Nowcasting forecast method for fast-moving quasi-linear convection

Abstract

The present invention discloses a near-term extrapolation forecasting method for fast-moving quasi-linear convection. The method comprises the following steps: within half an hour of the onset of the forecast, the echo is extrapolated using the semi-Lagrangian extrapolation method based on the mixed extrapolation vector field of the convective monomer; and half an hour after the onset of the forecast, the echo is extrapolated using the semi-Lagrangian extrapolation method based on the extrapolation vector field of the convective organization propagation characteristics. This method can improve the short-term extrapolation forecast accuracy of fast-moving quasi-linear convection and significantly enhance the early warning capability of fast-moving quasi-linear convection, representing a technological breakthrough for meteorological disaster prevention and mitigation of severe convection.

Description

Technical Field

[0001] The present invention relates to a forecasting method for quasi-linear convection, in particular to a nowcasting extrapolation forecasting method for fast-moving quasi-linear convection, and belongs to the field of atmospheric science. Background Art

[0002] Fast-moving quasi-linear convection is often closely related to large-scale wind and hail disasters in my country. Therefore, it is very important to provide short-term forecasts and warnings of strong winds caused by quasi-linear convection.

[0003] Research on the relationship between the quantification of the movement characteristics of quasi-linear convection and severe weather shows that the propagation direction of cold-pool-driven cells in linear convection systems is at a large angle to the extension direction of the quasi-linear convection, resulting in rapid movement and a tendency to produce strong winds. However, existing short-term severe convection forecasting methods, whether TREC or optical flow, lack consideration of the propagation characteristics of quasi-linear convection, resulting in large deviations between the predicted thunderstorm and strong wind areas and the actual forecast timeline, as well as short forecast validity. Some case studies of severe convection have attempted to improve the extrapolation of radar echoes for severe convection processes using machine learning and deep learning methods. These methods can achieve a certain degree of echo generation and dissipation prediction, but due to differences between the generation and dissipation models and the actual convective development and propagation characteristics, although the test indicators are superior to traditional methods, image blurring persists, and the application effect still needs further improvement. Based on the multi-scale storm identification and extrapolation method, by introducing larger-scale thunderstorm tracking information (indirectly reflecting the echo development and strengthening characteristics of more mature QLCSs under the combined influence of cold pools and vertical wind shear), the inverted extrapolated field represents the overall movement direction of the thunderstorm. Although it can improve the forecast accuracy of the thunderstorm and strong wind area in the mature convection stage to a certain extent, there is still a certain difference from the actual convective propagation direction.

[0004] Dual-polarization weather radar observation data can improve the ability to identify the morphology of precipitation particles in the air, thereby obtaining the microphysical structure characteristics of cloud precipitation within the convective system. During the evolution of convective precipitation, it can sensitively and meticulously reflect the fine structural characteristics of convective cloud bodies in advance, which helps to enhance and improve the short-term forecast and early warning capabilities of convective systems. Domestic and foreign scholars have conducted a large number of case studies on the analysis of dual-polarization radar echo characteristics for different types of disaster weather. The obvious polarization parameter characteristics of different parts of the convective cell in QLCSs are conducive to the observation and analysis of the dynamic structure characteristics of wind and hail clouds, such as ρ HV Small value area, Z DR Arc, Z DR Ring, ρ HV Ring, Z DR Column and K DPColumn. Quantitative research using dual-polarization radars shows that the microphysical structure of squall lines in eastern China can be distinguished at different stages of their development and maturity based on the vertical and horizontal structural characteristics of the polarization quantity. Using my country's first S-band operational dual-polarization radar, we analyzed the polarization observation characteristics of a typical supercell storm triggered by a spring cold front in South China in 2015, and also observed Z DR Arc, and Z DR Therefore, fast-moving quasi-linear convection exhibits distinct polarization variations at different stages and locations. Tracking these polarization characteristics to obtain motion vector information can more accurately approximate the motion vector information in the propagation direction of the convection feature.

[0005] In summary, existing technologies for short-term severe convective forecasting suffer from significant bias in forecasting fast-moving quasi-linear convection. Both the TREC and optical flow methods fail to consider the propagation characteristics of quasi-linear convection, resulting in significant deviations between predicted thunderstorm and windfall locations and short forecast times. Machine learning and deep learning methods can achieve a certain level of echo generation and dissipation prediction. However, due to discrepancies between generation and dissipation models and actual convective development and propagation characteristics, while their validation performance outperforms traditional methods, they still suffer from image blurring, and their effectiveness still requires further improvement. The extrapolated field derived from multi-scale storm identification and extrapolation methods characterizes the overall movement of thunderstorms, but due to discrepancies with the actual convective propagation direction, the predicted location still exhibits significant bias. Research on the polarization characteristics of quasi-linear convection has largely focused on observational and diagnostic analysis, without directly applying polarization to short-term extrapolation forecasting. Incorporating the microphysical properties represented by the polarization characteristics of quasi-linear convection into algorithmic models presents a complex challenge. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to establish a method for inverting the polarization characteristic extrapolation field based on quasi-linear convection. By using this technology, an extrapolation field that is closer to the actual convective propagation characteristics can be inverted to improve the short-term prediction capability of quasi-linear convection.

[0007] In order to solve the above technical problems, the fast-moving quasi-linear convection near-falling extrapolation prediction method of the present invention uses the semi-Lagrangian extrapolation method to extrapolate the echo within half an hour after the onset of the prediction based on the mixed extrapolation vector field of the convection monomer; and uses the semi-Lagrangian extrapolation method to extrapolate the echo half an hour after the onset of the prediction based on the extrapolation vector field of the convection organization propagation characteristics.

[0008] In the above technical solution, the convective monomer mixing extrapolation vector field is obtained by the following steps: dividing the 3km contour echo field into N square "areas" of equal size, and performing spatial multi-layer nesting of each "area" at the first moment T1 with all "areas" at the next moment T2 to calculate the convective distribution probability similarity ratio by the maximum CSI discrimination method; finding the "area" at the moment T2 corresponding to each "area" at the moment T1 with the largest convective distribution probability similarity ratio, and the center of the "area" is the end point of the echo movement vector in the "area" during the T2-T1 period; taking into account all the movement vectors of each "area" in the echo field to obtain the movement characteristics of the entire echo field; determining the movement vectors of each "area" of the radar echo, and using these obtained movement vectors to extrapolate the echo field of the corresponding "area"; and finally obtaining the entire initial convective monomer mixing extrapolation vector field.

[0009] In the above technical solution, the extrapolated vector field of the convective organization propagation characteristics is obtained through the following steps: based on the mixed extrapolated vector field of the convective monomer, the semi-elliptical filtering technology is used to perform an overall analysis of the convective propagation characteristics to determine whether the overall filtering direction is a right-propagation characteristic or a backward propagation characteristic; based on the propagation characteristic direction, the grid points are enhanced by the semi-elliptical filtering technology to form an estimated enhanced contour surface differential displacement after the convection development; based on the extrapolated mixed vector field of the convective monomer scale, the centroid of the differential displacement at the quasi-linear convective scale after extrapolation at the first moment T1 is compared with the centroid of the differential displacement at the quasi-linear convective scale after extrapolation at the next moment T2 to calculate the centroid displacement vector; the full-field displacement vector is inverted to obtain the extrapolated vector field of the convective organization propagation characteristics.

[0010] In the above technical solution, the semi-elliptical filtering technology is to construct a filtering template of half an ellipse in the filtering direction, and assign weights of all grid points inside the semi-elliptical area to 1 and those outside to 0; the filtering process is to sum up all differential phase displacements within the semi-elliptical area, and then calculate the average to obtain the differential phase displacement after filtering at the current grid point position.

[0011] In the above technical solution, the filtering direction is determined by the direction of the convection cell mixing extrapolation vector Vdir at the grid point. The direction of rightward filtering is +90°; the direction of backward filtering is +180°.

[0012] In the above technical solution, the grid point enhancement processing steps are as follows: if the echo intensity of the grid point is between 15Dbz and 42Dbz, and the gradient of the echo in the area where the grid point is located is greater than the threshold, it is determined that differential item displacement enhancement processing is required; the larger value of the current observed differential item displacement and the differential item displacement after filtering of the current grid point is selected as the enhanced contour surface differential item displacement value; if the echo intensity of the grid point is between 15Dbz and 42Dbz, and the gradient of the echo in the area where the grid point is located is less than the threshold, the time series change characteristics of the echo intensity and the differential item displacement are further analyzed. If the observed echo intensity corresponding to the adjacent time periods is enhanced by more than 5Dbz, and the observed differential item displacement value is enhanced by more than 0.1, it is determined that differential item displacement enhancement is required, and the larger value of the current observed differential item displacement and the differential item displacement after filtering of the current grid point is selected as the enhanced contour surface differential item displacement value; otherwise, differential item displacement enhancement processing is not required, and the enhanced contour surface differential item displacement is equal to the observed differential item displacement value.

[0013] In the above technical solution, the convective monomer mixing extrapolation vector field has been quality controlled.

[0014] In the above technical solution, the total variation vector correction processing method is used to perform quality control processing on the convective monomer mixing extrapolation vector field.

[0015] In the above technical solution, the specific steps of the total variation vector correction processing method are as follows: for all grid points with echoes but missing motion vectors, the average value of the motion vector in their domain range is used to replace them; then the vector smoothing factor of each grid point is calculated to determine whether it is less than the specified threshold; and the convective monomer mixing extrapolation vector field after quality control is obtained through cyclic iteration.

[0016] The fast-moving quasi-linear convection nowcasting prediction method of the present invention has the following beneficial effects.

[0017] 1. The extrapolation vector inverted by the extrapolation vector inversion method based on polarization characteristics can be closer to the direction of convective propagation characteristics. Extrapolation based on this can improve the short-term extrapolation prediction accuracy of fast-moving quasi-linear convection.

[0018] 2. It can significantly improve the early warning capability of fast-moving quasi-linear convection, which is a technological breakthrough for meteorological disaster prevention and mitigation of severe convection. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is the extrapolated vector field of convective cell mixing inverted at 03:02 UTC on May 15, 2021.

[0020] Figure 2 This is the extrapolated vector field of convective organization propagation characteristics inverted at 03:02 UTC on May 15, 2021.

[0021] Figure 3 This is a comparison between the radar echo and the actual situation in the first hour of the forecast, where Figure a is the predicted echo inverted by the CLTREC method in business, Figure b is the predicted echo inverted by the extrapolation method of the present invention, and Figure c is the actual echo.

[0022] Figure 4 This is a comparison between the radar echo predicted for the second hour and the actual situation, where Figure a is the predicted echo inverted by the CLTREC method in business, Figure b is the predicted echo inverted by the extrapolation method of the present invention, and Figure c is the actual echo.

[0023] Figure 5 This is a comparison between the radar echo predicted for the third hour and the actual situation, where Figure a is the predicted echo inverted by the CLTREC method in business, Figure b is the predicted echo inverted by the extrapolation method of the present invention, and Figure c is the actual echo. DETAILED DESCRIPTION

[0024] The nowcasting prediction methods for fast-moving quasi-linear convection include the extrapolation vector field inversion method at the convection cell scale, the extrapolation vector field inversion method of the convection organization propagation characteristics, and the multi-scale field mixed extrapolation method.

[0025] The extrapolated vector field inverted by the extrapolated vector field inversion method at the convection cell scale is called the convection cell mixed extrapolated vector field, which represents the mixed information of the overall moving speed of the quasi-linear convection and the environmental guidance field.

[0026] The steps of this method are as follows.

[0027] 1. Divide the 3km contour echo field into several equal-sized square "areas" containing the same number of data points. Then, perform the spatial multi-layer nested convection maximum CSI discrimination method on each "area" at the first moment T1 and all "areas" at the next moment T2. The formula for calculating the convection distribution probability similarity ratio C using the multi-layer nested convection maximum CSI discrimination method is as follows:

[0028]

[0029] Among them, C r It represents the effective hit rate when the echo intensity at the first moment T1 and the next moment T2 in the analysis area is greater than the echo intensity threshold r. r is the number of grid points whose echo intensity at two moments corresponds to the echo intensity threshold r and has a similar relationship, that is, both are greater than the threshold or both are less than the threshold; D r is the number of grid points at which the relationship between the corresponding grid point echo intensity and the echo intensity threshold r at two moments is different, that is, the relationship between the two is greater than or less than.

[0030] Finally, the "region" with the greatest similarity ratio of convective distribution probability at time T2 corresponding to each "region" at time T1 is identified. The center of the "region" with the greatest similarity ratio of convective distribution probability at time T2 is the endpoint of the echo motion vector within that "region" during the T2-T1 period. If the motion vectors of each "region" within the echo field are considered, the motion characteristics of the entire echo field can be obtained. Once the motion vectors of each radar echo "region" are determined, these motion vectors are used to extrapolate the echo field of the corresponding "region," ultimately obtaining the entire initial convective cell mixing extrapolation vector field.

[0031] 2. Use the quality control method in the CLTREC method to perform quality control on the initial convective monomer mixing extrapolation vector field, and finally obtain the quality-controlled convective monomer mixing extrapolation vector field.

[0032] The quality control method is total variation vector correction processing, which draws on the ideas of optical flow method and image vector total variation restoration. It minimizes the full-field vector smoothness S(u, v) through iterative calculation, corrects and fills in the inverted moving vector field.

[0033]

[0034] in, Ω is the area of ​​all valid radar data points within the horizontal wind field inversion analysis range, V m The variational correction process is as follows: For all grid points with echoes but missing motion vectors, the average value of the motion vector within their range is used to replace them. The smoothing factor of each grid point vector is then calculated and its value is determined to be less than a specified threshold. This process is repeated iteratively to obtain the quality-controlled extrapolated vector field of the convective cell mixture.

[0035] The extrapolated vector field inverted by the inversion method of the extrapolated vector field of convective organization propagation characteristics is called the extrapolated vector field of convective organization propagation characteristics, which represents the vector information of the convective propagation characteristics of quasi-linear convection.

[0036] The steps of this method are as follows.

[0037] 1. Based on the quality-controlled mixed extrapolated vector field of the convective monomer, the semi-elliptical filtering technique is used to perform an overall analysis of the convective propagation characteristics to determine whether the overall characteristics are right-propagation or backward propagation.

[0038] The principle of semi-ellipse filtering technology is to refer to the minimum ratio of quasi-linear convection as 1 / 4, construct a half ellipse in the filtering direction, and satisfy the filtering template with a length-to-width ratio of 1:4. All grid points inside the semi-ellipse area are assigned weights of 1, and those outside are assigned weights of 0. The filtering process is to sum up all differential phase shifts in the semi-ellipse area, and then calculate the average, which is the filtered quasi-linear convection value at the current grid point position, and use KDP as the average value. f express.

[0039] The semi-elliptical filtering technique is used to filter each grid point in the whole field in both the right and backward directions to calculate the KDP. f The filtering direction here depends on the direction of the extrapolated vector V of the convection monomer mixing after quality control of the grid point. dir The direction of rightward filtering is V dir +90°; the direction of the back-filter is V dir +180°.

[0040] The KDP of all grid points in the field f Taking the average, we can calculate the right-facing KDP f Average and Backward KDP f On average, compare the sizes of the two. If the former is larger, the current overall convection is characterized by rightward propagation, otherwise the current overall convection is characterized by backward propagation.

[0041] 2. Based on the propagation characteristic direction, the semi-elliptical filtering technology is used for enhancement processing to form the estimated enhanced contour surface quasi-linear convection value after convection development, using KDP e express.

[0042] The specific steps for enhancing each grid point are as follows.

[0043] If the echo intensity of the grid point is between 15dBZ and 42dBZ, and the gradient of the echo in the area where the grid point is located is greater than the threshold, it is determined that differential phase shift enhancement processing is required; select the current observed differential phase shift and the current grid point KDP f The larger of the two values ​​is the KDP e Numeric value.

[0044] If the echo intensity of the grid point is between 15dBz and 42dBZ, and the gradient of the echo in the area where the grid point is located is less than the threshold, the time series change characteristics of the echo intensity and differential phase shift will continue to be analyzed. If the observed echo intensity enhancement corresponding to the adjacent time is greater than 5dBz, and the observed differential phase shift value enhancement is greater than 0.1, it is determined that differential phase shift enhancement is required; the current observed differential phase shift and the current grid point KDP are selected. f The larger of the two values ​​is the KDP e Numeric value.

[0045] If it is not in the above two cases, there is no need to perform differential phase shift enhancement processing, KDP e is equivalent to observing the differential phase shift.

[0046] 3. Based on the quality-controlled extrapolated mixed vector field of the convective single body scale, the centroid of the differential phase displacement at the quasi-linear convective scale after extrapolation at the first moment T1 is compared with the centroid of the differential phase displacement at the quasi-linear convective scale at the next moment T2 to calculate the centroid movement vector. This velocity is the characteristic vector of convective propagation, and the inverted full-field movement vector is the characteristic extrapolated vector field of convective organization propagation.

[0047] The processing steps for each grid point are as follows.

[0048] Based on the quality control of the convection cell scale extrapolation mixing vector field, the KDP of the grid is e The backward semi-Lagrangian method is used with a step size of 1 minute to extrapolate for 6 minutes to obtain the extrapolated enhanced contour surface differential phase displacement. p Then calculate the KDP under the quasi-linear convection scale radius centered at the grid point at the first moment T1 p The center of mass position MP P , and calculate the KDP of the quasi-linear convection scale radius at the next moment T2 at the corresponding position e The center of mass MP c , MP c Subtract MP P The displacement vector calculated is the convective propagation characteristic vector of the current time grid point.

[0049] The multi-scale field mixing extrapolation method is to use the semi-Lagrangian extrapolation method to extrapolate the echo within half an hour of the start time based on the quality-controlled convective monomer mixing extrapolation vector field; half an hour after the start time, the semi-Lagrangian extrapolation method is used to extrapolate the echo based on the convective organization propagation characteristic extrapolation vector field.

[0050] The following is an explanation with specific examples.

[0051] See also Figures 1 to 5 The method was verified using a squall line process in Nanjing, which occurred on May 15, 2021 (UTC). Figure 1 and Figure 2 The inverted extrapolated vector field for the convective cell mixing and the extrapolated vector field for the convective organization propagation characteristics, respectively, were obtained at 03:02 UTC on May 15, 2021. The extrapolated vector field for the convective cell mixing reflects the movement of the thunderstorm cell and the direction of ambient airflow, which is generally southwestward. The extrapolated vector field for the convective organization propagation characteristics represents the direction of convective development and propagation. It can be seen that strong echoes in central and southern Taiwan have a southeastward development and propagation characteristic, indicating that echoes in these areas will gradually strengthen and may move faster in the future.

[0052] The project uses the CLTREC method and the extrapolation method proposed in this project for comparison. The CLTREC method is an improved correlation coefficient method. Figure 3-5 The comparison of different 1-3 hour extrapolation forecasting methods shows that, compared with the CLTREC method, the extrapolation forecasting method proposed in this invention predicts the location of the strong convective echo of the squall line to be significantly closer to the actual observation, and the effective forecast time for this strong process is longer.

Claims

1. A fast-moving quasi-linear convection nowcasting method, characterized by: Within half an hour of the onset of the alarm, the echo is extrapolated using the semi-Lagrangian extrapolation method based on the convective cell mixing extrapolation vector field; half an hour after the onset of the alarm, the echo is extrapolated using the semi-Lagrangian extrapolation method based on the convective organization propagation characteristic extrapolation vector field. The convective cell mixing extrapolation vector field has been quality controlled and obtained by the following steps: dividing the 3km contour echo field into N square "regions" of equal size, and performing spatial multi-layer nesting of each "region" at the first time T1 with all "regions" at the next time T2 to calculate the convective distribution probability similarity ratio; finding the "region" at time T2 with the largest convective distribution probability similarity ratio corresponding to each "region" at time T1, and the center of the "region" is the end point of the echo motion vector in the "region" during the T2-T1 period; taking into account all the motion vectors of each "region" in the echo field to obtain the motion characteristics of the entire echo field; Determine the motion vectors of each "region" of the radar echo, and use these obtained motion vectors to extrapolate the echo field of the corresponding "region"; ultimately, obtain the entire initial convective cell mixing extrapolation vector field. The convective organization propagation characteristic extrapolation vector field is obtained by the following steps: Based on the convective cell mixing extrapolation vector field, use the semi-elliptical filtering technique to perform an overall analysis of the convective propagation characteristics to determine whether the overall filtering direction is a right-propagation characteristic or a backward-propagation characteristic; Based on the propagation characteristic direction, the grid points are enhanced by semi-elliptical filtering technology to form the estimated enhanced contour surface differential displacement after convection development. Based on the extrapolated mixed vector field at the convective cell scale, the centroid displacement vector is calculated by comparing the centroid of the differential displacement term at the quasi-linear convective scale after extrapolation at the first moment T1 with the centroid of the differential displacement term at the quasi-linear convective scale after extrapolation at the next moment T2. The full-field displacement vector is inverted to obtain the extrapolated vector field of the convective organization propagation characteristics.

2. The fast-moving quasi-linear convection nowcasting method according to claim 1, characterized in that: The semi-elliptical filtering technology is to construct a filtering template of half an ellipse in the filtering direction, assigning weights of all grid points inside the semi-elliptical area to 1 and those outside to 0; the filtering process is to sum up all differential phase shifts within the semi-elliptical area, and then calculate the average to obtain the differential phase shift after filtering at the current grid point position.

3. The fast-moving quasi-linear convection nowcasting prediction method according to claim 1, characterized in that: The filtering direction depends on the direction of the convective monomer mixing extrapolation vector V at the grid point. dir , the direction of rightward filtering is V dir +90°; the direction of the back-filter is V dir +180°.

4. The fast-moving quasi-linear convection nowcasting prediction method according to claim 1, characterized in that: The grid point enhancement processing steps are as follows: if the echo intensity of the grid point is between 15Dbz and 42Dbz, and the gradient of the echo in the area where the grid point is located is greater than the threshold, it is determined that differential item displacement enhancement processing is required; the larger value of the current observed differential item displacement and the current grid point filtered differential item displacement is selected as the enhanced contour surface differential item displacement value; if the echo intensity of the grid point is between 15Dbz and 42Dbz, and the gradient of the echo in the area where the grid point is located is less than the threshold, then continue to analyze the time series change characteristics of the echo intensity and the differential item displacement, if the observed echo intensity enhancement corresponding to the adjacent time times is greater than 5Dbz, and the observed differential item displacement value enhancement is greater than 0.1, it is determined that differential item displacement enhancement is required, and the larger value of the current observed differential item displacement and the current grid point filtered differential item displacement is selected as the enhanced contour surface differential item displacement value; Otherwise, there is no need to perform differential item displacement enhancement processing, and the enhanced contour surface differential item displacement is equal to the observed differential item displacement value.

5. The fast-moving quasi-linear convection nowcasting method according to claim 1, 2, 3 or 4, characterized in that: The total variation vector correction method is used to perform quality control on the extrapolated vector field of convective cell mixing.

6. The fast-moving quasi-linear convection nowcasting prediction method according to claim 5, characterized in that: The specific steps of the total variation vector correction processing method are as follows: for all grid points with echoes but missing motion vectors, the average value of the motion vector in their range is used to replace them; then the vector smoothing factor of each grid point is calculated to determine whether it is less than the specified threshold; and the convective cell mixing extrapolation vector field after quality control is obtained through cyclic iteration.

Images