Strong convection weather tracking and early warning method based on radar data and deep learning

Abstract

The invention relates to the technical field of meteorological disaster monitoring, and particularly discloses a severe convective weather tracking and early warning method based on radar data and deep learning, and the method comprises the steps: collecting original radar data in real time through a laser radar and a spaceborne radar, obtaining the original radar data, and carrying out the cleaning and feature extraction processing, thereby obtaining real-time spatial-temporal feature data; real-time spatial-temporal characteristic data of radar data are automatically learned through a deep learning model, so that the deep learning model has the capability of automatically identifying and tracking a severe convection system, the boundary and strength of the severe convection system are directly output through the deep learning model, and a mask of the severe convection system is directly generated through one-time forward propagation. The calculation amount of radar data is effectively reduced, finally, the future influence area of the severe convection system is predicted based on the evolution characteristics of the severe convection system, early warning information is generated and output, the whole process from data processing to early warning release is free of manual intervention, and the automation degree and efficiency of early warning are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of meteorological disaster monitoring, and specifically relates to a severe convective weather tracking and early warning method based on radar data and deep learning. Background Art

[0002] Severe convective weather is one of the most sudden and disastrous weather types in meteorological monitoring. Severe convective weather mainly includes thunderstorms, strong winds, hail, tornadoes and short-term heavy rainfall. It has a short life span and a small spatial scale (up to 100 meters), posing a major threat to human property and safety. In order to reduce the damage to property and harm to life safety caused by severe convective weather, it is necessary to track severe convective weather.

[0003] At present, the method of tracking severe convective weather usually adopts observation means such as satellites and lidar, and uses the brightness temperature obtained by satellite observation to identify and track mesoscale convective systems (MCS), and then predict the occurrence and movement of severe convective weather. For example, the patent application number CN202310589785.X discloses a remote sensing-based severe convective weather monitoring and early warning method and system for power line. The method includes: using the brightness temperature obtained by satellite remote sensing to identify and track the occurrence and evolution process of mesoscale convective systems; according to the evolution process of the mesoscale convective system, extracting the main evolution characteristics of the mesoscale convective system, and based on its main evolution characteristics, using weighted average and Lagrange interpolation method to predict the impact area of ​​this mesoscale convective system at the next moment; according to the prediction results and the position of each transmission tower of the power line, issuing early warnings for the transmission towers in the impact area of ​​the mesoscale convective system, marking the transmission towers in the impact area, and outputting early warning results.

[0004] However, the above scheme has low computational efficiency when processing large amounts of satellite or radar data, and is not convenient for quickly outputting warning results. For example, when tracking the evolution process of a mesoscale convective system and extracting its evolution characteristics, the overlapping area method is used to track the cold cloud system. It is necessary to calculate the overlapping area ratio of all pairs of regions at adjacent moments, so that each overlapping area calculation requires traversing the pixels of two regions, which is highly complex. Therefore, we need to propose a severe convective weather tracking and warning method based on radar data and deep learning to solve the above problems, so that it can automatically process and analyze large amounts of data through deep learning technology, reduce manual intervention, and improve the automation and efficiency of warnings. Summary of the Invention

[0005] The purpose of the present invention is to provide a severe convective weather tracking and warning method based on radar data and deep learning, which can automatically process and analyze large amounts of data through deep learning technology, reduce manual intervention, and improve the automation and efficiency of warnings to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] The severe convective weather tracking and warning method based on radar data and deep learning includes the following steps:

[0008] S1, real-time acquisition of raw radar data of radar reflectivity, radial velocity, and velocity spectrum width through lidar and spaceborne radar;

[0009] S2. Obtaining raw radar data and performing cleaning and feature extraction to obtain real-time spatiotemporal feature data;

[0010] S3. Build a deep learning model suitable for tracking severe convective weather and train it with historical data to enable it to automatically identify and track severe convective systems.

[0011] S4. Input the real-time spatiotemporal feature data into the trained deep learning model to automatically identify severe convective systems in radar data and establish spatiotemporal correlation tracking to obtain the historical evolution characteristics of severe convective systems;

[0012] S5. Predict the future impact areas of severe convective systems based on their historical evolutionary characteristics;

[0013] S6. Generate and output warning information based on the predicted impact area of ​​the severe convective system.

[0014] Preferably, the processing flow of the raw radar data is as follows:

[0015] A1. Use the median filter algorithm to filter out noise in the original radar data and fill in missing values ​​through interpolation.

[0016] A2. Extract basic characteristics of radar reflectivity factor, radial velocity, and velocity spectrum width from filtered radar data, and simultaneously calculate characteristics related to severe convection, including storm top height, vertical liquid water content, mesocyclone characteristics, and hail index;

[0017] A3. Normalize the strong convection-related features.

[0018] Preferably, the deep learning model uses a spatiotemporal convolutional neural network combined with an attention mechanism to capture the spatial characteristics and temporal evolution laws of radar data. The attention mechanism includes spatial attention and channel attention. Spatial attention focuses on strong convection-related areas by calculating the spatial weights of feature maps; channel attention automatically learns the importance weights of radar features.

[0019] Preferably, the training process of the deep learning model is:

[0020] B1. Collect historical radar data containing radar reflectivity factor, radial velocity, and velocity spectrum width, as well as manually annotated severe convective system boundaries and intensity levels to establish a data set;

[0021] B2. Divide the data set into training set, validation set and test set in a ratio of 8:1:1;

[0022] B3. Use the cross-entropy loss function to train the deep learning model based on the training set data, and use the Adam optimizer to dynamically adjust the learning rate to accelerate convergence;

[0023] B4. Evaluate the performance of the deep learning model based on the validation set data. When the prediction accuracy reaches the preset threshold, the training is completed. Otherwise, continue to train and converge the deep learning model until the deep learning model training is completed.

[0024] Preferably, the process of real-time identification of severe convective systems is as follows:

[0025] C1. Segment the real-time spatiotemporal feature data according to time intervals or spatial ranges to form multiple data segments with independent spatiotemporal attributes;

[0026] C2, outputs the pixel-level segmentation bounding box of the strong convection system through forward propagation, and outputs the intensity score at the same time;

[0027] C3. Use non-maximum suppression to remove overlapping detection frames and retain the optimal detection results.

[0028] Preferably, the spatiotemporal correlation tracking process is as follows:

[0029] D1. Extract the spatiotemporal characteristics of the centroid coordinates, area, and mean reflectivity of the bounding box for each detected severe convective system;

[0030] D2. Calculate the spatial distance and characteristic distance between severe convective systems at adjacent moments and construct a cost matrix;

[0031] D3. Use the Hungarian algorithm to solve the optimal match, establish the continuity relationship of the system ID, and handle the complex situations of splitting and merging.

[0032] Preferably, the cost matrix formula is:

[0033] ,

[0034] ,

[0035] ,

[0036] in, is the first A strong convective system, is the jth severe convective system at time t+1, and for Center of mass coordinates, and for The centroid coordinates of for The mean reflectivity, for The mean reflectivity, is the weight coefficient used to adjust the importance of spatial distance and feature distance in the cost matrix, for and The cost of matching.

[0037] Preferably, the continuation relationship of the system ID is constructed by solving the cost matrix through the Hungarian algorithm to obtain the optimal matching pair of the strong convective system at adjacent moments. and Match, then inherit The system ID of the system to establish a continuation relationship.

[0038] Preferably, the process of impact area prediction is as follows:

[0039] E1. Extract the characteristics of velocity, acceleration and shape changes from the evolution trajectory of the severe convective system obtained by spatiotemporal correlation tracking to obtain historical characteristics;

[0040] E2. Use the self-attention mechanism to predict the location and range of future moments based on historical features;

[0041] E3. Evaluate the error based on the predicted location and range, and select the location and range with the smallest error as the affected area.

[0042] Preferably, when generating warning information, the predicted impact area and the location of the meteorological station are spatially superimposed and analyzed to determine which meteorological stations are located in the impact area. Different warning levels are divided according to the intensity and impact degree of the severe convective system. Finally, according to the warning level, warning information including the impact area, warning level and disaster prevention measures is generated and released to each impact area.

[0043] The severe convective weather tracking and early warning method proposed in this invention based on radar data and deep learning has the following advantages over existing technologies:

[0044] 1. The present invention automatically learns the real-time spatiotemporal feature data of radar data through a deep learning model, so that the deep learning model has the ability to automatically identify and track severe convective systems. The boundary and intensity of the severe convective system are directly output through the deep learning model, and the mask of the severe convective system is directly generated through a forward propagation, which effectively reduces the amount of calculation on the radar data. Finally, based on the evolution characteristics of the severe convective system, the future impact area of ​​the severe convective system is predicted, and early warning information is generated and output according to the impact area. There is no human intervention in the entire process from data processing to early warning issuance, which improves the automation and efficiency of early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A flowchart of a severe convective weather tracking and warning method according to an embodiment of the present invention is shown;

[0046] Figure 2 A flowchart of processing raw radar data according to an embodiment of the present invention is shown;

[0047] Figure 3 A flowchart of a deep learning model training process according to an embodiment of the present invention is shown;

[0048] Figure 4 A flowchart of real-time identification of a severe convective system according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0050] The present invention provides Figures 1-4 The severe convective weather tracking and warning method based on radar data and deep learning includes the following steps:

[0051] S1, real-time acquisition of raw radar data of radar reflectivity, radial velocity, and velocity spectrum width through lidar and spaceborne radar;

[0052] The satellite-borne radar is configured as a GPM satellite and is carried on the satellite, capable of realizing global observations, used to provide global background information, and helpful to grasp the overall movement trend and development direction of the weather system; the lidar is carried by a drone, and the drone is located at a weather station, and is used to provide high-precision three-dimensional spatial information, which helps to accurately track the internal structure and dynamic changes of convective weather; through the combination of lidar and satellite-borne radar, a more comprehensive understanding of the characteristics and evolution laws of severe convective weather can be achieved, thereby improving the tracking and early warning capabilities of severe convective weather.

[0053] S2. Obtaining raw radar data and performing cleaning and feature extraction to obtain real-time spatiotemporal feature data;

[0054] like Figure 2 As shown in the figure, the processing flow of raw radar data is as follows:

[0055] A1. Use the median filter algorithm to filter out noise in the original radar data, and interpolate to fill in missing values ​​to eliminate the impact of outliers on the original radar data analysis;

[0056] A2. Extract basic characteristics of radar reflectivity factor, radial velocity, and velocity spectrum width from filtered radar data, and simultaneously calculate characteristics related to severe convection, including storm top height, vertical liquid water content, mesocyclone characteristics, and hail index;

[0057] The basic feature extraction method is as follows:

[0058] The radar reflectivity factor extraction method is to directly read the reflectivity factor value from the radar raw data. The unit is dBZ. The radar reflectivity factor reflects the ability of precipitation particles within a unit volume to scatter electromagnetic waves and is used to judge the precipitation intensity.

[0059] The radial velocity is calculated based on the Doppler effect by the frequency shift of the echo signal. The radial velocity calculation formula is:

[0060] ,in, is the radial velocity, is the radar wavelength, For Doppler, is the distance from the target to the radar, and the radial velocity is the radial speed of the target relative to the radar. A positive value indicates that it is moving away from the radar, and a negative value indicates that it is moving closer to the radar.

[0061] The velocity spectrum width is the standard deviation obtained by calculating the power spectrum of the echo signal. The standard deviation calculation formula is:

[0062] ,in, is the velocity spectrum width, For the velocity components, is the average speed, For the The power corresponding to each speed, and the speed spectrum width are used to reflect the atmospheric turbulence intensity and the size distribution of precipitation particles;

[0063] The storm top height is the vertical development height of the strong convective cloud body, reflecting the convection intensity. The higher the height, the stronger the convection. The calculation method is to extract the vertical distribution of the reflectivity factor along the vertical direction and find the height layer where the vertical gradient of the reflectivity decreases significantly. The calculation formula is:

[0064] ,in, is the storm top height, is the vertical height, is a small change in height h, is the variation of reflectivity factor with height in the vertical direction, The gradient threshold used to reflect the critical value of reflectivity decreasing with altitude, usually taken as -5dBZ / km;

[0065] The vertical liquid water content is the total mass of precipitation particles per unit area and is used to assess contact potential and hail risk. It is calculated by converting the radar reflectivity factor into liquid water content using the reflectivity-liquid water content relationship and integrating it along the vertical direction. The calculation formula is:

[0066] ,in, is the vertical liquid water content, is the storm top height, is the lowest measurable height, is a small change in height h, is the liquid water content at height h, and the relationship between reflectivity and liquid water content is: Z=a*L b , a=200, b=1.6, Z is the radar reflectivity factor, L is the liquid water content corresponding to the radar reflectivity factor Z;

[0067] The mesocyclone characteristic quantity is used to reflect the rotating airflow within a severe convective system and is a precursor to severe weather such as tornadoes. The calculation formula for the mesocyclone characteristic quantity is:

[0068] ,in, and is the maximum value of positive and negative speed, is the horizontal distance of the velocity pair;

[0069] The hail index is used to assess the probability of hail in severe convective systems. The hail index calculation formula is:

[0070] ,in, is the hail index, is the vertical liquid water content, is the storm top height, is the height of the 0°C isotherm in the atmosphere;

[0071] A3. Normalize strong convection-related features to avoid interference from different feature dimensions on deep learning model training.

[0072] S3. Build a deep learning model suitable for tracking severe convective weather and train it with historical data to enable it to automatically identify and track severe convective systems.

[0073] The deep learning model uses a spatiotemporal convolutional neural network combined with an attention mechanism to capture the spatial characteristics and temporal evolution of radar data. The spatiotemporal convolutional neural network formula is:

[0074] ,

[0075] ,

[0076] ,

[0077] ,

[0078] ,

[0079] in, For input data, is the cell state at time t, is the cell state at time t-1, is the hidden state at time t, is the hidden state at time t-1, Input data for controlling the current moment There are multiple pieces of information stored in the cell state The input gate, To determine the cell state Which historical information is forgotten in the forget gate? To control cell state How much information is output as hidden state The output gate, and are cell state and hidden state respectively, is the sigmoid function, is the convolution operation, is element-wise multiplication, For input data The convolution weight matrix to the input gate, Hidden The convolution weights to the input gate, is the bias term of the input gate, is the bias term of the forget gate, is the bias term of the cell state, and are the convolution weights from input data and hidden state to candidate memory, is the hyperbolic tangent activation function, 、 and are the convolution weights from input data, hidden state, and cell state to the output gate, is the bias term of the output gate;

[0080] The attention mechanism includes spatial attention and channel attention. Spatial attention focuses on strong convection-related areas by calculating the spatial weight of the feature map. The calculation formula of spatial weight is:

[0081] ,in, is the spatial weight value, is the sigmoid function, is m*m convolution, X is the feature map, and is the maximum pooling and average pooling of the spatial dimension;

[0082] Channel attention is the importance weight of automatically learning radar features. The importance weight calculation formula is:

[0083] ,in, is the spatial weight value, is the sigmoid function, is m*m convolution, is the channel dimension feature, and is the maximum pooling and average pooling of the spatial dimension, For a multi-layer perceptron, weights are generated through global pooling and nonlinear transformation of the channel dimension ;

[0084] By introducing spatial attention and channel attention, the sensitivity to key areas of strong convection can be enhanced.

[0085] like Figure 3 As shown in Figure 2, the training process of the deep learning model is:

[0086] B1, collect historical radar data containing radar reflectivity factor, radial velocity and velocity spectrum width, and manually labeled data of severe convective system boundary, intensity level, etc. to establish a data set;

[0087] B2, divide the data in the data set into training set, validation set and test set in the ratio of 8:1:1 to ensure that the deep learning model performs stably on unseen data;

[0088] B3, train the deep learning model based on the training set data using the cross-entropy loss function, and use the Adam optimizer to dynamically adjust the learning rate to speed up convergence; in the art, the cross-entropy loss function and the Adam optimizer are used to train and optimize the deep learning model, which is a common technical knowledge, so it is not described in detail here;

[0089] B4, evaluate the performance of the deep learning model based on the validation set data, when the prediction accuracy reaches the preset threshold (the preset threshold is set according to the actual early warning demand), the training is completed, otherwise continue to train and converge the deep learning model until the deep learning model training is completed;

[0090] S4, input real-time spatio-temporal feature data into the trained deep learning model, automatically identify the severe convective system in the radar data, and establish spatio-temporal correlation tracking to obtain the historical evolution characteristics of the severe convective system;

[0091] Severe convective system is a mesoscale weather system caused by strong vertical motion in the atmosphere, which has the characteristics of suddenness, short life history and severe weather phenomena, often accompanied by disastrous weather such as thunderstorm, gale, hail, short-time heavy rain and even tornado. This kind of system shows specific reflectivity structure and spatio-temporal evolution characteristics in radar data, and is the key object of attention for meteorological monitoring and forecasting.

[0092] As shown in Figure 4 , the flow of real-time identification of severe convective system is as follows:

[0093] C1, divide the real-time spatio-temporal feature data according to time interval or spatial range to form multiple data segments with independent spatio-temporal attributes;

[0094] C2, output the pixel-level segmentation boundary box of the severe convective system through forward propagation, and output the intensity score, such as the probability of belonging to thunderstorm;

[0095] The forward propagation output formula is:

[0096] , wherein, is the output intensity score value, is the ReLU activation function, is the weight matrix used for linear transformation of input features, is the input eigenvalue, is the bias vector used to adjust the linear transformation output;

[0097] C3. Use non-maximum suppression to remove overlapping detection frames and retain the optimal detection results. The formula for non-maximum suppression is:

[0098] ,in, To retain the test results, are the detected severe convective system areas, is the intersection and union ratio, is the threshold;

[0099] The spatiotemporal correlation tracking process is as follows:

[0100] D1. Extract the spatiotemporal characteristics of the centroid coordinates, area, and mean reflectivity of the bounding box for each detected severe convective system;

[0101] D2. Calculate the spatial distance and characteristic distance between severe convective systems at adjacent moments (the spatial distance uses the Euclidean distance, and the characteristic distance uses the reflectivity difference value), and construct a cost matrix. The cost matrix formula is:

[0102] ,

[0103] ,

[0104] ,

[0105] in, is the first A strong convective system, is the jth severe convective system at time t+1, and for Center of mass coordinates, and for The centroid coordinates of for The mean reflectivity, for The mean reflectivity, is the weight coefficient used to adjust the importance of spatial distance and feature distance in the cost matrix, for and The cost of matching. The smaller the cost, the stronger the correlation.

[0106] D3. Use the Hungarian algorithm to find the optimal match, establish the continuity relationship of the system ID, and handle the complex situations of splitting and merging;

[0107] The formula of the Hungarian algorithm is:

[0108] ,

[0109] The constraints are: ,

[0110] ,

[0111] ,when When and Match, otherwise not match;

[0112] in, is the element in the cost matrix, that is and The matching cost, A matching strategy variable with a value of 0 or 1, used to indicate whether and match, is the number of severe convective systems detected at time t, is the number of severe convective systems detected at adjacent time t.

[0113] The continuation relationship of the system ID is constructed by solving the cost matrix through the Hungarian algorithm to obtain the optimal matching pair of the strong convective system at adjacent times. and Match, then inherit The system ID establishes a continuation relationship. The ID allocation rules are as follows:

[0114] like Matches the only When Continued use ID;

[0115] like No matches found When The ID is temporarily retained or marked as dead;

[0116] like No matches found , a new ID is assigned.

[0117] When splitting, if a system At time t, it splits into multiple systems 、 , ..., then based on the minimum cost, the Hungarian algorithm can The best match for one association, the rest of the unmatched Assign a new ID, and further judge in combination with the splitting rule. The splitting rule is set as the area and reflectivity difference threshold. If and The feature distance is less than the threshold, it is marked as Split offspring, and the derived branch sharing the same ID;

[0118] When multiple systems 、 , … merge into a system at time t, the Hungarian algorithm cannot directly match due to a pair that does not meet the constraint condition. At this time, by calculating and the cost of each , the ID continuation is established by selecting the with the smallest cost, and the rest is marked as merged and extinct. If multiple and The spatial distance and feature distance are both less than the threshold, then is considered as the merging result of these , a new ID is assigned and its historical source is recorded;

[0119] Real-time identification and tracking of strong convective systems are achieved through deep learning models and spatiotemporal association algorithms. Forward propagation is used for feature extraction and classification. The cost matrix combines spatial and feature distances to construct a matching relationship. The Hungarian algorithm is used to solve the optimal matching to establish ID continuation. The processing of splitting and merging needs to combine threshold judgment and logical rules to ensure the accuracy of system evolution characteristics. These processes collectively support the full-chain analysis of strong convective systems from detection to historical tracing.

[0120] S5, according to the historical evolution characteristics of strong convective systems, predict the future impact area of strong convective systems;

[0121] The process of impact area prediction is as follows:

[0122] E1, extract the features of velocity, acceleration and shape change from the evolution trajectory of strong convective systems obtained by spatiotemporal association tracking to obtain historical characteristics;

[0123] E2, use self-attention mechanism to predict the position and range at future time based on historical characteristics;

[0124] U, where, are query, key, and value matrices, is the dimension of the key;

[0125] E3, error evaluation according to the predicted position and range, select the position and range with small error as the impact area, the prediction error evaluation calculation formula is:

[0126] ,in, is the prediction error, To predict the number of samples, is the true value, is the predicted value;

[0127] S6. Generate and output warning information based on the predicted severe convective system impact area;

[0128] When generating warning information, the predicted impact area and the location of the meteorological station are spatially superimposed and analyzed to determine which meteorological stations are located in the impact area. Different warning levels are divided according to the intensity and impact of the severe convective system. Finally, warning information including the impact area, warning level and disaster prevention measures is generated according to the warning level and issued to each affected area. The warning level is based on the severe convective weather warning level standard of the China Meteorological Administration. From low to high, the severity is blue, yellow, orange and red. Blue means that severe convective weather may occur with less impact. Yellow means that the possibility of severe convective weather increases and the impact tends to be obvious. Orange means that severe convective weather is about to occur with a greater impact and may cause certain disasters. Red means that severe convective weather has seriously affected the local area and may cause major disasters.

[0129] The deep learning model automatically learns the real-time spatiotemporal feature data of radar data, enabling the deep learning model to automatically identify and track severe convective systems. The deep learning model directly outputs the boundary and intensity of the severe convective system, and directly generates the mask of the severe convective system through a forward propagation, which effectively reduces the computational complexity of the radar data. Finally, based on the evolutionary characteristics of the severe convective system, the future impact area of ​​the severe convective system is predicted, and early warning information is generated and output based on the impact area. There is no human intervention in the entire process from data processing to early warning issuance, which improves the automation and efficiency of the early warning.

[0130] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A severe convective weather tracking and warning method based on radar data and deep learning, characterized by: The steps include: S1, real-time acquisition of raw radar data of radar reflectivity, radial velocity, and velocity spectrum width through lidar and spaceborne radar; S2. Obtaining raw radar data and performing cleaning and feature extraction to obtain real-time spatiotemporal feature data; S3. Build a deep learning model suitable for tracking severe convective weather and train it with historical data to enable it to automatically identify and track severe convective systems. S4. Input the real-time spatiotemporal feature data into the trained deep learning model to automatically identify severe convective systems in radar data and establish spatiotemporal correlation tracking to obtain the historical evolution characteristics of severe convective systems; S5. Predict the future impact areas of severe convective systems based on their historical evolutionary characteristics; S6. Generate and output warning information based on the predicted impact area of ​​the severe convective system.

2. The severe convective weather tracking and early warning method based on radar data and deep learning according to claim 1 is characterized by: The processing flow of raw radar data is as follows: A1. Use the median filter algorithm to filter out noise in the original radar data and fill in missing values ​​through interpolation. A2. Extract basic characteristics of radar reflectivity factor, radial velocity, and velocity spectrum width from filtered radar data, and simultaneously calculate characteristics related to severe convection, including storm top height, vertical liquid water content, mesocyclone characteristics, and hail index; A3. Normalize the strong convection-related features.

3. The severe convective weather tracking and early warning method based on radar data and deep learning according to claim 2 is characterized by: The deep learning model uses a spatiotemporal convolutional neural network combined with an attention mechanism to capture the spatial characteristics and temporal evolution of radar data. The attention mechanism includes spatial attention and channel attention. Spatial attention focuses on areas related to strong convection by calculating the spatial weights of feature maps. Channel attention automatically learns the importance weights of radar features.

4. The severe convective weather tracking and early warning method based on radar data and deep learning according to claim 3 is characterized by: The training process of the deep learning model is: B1. Collect historical radar data containing radar reflectivity factor, radial velocity, and velocity spectrum width, as well as manually annotated severe convective system boundaries and intensity levels to establish a data set; B2. Divide the data set into training set, validation set and test set in a ratio of 8:1:1; B3. Use the cross-entropy loss function to train the deep learning model based on the training set data, and use the Adam optimizer to dynamically adjust the learning rate to accelerate convergence; B4. Evaluate the performance of the deep learning model based on the validation set data. When the prediction accuracy reaches the preset threshold, the training is completed. Otherwise, continue to train and converge the deep learning model until the deep learning model training is completed.

5. The severe convective weather tracking and early warning method based on radar data and deep learning according to claim 4 is characterized in that: The process of real-time identification of severe convective systems is as follows: C1. Segment the real-time spatiotemporal feature data according to time intervals or spatial ranges to form multiple data segments with independent spatiotemporal attributes; C2, outputs the pixel-level segmentation bounding box of the strong convection system through forward propagation, and outputs the intensity score at the same time; C3. Use non-maximum suppression to remove overlapping detection frames and retain the optimal detection results.

6. The severe convective weather tracking and early warning method based on radar data and deep learning according to claim 5, characterized in that: The spatiotemporal correlation tracking process is as follows: D1. Extract the spatiotemporal characteristics of the centroid coordinates, area, and mean reflectivity of the bounding box for each detected severe convective system; D2. Calculate the spatial distance and characteristic distance between severe convective systems at adjacent moments and construct a cost matrix; D3. Use the Hungarian algorithm to solve the optimal match, establish the continuity relationship of the system ID, and handle the complex situations of splitting and merging.

7. The severe convective weather tracking and early warning method based on radar data and deep learning according to claim 6, characterized in that: In step D2, the cost matrix formula is: ， ， ， in, is the first A strong convective system, is the jth severe convective system at time t+1, and for Center of mass coordinates, and for The coordinates of the center of mass, for The mean reflectivity, for The mean reflectivity, is the weight coefficient used to adjust the importance of spatial distance and feature distance in the cost matrix, for and The cost of matching.

8. The severe convective weather tracking and warning method based on radar data and deep learning according to claim 7 is characterized by: The continuation relationship of the system ID is constructed by solving the cost matrix through the Hungarian algorithm to obtain the optimal matching pair of the strong convective system at adjacent times. and Match, then inherit The system ID of the system to establish a continuation relationship.

9. The severe convective weather tracking and early warning method based on radar data and deep learning according to claim 8, characterized in that: The process of impact area prediction is as follows: E1. Extract the characteristics of velocity, acceleration and shape changes from the evolution trajectory of the severe convective system obtained by spatiotemporal correlation tracking to obtain historical characteristics; E2. Use the self-attention mechanism to predict the location and range of future moments based on historical features; E3. Evaluate the error based on the predicted location and range, and select the location and range with the smallest error as the affected area.

10. The severe convective weather tracking and early warning method based on radar data and deep learning according to claim 9, characterized in that: When generating warning information, the predicted impact area and the location of the meteorological stations are spatially superimposed and analyzed to determine which meteorological stations are located in the impact area. Different warning levels are divided according to the intensity and impact of the severe convective system. Finally, according to the warning level, warning information including the impact area, warning level and disaster prevention measures is generated and released to each affected area.

Images