Storm cell tracking and extrapolation methods, equipment and storage media

Abstract

This invention discloses a storm cell tracking and extrapolation method, device, and storage medium, relating to the field of weather tracking technology. The method includes: for the current storm profile, calculating its overlap score with each storm profile at the previous time step; for storm profile pairs with an overlap score greater than 0, calculating their comprehensive similarity score based on spatial location, morphological structure, and physical attributes, and selecting the storm profile with the highest comprehensive similarity score that is greater than a minimum acceptable threshold as the associated object, establishing a tracking chain relationship of storm cells in the time series; predicting the storm's position and morphology at the next time step based on echo data within historical storm profiles, thus achieving storm extrapolation. This invention effectively improves the storm's identification ability in densely active areas and during overlapping movements through multi-dimensional feature fusion and matching. Combined with the extrapolation results, the future impact range of the storm can be estimated, thereby improving the timeliness and accuracy of monitoring and forecasting.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of weather tracking, and particularly relates to a storm cell tracking and extrapolation method, device and storage medium. BACKGROUND

[0002] In the field of meteorological forecasting, especially in the short-impending forecasting of severe convective weather, continuous and stable tracking of storm cells is a key technical basis for generating their movement trajectories, predicting future paths, and issuing warnings. Currently, the widely used storm tracking method in operational systems is mainly based on the spatial correlation of storm targets between consecutive times.

[0003] Classic tracking algorithms, such as matching methods based on overlapping area or centroid nearest neighbor, are relatively efficient in computation, but have obvious limitations in practical application:

[0004] Firstly, these methods usually rely only on a single or a few features (such as centroid distance or overlapping area) for matching decisions, and the feature utilization is insufficient. When multiple storm targets are close in space, cross or move in parallel, relying on a single feature can easily lead to mis-matching, causing "jumping" of the tracking chain or exchange of identity identification (ID).

[0005] Secondly, existing methods generally lack effective use of complete life history information of storms. The matching process is usually independent between adjacent two times, without fully considering the movement trend and evolution consistency of storms over a longer time sequence. This leads to the generated tracking trajectory being jittery and not smooth enough, and when storms merge or split, it is difficult to accurately determine their kinship, easily leading to interruption or re-initialization of the tracking chain.

[0006] In addition, even after successful matching, directly using the original centroid position output by the recognition module to construct the trajectory will introduce positioning fluctuations caused by recognition noise. This non-smoothness of the centroid trajectory will directly impair the accuracy of subsequent storm extrapolation forecasts.

[0007] Therefore, there is an urgent need in the field for a storm tracking method that can comprehensively utilize multi-dimensional features of storms and effectively incorporate temporal context information, to improve the accuracy of matching, the smoothness of the trajectory, and the completeness of the life history record in complex weather scenarios. SUMMARY

[0008] In view of the above-mentioned defects in the prior art, the purpose of the present application is to provide a storm cell tracking and extrapolation method, device and storage medium, to solve the technical problems of low matching accuracy, poor trajectory smoothness and insufficient robustness in complex evolution scenarios caused by single feature utilization and lack of temporal continuity in existing tracking methods.

[0009] The application is to solve the above technical problems by the following technical scheme: a storm cell tracking and extrapolation method, comprising:

[0010] For any storm contour in the current storm cell identification result, calculate the overlap score between it and each storm contour in the last storm cell identification result;

[0011] For storm contour pairs with an overlap score greater than 0, calculate their comprehensive similarity score based on spatial position, morphological structure, and physical properties;

[0012] For a storm contour in the current storm cell identification result, select the storm contour in the last storm cell identification result with the highest comprehensive similarity score and greater than the minimum acceptance threshold as its associated object, and establish a tracking chain relationship of storm cells in time series;

[0013] Based on the echo data in the historical storm contour, predict the position and shape of the storm at the next time, and realize storm extrapolation.

[0014] The application does not rely on any single feature, but comprehensively calculates the similarity of multiple dimensions such as spatial position (such as centroid distance), morphological structure (such as area, shape), and physical properties (such as intensity, echo top height). This multi-dimensional evaluation greatly enhances the algorithm's ability to distinguish different storms. For example, two storms may have a very close centroid distance, but their shape or intensity may be completely different. Traditional methods may be confused, but the application can accurately associate the current storm with the most similar one in the historical storm as a whole by comprehensive judgment, thereby significantly reducing the mismatch rate in dense convective areas or complex motion patterns.

[0015] Selecting the historical storm contour with the highest comprehensive similarity score and greater than the minimum acceptance threshold, this principle itself will tend to select the target that evolves most continuously and consistently in multiple features. This is equivalent to embedding the consideration of motion continuity and evolution consistency in the matching logic, rather than just the nearest neighbor in space.

[0016] When storms merge or split, single features (such as overlap area) will fail. However, the multi-dimensional feature evaluation of the application can capture more subtle evolution clues. For example, in the splitting scenario, the sub-body and the parent body will have a higher similarity in shape and intensity characteristics; in the merging scenario, the merged entity and the multiple entities participating in the merging will have a inheritance relationship in spatial position and shape. By calculating the comprehensive similarity score, the application can more reasonably determine these complex kinship relationships, thereby more stably maintaining or ending the tracking chain, reducing storm ID disorder and trajectory breakage.

[0017] Further, for the storm profile pairs with overlap score greater than 0, a comprehensive similarity score based on spatial position, morphological structure and physical attribute is calculated, specifically including:

[0018] For the storm profile pairs with overlap score greater than 0, the center distance, area ratio, minimum distance between two profiles, shape difference degree, intensity difference degree, echo top height difference and intersection area ratio are calculated;

[0019] Based on the preset membership function, the center distance, area ratio, minimum distance between two profiles, shape difference degree, intensity difference degree, echo top height difference and intersection area ratio are respectively converted into corresponding similarity scores; wherein, the value range of each similarity score is [0, 1], and the greater the value is, the more similar it is in the corresponding feature;

[0020] All the similarity scores are weighted and summed to obtain the comprehensive similarity score.

[0021] The present application maps the original features (such as distance, area ratio, intensity difference) of different dimensions and meanings to the similarity scores in the interval [0, 1] through the preset membership function, solves the fundamental technical problem that different nature features cannot be directly compared and combined, and provides a mathematical basis for multi-feature decision-making.

[0022] The membership function allows nonlinear mapping and can more accurately depict what is similar. For example, it can define that the center distance is very similar within 5 kilometers and rapidly dissimilar when exceeding 10 kilometers, which is more consistent with the actual physical law and business experience of storm evolution than simple linear comparison, greatly improving the accuracy of matching.

[0023] Through weighted summation, the present application can comprehensively consider the overall similarity of storms in space, shape and physical intensity. When a certain feature is temporarily unreliable due to noise (such as temporary jump of centroid due to identification error), other stable features (such as shape and intensity) can still ensure the correctness of the overall judgment, thereby effectively dealing with complex situations such as storm approaching, crossing, merging / splitting, and greatly improving the stability and continuity of tracking.

[0024] Further, after establishing the tracking chain relationship of storm monomers in time series, the method further includes smoothing and predicting the centroid sequence on the tracking chain, specifically including:

[0025] When a certain storm is continuously tracked for at least three times, it is judged whether the spatial distance between the centroid of the current time and the centroid of the last time exceeds the distance threshold;

[0026] If it exceeds, the state vector of position and velocity is constructed according to each historical centroid on the tracking chain;

[0027] Based on the state vector, the centroid at the current moment is smoothed and predicted to obtain the candidate corrected centroid;

[0028] If the candidate correction centroid falls within the current storm profile, the spatial distances between the candidate correction centroid, the current centroid, and the previous centroid are calculated respectively, and the one with the smaller spatial distance is selected as the current correction centroid.

[0029] If the candidate centroid for correction falls outside the current storm profile, it will not be corrected, and the current centroid will be adopted.

[0030] In this embodiment, a state vector is constructed using a tracking chain and then subjected to Kalman filtering. This not only smooths the historical trajectory but also predicts the most likely position at the current moment. This dynamic-based estimation can more scientifically address storm changes, thus providing a more reasonable candidate corrected centroid.

[0031] Assuming both the candidate corrected centroid (based on dynamic prediction) and the current centroid are valid, by comparing their distances from the previous centroid and selecting the one with the smaller movement, we are essentially choosing the solution with better continuity and smoother flow with the historical centroid. This optimization strategy automatically selects the result that best maintains motion stability, maximizing the smoothing effect.

[0032] Furthermore, based on echo data within historical storm profiles, the location and shape of the storm at the next moment are predicted, specifically including:

[0033] Based on the echo data of the two-dimensional combined radar echo of the previous n frames within the current storm profile and the echo data within the current storm profile, the initial velocity at the current moment is calculated using the optical flow method.

[0034] The second decay coefficient is calculated based on the length of the tracking chain and the preset weight of the first moment;

[0035] The normalized weight coefficients at each moment on the tracking chain are calculated based on the second decay coefficient.

[0036] The combined velocity and direction at the current moment are calculated based on the initial velocity at the current moment, the velocity at each historical moment, and the normalized weighting coefficient at each moment.

[0037] Based on the current combined velocity and direction, as well as the coordinates of each point on the current storm profile, the current storm profile is spatially extrapolated to obtain the next storm profile.

[0038] This invention achieves more reliable short-term extrapolation forecasts by dynamically fusing instantaneous motion with historical trends through an adaptive weighting calculation mechanism.

[0039] By automatically calculating the attenuation coefficient and weighting coefficient based on the tracking chain length, this invention can intelligently adjust the level of trust in historical information. For emerging storms (short chains), the algorithm relies more on the instantaneous velocity calculated by the current optical flow method to ensure a fast response; for mature storms (long chains), it smoothly integrates more historical velocities to effectively suppress instantaneous fluctuations and enhance the stability of the trend.

[0040] Traditional methods often use instantaneous velocity directly for extrapolation, which is easily affected by local changes in echoes. The velocity and direction obtained by the present invention through weighted fusion are smoothed optimal estimates that better represent the overall movement trend of the storm. This provides a more reliable and accurate initial dynamic field for subsequent spatial extrapolation, enabling a more precise estimation of the future impact range of the storm and providing key evidence for refined early warning of severe convective weather.

[0041] Based on the same concept, the present invention also provides an electronic device, including a memory, a processor, and a computer program or instructions stored in the memory, wherein the processor executes the computer program or instructions to implement the storm cell tracking and extrapolation method as described above.

[0042] Based on the same concept, the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implements the storm cell tracing and extrapolation method as described above.

[0043] Compared with existing technologies, the storm cell tracking and extrapolation method provided by this invention brings the following significant benefits by introducing multi-feature fuzzy matching and trajectory smoothing techniques:

[0044] By employing multi-dimensional feature fusion matching based on spatial, morphological, and physical attributes, the ability to identify storms in densely populated areas and during cross-movement is significantly improved, effectively reducing mismatches and ID exchanges.

[0045] Through centroid correction and trajectory smoothing mechanisms, centroid jumps caused by identification noise were effectively suppressed, generating physically reasonable and temporally smooth storm motion trajectories, laying a solid foundation for accurate extrapolation forecasts.

[0046] The multi-feature complementary decision-making mechanism enables the present invention to maintain the stability and logical rationality of the tracking chain even when storms undergo complex evolution such as merging and splitting, significantly improving its adaptability in complex weather scenarios. Attached Figure Description

[0047] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only one embodiment of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart of the storm cell tracking and extrapolation method in an embodiment of the present invention;

[0049] Figure 2 This is a flowchart of the smoothing and prediction process for the storm centroid in an embodiment of the present invention;

[0050] Figure 3 This is a diagram showing the effect of identifying convective monomers with extrapolation direction at the current moment in an embodiment of the present invention. Detailed Implementation

[0051] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0053] Explanation of the name:

[0054] Storm profile: A connected region segmented from a single-time radar two-dimensional composite echo, continuous in two-dimensional or three-dimensional space and satisfying a preset reflectivity threshold. The storm profile defines the spatial shape and boundary of a storm at a specific moment, and is usually characterized by a set of polygon vertex coordinates.

[0055] Storm target: refers to an independent, discrete storm entity identified based on a single-time radar two-dimensional combined echo, using thresholding, spatial clustering, or image segmentation techniques. A storm target is a general term for a storm entity existing at a specific moment, and its core spatial attributes are defined by the storm profile at that moment. A storm target necessarily corresponds to a storm profile.

[0056] Storm Individual: Refers to a storm target that has been included in a continuous spatiotemporal tracking sequence. A storm individual is formed when a storm target is successfully associated with a tracking chain and assigned a unique identifier. A storm individual is a business entity with temporal continuity, and its complete life cycle is described by a series of temporally continuous storm profiles and their derived attributes (such as centroid, intensity, volume, etc.).

[0057] Example 1

[0058] like Figure 1 As shown, the storm cell tracking and extrapolation method provided in this embodiment of the invention includes the following steps:

[0059] Step S1: For any storm profile in the current storm cell identification result, calculate the overlap score between it and each storm profile in the previous storm cell identification result.

[0060] The current storm cell identification result refers to the current time T. i The storm cell identification result is a result consisting of at least one storm profile. The previous storm cell identification result refers to the result at the previous time T. i-1 The storm cell identification results are obtained, which consist of at least one storm profile. Each storm profile is characterized by a set of polygon vertex coordinates that define its boundaries.

[0061] For each storm profile in the current storm cell identification result, it is sequentially paired with each storm profile in the previous storm cell identification result to form storm profile pairs to be compared. For each storm profile pair, its overlapping area on the two-dimensional plane is calculated, and then the overlap score is calculated:

[0062] (1)

[0063] in, The overlap score is used to quantify the degree of spatial overlap between two storm profiles in a storm profile pair. This represents the ratio of the overlapping area of ​​the storm profile pair to the area of ​​the current storm profile (i.e., a storm profile in the current final storm cell identification result). This represents the ratio of the overlapping area of ​​the storm profile pair to the area of ​​the previous storm profile (i.e., a storm profile in the previous storm cell identification result).

[0064] As an alternative implementation, the overlap fraction can also be defined as the ratio of the overlap area to the area of ​​either of the two storm profiles, i.e. for or .

[0065] The overlap score is a specific value between [0,1]. If there is a complete overlap, the overlap score is 1; if there is no overlap, the overlap score is 0, indicating no match, and the pair is skipped and not included in the subsequent comprehensive similarity score calculation. If the overlap score is greater than 0, the corresponding storm contour pair proceeds to the next step and participates in the subsequent comprehensive similarity score calculation.

[0066] Step S2: For storm contour pairs with an overlap score greater than 0, calculate their comprehensive similarity score based on spatial location, morphological structure, and physical properties.

[0067] In one specific embodiment of the present invention, a multi-feature fuzzy matching algorithm is used to calculate the comprehensive similarity score, specifically including:

[0068] Step S2.1: For storm profile pairs with an overlap score greater than 0, calculate their center distance, area ratio, minimum distance between the two profiles, shape difference, intensity difference, echo top height difference, and intersection area ratio.

[0069] Center distance: The Euclidean distance between the centroids of two storm profiles is calculated to reflect the proximity of the storm positions.

[0070] Area ratio: The similarity of target size is measured by calculating the ratio of the areas of two storm profiles in a storm profile pair.

[0071] Minimum distance between two profiles: The proximity of storm positions is reflected by calculating the minimum Euclidean distance between two storm profiles in a storm profile pair.

[0072] Shape difference: Based on parameters such as contour eccentricity and aspect ratio, the differences in storm shape are depicted;

[0073] Intensity variation: Based on indicators such as maximum reflectivity and average reflectivity, assess the consistency of storm intensity evolution;

[0074] Echo top height difference: Based on calculated three-dimensional features, the difference in echo top height is compared to reflect the similarity of storms;

[0075] Intersection area ratio: The spatial continuity of a target is reflected by calculating the ratio of the overlapping areas of two storm profiles in a storm profile pair.

[0076] Step S2.2: Based on the preset membership function, the center distance, area ratio, minimum distance between two contours, shape difference, intensity difference, echo top height difference, and intersection area ratio are converted into corresponding similarity scores; where the value range of each similarity score is [0,1], and the larger the value, the more similar the feature is.

[0077] In this embodiment, a Gaussian function is used as the membership function. For different features (center distance, area ratio, minimum distance between two contours, eccentricity difference, maximum echo intensity difference, echo top height difference, and intersection area ratio), different means, standard deviations, and effective ranges are set to obtain the membership function corresponding to each feature. Then, the corresponding features are mapped to similarity scores using the membership function. The expression for the membership function is:

[0078] (2)

[0079] in, This represents the similarity score of the j-th feature; The j-th feature is represented by the center distance, area ratio, minimum distance between two contours, eccentricity difference, maximum echo intensity difference, echo top height difference, and intersection area ratio. Let represent the mean of the j-th feature; This represents the standard deviation of the j-th feature; This indicates the effective range of the j-th feature.

[0080] Mean of each feature Standard deviation and effective range Specifically:

[0081] center distance Mean 0, standard deviation The effective range is 2. for ;

[0082] Area ratio Mean =1, standard deviation The effective range is 0.35. for ;

[0083] Minimum distance between two contours Mean 0, standard deviation The effective range is 0.07. for ;

[0084] Eccentricity difference Mean 0, standard deviation The effective range is 0.15. for ;

[0085] Maximum echo intensity difference Mean 0, standard deviation The effective range is 5. for ;

[0086] Echo top height difference Mean 0, standard deviation The effective range is 3. for ;

[0087] Intersection area ratio Mean =1, standard deviation The effective range is 0.4. Effective across the entire region.

[0088] Step S2.3: Calculate the weighted sum of all similarity scores to obtain the comprehensive similarity score. The specific calculation formula is as follows:

[0089] (3)

[0090] Where S represents the overall similarity score; This represents the weight ratio of the j-th similarity score.

[0091] Weighting of each similarity score The settings can be adjusted based on experience. In this embodiment, the weight ratios of the similarity scores corresponding to center distance, area ratio, minimum distance between two contours, shape difference, intensity difference, echo top height difference, and intersection area ratio are 2:1:2:1:1:1:5.

[0092] Step S3: For a storm profile in the current storm cell identification result, select the storm profile with the highest comprehensive similarity score and greater than the minimum acceptance threshold in the previous storm cell identification result as its associated object, thereby establishing a tracking chain relationship of storm cells in the time series.

[0093] For a storm profile (C_current) in the current storm cell identification result, select all previous storm profiles (i.e., storm profiles C_prior_k in the previous storm cell identification result) that are paired with the current storm profile and have an overlap score greater than 0 from the output of the initial overlap score screening step, and form a candidate association list.

[0094] From the stored comprehensive similarity scores, the comprehensive similarity scores of the current storm profile C_current and each previous storm profile C_prior_k in the candidate association list are read. The previous storm profile C_bestmatch with the highest comprehensive similarity score that is greater than the minimum acceptance threshold is found and determined. The association object of the current storm profile C_current is identified as the previous storm profile C_bestmatch. The unique identifier (ID) of the storm unit corresponding to the previous storm profile C_bestmatch is assigned to the current storm profile C_current. The current storm profile C_current and all its attributes are added as a new state node to the life history tracking chain of this storm unit. Thus, the tracking chain of this storm unit is continued and updated in the time series.

[0095] If the candidate association list of the current storm profile C_current is empty, or the overall similarity score of all candidate pairs is lower than the minimum acceptance threshold, then the current storm profile C_current is determined to be a new storm entity. A brand new and unique identifier (ID) is assigned to the new storm entity, and a new storm entity tracking chain is initialized starting from it.

[0096] In one specific embodiment of the present invention, after establishing the tracking chain relationship of storm cells in the time series, the method of the present invention further includes step S4: smoothing and predicting the centroid sequence on the tracking chain to obtain a more stable corrected centroid at the current moment, thereby achieving temporal continuity correction of the centroid. Figure 2 As shown, it specifically includes:

[0097] Step S4.1: When a storm has been continuously tracked for at least three moments, determine whether the spatial distance between its current centroid and the centroid of the previous moment exceeds a distance threshold (e.g., 3km).

[0098] Extract the centroids of the storm profiles at each time point (3 or more times) from the tracking chain of a storm to form a centroid time series table for the storm. The centroid at the current time point refers to the centroid of the storm profile at the current time point (or the current storm profile), and the centroid at the previous time point refers to the centroid of the storm profile at the previous time point (or the previous storm profile).

[0099] If the spatial distance between the current centroid and the previous centroid does not exceed the distance threshold, the current centroid position is considered reasonable and will not be corrected.

[0100] If the spatial distance between the current centroid and the previous centroid exceeds the distance threshold, the centroid correction process is initiated, and the process proceeds to step S4.2.

[0101] Step S4.2: Construct the state vector of position and velocity based on each historical centroid on the tracking chain. The specific formula is as follows:

[0102] (4)

[0103] (5)

[0104] (6)

[0105] in, , F represents the state vectors at the current and previous time steps; F represents the state transition matrix. This represents the process noise at the current moment; Indicates the centroid coordinates within the current storm profile; This represents the velocity components of the centroid within the current storm profile in the x and y directions; the superscript T indicates matrix transpose. Indicates a time interval.

[0106] The observation model is defined as follows:

[0107] (7)

[0108] (8)

[0109] in, H represents the observation vector at the current moment; H represents the observation matrix. This represents the observation noise at the current moment.

[0110] Step S4.3: Smooth and predict the centroid at the current time based on the state vector to obtain the candidate corrected centroid.

[0111] Based on formulas (4) to (8), the Kalman filtering step is performed to extract the first two elements from the updated state vector, which are the candidate corrected centroids. The Kalman filtering execution step is a prior art technique.

[0112] Step S4.4: Determine whether the candidate correction centroid falls within the current storm profile.

[0113] If the candidate corrected centroid falls within the current storm profile, the spatial distances between the candidate corrected centroid, the current centroid, and the previous centroid are calculated separately, and the one with the smaller spatial distance is selected as the current corrected centroid.

[0114] Calculate the spatial distance between the candidate correction centroid and the centroid of the previous time step, and the spatial distance between the current centroid and the centroid of the previous time step. Select the centroid corresponding to the smaller of the two spatial distances as the correction centroid of the current time step. For example, if the spatial distance between the candidate correction centroid and the centroid of the previous time step is less than the spatial distance between the current centroid and the centroid of the previous time step, then the candidate correction centroid is the correction centroid of the current time step.

[0115] If the candidate centroid falls outside the current storm profile, it will not be corrected, and the current centroid will be adopted.

[0116] This invention utilizes a tracking chain to construct a state vector and applies Kalman filtering, which not only smooths the historical trajectory but also predicts the most likely position at the current moment. This dynamic-based estimation can more scientifically address storm changes, thus providing a more reasonable candidate corrected centroid.

[0117] Provided that the candidate corrected centroid (based on dynamic prediction) is valid, by comparing the distances of the candidate corrected centroid with the current centroid and the previous centroid, and selecting the one with the smaller movement, we are essentially selecting the solution with better continuity and smoother performance with the historical centroid. This optimization strategy automatically selects the result that best maintains motion stability, maximizing the smoothing effect.

[0118] Step S5: Predict the location and shape of the storm at the next moment based on echo data within the historical storm profile.

[0119] In one specific embodiment of the present invention, predicting the location and shape of a storm at the next moment based on echo data within historical storm profiles specifically includes:

[0120] Step S5.1: Based on the echo data of the radar two-dimensional combined echo in the current storm profile and the echo data in the current storm profile of the previous n-th frame, calculate the initial velocity at the current moment using the optical flow method.

[0121] Let the current time be T. i Then the first nth frame is T. i-n In this embodiment, n is 2. Calculating the initial velocity at the current moment using optical flow based on echo data is a prior art technique.

[0122] Step S5.2: Calculate the second attenuation coefficient based on the length of the tracking chain and the preset weight of the first time step (i.e., T1). The specific calculation formula is as follows:

[0123] (9)

[0124] in, Indicates the second attenuation coefficient; This represents the weight at the first time step (T1) in the tracking chain. The weights in this embodiment The value is 0.05; This indicates the length of the tracking chain, i.e., the number of moments in the tracking chain. If the tracking chain for a certain storm corresponds to moments T1, T2, ..., T... i-1 T i (At the current moment), then =i.

[0125] Step S5.3: Calculate the normalized weight coefficients at each time step on the tracking chain based on the second decay coefficient. The specific calculation formula is as follows:

[0126] (10)

[0127] (11)

[0128] in, This represents the weight coefficient at time k on the tracking chain; This represents the normalized weight coefficient at time k on the tracking chain; This represents the weight coefficient at time l on the tracking chain.

[0129] Step S5.4: Calculate the combined velocity and direction at the current moment based on the initial velocity at the current moment, the velocities at each historical moment, and the normalized weighting coefficients at each moment. The specific calculation formula is as follows:

[0130] (12)

[0131] in, Indicates the optimization speed at the current moment; This represents the velocity at time k on the tracking chain. When k = ... hour, That is, the initial velocity at the current moment.

[0132] The initial velocity at the current moment includes velocities in two different directions, such as eastward velocity and northward velocity. Therefore, the optimized velocity at the current moment calculated by formula (12) also includes optimized velocities in two different directions. Based on the optimized velocities in two different directions, the combined velocity and direction at the current moment can be calculated.

[0133] Step S5.5: Based on the current comprehensive velocity and direction and the coordinates of each point on the current storm profile, perform spatial extrapolation on the current storm profile to obtain the predicted storm profile.

[0134] Based on the current combined velocity and direction, the coordinates of a point on the current storm profile are calculated for the next moment, thus obtaining the coordinates of all points on the predicted storm profile. This allows for the prediction of the storm's potential evolution and movement path over short timescales. The specific calculation formula is as follows:

[0135] (13)

[0136] (14)

[0137] in, This represents the coordinates of a point on the current storm profile; It indicates the coordinates of a point on the current storm profile at the next moment, or the coordinates of the corresponding point on the storm profile at the next moment; This represents the overall velocity at the current moment; Indicates the direction at the current time, for example, the angle calculated clockwise from true north; Indicates a time interval.

[0138] Extrapolation results can be used to estimate the future impact range of storms, providing a basis for early identification and short-term warning of potential heavy rainfall, thunderstorms and other hazardous weather, thereby improving the timeliness and accuracy of monitoring and forecasting.

[0139] Figure 3 This diagram illustrates the current-moment convective cell identification effect of the method of the present invention, with extrapolated direction. The color bars represent reflectivity factor intensity, the black outline represents the identified storm outline, the black lines represent the storm centroids connecting different moments in the storm tracking chain, the black dashed lines represent the extrapolated storm outline, i.e., the future storm's influence range, and the arrows indicate the direction of movement and centroid position for the next half hour. Figure 3 It can be seen that the extrapolation stability and robustness of the method of the present invention are higher.

[0140] Example 2

[0141] This invention also provides an electronic device, which includes: a memory, a processor, and a computer program or instructions stored in the memory. The processor executes the computer program or instructions to implement the storm cell tracking and extrapolation method in this invention.

[0142] Although not shown, the electronic device includes a processor that can perform various appropriate operations and processes based on programs and / or data stored in read-only memory (ROM) or loaded from a storage portion into random access memory (RAM). The processor can be a multi-core processor or may contain multiple processors. In some embodiments, the processor may include a general-purpose main processor and one or more specialized coprocessors, such as a central processing unit, graphics processing unit (GPU), neural network processor (NPU), digital signal processor (DSP), etc. Various programs and data required for device operation are also stored in RAM. The processor, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0143] The processor and memory described above are used together to execute programs / instructions stored in the memory. When the program / instructions are executed by the computer, they can implement the methods, steps, or functions described in the above embodiments.

[0144] Although not shown, embodiments of the present invention also provide a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implements the storm cell tracing and extrapolation method of the present invention.

[0145] Readable storage media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0146] The above description only discloses specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or modifications that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A storm cell tracking and extrapolation method, characterized in that, The method includes: For any storm profile in the current storm cell identification result, calculate the overlap score between it and each storm profile in the previous storm cell identification result; For storm contour pairs with an overlap score greater than 0, calculate their comprehensive similarity score based on spatial location, morphological structure, and physical properties. For a storm profile in the current storm cell identification result, select the storm profile with the highest comprehensive similarity score and greater than the minimum acceptance threshold in the previous storm cell identification result as its associated object, and establish the tracking chain relationship of storm cells in the time series. Predicting the location and shape of a storm at the next moment based on echo data within historical storm profiles, thus achieving storm extrapolation; specifically including: Based on the echo data of the radar two-dimensional combined echo in the current storm profile and the echo data in the current storm profile of the nth frame, the initial velocity at the current moment is calculated using the optical flow method. The second attenuation coefficient is calculated based on the length of the tracking chain and the preset weight of the first moment; wherein, the length of the tracking chain refers to the number of moments in the tracking chain; The normalized weight coefficients at each moment on the tracking chain are calculated based on the second decay coefficient. The combined velocity and direction at the current moment are calculated based on the initial velocity at the current moment, the velocity at each historical moment, and the normalized weighting coefficient at each moment. Based on the current combined velocity and direction, as well as the coordinates of each point on the current storm profile, the current storm profile is spatially extrapolated to obtain the next storm profile.

2. The storm cell tracking and extrapolation method according to claim 1, characterized in that, The formula for calculating the overlap score is: ； in, Indicates overlapping fractions; This represents the ratio of the overlapping area of ​​the storm profile pairs to the area of ​​a single storm profile in the current final storm cell identification result. This represents the ratio of the overlapping area of ​​the storm profile pair to the area of ​​a storm profile in the previous storm cell identification result.

3. The storm cell tracking and extrapolation method according to claim 1, characterized in that, For storm contour pairs with an overlap score greater than 0, a comprehensive similarity score based on spatial location, morphological structure, and physical properties is calculated, specifically including: For storm profile pairs with an overlap score greater than 0, calculate their center distance, area ratio, minimum distance between the two profiles, shape difference, intensity difference, echo top height difference, and intersection area ratio. Based on the preset membership function, the center distance, area ratio, minimum distance between two contours, shape difference, intensity difference, echo top height difference, and intersection area ratio are converted into corresponding similarity scores; wherein, the value range of each similarity score is [0,1], and the larger the value, the more similar they are in the corresponding features; The weighted sum of all similarity scores yields the comprehensive similarity score.

4. The storm cell tracking and extrapolation method according to claim 3, characterized in that, The expression for the membership function is: ； in, This represents the similarity score of the j-th feature; Represents the j-th feature; Let represent the mean of the j-th feature; This represents the standard deviation of the j-th feature; This represents the effective range of the j-th feature; different features have different means, standard deviations, and effective ranges.

5. The storm cell tracking and extrapolation method according to any one of claims 1 to 4, characterized in that, After establishing the tracking chain relationship of storm cells in the time series, the method further includes smoothing and predicting the centroid sequence on the tracking chain, specifically including: When a storm is continuously tracked for at least three moments, determine whether the spatial distance between its current centroid and the centroid of the previous moment exceeds the distance threshold. If it exceeds the limit, then construct a state vector of position and velocity based on each historical centroid on the tracking chain; Based on the state vector, the centroid at the current moment is smoothed and predicted to obtain the candidate corrected centroid; If the candidate correction centroid falls within the current storm profile, the spatial distances between the candidate correction centroid, the current centroid, and the previous centroid are calculated respectively, and the one with the smaller spatial distance is selected as the current correction centroid. If the candidate centroid for correction falls outside the current storm profile, no correction will be made, and the current centroid will be adopted.

6. The storm cell tracking and extrapolation method according to claim 1, characterized in that, The specific formula for calculating the second attenuation coefficient is as follows: ； in, Indicates the second attenuation coefficient; This indicates the weight at the first moment in the tracking chain. ; Indicates the number of moments in the tracking chain; The specific formula for calculating the normalized weighting coefficient is as follows: ； ； in, This represents the weight coefficient at time k on the tracking chain; This represents the normalized weight coefficient at time k on the tracking chain; This represents the weight coefficient at time l on the tracking chain.

7. An electronic device comprising a memory, a processor, and a computer program or instructions stored in the memory, characterized in that, The processor executes the computer program or instructions to implement the storm cell tracking and extrapolation method according to any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by the processor, they implement the storm cell tracking and extrapolation method as described in any one of claims 1 to 6.

Images