:12 PM, a convective system developed in southern County B, with a maximum echo intensity exceeding 50 dBZ and sporadic lightning activity. Two other cells developed simultaneously to the left of the system. After more than 10 minutes of rapid development, the central intensity of the convective system increased further at 4:24 PM, exceeding 60 dBZ for the first time. At this time, lightning activity became active and concentrated near the center of the strong echo. At 4:48 PM, the convective system's intensity remained above 60 dBZ, with a maximum echo intensity of 65 dBZ, and lightning activity showed no significant weakening. However, the center of the strong echo rapidly descended and touched the ground, indicating the onset of hail at this time. Subsequently, the convective system began to weaken as it moved eastward. At 5:18 PM, the maximum echo intensity of the convective system dropped below 60 dBZ, and lightning activity intensified, with a significant increase in frequency.

[0141] Figure 2-Figure 4 The minute lightning frequency and the strongest echo of the convective system in County B, Province A on May 4, 2020 ( Figure 2 )、 Lightning jump condition judgment ( Figure 3 ) and support vector machine hail warning condition judgment ( Figure 4 ) time distribution diagram.

[0142] Depend on Figure 2 It can be seen that the echo intensity of the convective system increased rapidly and reached a peak before the hail fell, and then maintained above 60dBZ for about 6 minutes; as the hail occurred, the strongest echo decreased rapidly and reached a minimum of 57dBZ at 17:06. At the same time, the evolution of the convective system was also accompanied by obvious lightning activity, and its minute lightning frequency basically exceeded The activation threshold of the algorithm, and the maximum minute lightning frequency occurs after hail, more than 20 minutes -1 ; Among them, there were 7 lightning jumps in the convective system ( Figure 3 Actual hail data (hailfall time: 16:48) shows that the first five lightning surges (16:04, 16:16, 16:18, 16:20, and 16:42) were valid hail warnings, while the last two lightning surges (16:54 and 17:10) were invalid. Based on the earliest lightning surge, the lead time for this hailstorm was approximately 44 minutes.

[0143] Depend on Figure 4 It can be seen that the radar characteristic parameters (VIL max, GVIL, and VILD) are input into the support vector machine classification model, and the output is 1, indicating that the lightning surge is a valid hail warning. This embodiment provides a correct hail warning judgment. Similarly, the method of this embodiment also successfully determines that the second to fifth lightning surges are valid hail warnings. Among them, the third and fourth lightning surges have adjacent times and the same proximity to the radar time (16:18). In addition, the output of the support vector machine classification model for the sixth and seventh lightning surges is 0, indicating that the hail warning conditions are not met. This embodiment also successfully identifies these two invalid hail warnings, effectively reducing two false hail reports.

[0144] Table 1 shows the hail warning effectiveness evaluation results of the traditional 2σ lightning jump method and the improved method based on lightning jump and support vector machine in this embodiment based on 116 hail events in Province A from 2015 to 2023. The three parameters of hit rate (POD), false alarm rate (FAR), and critical success index (CSI) were used to evaluate the warning effectiveness of the two methods for hail weather in Province A, namely:

[0145] POD=NA / (NA+NC),

[0146] FAR= NB / (NA+NB),

[0147] CSI= NA / (NA+NB+NC),

[0148] Where NA is the number of samples with successful hail warnings, NB is the number of samples with hail warnings but no actual hail, and NC is the number of samples with actual hail but no hail warnings.

[0149] From the hail warning effect evaluation results of the traditional method of 2σ lightning jump based on 116 hail weather processes in Province A from 2015 to 2023 in Table 1 of the embodiment of the present application and the improved method of lightning jump and support vector machine of the present invention, it can be seen that: The traditional lightning surge method achieved POD, FAR, and CSI of 93.7%, 41.4%, and 56.4%, respectively. Its overall warning effectiveness is similar to the evaluation results of 177 hail events in a certain region (POD, FAR, and CSI of 80.2%, 41.6%, and 51.1%, respectively). Furthermore, compared with the results of this method applied in a certain country, the evaluation results in Province A showed a higher hit rate but a higher false alarm rate.

[0150] Table 1

[0151] .

[0152] Introducing radar characteristic parameters (VIL maxAfter the improved support vector machine method (GVIL, GVIL and VILD) was used, although the hail NA decreased from 208 to 180, the hit rate POD did not change significantly, only decreasing by 2.8%; on the contrary, the change in NB was particularly significant, decreasing from 147 to 67, and its false alarm rate FAR decreased from 41.4% to 27.1%, indicating that the improved method can effectively reduce the hail false alarm rate while ensuring the accuracy rate. In addition, the critical success index CSI increased from 56.4% to 67.9%, which also shows that the improved method has an overall better hail warning effect, that is, based on radar characteristic parameters (VIL max It is feasible to construct a support vector machine classification model based on GVIL, GVIL and VILD to assist in improving the hail warning method based on lightning jump.

[0153] This example uses the actual detection data of the convective system in County B, Province A on May 4, 2020. First, the radar network mosaic data is read to obtain the two-dimensional grid data of the combined reflectivity factor CR, the vertical integrated liquid water content VIL, and the echo top height TOP, and the CR data is binarized; then, the "8" neighborhood method is used to identify the connected area of strong convective echoes and calculate the maximum value VIL of the connected area of strong convective echoes. max Then, the lightning data after quality control is read, and the minute-by-minute lightning frequency in the strong convective echo connection area is counted, and based on The algorithm calculates the lightning time variation rate D and its standard deviation Then, the minute lightning frequency is greater than the threshold and D is greater than Finally, for the times that meet the lightning surge conditions, based on VIL max , GVIL and VILD were used to determine hail warning conditions using support vector machines, and hail warning services were provided. The following conclusions were initially drawn:

[0154] (1) Traditional The lightning jump method, which relies on the phenomenon of lightning frequency jumps before a hailstorm, provides early warning for hail, and has achieved good results in practice. However, lightning jumps can also occur during severe convective weather events other than hail, leading to a high incidence of false alarms.

[0155] (2) This embodiment involves a hail warning method based on lightning surge and support vector machine. Based on the flash jump method, VIL is introduced max, GVIL, VILD and other radar characteristic parameters reflecting hail indicators, and trained a support vector machine classification model based on the statistical results of a large number of historical hail processes, so as to improve the hail warning effect. Judging from the warning results of actual hail processes, compared with the traditional method, the improved method of this application can effectively reduce the false alarm rate FAR while maintaining its hit rate POD, further improve the critical success index CSI, and thus enhance the hail warning effect. The improved method of this application can provide strong technical support for the use of radar and lightning data in meteorological business to carry out hail forecast and warning services.

[0156] From the above conclusions, it can be seen that the method of this embodiment can extract the combined reflectivity factor, vertically integrated liquid water content and echo top height from the radar network mosaic data, and perform binarization processing on the combined reflectivity factor. Then, the "8" neighborhood method in image recognition technology is combined to identify the connected area of strong convective echoes; then, the lightning data is quality controlled and the minute lightning frequency in the connected area of strong convective echoes is counted; then, radar characteristic parameters such as the maximum value, jump amount and density of the vertically integrated liquid water content are calculated based on the identified strong convective connected area; finally, based on the judgment result of the lightning jump condition, a support vector machine classification model trained based on statistical data of historical hail processes is used to determine whether it is an effective hail warning, thereby realizing a hail warning service based on radar and lightning data.

[0157] Example 3

[0158] This embodiment also provides a hail warning system based on lightning surge and support vector machine, including: an acquisition module, a position acquisition module, a first calculation module, a second calculation module, a judgment module and an early warning module; the acquisition module is used to obtain two-dimensional grid data of radar combination reflectivity factor CR, vertical integrated liquid water content VIL and echo top height TOP based on radar network mosaic data in the early warning area; the position acquisition module is used to obtain grid position information of the connected area of strong convective echo based on the two-dimensional grid data of the radar combination reflectivity factor CR; the first calculation module is used to calculate the lightning time variability D and the standard deviation of the lightning time variability D based on the lightning data in the early warning area and the grid position information of the connected area of strong convective echo The second calculation module is used to obtain the maximum value VIL of the vertically integrated liquid water content VIL based on the two-dimensional grid data of the vertically integrated liquid water content VIL and the echo top height TOP and the grid position information of the strong convection echo connected area. max and VIL max The jump amount GVIL and density VILD; the judgment module is used based on the lightning time variation rate D and the standard deviation of the lightning time variation rate D Perform lightning surge condition judgment; the early warning module is used to judge the times when the lightning surge conditions are met, based on the maximum value VIL of the vertical integrated liquid water content VILmax and VIL max The jump increment GVIL and density VILD are used to judge the hail warning conditions using the support vector machine to achieve hail warning.

[0159] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A hail warning method based on lightning surge and support vector machine, characterized in that the steps include: Based on the radar network mosaic data of the warning area, obtain the two-dimensional grid data of the radar combination reflectivity factor CR, vertical integrated liquid water content VIL and echo top height TOP; Based on the two-dimensional grid point data of the radar combined reflectivity factor CR, grid point position information of the connected area of the strong convective echo is obtained; Calculate the lightning time variability D and the standard deviation σ of the lightning time variability D based on the lightning data in the warning area and the grid point location information of the connected area of the severe convective echo; Based on the two-dimensional grid data of the vertically integrated liquid water content VIL and the echo top height TOP and the grid position information of the strong convection echo connected area, the maximum value VIL of the vertically integrated liquid water content VIL is obtained. max And the VIL max The jump amount GVIL and density VILD; Performing lightning surge condition judgment based on the lightning time variation rate D and the standard deviation σ of the lightning time variation rate D; For the times that meet the lightning surge conditions, the maximum value VIL of the vertically integrated liquid water content VIL is max And the VIL max The jump increment GVIL and density VILD are used to judge the hail warning conditions using the support vector machine to achieve hail warning.

2. The hail early warning method based on lightning surge and support vector machine according to claim 1, characterized in that: The grid point location information of the connected area of the strong convective echo is obtained by using a neighborhood method, the steps comprising: performing binarization image processing on the two-dimensional grid point data of the radar combined reflectivity factor CR; The neighborhood method is used to identify the strong convective echo connected area of the data after the binary image processing, and the grid point position information of the strong convective echo connected area is obtained.

3. The hail early warning method based on lightning surge and support vector machine according to claim 2, characterized in that: The steps of identifying the connected area of strong convective echoes using the neighborhood method include: With the current strong convective echo grid point as the center, the eight adjacent grid points (upper, lower, left, right, upper left, upper right, lower left, and lower right) are used as the neighborhood of the current grid point. Determine whether any grid point in the neighborhood is a strong convective echo grid point. The judgment conditions include: If the grid point is a strong convective echo grid point, it belongs to the same connected area as the central grid point; If the grid point is not a strong convective echo grid point, it is a non-connected grid point; In this way, domain grid point judgment is performed on all grid points in the same connected area until all domain grid points in the connected area are non-connected grid points, and the judgment and identification of a connected area is completed.

4. The hail early warning method based on lightning surge and support vector machine according to claim 1, characterized in that: The step of calculating the lightning time variation rate D and the standard deviation σ of the lightning time variation rate D includes: Based on the lightning data in the warning area and the grid point location information of the strong convective echo connected area, the lightning frequency in the strong convective echo connected area is counted minute by minute; Based on the lightning frequency in the strong convective echo connected area, the lightning time variability D and the standard deviation σ of the lightning time variability D are calculated.

5. The hail early warning method based on lightning surge and support vector machine according to claim 4, characterized in that: The steps for calculating the lightning frequency in the area connected by severe convective echoes include: Based on the lightning data of the warning area and the grid point location information of the strong convective echo connected area, it is determined whether the lightning occurrence location is within the strong convective echo area, and if so, the lightning frequency count is increased by 1; wherein the lightning includes: cloud-to-ground lightning; The lightning frequency is calculated at minute intervals.

6. The hail early warning method based on lightning surge and support vector machine according to claim 4, characterized in that: The step of calculating the lightning time variation rate D and the standard deviation σ of the lightning time variation rate D includes: Step S61: Calculate the average minute lightning frequency within the last two time units based on the lightning frequency in the connected area of the severe convective echo; Step S62: Repeat step S61 to calculate the average lightning frequency of multiple minutes within a certain time unit period from the current time, where each average value corresponds to a specific time interval; Step S63: Based on the multiple minute lightning frequency averages obtained in steps S61 and S62, the differences between two adjacent minute lightning frequency averages are calculated to obtain multiple lightning time variation rates D; Step S64: Based on the set of lightning time variation rates D obtained in step S63, calculate the standard deviation σ that represents the degree of dispersion of these time variation rates.

7. The hail early warning method based on lightning surge and support vector machine according to claim 1, characterized in that: The method for judging hail warning conditions using a support vector machine includes: inputting radar characteristic parameters, if the output is 1, a hail warning is issued, and if the output is 0, no hail warning is issued; wherein, the support vector machine classification model adopts a radial basis kernel function and is trained based on lightning jump data samples.

8. A hail warning system based on lightning surge and support vector machine, the system being used to implement the method according to any one of claims 1 to 7, characterized in that: include: Acquisition module, location acquisition module, first calculation module, second calculation module, judgment module and early warning module; The acquisition module is used to obtain two-dimensional grid data of radar combined reflectivity factor CR, vertical integrated liquid water content VIL and echo top height TOP based on the radar network mosaic data of the warning area; The position acquisition module is used to obtain grid point position information of the strong convective echo connected area based on the two-dimensional grid point data of the radar combined reflectivity factor CR; The first calculation module is used to calculate the lightning time variation rate D and the standard deviation σ of the lightning time variation rate D based on the lightning data in the warning area and the grid point position information of the strong convective echo connected area; The second calculation module is used to obtain the maximum value VIL of the vertically integrated liquid water content VIL based on the two-dimensional grid data of the vertically integrated liquid water content VIL and the echo top height TOP and the grid position information of the strong convection echo connected area. max And the VIL max The jump amount GVIL and density VILD; The judgment module is used to judge the lightning jump condition based on the lightning time variation rate D and the standard deviation σ of the lightning time variation rate D; The warning module is used to detect the times when the lightning surge condition is met based on the maximum value VIL of the vertical integrated liquid water content VIL. max And the VIL max The jump increment GVIL and density VILD are used to judge the hail warning conditions using the support vector machine to achieve hail warning.

9. The hail early warning system based on lightning surge and support vector machine according to claim 8, characterized in that: The position acquisition module obtains the grid point position information of the strong convection echo connected area using a neighborhood method, and the process includes: performing binarization image processing on the two-dimensional grid point data of the radar combined reflectivity factor CR; The neighborhood method is used to identify the strong convective echo connected area of the data after the binary image processing, and the grid point position information of the strong convective echo connected area is obtained.

10. The hail early warning system based on lightning surge and support vector machine according to claim 9, characterized in that: The process of identifying the connected area of strong convective echoes using the neighborhood method includes: With the current strong convective echo grid point as the center, the eight adjacent grid points (upper, lower, left, right, upper left, upper right, lower left, and lower right) are used as the neighborhood of the current grid point. Determine whether any grid point in the neighborhood is a strong convective echo grid point. The judgment conditions include: If the grid point is a strong convective echo grid point, it belongs to the same connected area as the central grid point; If the grid point is not a strong convective echo grid point, it is a non-connected grid point; In this way, domain grid point judgment is performed on all grid points in the same connected area until all domain grid points in the connected area are non-connected grid points, and the judgment and identification of a connected area is completed.

Images