Radar echo image strong convective weather recognition method, electronic device and storage medium

Abstract

This application relates to the field of meteorological data processing technology, specifically to a method, electronic device, and storage medium for identifying severe convective weather using radar echo images. This application constructs a three-dimensional radar echo image of the target space through multi-elevation angle stereo scanning of radar. It then utilizes a three-dimensional convolutional neural network to extract and analyze multiple echo intensities, radial velocities, and velocity spectral widths contained in the three-dimensional radar echo image, outputting three-dimensional features such as echo hangs, bounded weak echo areas, and mesoscale cyclones upon which the confirmation of severe convective weather is based. This application can identify complex weather phenomena such as severe convection with high accuracy, providing technical support for meteorological station forecasters in nowcasting.

Description

Technical Field

[0001] This invention relates to the field of meteorological data processing technology, specifically to a method for identifying severe convective weather using radar echo images, an electronic device, and a storage medium. Background Technology

[0002] Radar echo images are visual images generated by processing electromagnetic waves emitted by radar and reflected signals from targets. Their core purpose is to visually display the location, intensity, structure, and movement trends of atmospheric targets such as precipitation and storms. Currently, the identification of severe convective weather in meteorological operations mainly relies on forecasters' subjective judgment of radar echo images. The accuracy rate for identifying high-impact weather systems such as supercell storms, squall lines, and mesoscale convective systems is not high, failing to meet the requirements of practical services. At the same time, the spatiotemporal characteristics of radar echo images are not effectively utilized, hindering the application of complex weather assessments. Summary of the Invention

[0003] The purpose of this application is to provide a method, electronic device and storage medium for identifying severe convective weather in radar echo images, which solves the technical problems of existing radar echo images relying on manual identification, having limited dimensional information, inaccurate subjective judgment, and inability to identify complex weather phenomena.

[0004] To solve the above-mentioned technical problems, this application adopts the following technical solution.

[0005] Firstly, this application provides a method for identifying severe convective weather using radar echo images, including:

[0006] Electromagnetic waves are emitted by a multi-elevation radar, and three-dimensional scanning data of the target space is constructed based on at least the radar echoes received at the first elevation angle and the second elevation angle. The three-dimensional scanning data includes at least one of the echo intensity, radial velocity, and velocity spectrum width at the corresponding elevation angle.

[0007] A three-dimensional convolutional neural network is used to extract at least one feature from the stereo scanning data, including echo overhang, bounded weak echo region, and mesoscale cyclone, and the severe convective weather conditions are determined based on the echo overhang, bounded weak echo region, and / or mesoscale cyclone.

[0008] In one alternative embodiment of the first aspect, the multi-elevation radar comprises 14.

[0009] In one alternative embodiment of the first aspect, the first elevation angle and the second elevation angle correspond to the lower layer of the target space;

[0010] The multi-elevation radar also includes a third elevation angle corresponding to the middle layer of the target space and a fourth elevation angle corresponding to the upper layer of the target space.

[0011] In one optional embodiment of the first aspect, the multi-elevation radar includes at least a first radar located at a first position and a second radar located at a second position, wherein the first position and the second position are adjacent positions facing the target space, so that the first radar and the second radar respectively obtain target space echo images from different perspectives.

[0012] In one alternative embodiment of the first aspect, the first radar and the second radar support time-division control.

[0013] In one optional embodiment of the first aspect, the first radar and the second radar respectively acquire target spatial echo images from different perspectives, including:

[0014] By default, the target space echo image obtained by the first radar is used for strong convection analysis.

[0015] In response to the echo overhang, bounded weak echo region and / or mesoscale cyclone being wholly or partially deviated from the field of view of the first radar, the second radar is triggered to acquire an echo image of the target space to supplement the analysis.

[0016] In one alternative embodiment of the first aspect, triggering the second radar to acquire an echo image of the target space to supplement the analysis includes:

[0017] The first elevation angle and the second elevation angle correspond to different height zones in the target space;

[0018] Based on the optimized regions of the echo overhang, bounded weak echo areas, and / or mesoscale cyclones, the echo images corresponding to the first elevation angle of the second radar and the echo images corresponding to the first elevation angle of the first radar are selected for analysis, and the echo images corresponding to the second elevation angle of the second radar are discarded. The first elevation angle corresponds to the height of the optimized region, and the second elevation angle deviates from the height of the optimized region.

[0019] In one alternative implementation of the first aspect, determining the severe convective weather conditions includes:

[0020] By using a convolutional long short-term memory network, the current stereo scanning data and stereo scanning data from adjacent time periods are analyzed in a time series to identify dynamic evolution features, including rapid echo enhancement, convective cell merging, and convective splitting, in order to output strong convective targets and confidence levels.

[0021] In one alternative embodiment of the first aspect, the stereoscopic scanning data of adjacent time periods includes 5 sets.

[0022] In an optional implementation of the first aspect, the output strong convection target and confidence level include:

[0023] The type of the severe convective target is determined based on the state of echo hangs, bounded weak echo regions, and / or mesoscale cyclones to predict future weather conditions.

[0024] In one alternative embodiment of the first aspect, the state of the echo overhang, the bounded weak echo region, and / or the mesoscale cyclone is determined by graphic appearance matching.

[0025] In one alternative embodiment of the first aspect, the type of the strong convective target includes supercell storms, squall lines, and mesoscale convective systems.

[0026] In an alternative implementation of the first aspect, determining the type of the severe convective target based on the state of echo overhang, bounded weak echo region, and / or mesoscale cyclone to predict future weather conditions includes:

[0027] The severe convective target is a supercell storm, and the forecast includes thunderstorms, large hail, strong gusts and / or short-duration heavy precipitation.

[0028] The strong convection target is a squall line, which is used to predict wind direction, temperature changes, and / or air pressure changes.

[0029] The severe convective target is a mesoscale convective system, and the forecast includes strong winds, short-duration heavy rainfall, and / or lightning.

[0030] In an alternative embodiment of the first aspect, the radar echo image severe convective weather identification method further includes:

[0031] Identify strong convective targets with a confidence level greater than a first threshold and mark them in the display area at the corresponding locations;

[0032] Calculate the extrapolated path of the strong convective target and mark it near the display area.

[0033] In an alternative implementation of the first aspect, calculating the extrapolated path of the strong convective target and marking it near the display area includes:

[0034] The multi-elevation radar includes at least a first radar located at a first position and a second radar located at a second position. By default, the first radar is used to obtain the target space echo image for strong convection analysis.

[0035] The extrapolated path of the strong convective target deviates from the direction of the first radar and approaches the direction of the second radar, triggering the second radar to obtain an echo image of the target space to correct the identification result of the strong convective target.

[0036] In an alternative implementation of the first aspect, calculating the extrapolated path of the strong convective target and marking it near the display area includes:

[0037] The warning level is determined based on the extrapolated speed of the strong convective target.

[0038] In one alternative implementation of the first aspect, determining the warning level includes:

[0039] The higher the warning level, the more conspicuous the color of the marker for the severe convective weather target.

[0040] In a second aspect, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the radar echo image severe convective weather identification method described in the first aspect.

[0041] Thirdly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the radar echo image severe convective weather identification method described in the first aspect.

[0042] Compared with existing technologies, this application constructs a three-dimensional radar echo image of the target space through multi-elevation angle stereo scanning of radar. It then utilizes a three-dimensional convolutional neural network to extract and analyze the multi-layered echo intensity, radial velocity, and velocity spectrum width contained in the three-dimensional radar echo image, outputting three-dimensional features such as echo hangs, bounded weak echo areas, and mesoscale cyclones upon which the confirmation of severe convective weather is based. This application can identify complex weather phenomena such as severe convection with high accuracy, providing technical support for meteorological station forecasters in nowcasting. Attached Figure Description

[0043] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description of the technical solution will be briefly introduced below. Obviously, the drawings described below are merely some examples recorded in this application, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0044] Figure 1 The flowcharts for radar echo image severe convective weather identification methods are shown in some examples of this application.

[0045] Figure 2 This is a schematic diagram of a multi-elevation, multi-station radar configuration in some examples of this application. Detailed Implementation

[0046] The present application will be described in detail below with reference to the accompanying drawings. However, the description is only a few examples recorded in the present application and does not limit the present application. Any changes in structure, method or function made by those skilled in the art based on these examples are included within the protection scope of the present application.

[0047] It should be noted that while the same labels or markers may be used in different examples, these do not represent an absolute structural or functional relationship. Furthermore, the use of terms such as "first," "second," etc., in the examples is merely for descriptive convenience and does not represent an absolute structural or functional distinction, nor should it be interpreted as indicating or implying relative importance or the number of corresponding objects. Unless otherwise specified, "at least one" in the description refers to one or more, and "more than one" refers to two or more.

[0048] Furthermore, when representing features, the character " / " can indicate an OR relationship between the preceding and following objects; for example, A / B can be represented as A or B. When representing operations, the character " / " can indicate a division relationship between the preceding and following objects; for example, C=A / B can be represented as C equal to A divided by B. Moreover, the use of "and / or" in different examples is merely to describe the relationship between the preceding and following objects, which can include three cases: for example, A and / or B can be represented as A existing alone, B existing alone, or both A and B existing simultaneously.

[0049] In some examples, such as Figure 1 As shown, the method for identifying severe convective weather using radar echo images specifically includes the following steps:

[0050] Step S1: Transmit electromagnetic waves using a multi-elevation radar. Construct a three-dimensional scanning data of the target space based on at least the received radar echoes from the first and second elevation angles. The three-dimensional scanning data includes at least one of the echo intensity, radial velocity, and velocity spectral width at the corresponding elevation angle. Weather radar can identify components in the target space by transmitting and receiving electromagnetic waves, based on phenomena such as reflection and scattering of electromagnetic waves in the target space. Specifically, when the radar transmits electromagnetic waves directionally into the air, the electromagnetic waves will be scattered when they encounter meteorological objects such as rain, hail, and snowflakes. Some of the scattered electromagnetic waves will return and be received by the radar as echoes. The corresponding radar echo image is generated by processing the echo intensity, time difference, frequency, etc., and converting it into a corresponding image. Furthermore, the parameters in the radar echo image implicitly contain corresponding feature information, which can be used to conduct detailed analysis of severe convective weather. Specifically, calculations can be performed based on echo intensity, radial velocity, velocity spectral width, etc. Echo intensity is the signal strength received by the radar antenna after the electromagnetic waves emitted by the radar encounter precipitation particles (such as raindrops, hail, and snowflakes) in the atmosphere. It primarily reflects the concentration, size, and distribution of precipitation particles, and is typically measured in dBZ (decibels per second). In this example, echo intensity is a fundamental parameter for identifying strong convective targets. Higher echo intensity usually corresponds to stronger precipitation and more intense convection, and it is also a core carrier of three-dimensional features such as echo overhang and bounded weak echo areas. Radial velocity refers to the radial velocity of a specific target (such as precipitation particles or airflow) relative to the radar antenna, i.e., the velocity along the direction of the radar's emitted electromagnetic waves. Positive numbers indicate that the target is moving away from the radar, while negative numbers indicate that the target is moving towards the radar. In this system, it is used to capture airflow motion characteristics such as mesoscale cyclones and backflow jets, and is a key parameter for dynamic spatial features in three-dimensional feature extraction. By observing differences in radial velocity distribution, key signals of strong convection such as airflow rotation, convergence, and divergence can be identified. Velocity spectral width reflects the dispersion of the radial velocity of a specific target (such as precipitation particles or airflow) near the same detection point within the radar's detection range. The wider the velocity spectrum, the more turbulent and uneven the airflow in the region. It is an important parameter for judging the intensity of airflow inside a severe convective storm. In this example, it can be analyzed in conjunction with echo intensity and radial velocity to improve the accuracy of three-dimensional feature extraction of severe convective targets and help identify the core region of a severe convective storm.

[0051] Furthermore, the radar employs multi-elevation angles to scan the target space at different heights layer by layer. Correspondingly, this multi-elevation angle is achieved by forming multiple independent and partially overlapping beams within the same vertical plane. For example, it can scan the sky of the target space layer by layer at 14 elevation angles, such as 0.5 degrees, 1.5 degrees, 2.4 degrees, etc., forming a three-dimensional scan data. A higher elevation angle indicates a higher altitude in the target space, thus increasing the layered complexity of the radar echo image, especially in reconstructing the three-dimensional layered features of strong convective targets. Specifically, the target space can also be divided according to altitude, specifically based on the internal characteristics of different types of strong convective targets. For example, it can be divided into three layers: high-level, low-level, and a middle layer in between, each configured with a corresponding elevation angle for monitoring. This not only obtains three-dimensional layered features but also labels the inherent attributes of the features with approximate altitude tags. Optionally, the multi-elevation angle radar can be implemented using a phased array antenna. By electronically controlling the phase of each element in the antenna array, it achieves rapid and flexible scanning of the beam in the elevation direction without mechanical rotation, making switching between elevation angles more efficient. Optionally, such as Figure 2 As shown, the radar can include radars 201 and 202 from different locations. In more examples, radars from even more locations can be used for combined monitoring. Although radars 201 and 202 face the same target space, their relative positions to the target space differ. Therefore, when scanning meteorological targets in the target space, they may obtain different scanning results. When they can enhance each other, coordinated monitoring can improve the accuracy of identification. Accordingly, radars 201 and 202 communicate with each other to achieve time-division control, reducing mutual interference through time division. More importantly, a default scanning radar and a backup scanning radar can be configured. As mentioned above, by creating a three-dimensional network mosaic of these multi-elevation, multi-station radar data to form a complete three-dimensional atmospheric data volume, instead of analyzing only the meteorological conditions in the layer near the ground, the accuracy of identifying strong convective targets is greatly improved, which will be detailed below.

[0052] Step S2: Utilize a three-dimensional convolutional neural network to extract at least one feature from the stereoscopic scanning data: echo overhang, bounded weak echo region, and mesoscale cyclone. Determine the severe convective weather conditions based on the echo overhang, bounded weak echo region, and / or mesoscale cyclone. In Step S1, the stereoscopic scanning data obtained through multiple elevation angles contains a lot of implicit information, especially the echo intensity, radial velocity, and velocity spectrum width at different elevation angles, which have a strong correlation with severe convection. The identification process involves establishing a related mapping relationship between the input of the stereoscopic scanning data and the output of the identification result. In this example, an artificial intelligence model is introduced to assist in data analysis, which can eliminate the need for complex logical analysis of the internal mechanisms and directly deduce the parameter relationship between the input and output through data training. In this example, a parameter-trained 3D convolutional neural network (CNN) is employed. The core characteristic of 3D CNNs compared to other neural networks is their use of local connections, weight sharing, and hierarchical feature extraction mechanisms designed for spatial data such as images. This makes them more efficient and better at generalizing when processing high-dimensional grid-structured data. Unlike traditional neural networks that need to flatten images into one-dimensional vectors, thus destroying spatial information, CNNs directly process 3D tensors, preserving the original spatial structure. This is particularly useful in this application, where features such as echo overhangs, bounded weak echo regions, and mesoscale cyclones need to be extracted from different echo layers (high, medium, and low). Furthermore, due to the smaller number of parameters and the more rational structure, the output is faster and less prone to overfitting. In the specific example, the 3D CNN model receives stereo scanning data input from various time periods, preserving complete 3D spatial information. Multi-scale 3D convolutional kernels are used to process the stereo scanning data for CNN output. Optionally, the convolutional neural network can be designed with multiple layers of convolution and pooling layers. Shallow convolution can extract basic spatial features such as echo intensity, radial velocity, and velocity spectrum width, including local echo intensity differences and velocity gradients. Deep convolution can capture three-dimensional spatial correlation information across elevation angles and distances by fusing features from three-channel data through the spatial correlation of three-dimensional convolution kernels. It focuses on mining the three-dimensional structural features of echo overhangs, bounded weak echo regions, and mesoscale cyclones. Then, through fully connected layers or attention mechanisms, the three-dimensional structural features extracted from the three channels are jointly fused to strengthen the recognition weight of the iconic three-dimensional structures of strong convection targets, such as the tilt features of echo overhangs, weak echo boundary features, and the rotational speed of mesoscale cyclones. By extracting features such as echo overhangs, bounded weak echo areas, and mesoscale cyclones from the 3D scanning data, these features can be further used to identify strong convective targets. Compared to existing methods that only use the lowest elevation angle plane radar to scan and analyze meteorological conditions, this method additionally captures three-dimensional structural features such as echo overhangs, bounded weak echo areas, and mesoscale cyclones, thus enabling the identification of more detailed meteorological conditions.

[0053] Among them, echo hanging is a typical feature of the vertical structure of strong convective storms (such as supercell storms). It refers to the strong mid-to-high-level echoes (corresponding to precipitation particles) in radar echoes extending obliquely towards the storm inflow side, forming a "hanging" shape. It is often accompanied by a weak echo area below. It is a direct manifestation of precipitation particles being carried to the mid-to-high levels under the influence of strong updrafts and is an important identification mark of strong convection (such as hail). Bounded weak echo areas, also known as domes, are one of the core features of supercell storms. They refer to weak or no echo areas appearing inside or below the strong echo area, surrounded by mid-level echo hanging. Essentially, they are areas of concentrated strong updrafts, containing only cloud particles and no obvious precipitation particles. Persistent bounded weak echo areas are usually accompanied by mesocyclones and are a key signal of the rotation of strong updrafts. Mesoscale cyclones are deep, continuously rotating updraft systems, which are the essential characteristics of supercell storms. They typically have a horizontal scale of several to tens of kilometers and a deep vertical extension. Obvious cyclonic rotation pairs can be identified on radar radial velocity maps. They provide the driving force for the continuous development of severe convective storms and are also important breeding grounds for extreme disasters such as tornadoes and large hail.

[0054] Furthermore, the attributes of echo overhang, bounded weak echo regions, and mesoscale cyclones, as intermediate parameters, have a strong internal correlation with the identification of strong convective targets and are more direct than the parameters in stereo scanning data. Therefore, data analysis can be performed on parameters such as echo overhang, bounded weak echo regions, and mesoscale cyclones. Accordingly, historical stereo scanning data will be introduced on the basis of transverse cross-sectional data for time-series analysis. In this example, a convolutional long short-term memory network is used. The convolutional long short-term memory network has both the spatial convolution characteristics of convolutional neural networks and the temporal memory capability of LSTM (Long Short-Term Memory). It can not only process the spatial structure of radar echoes but also remember the historical changes of multiple consecutive frames, thereby realizing spatiotemporal sequence data analysis. Its core lies in introducing the convolution operation in convolutional neural networks into the gating structure of LSTM, so that the model can not only capture long-term temporal dependencies but also effectively extract local spatial features, unlike LSTM which needs to flatten two-dimensional data into vectors when processing image sequences, thus destroying the spatial structure. In a specific example, multi-elevation angle stereo scan data from adjacent time periods can be merged and input. For instance, with six scanning time periods, executed according to a radar scanning cycle of 5-6 minutes, the current stereo scan data is correlated and analyzed with the stereo scan data from the previous five adjacent time periods. This allows for the analysis of a meteorological evolution process spanning approximately half an hour, identifying dynamic evolutionary features including rapid echo enhancement, convective cell merging, and convection splitting, to output strong convective targets and their confidence levels. Correspondingly, for each of the six time periods of stereo scan data, a spatial feature map can be extracted using a 3D convolutional neural network, forming a temporal sequence which is then input into a convolutional long short-term memory network. Internally, convolution operations are performed to preserve the spatial structure, and a gating mechanism is used to memorize the echo state changes from previous time periods. Finally, the temporal hidden features are output, thus fusing six static images into a dynamically evolving spatiotemporal feature. The fused features are fed into the fully connected layer and the activation function of logistic regression, and finally output the confidence score (confidence score = f(spatial features ⊕ temporal features). The confidence score can be the credible probability that the corresponding region is a strong convective target. That is, under the joint support of the current structural features and the historical evolution trend, the output credible probability meets the requirements of a strong convective target.

[0055] Optionally, for identified severe convective targets, their types will also be identified, specifically including supercell storms, squall lines, and mesoscale convective systems. Among these, supercell storms are the most violently developing and structurally complete local severe convective systems. Their core characteristic is a persistent and deep mesocyclone (deep, continuously rotating updrafts). They often exist in isolation from other thunderstorms, with a horizontal scale of approximately 20-40 kilometers and a vertical scale of up to 18 kilometers. Their lifespan is typically several hours, making them one of the most destructive types of severe convective weather. They are often accompanied by strong lightning, large hail, strong gusts, and short-duration heavy rainfall. Under favorable environmental conditions, they can develop into tornadoes, with an impact range reaching up to 32 kilometers. They cause severe damage to buildings, crops, power lines, and transportation, making them a key focus of severe convective disaster warnings. In radar echo images, the plan position indicator (PPI) shows a single-celled elliptical structure with hook-shaped echoes and inflow gaps visible at lower levels. The echo intensity slopes towards the inflow side from lower to higher levels. The range height indicator (RHI) shows a clearly defined bounded weak echo zone and echo overhang structures. The strongest echo is located to the left of the bounded weak echo zone and is often accompanied by strong echo characteristics associated with large hailstorms, such as three-body scattering echoes. On the radial velocity map, a clear cyclonic rotation can be identified in the middle and lower levels, while the upper levels exhibit obvious storm top divergence characteristics.

[0056] A squall line is a linear, rapidly moving band of strong convective storms that advances like a wall. Accompanying the rapid advance of severe convective weather, it causes drastic weather changes and a sharp increase in wind speed upon its passage, making it one of the main causes of widespread severe convective disasters in mid-latitude regions during summer. Before its passage, warm and humid southerly or southerly winds prevail, but after passing, they quickly turn into cold and dry northerly or northerly winds, with the wind direction change angle usually exceeding 90 degrees. It also manifests as a sudden drop in temperature of 5-10 degrees Celsius within 1-2 hours, with some areas experiencing a drop of more than 15 degrees Celsius. Air pressure also rises rapidly, forming a significant "barosurgical surge." Simultaneously, it is accompanied by short-duration heavy rainfall and dense thunderstorms. Some squall lines can produce hailstones with a diameter of ≥20mm. Some strong squall lines, especially those developing into mesoscale convective systems, can breed tornadoes, which often occur at the leading edge of the squall line or the vortex at the end of the bow echo. While less frequent than isolated supercell storms, they are extremely destructive.

[0057] Mesoscale convective systems are organized collections of convective storms, typically forming in environments with moderate to strong wind shear at the lower 2-3 km levels. They often evolve from weakly organized squall lines or supercells. Their movement is primarily controlled by internal airflow, exhibiting an outward-spreading trend and gradually dissipating in the later stages of development. Their hazards include strong straight-line winds, with gusts reaching gale or strong wind levels, easily causing trees to fall, billboards to collapse, and buildings to be damaged. They are also accompanied by short-duration heavy rainfall and lightning. Occasionally, tornadoes or gust fronts may form due to terminal vortices, but the probability of tornadoes is lower than that of supercell storms, and the affected area is usually more extensive. In radar echo images, they appear as a clear bow shape on a plane, with the bulge pointing in the direction of storm movement, hence the name "bow echo." Unlike supercells, bow echoes lack obvious mesocyclonic and hook-shaped echo characteristics; the bow-shaped bulge of the linear echo band is the core identifying feature.

[0058] Accordingly, since different types of severe convective targets will have different echo overhangs, bounded weak echo regions, and mesoscale cyclones, the type of severe convective target can be determined by matching these features. For example, feature maps of echo overhangs, bounded weak echo regions, and / or mesoscale cyclones of different types of severe convective targets can be stored in advance and determined through similarity analysis. Preferably, this feature map can also be fed into an artificial intelligence model for training, and then directly input into the artificial intelligence model to output the type result. Since different severe convective targets are accompanied by different weather conditions, future weather conditions can also be predicted based on the type judgment results. Referring to the above analysis of severe convective target types such as supercell storms, squall lines, and mesoscale convective systems, specifically, if the severe convective target is a supercell storm, lightning, large hail, strong gusts, and / or short-duration heavy precipitation can be predicted; if the severe convective target is a squall line, wind direction, temperature changes, and / or air pressure changes can be predicted; and if the severe convective target is a mesoscale convective system, gale-force winds, short-duration heavy precipitation, and / or lightning can be predicted.

[0059] In some examples, after obtaining the confidence level of a severe convective target, it is compared with a first threshold, which measures a high probability of success. When the output confidence level is greater than the first threshold, the severe convective target is marked, and its corresponding position in the display area is highlighted. Optionally, the predicted extrapolated path of the severe convective target's movement is also displayed near the corresponding location to visually show the future development trend of the severe convective target. The predicted extrapolated movement speed is also mapped to the warning level and the marking color. Specifically, the faster the extrapolated movement speed, the higher the warning level, prompting relevant personnel to pay attention to information dissemination and disaster prevention measures; the higher the warning level, the more conspicuous the marking color, thereby helping relevant personnel to promptly detect important situations.

[0060] In some examples, severe convective weather identification also employs a closed-loop optimization mechanism. For instance, after identifying a specific severe convective target based on confidence levels, the parameters are adjusted according to the identification results. In this example, multiple radar stations will be configured; taking the first and second radars as examples... Figure 1 In this example, the stereo scanning data acquired in step S1 is transmitted and received via electromagnetic waves by the first radar by default. After extracting echo overhangs, bounded weak echo regions, and / or mesoscale cyclones in step S2, the completeness or distortion of these features can be determined by matching the corresponding feature maps. These distortions may be caused by deviations in the field of view of the first radar. Optionally, the relative direction between the final determined location of the strong convective target and the location of the first radar can be used to determine whether all or part of the echo overhangs, bounded weak echo regions, and / or mesoscale cyclones deviate from the field of view of the first radar. If not, the default state is maintained. If a deviation occurs, the second radar is triggered to scan the target space, thereby obtaining the corresponding layered stereo scanning data. Preferably, the second radar also has the capability to scan at different elevation angles. In this example, not all elevation angles are scanned. Instead, an appropriate elevation angle is selected based on the optimized area height of the echo overhangs, bounded weak echo regions, and / or mesoscale cyclones. For example, if the echo overhang needs optimization, scanning at mid-to-high-altitude elevation angles can be triggered, and the data is then merged into the specified model for calculation. In some examples, once the location and extrapolated movement path of a strong convective target are determined, a nearby radar can be triggered to scan based on the initially determined location of the target, leveraging the advantages of close-range scanning and identification. For instance, if the extrapolated movement path of the strong convective target deviates from the direction of the first radar (the one that defaults to scanning) and instead moves closer to the direction of the second radar, the second radar will be triggered to obtain an echo image of the target space to correct the identification result of the strong convective target. If the corrected location of the strong convective target changes, the optimal radar will be continuously selected for iterative optimization.

[0061] This application integrates radar volume scan data from multiple stations and elevation angles, eliminating data redundancy and bias, and forming a spatially continuous and parameter-complete three-dimensional data volume. It has a high accuracy rate in identifying three types of high-impact strong convection: supercell storms, squall lines, and mesoscale convective systems, providing technical support for intelligent radar image recognition to help meteorological station forecasters with their nowcasting operations.

[0062] Based on the above examples, the technical solutions involved in this application can be directly embodied in hardware, software modules executed by a control unit, or a combination of both, i.e., one or more steps and / or combinations of one or more steps. These can correspond to various software modules in a computer program flow, or to various hardware modules, such as ASICs (Application Specific Integrated Circuits), FPGAs (Field-Programmable Gate Arrays), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or any suitable combination thereof. For ease of description, the above description divides the functions into various modules. Of course, in implementing this application, the functions of each module can be implemented in one or more software and / or hardware components.

[0063] Through the above description of examples, those skilled in the art can clearly understand that this application can be implemented using software plus the necessary general-purpose hardware platform. Based on this understanding, the technical solution involved in this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This software is executed by a microcontroller unit and, depending on the required configuration, can include one or more microcontroller units of any type, including but not limited to microcontroller units, microcontrollers, DSPs (Digital Signal Processors), or any combination thereof. The software is stored in memory, such as volatile memory (e.g., random access memory), non-volatile memory (e.g., read-only memory, flash memory), or any combination thereof.

[0064] In summary, this application constructs a three-dimensional radar echo image of the target space through multi-elevation angle stereo scanning of radar. It then utilizes a three-dimensional convolutional neural network to extract and analyze multiple echo intensities, radial velocities, and velocity spectral widths contained within the three-dimensional radar echo image, outputting stereoscopic features such as echo hangs, bounded weak echo areas, and mesoscale cyclones as the basis for confirming severe convective weather. This application can identify complex weather phenomena such as severe convection with high accuracy, providing technical support for meteorological forecasters in nowcasting.

[0065] It should be understood that although this specification includes some examples, none of these examples constitutes a single, independent technical solution. This descriptive style is merely for clarity. Those skilled in the art should consider this specification as a whole, and the technical solutions in the examples can be appropriately combined to form other embodiments that can be understood by those skilled in the art.

[0066] The detailed descriptions listed above are merely specific descriptions of feasible implementation methods of this application, and are not intended to limit the scope of protection of this application. All equivalent implementation methods or modifications that do not depart from the teachings of this application should be included within the scope of protection of this application.

Claims

1. A method for identifying severe convective weather using radar echo images, characterized in that, include: Electromagnetic waves are emitted by a multi-elevation radar, and three-dimensional scanning data of the target space is constructed based on at least the radar echoes received at the first elevation angle and the second elevation angle. The three-dimensional scanning data includes at least one of the echo intensity, radial velocity, and velocity spectrum width at the corresponding elevation angle. A three-dimensional convolutional neural network is used to extract at least one feature from the stereo scanning data, including echo overhang, bounded weak echo region, and mesoscale cyclone, and the severe convective weather conditions are determined based on the echo overhang, bounded weak echo region, and / or mesoscale cyclone.

2. The method for identifying severe convective weather using radar echo images according to claim 1, characterized in that, The multi-elevation radar comprises 14 units.

3. The method for identifying severe convective weather using radar echo images according to claim 1, characterized in that, The multi-elevation radar includes at least a first radar located at a first position and a second radar located at a second position. The first position and the second position are adjacent positions facing the target space, so that the first radar and the second radar can obtain target space echo images from different perspectives.

4. The method for identifying severe convective weather using radar echo images according to claim 3, characterized in that, The first radar and the second radar respectively obtained target space echo images from different perspectives, including: By default, the target space echo image obtained by the first radar is used for strong convection analysis. In response to the echo overhang, bounded weak echo region and / or mesoscale cyclone being wholly or partially deviated from the field of view of the first radar, the second radar is triggered to acquire an echo image of the target space to supplement the analysis.

5. The method for identifying severe convective weather using radar echo images according to claim 1, characterized in that, The circumstances under which severe convective weather is determined include: By using a convolutional long short-term memory network, the current stereo scanning data and stereo scanning data from adjacent time periods are analyzed in a time series to identify dynamic evolution features, including rapid echo enhancement, convective cell merging, and convective splitting, in order to output strong convective targets and confidence levels.

6. The radar echo image severe convective weather identification method according to claim 5, characterized in that, The output strong convection targets and confidence levels include: The type of the severe convective target is determined based on the state of echo hangs, bounded weak echo regions, and / or mesoscale cyclones to predict future weather conditions.

7. The radar echo image severe convective weather identification method according to claim 5, characterized in that, The radar echo image severe convective weather identification method also includes: Identify strong convective targets with a confidence level greater than a first threshold and mark them in the display area at the corresponding locations; Calculate the extrapolated path of the strong convective target and mark it near the display area.

8. The method for identifying severe convective weather using radar echo images according to claim 7, characterized in that, The calculation of the extrapolated path of the strong convective target and its marking near the display area includes: The multi-elevation radar includes at least a first radar located at a first position and a second radar located at a second position. By default, the first radar is used to obtain the target space echo image for strong convection analysis. The extrapolated path of the strong convective target deviates from the direction of the first radar and approaches the direction of the second radar, triggering the second radar to obtain an echo image of the target space to correct the identification result of the strong convective target.

9. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the radar echo image severe convective weather identification method according to any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the radar echo image severe convective weather identification method according to any one of claims 1-8.

Images