Severe convective weather forecasting method and device based on spatiotemporal cross-convergence mapping

Abstract

The present application relates to a method and device for forecasting severe convective weather based on spatiotemporal cross-convergence mapping. The method comprises: first, pre-processing radar reflectivity data and meteorological element data at the onset of convection; then, calculating the correlation between the meteorological element data and the area of ​​strong echo in the radar reflectivity, screening candidate factors based on the significance level, and constructing a candidate factor set; finally, calculating the causal relationship between the candidate factors and the radar reflectivity through spatiotemporal cross-convergence mapping; if a causal relationship exists, further calculating the strength of the causal relationship, and using the candidate factor as a severe convection forecast factor. This method provides input for severe convection forecasting by analyzing the severe convection forecast factors, which helps to improve the accuracy and timeliness of severe convective weather forecasts.

Description

Technical Field

[0001] The present application relates to the technical field of nonlinear time series analysis, and in particular to a method and device for forecasting severe convective weather based on spatiotemporal cross-convergence mapping. Background Art

[0002] Severe convective phenomena (such as thunderstorms, heavy rains, and tornadoes) are important research areas in atmospheric science, and identifying causal relationships is crucial for weather forecasting and disaster prevention. Therefore, causal inference has always been a core issue in severe convection research. In statistics, correlation indicates a relationship or trend of joint variation between two variables. However, even strong correlations, influenced by confounding variables, cannot confirm causality. More causal models are needed to identify the characteristics between data. In recent years, convergent cross-mapping (CCM) has provided a new approach to identifying causal relationships. Its core element is to abstract time series into chaotic systems. Shadow manifolds, based on phase space reconstruction attractors, can effectively characterize asymmetric causal relationships between system state variables. However, the field of severe convection requires high timeliness, and this method is limited in processing severe convection data by low computational efficiency and difficulty scalable to spatiotemporal data. Summary of the Invention

[0003] Based on this, it is necessary to provide a severe convective weather forecasting method and device based on spatiotemporal cross-convergence mapping to address the above technical problems.

[0004] A severe convective weather forecasting method based on spatiotemporal cross-convergence mapping, the method comprising:

[0005] Obtain radar reflectivity data and meteorological element data at the onset of convection and perform preprocessing.

[0006] The preprocessed radar reflectivity data and meteorological element data were subjected to correlation analysis, candidate factors were screened out according to the significance level, and a candidate factor set was constructed.

[0007] A prediction pixel is randomly selected from the radar reflectivity. Based on the delayed embedding of each candidate factor pixel and the prediction pixel, a spatiotemporal cross-convergence mapping algorithm is used to calculate whether there is a causal relationship between each candidate factor and the radar reflectivity. If there is a causal relationship, the strength of the causal relationship is further calculated, and the candidate factor is used as a severe convection forecast factor.

[0008] All severe convective forecast factors are used in severe convective weather forecast to obtain severe convective weather forecast results.

[0009] In one embodiment, obtaining radar reflectivity data and meteorological element data at the time of convection initiation and performing preprocessing include:

[0010] Obtain radar reflectivity data and meteorological element data when convection is incipient.

[0011] The radar reflectivity data and meteorological element data are temporally and spatially resampled to keep their temporal and spatial resolutions consistent.

[0012] Pixels with radar reflectivity greater than 35dBz are marked as having initial convection; pixels are the smallest spatial and temporal units of radar reflectivity data and meteorological element data.

[0013] In one embodiment, correlation analysis is performed on the pre-processed radar reflectivity data and meteorological element data, and candidate factors are screened out according to the significance level to construct a candidate factor set, including:

[0014] The meteorological element data and the radar reflectivity values ​​of the convection primary area were extracted, and the Pearson correlation coefficient was calculated using the numerical relationship between the pixels.

[0015] The statistics of the Student distribution were constructed based on the Pearson correlation coefficient to conduct significance test.

[0016] Candidate factors are screened out according to the significance level to obtain a candidate factor set.

[0017] In one embodiment, a prediction pixel is randomly selected from the radar reflectivity. Based on the delayed embedding of each candidate factor pixel and the prediction pixel, a spatiotemporal cross-convergence mapping algorithm is used to calculate whether there is a causal relationship between each candidate factor and the radar reflectivity. If there is a causal relationship, the strength of the causal relationship is further calculated, and the candidate factor is used as a severe convection prediction factor, including:

[0018] Randomly select several prediction pixels in the radar reflectivity to form the prediction pixel set:

[0019] ;

[0020] in, Represents the set of predicted pixel true values, Indicates the The radar reflectivity value of the predicted pixel, Indicates the number of randomly selected prediction pixels, Indicates the The spatiotemporal index of the predicted pixels, , Lon 、 Lat and Time Represents longitude, latitude and time respectively.

[0021] Traverse each candidate factor and use the spatiotemporal index of the predicted pixel to find the corresponding pixel in the current candidate factor; calculate the Euclidean distance between each pixel in the candidate factor and the delayed embedding of the corresponding pixel to obtain the delayed embedding distance.

[0022] According to the current candidate factors, predicted pixels and the delayed embedding distance of the corresponding pixels, spatiotemporal cross-convergence mapping is used to determine the causal relationship.

[0023] If there is a causal relationship between the current candidate factor and radar reflectivity, the strength of the causal relationship is calculated, and the candidate factor is used as a severe convection forecast factor.

[0024] Continue to make causal judgment on the next candidate factor until all candidate factors are traversed.

[0025] In one embodiment, a causal relationship is determined using a spatiotemporal cross-convergence mapping based on the current candidate factor, the predicted pixel, and the delayed embedding distance of the corresponding pixel, including:

[0026] Traverse each predicted pixel and find the closest distance to each predicted pixel from the current candidate factors according to the delayed embedding distance of the corresponding pixel k pixels, which are called similar pixels; obtain the spatiotemporal index of similar pixels and the delayed embedding distance with the current predicted pixel.

[0027] According to the delayed embedding distance between the spatiotemporal index of similar pixels and the current predicted pixel, the weight of similar pixels is calculated as:

[0028] ;

[0029] in, Yes The normalized weights, Indicates selected The delayed embedding distance between the pixel and the current predicted pixel, represents the unnormalized weights.

[0030] The estimated value of each prediction pixel is calculated based on the weight:

[0031] ;

[0032] in, represents the estimated radar reflectivity value predicted based only on the numerical relationship between the current candidate factor and the radar reflectivity; express The radar reflectivity value corresponding to the similar pixels under the spatiotemporal index.

[0033] After completing the traversal of the predicted pixels, calculate the difference between the estimated value of the predicted pixel and the true value. PearsonCorrelation coefficient and calculate the significance level corresponding to the correlation coefficient.

[0034] A significance test is performed according to the significance level corresponding to the correlation coefficient. If the significance test is passed, there is a causal relationship between the current candidate factor and the radar reflectivity; otherwise, there is no causal relationship between the current candidate factor and the radar reflectivity.

[0035] In one embodiment, the process of calculating the strength of the causal relationship includes: traversing the current candidate factors with a window of a preset size, randomly selecting m windows of corresponding sizes, and use spatiotemporal cross-convergence mapping to evaluate in each window, and obtain m The default window size is three-dimensional, including all time segments and part of the latitude and longitude grid. The shape of the window is L×L×T In the form of L×L represents the geographic space range, and T represents all time segments.

[0036] A significance test is performed on the statistics of the Student distribution constructed by the mean of the correlation coefficient; the statistics of the Student distribution are:

[0037] ;

[0038] in, is the statistic representing the Student distribution, represents the mean of the correlation coefficient, is the number of randomly selected prediction pixels.

[0039] Iterate the window selection and significance test in a binary search manner until the minimum window when the correlation coefficient mean passes the significance test is found. L Value as the strength of the causal relationship.

[0040] A severe convective weather forecasting device based on spatiotemporal cross-convergence mapping, the device comprising:

[0041] The data preprocessing module is used to obtain radar reflectivity data and meteorological element data at the onset of convection and perform preprocessing.

[0042] The candidate factor set construction module is used to perform correlation analysis on the preprocessed radar reflectivity data and meteorological element data, screen out candidate factors according to the significance level, and construct a candidate factor set.

[0043] The causal relationship judgment and severe convection prediction factor determination module is used to randomly select prediction pixels in the radar reflectivity. Based on the delayed embedding of each candidate factor pixel and the prediction pixel, a spatiotemporal cross-convergence mapping algorithm is used to calculate whether there is a causal relationship between each candidate factor and the radar reflectivity. If there is a causal relationship, the strength of the causal relationship is further calculated, and the candidate factor is used as a severe convection prediction factor.

[0044] The severe convective weather forecast module is used to apply all severe convective weather forecast factors to severe convective weather forecast to obtain severe convective weather forecast results.

[0045] The above-mentioned severe convective weather forecasting method and device based on spatiotemporal cross-convergence mapping includes: first, preprocessing radar reflectivity data and meteorological element data at the onset of convection; then, calculating the correlation between the meteorological element data and the areas of strong echoes in the radar reflectivity, screening candidate factors based on the significance level, and constructing a candidate factor set; finally, calculating the causal relationship between the candidate factors and the radar reflectivity through spatiotemporal cross-convergence mapping; if a causal relationship exists, further calculating the strength of the causal relationship, and using the candidate factor as a severe convection forecast factor. This method provides input for severe convection forecasts by analyzing the severe convection forecast factors, helping to improve the accuracy and timeliness of severe convective weather forecasts. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 1 is a flow chart of a severe convective weather forecasting method based on spatiotemporal cross-convergence mapping in one embodiment;

[0047] Figure 2 A schematic diagram of delayed embedding in one embodiment;

[0048] Figure 3 A flow chart of spatiotemporal cross-convergence mapping in one embodiment;

[0049] Figure 4 Schematic diagram of the overall process of a severe convective weather forecasting method based on spatiotemporal cross-convergence mapping in another embodiment. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0051] In one embodiment, Figure 1 As shown, a severe convective weather forecasting method based on spatiotemporal cross-convergence mapping is provided, which includes the following steps:

[0052] Step 100: Obtain radar reflectivity data and meteorological element data at the time of convection initiation and perform preprocessing.

[0053] Specifically, in an observation of the incipient convection, operations such as reading, resolution scaling, and annotation are performed on the radar reflectivity data and meteorological element data respectively.

[0054] Step 102: Perform correlation analysis on the pre-processed radar reflectivity data and meteorological element data, screen out candidate factors based on the significance level, and construct a candidate factor set.

[0055] Step 104: Randomly select prediction pixels from the radar reflectivity, and use the spatiotemporal cross-convergence mapping algorithm to calculate whether there is a causal relationship between each candidate factor and the radar reflectivity based on the delayed embedding of each candidate factor pixel and the prediction pixel; if there is a causal relationship, further calculate the strength of the causal relationship, and use the candidate factor as a severe convection prediction factor.

[0056] Specifically, the predicted pixels are randomly selected in the radar reflectivity; the spatiotemporal index of the pixels includes: longitude, latitude and time, that is, , where i is the spatiotemporal index of the pixel, Lon 、 Lat and Time Represents longitude, latitude and time respectively.

[0057] Delayed embedding distance calculation: Calculate the delayed embedding of each pixel in the current candidate factor and use the spatiotemporal index of the predicted pixel to find the corresponding pixel. The delayed embedding distance can be calculated using the Euclidean distance of the delayed embeddings, which measures the similarity between pixels.

[0058] Causal relationship judgment: Use spatiotemporal cross-convergence mapping to judge causal relationships. Specifically, the global region of the current candidate factor is used as the calculation area. For each predicted pixel, similar pixels are searched in the candidate factor. The value of each predicted pixel is estimated using the delayed embedding distance as the weight. The difference between the estimated value and the true value is calculated. Pearson Correlation coefficient and significance level. If the significance test is passed, it indicates that there is a causal relationship between the current candidate factor and radar reflectivity, otherwise there is no causal relationship.

[0059] Causal relationship strength calculation: After determining a causal relationship, the candidate factor's window is used as the calculation area, and spatiotemporal cross-convergence mapping is used to calculate the results for windows of different sizes. The minimum significant window size is used to characterize the strength of the causal relationship. This candidate factor is then used as a severe convection predictor.

[0060] Step 106: All severe convection prediction factors are used for severe convective weather forecasting to obtain severe convective weather forecast results.

[0061] The above-mentioned severe convective weather forecasting method based on spatiotemporal cross-convergence mapping includes: first, preprocessing radar reflectivity data and meteorological element data at the onset of convection; then, calculating the correlation between the meteorological element data and the areas of strong echoes in the radar reflectivity, screening candidate factors based on the significance level, and constructing a candidate factor set; finally, calculating the causal relationship between the candidate factors and the radar reflectivity through spatiotemporal cross-convergence mapping; if a causal relationship exists, further calculating the strength of the causal relationship, and using the candidate factor as a severe convection forecast factor. By analyzing the severe convection forecast factors, this method provides input for severe convection forecasting, helping to improve the accuracy and timeliness of severe convective weather forecasts.

[0062] In one embodiment, step 100 includes: obtaining radar reflectivity data and meteorological element data at the time of convection initiation; performing spatiotemporal resampling on the radar reflectivity data and meteorological element data so that the spatiotemporal resolution of the two remains consistent; marking pixels with radar reflectivity greater than 35 dBz as the occurrence of convection initiation; the pixel is the minimum spatiotemporal unit of the radar reflectivity data and meteorological element data.

[0063] Specifically, radar reflectivity data and meteorological element data at the onset of convection are obtained and preprocessed, including:

[0064] Step 1-1, data reading: Based on the observation of the initiation of convection, read the radar reflectivity data and meteorological element data according to time and space.

[0065] Step 1-2, marking of convection initiation: The smallest unit of radar reflectivity and meteorological element data is called pixel. Pixels with radar reflectivity greater than 35dBz are marked as experiencing convection initiation, otherwise pixels are marked as not experiencing convection initiation.

[0066] Steps 1-3, data scaling: resample the radar reflectivity and meteorological element data in time and space to keep the temporal and spatial resolutions of the two consistent.

[0067] In one embodiment, step 102 includes: extracting meteorological element data and radar reflectivity values ​​of the convection initiation area, and calculating the Pearson correlation coefficient using the numerical relationship between pixels; constructing a Student distribution statistic based on the Pearson correlation coefficient to perform a significance test; and screening candidate factors based on the significance level to obtain a candidate factor set.

[0068] Specifically, Pearson Correlation coefficient calculation: extract meteorological element data and radar reflectivity values ​​in the initial convection area, and calculate the correlation coefficient using the numerical relationship between pixels. Pearson Correlation coefficient, Pearson The correlation coefficient calculation formula is:

[0069] ;

[0070] in, X and Y They represent the meteorological elements and radar reflectivity values ​​in the convection initiation area respectively. cov is the covariance, and They represent the standard deviation of meteorological elements and radar reflectivity values ​​in the flow initiation area respectively. r express Pearson Correlation coefficient.

[0071] Correlation coefficient significance test: calculate Pearson After obtaining the correlation coefficient, a significance test can be performed by constructing the statistics of the Student distribution.

[0072] ;

[0073] in, represents the statistics of the student distribution, num Represents the number of pixels in the convection initiation area.

[0074] In one embodiment, step 104 includes randomly selecting a number of prediction pixels in the radar reflectivity to form a set of prediction pixels:

[0075] ;

[0076] in, Represents the set of predicted pixel true values, Indicates the The radar reflectivity value of the predicted pixel, n Indicates the number of randomly selected prediction pixels, Indicates the The spatiotemporal index of the predicted pixels, , Lon 、 Lat and Time Represents longitude, latitude and time respectively;

[0077] Traverse each candidate factor and use the spatiotemporal index of the predicted pixel to find the corresponding pixel in the current candidate factor. Calculate the Euclidean distance between each pixel in the candidate factor and the delayed embedding of the corresponding pixel to obtain the delayed embedding distance. Use spatiotemporal cross-convergence mapping to determine the causal relationship based on the delayed embedding distance of the current candidate factor, the predicted pixel, and the corresponding pixel. If there is a causal relationship between the current candidate factor and the radar reflectivity, calculate the strength of the causal relationship and use the candidate factor as a severe convection forecast factor. Continue causal judgment on the next candidate factor until all candidate factors have been traversed.

[0078] Specifically, delay embedding means that the delay embedding of each pixel includes all pixels adjacent to it in space and time. After the delay embedding operation, each pixel corresponds to a three-dimensional matrix, where the three dimensions correspond to all adjacent pixels on the longitude, latitude, and time axes respectively. Delay embedding is as follows Figure 2 shown.

[0079] Spatiotemporal index matching: Find the spatiotemporal index of the predicted pixel in the current candidate factor The corresponding pixels are called “corresponding pixels”.

[0080] Euclidean distance quantization: store the delay embedding of each corresponding pixel, obtain the delay embedding of each pixel in the candidate factor, and calculate the Euclidean distance between them. Formally, the value of the Euclidean distance can be expressed as the second-order norm of the matrix:

[0081] ;

[0082] in, It represents the distance between two pixels. X 1 and X 2 represents the delayed embedding of two pixels.

[0083] In one embodiment, a causal relationship is determined using a spatiotemporal cross-convergence mapping based on the current candidate factor, the predicted pixel, and the delayed embedding distance of the corresponding pixel, including:

[0084] Traverse each predicted pixel and find the closest distance to each predicted pixel from the current candidate factors according to the delayed embedding distance of the corresponding pixel Pixels are called similar pixels; the spatiotemporal index of similar pixels and the delayed embedding distance with the current predicted pixel are obtained; according to the spatiotemporal index of similar pixels and the delayed embedding distance with the current predicted pixel, the weight of similar pixels is calculated as:

[0085] ;

[0086] in, Yes The normalized weights, Indicates selected The delayed embedding distance between the pixel and the current predicted pixel, represents the unnormalized weights.

[0087] The estimated value of each prediction pixel is calculated based on the weight:

[0088] ;

[0089] in, represents the estimated radar reflectivity value predicted based only on the numerical relationship between the current candidate factor and the radar reflectivity; express The radar reflectivity value corresponding to the similar pixels under the spatiotemporal index.

[0090] After completing the traversal of the predicted pixels, calculate the difference between the estimated value of the predicted pixel and the true value. Pearson Correlation coefficient, and calculate the significance level corresponding to the correlation coefficient; perform a significance test based on the significance level corresponding to the correlation coefficient. If the significance test passes, there is a causal relationship between the current candidate factor and the radar reflectivity; otherwise, there is no causal relationship between the current candidate factor and the radar reflectivity.

[0091] The process of spatiotemporal cross-convergence mapping is as follows Figure 3 shown.

[0092] Specifically, the steps for determining causality include:

[0093] (1) Similar pixel search: traverse each predicted pixel and find the closest pixel to each predicted pixel from the candidate factors according to the calculated delayed embedding distance. pixels and obtain their spatiotemporal indexes and distances from the current predicted pixel.

[0094] (2) Distance weight solution: According to The delayed embedding distance of pixels is calculated using the weight expression of similar pixels. .

[0095] (3) Prediction pixel estimation: calculation Estimated value of , It is through this The radar reflectivity value of the pixel under the corresponding spatiotemporal index is multiplied by the corresponding weight The calculation formula is shown in the above expression of the estimated value of the predicted pixel.

[0096] (4) Significance test: After completing the traversal of the predicted pixels, each composition ,calculate and between Pearson Correlation coefficient , and calculate The significance level of express estimated value.

[0097] In one embodiment, the process of calculating the strength of the causal relationship in step 104 includes: traversing the current candidate factors with a window of a preset size, randomly selecting m windows of corresponding sizes, and use spatiotemporal cross-convergence mapping to evaluate in each window, and obtain m The default window size is three-dimensional, including all time segments and part of the latitude and longitude grid. The shape of the window is L×L×T In the form of L×L Represents the geographic space range, T represents all time segments; the significance test is performed on the statistics of the student distribution constructed by the mean of the correlation coefficient; the statistics of the student distribution are:

[0098] ;

[0099] in, is the statistic representing the Student distribution, represents the mean of the correlation coefficient, is the number of randomly selected prediction pixels.

[0100] Iterate the window selection and significance test in a binary search manner until the minimum window when the correlation coefficient mean passes the significance test is found. L Value as the strength of the causal relationship.

[0101] Specifically, the window setting is: the window is three-dimensional, including all time segments and part of the latitude and longitude grid. Its shape can be expressed in the form of L×L×T, where L×L represents the geographic space range and T represents all time segments. After determining the window size, randomly select from the current candidate factors m A window of corresponding size, and traverse this m window.

[0102] Window prediction result calculation: After a window is selected, repeat the spatiotemporal cross convergence mapping. m After all the areas of the same size are traversed, take this m indivual The mean .right Construct the statistics of the student distribution to perform a significance test. The statistics of the student distribution are shown in the above statistical expression of the student distribution.

[0103] Causal relationship strength output: Iterate the window size setting and window prediction result calculation in a binary search manner until the The minimum window when passing the significance test. Specifically, if the current window calculates Significantly, then reduce L, on the contrary increases L After the traversal is completed, the output L The value of , as a representation of the strength of the causal relationship.

[0104] The overall process of the severe convective weather forecast method based on spatiotemporal cross-convergence mapping is as follows: Figure 4 shown.

[0105] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0106] In one embodiment, a severe convective weather forecasting device based on spatiotemporal cross-convergence mapping is provided, comprising: a data preprocessing module, a candidate factor set construction module, a causal relationship judgment and severe convective forecast factor determination module, and a severe convective weather forecasting module, wherein:

[0107] The data preprocessing module is used to obtain radar reflectivity data and meteorological element data at the onset of convection and perform preprocessing.

[0108] The candidate factor set construction module is used to perform correlation analysis on the preprocessed radar reflectivity data and meteorological element data, screen out candidate factors according to the significance level, and construct a candidate factor set.

[0109] The causal relationship judgment and severe convection prediction factor determination module is used to randomly select prediction pixels in the radar reflectivity. Based on the delayed embedding of each candidate factor pixel and the prediction pixel, a spatiotemporal cross-convergence mapping algorithm is used to calculate whether there is a causal relationship between each candidate factor and the radar reflectivity. If there is a causal relationship, the strength of the causal relationship is further calculated, and the candidate factor is used as a severe convection prediction factor.

[0110] The severe convective weather forecast module is used to apply all severe convective weather forecast factors to severe convective weather forecast to obtain severe convective weather forecast results.

[0111] In one embodiment, the data preprocessing module is further used to obtain radar reflectivity data and meteorological element data at the time of convection initiation; perform spatiotemporal resampling on the radar reflectivity data and meteorological element data to keep the spatiotemporal resolution of the two consistent; mark pixels with radar reflectivity greater than 35dBz as having occurred convection initiation; and pixels are the minimum spatiotemporal units of radar reflectivity data and meteorological element data.

[0112] In one embodiment, the candidate factor set construction module is also used to extract meteorological element data and radar reflectivity values ​​of the convection initiation area, calculate the Pearson correlation coefficient using the numerical relationship between pixels; construct a Student distribution statistic based on the Pearson correlation coefficient to perform a significance test; and screen out candidate factors based on the significance level to obtain a candidate factor set.

[0113] In one embodiment, the causal relationship judgment and severe convection prediction factor determination module is also used to randomly select prediction pixels in the radar reflectivity; the prediction pixels are as shown in the above-mentioned prediction pixel expression; each candidate factor is traversed, and the corresponding pixel in the current candidate factor is found using the spatiotemporal index of the prediction pixel; the Euclidean distance between each pixel in the candidate factor and the delayed embedding of the corresponding pixel is calculated to obtain the delayed embedding distance; based on the current candidate factor, the prediction pixel and the delayed embedding distance of the corresponding pixel, the causal relationship is judged using spatiotemporal cross-convergence mapping; if there is a causal relationship between the current candidate factor and the radar reflectivity, the strength of the causal relationship is calculated, and the candidate factor is used as the severe convection prediction factor; and the causal judgment is continued for the next candidate factor until all candidate factors are traversed.

[0114] In one embodiment, the causal relationship judgment and severe convection prediction factor determination module is further used to traverse each prediction pixel and find the closest distance to each prediction pixel from the current candidate factors according to the delayed embedding distance of the corresponding pixel. k pixels, which are called similar pixels; obtain the spatiotemporal index of the similar pixel and the delayed embedding distance with the current predicted pixel; according to the spatiotemporal index of the similar pixel and the delayed embedding distance with the current predicted pixel, the weight of the similar pixel is calculated using the weight expression of the similar pixel as mentioned above.

[0115] According to the weight, the estimated value of each predicted pixel is calculated using the estimated value expression of the predicted pixel as above; after completing the traversal of the predicted pixels, the difference between the estimated value of the predicted pixel and the true value is calculated. Pearson Correlation coefficient, and calculate the significance level corresponding to the correlation coefficient; perform a significance test based on the significance level corresponding to the correlation coefficient. If the significance test passes, there is a causal relationship between the current candidate factor and the radar reflectivity; otherwise, there is no causal relationship between the current candidate factor and the radar reflectivity.

[0116] In one embodiment, the process of calculating the strength of the causal relationship in the causal relationship judgment and severe convection forecast factor determination module includes: traversing the current candidate factors with a window of a preset size, randomly selecting m windows of corresponding sizes, and use spatiotemporal cross-convergence mapping to evaluate in each window, and obtain m The default window size is three-dimensional, including all time segments and part of the latitude and longitude grid. The shape of the window is L×L×T In the form of L×L Represents the geographic space range, T represents all time segments; construct the statistics of the student distribution of the correlation coefficient mean and perform a significance test; the statistics of the student distribution are shown in the above statistical expression of the student distribution; iterate the window selection and significance test in a binary search manner until the minimum window when the correlation coefficient mean passes the significance test is found, and use the minimum window L Value as the strength of the causal relationship.

[0117] Regarding the specific limitations of the severe convective weather forecasting device based on spatiotemporal cross-convergence mapping, please refer to the limitations of the severe convective weather forecasting method based on spatiotemporal cross-convergence mapping above, which will not be repeated here. The various modules in the above-mentioned severe convective weather forecasting device based on spatiotemporal cross-convergence mapping can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0118] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0119] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A severe convective weather forecasting method based on spatiotemporal cross-convergence mapping, characterized in that: The method comprises: Obtain radar reflectivity data and meteorological element data at the onset of convection and perform preprocessing; Perform correlation analysis on the pre-processed radar reflectivity data and meteorological element data, screen out candidate factors based on the significance level, and construct a candidate factor set; A prediction pixel is randomly selected from the radar reflectivity. Based on the delayed embedding of each candidate factor pixel and the prediction pixel, a spatiotemporal cross-convergence mapping algorithm is used to calculate whether there is a causal relationship between each candidate factor and the radar reflectivity. If there is a causal relationship, the strength of the causal relationship is further calculated, and the candidate factor is used as a severe convection prediction factor. Using all the severe convection prediction factors for severe convective weather forecasting to obtain a severe convective weather forecast result; The process of calculating the strength of causal relationships includes: Traverse the current candidate factors with a window of preset size and randomly select from the current candidate factors m windows of corresponding sizes, and use spatiotemporal cross-convergence mapping to evaluate in each window, and obtain m The window of the preset size is three-dimensional, including all time segments and part of the longitude and latitude grid, and the shape of the window is L×L×T In the form of L×L represents the geographic space range, and T represents all time segments; A significance test is performed on the statistics of the student distribution constructed by the mean of the correlation coefficient; the statistics of the student distribution are: ； in, t is the statistic representing the Student distribution, represents the mean of the correlation coefficient, n is the number of randomly selected prediction pixels; Iterate the window selection and significance test in a binary search manner until the minimum window when the correlation coefficient mean passes the significance test is found. L Value as the strength of the causal relationship.

2. The severe convective weather forecasting method based on spatiotemporal cross-convergence mapping according to claim 1, characterized in that: Obtain radar reflectivity data and meteorological element data at the onset of convection and perform preprocessing, including: Obtain radar reflectivity data and meteorological element data at the onset of convection; The radar reflectivity data and meteorological element data are temporally and spatially resampled to keep the temporal and spatial resolutions of the two consistent; Pixels with radar reflectivity greater than 35 dBz are marked as having incipient convection; the pixels are the smallest space-time units of radar reflectivity data and meteorological element data.

3. The severe convective weather forecasting method based on spatiotemporal cross-convergence mapping according to claim 1, characterized in that: Correlation analysis is performed on the pre-processed radar reflectivity data and meteorological element data. Candidate factors are screened out according to the significance level and a candidate factor set is constructed, including: Extract meteorological element data and radar reflectivity values ​​in the convection initiation area, and calculate the Pearson correlation coefficient using the numerical relationship between pixels; Constructing a student distribution statistic based on the Pearson correlation coefficient to perform a significance test; Candidate factors are screened out according to the significance level to obtain a candidate factor set.

4. The severe convective weather forecasting method based on spatiotemporal cross-convergence mapping according to claim 1, characterized in that: A prediction pixel is randomly selected from the radar reflectivity, and a spatiotemporal cross-convergence mapping algorithm is used to calculate whether there is a causal relationship between each candidate factor and the radar reflectivity based on the delayed embedding of each candidate factor pixel and the prediction pixel; If there is a causal relationship, the strength of the causal relationship is further calculated, and the candidate factor is used as a severe convection prediction factor, including: Randomly select several prediction pixels in the radar reflectivity to form the prediction pixel set: ； in, Represents the set of predicted pixel true values, Y i Indicates the i The radar reflectivity value of the predicted pixel, n Indicates the number of randomly selected prediction pixels, i Indicates the i The spatiotemporal index of the predicted pixels, , Lon 、 Lat and Time Represents longitude, latitude and time respectively; Traverse each candidate factor and use the spatiotemporal index of the predicted pixel to find the corresponding pixel in the current candidate factor; calculate the Euclidean distance between each pixel in the candidate factor and the corresponding pixel delay embedding to obtain the delay embedding distance; Based on the current candidate factors, predicted pixels and the delayed embedding distance of the corresponding pixels, the spatiotemporal cross-convergence mapping is used to determine the causal relationship; If there is a causal relationship between the current candidate factor and the radar reflectivity, the strength of the causal relationship is calculated, and the candidate factor is used as a severe convection prediction factor; Continue to make causal judgment on the next candidate factor until all candidate factors are traversed.

5. The severe convective weather forecasting method based on spatiotemporal cross-convergence mapping according to claim 1, characterized in that: Based on the current candidate factors, predicted pixels, and the delayed embedding distance of the corresponding pixels, the spatiotemporal cross-convergence mapping is used to determine the causal relationship, including: Traverse each predicted pixel and find the closest distance to each predicted pixel from the current candidate factors according to the delayed embedding distance of the corresponding pixel k pixels, which are called similar pixels; obtain the spatiotemporal index of similar pixels and the delayed embedding distance with the current predicted pixel; According to the delayed embedding distance between the spatiotemporal index of similar pixels and the current predicted pixel, the weight of similar pixels is calculated as: ； in, w k Yes u k The normalized weights, d k Indicates selected k The delayed embedding distance between the pixel and the current predicted pixel, u k represents the unnormalized weight; The estimated value of each predicted pixel calculated based on the weights is: ； in, represents the estimated radar reflectivity value predicted based only on the numerical relationship between the current candidate factor and the radar reflectivity; express k The radar reflectivity value corresponding to similar pixels under the spatiotemporal index; After completing the traversal of the predicted pixels, calculate the difference between the estimated value of the predicted pixel and the true value. Pearson Correlation coefficient, and calculate the significance level corresponding to the correlation coefficient; A significance test is performed according to the significance level corresponding to the correlation coefficient. If the significance test is passed, there is a causal relationship between the current candidate factor and the radar reflectivity; otherwise, there is no causal relationship between the current candidate factor and the radar reflectivity.

6. A severe convective weather forecasting device based on spatiotemporal cross-convergence mapping, characterized in that: The device comprises: The data preprocessing module is used to obtain radar reflectivity data and meteorological element data at the onset of convection and perform preprocessing; The candidate factor set construction module is used to perform correlation analysis on the pre-processed radar reflectivity data and meteorological element data, screen out candidate factors according to the significance level, and construct a candidate factor set; A causal relationship judgment and severe convection prediction factor determination module is used to randomly select prediction pixels from the radar reflectivity, and use a spatiotemporal cross-convergence mapping algorithm to calculate whether there is a causal relationship between each candidate factor and the radar reflectivity based on the delayed embedding of each candidate factor pixel and the prediction pixel; if there is a causal relationship, the strength of the causal relationship is further calculated, and the candidate factor is used as a severe convection prediction factor; A severe convective weather forecast module, configured to apply all of the severe convective weather forecast factors to severe convective weather forecasting to obtain a severe convective weather forecast result; The process of calculating the strength of causal relationship in the causal relationship judgment and severe convection forecast factor determination module includes: traversing the current candidate factors with a window of preset size, randomly selecting from the current candidate factors m windows of corresponding sizes, and use spatiotemporal cross-convergence mapping to evaluate in each window, and obtain m The window of the preset size is three-dimensional, including all time segments and part of the longitude and latitude grid, and the shape of the window is L×L×T In the form of L×L Represents the geographic space range, T represents all time segments; the significance test is performed on the statistics of the student distribution constructed by the mean of the correlation coefficient; the statistics of the student distribution are: ； in, t is the statistic representing the Student distribution, represents the mean correlation coefficient, n is the number of randomly selected prediction pixels; Iterate the window selection and significance test in a binary search manner until the minimum window when the correlation coefficient mean passes the significance test is found. L Value as the strength of the causal relationship.

Images