A phased array radar reflectivity factor simulation method based on sparse sounding data

Abstract

The present application relates to a kind of based on sparse sounding data phased array radar reflectivity factor simulation method, belong to radio field, based on the sparse sounding data of multiple point multiple height layer collected, sparse reconstruction simulation scheme is designed, the high-quality restoration of original sparse observation data and three-dimensional continuous field construction are realized.Sparse reconstruction simulation method based on sparse sounding data is provided, and the method comprises the following steps: constructing multivariate dictionary base matrix;By sparse representation theory and bayesian optimization model, the sparse coefficient reflecting the essential characteristics of precipitation is obtained, and the high spatial resolution reconstruction of temperature and humidity field and precipitation microphysical parameter field is realized.Afterwards, the phase state determination and probability field calculation model driven by meteorological parameters is constructed, the three-dimensional simulation of liquid, solid and mixed state precipitation echo is realized.Finally, the unit direction vector and azimuth guide code matrix of phased array radar beam ray are calculated, the radar volume scanning data implicit sampling simulator is introduced, and the high-precision reflectivity factor simulation under the stereoscopic scanning mode of phased array radar is realized.

Description

Technical Field

[0001] This invention relates to the field of radio, and more particularly to a method for simulating the reflectivity factor of a phased array radar based on sparse radiosonde data. Background Technology

[0002] In modern meteorological monitoring and forecasting systems, phased array weather radar, with its advantages of rapid scanning, flexible beam scheduling, and strong three-dimensional volume detection capabilities, is gradually becoming an important technical equipment for severe weather identification, quantitative precipitation estimation, and radar networking. Radar reflectivity factor, as the most crucial observational product, directly affects the identification and quantification of precipitation intensity and meteorological disasters. However, in practical applications, how to accurately simulate or invert reflectivity factor data of phased array radar at different radial and range libraries with multiple antenna elevation angles based on limited environmental field information is an important research topic in the field of radar remote sensing and numerical simulation, possessing significant scientific value and engineering significance.

[0003] Currently, methods for simulating reflectivity factors of phased array weather radars are mainly divided into two categories: one is the parameterized simulation method based on empirical regression relationships, such as constructing empirical models like ZR and Z-Nw-Dm based on the statistical relationship between temperature and humidity fields and raindrop spectral parameters to estimate reflectivity; the other is the microphysical simulation method based on physical processes, such as using Mie scattering theory or T-matrix theory to calculate the intensity of radar scattering echoes given the meteorological profile and raindrop spectral distribution. These methods have promoted radar data simulation research to some extent, but they also have significant shortcomings: First, most methods rely on single-point ground observations or weather balloon profile data, resulting in low spatial coverage and difficulty in reflecting the continuous and three-dimensional changes in the near-surface upper atmosphere meteorological field; Second, radar simulation processes often lack the ability to characterize spatial non-uniformity, leading to coarse echo simulation results that are difficult to support the simulation requirements of high-precision radar data; Third, existing methods tend to idealize the characterization of different particle phases and lack probabilistic modeling mechanisms, making it difficult to handle the simulation of precipitation systems with complex mixed phases; Fourth, traditional methods often rely on manually set echo functions, making it difficult to embed prior perturbation information under sparse detection data, thus limiting simulation resolution and accuracy. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for simulating the reflectivity factor of phased array radar based on sparse radiosonde data, thus solving the deficiencies of the prior art.

[0005] The objective of this invention is achieved through the following technical solution: a method for simulating the reflectivity factor of a phased array radar based on sparse radiosonde data, the simulation method comprising:

[0006] S1. Collect sparse sounding profile datasets and auxiliary data, as well as spatial location and operational status data of phased array weather radars; generate highly uniform vectors and generate spatial basis matrices based on the collected data;

[0007] S2. Using the data collected in S1 as input, the normalized atmospheric temperature profile is obtained by inverse distance weighted interpolation and normalization. The attribute perturbation spectrum basis matrix is ​​obtained by combining the 1D discrete Fourier transform model. A multi-scale perturbation basis matrix is ​​generated based on the normalized atmospheric temperature profile and wavelet transform model. The complete dictionary basis matrix of atmospheric temperature is generated based on the spatial basis matrix, the attribute perturbation spectrum basis matrix, and the multi-scale perturbation basis matrix. The atmospheric temperature profile is updated with profile data of atmospheric humidity, mass-weighted average diameter of precipitation particles, and normalized intercept of precipitation particle spectrum. The complete dictionary basis matrix of atmospheric humidity, mass-weighted average diameter of precipitation particles, and normalized intercept of precipitation particle spectrum is obtained.

[0008] S3. Based on the data obtained in S2, the sparse coefficient vector of each observation point is obtained by using the sparse representation model and combining the calculation model of the likelihood function and sparse prior. The sparse representation model is then used for optimization to achieve the optimization of atmospheric temperature, atmospheric humidity, mass-weighted average diameter of precipitation particles and normalized intercept data of precipitation particle spectrum.

[0009] S4. The optimized data from S3 is processed and combined with the structural similarity weight coefficient calculation model and the spatial weight coefficient calculation model to obtain the calculation results. The calculation results are multiplied and normalized. Then, according to the weighted projection formula, a three-dimensional dataset with high spatial resolution of atmospheric temperature, atmospheric humidity, mass-weighted average diameter of precipitation particles and normalized intercept of precipitation particle spectrum is obtained.

[0010] S5. Filter the data obtained in S4, process the filtered data and combine it with the data in S4 and the formula for calculating the phase probability of precipitation particles and the formula for calculating the multiphase radar reflectivity factor to calculate the three-dimensional data of the reflectivity factor of the high spatial resolution phased array weather radar.

[0011] S6. Generate a spatial position matrix based on the data collected in S1. Combine the calculation of each point in the spatial position matrix with the azimuth guidance coding vector of each point column by column to form an azimuth guidance coding matrix. Combine the data collected in S1 and the three-dimensional data of high spatial resolution phased array weather radar reflectivity factor obtained in S5 to obtain radar reflectivity factor data in the stereo scanning mode of phased array weather radar.

[0012] The collection of sparse sounding profile datasets and auxiliary data, as well as spatial location and operational status data of phased array weather radars, specifically includes the following:

[0013] Collect sparse sounding profile datasets and auxiliary data obtained by a meteorological UAV array. The sparse sounding profile datasets obtained by the meteorological UAV array include: a dataset T_TR of atmospheric temperature profiles at different altitudes at the i-th meteorological UAV ground fixed point, with a total number of M ground fixed points corresponding to the meteorological UAV array. i Atmospheric humidity profile dataset Q_TR i Mass-weighted average diameter profile dataset of precipitation particles Dm_TR i Normalized intercept profile dataset of precipitation particle spectra Nw_TR i , i=1,2,3,…M; the auxiliary data for the meteorological UAV array includes: the spatial location information of observation points at different altitudes on the i-th ground fixed point of the meteorological UAV, P i = (Lon_TR) i Lat_TR i Hig_TR i ), where Lon_TR i For longitude points, Lat_TR i Latitude point, Hig_TR i A vector representing different heights, i = 1, 2, 3, ..., M;

[0014] The collection of spatial location and operational status data for phased array weather radar mainly includes: spatial location information of the phased array weather radar antenna feed P_radar=[lon_radar,lat_radar,hig_radar], where lon_radar, lat_radar, and hig_radar are longitude, latitude, and altitude, respectively; radar wavelength, radar radial range resolution delt_R, beamwidth delt_sita, and radar scanning sequence el_array=[el0,el1,el2,…,el] at the elevation layer. max ], el max Rmax_radar represents the maximum scanning elevation angle and the maximum unambiguous distance.

[0015] The generation of highly uniform vectors and the generation of spatial basis matrices based on the collected data specifically include the following:

[0016] A1. Set the resolution and multiplier of the altitude layer to delt_H and Num_H respectively, generate the maximum altitude layer as Hig_max = Num_H × delt_H, and then generate a uniform altitude vector Z at each UAV ground fixed point. i =h×delt_H, where h=1,2,3,…,Num_H, i=1,2,3,…,M;

[0017] A2. Based on the collected Pi = (Lon_TR) i Lat_TR i Hig_TR i ), P i Hig_TR in i Change to Z i Generate highly uniform profile data spatial location information P for each meteorological drone's ground fixed point. i_uni = (Lon_TR) i Lat_TR i Z i );

[0018] A3. Set the observation point P j For P i_uni For any point in space, relative to the observation point P j The longitude, latitude, and altitude are normalized along each of the longitude, latitude, and altitude dimensions to obtain the normalized longitude. ,latitude and height and normalized =( ), j=1,2,3,..,M×Num_H;

[0019] A4. Generate the observation point P based on the second-order polynomial orthogonal basis functions. j Spatial structure vector , Denotes the first orthogonal basis function of the second-order polynomial;

[0020] A5. Following the methods in steps A3 and A4, process M×Num_H P... j The process involves traversing the vectors and combining the resulting M×Num_H spatial structure vectors to generate the spatial basis matrix Φs=[ ], its dimensions are M×Num_H×10.

[0021] The process of using the data collected by S1 as input, employing inverse distance weighted interpolation and normalization to obtain the normalized atmospheric temperature profile, and combining it with a 1D discrete Fourier transform model to obtain the attribute perturbation spectrum basis matrix specifically includes the following:

[0022] B1. Collect the atmospheric temperature profile dataset T_TR from the i-th ground-fixed point of the meteorological UAV at different altitudes. i The spatial location information P corresponding to this profile line i As input, the inverse distance weighted interpolation method is used to input T_TR. i Interpolation to the generated spatial information P i_uniFor the spatial location of the target profile, interpolated atmospheric temperature profile data are obtained at each fixed ground point of the meteorological UAV, denoted as . ;

[0023] B2. Calculate the profile data. The mean and standard deviation are denoted as mean( ) and std( Then, according to the Z-Score normalization formula, the profile data is normalized. After normalization, the normalized atmospheric temperature profile is obtained. ;

[0024] B3. Atmospheric temperature profile The data is input into a 1D discrete Fourier transform model, the model is run, and the modulus of the model's output is taken to obtain the atmospheric temperature profile. Spectral mode sequence DFT i =|DFT( )|;

[0025] B4. DFT i The modulus of the mid-to-high frequency range is set to zero, while the modulus of the low frequency range is retained, and the DFT is selected. i The first Ka low-frequency components, excluding the DC component, are denoted as DFT. i,2 DFT i,3 ..., DFT i,Ka+1 These Ka low-frequency components are then combined into a one-dimensional vector, forming the perturbation mode vector c. i Ka is the number of low-frequency components selected;

[0026] B5. Apply the methods from steps B1 to B4 to atmospheric temperature profile data at different altitudes from M fixed ground points of meteorological UAVs, obtaining M disturbance mode vectors, denoted as c1, c2, c3, ..., c M And these M perturbation mode vectors are concatenated into an attribute perturbation spectrum basis matrix Φc=[c1;c2;c3;…;c M ], whose dimensions are M×Num_H×Ka.

[0027] The process involves generating a multi-scale perturbation basis matrix based on the normalized atmospheric temperature profile and wavelet transform model, and then generating a complete dictionary basis matrix for atmospheric temperature based on the spatial basis matrix, attribute perturbation spectrum basis matrix, and multi-scale perturbation basis matrix. The atmospheric temperature profile is then updated with atmospheric humidity, the mass-weighted average diameter of precipitation particles, and the normalized intercept profile of the precipitation particle spectrum. The resulting complete dictionary basis matrix for atmospheric humidity, the mass-weighted average diameter of precipitation particles, and the normalized intercept of the precipitation particle spectrum specifically includes the following:

[0028] C1. Generate the normalized atmospheric temperature profile Input the data into the wavelet transform model, select the wavelet basis type, set the decomposition level to La, run the model, and the model output is the atmospheric temperature profile at point i. and ,in These are the approximation coefficients for the La-th layer. Let La be the detail coefficient of the r-th layer, where r = 1, 2, 3, ..., La;

[0029] C2. Based on the calculation formula for extracting perturbation energy using detail coefficients, extract the perturbation energy b of the La decomposition layer at the i-th point. i,r r=1,2,3,…,La; then, the perturbation energy b of all La decomposition layers at point i is... i,1 b i,2 b i,3 , ..., b i,La Combined into a multi-scale perturbation feature vector m i ;

[0030] C3. Apply the methods from steps C1 and C2 to the normalized atmospheric temperature profiles at different altitudes on M ground-based fixed points of meteorological UAVs, obtaining M multi-scale eigenvectors. Then, concatenate these M multi-scale eigenvectors into a multi-scale perturbation basis matrix Φm=[m1;m2;m3;…;m…]. M ], m M Let be the Mth multi-scale feature vector, with dimensions M×Num_H×La;

[0031] C4. Combine the generated Φs, Φc, and Φm from this step to construct a complete dictionary base matrix Φ for atmospheric temperature. T =[Φs||Φc||Φm], update the atmospheric temperature profile dataset to the atmospheric humidity profile dataset, the mass-weighted average diameter profile dataset of precipitation particles, and the normalized intercept profile dataset of precipitation particle spectra, respectively. Then, combine Φs generated by A5 in S1 to generate the complete dictionary basis matrix Φ of the normalized intercept of atmospheric humidity. Q The complete dictionary basis matrix Φ of the normalized intercept of the mass-weighted average diameter of precipitation particles. Dm The complete dictionary basis matrix Φ of the normalized intercept of precipitation particle spectrum Nw .

[0032] S3 specifically includes the following:

[0033] S301. Applying step B1 to the atmospheric temperature profile data from M ground-fixed points on a meteorological UAV, we can obtain M interpolated atmospheric temperature profile data, denoted as... ;

[0034] S302, Based on the generated complete dictionary base matrix of atmospheric temperature Φ T Using a sparse representation model, The observation data d at each observation point g g Use Φ T and sparse coefficient vector a g Representing d in the form of a linear combination g The input is fed into a computational model that combines the likelihood function and sparse priors. Running the model yields a sparse coefficient vector a for each observation point. g , will a g The input is fed into a Bayesian inference and maximum a posteriori estimation model with sparse constraints, and the model is run to obtain the optimized sparse coefficient vector. ;

[0035] S303, based on the generated dictionary base Φ and the optimized An optimized sparse representation model is used to optimize the atmospheric temperature dataset. The optimized atmospheric temperature dataset is denoted as... ;

[0036] S304, Φ in this step T The atmospheric temperature profile dataset was updated to Φ. Q And atmospheric humidity profile dataset, Φ Dm And the mass-weighted average diameter profile dataset of precipitation particles, Φ Nw The datasets of normalized intercept profiles of precipitation particle spectra are used to optimize atmospheric humidity, mass-weighted average diameter of precipitation particles, and normalized intercept data of precipitation particle spectra, respectively. These are denoted as... , , .

[0037] S4 specifically includes the following:

[0038] S401. Based on the generated optimized atmospheric temperature dataset The dataset is grouped according to height levels to form a two-dimensional sparse plane dataset T_Z with multiple height levels. h h=1,2,3,…,Num_H;

[0039] S402. Based on the set height layer resolution delt_H, further set the horizontal space height resolution to delt_R_H, and generate a target 3D mesh Tar_grid with a spatial resolution of delt_R_H×delt_R_H×delt_H. For T_Z h Let T_Z be the k-th height layer dataset of Tar_grid. k And Tar_grid_k, calculate T_Zk The maximum, median, and minimum values ​​are denoted as max(T_Z). k ), med(T_Z k ), min(T_Z k );

[0040] S403, T_Z k With max(T_Z) k ), med(T_Z k ), min(T_Z k The inputs are combined with the inputs into the structural similarity weight coefficient calculation model. Running the model outputs T_Z. k The set of structural weight coefficients W1_T_Z for each sparse point k Calculate the relationship between a point pt and T_Z in Tar_grid_k. k The spatial distances between all sparse observation points are calculated, and the results are input into the spatial weight coefficient calculation model. The model is then run, and the output is T_Z. k The set of spatial weight coefficients for each sparse point in W2_T_Z k W1_T_Z k With W2_T_Z k Multiply and normalize to form the total weight coefficient set W_T_Z k According to the weighted projection formula, the atmospheric temperature value T_grid_pt at point pt in Tar_grid_k is calculated;

[0041] S404. Apply the method of step S403 to all points in Tar_grid_k, and apply this step to all height layers in Tar_grid to obtain a high spatial resolution atmospheric temperature 3D dataset T_3D. Update the atmospheric temperature dataset of this step sequentially to atmospheric humidity, mass-weighted average diameter of precipitation particles, and normalized intercept dataset of precipitation particle spectrum, to obtain high spatial resolution atmospheric humidity, mass-weighted average diameter of precipitation particles, and normalized intercept dataset of precipitation particle spectrum, denoted as Q_3D, Dm_3D, and Nw_3D, respectively.

[0042] S5 specifically includes the following:

[0043] S501. Based on the generated T_3D, Q_3D, Dm_3D and Nw_3D datasets, set the variation ranges of Dm_3D and Nw_3D to [Dm_min, Dm_max] and [Nw_min, Nw_max] respectively, and remove the data in the Dm_3D and Nw_3D datasets that exceed the variation range;

[0044] S502. Input the Dm_3D and Nw_3D datasets after removing outliers into a 3-point × 3-point × 3-point median filter, and denote the filtered outputs as Dm_3D_f and Nw_3D_f, respectively. Input the T_3D and Q_3D datasets into a spatial Gaussian smoothing filter, and denote the filtered atmospheric temperature and atmospheric humidity datasets as T_3D_f and Q_3D_f, respectively.

[0045] S503. Set the atmospheric humidity threshold for precipitation to Q_th, mark all grid points in Q_3D_f with atmospheric humidity values ​​greater than Q_th as 1, mark the remaining grid points as 0, generate a precipitation identification mask matrix Rain_Mask, and set the phase state of precipitation particles to three types: liquid, solid-liquid mixture and solid.

[0046] S504. For any point P_s in the established target 3D mesh Tar_grid, combine T_3D_f and Q_3D_f to obtain the atmospheric temperature and humidity data of that point. and Then, using the formula for calculating the phase probability of precipitation particles, the probabilities of the spatial point P_s being in the liquid, solid-liquid mixture, and solid states are calculated respectively, and denoted as P_L. s P_M s P_I s ;

[0047] S505. Apply the steps of S504 to all spatial points in the target 3D mesh Tar_grid to generate probability matrices for the three phase states: liquid, solid-liquid mixture, and solid, denoted as P_L, P_M, and P_I, respectively. Set the correlation indices of the liquid, solid-liquid mixture, and solid particles to alph_L, alph_M, and alph_I, respectively. Then, assign the following values ​​to the given values: [alph_L, T_3D, Q_3D, Dm_3D, Nw_3D, Rain_Mask], [alph_M, T_3D, Q_3D, Dm_3D, Nw_3D, Rain_Mask], and [alph_I, T_3D, ... Q_3D, Dm_3D, Nw_3D, Rain_Mask] and the collected radar wavelength are input together into a phase-sensitive radar echo simulation system. The system is run and outputs a three-dimensional dataset of radar reflectivity factors Z_L, Z_M, and Z_I for all points in the target's three-dimensional grid Tar_grid when the precipitation particles are in liquid, solid-liquid mixture, and solid states. Based on the calculated P_L, P_M, P_I and Z_L, Z_M, and Z_I, and according to the multi-phase radar reflectivity factor joint calculation formula, the high spatial resolution phased array weather radar reflectivity factor three-dimensional data Z_grid is calculated.

[0048] S6 specifically includes the following:

[0049] S601. Based on the collected radial range resolution delt_R, beamwidth delt_sita, maximum unambiguous range Rmax_radar, and radar elevation scan sequence el_array of the phased array weather radar, generate the range library sequence bins_array=[delt_R,2×delt_R,3×delt_R,…,Rmax_radar] and the azimuth sequence az_array=[delt_sita,2×delt_sita,3×delt_sita,…,360︒]. Combine el_array, az_array, and bins_array to generate the spatial position matrix P_echo=[el_array,az_array,bins_array] of the echo in the stereo scanning mode of the phased array weather radar.

[0050] S602. Let a point in P_echo be P0, and let the spatial position of the point be [el_0, az_0, bin_0], where el_0, az_0, and bin_0 are the elevation angle, azimuth angle, and range of the point, respectively. Then, according to the unit direction vector calculation formula, calculate the unit direction vector dr of the radar beam ray, and according to the polar coordinate angle encoding formula, calculate the polar coordinate angle code E0_dir. According to the three-dimensional relative coordinate encoding calculation formula, calculate the three-dimensional relative coordinate code E0_rel. Combine the calculated E0_rel, E0_dir, and dr to generate the azimuth guidance encoding vector of point P0, E0 = [E0_rel||E0_dir||dr].

[0051] S603. Apply the steps of S602 to each point in P_echo, and combine the generated azimuth guidance coding vectors of each point by column to form the azimuth guidance coding matrix E_ALL. Input the collected spatial location information of the radar antenna feed P_radar, the generated Z_grid, and the azimuth guidance coding matrix E_ALL into the volumetric scan data implicit sampling simulator, run the simulator, and output the radar reflectivity factor data in the phased array weather radar stereo scanning mode.

[0052] The present invention has the following advantages:

[0053] 1. Regarding data sources, this invention utilizes a meteorological UAV array to acquire multivariate vertical profile observation data over multiple fixed ground points, including atmospheric temperature, atmospheric humidity, mass-weighted average diameter (Dm) of precipitation particles, and normalized intercept (Nw). This enables spatial expansion and refined vertical characterization of radiosonde data, significantly improving the physical basis and spatial coverage of reflectivity simulation. Secondly, in terms of modeling methods, this invention integrates three types of dictionary bases: spatial location basis functions, attribute perturbation spectrum basis, and multi-scale wavelet energy basis. This constructs a high-dimensional expression system with strong physical interpretability. Through sparse representation and Bayesian optimization, it achieves high-precision reconstruction of the original sparse data, effectively solving the problems of data sparsity and discontinuity in traditional methods.

[0054] 2. This method introduces a particle phase state probabilistic modeling mechanism, combining different microphysical response models for liquid, mixed, and solid states. A phase-sensitive radar echo simulation module is introduced, and a more realistic and continuous three-dimensional reflectivity field is generated through a multi-phase weighted fusion formula, overcoming the shortcomings of traditional methods that involve coarse or static phase state handling. Finally, by constructing an azimuth guidance coding matrix, this method fully considers the spatial sampling characteristics of phased array radar under different elevation, azimuth, and range resolutions, enabling the simulation output to accurately match the actual radar observation mode and improving the spatial consistency and system adaptability of reflectivity simulation. Attached Figure Description

[0055] Figure 1 This is a schematic diagram of the process of the present invention;

[0056] Figure 2 This is a flowchart illustrating the implementation of the spatial basis matrix generation method of the present invention.

[0057] Figure 3 This is a flowchart illustrating the implementation of the attribute perturbation spectrum basis matrix generation method of the present invention.

[0058] Figure 4 This is a flowchart illustrating the implementation of generating the complete dictionary base matrix of this invention.

[0059] Figure 5 This is a flowchart illustrating the implementation of the meteorological unmanned aerial vehicle (UAV) array observation data optimization of the present invention.

[0060] Figure 6 This is a flowchart illustrating the implementation process of generating a high spatial resolution 3D dataset based on meteorological UAV observation data according to the present invention.

[0061] Figure 7 A flowchart illustrating the implementation of the three-dimensional data generation of reflectivity factor for the high spatial resolution phased array weather radar of the present invention.

[0062] Figure 8This is a flowchart illustrating the implementation of radar reflectivity factor data generation in the phased array weather radar stereo scanning mode of the present invention. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Therefore, the detailed description of the embodiments of this application provided below with reference to the accompanying drawings is not intended to limit the scope of protection of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. The present invention will be further described below with reference to the accompanying drawings.

[0064] This invention specifically relates to a method for simulating the reflectivity factor of phased array weather radar based on sparse radiosonde data from meteorological UAV arrays. It utilizes multiple meteorological UAVs to construct a sparse upper-air profile data grid, acquiring atmospheric temperature profiles, atmospheric humidity profiles, mass-weighted average diameters of precipitation particles, and normalized intercepts, among other environmental field and microphysical variables, over different fixed ground points. A spatial-attribute-multi-scale joint perturbation dictionary basis matrix is ​​then constructed based on the spatial location. This method not only achieves high-quality reconstruction of the original sparse data through sparse representation and Bayesian inference, but also realizes high-precision three-dimensional simulation of radar reflectivity for complex precipitation fields by constructing a particle phase probability field and integrating sensitive reflectivity models for liquid, solid, and mixed states.

[0065] like Figure 1 As shown, it specifically includes the following:

[0066] Step 1: Collect sparse sounding profile datasets and auxiliary data obtained by meteorological drone arrays; collect spatial location and operational status data of phased array weather radars;

[0067] The sparse sounding profile dataset acquired by the meteorological UAV array mainly includes: the total number of ground fixed points corresponding to the meteorological UAV array is M=100; and the atmospheric temperature profile dataset T_TR at different altitudes at the i-th meteorological UAV ground fixed point. i Atmospheric humidity profile dataset Q_TR i Mass-weighted average diameter profile dataset of precipitation particles Dm_TR i Normalized intercept profile dataset of precipitation particle spectra Nw_TR i , i=1,2,3,…M;

[0068] The auxiliary data for the meteorological UAV array mainly includes: the spatial location information of observation points at different altitudes on the i-th ground fixed point of the meteorological UAV, denoted as P. i = (Lon_TR) i Lat_TR i Hig_TR i ), where Lon_TR i For longitude points, Lat_TR i Latitude point, Hig_TR i A vector representing different heights, i = 1, 2, 3, ..., M;

[0069] The spatial location and operational status data of the phased array weather radar mainly include: the spatial location information of the phased array weather radar antenna feed P_radar=[lon_radar,lat_radar,hig_radar], where lon_radar, lat_radar and hig_radar are longitude, latitude and altitude, respectively; radar wavelength=3cm, radar radial range resolution delt_R=30m, beamwidth delt_sita=1 degree; radar scanning sequence on the elevation layer el_array=[0.5:0.5:18], maximum antenna scanning elevation angle el_max=18 degrees, maximum unambiguous range Rmax_radar=60km;

[0070] Step 2: As Figure 2 As shown, firstly, the resolution and multiplier of the height layer are set to delt_H = 30 meters and Num_H = 200, respectively, generating the maximum height layer as Hig_max = Num_H × delt_H, and then generating a uniform height vector Z at each UAV ground fixed point. i =h×delt_H (h=1,2,3,…,Num_H, i=1,2,3,…,M); Secondly, based on the P collected in step 1 i = (Lon_TR) i Lat_TR i Hig_TR i ), P i Hig_TR in i Change to Z i Generate highly uniform profile data spatial location information P for each meteorological drone's ground fixed point. i_uni = (Lon_TR) i Lat_TR i Z i Then, assuming the observation point P... j (j=1,2,3,..,M×Num_H) is P i_uniFor any point in space, relative to the observation point P j The longitude, latitude, and altitude are normalized along each of the three dimensions to obtain the normalized longitude. ,latitude and height and normalized =( Then, based on the second-order polynomial orthogonal basis functions, the observation point P is generated. j Spatial structure vector Finally, following the above method, process M×Num_H P... j The process involves traversing the vectors and combining the resulting M×Num_H spatial structure vectors to generate the spatial basis matrix Φs=[ ], its dimensions are M×Num_H×10;

[0071] Furthermore, normalization is performed across all dimensions as follows: =(f j -f min ) / (f max -f min ), where f j and These are the dimension variables to be normalized and the normalized ones, respectively, f max and f min f j The maximum and minimum values;

[0072] The orthogonal basis functions of the second-order polynomial are: ;

[0073] Step 3: As Figure 3 As shown, the atmospheric temperature profile dataset T_TR at different altitudes at the i-th ground-fixed point of the meteorological UAV collected in step 1 is... i The spatial location information P corresponding to this profile line i As input, the inverse distance weighted interpolation method is used to input T_TR. i Interpolate to the spatial information P generated in step 2 i_uni For the spatial location of the target profile, interpolated atmospheric temperature profile data are obtained at each fixed ground point of the meteorological UAV, denoted as . Then, the profile data was calculated. The mean and standard deviation are denoted as mean( ) and std( Then, according to the Z-Score normalization formula, the profile data is normalized. After normalization, the normalized atmospheric temperature profile is obtained. Then, the atmospheric temperature profile was... The data is input into a 1D discrete Fourier transform model, the model is run, and the modulus of the model's output is taken to obtain the atmospheric temperature profile. Spectral mode sequence DFT i =|DFT( After that, the DFT will be... i The modulus of the mid-to-high frequency range is set to zero, while the modulus of the low frequency range is retained, and the DFT is selected. i The first Ka low-frequency components, excluding the DC component, are denoted as DFT. i,2 DFT i,3 ..., DFT i,Ka+1 These Ka low-frequency components are then combined into a one-dimensional vector, forming the perturbation mode vector c. i Finally, the above method is applied to atmospheric temperature profile data at different altitudes from M fixed ground points of meteorological UAVs to obtain M disturbance mode vectors. These M disturbance mode vectors are then concatenated into an attribute disturbance spectrum basis matrix Φc=[c1;c2;c3;…;c…]. M The dimension is M×Num_H×Ka, where Ka is the number of low-frequency components selected;

[0074] The Z-Score normalization formula is as follows: ;

[0075] Perturbation mode vector c i =[DFT i,2 DFT i,3 ,…,DFT i,Ka+1 ], i=1,2,3,…,M;

[0076] Step 4: As Figure 4 As shown, firstly, the normalized atmospheric temperature profile generated in step 3 is... The data is input into a wavelet transform model. The wavelet basis type is selected as multi-scale sliding window residual basis, and the number of decomposition levels is set to La=3. The model is run, and the output is the atmospheric temperature profile at point i. and ,in These are the approximation coefficients for the La-th layer. Let be the detail coefficients of the r-th layer, where r = 1, 2, 3, ..., La; then, based on the calculation formula for extracting perturbation energy from the detail coefficients, extract the perturbation energy b of the La-th decomposition layer at the i-th point. i,r r=1,2,3,…,La; then, the perturbation energies of all La decomposition layers at point i are combined into a multi-scale perturbation eigenvector m. iFinally, the above method is applied traversally to the normalized atmospheric temperature profiles at different altitudes on M ground-based fixed points of meteorological UAVs to obtain M multi-scale eigenvectors. These M multi-scale eigenvectors are then concatenated into a multi-scale perturbation basis matrix Φm=[m1;m2;m3;…;m…]. M ], m M The system consists of M low-scale feature vectors with dimensions M×Num_H×La. The Φs and Φc generated in steps 2 and 3, along with the Φm generated in this step, are then combined to construct the complete dictionary basis matrix Φ for atmospheric temperature. T =[Φs||Φc||Φm]. The atmospheric temperature profile datasets from steps 3 and 4 are updated to atmospheric humidity profile datasets, precipitation particle mass-weighted average diameter profile datasets, and precipitation particle spectrum normalized intercept profile datasets, respectively. Combined with the Φs generated in step 2, the complete dictionary basis matrix Φ for the atmospheric humidity normalized intercept can be generated. Q The complete dictionary basis matrix Φ of the normalized intercept of the mass-weighted average diameter of precipitation particles. Dm The complete dictionary basis matrix Φ of the normalized intercept of precipitation particle spectrum Nw ;

[0077] The formula for calculating the perturbation energy extracted from the detail coefficients is: the perturbation energy at the i-th point at the r-th decomposition layer scale. ;

[0078] Multi-scale perturbation eigenvector m i =[b i,1 ,b i,2 ,b i,3 ,…,b i,La ];

[0079] Step 5: As Figure 5 As shown, firstly, step 3 is applied to the atmospheric temperature profile data from M fixed ground points of meteorological UAVs, resulting in M ​​interpolated atmospheric temperature profile data, denoted as... Secondly, based on the complete atmospheric temperature dictionary basis matrix Φ generated in step 4... T Using a sparse representation model, The observation data d for each observation point g (g=1,2,3,..,M×Num_H) g Use Φ T and sparse coefficient vector a g It is represented in the form of a linear combination; then, d g The input is fed into a computational model that combines the likelihood function and sparse priors. Running the model yields a sparse coefficient vector a for each observation point. g After that, a gThe input is fed into a Bayesian inference and maximum a posteriori estimation model with sparse constraints, and the model is run to obtain the optimized sparse coefficient vector. Finally, based on the generated dictionary base Φ and the optimized... An optimized sparse representation model is used to optimize the atmospheric temperature dataset. The optimized atmospheric temperature dataset is denoted as... The complete dictionary base matrix Φ used in this step. T The atmospheric temperature profile dataset was updated to Φ. Q And atmospheric humidity profile dataset, Φ Dm And the mass-weighted average diameter profile dataset of precipitation particles, Φ Nw The dataset of normalized intercept profiles of precipitation particle spectra can be used to optimize atmospheric humidity, mass-weighted average diameter of precipitation particles, and normalized intercept data of precipitation particle spectra, respectively denoted as . , , ;

[0080] The sparse representation model is as follows: ,in Gaussian noise with zero mean.

[0081] The sparsity constraint is: , for Norm regularization term, T denotes transpose;

[0082] The optimized sparse representation model is as follows: ;

[0083] Step 6: As Figure 6 As shown, firstly, based on the optimized atmospheric temperature dataset generated in step 5... The dataset is grouped according to height levels to form a two-dimensional sparse plane dataset T_Z with multiple height levels. h h=1,2,3,…,Num_H; secondly, based on the height layer resolution delt_H set in step 2, the horizontal space height resolution is set to delt_R_H, which generates a target 3D mesh Tar_grid with a spatial resolution of delt_R_H×delt_R_H×delt_H; then, for T_Z h Let T_Z be the k-th height layer dataset of Tar_grid. k And Tar_grid_k, calculate T_Z k The maximum, median, and minimum values ​​are denoted as max(T_Z). k ), med(T_Z k ), min(T_Z kAfter that, T_Z k With max(T_Z) k ), med(T_Z k ), min(T_Z k The inputs are combined with the inputs into the structural similarity weight coefficient calculation model. Running the model outputs T_Z. k The set of structural weight coefficients W1_T_Z for each sparse point k Then, calculate the relationship between point pt and T_Z in Tar_grid_k. k The spatial distances between all sparse observation points are calculated, and the results are input into the spatial weight coefficient calculation model. The model is then run, and the output is T_Z. k The set of spatial weight coefficients for each sparse point in W2_T_Z k After that, W1_T_Z k With W2_T_Z k Multiply and normalize to form the total weight coefficient set W_T_Z k Finally, based on the weighted projection formula, the atmospheric temperature value T_grid_pt at point pt in Tar_grid_k is calculated. Applying the above steps to all points in Tar_grid_k and iterating through all height layers in Tar_grid yields a high spatial resolution 3D atmospheric temperature dataset T_3D. This atmospheric temperature dataset is then sequentially updated with atmospheric humidity, the mass-weighted average diameter of precipitation particles, and the normalized intercept of the precipitation particle spectrum. This results in high spatial resolution 3D datasets for atmospheric humidity, the mass-weighted average diameter of precipitation particles, and the normalized intercept of the precipitation particle spectrum, denoted as Q_3D, Dm_3D, and Nw_3D, respectively.

[0084] The weighted projection formula is: , and T_Z k and W_T_Z k The value of the m-th point;

[0085] Step 7: As Figure 7 As shown, firstly, based on the T_3D, Q_3D, Dm_3D, and Nw_3D datasets generated in step 6, the reasonable variation ranges [Dm_min, Dm_max] and [Nw_min, Nw_max] of Dm_3D and Nw_3D are set to [0.1mm, 3.5mm] and [10mm, 3.5mm], respectively. 2 10 6.5First, data in the Dm_3D and Nw_3D datasets that exceed the reasonable range of variation are removed. Next, the Dm_3D and Nw_3D datasets after outlier removal are input into a 3×3×3 median filter, and the filtered outputs are denoted as Dm_3D_f and Nw_3D_f, respectively. Then, the T_3D and Q_3D datasets are input into a spatial Gaussian smoothing filter, and the outputs are the filtered atmospheric temperature and humidity datasets, denoted as T_3D_f and Q_3D_f, respectively. Next, the atmospheric humidity threshold for precipitation is set to Q_th, and all grid points in Q_3D_f with atmospheric humidity values ​​greater than Q_th are marked as 1, while the remaining grid points are marked as 0, generating a precipitation identification mask matrix Rain_Mask. Finally, the phase states of precipitation particles are set to three types: liquid, solid-liquid mixture, and solid. For any point P_s in the target 3D mesh Tar_grid established in step 6, by combining T_3D_f and Q_3D_f, the atmospheric temperature and humidity data of that point can be obtained. and Then, using the formula for calculating the phase probability of precipitation particles, the probabilities of the spatial point P_s being in the liquid, solid-liquid mixture, and solid states are calculated respectively, and denoted as P_L. s P_M s P_I s The above method is applied to all spatial points in the target 3D mesh Tar_grid to generate probability matrices for three phase states: liquid, solid-liquid mixture, and solid, denoted as P_L, P_M, and P_I, respectively. Then, the correlation indices for liquid, solid-liquid mixture, and solid particles are set to alph_L=7.0, alph_M=6.0, and alph_I=5.5, respectively. Finally, [alph_L, T_3D, Q_3D, Dm_3D, Nw_3D, Rain_Mask], [alph_M, T_3D, Q_3D, Dm_3D, Nw_3D, Rain_Mask], and [alph_I, T_3D, Q_3D, Dm_3D, Nw_3D, Rain_Mask] are compared with those collected in step 1. The radar wavelength is input into the phase-sensitive radar echo simulation system. The system runs and outputs a three-dimensional dataset of radar reflectivity factors Z_L, Z_M, and Z_I for all points in the target's three-dimensional grid Tar_grid when the precipitation particles are in liquid, solid-liquid mixture, and solid states. Finally, based on the calculated P_L, P_M, P_I and Z_L, Z_M, and Z_I, the three-dimensional data Z_grid of high spatial resolution phased array weather radar reflectivity factors is calculated according to the multi-phase radar reflectivity factor joint calculation formula.

[0086] The formula for calculating the phase probability of precipitation particles is as follows: ,

[0087] ,

[0088] ,

[0089] ,

[0090] in The fuzziness control factor, when a is 1 or 2, is used to distinguish whether the input is temperature or humidity to calculate the probability. F represents the function, and X is the input variable.

[0091] The joint calculation formula for multiphase radar reflectivity factor is: Z_grid=Z_L×P_L+Z_M×P_M+Z_I×P_I;

[0092] Step 8: As Figure 8As shown, firstly, based on the radial range resolution delt_R, beamwidth delt_sita, maximum unambiguous range Rmax_radar, and the radar elevation scan sequence el_array collected in step 1, a range library sequence bins_array=[delt_R,2×delt_R,3×delt_R,…,Rmax_radar] and an azimuth sequence az_array=[delt_sita,2×delt_sita,3×delt_sita,…,360︒] are generated. Then, el_array, az_array, and bins_array are combined to generate the spatial position matrix P_echo=[el_array,az_array,bins_array] of the echo in the stereo scanning mode of the phased array weather radar. Secondly, assuming a point in P_echo is P0, and its spatial position is [el_0, az_0, bin_0], where el_0, az_0, and bin_0 are the elevation, azimuth, and range of the point, respectively, the unit direction vector dr of the radar beam is calculated according to the unit direction vector calculation formula. Then, the polar coordinate angle code E0_dir is calculated according to the polar coordinate angle encoding formula, and the three-dimensional relative coordinate code E0_rel is calculated according to the three-dimensional relative coordinate encoding calculation formula. Next, the calculated E0_rel, E0_dir, and dr are combined to generate the azimuth guidance encoding vector E0 = [E0_rel||E0_dir||dr] for point P0. This process is applied to every point in P_echo, and the generated azimuth guidance encoding vectors for each point are combined column-wise to form the azimuth guidance encoding matrix E_ALL. Next, the spatial location information P_radar of the radar antenna feed collected in step 1, the Z_grid generated in step 7, and the azimuth guidance coding matrix E_ALL generated in this step are input into the volume scan data implicit sampling simulator. The simulator is run and the output is the radar reflectivity factor data in the phased array weather radar stereo scanning mode.

[0093] The formula for calculating the unit direction vector is: dr=[cos(el_0)×sin(az_0),cos(el_0)×cos(az_0),sin(el_0)].

[0094] The polar coordinate angle encoding formula is: E0_dir=[sin(k1×az_0),cos(k1×az_0),sin(k1×el_0),cos(k1×el_0),…,sin(k n ×el_0),cos(k n ×el_0)], where k1 and k n 2 and 2 respectively nFrequency increment of the frequency base.

[0095] The formula for calculating the three-dimensional relative coordinate encoding is: E0_rel=bin_0×dr.

[0096] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and improvements, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for simulating reflectivity factors of phased array radar based on sparse radiosonde data, characterized in that: The simulation method comprises: S1, collecting sparse sounding profile data sets and auxiliary data, and phased array weather radar spatial position and operating state data; generating height uniform vectors, and generating a spatial base matrix according to the collected data; S2, taking the data collected in S1 as input, obtaining normalized atmospheric temperature profiles by adopting inverse distance weighted interpolation and normalization processing, and obtaining an attribute disturbance frequency spectrum base matrix in combination with a 1D discrete Fourier transform model; generating a multi-scale disturbance base matrix according to the normalized atmospheric temperature profiles and a wavelet transform model, and generating a complete dictionary base matrix of atmospheric temperature according to the spatial base matrix, the attribute disturbance frequency spectrum base matrix and the multi-scale disturbance base matrix; updating the atmospheric temperature profile into normalized intercept profile data of atmospheric humidity, mass-weighted average diameter of precipitation particles and precipitation particle spectrum to obtain a complete dictionary base matrix of the normalized intercept of the atmospheric humidity, the mass-weighted average diameter of the precipitation particles and the precipitation particle spectrum; S3, obtaining sparse coefficient vectors of each observation point by using a sparse representation model in combination with a likelihood function and a sparse prior calculation model according to the data obtained in S2, and optimizing by using the sparse representation model to realize optimization of the atmospheric temperature, the atmospheric humidity, the mass-weighted average diameter of the precipitation particles and the normalized intercept data of the precipitation particle spectrum; S4, processing the data optimized in S3 and obtaining a calculation result in combination with a structural similarity weight coefficient calculation model and a spatial weight coefficient calculation model, multiplying the calculation result and normalizing, and then obtaining a high spatial resolution three-dimensional data set of the atmospheric temperature, the atmospheric humidity, the mass-weighted average diameter of the precipitation particles and the normalized intercept of the precipitation particle spectrum according to a weighted projection formula; S5, filtering the data obtained in S4, processing the filtered data and calculating high spatial resolution phased array weather radar reflectivity factor three-dimensional data in combination with the data of S4 and a precipitation particle phase state probability calculation formula and a multi-phase radar reflectivity factor joint calculation formula; S6, generating a spatial position matrix according to the data collected in S1, combining the calculation of each point in the spatial position matrix with a direction guide coding vector column by column to form a direction guide coding matrix, and obtaining radar reflectivity factor data in a phased array weather radar stereoscopic scanning mode in combination with the data collected in S1 and the high spatial resolution phased array weather radar reflectivity factor three-dimensional data obtained in S5.

2. The method of claim 1, wherein: The collection of sparse sounding profile data sets and auxiliary data, and phased array weather radar spatial position and operating state data specifically comprises the following contents: Collecting sparse sounding profile data sets and auxiliary data obtained by an array of meteorological unmanned aerial vehicles; The sparse sounding profile dataset acquired by the meteorological UAV array includes: a total of M ground fixed points corresponding to the meteorological UAV array, and an atmospheric temperature profile dataset T_TR at different altitudes at the i-th meteorological UAV ground fixed point. i Atmospheric humidity profile dataset Q_TR i Mass-weighted average diameter profile dataset of precipitation particles Dm_TR i Normalized intercept profile dataset of precipitation particle spectra Nw_TR i , i=1,2,3,…M; the auxiliary data for the meteorological UAV array includes: the spatial location information of observation points at different altitudes on the i-th ground fixed point of the meteorological UAV, P i = (Lon_TR) i Lat_TR i Hig_TR i ), where Lon_TR i For longitude points, Lat_TR i Latitude point, Hig_TR i A vector representing different heights, i = 1, 2, 3, ..., M; The collection of spatial location and operational status data for phased array weather radar mainly includes: spatial location information of the phased array weather radar antenna feed P_radar=[lon_radar,lat_radar,hig_radar], where lon_radar, lat_radar, and hig_radar are longitude, latitude, and altitude, respectively; radar wavelength, radar radial range resolution delt_R, beamwidth delt_sita, and radar scanning sequence el_array=[el0,el1,el2,…,el] at the elevation layer. max ], el max Rmax_radar represents the maximum scanning elevation angle and the maximum unambiguous distance.

3. The method of claim 2, wherein: The generation of height uniform vectors and the generation of a spatial base matrix according to the collected data specifically comprises the following contents: A1, set the resolution and multiple of height layer respectively as delt_H and Num_H, generate the maximum height layer as Hig_max=Num_Hxdelt_H, and further generate the height uniform vector Z on each unmanned aerial vehicle ground fixed point i =hxdelt_H, wherein h=1, 2, 3, …, Num_H, i=1, 2, 3, …, M. A2, according to the collected P i = (Lon_TR i , Lat_TR i , Hig_TR i ), replace Hig_TR i in P i with Z i , generate each weather unmanned aerial vehicle ground fixed point after the height unified after the profile data space position information P i_uni = (Lon_TR i , Lat_TR i , Z i ); A3, set the observation point P j For any one of the spatial points, the longitude, latitude and height of the observation point P i_uni j are normalized in longitude, latitude and height dimensions respectively to obtain the normalized longitude , latitude and height , and the normalized ( ), j=1,2,3,..,MxNum_H;​ A4. Generating the observation point P according to the second order polynomial orthogonal basis function j of the spatial structure vector , denotes the first second order polynomial orthogonal basis function; A5, respectively, according to the method of steps A3 and A4, MxNum_H P j Perform traversal processing, and combine the processed MxNum_H spatial structure vectors to generate a spatial basis matrix Φs[ ] with a dimension of MxNum_Hx10.

4. The method of claim 3, wherein: The taking of the data collected in S1 as input, the obtaining of normalized atmospheric temperature profiles by adopting inverse distance weighted interpolation and normalization processing, and the obtaining of an attribute disturbance frequency spectrum base matrix in combination with a 1D discrete Fourier transform model specifically comprises the following contents: B1. Collecting the atmospheric temperature profile data set T_TR of different height layers of the i-th meteorological unmanned aerial vehicle ground fixed point i and the spatial position information P corresponding to the profile i As input, the inverse distance weighted interpolation method is used to interpolate T_TR i to the generated spatial information P i_uni The atmospheric temperature profile data after interpolation at each meteorological unmanned aerial vehicle ground fixed point is obtained at the spatial position of the target profile, denoted as ; B2. Calculate the profile data. The mean and standard deviation are denoted as mean( ) and std( Then, according to the Z-Score normalization formula, the profile data is normalized. After normalization, the normalized atmospheric temperature profile is obtained. ; B3, the atmospheric temperature profile is input into a 1D discrete Fourier transform model, the model is run, and the output of the model is taken modulo to obtain a spectrum modulo sequence DFT of the atmospheric temperature profile i = |DFT( )|; B4, the DFT i The modulus of the middle and high frequency part is set to zero, and the modulus of the low frequency part is reserved. The DFT i The first Ka low frequency components except the direct current component are respectively recorded as DFT i,2 , DFT i,3 , …, DFT i,Ka+1 , and the Ka low frequency components are combined into a one-dimensional vector, that is, the disturbance modal vector c i , Ka is the number of selected low frequency components; B5, traverse the method of B1-B4 steps to the atmospheric temperature profile data of M meteorological unmanned aerial vehicle ground fixed points on different height layers, obtain M disturbance modal vectors, respectively denoted as c1, c2, c3, …, c M , and splice the M disturbance modal vectors into the attribute disturbance frequency spectrum basis matrix Φc=[c1;c2;c3;…;c M ], the dimension of which is MxNum_HxKa.

5. The method of claim 4, wherein: The complete dictionary matrix of the atmospheric temperature, humidity, mass-weighted mean diameter of the precipitation particles and the normalized intercept of the precipitation particle spectrum is obtained by updating the atmospheric temperature profile to the normalized intercept profile of the atmospheric humidity, mass-weighted mean diameter of the precipitation particles and the precipitation particle spectrum. C1, the generated normalized atmospheric temperature profile into the wavelet transform model, select the type of wavelet basis, set the decomposition level La, run the model, and the model output is the atmospheric temperature profile at the i-th point and wherein is the La-th layer approximation coefficient, is the r-th layer detail coefficient, r = 1, 2, 3, …, La; C2, according to the detail coefficient extraction disturbance energy calculation formula, extract the disturbance energy b of the i-th point at La decomposition layers i,r , r = 1, 2, 3, …, La; then, the disturbance energy b of all La decomposition layers at the i-th point i,1 , b i,2 , b i,3 , …, b i,La Combined into a multi-scale disturbance feature vector m i ; C3, the method of C1 and C2 steps is applied to the normalized atmospheric temperature profile of different height layers of M meteorological unmanned aerial vehicle ground fixed points, M multi-scale feature vectors are obtained, and the M multi-scale feature vectors are spliced into a multi-scale disturbance basis matrix Φm=[m1;m2;m3;…;m M ],m M is the Mth multi-scale feature vector, and the dimension is MxNum_HxLa; C4, the generated Φs, Φc and Φm generated in this step are combined to form the complete dictionary basis matrix Φ of atmospheric temperature T =[Φs||Φc||Φm], the atmospheric temperature profile data set is updated to the atmospheric humidity profile data set, the mass-weighted mean diameter profile data set of the precipitation particles, and the normalized intercept profile data set of the precipitation particle spectrum, respectively, and combined with Φs generated by A5 of S1 to generate the complete dictionary basis matrix Φ of the normalized intercept of atmospheric humidity Q , the complete dictionary basis matrix Φ of the normalized intercept of the mass-weighted mean diameter of the precipitation particles Dm , and the complete dictionary basis matrix Φ of the normalized intercept of the precipitation particle spectrum Nw .

6. The method of claim 5, wherein: The S3 specifically includes the following contents: S301, applying the step of B1 to the atmospheric temperature profile data on the M meteorological unmanned aerial vehicle ground fixed points, M interpolated atmospheric temperature profile data can be obtained, denoted as ; S302, Based on the generated complete dictionary base matrix of atmospheric temperature Φ T Using a sparse representation model, The observation data d at each observation point g g Use Φ T and sparse coefficient vector a g Representing d in the form of a linear combination g The input is fed into a computational model that combines the likelihood function and sparse priors. Running the model yields a sparse coefficient vector a for each observation point. g , will a g The input is fed into a Bayesian inference and maximum a posteriori estimation model with sparse constraints, and the model is run to obtain the optimized sparse coefficient vector. ; S303、based on the generated dictionary base Φ and the optimized , the atmospheric temperature data set is optimized by using the optimized sparse representation model, and the optimized atmospheric temperature data set is denoted as ; S304, Φ in this step T The atmospheric temperature profile dataset was updated to Φ. Q And atmospheric humidity profile dataset, Φ Dm And the mass-weighted average diameter profile dataset of precipitation particles, Φ Nw The datasets of normalized intercept profiles of precipitation particle spectra are used to optimize atmospheric humidity, mass-weighted average diameter of precipitation particles, and normalized intercept data of precipitation particle spectra, respectively. These are denoted as... , , .

7. The method of claim 6, wherein: The S4 specifically includes the following contents: S401、According to the generated optimized atmospheric temperature dataset The dataset is grouped according to the height layer to form a plurality of height layer two-dimensional sparse plane datasets T_Z h h = 1, 2, 3, …, Num_H; S402、According to the resolution of the set height layer delt_H, the high spatial resolution of the horizontal space is set as delt_R_H, a target three-dimensional grid Tar_grid with a spatial resolution of delt_R_H*delt_R_H*delt_H is generated, and T_Z h and the kth height layer data set of Tar_grid are denoted as T_Z k and Tar_grid_k, respectively, the maximum, median and minimum of T_Z k are calculated, and are denoted as max(T_Z k ), med(T_Z k ), and min(T_Z k ), respectively. S403, T_Z k With max(T_Z) k ), med(T_Z k ), min(T_Z k The inputs are combined with the inputs into the structural similarity weight coefficient calculation model. Running the model outputs T_Z. k The set of structural weight coefficients W1_T_Z for each sparse point k Calculate the relationship between a point pt and T_Z in Tar_grid_k. k The spatial distances between all sparse observation points are calculated, and the results are input into the spatial weight coefficient calculation model. The model is then run, and the output is T_Z. k The set of spatial weight coefficients for each sparse point in W2_T_Z k W1_T_Z k With W2_T_Z k Multiply and normalize to form the total weight coefficient set W_T_Z k According to the weighted projection formula, the atmospheric temperature value T_grid_pt at point pt in Tar_grid_k is calculated; S404, the method of S403 is applied to all points in Tar_grid_k, and the step is applied to all height layers in Tar_grid to obtain a high spatial resolution atmospheric temperature three-dimensional data set T_3D, and the atmospheric temperature data set of the step is sequentially updated to the atmospheric humidity, mass-weighted mean diameter of the precipitation particles and the normalized intercept of the precipitation particle spectrum data set, and high spatial resolution atmospheric humidity, mass-weighted mean diameter of the precipitation particles and normalized intercept of the precipitation particle spectrum three-dimensional data sets are obtained, respectively, and are denoted as Q_3D, Dm_3D and Nw_3D.

8. The method of claim 7, wherein the method is characterized by: The S5 specifically includes the following contents: S501, according to the generated T_3D, Q_3D, Dm_3D and Nw_3D data sets, setting the change range of Dm_3D and Nw_3D to [Dm_min, Dm_max] and [Nw_min, Nw_max] respectively, and removing the data in the Dm_3D and Nw_3D data sets that exceed the change range; S502, input the Dm_3D and Nw_3D data sets after removing the outliers into a 3-point × 3-point × 3-point median filter, and the filtered outputs are denoted as Dm_3D_f and Nw_3D_f, respectively; input the T_3D and Q_3D data sets into a spatial Gaussian smoothing filter, and the outputs are the filtered atmospheric temperature and humidity data sets, denoted as T_3D_f and Q_3D_f, respectively; S503, setting the atmospheric humidity threshold for precipitation occurrence as Q_th, marking all grid points with atmospheric humidity value greater than Q_th in Q_3D_f as 1, and marking the remaining grid points as 0 to generate a precipitation identification mask matrix Rain_Mask, and setting the phase of the precipitation particles as liquid, solid-liquid mixed and solid. S504, for any point P_s in the established target three-dimensional grid Tar_grid, combined with T_3D_f and Q_3D_f, obtain the atmospheric temperature and atmospheric humidity data of the point and , and then calculate the probabilities of the three phase states of liquid, solid-liquid mixed state and solid state of the space point P_s respectively using the precipitation particle phase state probability calculation formula, and record them as P_L s , P_M s , and P_I s respectively; S505, apply the step of S504 to all spatial points in the target three-dimensional grid Tar_grid, generate the probability matrixes of liquid, solid-liquid mixed and solid three phases, respectively recorded as P_L, P_M and P_I, set the related indexes of liquid, solid-liquid mixed and solid particles as alph_L, alph_M and alph_I, respectively, and then input [alph_L, T_3D, Q_3D, Dm_3D, Nw_3D, Rain_Mask], [alph_M, T_3D, Q_3D, Dm_3D, Nw_3D, Rain_Mask], [alph_I, T_3D, Q_3D, Dm_3D, Nw_3D, Rain_Mask] and the collected radar wavelength wavelength into the phase state sensitive radar echo simulation system, run the system, and output the radar reflectivity factor three-dimensional data set Z_L, Z_M and Z_I of all points in the target three-dimensional grid Tar_grid when the precipitation particle phase is liquid, solid-liquid mixed and solid, respectively, calculate the high spatial resolution phased array weather radar reflectivity factor three-dimensional data Z_grid according to the calculated P_L, P_M and P_I, and Z_L, Z_M and Z_I, and according to the multi-phase radar reflectivity factor joint calculation formula.

9. The method of claim 8, wherein: The S6 specifically includes the following contents: S601, according to the collected phased array weather radar radial distance resolution delt_R, beam width delt_sita, maximum unambiguous distance Rmax_radar and radar elevation scanning sequence el_array, generate distance bin sequence bins_array=[delt_R, 2×delt_R, 3×delt_R, …, Rmax_radar], azimuth sequence az_array=[delt_sita, 2×delt_sita, 3×delt_sita, …, 360°], combine el_array, az_array and bins_array to generate the spatial position matrix P_echo=[el_array, az_array, bins_array] of the phased array weather radar echo in the three-dimensional scanning mode; S602, set a point in P_echo as P0, the spatial position of the point is [el_0, az_0, bin_0], wherein el_0, az_0 and bin_0 are the elevation angle, azimuth angle and distance bin of the point, respectively, then calculate the unit direction vector dr of the radar beam ray according to the unit direction vector calculation formula, calculate the polar coordinate angle coding E0_dir according to the polar coordinate angle coding formula, and calculate the three-dimensional relative coordinate coding E0_rel according to the three-dimensional relative coordinate coding calculation formula, combine the calculated E0_rel, E0_dir and dr to generate the azimuth direction coding vector E0 of the point P0. S603, apply the step of S602 to each point in P_echo, and combine the generated azimuthally steered encoding vector of each point by column to form an azimuthally steered encoding matrix E_ALL, input the collected spatial position information P_radar of the radar antenna feed, the generated Z_grid and the azimuthally steered encoding matrix E_ALL together into a volumetric scanning data implicit sampling simulator, run the simulator, and output the radar reflectivity factor data under the phased array weather radar stereoscopic scanning mode.

Images