Vertical velocity quantitative estimation method and system based on dual-polarization radar

Abstract

The invention provides a vertical velocity quantitative estimation method and system based on a dual-polarization radar, and the method employs the multi-polarization parameters of the dual-polarization radar, combines a deep learning technology with radar meteorological physical knowledge, builds a nonlinear relation between the multi-polarization parameters of the dual-polarization radar and the vertical velocity, and achieves the quantitative estimation of the vertical velocity. Quantitative estimation of the three-dimensional vertical velocity profile is realized, and a quantitative analysis product is provided for convective weather forecast and early warning. According to the method, a differential reflectivity column, a differential phase column recognition algorithm and a dual-polarization radar hydrogel classification algorithm are combined to screen an upflow area of convective weather, and a data set of dual-polarization radar observation characteristic data and vertical velocity of the upflow area is established. Transform is used as a prediction model, a space-time attention mechanism is added, and multi-time step information is used as input, so that extraction of time features observed by the dual-polarization radar is facilitated, and the vertical velocity prediction accuracy is improved.

Description

Technical Field

[0001] This invention belongs to the field of atmospheric science and technology, and relates to a method and system for quantitative estimation of vertical velocity based on dual-polarization radar. Background Technology

[0002] Vertical velocity directly influences the formation and development of convective clouds and precipitation, and plays a crucial role in the parameterization schemes of thermodynamic and microphysical processes. Vertical velocity not only provides the driving force for condensation formation but also determines the degree of atmospheric instability and the development potential of convection; therefore, it is essential for the forecasting and early warning of severe convective weather.

[0003] Although vertical velocity is a core physical quantity in the dynamic processes of convective systems, its direct measurement still faces technical obstacles. Therefore, some studies have used remote sensing observations, such as the reflectivity factor of weather radar, to retrieve vertical velocity information. The three-dimensional structure of the reflectivity factor (e.g., centroid, spatial gradient) reflects the interaction between updrafts and precipitation particles. Reference 1: Chase et al. (2024) used the UNET deep learning method to retrieve the maximum vertical velocity from the reflectivity factor. Reference 1: Chase, RJ, A. McGovern, C.R. Homeyer, P.J. Marinascu, and C.K. Potvin. Machine Learning Estimation of Maximum Vertical Velocity from Radar. Artificial Intelligence for the Earth Systems, 2024, 3, 230095. However, existing techniques retrieve vertical velocity profiles using only the reflectivity factor, lacking sufficient physical constraints and resulting in significant errors; furthermore, this technique can only retrieve the maximum vertical velocity across the entire atmosphere.

[0004] Compared to traditional weather radar, dual-polarization radar can detect additional information about particle shape, size, and orientation. Through features such as differential reflectivity bars, differential phase bars, and weak correlation coefficient loops, it can effectively indicate the location, intensity, and evolution trend of updrafts during convection development, providing a new approach for vertical velocity inversion. However, there is still no effective method to utilize this information for efficient inversion. Existing technical solutions are still limited to statistically analyzing the correlation of single variables, and in practical operations, updraft interpretation still relies on forecasters' experience.

[0005] Based on the above analysis, it is still necessary to address the shortcomings of existing quantitative inversion techniques for vertical velocity profiles, as well as the problem of large errors in inverting vertical velocity profiles using a single variable such as reflectivity factor. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a method and system for quantitative estimation of vertical velocity based on dual-polarization radar.

[0007] The specific technical solution of the present invention is as follows:

[0008] A method for quantitative estimation of vertical velocity based on dual-polarization radar includes the following steps:

[0009] Historical dual-polarization radar data of the study area was acquired, and the historical data was preprocessed to obtain dual-polarization radar observation data with three-dimensional grid coordinates.

[0010] The updraft regions of convective weather are selected from the dual-polarization radar observation data, and the dual-polarization radar characteristic data of the updraft regions are obtained.

[0011] Historical reanalysis data of long time series of the study area are obtained. Multi-source observation data are assimilated by radar variational analysis system combined with short-term forecasts to generate dynamic analysis field and obtain three-dimensional vertical velocity profile. Spatiotemporal matching is performed between the updraft region selected by dual-polarization radar observation data and the three-dimensional vertical velocity profile to obtain matched three-dimensional vertical velocity profile data.

[0012] A dataset is constructed using the dual-polarization radar eigenvalue data and vertical velocity profile data. The dataset is then divided into a training set, a validation set, and a test set. Using the vertical velocity profile data as label values, a multivariate fusion vertical velocity quantitative prediction model is constructed. The model is then trained and validated to obtain the trained multivariate fusion vertical velocity quantitative prediction model.

[0013] Based on the trained multivariate fusion vertical velocity quantitative prediction model, the vertical velocity profile is quantitatively estimated using real-time observation data from dual-polarization radar as input.

[0014] Furthermore, the preprocessing of historical data includes interpolating historical data from dual-polarization radar to form dual-polarization radar observation data with three-dimensional grid coordinates; the multi-source observation data includes historical data of severe convective weather, radiosonde data, radar data, and ground station data.

[0015] Furthermore, the step of filtering updraft regions of convective weather from the dual-polarization radar observation data specifically involves...

[0016] First, quality control is performed on the dual-polarization radar observation data, and then the differential reflectivity Z is combined. DR Column, ratio differential phase K DRA weighted scoring model is constructed using a column recognition algorithm and a dual-polarization radar hydrophobic classification algorithm. The criterion for determining whether an area is an updraft region is whether the comprehensive score of the updraft is ≥2. The weighted scoring model is as follows:

[0017] Uscore = w1s1 + w2s2 + w3s3 + w4s 43

[0018] In the formula, s1, s2, s3, and s4 represent Z respectively. DR Column height score, Z DR Column strength score, K DR Column strength score, particle type score, w1, w2, w3, w4 represent Z respectively. DR Weighting coefficient of column height, Z DR Weighting coefficient for column strength, K DR The weighting coefficients for column strength and particle type are determined by their importance to the updraft.

[0019] Furthermore, the scoring rules for each parameter in the weighted scoring model are as follows:

[0020] S1 scores 1 point for every 0.5km above the 0°C level in column top height.

[0021] The score of s2 is based on Z DR Z within a 3km grid around the pillar DR For every 20% of the area with a depth >1dB, 1 point is awarded.

[0022] The score for S3 is based on the distance from the ground to K. DR Top of column, K DR One point is awarded for every 10° / km increase in vertical integral.

[0023] The score for s4 is based on the identification of supercooled water above the melt layer, indicating that raindrops are lifted, and it scores 1 point.

[0024] Furthermore, the method for determining the weight coefficients of each parameter in the weighted scoring model is as follows:

[0025] Based on the importance of each parameter to the updraft, Z DR Column height, Z DR Column strength, K DR The column strength and particle type are configured with weighting coefficients of 0.5, 0.25, 0.15, and 0.1, respectively.

[0026] Furthermore, the quality control of the dual-polarization radar observation data includes removing data with Z < 10 dB and correlation coefficient ρ. hvData with a value <0.85 were removed, and data with an ambient temperature greater than -20℃ were smoothed and filtered for dual-polarization radar observations.

[0027] Furthermore, the acquisition of dual-polarization radar characteristic data of the updraft region specifically involves...

[0028] For the grid points in the updraft region at time T, select the radar reflectivity Z and Z0 within a 3×3 grid point around the time (T-12, T-6, T). DR Column height, K DP Column height, profile ρ hv and Z DR Column depth, K DP Column depth and volume characteristic data are used as input.

[0029] An importance sampling method is used to resample the updraft samples in the dataset, thereby mitigating the sample imbalance problem.

[0030] The multivariate fusion vertical velocity quantitative prediction model uses the Transformer model as the prediction model and adopts a spatiotemporal attention mechanism to improve the model structure for multivariate feature information extraction.

[0031] This invention provides a vertical velocity quantitative estimation system based on dual-polarization radar, comprising the following parts:

[0032] The data processing module acquires historical data of dual-polarization radar in the study area and preprocesses the historical data to obtain dual-polarization radar observation data with three-dimensional grid coordinates.

[0033] The feature extraction module filters upflow regions of convective weather from the dual-polarization radar observation data and obtains dual-polarization radar feature value data of the upflow regions.

[0034] The vertical velocity profile module acquires historical reanalysis data of the study area over a long period of time. It then uses a radar variational analysis system combined with short-term forecasts to assimilate the reanalysis data, generating a dynamic analysis field and extracting a three-dimensional vertical velocity profile. The module then uses the updraft region selected from dual-polarization radar observation data as a benchmark to perform spatiotemporal matching with the three-dimensional vertical velocity profile, resulting in matched three-dimensional vertical velocity profile data.

[0035] The model building module constructs a multivariate fusion vertical velocity quantitative prediction model. It uses the dual-polarization radar feature value data as input and the vertical velocity profile data as output to construct a dataset. The dataset is then divided into a training set, a validation set, and a test set for model training, validation, and optimization to obtain the multivariate fusion vertical velocity quantitative prediction model.

[0036] The quantitative estimation module, based on the trained multivariate fusion vertical velocity quantitative prediction model, uses real-time observation data from dual-polarization radar as input to quantitatively estimate the vertical velocity profile.

[0037] Compared with the prior art, the present invention has the following advantages:

[0038] This invention provides a quantitative estimation method for vertical velocity based on dual-polarization radar. This method utilizes the multiple polarization parameters of dual-polarization radar, combined with deep learning technology and radar meteorological physics knowledge, to establish a nonlinear relationship between the multiple parameters of dual-polarization radar and vertical velocity, thereby realizing the quantitative estimation of the three-dimensional vertical velocity profile and providing quantitative analysis products for convective weather forecasting and early warning.

[0039] This invention combines differential reflectivity bar and differential phase bar identification algorithms with a dual-polarization radar hydrophobic classification algorithm to screen updraft regions in convective weather, taking into account Z... DR Column height, Z DR Column strength, K DR The weights of each parameter, including column intensity and particle type, on the updraft in the convective region are determined, and a weighted average model is used for scoring and screening to accurately and reliably select the updraft region.

[0040] By extracting feature values ​​related to the vertical characteristics of updraft regions from dual-polarization radar observation data, the correlation between the feature value data and the distribution characteristics of updraft regions is improved. These feature values ​​can fully reflect the vertical motion state of the airflow. The data includes forward time data (T-12, T-6, T) and surrounding data (within a 3×3 grid). Utilizing multi-timestep, multi-polarization parameter features as input is beneficial for extracting the temporal characteristics of dual-polarization radar observations and improving the accuracy of vertical velocity prediction.

[0041] This invention proposes a method that combines deep learning methods (Transformer and attention mechanism) with radar meteorological knowledge (Z). DR K DP An intelligent prediction model based on the microphysical-dynamic feedback mechanism between column depth and volume and vertical velocity enables more accurate estimation of vertical velocity profiles. A spatiotemporal attention mechanism is employed to improve the model structure for multivariate feature extraction. Using updraft regions selected from dual-polarization radar observation data as a benchmark and spatiotemporally matched data from a three-dimensional vertical velocity profile as input, a complex nonlinear relationship between dual-polarization radar multi-parameters and vertical velocity is established. Deep learning techniques are then used to capture the highly complex relationship between input data and target parameters, providing an effective method for predicting vertical wind speed profiles.

[0042] This invention overcomes the problem of large errors in existing methods that rely on a single variable, the radar reflectivity factor, to establish vertical velocity relationships. This invention innovatively combines the physical understanding of multiple observation variables in dual-polarization radar, and its vertical velocity quantitative estimation method based on dual-polarization radar improves the accuracy of vertical velocity inversion.

[0043] When screening updraft regions, this invention uses Z... DR Column and K DP A column is a dynamic characteristic of a strong convective cloud, K DP The column is related to the liquid water content, indirectly indicating the location of the updraft. Z DR The column is related to the updraft, Z DR The vertical distribution of the column may reflect the process of raindrops being carried into upper levels by updrafts and freezing. The vertical distribution of these parameters can also reflect the vertical movement of airflow, such as potentially carrying larger water droplets upwards, leading to a higher Z-axis. DR A characteristic distribution appears at certain altitude levels. Attached Figure Description

[0044] Figure 1 This is a control flowchart of the method in an embodiment of the present invention;

[0045] Figure 2 This is a flowchart of the deep learning model for quantitative estimation of vertical velocity based on dual-polarization radar in an embodiment of the present invention. Detailed Implementation

[0046] Example 1:

[0047] like Figure 1 As shown, the present invention provides a quantitative estimation method for vertical velocity based on dual-polarization radar, comprising the following steps:

[0048] Historical dual-polarization radar data of the study area was acquired, and the historical data was preprocessed to obtain dual-polarization radar observation data with three-dimensional grid coordinates.

[0049] The updraft regions of convective weather are selected from the dual-polarization radar observation data, and the dual-polarization radar characteristic data of the updraft regions are obtained.

[0050] Historical reanalysis data of long time series of the study area are obtained. Multi-source observation data are assimilated by radar variational analysis system combined with short-term forecasts to generate dynamic analysis field and obtain three-dimensional vertical velocity profile. Spatiotemporal matching is performed between the updraft region selected by dual-polarization radar observation data and the three-dimensional vertical velocity profile to obtain matched three-dimensional vertical velocity profile data.

[0051] A dataset is constructed using the dual-polarization radar eigenvalue data and vertical velocity profile data. The dataset is then divided into a training set, a validation set, and a test set. Using the vertical velocity profile data as label values, a multivariate fusion vertical velocity quantitative prediction model is constructed. The model is then trained and validated to obtain the trained multivariate fusion vertical velocity quantitative prediction model.

[0052] Based on the trained multivariate fusion vertical velocity quantitative prediction model, the vertical velocity profile is quantitatively estimated using real-time observation data from dual-polarization radar as input.

[0053] Preprocessing of historical data includes interpolating historical data from dual-polarization radar to form dual-polarization radar observation data with three-dimensional grid coordinates; the multi-source observation data includes historical data of severe convective weather, radiosonde data, radar data, and ground station data.

[0054] Example 2:

[0055] This example, based on Example 1, further illustrates the specific process of filtering updraft regions of convective weather from dual-polarization radar observation data:

[0056] First, quality control is performed on the dual-polarization radar observation data, and then the differential reflectivity Z is combined. DR Column, ratio differential phase K DR A weighted scoring model is constructed using a column recognition algorithm and a dual-polarization radar hydrophobic classification algorithm. The criterion for determining whether an area is an updraft region is whether the comprehensive score of the updraft is ≥2. The weighted scoring model is as follows:

[0057] Uscore = w1s1 + w2s2 + w3s3 + w4s 43

[0058] In the formula, s1, s2, s3, and s4 represent Z respectively. DR Column height score, Z DR Column strength score, K DR Column strength score, particle type score, w1, w2, w3, w4 represent Z respectively. DR Weighting coefficient of column height, Z DR Weighting coefficient for column strength, K DR The weighting coefficients for column intensity and particle type are determined by their importance to the updraft. A comprehensive score of 2 or higher indicates the presence of significant updrafts, and the region is designated as an updraft region.

[0059] Example 3:

[0060] Based on Example 2, this example further illustrates the scoring rules for each parameter in the weighted scoring model as follows:

[0061] S1 scores 1 point for every 0.5km above the 0°C level in column top height.

[0062] The score of s2 is based on Z DR Z within a 3km grid around the pillar DR For every 20% of the area with a depth >1dB, 1 point is awarded.

[0063] The score for S3 is based on the distance from the ground to K. DR Top of column, K DR One point is awarded for every 10° / km increase in vertical integral.

[0064] The score for s4 is based on the identification of supercooled water above the melt layer, indicating that raindrops are lifted, and it scores 1 point.

[0065] The method for determining the weight coefficients of each parameter in the weighted scoring model is as follows:

[0066] Based on the importance of each parameter to the updraft, Z DR Column height, Z DR Column strength, K DR The column strength and particle type are configured with weighting coefficients of 0.5, 0.25, 0.15, and 0.1, respectively.

[0067] Example 4:

[0068] This example provides a vertical velocity quantitative estimation system based on dual-polarization radar, which includes the following modules:

[0069] The data processing module acquires historical data of dual-polarization radar in the study area and preprocesses the historical data to obtain dual-polarization radar observation data with three-dimensional grid coordinates.

[0070] The feature extraction module filters upflow regions of convective weather from the dual-polarization radar observation data and obtains dual-polarization radar feature value data of the upflow regions.

[0071] The vertical velocity profile module acquires historical reanalysis data of the study area over a long period of time. It assimilates multi-source observation data by combining radar variational analysis system with short-term forecasts to generate a dynamic analysis field and obtain a three-dimensional vertical velocity profile. The updraft region selected by dual-polarization radar observation data is used as a benchmark to perform spatiotemporal matching with the three-dimensional vertical velocity profile to obtain matched three-dimensional vertical velocity profile data.

[0072] The model building module constructs a multivariate fusion vertical velocity quantitative prediction model. It uses the dual-polarization radar feature value data as input and the vertical velocity profile data as output to construct a dataset. The dataset is then divided into a training set, a validation set, and a test set for model training, validation, and optimization to obtain the multivariate fusion vertical velocity quantitative prediction model.

[0073] The quantitative estimation module, based on the trained multivariate fusion vertical velocity quantitative prediction model, uses real-time observation data from dual-polarization radar as input to quantitatively estimate the vertical velocity profile.

[0074] Example 5:

[0075] like Figure 2 As shown, the vertical velocity quantitative estimation method based on dual-polarization radar of the present invention specifically includes the following steps:

[0076] Step S1: Prepare historical dual-polarization radar baseline data and perform quality control on it using radar quality control software to ensure the quality of the input observations. Use radar processing software (such as NCAR-RDAX) to interpolate the radar observation data to form dual-polarization radar observation data with three-dimensional grid coordinates.

[0077] Step S2: Filter updraft regions of convective weather from dual-polarization radar observation data and obtain dual-polarization radar characteristic data of the updraft regions; characteristic data includes: radar reflectivity Z, Z0 DR Column height, K DP Column height, profile ρ hv and Z DR Column depth, K DP Column depth and volume characteristics.

[0078] The specific areas of rising airflow for convective weather selected from dual-polarization radar observation data are as follows:

[0079] Joint differential reflectivity column (Z) DR The column identification algorithm and the dual-polarization radar condensate classification algorithm identified the updraft regions of convective weather in the dual-polarization radar observation data. First, data with Z < 10 dB and ρhv < 0.85 were removed to ensure sufficient signal-to-noise ratio and data quality. Data corresponding to ambient temperatures greater than -20℃ were also removed to exclude large Z values ​​caused by ice crystals. DR The interference of Z values. To reduce Z... DR The effects of noise make Z DR The identification of columns is more reliable, especially for Z. DR Perform smoothing and filtering processing. DR The column recognition algorithm is based on Chen et al. (2024). It utilizes Z... DR While the column identification method determines the updraft region, the distribution of supercooled water in the cloud is determined by the dual-polarization radar phase state classification algorithm, further filtering out more reliable input data.

[0080] The overall score for updrafts is calculated using the weighted scoring model (Uscore model):

[0081]

[0082] In the formula, s1, s2, s3, and s4 represent Z respectively. DR Column height score, Z DR Column strength score, K DR Column strength score, particle type score, w1, w2, w3, w4 represent Z respectively. DR Weighting coefficient of column height, Z DR Weighting coefficient for column strength, K DR Weighting coefficients for column strength and particle type.

[0083] The weighting coefficients for each parameter are determined as follows:

[0084] Based on the importance of each parameter to the updraft, Z DR Column height, Z DR Column strength, K DR The column strength and particle type are configured with weighting coefficients of 0.5, 0.25, 0.15, and 0.1, respectively.

[0085] The weighting coefficients for this example are determined in the table below, and the scoring rules for each parameter are also shown in the table below. If the overall score of the Uscore weighted scoring model is greater than or equal to 2, then there is a significant updraft.

[0086]

[0087] Step S3: Select severe convective weather cases that have occurred in the study area within the past 5 years. Using ERA5 data or domestically produced high-resolution real-time atmospheric analysis data (ART) as initial boundary conditions, the network observation data of the severe convective weather cases from the high-resolution numerical simulation are assimilated and short-term forecasted using the Radar Variational Analysis System (VDRAS). The observation data includes conventional observations, dense automatic weather stations, and radar observations, forming a dynamic analysis field (three-dimensional vertical profile) with a resolution of 6 minutes and 1 kilometer. The 6-minute temporal resolution is used to match the 6-minute observation frequency of my country's operational weather radar.

[0088] Step S4: Select severe convective weather cases that have occurred in the study area in the past 5 years. Four typical severe convective weather events will be selected each year, totaling 20 cases over 5 years. Taking a duration of 6 hours as an example, the data will be output every 6 minutes, resulting in 4 × 5 × 6 × 10 = 1200 time points. Considering the high similarity of data samples at individual time points, data will be extracted in tiers based on the magnitude of the maximum upward velocity. 50-100 vertical velocity profile samples will be retained for each time point to ensure the model's generalization ability.

[0089] The updraft region selected in S2 is spatiotemporally matched with the vertical velocity profile in S3 using dual-polarization radar observation data, and the matched vertical velocity profile is used as the model output. The selection method for eigenvalues ​​in the dual-polarization radar observation data is as follows: for grid points in the updraft region at time T, Z1 and Z2 values ​​are selected from the surrounding 3×3 grid points at times (T-12, T-6, T). DR K DP ρhv profile and Z DR K DP The column depth and volume features are used as model inputs, and the dataset is constructed using the input and output data as samples.

[0090] Step S5: Considering the imbalance in the distribution of strong updraft samples, the importance sampling method is used to resample the updraft samples in the dataset to alleviate the sample imbalance problem; the dataset is divided into training set, validation set and test set based on the sampled strong convective event samples, and it is ensured that the three sub-datasets do not overlap in time and are independent of each other.

[0091] Since high-resolution numerical simulations provide a sufficient sample size, the Transformer model, which has a high upper limit of fitting ability, is used as the prediction model. A spatiotemporal attention mechanism is adopted to design an improved model structure for extracting multivariate feature information, effectively fusing dual polarization variable feature input information, and establishing a multivariate fusion vertical velocity quantitative prediction model.

[0092] Step S6: Input the actual observations from the dual-polarization radar. After preprocessing in S1 and identifying the rising zone in S2, if the rising airflow discrimination conditions are met, process it into the input format in S4, and use the deep learning model established in S5 to perform quantitative prediction of vertical velocity. The final output is a three-dimensional profile product.

[0093] The input parameters in S4 are the suggested combinations provided by this invention, including the selection of the time step and Z. DR K DP Features such as column depth and volume can be used as inputs and adjusted accordingly based on the actual observation time resolution, observation quality, and other factors.

Claims

1. A method for quantitative estimation of vertical velocity based on dual-polarization radar, characterized in that, Includes the following steps: Historical dual-polarization radar data of the study area was acquired, and the historical data was preprocessed to obtain dual-polarization radar observation data with three-dimensional grid coordinates. The updraft regions of convective weather are selected from the dual-polarization radar observation data, and the dual-polarization radar characteristic data of the updraft regions are obtained. Historical reanalysis data of long time series of the study area are obtained. Multi-source observation data are assimilated by radar variational analysis system combined with short-term forecasts to generate dynamic analysis field and obtain three-dimensional vertical velocity profile. Spatiotemporal matching is performed between the updraft region selected by dual-polarization radar observation data and the three-dimensional vertical velocity profile to obtain matched three-dimensional vertical velocity profile data. A dataset is constructed using the dual-polarization radar eigenvalue data and vertical velocity profile data. The dataset is then divided into a training set, a validation set, and a test set. Using the vertical velocity profile data as label values, a multivariate fusion vertical velocity quantitative prediction model is constructed. The model is then trained and validated to obtain the trained multivariate fusion vertical velocity quantitative prediction model. Based on the trained multivariate fusion vertical velocity quantitative prediction model, the vertical velocity profile is quantitatively estimated using real-time observation data from dual-polarization radar as input.

2. The method for quantitative estimation of vertical velocity based on dual-polarization radar according to claim 1, characterized in that, The preprocessing of historical data includes interpolating historical data from dual-polarization radar to form dual-polarization radar observation data with three-dimensional grid coordinates; the multi-source observation data includes historical radiosonde data, radar data, and ground station data of severe convective weather.

3. The method for quantitative estimation of vertical velocity based on dual-polarization radar according to claim 1, characterized in that, The process of filtering updraft regions of convective weather from the dual-polarization radar observation data specifically involves first performing quality control on the dual-polarization radar observation data, and then combining the differential reflectivity Z... DR Column, ratio differential phase K DR A weighted scoring model is constructed using a column recognition algorithm and a dual-polarization radar hydrophobic classification algorithm. The criterion for determining whether an area is an updraft region is whether the comprehensive score of the updraft is ≥2. The weighted scoring model is as follows: Uscore=w1s1+w2s2+w3s3+w4s 43 In the formula, s1, s2, s3, and s4 represent Z respectively. DR Column height score, Z DR Column strength score, K DR Column strength score, particle type score, w1, w2, w3, w4 represent Z respectively. DR Weighting coefficient of column height, Z DR Weighting coefficient for column strength, K DR Weighting coefficients for column strength and particle type.

4. The method for quantitative estimation of vertical velocity based on dual-polarization radar according to claim 3, characterized in that, The scoring rules for each parameter in the weighted scoring model are as follows: S1 scores 1 point for every 0.5km above the 0°C level in column top height. The score of s2 is based on Z DR Z within a 3km grid around the pillar DR For every 20% of the area with a depth >1dB, 1 point is awarded. The score for S3 is based on the distance from the ground to K. DR Top of column, K DR One point is awarded for every 10° / km increase in vertical integration. The score for s4 is based on the identification of supercooled water above the melt layer, indicating that raindrops are lifted, and it scores 1 point.

5. The method for quantitative estimation of vertical velocity based on dual-polarization radar according to claim 4, characterized in that, The method for determining the weight coefficients of each parameter in the weighted scoring model is as follows: Based on the importance of each parameter to the updraft, Z DR Column height, Z DR Column strength, K DR The column strength and particle type are configured with weighting coefficients of 0.5, 0.25, 0.15, and 0.1, respectively.

6. The method for quantitative estimation of vertical velocity based on dual-polarization radar according to claim 5, characterized in that, The quality control of dual-polarization radar observation data includes removing data with Z < 10 dB and the correlation coefficient ρ. hv Data with a value <0.85 were removed, and data with an ambient temperature greater than -20℃ were smoothed and filtered for dual-polarization radar observations.

7. The method for quantitative estimation of vertical velocity based on dual-polarization radar according to any one of claims 1-6, characterized in that, The acquisition of dual-polarization radar characteristic data of the updraft region specifically involves selecting, for the grid points in the updraft region at time T, the radar reflectivity Z0 and Z1 values ​​within a 3×3 grid point surrounding the (T-12, T-6, T) time zone. DR Column height, K DP Column height, profile ρ hv and Z DR Column depth, K DP Column depth and volume characteristic data are used as input.

8. The method for quantitative estimation of vertical velocity based on dual-polarization radar according to claim 1, characterized in that, An importance sampling method is used to resample the updraft samples in the dataset, thereby mitigating the sample imbalance problem.

9. The method for quantitative estimation of vertical velocity based on dual-polarization radar according to claim 1, characterized in that, The multivariate fusion vertical velocity quantitative prediction model uses the Transformer model as the prediction model and adopts a spatiotemporal attention mechanism to improve the model structure for multivariate feature information extraction.

10. A vertical velocity quantitative estimation system based on dual-polarization radar, characterized in that, Includes the following parts: The data processing module acquires historical data of dual-polarization radar in the study area and preprocesses the historical data to obtain dual-polarization radar observation data with three-dimensional grid coordinates. The feature extraction module filters upflow regions of convective weather from the dual-polarization radar observation data and obtains dual-polarization radar feature value data of the upflow regions. The vertical velocity profile module acquires historical reanalysis data of the study area over a long period of time. It then uses a radar variational analysis system combined with short-term forecasts to assimilate the reanalysis data, generating a dynamic analysis field and extracting a three-dimensional vertical velocity profile. The module then uses the updraft region selected from dual-polarization radar observation data as a benchmark to perform spatiotemporal matching with the three-dimensional vertical velocity profile, resulting in matched three-dimensional vertical velocity profile data. The model building module constructs a multivariate fusion vertical velocity quantitative prediction model. It uses the dual-polarization radar feature value data as input and the vertical velocity profile data as output to construct a dataset. The dataset is then divided into a training set, a validation set, and a test set for model training, validation, and optimization to obtain the multivariate fusion vertical velocity quantitative prediction model. The quantitative estimation module, based on the trained multivariate fusion vertical velocity quantitative prediction model, uses real-time observation data from dual-polarization radar as input to quantitatively estimate the vertical velocity profile.

Images