Hail point and quantity recognition and early warning method based on radar and lightning data analysis

Abstract

The application discloses a hail point and quantity identification early warning method based on radar and lightning data analysis, relates to the technical field of radar meteorology, and comprises the following steps: calculating the posterior probability of each grid point belonging to the graupel phase state in a three-dimensional space grid, extracting a hail risk field, marking cloud lightning discharge points falling in the hail risk field as effective anchor points, calculating the edge weight value of each undirected edge in the undirected graph based on the posterior probability, performing weight propagation and updating to generate a space weight field, calculating a modified vertical cumulative liquid water content of each horizontal position, and determining a hail early warning drop area and a pre-hail early warning grade index. Through the construction of a three-dimensional space dynamic network model based on a lightning triggering mechanism and undirected graph weight propagation, the application realizes the deep space topological association of a hail risk field and lightning discharge points, introduces the space normalization weight generated by graph propagation into the calculation of a vertical cumulative liquid water content, accurately removes the liquid water interference in a non-hail area, and realizes the quantitative correction of a hail intensity.

Description

Technical Field

[0001] This invention relates to the field of radar meteorology technology, and in particular to a method for targeted and quantitative identification and early warning of hail based on radar and lightning data analysis. Background Technology

[0002] In recent years, hail has become increasingly sudden and destructive. In the detection of severe convection, dual-polarization radar can identify the phase state of water precipitates and the three-dimensional microphysical structure within clouds by acquiring parameters such as differential reflectivity. Meanwhile, three-dimensional lightning location data can reflect the charge activity inside storms, and the cloud lightning discharge point is usually closely related to the hail growth area. These two types of detection data provide an important data foundation for analyzing hail cloud evolution and hail warning from microphysical and electrical dimensions.

[0003] Currently, Chinese invention patent CN120233320A discloses a hail warning method and system based on lightning jump and support vector machine, including: acquiring combined reflectivity factor, vertical integral liquid water content, and echo top height data through radar network data; identifying strong convection echo areas and calculating lightning temporal variability and its standard deviation; screening potential hail events based on lightning jump conditions; further extracting the maximum vertical integral liquid water content (VILmax), jump increment (GVIL), and density (VILD) as feature parameters; and inputting them into a support vector machine classification model trained based on historical samples for secondary judgment. While this approach effectively reduces the false alarm rate of traditional 2σ algorithms and improves the accuracy of hail warnings, current technologies often rely on single reflectivity or vertical cumulative liquid water content thresholds for identification, without considering the spatial evolution of particle phases, resulting in a high false alarm rate. Methods incorporating lightning data often remain at the level of simple frequency or location references, failing to explore the three-dimensional spatial topological relationship between lightning discharge points and the hail risk field, and lacking dynamic network modeling. Furthermore, conventional vertical cumulative liquid water content algorithms do not consider the uneven distribution of hail spatial risks, and the use of equal-weighted accumulation makes it impossible to accurately separate interference from non-hail areas, thus failing to meet the high-precision, fixed-point, and quantitative early warning requirements. Summary of the Invention

[0004] The technical problem solved by this invention is that: the identification of hail relies on a single reflectivity or vertical cumulative liquid water content threshold, without considering the spatial evolution of particle phases, resulting in a high false alarm rate; the methods for introducing lightning data mostly remain at the level of simple frequency or location reference, without exploring the three-dimensional spatial topological relationship between lightning discharge points and hail risk fields, lacking dynamic network modeling; the conventional vertical cumulative liquid water content algorithm does not consider the non-uniformity of hail spatial risk distribution, and the use of equal weight accumulation makes it impossible to accurately separate interference from non-hail areas, making it difficult to meet the requirements of high-precision fixed-point quantitative early warning.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a method for targeted and quantitative identification and early warning of hail based on radar and lightning data analysis, comprising the following steps: Step S1: Obtain the registered dual-polarization radar three-dimensional volume scan data and three-dimensional lightning location data, calculate the posterior probability of each grid point in the three-dimensional spatial grid belonging to the grenade phase state, and extract the hail risk field based on the posterior probability. Step S2: Extract cloud flash discharge points from the three-dimensional lightning location data as trigger events, and mark the trigger events that fall into the hail risk field as valid anchor points; Step S3: Take the effective anchor points and grid points other than the effective anchor points in the hail risk field as nodes of the undirected graph, calculate the edge weights of each undirected edge in the undirected graph based on the posterior probability, and perform weight propagation and update on each node in the undirected graph to generate a spatial weight field. Step S4: Calculate the liquid water content of each grid point in the three-dimensional volume scan data of the dual polarization radar, and multiply it with the normalized weight value of the corresponding grid point in the spatial weight field to obtain the corrected liquid water content. Then, sum the corrected liquid water content of each grid point at the same horizontal position along the vertical direction to obtain the corrected vertical cumulative liquid water content at each horizontal position. Step S5: Determine the hail warning area and pre-hail warning level index based on the modified vertical cumulative liquid water content, and generate warning information.

[0006] As a preferred embodiment of the hail location-based quantitative identification and early warning method based on radar and lightning data analysis described in this invention, step S1 specifically includes: The target detection area is divided into three-dimensional grids using a preset spatial resolution to obtain a three-dimensional spatial grid matrix; The smallest spatial unit in the three-dimensional spatial grid matrix is ​​taken as the grid point; Acquire three-dimensional volume scan data and three-dimensional lightning location data from dual-polarization radar, convert the polar coordinates of the three-dimensional volume scan data from dual-polarization radar to geodetic coordinates, and convert the three-dimensional lightning location data to geodetic coordinates that are the same as the three-dimensional spatial grid matrix; The transformed geodetic coordinates are mapped to the grid points to generate the registered dual-polarization radar 3D volume scan data and 3D lightning location data. The differential reflectivity, differential phase shift rate, correlation coefficient, and horizontal reflectivity in the registered dual-polarization radar three-dimensional volume scan data are used as dual-polarization parameters. The dual polarization parameters are input into a preset fuzzy logic classifier to calculate the membership degree of each grid point to the grenade phase state; The membership degree is normalized to obtain the posterior probability of each grid point; Hail risk field is extracted based on the posterior probability.

[0007] As a preferred embodiment of the hail location-based quantitative identification and early warning method based on radar and lightning data analysis described in this invention, the step of extracting the hail risk field based on the posterior probability specifically includes: Grid points with a posterior probability greater than a preset first probability threshold are selected as candidate grid points. Traverse the three-dimensional spatial mesh matrix and extract the adjacent candidate grid points that have a common surface in space for each candidate grid point; Merge adjacent candidate grid points into a connected domain; Connected regions with a volume greater than a preset volume threshold are designated as hail risk fields.

[0008] As a preferred embodiment of the hail location-based quantitative identification and early warning method based on radar and lightning data analysis described in this invention, step S2 specifically includes: Extract the three-dimensional coordinates of cloud-to-cloud lightning discharge points from the registered three-dimensional lightning location data, and use the cloud-to-cloud lightning discharge points as trigger events; The longitude, latitude, and altitude ranges of each grid point in the hail risk field are used as the three-dimensional coordinate ranges. Extract triggering events whose three-dimensional coordinates are within the range of the three-dimensional coordinates, and use them as target triggering events. Also, use the grid point where the three-dimensional coordinates of the target triggering event are located as the target grid point. If the posterior probability of the target grid point is greater than a preset second probability threshold, then the target triggering event corresponding to the target grid point is marked as a valid anchor point.

[0009] As a preferred embodiment of the hail location-based quantitative identification and early warning method based on radar and lightning data analysis described in this invention, step S3 specifically includes: Each effective anchor point is designated as an anchor point node, and the remaining grid points in the hail risk field, excluding the anchor point nodes, are designated as ordinary nodes. Any two nodes that share a common surface in the space within the hail risk field are considered as adjacent nodes; An undirected graph is obtained by establishing undirected edge connections between the adjacent nodes. The preset first initial value is used as the initial anchor weight of each anchor node in the undirected graph; Use the preset second initial value as the initial ordinary weight of each ordinary node in the undirected graph; The edge weights of each undirected edge in the undirected graph are calculated based on the posterior probability. Based on the edge weights, the weights of each node in the undirected graph are propagated and updated to obtain the target weights of each node. The target weights are subjected to maximum and minimum normalization to obtain normalized weight values; A spatial weight field is generated based on the normalized weight values ​​of each node in the undirected graph.

[0010] As a preferred embodiment of the hail location-based quantitative identification and early warning method based on radar and lightning data analysis described in this invention, the step of calculating the edge weights of each undirected edge in the undirected graph based on the posterior probability specifically includes: Traverse each undirected edge in the undirected graph and multiply the posterior probabilities of the adjacent nodes connected by the current undirected edge to obtain the product result; Use the product as the edge weight of the current undirected edge.

[0011] As a preferred embodiment of the hail location-based quantitative identification and early warning method based on radar and lightning data analysis described in this invention, the step of weight propagation and updating of each node in the undirected graph based on the edge weights specifically includes: Based on the edge weights, the initial anchor weights of the anchor nodes are iteratively passed to the adjacent ordinary nodes. In each iteration, calculate the absolute value of the weight difference before and after the update of each node at the current iteration number; The iteration is stopped when the absolute value of the weight difference is less than the preset convergence threshold, or when the current iteration number is greater than or equal to the preset maximum iteration number. When the absolute value of the weight difference is greater than or equal to the preset convergence threshold, and the current iteration number is less than the preset maximum iteration number, the next iteration continues. Extract the current weight of each node in the undirected graph at the time of stopping iteration, and use it as the target weight; The iterative transmission specifically includes: For a regular node in the undirected graph, the edge weights of each undirected edge connected to the regular node are multiplied by the current weights of the corresponding adjacent nodes and then summed. The sum is divided by the sum of the edge weights of all undirected edges connected to the regular node to obtain a quotient, which is then used as the updated weight of the regular node. For the anchor nodes in the undirected graph, the initial anchor weight of the current anchor node remains unchanged.

[0012] As a preferred embodiment of the hail location-based quantitative identification and early warning method based on radar and lightning data analysis described in this invention, step S4 specifically includes: Read the horizontal reflectivity factor of each grid point in the three-dimensional volume scan data of the dual-polarization radar; Substitute the horizontal reflectivity factor into the preset empirical formula for water content to calculate the liquid water content at each grid point. The modified liquid water content of each grid point is obtained by multiplying the liquid water content of each grid point with the normalized weight value of the corresponding grid point in the spatial weight field. The modified liquid water content at each grid point at the same horizontal position is summed along the vertical direction to obtain the modified vertical cumulative liquid water content at each horizontal position.

[0013] As a preferred embodiment of the hail location-based quantitative identification and early warning method based on radar and lightning data analysis described in this invention, step S5 specifically includes: The horizontal position where the modified vertical cumulative liquid water content is greater than the preset water content threshold is used as the warning position; Merge adjacent warning locations that share a common edge into a two-dimensional connected region; Calculate the area of ​​each two-dimensional connected region; Extract the outer edge contour of two-dimensional connected regions whose area is greater than a preset area threshold; The horizontal area enclosed by the outer edge outline is designated as the hail warning area; The hail warning level index is determined based on the value of the modified vertical liquid water content, and warning information including the hail warning area and the hail warning level index is generated.

[0014] As a preferred embodiment of the hail location-based quantitative identification and early warning method based on radar and lightning data analysis described in this invention, the generation of early warning information including the hail warning area and the hail warning level index specifically includes: Traverse all horizontal positions within the hail warning area and extract the maximum value of the corrected vertical cumulative liquid water content corresponding to each horizontal position; Based on a preset mapping table, the hail diameter range corresponding to the maximum value is found in the preset mapping table; Output the level label corresponding to the hail diameter range as a hail warning level indicator; Obtain the set of latitude and longitude coordinates of the outer edge contour of the hail warning area, and obtain the warning color code and warning text description corresponding to the hail warning level index; The latitude and longitude coordinates, the warning color code, and the warning text description are encapsulated to generate warning information.

[0015] The beneficial effects of this invention are as follows: By constructing a three-dimensional spatial dynamic network model based on lightning triggering mechanism and undirected graph weight propagation, this invention achieves a deep spatial topological association between hail risk field and lightning discharge point, effectively overcoming the problem of high false alarm rate caused by traditional methods relying only on a single threshold or simple lightning frequency reference. It significantly improves the accuracy of hail core area identification. By introducing the spatial normalized weight generated by graph propagation into the calculation of vertical cumulative liquid water content, it breaks through the limitations of conventional algorithms' equal weight accumulation, accurately removes liquid water interference in non-hail areas, and achieves quantitative correction of hail intensity. In real-world meteorological disaster prevention and mitigation applications, it upgrades hail warnings from macroscopic and vague regional indications to high-precision three-dimensional fixed-point landing area and level quantitative forecasts, directly providing more accurate decision support for agricultural disaster reduction, urban emergency response, and transportation security. It significantly reduces the social response costs caused by false alarms and effectively improves the refined prevention capabilities of severe convective hail disasters. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the basic process of a hail location-based quantitative identification and early warning method based on radar and lightning data analysis, provided as an embodiment of the present invention. Detailed Implementation

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] Reference Figure 1 As an embodiment of the present invention, a method for targeted and quantitative identification and early warning of hail based on radar and lightning data analysis is provided, comprising the following steps: Step S1: Obtain the registered dual-polarization radar three-dimensional volume scan data and three-dimensional lightning location data, calculate the posterior probability of each grid point in the three-dimensional spatial grid belonging to the grenade phase state, and extract the hail risk field based on the posterior probability. Step S2: Extract cloud flash discharge points from the three-dimensional lightning location data as trigger events, and mark the trigger events that fall into the hail risk field as valid anchor points; Step S3: Take the effective anchor points and grid points other than the effective anchor points in the hail risk field as nodes of the undirected graph, calculate the edge weights of each undirected edge in the undirected graph based on the posterior probability, and perform weight propagation and update on each node in the undirected graph to generate a spatial weight field. Step S4: Calculate the liquid water content of each grid point in the three-dimensional volume scan data of the dual polarization radar, and multiply it with the normalized weight value of the corresponding grid point in the spatial weight field to obtain the corrected liquid water content. Then, sum the corrected liquid water content of each grid point at the same horizontal position along the vertical direction to obtain the corrected vertical cumulative liquid water content at each horizontal position. Step S5: Determine the hail warning area and pre-hail warning level indicators based on the modified vertical cumulative liquid water content, and generate warning information.

[0019] This invention achieves precise and quantitative identification and early warning of hail, significantly improving the spatial accuracy and confidence of the warning. Existing technologies typically rely solely on radar reflectivity factors for identification, which are susceptible to interference from terrain clutter or non-meteorological targets, resulting in a high false alarm rate. This invention introduces three-dimensional lightning location data and utilizes the strong correlation between lightning activity and the presence of graupel particles during hail formation to construct a hail risk field. In particular, this invention uses cloud-to-cloud lightning discharge points as triggering events and marks them as effective anchor points. Combined with graph theory algorithms for weight propagation, the instantaneous and precise positioning characteristics of lightning data are used to dynamically correct radar identification results, effectively compensating for the shortcomings of single radar data in identifying the initial development characteristics of hail and enhancing the physical basis of the identification results. By generating early warning information through modified vertical cumulative liquid water content, the bias problem of traditional vertical cumulative liquid water content estimation of hail landing areas is avoided, achieving precise delineation of hail landing areas and quantitative assessment of hail intensity.

[0020] Step S1 specifically includes: The target detection area is divided into three-dimensional grids using a preset spatial resolution to obtain a three-dimensional spatial grid matrix; The smallest spatial unit in the three-dimensional spatial grid matrix is ​​taken as the grid point; Acquire 3D volume scan data and 3D lightning location data from dual-polarization radar, convert the polar coordinates of the 3D volume scan data from dual-polarization radar to geodetic coordinates, and convert the 3D lightning location data to geodetic coordinates that are identical to the 3D spatial grid matrix; Dual-polarization radar three-dimensional volume scan data is a raw set of observations acquired by radar through stereo scanning at various elevation angles. Its core components include the horizontal reflectivity factor reflecting precipitation intensity, Doppler velocity and spectral width detecting particle motion, and dual-polarization characteristic parameters that can identify particle phase states: differential reflectivity for determining particle shape, correlation coefficient characterizing particle uniformity, and differential phase shift rate reflecting liquid precipitation content. These data collectively construct a three-dimensional microphysical structure image of precipitation particles within the cloud. Three-dimensional lightning location data, on the other hand, utilizes real-time lightning discharge information acquired through a multi-station detection network. This data primarily includes the occurrence time accurate to milliseconds, lightning type (distinguishing between cloud-to-ground lightning), polarity, peak current intensity, and precise longitude, latitude, and altitude three-dimensional spatial coordinates of the discharge point. The combination of these two types of data enables dual monitoring of cloud physical properties and electrical activity, providing comprehensive observational evidence for hail identification.

[0021] The transformed geodetic coordinates are mapped to grid points to generate registered dual-polarization radar 3D volume scan data and 3D lightning location data. The differential reflectivity, differential phase shift rate, correlation coefficient, and horizontal reflectivity in the registered dual-polarization radar three-dimensional volume scan data are used as dual-polarization parameters. The dual polarization parameters are input into a preset fuzzy logic classifier to calculate the membership degree of each grid point to the grenade phase state; The membership degree is normalized to obtain the posterior probability of each grid point; Hail risk field is extracted based on posterior probability.

[0022] In this embodiment, the construction of the three-dimensional spatial grid matrix is ​​based on the volume scan resolution setting of the dual-polarization radar. The preset spatial resolution includes a horizontal resolution of 1km × 1km and a vertical resolution of 0.5km, forming a three-dimensional Cartesian coordinate system covering a radar detection radius of 150km and a vertical height of 0-20km. The preset fuzzy logic classifier uses a combination of trapezoidal and triangular membership functions. Specifically, differential reflectivity, differential phase shift rate, correlation coefficient, and horizontal reflectivity are used as four input feature dimensions. Each dimension is assigned a membership function corresponding to the grenadier phase. For example, for differential reflectivity, 2.0 dB to 5.0 dB is set as the high membership interval; for horizontal reflectivity, 40 dBZ to 55 dBZ is set as the high membership interval. After calculating the membership degree of each parameter, the final membership degree is synthesized by weighted averaging. The Softmax normalization method is used to map the synthesized membership degree to a posterior probability value between 0 and 1. The weight of high membership degree grid points is enhanced by exponential transformation, thereby accurately quantifying the probability that each grid point belongs to the grenadier phase.

[0023] This invention solves the problem of registration and fusion of multi-source heterogeneous data in three-dimensional space, laying a data foundation for subsequent accurate identification. Existing technologies often simply superimpose radar and lightning data in a two-dimensional plane, ignoring the three-dimensional structural features of hail occurrence and development. This invention establishes a unified three-dimensional spatial grid matrix to transform the polar coordinate data of dual-polarization radar and lightning positioning data into a unified geodetic coordinate system, eliminating positioning errors caused by coordinate system differences. By using dual-polarization parameters to calculate the posterior probability of graupel particles through a fuzzy logic classifier, compared with the method of using only horizontal reflectivity for threshold judgment, it can more effectively distinguish different phase particles such as hail, raindrops, and snowflakes. From a physical mechanism perspective, it improves the accuracy of hail main phase identification and reduces misjudgments caused by environmental interference.

[0024] Extracting the hail risk field based on posterior probability specifically includes: Grid points with a posterior probability greater than a preset first probability threshold are selected as candidate grid points. Traverse the three-dimensional spatial mesh matrix and extract the adjacent candidate grid points that have a common surface in space for each candidate grid point; Merge adjacent candidate grid points into a connected domain; Connected regions with a volume greater than a preset volume threshold are designated as hail risk fields.

[0025] In this embodiment, the specific process of extracting the hail risk field based on the posterior probability is as follows: First, a preset first probability threshold is set to 0.7, and grid points with a posterior probability greater than the threshold are selected as candidate grid points to eliminate pixels that are likely to belong to raindrops or other interference.

[0026] The existence of a common surface refers to two candidate grid points being directly adjacent along the X, Y, and Z axes in a 3D grid coordinate system, sharing a complete square surface. A 3D flooding filling algorithm is used to traverse all candidate grid points, aggregating spatially adjacent grid points into connected domains. Subsequently, the number of grid points contained in each connected domain is counted and multiplied by the volume of a single grid point to obtain the volume of the connected domain. The preset volume threshold is set to 10 cubic kilometers, which is based on the smallest scale of convective cell development. Only connected domains with a volume exceeding this threshold are identified as hail risk fields, thus effectively filtering out isolated noise points and non-meteorological echoes.

[0027] This invention improves the anti-interference capability and integrity of hail risk field extraction. Traditional threshold segmentation methods are prone to misclassifying isolated noise points as risk areas or fragmenting what should be continuous hail clouds. This invention filters candidate grid points by setting a probability threshold and performs secondary filtering by combining a connected component volume threshold. This effectively eliminates sporadic false signal points and retains continuous convective cell structures with actual meteorological significance. This extraction method based on posterior probability connected components ensures that the basic data source for subsequent weight propagation calculations has sufficient spatial scale and structural integrity, avoiding early warning failures caused by fragmented input data.

[0028] Step S2 specifically includes: Extract the three-dimensional coordinates of cloud-to-cloud lightning discharge points from the registered three-dimensional lightning location data, and use the cloud-to-cloud lightning discharge points as trigger events; The longitude, latitude, and altitude ranges of each grid point in the hail risk field are used as the three-dimensional coordinate ranges. Extract trigger events whose 3D coordinates are within the 3D coordinate range, and use them as target trigger events. Then, use the grid point where the 3D coordinates of the target trigger event are located as the target grid point. If the posterior probability of the target grid point is greater than the preset second probability threshold, then the target triggering event corresponding to the target grid point is marked as a valid anchor point.

[0029] If the posterior probability of the target grid point is less than or equal to the preset second probability threshold, it is determined that the phase characteristic uncertainty of the target grid point is high or that it is more likely to belong to a non-hail phase. This indicates that although the triggering event occurs within the risk field, it is not located in the core area of ​​the hail. The target triggering event does not meet the conditions of being a high-confidence risk source. Therefore, it is not marked as a valid anchor point and is excluded from the anchor point node set when constructing the undirected graph. It is treated as an ordinary node or ignored directly, thereby preventing low-probability risk points from misleading the graph propagation algorithm and ensuring the high purity and representativeness of valid anchor points.

[0030] In this embodiment, extracting cloud-to-cloud lightning discharge points as triggering events is based on the microphysical mechanism of hail generation: in the early stages of hail cloud development, frequent collisions of ice phase particles within the cloud mainly induce cloud-to-cloud lightning. Therefore, cloud-to-cloud lightning has a higher indicative significance for early hail identification. Specifically, the type flag is read from the three-dimensional lightning location data, and only data points marked as "cloud-to-cloud lightning" are retained. Coordinate matching adopts a three-dimensional bounding box detection algorithm. If the longitude, latitude, and altitude coordinates of the cloud-to-cloud lightning point fall within the three-dimensional coordinate range corresponding to the hail risk field, it is initially identified as a target triggering event. Subsequently, the posterior probability of the grid point where the target triggering event is located is further queried, and a preset second probability threshold is set to 0.85. Only when the probability value exceeds this threshold is it marked as a valid anchor point, ensuring that the anchor point position is in the high-risk core area of ​​hail and avoiding edge interference.

[0031] A physical triggering mechanism for deep fusion of lightning data and radar data has been established. Existing technologies often simply use lightning density as a weighting factor, lacking the ability to extract the physical meaning of lightning occurrence locations. This invention innovatively extracts cloud-to-cloud lightning discharge points as triggering events, because cloud-to-cloud lightning usually occurs inside strong convective clouds and is directly related to the collision and electrification process of graupel particles. By screening triggering events that fall into the hail risk field and meet the probability standard as effective anchor points, the key nodes involved in the calculation have extremely high credibility and representativeness, enabling subsequent algorithms to focus on the core area of ​​hail occurrence, greatly improving the directionality and noise resistance of identification.

[0032] Step S3 specifically includes: Each valid anchor point is designated as an anchor node, and the remaining grid points in the hail risk field, excluding anchor nodes, are designated as ordinary nodes. Any two nodes that share a common surface in the hail risk field are considered as adjacent nodes; The existence of a common face refers to two nodes being directly adjacent in the three-dimensional Cartesian coordinate system within a three-dimensional mesh matrix. Specifically, this means that two unit cube grid points share a complete square side face. This implies that the center coordinates of the two nodes differ by only one grid unit in one of the X, Y, and Z axes, while being identical in the other two axes. This eliminates the possibility of intersection only at vertices or edges. This strict six-adjacency definition ensures the direct physical connectivity of adjacent nodes, providing an accurate topological basis for constructing undirected graphs.

[0033] By establishing undirected edges between adjacent nodes, an undirected graph is obtained. The preset first initial value is used as the initial anchor weight of each anchor node in the undirected graph; Use the preset second initial value as the initial ordinary weight of each ordinary node in the undirected graph; Calculate the edge weights of each undirected edge in an undirected graph based on posterior probability; The target weight of each node in the undirected graph is obtained by propagating and updating the weights of each node based on the edge weights. The target weights are subjected to max-min normalization to obtain normalized weight values; Max-min normalization is a linear data transformation method designed to map target weight data to a specific numerical range. In practice, it iterates through all nodes in the spatial weight field, finds the maximum and minimum values ​​of the target weight values, calculates the difference between the target weight and the minimum value for each node, as well as the difference between the maximum and the minimum value, and uses the ratio of the aforementioned difference to the following difference as the normalization result. This process uniformly scales the original values ​​to a standard range of 0 to 1, eliminating the influence of differences in the dimensionality of weight values ​​between different iteration cycles or different storm cells. This allows the normalized weight values ​​to intuitively reflect the relative risk level of grid points in the current risk field, providing a dimensionless and comparable correction coefficient for subsequent calculations of corrected vertical cumulative liquid water content.

[0034] A spatial weight field is generated based on the normalized weight values ​​of each node in the undirected graph.

[0035] In this embodiment, the undirected graph is constructed using a six-adjacency structure. Each grid point within the hail risk field is considered a node. If two grid points share a common surface in three-dimensional space, an undirected edge is established to connect them. The initial weight assignment strategy is as follows: For valid anchor point nodes that have undergone rigorous screening, a preset first initial value of 1.0 is assigned, representing their highest risk confidence. For ordinary nodes within the risk field, a preset second initial value of 0.0 is assigned, representing the unknown risk in the initial state. This assignment method establishes a clear information source. Through the subsequent weight propagation algorithm, the high confidence information of the anchor points is diffused to the surrounding ordinary nodes, achieving the goal of enhancing the recognition effect of continuous radar data using discrete lightning data.

[0036] A weighted network model capable of adaptively propagating risk information was constructed. Most existing identification methods are based on static thresholds or linear combinations, lacking the utilization of spatial correlation information. This invention abstracts the hail risk field as an undirected graph, uses effective anchor points as information sources with high confidence, and diffuses risk information to surrounding grid points through a weight propagation algorithm. This method not only preserves the high-risk characteristics of the anchor point locations, but also utilizes the spatial adjacency relationship between nodes. Through iterative updates, the weights of surrounding grid points with similar hail characteristics are enhanced, effectively smoothing abrupt noise in the data, improving the overall continuity and internal consistency of the risk field, and making the final generated spatial weight field more realistically reflect the spatial distribution pattern of hail.

[0037] Calculating the edge weights of undirected edges in an undirected graph based on posterior probability specifically includes: Traverse each undirected edge in the undirected graph and multiply the posterior probabilities of the adjacent nodes connected by the current undirected edge to obtain the product result; Use the product as the edge weight of the current undirected edge.

[0038] The physical significance of edge weight design lies in the fact that the larger the product of the posterior probabilities of two nodes, the more consistent they are in phase characteristics, the greater the probability that they both belong to the graupel core region, and the "wider" and more reliable the information transmission channel between them. Conversely, if the product is small, it indicates that the attributes of the two nodes are significantly different, and information transmission is hindered. By directly using the probability product as the edge weight, a weighted network model based on the similarity of microphysical properties is constructed. This allows risk information to flow and be amplified efficiently in regions where graupel particles are concentrated with a high probability, while automatically suppressing excessive diffusion to non-target regions, effectively ensuring the convergence accuracy of the algorithm when identifying the hail core boundary.

[0039] This invention optimizes the influence transmission rules between nodes in the graph model. In traditional graph propagation algorithms, edge weights are often set based solely on spatial distance or common geometric connections. This invention uses the posterior probability product as the edge weight, directly quantifying the similarity of microphysical properties between adjacent nodes. A larger posterior probability product indicates a higher confidence level that two grid points belong to the same grenade phase, resulting in smoother information transmission; conversely, a smaller product indicates phase differences and hindered propagation. This design aligns with the relatively concentrated distribution of grenade within hail clouds, enabling the construction of connected domains using phase purity. It effectively prevents the ineffective diffusion of risk weights to non-hail areas (such as liquid precipitation areas), ensuring efficient transmission of risk information in areas with similar physical properties and improving the model's ability to define the boundaries of the hail core area.

[0040] The specific steps of weight propagation and update for each node in an undirected graph based on edge weights include: Based on the edge weight, the initial anchor weight of the anchor node is iteratively passed to the adjacent ordinary nodes. In each iteration, calculate the absolute value of the weight difference before and after the update of each node at the current iteration number; The iteration propagation stops when the absolute value of the weight difference is less than the preset convergence threshold, or when the current iteration number is greater than or equal to the preset maximum iteration number. When the absolute value of the weight difference is greater than or equal to the preset convergence threshold, and the current iteration number is less than the preset maximum iteration number, the next iteration propagation continues. Extract the current weight of each node in the undirected graph when the iteration stops, and use it as the target weight; Iterative propagation specifically includes: For a regular node in an undirected graph, the weights of the undirected edges connected to the regular node are multiplied by the current weights of the corresponding adjacent nodes and then summed. The sum is divided by the sum of the weights of all the undirected edges connected to the regular node to obtain the quotient, which is then used as the updated weight of the regular node. For anchor nodes in an undirected graph, keep the initial anchor weight of the current anchor node unchanged.

[0041] In this embodiment, the weight propagation and update adopts an improved version of the graph-based label propagation algorithm. In each iteration, the updated weight of a normal node is equal to the weighted sum of the products of the weights passed from all its neighboring nodes and the corresponding edge weights, and is normalized by dividing by the sum of edge weights. The anchor node always maintains its initial weight of 1.0 as a constant constraint. A preset convergence threshold of 0.0001 is set, that is, when the absolute value of the weight difference before and after the update of all nodes is less than this threshold, the calculation is considered to have converged. At the same time, a preset maximum number of iterations is set to 50 to prevent infinite loops in extreme cases. This iterative mechanism smoothly expands the discrete risk points triggered by lightning into a continuous risk distribution field through multiple convolution operations, generating a spatial weight field with clear physical meaning.

[0042] The algorithm's stability and computational efficiency are ensured by establishing a rule that keeps the anchor point weights constant, thus establishing a source benchmark for risk propagation and preventing the dissipation of key information during propagation. This ensures that the calculation results always converge around the effective anchor point. At the same time, the absolute value of the weight difference and the maximum number of iterations are introduced as dual convergence conditions. This ensures that the calculation results reach a stable state while avoiding the waste of resources caused by unlimited iterative calculations. This mechanism enables the invention to respond quickly to changes in input data, meeting the strict timeliness requirements of hail warnings and significantly shortening the processing time while ensuring recognition accuracy.

[0043] Step S4 specifically includes: Read the horizontal reflectivity factor of each grid point in the three-dimensional volume scan data of dual-polarization radar; Substitute the horizontal reflectivity factor into the preset empirical formula for water content to calculate the liquid water content at each grid point. The modified liquid water content of each grid point is obtained by multiplying the liquid water content of each grid point with the normalized weight value of the corresponding grid point in the spatial weight field. The modified liquid water content at each grid point at the same horizontal position is summed along the vertical direction to obtain the modified vertical cumulative liquid water content at each horizontal position.

[0044] In this embodiment, the preset moisture content empirical formula adopts the general ZM empirical formula: Where Z is the horizontal reflectivity factor, in dBZ, and M represents the calculated liquid water content, in g / m³. This invention overcomes the limitations of traditional vertical cumulative liquid water content in hail identification. Traditional methods typically assume all water particles within clouds are liquid, neglecting the influence of hail and graupel on reflectivity, and weighting the water content across the entire atmospheric column, leading to systematic biases in the calculation results. This invention utilizes a spatial weight field incorporating lightning and dual-polarization information. It first weights and corrects the liquid water content at each grid point, eliminating interference from liquid water in non-hail phase layers, and then performs vertical integration to obtain a corrected index. The corrected index is no longer constrained by the weighted accumulation of traditional algorithms, achieving three-dimensional coupling between phase identification and liquid water content in the vertical direction. This effectively suppresses misjudgments in non-hail high-reflectivity areas (such as areas of heavy rainfall), making the physical meaning of the hail warning index clearer and the numerical value more accurately reflecting the true potential intensity of hail.

[0045] Step S5 specifically includes: The horizontal position where the modified vertical cumulative liquid water content is greater than the preset water content threshold is used as the warning position; Merge adjacent warning locations that share a common edge into a two-dimensional connected region; Calculate the area of ​​each two-dimensional connected region; Extract the outer edge contour of two-dimensional connected regions whose area is greater than a preset area threshold; The horizontal area enclosed by the outer edge outline is designated as the hail warning area; The hail warning level index is determined based on the numerical value of the modified vertical liquid water content, and warning information including the hail warning area and the hail warning level index is generated.

[0046] In this embodiment, the determination of the hail warning area adopts a three-dimensional morphological processing method. First, a preset water content threshold is set. High-value grid points are selected, and then a three-dimensional region growing algorithm is used to merge adjacent warning grid points into a two-dimensional connected region in the horizontal direction. The actual coverage area on the horizontal plane is calculated, and a preset area threshold of 5 square kilometers is set to eliminate small areas caused by noise. Finally, the outer edge contour of the three-dimensional connected body that meets the conditions is extracted. The area enclosed by this contour is the hail warning landing area. This method avoids the spatial information loss caused by two-dimensional planar projection through three-dimensional screening and ensures the accuracy of the landing area boundary.

[0047] The preset water content threshold is set to 10 kg / m². This threshold is recalibrated based on the statistical characteristics of the modified vertical cumulative liquid water content. Since the modified value is lower than the traditional vertical cumulative liquid water content after weight modulation, this threshold is lower than the threshold of the traditional method. This invention achieves the extraction of accurate two-dimensional warning areas from three-dimensional volumetric scanning data. Existing methods often directly utilize horizontal slicing or simple threshold projection, resulting in irregularly shaped warning areas that easily contain invalid regions. This invention employs a progressive filtering strategy, from warning grid points to two-dimensional connected regions and then to three-dimensional connected volumes. By using projection area thresholds to eliminate minor interference areas, the outer edge contour is finally extracted as the warning area. This projection method based on three-dimensional morphological processing can eliminate isolated noise points and generate warning areas with complete shapes and clear boundaries, greatly facilitating meteorologists' intuitive judgment of the hail impact range and subsequent warning issuance.

[0048] The generation of warning information, including hail warning areas and hail warning level indicators, specifically includes: Traverse all horizontal positions within the hail warning area and extract the maximum value of the corrected vertical cumulative liquid water content corresponding to each horizontal position; Based on the preset mapping table, find the hail diameter range corresponding to the maximum value in the preset mapping table; The preset mapping table is a set of discrete numerical correspondences built based on prior meteorological knowledge. Its establishment logic lies in setting mapping rules between dual-polarization radar parameters and hail probabilities in different numerical ranges according to the microphysical characteristics of hail occurrence and development. In actual calculation, by performing a matching query on the table for the real-time observation parameter values ​​of the target grid points, the corresponding probability values ​​are directly read, thereby transforming the complex nonlinear changes of physical quantities into a queryable standard probability assignment process.

[0049] Output the level labels corresponding to the hail diameter range as the hail warning level index; Obtain the set of latitude and longitude coordinates of the outer edge contour of the hail warning area, and obtain the warning color code and warning text description corresponding to the hail warning level index; The data, consisting of latitude and longitude coordinates, warning color codes, and warning text descriptions, is encapsulated to generate warning information.

[0050] In this embodiment, the preset mapping table is established based on the retrospective statistics of historical hail disaster reports and radar data at the corresponding time. Specifically, the maximum value of the modified vertical cumulative liquid water content is divided into several intervals. For example, [20,30) corresponds to small hailstones with a diameter of less than 1cm, [30,50) corresponds to medium hailstones with a diameter of 1-2cm, and [50, +∞) corresponds to large hailstones with a diameter of more than 2cm.

[0051] When generating early warning information, all latitude and longitude coordinates of the outer edge contour of the hail warning area are extracted to form a closed polygon data packet. Combined with the early warning color codes (yellow, orange, and red) obtained from the lookup table, as well as the text description containing the location of the landing area and the predicted hail diameter, the data is encapsulated according to meteorological data exchange standard formats, including JSON and MICAPS, to generate standard early warning information that can be directly transmitted to the meteorological early warning release system.

[0052] This invention achieves the quantitative and standardized output of hail warning information. Existing warning methods mostly remain at the level of qualitative description and lack quantitative prediction of hail diameter. This invention establishes a mapping relationship between the maximum value of modified vertical cumulative liquid water content and the hail diameter range. It outputs specific hail diameter ranges and level labels through a lookup table. This processing method transforms the complex radar and lightning data analysis results into intuitive numerical ranges and level indicators, which not only improves the readability of warning information, but also provides a more accurate reference for disaster prevention and mitigation decision-making, significantly enhancing the practical value of meteorological warning services.

[0053] This invention constructs a three-dimensional spatial dynamic network model based on lightning triggering mechanisms and undirected graph weight propagation, realizing a deep spatial topological association between hail risk fields and lightning discharge points. This effectively overcomes the high false alarm rate caused by traditional methods relying solely on a single threshold or simple lightning frequency reference, significantly improving the accuracy of hail core area identification. By introducing spatially normalized weights generated by graph propagation into the calculation of vertically accumulated liquid water content, it overcomes the limitations of conventional algorithms' equal-weighted accumulation, accurately removing liquid water interference from non-hail areas and achieving quantitative correction of hail intensity. In real-world meteorological disaster prevention and mitigation applications, hail warnings are upgraded from macroscopically vague regional indications to high-precision three-dimensional pinpoint landing area and level-based quantitative forecasts, directly providing more accurate decision support for agricultural disaster reduction, urban emergency response, and transportation security. This significantly reduces the social response costs caused by false alarms and effectively enhances the refined prevention capabilities against severe convective hail disasters.

[0054] The various embodiments of the techniques described above can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0055] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0056] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), and erasable programmable read-only memory (EPROM). Or flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.

Claims

1. A method for targeted and quantitative identification and early warning of hail based on radar and lightning data analysis, characterized in that, Includes the following steps: Step S1: Obtain the registered dual-polarization radar three-dimensional volume scan data and three-dimensional lightning location data, calculate the posterior probability of each grid point in the three-dimensional spatial grid belonging to the grenade phase state, and extract the hail risk field based on the posterior probability. Step S2: Extract cloud flash discharge points from the three-dimensional lightning location data as trigger events, and mark the trigger events that fall into the hail risk field as valid anchor points; Step S3: Take the effective anchor points and grid points other than the effective anchor points in the hail risk field as nodes of the undirected graph, calculate the edge weights of each undirected edge in the undirected graph based on the posterior probability, and perform weight propagation and update on each node in the undirected graph to generate a spatial weight field. Step S4: Calculate the liquid water content of each grid point in the three-dimensional volume scan data of the dual polarization radar, and multiply it with the normalized weight value of the corresponding grid point in the spatial weight field to obtain the corrected liquid water content. Then, sum the corrected liquid water content of each grid point at the same horizontal position along the vertical direction to obtain the corrected vertical cumulative liquid water content at each horizontal position. Step S5: Determine the hail warning area and pre-hail warning level index based on the modified vertical cumulative liquid water content, and generate warning information.

2. The hail location-based quantitative identification and early warning method based on radar and lightning data analysis as described in claim 1, characterized in that, Step S1 specifically includes: The target detection area is divided into three-dimensional grids using a preset spatial resolution to obtain a three-dimensional spatial grid matrix; The smallest spatial unit in the three-dimensional spatial grid matrix is ​​taken as the grid point; Acquire three-dimensional volume scan data and three-dimensional lightning location data from dual-polarization radar, convert the polar coordinates of the three-dimensional volume scan data from dual-polarization radar to geodetic coordinates, and convert the three-dimensional lightning location data to geodetic coordinates that are the same as the three-dimensional spatial grid matrix; The transformed geodetic coordinates are mapped to the grid points to generate the registered dual-polarization radar 3D volume scan data and 3D lightning location data. The differential reflectivity, differential phase shift rate, correlation coefficient, and horizontal reflectivity in the registered dual-polarization radar three-dimensional volume scan data are used as dual-polarization parameters. The dual polarization parameters are input into a preset fuzzy logic classifier to calculate the membership degree of each grid point to the grenade phase state; The membership degree is normalized to obtain the posterior probability of each grid point; Hail risk field is extracted based on the posterior probability.

3. The hail location-based quantitative identification and early warning method based on radar and lightning data analysis as described in claim 2, characterized in that, The extraction of the hail risk field based on the posterior probability specifically includes: Grid points with a posterior probability greater than a preset first probability threshold are selected as candidate grid points. Traverse the three-dimensional spatial mesh matrix and extract the adjacent candidate grid points that have a common surface in space for each candidate grid point; Merge adjacent candidate grid points into a connected domain; Connected regions with a volume greater than a preset volume threshold are designated as hail risk fields.

4. The hail location-based quantitative identification and early warning method based on radar and lightning data analysis as described in claim 3, characterized in that, Step S2 specifically includes: Extract the three-dimensional coordinates of cloud-to-cloud lightning discharge points from the registered three-dimensional lightning location data, and use the cloud-to-cloud lightning discharge points as trigger events; The longitude, latitude, and altitude ranges of each grid point in the hail risk field are used as the three-dimensional coordinate ranges. Extract triggering events whose three-dimensional coordinates are within the range of the three-dimensional coordinates, and use them as target triggering events. Also, use the grid point where the three-dimensional coordinates of the target triggering event are located as the target grid point. If the posterior probability of the target grid point is greater than a preset second probability threshold, then the target triggering event corresponding to the target grid point is marked as a valid anchor point.

5. The hail location-based and quantitative identification and early warning method based on radar and lightning data analysis as described in claim 4, characterized in that, Step S3 specifically includes: Each effective anchor point is designated as an anchor point node, and the remaining grid points in the hail risk field, excluding the anchor point nodes, are designated as ordinary nodes. Any two nodes that share a common surface in the space within the hail risk field are considered as adjacent nodes; An undirected graph is obtained by establishing undirected edge connections between the adjacent nodes. The preset first initial value is used as the initial anchor weight of each anchor node in the undirected graph; Use the preset second initial value as the initial ordinary weight of each ordinary node in the undirected graph; The edge weights of each undirected edge in the undirected graph are calculated based on the posterior probability. Based on the edge weights, the weights of each node in the undirected graph are propagated and updated to obtain the target weights of each node. The target weights are subjected to maximum and minimum normalization to obtain normalized weight values; A spatial weight field is generated based on the normalized weight values ​​of each node in the undirected graph.

6. The hail location-based and quantitative identification and early warning method based on radar and lightning data analysis as described in claim 5, characterized in that, The calculation of the edge weights of each undirected edge in the undirected graph based on the posterior probability specifically includes: Traverse each undirected edge in the undirected graph and multiply the posterior probabilities of the adjacent nodes connected by the current undirected edge to obtain the product result; Use the product as the edge weight of the current undirected edge.

7. The hail location-based quantitative identification and early warning method based on radar and lightning data analysis as described in claim 6, characterized in that, The specific steps of weight propagation and updating of each node in the undirected graph based on the edge weights include: Based on the edge weights, the initial anchor weights of the anchor nodes are iteratively passed to the adjacent ordinary nodes. In each iteration, calculate the absolute value of the weight difference before and after the update of each node at the current iteration number; The iteration is stopped when the absolute value of the weight difference is less than the preset convergence threshold, or when the current iteration number is greater than or equal to the preset maximum iteration number. When the absolute value of the weight difference is greater than or equal to the preset convergence threshold, and the current iteration number is less than the preset maximum iteration number, the next iteration continues. Extract the current weight of each node in the undirected graph at the time of stopping iteration, and use it as the target weight; The iterative transmission specifically includes: For a regular node in the undirected graph, the edge weights of each undirected edge connected to the regular node are multiplied by the current weights of the corresponding adjacent nodes and then summed. The sum is divided by the sum of the edge weights of all undirected edges connected to the regular node to obtain a quotient, which is then used as the updated weight of the regular node. For the anchor nodes in the undirected graph, the initial anchor weight of the current anchor node remains unchanged.

8. The hail location-based quantitative identification and early warning method based on radar and lightning data analysis as described in claim 7, characterized in that, Step S4 specifically includes: Read the horizontal reflectivity factor of each grid point in the three-dimensional volume scan data of the dual-polarization radar; Substitute the horizontal reflectivity factor into the preset empirical formula for water content to calculate the liquid water content at each grid point. The modified liquid water content of each grid point is obtained by multiplying the liquid water content of each grid point with the normalized weight value of the corresponding grid point in the spatial weight field. The modified liquid water content at each grid point at the same horizontal position is summed along the vertical direction to obtain the modified vertical cumulative liquid water content at each horizontal position.

9. The hail location-based quantitative identification and early warning method based on radar and lightning data analysis as described in claim 8, characterized in that, Step S5 specifically includes: The horizontal position where the modified vertical cumulative liquid water content is greater than the preset water content threshold is used as the warning position; Merge adjacent warning locations that share a common edge into a two-dimensional connected region; Calculate the area of ​​each two-dimensional connected region; Extract the outer edge contour of two-dimensional connected regions whose area is greater than a preset area threshold; The horizontal area enclosed by the outer edge outline is designated as the hail warning area; The hail warning level index is determined based on the value of the modified vertical liquid water content, and warning information including the hail warning area and the hail warning level index is generated.

10. The hail location-based quantitative identification and early warning method based on radar and lightning data analysis as described in claim 9, characterized in that, The generation of warning information, including the hail warning area and the hail warning level index, specifically includes: Traverse all horizontal positions within the hail warning area and extract the maximum value of the corrected vertical cumulative liquid water content corresponding to each horizontal position; Based on a preset mapping table, the hail diameter range corresponding to the maximum value is found in the preset mapping table; Output the level label corresponding to the hail diameter range as a hail warning level indicator; Obtain the set of latitude and longitude coordinates of the outer edge contour of the hail warning area, and obtain the warning color code and warning text description corresponding to the hail warning level index; The latitude and longitude coordinates, the warning color code, and the warning text description are encapsulated to generate warning information.

Images