Meteorological radar three-dimensional wind field inversion method and system based on multi-source data fusion

Abstract

This invention relates to the field of three-dimensional wind field inversion technology using meteorological radar, specifically disclosing a method and system for inverting three-dimensional wind fields based on multi-source data fusion. The invention involves acquiring multi-source data from ground, radar, and satellite sources, performing data cleaning and spatiotemporal alignment; extracting multi-source feature data, performing weighted fusion to construct an initial three-dimensional wind field; performing physical constraint optimization to generate an optimized three-dimensional wind field; performing spatial interpolation and multi-scale feature fusion to expand the optimized three-dimensional wind field, generating a full-space three-dimensional wind field; and verifying and updating the full-space three-dimensional wind field. This invention enables multi-source data acquisition from ground, radar, and satellite sources, performing multi-source feature extraction and weighted fusion to construct an initial three-dimensional wind field, and performing constraint optimization and wind field expansion to generate a full-space three-dimensional wind field. It effectively avoids uncertainties in the lateral and vertical directions of the inversion results and solves the problem of large inversion errors in local areas.

Description

Technical Field

[0001] This invention belongs to the field of three-dimensional wind field inversion technology of meteorological radar, and particularly relates to a method and system for three-dimensional wind field inversion of meteorological radar based on multi-source data fusion. Background Technology

[0002] Three-dimensional wind field inversion by weather radar is a technique that uses weather radar to observe the atmosphere and, through physical constraints and mathematical inversion models, calculates the wind speed and direction distribution in three-dimensional space. Doppler weather radar is the main observation method. By acquiring observation data such as radar radial velocity and reflectivity, and combining them with radar geometry, the radial wind information that radar can only directly observe is inverted into a complete three-dimensional wind vector field that includes east-west, north-south, and vertical directions.

[0003] The three-dimensional wind field inversion results of meteorological radar are important basic data for monitoring severe convective weather, short-term forecasting, severe weather warning and numerical weather prediction assimilation. They are of great significance for improving the understanding of atmospheric motion mechanisms and the level of precision and intelligence of meteorological operations.

[0004] In existing technologies, three-dimensional wind field inversion mainly relies on single radar data. Since radar can only observe the radial wind component and cannot directly obtain the complete three-dimensional wind vector, the inversion results have large uncertainties in the lateral and vertical directions, and are prone to large inversion errors in local areas. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for inverting three-dimensional wind fields from meteorological radar based on multi-source data fusion, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions: A method for retrieving three-dimensional wind fields from meteorological radar based on multi-source data fusion, the method specifically includes the following steps: Multi-source data acquisition is performed from ground, radar, and satellite sources to obtain multi-source acquired data. The multi-source acquired data is then cleaned and spatiotemporally aligned to obtain multi-source aligned data. Multi-source feature data is extracted from the multi-source aligned data, and the multi-source feature data is weighted and fused to construct an initial three-dimensional wind field; Using multi-source aligned data, the initial three-dimensional wind field is sequentially subjected to momentum conservation constraints, vertical motion correction, and mass conservation constraints to generate an optimized three-dimensional wind field; Based on the optimized three-dimensional wind field, spatial interpolation and multi-scale feature fusion are performed to expand the optimized three-dimensional wind field and generate a full-space three-dimensional wind field. Multi-source comparison data is acquired to verify and update the full-space three-dimensional wind field.

[0007] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention effectively overcomes the limitations of insufficient lateral and vertical inversion information from single radar data by fusing multi-source data from ground, radar, and satellite, and performing feature extraction and weighted fusion, thereby improving the integrity and spatial coverage of the three-dimensional wind field.

[0008] 2. This invention introduces multiple physical constraints such as momentum conservation, vertical motion correction, and mass conservation, and combines them with a stepwise optimization and local adjustment mechanism to significantly enhance the physical consistency and vertical motion accuracy of the inverted wind field.

[0009] 3. This invention achieves wind field expansion based on spatial interpolation and multi-scale feature fusion. Combined with terrain correction and multi-source verification updates, it reduces local inversion errors and improves the reliability and applicability of the full-space three-dimensional wind field under complex weather conditions. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention.

[0011] Figure 1 A flowchart of the method provided by an embodiment of the present invention is shown.

[0012] Figure 2 A flowchart of multi-source acquisition and data processing in the method provided by an embodiment of the present invention is shown.

[0013] Figure 3 A flowchart illustrating the generation of a full-space three-dimensional wind field in the method provided by an embodiment of the present invention is shown.

[0014] Figure 4 An application architecture diagram of the system provided in an embodiment of the present invention is shown.

[0015] Figure 5 A structural block diagram of the multi-source acquisition and processing unit in the system provided by an embodiment of the present invention is shown.

[0016] Figure 6 A structural block diagram of the initial wind field construction unit in the system provided by an embodiment of the present invention is shown.

[0017] Figure 7 A structural block diagram of the wind field expansion processing unit in the system provided by an embodiment of the present invention is shown. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] Understandably, the three-dimensional wind field inversion results from meteorological radar are crucial foundational data for severe convective weather monitoring, short-term forecasting, hazardous weather warnings, and numerical weather prediction assimilation. They are of great significance for improving our understanding of atmospheric motion mechanisms and enhancing the precision and intelligence of meteorological operations. However, current technologies primarily rely on single radar data for three-dimensional wind field inversion. Since radar can only observe the radial wind component and cannot directly obtain the complete three-dimensional wind vector, the inversion results exhibit significant uncertainties in the lateral and vertical directions, easily leading to large inversion errors in localized areas.

[0020] To address the aforementioned issues, this invention employs multi-source data acquisition from ground, radar, and satellite sources to obtain multi-source acquired data. This data is then cleaned and spatiotemporally aligned to obtain multi-source aligned data. Multi-source feature data is extracted from the aligned data and weighted fusion is performed to construct an initial 3D wind field. Physical constraints are applied to the initial 3D wind field to generate an optimized 3D wind field. Based on the optimized 3D wind field, spatial interpolation and multi-scale feature fusion are performed to expand the wind field, generating a full-space 3D wind field. Multi-source comparison data is acquired to verify and update the full-space 3D wind field. This method enables multi-source data acquisition from ground, radar, and satellite sources, performing feature extraction and weighted fusion to construct an initial 3D wind field, and then performing constraint optimization and wind field expansion to generate a full-space 3D wind field. This effectively avoids uncertainties in the lateral and vertical directions of the inversion results and solves the problem of large inversion errors in local areas.

[0021] Figure 1 A flowchart of the method provided by an embodiment of the present invention is shown.

[0022] Specifically, the method for retrieving three-dimensional wind fields from meteorological radar based on multi-source data fusion includes the following steps: Step S101: Perform multi-source data acquisition from ground, radar, and satellite to obtain multi-source acquired data, and perform data cleaning and spatiotemporal alignment on the multi-source acquired data to obtain multi-source aligned data.

[0023] In this embodiment of the invention, multi-source data acquisition is performed from ground, radar, and satellite sources to obtain multi-source acquired data including ground-acquired data, radar-acquired data, and satellite-acquired data. Ground-acquired data includes meteorological wind speed, meteorological wind direction, air pressure, temperature, and vertical velocity of a wind profiler; radar-acquired data includes Doppler radar radial velocity, reflectivity factor, and phased array radar scan data; satellite-acquired data includes satellite cloud image data, atmospheric motion vectors, and water vapor channel brightness and temperature data. The multi-source acquired data is then subjected to anomaly identification and cleaning processing to obtain multi-source cleaned data. Subsequently, the multi-source cleaned data is time-aligned and spatially aligned to obtain multi-source aligned data. Specifically, the anomaly identification and cleaning processing includes: removing outliers in the ground-acquired data where wind speed exceeds the instrument's range; removing noise points in the radar-acquired data where signal strength is below a preset strength threshold; and filtering areas in the satellite-acquired data where cloud cover is below a preset percentage.

[0024] Specifically, Figure 2 A flowchart of multi-source acquisition and data processing in the method provided by an embodiment of the present invention is shown.

[0025] In a preferred embodiment of the present invention, the step of acquiring multi-source data from ground, radar, and satellite sources, obtaining multi-source acquired data, and performing data cleaning and spatiotemporal alignment on the multi-source acquired data to obtain multi-source aligned data specifically includes the following steps: Step S1011: Perform multi-source data acquisition from ground, radar, and satellite to obtain multi-source acquired data; Step S1012: Perform anomaly identification and cleaning processing on the multi-source collected data to obtain multi-source cleaned data; Step S1013: Perform time alignment and spatial alignment on the multi-source cleaning data to obtain multi-source aligned data.

[0026] Specifically, the multi-source collected data undergoes anomaly identification and cleaning processing to obtain multi-source cleaned data, which includes the following steps: The instrument's measurement range is preset to indicate the normal ranges for meteorological wind speed, air pressure, and temperature. Based on the normal ranges for meteorological wind speed, values ​​outside the normal ranges are removed from the ground-collected data. Similarly, based on the normal ranges for air pressure, values ​​outside the normal ranges for air pressure are removed from the ground-collected data. Furthermore, based on the normal ranges for temperature, values ​​outside the normal ranges for temperature are removed from the ground-collected data, thus obtaining ground-collected data that has been filtered to remove abnormalities. Reflectivity factor values ​​below a preset reflectivity factor intensity threshold are removed from the radar data. At the same time, Doppler radar radial velocity values ​​and phased array radar scan data corresponding to reflectivity factor values ​​below the reflectivity factor intensity threshold are also removed to obtain noise-removed radar data. The satellite cloud images in the satellite acquisition data are divided into grid areas according to latitude and longitude. Based on the preset cloud cover percentage threshold, grid areas with cloud cover percentages lower than the preset cloud cover percentage threshold are selected to obtain the selected grid areas. All atmospheric motion vectors and water vapor channel brightness and temperature data in the selected grid areas are removed to obtain satellite data with clear sky areas removed. Ground-based data with anomalies removed, radar data with noise removed, and satellite data with clear-sky areas removed are matched by timestamps and geographic coordinates to obtain multi-source data groups. The absolute difference between the meteorological wind direction vector and the atmospheric motion vector in each group is calculated. When the absolute difference exceeds the preset angle tolerance range, the data in the current group is removed to obtain multi-source data after direction verification. According to the time sequence, the meteorological wind speed value and meteorological wind direction vector in the multi-source data after direction verification are continuously taken, and the rate of change of meteorological wind speed value and the amount of change of meteorological wind direction vector are calculated. When the rate of change of meteorological wind speed value exceeds the preset wind speed value mutation threshold or the amount of change of meteorological wind direction vector exceeds the preset direction mutation threshold, the meteorological wind speed value and meteorological wind direction vector corresponding to the intermediate time in the continuous taking are removed to obtain the time-series smoothed ground data. The time-smoothed ground data and the radar data portion of the multi-source data after direction verification are spatially overlaid. At each overlapping spatial coordinate, when the absolute value of the Doppler radar radial velocity value is greater than the meteorological wind speed value and exceeds the preset reasonable deviation range, the radar data acquired at the current overlapping spatial coordinate is removed to obtain the radar data after ground constraints. The radar data after ground constraint, the satellite data portion of the multi-source data after direction verification, and the ground data after time-series smoothing are collected and integrated to obtain multi-source cleaned data.

[0027] Furthermore, this step effectively eliminates anomalies and noise in the data by performing multi-level anomaly identification and cleaning on the multi-source acquired data, including range removal, noise filtering, clear sky removal, direction verification, time series smoothing, and ground constraints. This improves the accuracy and spatiotemporal consistency of the data and provides a high-quality data foundation for the subsequent construction of the three-dimensional wind field.

[0028] It should be noted that the preset angle tolerance range in this step is determined statistically based on the deviation of the wind direction system between the ground station and satellite observations. This invention sets it to ±30° and ±45°. If it exceeds this range, it is considered that there is inconsistency between the two observation sources, and it needs to be raised or verified.

[0029] Furthermore, the preset cloud cover percentage threshold in this step is used to identify invalid satellite observation data in clear sky areas. It can be set according to the sensitivity of the satellite channel and seasonal climate characteristics. In this invention, the value is set to 10% to 20%.

[0030] Furthermore, the meteorological radar three-dimensional wind field inversion method based on multi-source data fusion also includes the following steps: Step S102: Extract multi-source feature data from the multi-source aligned data, and perform weighted fusion on the multi-source feature data to construct an initial three-dimensional wind field.

[0031] In this embodiment of the invention, multi-source feature data, including ground feature data, radar feature data, and satellite feature data, is extracted from multi-source aligned data. The weights of the ground data, radar data, and satellite data are determined, and then imported into a preset weighted fusion model. The multi-source feature data is then fused according to the weighted fusion model, the ground data weights, the radar data weights, and the satellite data weights to construct an initial three-dimensional wind field. Specifically, the preset weighted fusion model can employ Kalman filtering or random forest regression. Among the determined ground data weights, radar data weights, and satellite data weights, the ground data weight is fixed at a high weight; the radar data weight is inversely proportional to distance; and the satellite data weight is directly proportional to cloud cover.

[0032] Specifically, in the preferred embodiment provided by the present invention, the step of extracting multi-source feature data from the multi-source aligned data and performing weighted fusion on the multi-source feature data to construct an initial three-dimensional wind field specifically includes the following steps: From the multi-source aligned data, extract multi-source feature data including ground feature data, radar feature data, and satellite feature data; Determine the weights for ground data, radar data, and satellite data; The multi-source feature data are fused according to the preset weighted fusion model, the ground data weight, the radar data weight, and the satellite data weight to construct an initial three-dimensional wind field.

[0033] Specifically, the multi-source feature data is fused according to the preset weighted fusion model, the ground data weights, the radar data weights, and the satellite data weights to construct an initial three-dimensional wind field. The specific steps are as follows: Meteorological wind speed values, meteorological wind direction vectors, and spatial coordinates are obtained from ground feature data. Using the spatial coordinates, the distance between grid points is calculated. Taking the position corresponding to the spatial coordinate as the grid point, all grid points are traversed. The product of the ground data weight and the reciprocal of the distance between grid points is used as the comprehensive weight to perform a weighted average on the wind speed values ​​and wind direction vectors to generate a preliminary ground wind field distribution. Based on the preliminary surface wind field distribution, Doppler radar radial velocity values, their corresponding spatial locations, and radar station coordinates are extracted from radar characteristic data. At each spatial location corresponding to a Doppler radar radial velocity value, the projection of the meteorological wind direction vector onto the radar line is calculated using the meteorological wind direction vector and the radar station coordinates. The relative deviation value is obtained by comparing the projection of the meteorological wind direction vector onto the radar line with the Doppler radar radial velocity value. Using the relative deviation value and the wind direction data of the grid points in the preliminary surface wind field distribution within the radar data area, after rotational adjustment, spatial interpolation is performed to generate the radar-corrected wind field data. Atmospheric motion vectors and radar altitude layer information parameters are obtained from radar-corrected wind field data; grid points lacking radar coverage in the upper atmosphere are located using radar-corrected wind field data; based on altitude layer information parameters, the grid points lacking radar coverage in the upper atmosphere are matched with atmospheric motion vectors according to altitude layer to obtain matching results; using satellite data weights, the overlapping areas of radar and satellite data in the matching results are weighted and fused to obtain satellite-fused upper-air wind field data; Using radar height layer information parameters, the vertical velocity of the wind profiler in the satellite-fused upper-air wind field data is matched with the grid points in the satellite-fused upper-air wind field data, and the vertical velocity component of the atmosphere is estimated by spatial interpolation to obtain a preliminary three-dimensional wind field grid. Based on the altitude layer information parameters, the air density of each altitude layer is calculated and obtained using the air pressure value in the ground feature data; the horizontal wind speed in the preliminary three-dimensional wind field grid is corrected using the air density of each altitude layer and the temperature value in the ground feature data to obtain the thermally adjusted three-dimensional wind field. Based on the reflectivity factor value in the thermally adjusted three-dimensional wind field, the vertical velocity increment caused by particle drag is determined; the vertical velocity increment is superimposed on the vertical velocity component contained in each grid in the thermally adjusted three-dimensional wind field, and the horizontal wind speed is adjusted according to mass conservation to obtain the three-dimensional wind field under the influence of precipitation particles. The brightness and temperature data of water vapor channels in the satellite feature data are converted into water vapor content values ​​for each grid point; the horizontal wind speed in the three-dimensional wind field under the influence of precipitation particles is multiplied by the water vapor content value to obtain the water vapor flux distribution; the three-dimensional wind field under the influence of precipitation particles is adjusted using the water vapor flux distribution to obtain the initial three-dimensional wind field.

[0034] Furthermore, this step constructs a preliminary wind field from ground data, and then uses radar radial velocity for directional correction, satellite data to fill in the upper-level gaps, and combines thermal effects to adjust density, precipitation particle drag to correct vertical motion, and water vapor flux adjustment to achieve a layer-by-layer fine fusion of multi-source characteristics, thereby constructing an initial three-dimensional wind field with a more complete physical mechanism and more comprehensive elements.

[0035] It should be noted that the "preset weighted fusion model" in this step can use Kalman filtering or random forest regression; When using Kalman filtering for weighted fusion, ground feature data, radar feature data, and satellite feature data are treated as different observation sources. The observation noise covariance matrix is ​​set according to their respective preset weights (ground data weight is fixed at a high level, radar data weight is inversely proportional to distance, and satellite data weight is directly proportional to cloud cover). Through state prediction equations and observation update equations, the fused state estimate is dynamically adjusted using Kalman gain, thereby achieving optimal fusion of multi-source data with minimum mean square error, resulting in a more accurate initial three-dimensional wind field. When using random forest regression for weighted fusion, a training sample set is first constructed using historical meteorological data. Each sample contains ground, radar, and satellite feature data for the same spatiotemporal point, along with its corresponding true wind field value (or high-precision reanalysis data). The random forest regression model is trained using multi-source feature data as input and wind field components (U, V, W) as output. In real-time applications, the current multi-source aligned data is input into the trained random forest model. The model automatically learns the importance and nonlinear relationships of different data sources by weighted averaging of the prediction results from multiple decision trees, thereby generating the fused initial three-dimensional wind field.

[0036] Furthermore, the meteorological radar three-dimensional wind field inversion method based on multi-source data fusion also includes the following steps: Step S103: Using multi-source aligned data, momentum conservation constraints, vertical motion correction and mass conservation constraints are sequentially applied to the initial three-dimensional wind field to generate an optimized three-dimensional wind field.

[0037] In this embodiment of the invention, the initial three-dimensional wind field is subject to mass conservation constraints. The wind field divergence is calculated and compared with a preset divergence threshold, and then the components are adjusted and corrected. The initial three-dimensional wind field is also subject to vertical motion correction and momentum conservation constraints. The wind field intensity is calculated to determine whether there is vorticity anomaly. If there is vorticity anomaly, smoothing filtering is performed to suppress small-scale noise and generate an optimized three-dimensional wind field.

[0038] Specifically, in the preferred embodiment provided by the present invention, the step of using multi-source aligned data to sequentially apply momentum conservation constraints, vertical motion corrections, and mass conservation constraints to the initial three-dimensional wind field to generate an optimized three-dimensional wind field includes the following steps: The reflectivity factor values ​​of the initial three-dimensional wind field are filtered according to a preset reflectivity screening threshold to select grid points that are higher than the preset reflectivity screening threshold. All grid points that are higher than the preset reflectivity screening threshold are traversed, and the angle between the meteorological wind direction vector and the radar beam is calculated using the meteorological wind direction vector and the Doppler radar radial velocity value of the initial three-dimensional wind field. When the angle between the meteorological wind direction vector and the radar beam exceeds the preset deviation range, the wind direction vector of the initial three-dimensional wind field is rotated to make the projection of the wind direction vector zero and obtain an intermediate wind field with momentum coordination. By utilizing the vertical velocity of the wind profiler in the momentum-coordinated intermediate wind field, grid points with and without wind profiler vertical velocities are selected. In the grid points with wind profiler vertical velocities, the wind profiler vertical velocities in the multi-source aligned data are replaced with vertical velocity components. Meanwhile, in the grid points without wind profiler vertical velocities, the vertical velocity is linearly extrapolated based on the gradient of meteorological wind speed values ​​with decreasing height in the multi-source aligned data according to adjacent grid points, so as to obtain the intermediate wind field after vertical motion correction. Using the air pressure and temperature values ​​in the multi-source aligned data, the air density of each grid in the intermediate wind field after vertical motion correction is calculated by the ideal gas law. Using the air density of each grid, the mass flux of adjacent grid points is calculated and the net mass flux is obtained. For grid points where the absolute value of the net mass flux exceeds a preset threshold, the wind speed and air volume are adjusted to make the net mass flux zero, and the intermediate wind field after mass conservation constraints is obtained. In the intermediate wind field after mass conservation constraints, at the grid points covered by satellite data, the horizontal wind direction is compared with the atmospheric motion vector in the multi-source aligned data to obtain the comparison deviation. When the comparison deviation of continuous grid points exceeds the preset tolerance value, all grid points of the current wind guide layer are rotated as a whole based on the average direction of the atmospheric motion vector to obtain the calibrated intermediate wind field. The reflectivity factor value is obtained from the multi-source aligned data; based on the calibrated intermediate wind field, continuous regions with reflectivity factor values ​​higher than the preset convection threshold are marked to obtain the marked regions; the vertical velocity component is multiplied by the corresponding preset enhancement coefficient according to the magnitude of the reflectivity factor value, and smoothing constraint is applied through the vertical velocity of adjacent grid points to obtain the intermediate wind field with enhanced marking features. The meteorological wind speed and direction vectors of each grid point in the feature-enhanced intermediate wind field are compared with the time-smoothed ground data to obtain the comparison results. When there is a deviation in the comparison results, the radar radial velocity in the feature-enhanced intermediate wind field and the satellite atmospheric motion vector in the multi-source aligned data are used to calculate the correction value through spatial interpolation. The correction value is used to correct the feature-enhanced intermediate wind field to obtain the ground-calibrated intermediate wind field. The meteorological wind direction vector of each grid point in the intermediate wind field calibrated by the stratum is projected onto the radar beam direction and compared point by point with the Doppler radar radial velocity value in the radar data after ground constraint. Grid points with differences exceeding the preset difference deviation range are selected to obtain control points. Based on the control points, Kriging interpolation is used to locally adjust the intermediate wind field calibrated by the stratum to obtain an optimized three-dimensional wind field.

[0039] Furthermore, this step eliminates wind direction projection deviation through momentum coordination, corrects vertical motion using measured vertical velocity and linear extrapolation, and adjusts flux balance by combining mass conservation; it also introduces satellite data to calibrate wind direction and enhances upward motion in convective regions; finally, it improves the physical consistency, vertical motion accuracy, and overall reliability of the three-dimensional wind field through stepwise constraints of ground data and radar radial velocity and local optimization using Kriging interpolation.

[0040] It should be noted that the preset reflectivity screening threshold in this step is set according to the characteristics of the radar band and seasonal changes, preferably between 15dBZ and 35dBZ; while for the identification of severe convective weather, the preset reflectivity screening threshold can be dynamically adjusted to above 40dBZ to distinguish between precipitation particles and non-precipitation echoes.

[0041] Furthermore, the preset deviation range in this step is used to determine whether the angle between the wind direction vector and the radar beam is within an acceptable range. This range can be set according to the error characteristics of radar observation. When the relative error between the wind direction projection caused by the angle deviation and the measured radial velocity exceeds the radar's radial velocity observation error (usually 1-2 m / s), it is determined to be out of range. In this invention, when the angle exceeds ±15°, wind direction rotation adjustment is initiated. When correcting the ground wind field using radar radial velocity, a least-squares variational adjustment method is employed. For each grid point with radar radial velocity observations, a cost function is constructed, consisting of two terms: the first term is the square of the difference between the adjusted wind direction and the initial wind direction obtained by interpolation from ground data; the second term is the square of the difference between the radar-observed radial velocity and the radial velocity calculated from the wind direction and the horizontal wind speed, multiplied by a weighting coefficient. The horizontal wind speed is obtained by interpolation from ground data, assuming that the wind speed remains constant during the adjustment process. The weighting coefficient represents the degree of confidence in the radar observations and is taken as the reciprocal of the variance of the radial velocity observation error. By setting the derivative of the cost function with respect to the wind direction to be determined to zero, a nonlinear equation for the wind direction to be determined can be obtained, which can be solved using the Newton-Raphson iteration method. The initial value of the iteration is the initial wind direction, and the result obtained after the iteration converges is the wind direction after correction by the radar radial velocity. If the calculated wind direction differs from the initial wind direction by more than 90°, it is considered that the radar observation and the background field are in serious conflict. In this case, the background field value is retained, and the point is marked as a quality control point for reference in subsequent steps.

[0042] Specifically, the preset difference deviation range in this step is used to select grid points as control points for Kriging interpolation and to set the radial velocity measurement accuracy based on Doppler radar. This range is usually set to 2-3 times the radar radial velocity observation error (e.g., 3-5 m / s). If it exceeds this range, it is considered that the inverted value of the wind field at that point is significantly inconsistent with the measured value, and local adjustment is required.

[0043] Furthermore, in this step, "adjusting the wind speed and airflow to bring the net mass flux to zero for grid points where the absolute value of the net mass flux exceeds a preset threshold" means that when the absolute value of the net mass flux at a certain grid point exceeds the preset threshold (10⁻⁻⁶), the net mass flux is reduced to zero. 5 When the mass velocity is measured to be kg·m⁻²·s⁻¹, it is determined that there is a mass non-conservation phenomenon in the region. The horizontal wind speed component needs to be adjusted to restore mass balance. The specific adjustment method is as follows: First, calculate the mass flux difference between the grid point and its adjacent grid points to determine the direction of the net mass flux (i.e., net mass inflow or net mass outflow). If the net mass flux is positive (net inflow), it indicates that the air mass flowing into the grid point is greater than the air mass flowing out, and the wind speed component in the inflow direction should be appropriately reduced or the wind speed component in the outflow direction should be increased. If the net mass flux is negative (net outflow), an iterative correction strategy is adopted during the actual adjustment process. Specifically, the required wind speed correction amounts ΔU and ΔV are calculated based on the magnitude of the net mass flux and the air density of adjacent grid points. The correction amounts are distributed to the boundaries adjacent to the grid point according to the proportion of mass flux imbalance, prioritizing the adjustment of the component whose wind direction is consistent with the direction of the net mass flux. The net mass flux is recalculated after each adjustment until its absolute value drops below the preset threshold or reaches the maximum number of iterations (10 times). To avoid new imbalances in adjacent areas caused by local adjustments, the continuous super-relaxation iterative method (SOR) is used to synchronously correct the overall wind field, ensuring the balance of mass flux at each grid point while maintaining the spatial continuity and physical rationality of the wind field. Through the above adjustments, the intermediate wind field after mass conservation constraints is finally obtained.

[0044] Furthermore, the meteorological radar three-dimensional wind field inversion method based on multi-source data fusion also includes the following steps: Step S104: Based on the optimized three-dimensional wind field, perform spatial interpolation and multi-scale feature fusion to expand the optimized three-dimensional wind field and generate a full-space three-dimensional wind field.

[0045] In this embodiment of the invention, based on the optimized three-dimensional wind field, spatial interpolation is performed in the horizontal and vertical directions to obtain a digital elevation model. The interpolated wind field is then corrected for terrain occlusion using the digital elevation model to obtain the corrected wind field. A large-scale background field and small-scale turbulence are then fused and superimposed to generate a full-space three-dimensional wind field with the required resolution and format. Specifically, the required resolution can be 250 meters horizontally, 500 meters vertically, and 6 minutes in time. The required resolution can be a NetCDF file containing U / V / W components, timestamps, spatial coordinates, uncertainty estimates, etc.

[0046] Specifically, Figure 3 A flowchart illustrating the generation of a full-space three-dimensional wind field in the method provided by an embodiment of the present invention is shown.

[0047] In a preferred embodiment of the present invention, the step of performing spatial interpolation and multi-scale feature fusion based on the optimized three-dimensional wind field to expand the optimized three-dimensional wind field and generate a full-space three-dimensional wind field specifically includes the following steps: Step S1041: Perform spatial interpolation in the horizontal and vertical directions on the optimized three-dimensional wind field to obtain the interpolated wind field; Step S1042: Obtain the digital elevation model, and use the digital elevation model to correct the terrain occlusion of the interpolated wind field to obtain the corrected wind field. Step S1043: The corrected wind field is fused and superimposed with the large-scale background field and the small-scale turbulence to obtain the full-space three-dimensional wind field.

[0048] Specifically, the corrected wind field is fused and superimposed with the large-scale background field and small-scale turbulence to obtain a three-dimensional wind field in the entire space. The specific steps are as follows: The region in the corrected wind field is filtered using water vapor channel brightness and temperature data from multi-source aligned data to identify regions where the water vapor channel brightness and temperature data parameters are greater than a preset brightness threshold. The atmospheric motion vectors in the selected regions are used as control points for the background field. Based on the meteorological wind speed values ​​from the multi-source aligned data, control points of the background field whose meteorological wind speed values ​​exceed the preset wind speed tolerance range are removed. At the same time, spatial interpolation is performed on the remaining control points of the background field to generate a large-scale background field. Based on the reflectivity factor values ​​in the multi-source aligned data and the phased array radar scan data, the differences in reflectivity factor values ​​of each grid point are compared in the corrected wind field. The areas formed by grids whose differences exceed the preset fluctuation threshold are marked as fluctuation feature areas. Within the fluctuation feature areas, when the deviation between the meteorological wind direction vector of the large-scale background field and the meteorological wind direction vector of the ground-collected data exceeds the preset direction tolerance range, the change in the meteorological wind direction vector and the fluctuation in the meteorological wind speed value of the current fluctuation feature area are extracted as disturbance information. A small-scale turbulent fluctuation field is generated based on the disturbance information. The large-scale background field and the small-scale turbulent fluctuation field are superimposed on each grid point in the corrected wind field. The weight of the small-scale turbulent fluctuation field is dynamically adjusted according to the reflectivity factor value and the brightness and temperature data of the water vapor channel. The preliminary fused wind field is generated by weighted fusion point by point. The larger the reflectivity factor value and the higher the water vapor content, the greater the weight. The vertical velocity of the wind profiler in the multi-source aligned data is compared point by point with the vertical velocity component of the preliminary fused wind field to obtain the point-by-point comparison result; when there is a deviation in the point-by-point comparison result, the convergence and divergence distribution of the horizontal wind field is calculated using the atmospheric motion vector in the multi-source aligned data; the vertical velocity component of the preliminary fused wind field is adjusted according to the convergence and divergence distribution to obtain a vertically coordinated fused wind field. Based on the reflectivity factor values ​​in the multi-source aligned data, the evolution stage of convective clouds is determined by the increasing and decreasing trends of the reflectivity factor values ​​over consecutive time intervals. The evolution stage with continuously increasing reflectivity factor values ​​is identified as the development stage, and the evolution stage with continuously decreasing reflectivity factor values ​​is identified as the dissipation stage. In the development stage, the pulsation characteristics of wind direction and wind speed in the vertically coordinated fused wind field are preserved. In the dissipation stage, a moving average is used to suppress high-frequency fluctuations in wind direction and wind speed in the vertically coordinated fused wind field, and a time-smoothed fused wind field is obtained. The horizontal wind direction in the time-smoothed fused wind field is projected onto the radar beam direction and verified point by point with the Doppler radar radial velocity value in the multi-source aligned data to obtain the verification result. No processing is performed on grid points with consistent verification results. For grid points with inconsistent verification results, the horizontal wind speed magnitude in the time-smoothed fused wind field is constrained by the meteorological wind speed value in the multi-source aligned data to obtain the verified full-space three-dimensional wind field. The meteorological wind speed, meteorological wind direction vector, and vertical velocity component of each grid point in the verified full-space three-dimensional wind field, as well as the reflectivity factor value in the multi-source aligned data, are formatted and encapsulated according to the timestamp and spatial coordinates in the multi-source aligned data to generate the full-space three-dimensional wind field.

[0049] Furthermore, this step involves separating and constructing a large-scale background field and a small-scale turbulent fluctuation field, and dynamically fusing reflectivity and water vapor data to adjust the weights. Combined with vertical velocity coordination and adaptive smoothing of the convection evolution stage, and then through radar radial velocity verification and formatting encapsulation, a full-space three-dimensional wind field with high spatiotemporal resolution and consistent physical mechanisms is finally generated.

[0050] It should be noted that the preset wind speed tolerance range in this step is used to screen control points of the background field. The range is set based on the error statistical distribution of ground wind speed and reanalysis background field wind speed in the same historical observation. It is preferred to use 3 times the mean error or a preset percentile (such as the 90th percentile) as the rejection criterion.

[0051] Furthermore, the setting rules for the preset direction tolerance range in this step are consistent with the preset angle tolerance range in this invention.

[0052] Specifically, in this step, "dynamically adjusting the weights of the small-scale turbulent fluctuation field based on the reflectivity factor value and the brightness and temperature data of the water vapor channel" refers to: When superimposing and fusing the large-scale background field with the small-scale turbulent fluctuation field, the weight of the small-scale turbulent fluctuation field at each grid point is not fixed, but is adaptively determined according to the reflectivity factor value and water vapor channel brightness and temperature data corresponding to that grid point. Specifically, the larger the reflectivity factor value, the stronger the convective activity and the higher the precipitation particle concentration in the region, and the more significant the impact of small-scale turbulent fluctuations on the wind field. Therefore, it should be given a higher weight. Similarly, the lower the brightness temperature data of the water vapor channel (i.e., the higher the water vapor content), the more it indicates a region with vigorous convection development, and the contribution of the small-scale turbulent fluctuation field should also be increased. In this invention, the reflectivity factor value and the brightness and temperature data of the water vapor channel are normalized to obtain two dimensionless coefficients between 0 and 1. The weighted average (or maximum value) of the two coefficients is then taken as the basic weighting coefficient of the small-scale turbulent fluctuation field. To ensure the physical rationality and numerical stability of the merged wind field, the weighting coefficient also needs to be truncated by an upper threshold (0.7) to avoid small-scale disturbances from excessively dominating the wind field in extreme cases. By using this dynamic weight allocation per grid point, the adaptive fusion of small-scale turbulence features and background large-scale flow field can be achieved, so that the final three-dimensional wind field in the whole space can reflect both the large-scale circulation background and capture the small-scale turbulence details in the strong convection region.

[0053] Furthermore, the meteorological radar three-dimensional wind field inversion method based on multi-source data fusion also includes the following steps: Step S105: Obtain multi-source comparison data to verify and update the full-space three-dimensional wind field.

[0054] It should be noted that multi-source comparison data can be used as a reference true value for independent verification; the comparison data is compared point by point with the full-space three-dimensional wind field data of the corresponding spatiotemporal points, and statistical indicators such as root mean square error (RMSE), mean absolute error (MAE) and systematic bias (Bias) are calculated. At the same time, the spatial distribution field of the error is generated to identify areas with large errors. Based on the verification results, the key parameters of the system are dynamically adjusted, specifically including the following steps: For key parameters such as ground data weight, radar data weight, and satellite data weight in step S102, preset reflectivity screening threshold, preset deviation range, preset convection threshold, and preset difference deviation range in step S103, and preset brightness threshold, preset wind speed tolerance range, preset fluctuation threshold, and preset direction tolerance range in step S104, parameter-error response relationships are established using historical data or offline simulations to determine the sensitivity of each parameter to inversion errors and provide a basis for subsequent adjustments. In regions with large errors, the contribution of each data source to the inversion results is analyzed. If the error of a certain data source (such as satellite data) is significantly higher in this region, its fusion weight in this region is appropriately reduced, while the weights of other data sources (such as radar and ground data) are increased accordingly. The weight adjustment can be achieved using a covariance matching method based on Kalman filtering: treating multi-source comparison data as observations and the inverted wind field as state estimates, the observation noise covariance matrix is ​​dynamically adjusted by comparing the theoretical covariance and the actual covariance of the innovation sequence, thereby correcting the fusion weights. Alternatively, fuzzy logic can be used to correct the weight coefficients in real time based on error feedback, making the fusion results closer to high-precision observations. For the preset angle tolerance and velocity mutation threshold, corrections are made according to the error distribution. For example, if multiple consecutive verifications find that the angle between the wind direction and the radar beam exceeds the original preset range but the actual wind field is consistent with the comparison data, the angle tolerance is appropriately relaxed. If the error is concentrated near certain thresholds, the thresholds are tightened to improve the accuracy of screening. The correction adopts a sliding window statistical method, aiming to minimize the recent verification error, and updates the thresholds to the optimal range in real time. If a machine learning model (such as random forest) is used as the weighted fusion model in step S102, the accumulated multi-source aligned data and the corresponding comparison data are periodically used as new samples to perform incremental training or retraining on the model, optimize the internal parameters of the model (such as decision tree structure, split threshold, etc.), and improve the fusion accuracy; if Kalman filtering is used, the covariance matrix of process noise and observation noise is updated according to long-term error statistics to make the filter more adaptable to the actual environment. The preset enhancement coefficients and preset thresholds involved in the momentum conservation, vertical motion correction, and mass conservation constraints in step S103 can be fine-tuned according to the verification error. For example, if the vertical velocity is generally underestimated in the convection region, the preset enhancement coefficients can be appropriately increased. If there is still an imbalance after adjusting the mass flux, the net mass flux threshold can be tightened and the iteration termination condition can be adjusted. The adjusted parameters are updated to the system configuration for use in subsequent wind field inversion processes; the time of each update, the adjusted parameters, and the changes in verification error are recorded to form a parameter evolution log, providing a basis for long-term system optimization; at the same time, the updated parameters are applied to the next round of data cleaning, feature fusion, and physical constraint processes, forming a continuous optimization closed loop of "inversion → verification → parameter update → re-inversion"; Through the above verification and update mechanism, the system can dynamically adjust key parameters based on actual observation data, effectively reducing inversion errors and improving the adaptability and accuracy of the three-dimensional wind field under different weather conditions and geographical environments.

[0055] In this embodiment of the invention, multi-source comparative data, including vertical velocity of lidar and horizontal wind speed of wind profiler, are acquired, the root mean square error is calculated, and the indicators are compared. Then, the three-dimensional wind field in the whole space is verified and the parameters are adjusted. In subsequent use, continuous verification and updating are carried out.

[0056] Furthermore, Figure 4 An application architecture diagram of the system provided in an embodiment of the present invention is shown.

[0057] In another preferred embodiment of the present invention, the meteorological radar three-dimensional wind field inversion system based on multi-source data fusion includes: The multi-source acquisition and processing unit 101 is used to acquire multi-source data from ground, radar and satellite, obtain multi-source acquired data, and perform data cleaning and spatiotemporal alignment on the multi-source acquired data to obtain multi-source aligned data.

[0058] In this embodiment of the invention, the multi-source acquisition and processing unit 101 acquires multi-source data from ground, radar, and satellite sources, obtaining multi-source acquired data including ground acquisition data, radar acquisition data, and satellite acquisition data. The ground acquisition data includes meteorological wind speed, meteorological wind direction, air pressure, temperature, and vertical velocity of a wind profiler; the radar acquisition data includes Doppler radar radial velocity, reflectivity factor, and phased array radar scan data; and the satellite acquisition data includes satellite cloud image data, atmospheric motion vectors, and water vapor channel brightness and temperature data. The multi-source acquired data is then subjected to anomaly identification and cleaning processing to obtain multi-source cleaned data. Subsequently, the multi-source cleaned data is time-aligned and spatially aligned to obtain multi-source aligned data. Specifically, the anomaly identification and cleaning processing includes: removing outliers in the ground acquisition data where wind speed exceeds the instrument's range; removing noise points in the radar acquisition data where signal strength is below a preset strength threshold; and filtering areas in the satellite acquisition data where cloud cover is below a preset percentage.

[0059] Specifically, Figure 5 A structural block diagram of the multi-source acquisition and processing unit 101 in the system provided by an embodiment of the present invention is shown.

[0060] In a preferred embodiment provided by the present invention, the multi-source acquisition and processing unit 101 specifically includes: The multi-source data acquisition module 1011 is used to acquire multi-source data from ground, radar and satellite sources, and obtain multi-source acquired data. Anomaly identification and cleaning module 1012 is used to perform anomaly identification and cleaning processing on the multi-source collected data to obtain multi-source cleaned data; The time-space alignment module 1013 is used to perform time and space alignment on the multi-source cleaned data to obtain multi-source aligned data.

[0061] Furthermore, the meteorological radar three-dimensional wind field inversion system based on multi-source data fusion also includes: The initial wind field construction unit 102 is used to extract multi-source feature data from the multi-source aligned data and perform weighted fusion on the multi-source feature data to construct an initial three-dimensional wind field.

[0062] In this embodiment of the invention, the initial wind field construction unit 102 extracts multi-source feature data, including ground feature data, radar feature data, and satellite feature data, from multi-source aligned data, determines the weights of ground data, radar data, and satellite data, and then imports them into a preset weighted fusion model. The multi-source feature data is then fused according to the weighted fusion model, ground data weights, radar data weights, and satellite data weights to construct an initial three-dimensional wind field. Specifically, the preset weighted fusion model can employ Kalman filtering or random forest regression. Among the determined ground data weights, radar data weights, and satellite data weights, the ground data weight is fixed at a high weight; the radar data weight is inversely proportional to the distance; and the satellite data weight is directly proportional to the cloud cover.

[0063] Specifically, Figure 6 A structural block diagram of the initial wind field construction unit 102 in the system provided by an embodiment of the present invention is shown.

[0064] In a preferred embodiment of the present invention, the initial wind field construction unit 102 specifically includes: The feature extraction module 1021 is used to extract multi-source feature data, including ground feature data, radar feature data and satellite feature data, from the multi-source aligned data; The weight determination module 1022 is used to determine the weights of ground data, radar data, and satellite data. The initial wind field construction module 1023 is used to fuse the multi-source feature data according to the preset weighted fusion model, the ground data weight, the radar data weight and the satellite data weight to construct an initial three-dimensional wind field.

[0065] Furthermore, the meteorological radar three-dimensional wind field inversion system based on multi-source data fusion also includes: The physical constraint optimization unit 103 is used to apply momentum conservation constraints, vertical motion correction and mass conservation constraints to the initial three-dimensional wind field in sequence using multi-source aligned data, so as to generate an optimized three-dimensional wind field.

[0066] In this embodiment of the invention, the physical constraint optimization unit 103 performs mass conservation constraints on the initial three-dimensional wind field, calculates the wind field divergence, compares it with a preset divergence threshold, and then performs component adjustment correction. It also performs vertical motion correction on the initial three-dimensional wind field and momentum conservation constraints on the initial three-dimensional wind field. By calculating the wind field intensity, it determines whether there is vortex anomaly. When there is vortex anomaly, it performs smoothing filtering to suppress small-scale noise and generates an optimized three-dimensional wind field.

[0067] The wind field expansion processing unit 104 is used to perform spatial interpolation and multi-scale feature fusion based on the optimized three-dimensional wind field, expand the optimized three-dimensional wind field, and generate a full-space three-dimensional wind field.

[0068] In this embodiment of the invention, the wind field extension processing unit 104 performs spatial interpolation in the horizontal and vertical directions based on the optimized three-dimensional wind field, and obtains a digital elevation model. The digital elevation model is then used to correct the terrain occlusion of the interpolated wind field to obtain the corrected wind field. Then, a large-scale background field and a small-scale turbulence are fused and superimposed to generate a full-space three-dimensional wind field with the required resolution and format. Specifically, the required resolution can be 250 meters horizontally, 500 meters vertically, and 6 minutes in time. The required resolution can be a NetCDF file containing U / V / W components, timestamps, spatial coordinates, uncertainty estimates, etc.

[0069] Specifically, Figure 7 A structural block diagram of the wind field expansion processing unit 104 in the system provided in an embodiment of the present invention is shown.

[0070] In a preferred embodiment provided by the present invention, the wind field expansion processing unit 104 specifically includes: Spatial interpolation module 1041 is used to perform spatial interpolation of the optimized three-dimensional wind field in the horizontal and vertical directions to obtain the interpolated wind field. The occlusion correction module 1042 is used to acquire a digital elevation model and use the digital elevation model to perform terrain occlusion correction on the interpolated wind field to obtain the corrected wind field. The fusion and overlay module 1043 is used to fuse and overlay the large-scale background field and small-scale turbulence of the corrected wind field to obtain a three-dimensional wind field in the whole space.

[0071] Furthermore, the meteorological radar three-dimensional wind field inversion system based on multi-source data fusion also includes: The wind field verification and update unit 105 is used to acquire multi-source comparison data and verify and update the full-space three-dimensional wind field.

[0072] In this embodiment of the invention, the wind field verification and update unit 105 acquires multi-source comparison data including lidar vertical velocity and wind profiler horizontal wind speed, calculates root mean square error, compares indicators, and then verifies and adjusts the parameters of the three-dimensional wind field in the whole space. In subsequent use, it performs continuous verification and update processing.

[0073] It should be understood that although the steps in the flowcharts of the various embodiments of the present invention are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the various embodiments may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0074] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0075] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0076] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0077] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for retrieving three-dimensional wind field from meteorological radar based on multi-source data fusion, characterized in that, The method specifically includes the following steps: Multi-source data acquisition is performed from ground, radar, and satellite sources to obtain multi-source acquired data. The multi-source acquired data is then cleaned and spatiotemporally aligned to obtain multi-source aligned data. Multi-source feature data is extracted from the multi-source aligned data, and the multi-source feature data is weighted and fused to construct an initial three-dimensional wind field; Using multi-source aligned data, the initial three-dimensional wind field is sequentially subjected to momentum conservation constraints, vertical motion correction, and mass conservation constraints to generate an optimized three-dimensional wind field; Based on the optimized three-dimensional wind field, spatial interpolation and multi-scale feature fusion are performed to expand the optimized three-dimensional wind field and generate a full-space three-dimensional wind field. Multi-source comparison data is acquired to verify and update the full-space three-dimensional wind field.

2. The method for inverting three-dimensional wind field from meteorological radar based on multi-source data fusion according to claim 1, characterized in that, The process of acquiring multi-source data from ground, radar, and satellite sources, obtaining multi-source acquired data, and performing data cleaning and spatiotemporal alignment on the multi-source acquired data to obtain multi-source aligned data specifically includes the following steps: Conduct multi-source data acquisition from ground, radar, and satellite sources to obtain multi-source data; The multi-source collected data is subjected to anomaly identification and cleaning processing to obtain multi-source cleaned data; The multi-source cleaned data is time-aligned and spatial-aligned to obtain multi-source aligned data.

3. The method for inverting three-dimensional wind field from meteorological radar based on multi-source data fusion according to claim 2, characterized in that, The multi-source acquired data includes: ground-based acquired data, radar-based acquired data, and satellite-based acquired data; The ground-based data includes meteorological wind speed, meteorological wind direction vector, air pressure, temperature, and vertical velocity from a wind profiler; the radar data includes Doppler radar radial velocity, reflectivity factor, and phased array radar scan data; and the satellite data includes satellite cloud imagery, atmospheric motion vectors, and water vapor channel brightness and temperature data. The anomaly identification and cleaning process includes: removing abnormal values ​​in the ground-collected data where the wind speed exceeds the instrument's range; removing noise points in the radar-collected data where the signal strength is lower than a preset strength threshold; and filtering areas in the satellite-collected data where the cloud cover is lower than a preset percentage.

4. The method for inverting three-dimensional wind field from meteorological radar based on multi-source data fusion according to claim 3, characterized in that, The multi-source collected data is subjected to anomaly identification and cleaning processing to obtain multi-source cleaned data, specifically including the following steps: By removing meteorological wind speed, air pressure, and temperature values ​​that exceed the preset normal range from the ground-collected data, the abnormal ground-collected data is obtained. The radar data with noise removed is obtained by removing reflectivity factor values ​​below a preset intensity threshold and their corresponding Doppler radar radial velocity values ​​and phased array radar scan data. The satellite cloud image is divided into grid areas according to latitude and longitude. The atmospheric motion vector and water vapor channel brightness and temperature data with cloud cover ratios below a preset threshold are filtered and removed to obtain satellite data with clear sky areas removed. Ground-based data (excluding anomalies), radar data (excluding noise), and satellite data (excluding clear-sky areas) are matched into multi-source data groups based on timestamps and geographic coordinates. The absolute difference between the meteorological wind direction vector and the atmospheric motion vector within each group is calculated. When the absolute difference exceeds the preset angle tolerance range, the entire data group is discarded to obtain multi-source data after direction verification. The meteorological wind speed and meteorological wind direction vector in the multi-source data after direction verification are continuously taken in time order. The rate of change of wind speed and the amount of change of wind direction vector are calculated. When the rate of change of wind speed exceeds the preset wind speed change threshold or the amount of change of wind direction vector exceeds the preset direction change threshold, the data corresponding to the middle time in the continuous value is removed to obtain the time-smoothed ground data. The time-smoothed ground data and the radar data portion of the multi-source data after direction verification are spatially overlaid. At the overlapping spatial coordinates, when the absolute value of the Doppler radar radial velocity value is greater than the meteorological wind speed value and exceeds the preset deviation range, the radar data at the current coordinates is removed to obtain the radar data after ground constraints. The radar data after ground constraint, the satellite data portion of the multi-source data after direction verification, and the ground data after time-series smoothing are collected and integrated to obtain multi-source cleaned data.

5. The method for inverting three-dimensional wind field from meteorological radar based on multi-source data fusion according to claim 4, characterized in that, The step of extracting multi-source feature data from the multi-source aligned data and weighting and fusing the multi-source feature data to construct the initial three-dimensional wind field specifically includes the following steps: From the multi-source aligned data, extract multi-source feature data including ground feature data, radar feature data, and satellite feature data; Determine the weights for ground data, radar data, and satellite data; The multi-source feature data are fused according to the preset weighted fusion model, the ground data weight, the radar data weight, and the satellite data weight to construct an initial three-dimensional wind field.

6. The method for inverting three-dimensional wind field from meteorological radar based on multi-source data fusion according to claim 5, characterized in that, According to the preset weighted fusion model, the ground data weights, the radar data weights, and the satellite data weights, the multi-source feature data are fused to construct an initial three-dimensional wind field. The specific steps are as follows: Meteorological wind speed values, meteorological wind direction vectors, and spatial coordinates are obtained from ground feature data. The comprehensive weight is the product of the ground data weight and the inverse of the distance between grid points. The meteorological wind speed values ​​and meteorological wind direction vectors are weighted and averaged to generate a preliminary ground wind field distribution. Based on the preliminary ground wind field distribution, Doppler radar radial velocity values, corresponding spatial locations, and radar station coordinates are extracted from radar feature data. The projection of the meteorological wind direction vector in the radar line direction is calculated and compared with the Doppler radar radial velocity values ​​to obtain the relative deviation value. The relative deviation value is used to rotate and adjust the wind direction data of grid points in the preliminary ground wind field distribution, and then spatial interpolation is used to generate radar-corrected wind field data. The atmospheric motion vector and radar height layer information parameters are obtained from the radar-corrected wind field data. The grid points lacking radar coverage in the upper atmosphere are located, and the grid points are matched with the atmospheric motion vector according to the height layer. The overlapping area of ​​radar and satellite is weighted and fused using satellite data weights to obtain the upper-air wind field data after satellite fusion. Using radar height layer information parameters, the vertical velocity of the wind profiler in the satellite-fused upper-air wind field data is matched with grid points, and the atmospheric vertical motion velocity component is estimated by spatial interpolation to obtain a preliminary three-dimensional wind field grid. Based on the radar altitude information parameters, the air density at each altitude level is calculated using the air pressure values ​​in the ground feature data, and the horizontal wind speed in the preliminary three-dimensional wind field grid is corrected for density by combining the temperature values, so as to obtain the thermally adjusted three-dimensional wind field. The vertical velocity increment caused by particle drag is determined based on the reflectivity factor value in the thermally adjusted three-dimensional wind field, and then superimposed on the vertical velocity component of each grid. At the same time, the horizontal wind speed is adjusted according to the mass conservation to obtain the three-dimensional wind field under the influence of precipitation particles. The brightness and temperature data of water vapor channels in the satellite feature data are converted into water vapor content values ​​for each grid point. These values ​​are then multiplied by the horizontal wind speed in the three-dimensional wind field under the influence of precipitation particles to obtain the water vapor flux distribution. The wind field is then adjusted using the water vapor flux distribution to obtain the initial three-dimensional wind field.

7. The method for inverting three-dimensional wind field from meteorological radar based on multi-source data fusion according to claim 6, characterized in that, The process of using multi-source aligned data to sequentially apply momentum conservation constraints, vertical motion corrections, and mass conservation constraints to the initial three-dimensional wind field to generate an optimized three-dimensional wind field specifically includes the following steps: The initial three-dimensional wind field grid points are selected based on the reflectivity threshold. The angle between the wind direction vector and the radar beam is calculated. When the angle between the wind direction vector and the radar beam exceeds the range of the angle deviation, the projection is rotated to zero, and the momentum-coordinated intermediate wind field is obtained. By using the momentum-coordinated vertical velocity of the wind profiler in the intermediate wind field, the grid with data points is replaced with the vertical velocity components of multi-source aligned data, and the grid without data points is linearly extrapolated according to the decreasing gradient of wind speed and height of adjacent grid points to obtain the vertical motion corrected intermediate wind field. By using the air pressure and temperature in the multi-source aligned data, the air density of each grid in the vertical motion corrected intermediate wind field is calculated, and the net mass flux of adjacent grid points is calculated using the air density of each grid to obtain the mass conservation constrained intermediate wind field. At the satellite-covered grid points in the intermediate wind field under mass conservation constraints, the horizontal wind direction is compared with the atmospheric motion vector of the multi-source aligned data. When the tolerance is exceeded, the average direction of the atmospheric motion vector is used as the reference to rotate all grid points of the current wind guide layer to obtain the calibrated intermediate wind field. Based on the calibrated intermediate wind field, for continuous regions where the reflectivity factor value exceeds the preset convection threshold, the vertical velocity component of the multi-source aligned data is multiplied by the corresponding enhancement coefficient according to the reflectivity factor value, and the intermediate wind field with marked features is obtained by smoothing the adjacent grid points. The wind speed and wind direction vectors in the feature-enhanced intermediate wind field are compared with the time-smoothed ground data. When there is a deviation between the wind speed and wind direction vectors in the feature-enhanced intermediate wind field and the time-smoothed ground data, the radar radial velocity of the feature-enhanced intermediate wind field and the satellite atmospheric motion vector in the multi-source aligned data are used to calculate the correction amount through spatial interpolation. The correction amount is then used to adjust the wind speed and wind direction vectors in the feature-enhanced intermediate wind field to obtain the ground-calibrated intermediate wind field. The wind direction vector of the intermediate wind field calibrated by the stratum is projected onto the radar beam direction and compared point by point with the Doppler radar radial velocity in the radar data after ground constraint. Grid points with excessive difference are selected as control points, and Kriging interpolation is used for local adjustment to obtain the optimized three-dimensional wind field.

8. The method for inverting three-dimensional wind field from meteorological radar based on multi-source data fusion according to claim 7, characterized in that, The process of spatial interpolation and multi-scale feature fusion based on the optimized three-dimensional wind field to expand the wind field and generate a full-space three-dimensional wind field specifically includes the following steps: The optimized three-dimensional wind field is spatially interpolated in the horizontal and vertical directions to obtain the interpolated wind field. Obtain the digital elevation model, and use the digital elevation model to correct the terrain occlusion of the interpolated wind field to obtain the corrected wind field; The corrected wind field is fused and superimposed with the large-scale background field and the small-scale turbulence to obtain a three-dimensional wind field in the whole space.

9. The method for inverting three-dimensional wind field from meteorological radar based on multi-source data fusion according to claim 8, characterized in that, The corrected wind field is then fused and superimposed with the large-scale background field and small-scale turbulence to obtain a three-dimensional wind field across the entire space. The specific steps are as follows: Using water vapor channel brightness and temperature data, regions in the corrected wind field that are greater than the preset brightness threshold are selected. The atmospheric motion vectors of the selected regions are used as control points. After removing control points whose wind speeds exceed the tolerance range, interpolation is performed to generate a large-scale background field. Using reflectivity factor values ​​from multi-source aligned data and phased array radar data, regions with reflectivity differences exceeding a preset fluctuation threshold are marked in the corrected wind field. When the deviation between the large-scale background wind direction and the ground wind direction exceeds the directional tolerance, the wind direction change and wind speed fluctuation are extracted as disturbance information to generate a small-scale turbulent pulsation field. The large-scale background field and the small-scale turbulent fluctuation field are superimposed point by point in the corrected wind field, and the weight of the small-scale turbulent fluctuation field is dynamically adjusted according to the reflectivity factor value and the brightness and temperature data of the water vapor channel to generate a preliminary fused wind field. The vertical velocity of the wind profiler in the multi-source aligned data is compared point by point with the vertical velocity component of the preliminary fused wind field. When there is a deviation, the convergence and divergence distribution is calculated using the atmospheric motion vector in the multi-source aligned data and the vertical velocity component is adjusted to obtain a vertically coordinated fused wind field. Based on the continuous temporal increase and decrease trend of reflectivity factor values ​​in multi-source aligned data, the evolution stage of the cloud cluster is determined, and the vertically coordinated fused wind field is temporally smoothed to obtain the temporally smoothed wind field. The horizontal wind direction of the time-smoothed wind field is projected onto the radar beam direction and verified point by point with the Doppler radar radial velocity in the multi-source aligned data to obtain the verified three-dimensional wind field in the whole space. The meteorological wind speed, meteorological wind direction vector, and vertical velocity component of each grid point in the verified full-space three-dimensional wind field, as well as the reflectivity factor value in the multi-source aligned data, are formatted and encapsulated according to the timestamp and spatial coordinates in the multi-source aligned data to generate the full-space three-dimensional wind field.

10. A three-dimensional wind field inversion system based on multi-source data fusion from meteorological radar, characterized in that, The system employs the meteorological radar three-dimensional wind field inversion method based on multi-source data fusion as described in any one of claims 1 to 9. The system includes a multi-source acquisition and processing unit, an initial wind field construction unit, a physical constraint optimization unit, a wind field expansion processing unit, and a wind field verification and update unit, wherein: The multi-source acquisition and processing unit is used to acquire multi-source data from ground, radar and satellite, obtain multi-source acquired data, and perform data cleaning and spatiotemporal alignment on the multi-source acquired data to obtain multi-source aligned data. An initial wind field construction unit is used to extract multi-source feature data from the multi-source aligned data and perform weighted fusion on the multi-source feature data to construct an initial three-dimensional wind field. The physical constraint optimization unit is used to apply momentum conservation constraints, vertical motion correction and mass conservation constraints to the initial three-dimensional wind field in sequence using multi-source aligned data, so as to generate an optimized three-dimensional wind field. The wind field expansion processing unit is used to perform spatial interpolation and multi-scale feature fusion based on the optimized three-dimensional wind field, expand the optimized three-dimensional wind field, and generate a full-space three-dimensional wind field. The wind field verification and update unit is used to acquire multi-source comparison data and verify and update the full-space three-dimensional wind field.

11. The meteorological radar three-dimensional wind field inversion system based on multi-source data fusion according to claim 10, characterized in that, The multi-source acquisition and processing unit specifically includes: The multi-source data acquisition module is used to acquire multi-source data from ground, radar, and satellite sources, and obtain multi-source acquired data. An anomaly identification and cleaning module is used to perform anomaly identification and cleaning processing on the multi-source collected data to obtain multi-source cleaned data. The time-space alignment module is used to perform time and space alignment on the multi-source cleaned data to obtain multi-source aligned data. Specifically, the initial wind field construction unit includes: The feature extraction module is used to extract multi-source feature data, including ground feature data, radar feature data and satellite feature data, from the multi-source aligned data; The weight determination module is used to determine the weights of ground data, radar data, and satellite data; The initial wind field construction module is used to fuse the multi-source feature data according to the preset weighted fusion model, the ground data weight, the radar data weight, and the satellite data weight to construct an initial three-dimensional wind field.

Images